Surveys on stand-alone solar PuE appliances. In: Rauschenbach et al. (2024): Access to (Green) Energy in Rural Africa

Full text

ACCESS TO (GREEN) ENERGY
IN RURAL AFRICA
Online Appendix
2024
Mascha Rauschenbach
Alexandra Köngeter
Kevin Moull
Anna Warnholz

CONTENT
Abbreviations and acronyms ............................................................................................................................viii
1. Literature reviews ...................................................................................................................................... 1
1.1 Are rural energy access programmes pro-poor interventions? ...................................................... 1
1.2Rural energy access and women’s empowerment..........................................................................7
1.3Cost-effectiveness of rural energy access strategies .................................................................... 10
2. Portfolio analysis ...................................................................................................................................... 17
2.1Introduction ................................................................................................................................... 17
2.2Evaluation subject ......................................................................................................................... 17
2.3Reconstruction of the portfolio ..................................................................................................... 18
2.3.1 Intervention identification process .................................................................................. 19
2.3.2 Limitations ........................................................................................................................ 20
2.4Additional findings ......................................................................................................................... 21
2.4.1 Descriptive overview of the portfolio ............................................................................... 21
2.4.2 Overview of technical off-grid approaches ...................................................................... 21
2.4.3 Initial access to modern energy in rural areas.................................................................. 23
2.4.4 Gender equality and women’s empowerment ................................................................ 26
2.4.5 Climate change mitigation ................................................................................................ 27
2.5Conclusion ..................................................................................................................................... 30
3. Data collection of focus groups ............................................................................................................... 32
3.1Benin .............................................................................................................................................. 32
3.2Senegal .......................................................................................................................................... 32
3.3Uganda ........................................................................................................................................... 32
4. Surveys on stand-alone solar PuE appliances .......................................................................................... 33
4.1Identification strategy ................................................................................................................... 33
4.1.1 Before-after comparisons ................................................................................................. 33
4.1.2 Constructing and sampling a control group and an alternative treatment group ........... 34
4.1.3 Cross-sectional analyses ................................................................................................... 35
4.1.4 Difference-in-differences (DiD) analyses .......................................................................... 36
4.2Outcome variables ......................................................................................................................... 36
4.3Causal pathways and operationalisation ....................................................................................... 37
4.4Results for Benin ............................................................................................................................ 39
4.4.1 Descriptive results ............................................................................................................ 39

4.4.2 Before-after comparison .................................................................................................. 40
4.4.3 Matching ........................................................................................................................... 41
4.4.4 Cross-sectional analysis .................................................................................................... 46
4.4.5 Difference-in-differences (DiD) analysis ........................................................................... 52
4.4.6 Perception of outcomes and impacts ............................................................................... 62
4.5Results for Senegal ........................................................................................................................ 63
4.5.1 Descriptive results ............................................................................................................ 63
4.5.2 Matching ........................................................................................................................... 65
4.5.3 Cross-sectional analysis .................................................................................................... 69
4.5.4 Difference-in-differences (DiD) analysis ........................................................................... 76
4.5.5 Perception of outcomes and impacts ............................................................................... 82
5. Mini-grid survey ....................................................................................................................................... 84
5.1Relevance for SDG 7.1 – energy for all by 2030 ............................................................................ 84
5.2Functionality of the mini-grids ...................................................................................................... 85
5.3Economic activities in mini-grid villages ........................................................................................ 89
5.4Productive use potential of mini-grids .......................................................................................... 92
5.5Sustainability and maintenance of the mini-grids ......................................................................... 93
6. References ............................................................................................................................................... 97
Figures
Figure 1 Female employment effects ............................................................................................... 8
Figure 2 Data sources used in the portfolio analysis ...................................................................... 17
Figure 3 PRISMA diagram of case selection for portfolio analysis ................................................. 20
Figure 4 Implemented technical approaches in off-grid interventions .......................................... 22
Figure 5 Share of off-grid and cooking energy interventions in overall energy portfolio
in the years 2000-2022 ..................................................................................................... 23
Figure 6 Tier level targeted by off-grid interventions .................................................................... 23
Figure 7 Lowest targeted tier level of off-grid interventions between the years 2000-2022 ........ 24
Figure 8 Share of rural development in energy and cooking energy interventions
in the years 2000-2022 ..................................................................................................... 25
Figure 9 Share of technical approaches with the marker ´rural development´
within the off-grid portfolio in the years 2000-2022 ........................................................ 25
Figure 10 Gender equality (GG) as an objective in energy and cooking energy interventions ........ 26
Figure 11 Gender equality (GG) objectives in off-grid interventions in the years 2011-2022 ......... 27
Figure 12 Share of interventions contributing to climate change mitigation
in the years 2011-2022 ..................................................................................................... 28

Figure 13 Share of technical approaches in off-grid interventions contributing to
climate change mitigation in the years 2011-2022 .......................................................... 28
Figure 14 Financial commitments for climate change mitigation (KLM-coded)
in the years 2011-2022 ..................................................................................................... 30
Figure 15 Density plot for pre-treatment outcome energy expenses for before and after
matching state, GIZ and non-GIZ treated versus control group, Benin ......................... 45
Figure 16 Density plot for outcome energy expenses for before and after matching state,
GIZ treatment sample, Benin ............................................................................................ 46
Figure 17 Perceived changes in enterprise-related outcomes for farmers, Benin ........................... 62
Figure 18 Perceived changes in enterprise-related outcomes for trade and other, Benin .............. 62
Figure 19 Perceived changes in gender-related outcomes and impacts, Benin .............................. 63
Figure 20 Density plot for outcome energy expenses for before and after matching state,
GIZ and non-GIZ treated, Senegal ..................................................................................... 68
Figure 21 Density plot for outcome energy expenses for before and after matching state,
GIZ treatment sample, Senegal ........................................................................................ 68
Figure 22 Perceived changes in enterprise related outcomes for farmers, Senegal........................ 83
Figure 23 Perceived changes in enterprise related outcomes for trade and other, Senegal ........... 83
Figure 24 Perceived changes in gender-related outcomes and impacts, Senegal ........................... 84
Figure 25 Share of villages by type of energy source before the mini-grids were installed ............ 85
Figure 26 Energy used before the mini-grids were installed by area ............................................... 85
Figure 27 Year of installation and functionality in 2023 ................................................................... 86
Figure 28 Frequency of power outages of the mini-grids during the dry season ............................ 87
Figure 29 Sources of alternative electrical energy used by enterprises and households
(when not connected to mini-grids) ................................................................................. 87
Figure 30 Share of villages by mini-grid operating hours (left)
and by reasons for not operating all day (right) ............................................................... 88
Figure 31 Share of villages by reasons for not being connected to the mini-grids .......................... 89
Figure 32 Types of road connecting the villages .............................................................................. 92
Figure 33 Share of villages by level of accessibility during the rainy season.................................... 93
Figure 34 Share of villages entirely or partially electrified by central grid ....................................... 93
Figure 35 Share of villages by tasks of the mini-grid managers ....................................................... 94
Figure 36 Share of villages by frequency of maintenance of the mini-grid ...................................... 94
Figure 37 Share of villages supplied with fuel on a regular basis by the mini-grid operator ........... 95
Figure 38 Share of villages by repair entity of the mini-grid ............................................................ 95
Figure 39 Time it takes the operator to repair the mini-grid ........................................................... 96

Tables
Table 1 Cost-effectiveness appraisal for rural energy access technologies .................................. 12
Table 2 Data sources ..................................................................................................................... 18
Table 3 Operationalisation of key variables .................................................................................. 19
Table 4 Non-renewable (fossil) energy interventions by financial year........................................ 29
Table 5 Mean comparison tests of outcomes in 2015/2016 (pre-treatment),
GIZ beneficiaries vs. control group, Benin ........................................................................ 39
Table 6 Mean comparison tests of outcomes in 2022/2023 (post-treatment),
GIZ beneficiaries vs. control group, Benin ........................................................................ 40
Table 7 Paired Student’s t-test after vs. before within the GIZ treatment group, Benin.............. 41
Table 8 Mean comparison tests for the matching variables 2015,
GIZ beneficiaries vs. control group, Benin ........................................................................ 42
Table 9 Mean comparison tests for the matching variables 2015,
GIZ and non-GIZ treated vs. control group, Benin ............................................................ 43
Table 10 Results of the matching per outcome in 2015 and 2023,
GIZ beneficiaries vs. control group, Benin ........................................................................ 44
Table 11 Results of the matching per outcome in 2015 and 2023,
GIZ and non-GIZ treated vs. control group, Benin ............................................................ 44
Table 12 Cross-sectional treatment effect on revenue, processing, energy expenses,
number of employees, customer origin, GIZ beneficiaries vs. control group, Benin ....... 47
Table 13 Cross-sectional treatment effect on revenue, processing, energy expenses, number of
employees, customer origin, GIZ and non-GIZ treated vs. control group, Benin ............. 47
Table 14 Cross-sectional treatment effect on sales, assets and food security,
GIZ beneficiaries vs. control group, Benin ........................................................................ 48
Table 15 Cross-sectional treatment effect on revenue, processing, energy expenses, number of
employees, GIZ and non-GIZ treated vs. control group, Benin, farmers only .................. 49
Table 16 Cross-sectional treatment effect on revenue, processing, energy expenses,
number of employees, GIZ beneficiaries vs. control group, Benin, farmers only ............ 49
Table 17 Cross-sectional treatment effect on planted area, assets and food security,
GIZ and non-GIZ treated vs. control group, Benin, farmers only ..................................... 50
Table 18 Cross-sectional treatment effect on planted area, assets and food security,
GIZ beneficiaries vs. control group, Benin, farmers only .................................................. 50
Table 19 Cross-sectional treatment effect on revenue, processing, energy expenses
and number of employees, GIZ beneficiaries vs. control group, Benin, women only ..... 51
Table 20 Cross-sectional treatment effect on revenue, processing, energy expenses
and number of employees, GIZ and non-GIZ treated vs.
control group, Benin, women only ................................................................................... 51
Table 21 Cross-sectional treatment effect on sales, assets and food security,
GIZ beneficiaries vs. control group, Benin, women only .................................................. 52
Table 22 Cross-sectional treatment effect on sales, assets and food security,
GIZ and non-GIZ treated vs. control group, Benin, women only ...................................... 52

Table 23 DiD treatment effect on revenue, processing, energy expenses,
number of employees, Benin ........................................................................................... 53
Table 24 DiD treatment effect on sales, assets, customer origin and food security, Benin ........... 55
Table 25 DiD treatment effect on revenue, processing, energy expenses,
number of employees, Benin, farmers only ..................................................................... 56
Table 26 DiD treatment effect on assets, customer origin and food security,
Benin, farmers only ........................................................................................................... 58
Table 27 Treatment effect on revenue, processing, energy expenses,
number of employees, Benin, women only ...................................................................... 59
Table 28 Treatment effect on sales, assets, customer origin and food security,
Benin, women only ........................................................................................................... 61
Table 29 Mean comparison tests of outcomes in 2019 (pre-treatment),
GIZ beneficiaries vs. control group, Senegal..................................................................... 63
Table 30 Mean comparison tests of outcomes in 2023 (post-treatment),
GIZ beneficiaries vs. control group, Senegal..................................................................... 64
Table 31 Mean comparison tests for the matching variables 2019,
GIZ beneficiaries vs. control group, Senegal..................................................................... 65
Table 32 Mean comparison tests for matching variables, GIZ and non-GIZ treated vs. control
group, Senegal .................................................................................................................. 66
Table 33 Results of the matching per outcome in 2019 and 2023,
GIZ beneficiaries vs. control group, Senegal..................................................................... 67
Table 34 Results of the matching per outcome in 2019 and 2023,
GIZ and non-GIZ treated vs. control group, Senegal ........................................................ 67
Table 35 Cross-sectional treatment effect on assets, food security, quantity sold and
cultivation during dry season, GIZ beneficiaries vs. control group, Senegal .................... 69
Table 36 Cross-sectional treatment effect on assets, food security, quantity sold and
cultivation during dry season, GIZ and non-GIZ treated vs. control group, Senegal ........ 69
Table 37 Cross-sectional treatment effect on revenue, processing, energy expenses,
number of employees, customer origin, GIZ beneficiaries vs. control group, Senegal .... 70
Table 38 Cross-sectional treatment effect on revenue, processing, energy expenses, number of
employees, customer origin, GIZ and non-GIZ treated vs. control group, Senegal ......... 71
Table 39 Cross-sectional treatment effect on planted area, food security,
quantity sold and cultivation during dry season,
GIZ beneficiaries vs. control group, Senegal, farmers only .............................................. 71
Table 40 Cross-sectional treatment effect on planted area, food security,
quantity sold and cultivation during dry season,
GIZ and non-GIZ treated vs. control group, Senegal, farmers only .................................. 72
Table 41 Cross-sectional treatment effect on revenue, processing, energy expenses,
number of employees, customer origin,
GIZ beneficiaries vs. control group, Senegal, farmers only .............................................. 72
Table 42 Cross-sectional treatment effect on revenue, processing, energy expenses,
number of employees, customer origin,
GIZ and non-GIZ treated vs. control group, Senegal, farmers only .................................. 73

Table 43 Cross-sectional treatment effect on planted area, assets, food security
and quantity sold, GIZ beneficiaries vs. control group, Senegal, women only................. 74
Table 44 Cross-sectional treatment effect on assets, food security and quantity sold,
GIZ and non-GIZ treated vs. control group, Senegal, women only .................................. 74
Table 45 Cross-sectional treatment effect on revenues, processing, energy expenses and
number of employees, GIZ beneficiaries vs. control group, Senegal, women only ......... 75
Table 46 Cross-sectional treatment effect on revenues, processing, energy expenses
and number of employees, GIZ and non-GIZ treated vs.
control group, Senegal, women only ................................................................................ 75
Table 47 DiD treatment effect on revenue, energy expenses, number of employees
and food security, Senegal................................................................................................ 77
Table 48 DiD treatment effect on assets, share of customers from outside the municipality
and quantity sold, Senegal................................................................................................ 78
Table 49 DiD treatment effect on revenue, energy expenses, number of employees
and food security, Senegal, farmers only ......................................................................... 79
Table 50 DiD treatment effect on assets, customer origin and quantity sold,
Senegal, farmers only ....................................................................................................... 80
Table 51 DiD treatment effect on revenue, energy expenses and number of employees,
Senegal, women only ........................................................................................................ 81
Table 52 DiD treatment effect on assets, quantity sold and food security,
Senegal, women only ........................................................................................................ 82
Table 53 Share of villages by income-generating activities ............................................................ 89
Table 54 Share of villages by appliances used ................................................................................ 90
Table 55 Main income-generating activities for enterprises connected to the mini-grids ............. 91
Table 56 Share of villages by number of enterprises ...................................................................... 91

ABBREVIATIONS AND ACRONYMS
ATT Average Treatment Effect on the Treated
BMZ Federal Ministry for Economic Cooperation and Development
C Control group
CRS Creditor Reporting System
DAC Development Assistance Committee
DC Development cooperation
DEval German Institute for Development Evaluation
DiD Difference-in-differences
EEBC Energy-efficient biomass cookstove
EnDev Energising Development
ESMAP Energy Sector Management Assistance Program
FE Fixed effects
FOKG Focus group discussion
GBE Green People's Energy for Africa
GG Gender equality marker
GIZ Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH
IQR Interquartile range
IV Instrumental Variables
KfW KfW Development Bank
KLA Rio marker for climate change adaptation
KLM Rio marker for climate change mitigation
LPG Liquefied petroleum gas
MeMFIS Management, finance and information system of the BMZ
MSME Micro, small and medium-sized enterprise
MTF Multi-Tier Framework
O&M Operation and maintenance
ODA Official Development Assistance
OECD Organisation for Economic Co-operation and Development
OLS Ordinary least squares
PRISMA Preferred Reporting Items for Systematic Reviews and Meta-Analyses
PSM Propensity Score Matching
PuE Productive use of energy
PV Photovoltaic
RD Regression Discontinuity
SDG Sustainable Development Goal
SHS Solar Home System
SSA Sub-Saharan Africa
T Treatment group
UN United Nations

1. | Literature reviews 1
1. LITERATURE REVIEWS
1.1Are rural energy access programmes pro-poor interventions?1
Abstract
This paper discusses whether energy access programmes in rural Sub-Saharan Africa (SSA) actually reach the
poor. We examine on and off-grid electrification as well as improved cooking. Pro-poor development requires
that the programmes enable the poor to unlock their productive potential. We therefore focus on the
productive use potentials triggered by energy access programmes such as irrigation.
Our review of the recent evaluation literature informed by our sector and evaluation experience on the topic
also comprehensively covers other potential channels, including education and health. In doing so, we
consider both direct economic benefits to the poor as well as whether indirect effects accrue to the
unconnected via spillovers from among the connected.
We conclude by emphasising that energy access is beneficial for the poor if connections are made affordable
through subsidisation, but indirect effects from productive use and income generation are largely absent.
From a pro-poor perspective, energy efficient biomass cookstoves offer the greatest potential.
Introduction
It is the general consensus that access to electricity is a prerequisite to the provision of basic services and
economic growth. Access to affordable and clean energy, which also includes access to improved cooking
technologies, is therefore envisioned by the Sustainable Development Goals (SDGs) to improve livelihoods in
low-income countries, not least in rural areas of SSA where energy access deficits are most pronounced. But
what does the recent experience tell us about whether energy access programmes lead to pro-poor
development?
In this perspectives article, we discuss the latest evidence from the energy access literature on whether rural
energy access programmes typically reach the poor. This discussion is crucially informed by our experience
working in the energy sectors of different SSA countries and several impact evaluations we have conducted.
We use our sector expertise to critically review the most relevant literature.
We focus this analysis on the question of whether energy access provides the poor with the opportunities to
release their productive potential. Productive use potential to promote capabilities are a key concern of the
different understandings of “pro-poorness” (Kakwani and Pernia 2000; Ravallion and Chen 2003; Day et al.
2016), contributing to the distributional dimension of energy justice (Sovacool and Dworkin 2014; Jenkins et
al. 2016; Munro et al. 2017).
Pro-poor development effects can unfold by providing direct benefits to those gaining access to affordable
and clean energy, who may still be poor despite typically being among the better-off in their communities.
Alternatively, the pro-poor effects can unfold via indirect spillover effects from those who have received
access to those who still lack energy access. For example, small enterprises or health centres that are newly
1 Authors: Jörg Ankel-Peters12*, Gunther Bensch1, Alexandra Köngeter3, Mascha Rauschenbach3 and Maximiliane Sievert1.
Institutions: 1 RWI – Leibniz Institute for Economic Research, 2 University of Passau, 3 DEval - German Institute for Development Evaluation.
*Corresponding author: peters@rwi-essen.de, Hohenzollernstr. 1-3, 45128 Essen, Germany.
Acknowledgements: We are grateful for valuable comments and suggestions by Gerald Leppert and Sven Harten.
Status: Under review. April 2024.

1. | Literature reviews 2
connected to the electricity grid can generate externalities leading to higher income or improved health even
among poor and non-connected households.
We therefore first examine a key factor that determines which socio-economic strata of the population
obtain access in energy access programmes: the typical cost burden of different energy access options. We
then turn to our core discussion of the productive use of energy among enterprises and households in access
programmes with no particular targeting, followed by a discussion of targeted productive use programmes
and their impact potentials.
Productive potentials among the poor can also be enhanced through educational and health impacts,
through direct access or otherwise indirectly through schools and health infrastructure. Lastly, we describe
the impact dimensions for which there is the most evidence regarding their pro-poor impact, namely savings
in terms of money and time.
In discussing these points, we distinguish between improved cooking technologies, more specifically energy-
efficient biomass cookstoves (EEBCs) and clean fuels2, as well as electrification in the form of the centralised
grid, mini-grids and stand-alone solar.
The cost burden of different energy access options
Two features of energy access programmes determine the extent to which poorer strata of the population
are reached: a) the costs of the provided technology and b) the cost-sharing ambition of the programme,
that is, how much the end users must contribute to acquisition costs via fees and prices.
For on-grid electrification programmes, the cost-sharing ambition is generally low but fees are nevertheless
often too high for considerable portions of the target population because of the extremely high costs.
Therefore, only the relatively better-off households get connected. Connection rates “under the grid” across
various countries are typically way below 100 % (Golumbeanu and Barnes, 2013). The impact evaluations
document connection rates in recently-connected areas of 60 % in Rwanda (Lenz et al., 2017), 39 % in
Ethiopia (Bernard and Torero, 2009), less than 30 % in Tanzania (Chaplin et al., 2017) and below 10 % in Kenya
(Lee et al., 2020a) and Burkina Faso (Schmidt and Moradi, 2023).
To assess how responsive households are to connection fees, Lee et al. (2020a) randomised transformers
across 150 communities in Kenya and, subsequently, randomised different subsidy levels for connection fees.
They diagnose a sharp decrease in connection rates as fees increase: while almost all households connected
if connection was free of charge, a subsidy equivalent to 57 % or 29 % increased connection rates by only 23
and 6 percentage points respectively. Therefore, considerable parts of the target population – not least the
poorer sections – do not directly benefit from grid extension programmes.
Programmes promoting stand-alone solar face lower technology costs than grid extension, yet most
programmes follow a market-based paradigm requiring end users to pay cost-covering prices. Programmes
often only subsidise the marketing and perhaps also the market expansion and after-sales infrastructure of
a solar company.
Since the demand for stand-alone solar is very sensitive to the price (see Grimm et al., 2020 and Meriggi et
al., 2021), customers of such programmes are usually from wealthier backgrounds (see Barry and Creti, 2020,
Bensch et al. 2018, and Mukoro et al., 2022).
Improved cookstove promotion programmes are also mostly implemented under the market-based
paradigm but the costs of the technology are much lower, especially for EEBCs. At the same time, the price-
responsiveness of improved cookstove demand is well established – including the diagnosis that many
2 In rural SSA, providing access to improved cooking technologies usually implies the dissemination of low-cost EEBCs. Liquefied Petroleum Gas (LPG)
is barely available in rural SSA and establishing supply chains is prohibitively expensive. Pilot interventions for other clean stoves like gasifier stoves
or biogas have largely failed (Carrión et al., 2021; Puzzolo et al., 2016; Rupf et al., 2015).

1. | Literature reviews 3
households in rural areas cannot afford the investment (Bensch and Peters, 2020; Beltramo et al., 2015;
Munyehirwe et al., 2022; Pattanayak et al., 2019).
Therefore, the cost-sharing approach screens out poorer households, especially in rural areas where the
woodfuel is collected and not purchased, meaning that no monetary savings can occur.
Productive use in energy access interventions
Technically, the grid and sufficiently-sized mini-grids provide powerful electricity that can be used for energy-
intensive machinery and three-phase current. Productive use potentials for grid electrification therefore
constitute the upper bound of productive use potentials related to energy access. If productive use does not
emerge in the wake of grid connection, it is unlikely to emerge when lower-powered stand-alone solar
systems or smaller-scaled mini-grids are promoted.
Overall, recent impact evaluations suggest that the productive use of electricity in newly connected regions
is very limited. Technical potentials are not exploited and consumption remains on a very basic level. What
is more, most enterprises in grid-covered rural areas are shops, bars, tailors and hairdressers, and home
enterprises in households are rare.
They use electricity for lighting and small appliances like electric shavers, entertainment devices and fridges,
sometimes complemented by offering phone charging. The use of grid electricity for irrigation is rare, since
pumps are mostly needed in plots that are too far away from the grid.
Only few enterprises are typically found in rural areas that use powerful electric machinery (in most cases
welders, carpenters and mills). What these enterprises have in common is that they mostly serve local
demand. Products are very rarely sold to regional or urban markets.
These patterns have been observed in several impact evaluations covering both enterprises and households
in different countries. Chaplin et al. (2017) use a difference-in-differences (DiD) design to evaluate a large-
scale grid extension programme in Tanzania, and observe very little productive take-up in enterprises or
through home enterprises.
In a different part of Tanzania but also using a DiD design, Bensch et al. (2019) confirm these findings. Lenz
et al. (2017) evaluate a country-wide grid roll-out programme in Rwanda, using a mixed-methods DiD
identification strategy. They also observe very low consumption levels among enterprises and households,
mostly for lighting, which is generally in line with literature reviews on commercial and domestic productive
use potentials (see, for example, Terrapon-Pfaff et al., 2018, Kizilcec and Parikh, 2020 or Radley and
Lehmann-Grube, 2022). Lee et al.’s (2020a) study also does not find a productive take-up among households
in their sample in Kenya.
Again in Kenya, Taneja (2018) documents another remarkable pattern: even when accounting for the time
since grid connection, electricity consumption levels in areas which have been newly connected to the grid
are drastically lower for households and also notably lower for small enterprises. This underpins the idea that
with progressive electrification, poorer and more remote and poorer regions are gaining access to the
centralised grid.
One valid concern about all these studies is the short-term evaluation horizon: they examine adoption and
impacts two to five years after connection. Masselus et al. (2024) therefore provide a 10-year long-term
evaluation of the Lenz et al. (2017) sample, and find that the consumption and take-up patterns have not
changed. A very modest productive take-up among enterprises was also documented in Benin seven years
after the grid connection (Peters et al., 2011).
Contrary to these studies which use primary, self-collected data, the literature based on secondary data
diagnoses positive impacts of grid-extension electrification (see Lee et al., 2020b for a review). Ankel-Peters
and Schmidt (2023) argue that the key difference is that secondary-data studies cannot use well-specified
interventions but instead have to rely on proxy interventions to identify where electrification happened; they
also note a higher risk of publication bias in secondary-data studies due to there being less incentive to
publish with a null result than with primary-data studies given the high costs of data collection.

1. | Literature reviews 4
Moreover, only few secondary-data studies examine countries or regions in SSA (Hamburger et al., 2019;
Peters and Sievert, 2016). Regardless of the deeper reasons for the divide in findings between these two
types of studies, we argue that the impact evaluation literature referred to above is more relevant for
programme evaluation purposes and intervention specific cost-benefit analysis in rural SSA.
The literature on stand-alone solar and micro-grids confirms our prior of modest productive use and the
impacts of programmes promoting these technologies (see Grimm et al., 2017, Bensch et al., 2018, Kizilec
and Parikh, 2020, and Radley and Lehmann-Grube, 2022 for stand-alone solar and Aklin et al., 2017 for micro-
grids).
Most improved cooking programmes target households, not enterprises (see Grimm and Peters, 2015 for an
efficient cookstove intervention targeting local beer breweries in urban Burkina Faso). One widespread
productive application of improved cookstoves is in restaurants. Here they likely lead to higher productivity,
but probably not in a transformative way – also since, as discussed above, these enterprises mainly cater to
local demand.
Targeted productive use interventions
The previous section showed that programmes providing energy access to the broader population of
households and enterprises in SSA yield no significant impacts on productive uses. We now turn to energy
access interventions that specifically target certain users with high productive potential (see Lukuyu and
Taneja, 2023). For example, a mini-grid intervention could select only villages that host a so-called anchor
customer.
In practice, however, this proves difficult, as such anchor customers in remote areas are rare and can often
not be identified (see Duthie et al., 2023 for a large programme pursuing this approach in Indonesia; see also
Peters et al., 2019). Mini-grid placement according to irrigation potentials is another option (Wamalwa et al.,
2023). Lukuyu et al. (2022) propose a technique to detect such potentials based on existing diesel-fed
irrigation pumps using remote sensing data.
Another approach is to target potential productive users with stand-alone solar-powered machinery. Their
portability makes solar-powered water pumps particularly interesting. Increasing agricultural productivity via
irrigation additionally circumvents the barrier to many other productive uses in rural areas, which is a lack of
market access.
Most parts of rural SSA are well integrated into markets for agricultural products, which means that
expanding agricultural production is more straightforward than for artisanal or manufactured products
(Peters and Sievert, 2016).
While proof-of-concept evidence for solar-powered irrigation does exist (Burney et al., 2010), a broader view
of the thin literature available suggests that promotion at scale is faced with various problems: groundwater
depletion, operational problems ranging from maintenance problems to lack of power on cloudy days
(see Closas and Rap, 2017) and unresolved regulatory questions such as land tenure (Chokkakula and
Giordano, 2013).
Solar-powered water pumps also compete with diesel, which under many circumstances is the more
economically viable energy source from the farmers’ perspective (see Smith and Urpelainen, 2016
and Xie et al., 2021).
In general, diesel is the most important hazard to the impact potentials of this targeting approach, because
potentials for standard productive uses such as milling and pumping are typically already exploited in regions
which are not covered by the grid by using diesel-powered generators, pumps or mills. Less obvious
productive potentials are much harder to identify.
As a result, targeting programmes must either increase the risk they take and aim at not-so-obvious
productive potential which has previously been untapped by diesel-driven appliances. If the programme
otherwise supports the conversion of existing productive uses from diesel to solar or other sources of

1. | Literature reviews 5
electricity, impact potentials are limited to potential reductions in fuel costs (if the electricity is cheaper than
diesel) and environmental benefits.
In sum, reliable evidence on targeted energy access programmes is very scarce. Development practitioners’
priors on such programmes are often shaped by experience and anecdotes from small-scale pilot
interventions. Tacit knowledge like this is not irrelevant, but it needs to be considered that small-scale pilot
interventions are often successful because they are small-scale and pilot. That is, the level of care the
intervention receives from implementers typically cannot be replicated at scale, and once at scale, the
increased supply of the supported production (e.g. irrigated vegetables) may find it harder to find sufficient
demand.
Nevertheless, productive use impact prospects for a programme featuring a well-crafted targeting are
certainly higher than for the typical non-targeted electrification programme. In case they prove successful, a
pro-poor effect is also plausible via external effects on local employment and income. In any case, targeted
approaches will also require subsidies to overcome the limited purchasing power and liquidity constraints in
the high-risk investment setting that rural entrepreneurs typically face.
Health and education
Energy access may not only reach the poor through economic development in the narrow sense, but also by
improving their health and educational status. This might happen through immediate impacts on households
with energy access or indirectly via improved educational and health services in public institutions.
Positive health effects at the household level are possible if dirty kerosene lamps are replaced by electric
lighting (Barron and Torero, 2017). However, the LED lighting transition that rural SSA has experienced over
the past 15 years has changed the baseline situation. That is, kerosene is rarely used any more in rural SSA
and has been replaced by LED torches and small solar lamps (Bensch et al., 2017).
Household-level health effects are most widely discussed for improved cooking technologies. Simple EEBCs
do not reduce smoke emissions to a level that is sufficient to prevent significant health hazards according to
the guidelines of the World Health Organisation.
While there is some suggestive evidence that positive health effects might nevertheless materialise
(La Fave et al., 2021), for example because of a reduced cooking time and hence less exposure to smoke
(see Bensch and Peters, 2015), it seems more prudent not to expect substantive positive health effects in
EEBC dissemination interventions (Bensch et al., 2023).
The alternative option of disseminating clean stoves like LPG or gasifier stoves has proven to be very difficult
in rural SSA. Even in efficacy studies where the clean stoves (or fuels) were delivered free of charge to
households, no health improvements could be observed because many households continued to use dirty
fuels for at least part of their food preparation (Jack et al., 2021; Mortimer et al., 2017).
Educational effects at the household level are most likely to materialise because of improved lighting
conditions for studying at home. This has indeed been observed in Rwanda (Grimm et al., 2017), for example,
but could not be confirmed in Malawi (Stojanowski et al., 2021). In sum, while positive educational and health
effects might materialise at the household level under certain circumstances, it is unlikely that they will be
very pronounced among the poorer strata of the population – also because adoption rates are low.
At the institutional level, it has often been claimed that the lack of electricity in rural health facilities and
schools is a barrier for service provision (see for example Moner-Girona et al., 2021; IEG, 2008). In our
experience (see Lenz et al., 2017, for example), however, even grid-connected schools barely use electricity
for educational purposes.
Schools only operate during daytime hours and computers are not used in class. Teachers – often civil
servants from urban areas – benefit, and anecdotes suggest they tend to stay longer in a village, if connection
to the grid is available. Rural health centres in regions beyond the reach of the grid mostly use solar panels
to fuel basic appliances like a fridge, a steriliser and lighting. Fridges are otherwise also run on kerosene or
gas.

1. | Literature reviews 6
Electricity therefore facilitates services (and lowers costs), but it is not key. Both health services and
education are incredibly important for empowering the poor to develop out of poverty. However, both
services are mainly hampered by the combination of other bottlenecks instead of the lack of access to
electricity alone, including limited budgets and the lack of skilled staff and equipment.
Main impacts on the poor: savings of time and money
We have outlined in the previous sections that in most programmes the poorer households in the respective
target population do not obtain the relevant access, largely for reasons of affordability, and that the main
transformative development effects on income, health and education are limited. We now intend to discuss
the other pro-poor effects which can be identified, at least for those with direct access to energy.
For electricity, most important direct effects are on quality of life and convenience rather than dimensions
with transformative potential. Electricity access can lead to monetary savings if electricity is cheaper per kWh
than what was used at the baseline. Even then, the net savings effect on the household budget is either
negligible or even negative, since new and often consumptive (not productive) energy services are used such
as television.
Savings potentials are also limited for households who only consume lighting, because people use cheap LED
torches or non-branded solar products to meet their basic lighting needs in the absence of access to
electricity (Bensch et al., 2017; Grimm and Peters, 2016; Groenewoudt et al., 2020).
The most accentuated impacts with some transformative potential occur for households which gain access
to an EEBC. Here, the ratio between what can be expected in terms of monetary or time savings and the costs
of an EEBC is clearly higher than for electrification. Firewood savings in rural areas for appropriate EEBCs
which are also regularly used are between 15 and 40 % (see Bensch and Peters, 2015;
Munyehirwe et al., 2022, Mekonnen et al., 2022, and Usmani et al., 2017).
Since the amount of time spent collecting firewood is often in the ballpark of 8 to 12 hours per week,
especially in biomass-scarce regions, it is easy to see that such savings have a noteworthy impact on people’s
time constraints (see for example Krishnapriya et al., 2021). Purchasing firewood is much rarer in rural SSA,
but for those who purchase firewood the time savings are considerable and in the order of the savings
mentioned above.
Conclusion
Energy access is important for a decent living, and providing everybody with electricity and proper cooking
opportunities is a first-order policy priority as reflected in the Sustainable Development Goals (SDG 7).
However, rural energy access alone does probably not lead to transformative impacts in terms of economic
development – even if the non-renewable wisdom also holds true in that electricity is required to enable
endogenous growth.
Most programmes also require the target group to share parts of the costs through connection fees or even
cost-covering market prices, which excludes significant shares of the population from the service. Even for
EEBC programmes, where the disseminated technology comes at much lower costs than for stand-alone solar
or grid extension programmes, the poorer strata of the population mostly abstain from making the
investment.
If these groups at the bottom of the pyramid can be reached, significant subsidies are necessary to bring
down end user prices to affordable levels. The interest of poorer households in improved energy services has
been widely documented. It is affordability that hampers them from adopting these services. Subsidies
targeted at potential productive users are probably also required if productive take-up is to reach levels that
might trigger transformational economic impacts.
Simply replacing existing diesel applications (for example for irrigation pumps) may be a worthwhile
undertaking for grid planning and environmental reasons, but also to develop the market for such solar
appliances. However, in order to have a noticeable impact, targeting procedures need to identify productive

1. | Literature reviews 7
potentials that had not been profitable enough with diesel machinery. The extent to which such income
generation activities then reach the poor or otherwise spill over to poorer sections of the population is
another open question. In fact, from a pro-poor perspective, affordable and therefore subsidised EEBCs offer
the greatest potential.
1.2 Rural energy access and women’s empowerment3
Abstract
In this note, we discuss the link between energy access and women empowerment. The focus is on access to
electricity and improved cooking technologies in rural Sub-Saharan Africa (SSA), where access rates are low.
The discussion is informed by our experience working in the energy sectors of different sub-Sahara African
(SSA) countries, several impact evaluations we have conducted and a critical reading of the literature. We
examine impact dimensions that are not necessarily women empowerment indicators in the narrow sense,
but that are plausibly related to the degrees of freedom of women in SSA societies – and hence changes on
these levels would plausibly empower women within and beyond the household, and to aspire for more
agency in society (see as well Das et al., 2023).
Electrification
It is common sense that electrification of household tasks in today’s industrialized countries contributed to
women empowerment by freeing up time that could be used for other purposes including labor market
participation (Greenwood et al., 2005; Lewis, 2018). Similar effects are generally plausible for contemporary
electrification interventions in the Global South. Several policy reviews and systematic reviews summarize
the academic literature on gender and rural electrification (Das et al., 2020, 2023; Grantham, 2022; Pueyo
and Maestre, 2019; Rewald, 2017; Wilhite, 2017). According to these reviews, three categories of potential
impacts have emerged in the literature: labor market participation and non-agricultural employment, gender
norms, and fertility.
First and foremost, the most influential and most cited studies find considerable effects of rural electrification
on female employment. The seminal paper by Dinkelman (2011) finds an increase of 9 percentage points in
female labor market participation in rural South Africa, relative to a very low baseline of 6 percent
(see Figure 1, Panel A). Another study from rural Nicaragua finds positive statistically significant results on
paid employment of over 20 percentage points (Grogan and Sadanand, 2013; see Figure 1, Panel B). The
broader picture that emerges from the literature is less clear. The two panels in Figure 1 compile all impact
estimates from published work using at least some form of quasi-experimental method, retrieved from the
abovementioned reviews, additional systematic reviews that generally assess electrification impacts
(Ayana et al., 2022; Raitzer et al., 2019), and an additional search among the most recent research. Regarding
female labor market participation presented in Panel A, the remaining literature finds lower and often
insignificant effects (note also that the right-hand scales for studies using Instrumental Variables (IV) are
double the left-hand scales for other studies). Interestingly, even zooming into more specific types of
employment, such as paid and off-farm employment, does not yield the strong and significant effects as
found in the flagship reference, Grogan and Sadanand (2013). All these studies assess rather short to medium
term effects, with few exceptions covering impacts accruing more than 6 years after electrification.
3 Authors: Jörg Ankel-Peters1,2, Gunther Bensch1, & Maximiliane Sievert1, Mascha Rauschenbach3, Anna Warnholz3.
Institutions: 1 RWI – Leibniz Institute for Economic Research, 2 University of Passau, 3 DEval – German Institute for Development Evaluation.
Status: Under review, June 2024.

1. | Literature reviews 8
Figure 1 Female employment effects
Panel A: Labour market participation
10
percentage points
510
main coefficient,
0
-5
0
-10
[FE] [DiD] [PSM] [FE] [PSM] [FE] [RD] [IV] [IV]
x = 72%x = 58%x = 24%x = 16%x = 6% x = 71%
Peru India South Africa India South Nigeria
Africa
Dasso &
Fernandez
Rathi & Vermaak Fetter & Usmani Dinkelman Salmon &
Tanguy
Panel B: More specific forms of female employment
20
-50
-25
0
25
[FE] [DiD] [PSM] [PSM] [FE] [IV] [PSM] [FE]
x = 25
hrs/week
x = 5.5 hrs/day x = 1027 hrs/year x = 1270
hrs/year
Peru Nepal India South Africa
Dasso &
Fernandez
Banerjee et al. Rathi & Vermaak +
Khandker et al.
Rathi &
Vermaak
main coefficient (as % of control mean)
-1500
-750
0
750
1500
[IV]
x = 30.7
min/day
[IV]
x = 1 day/year
Nicaragua India
Grogan &
Sadanand
van de Walle et
al.
50
Source: DEval, own figure
Note: The charts differentiate between Instrumental Variables (IV) studies on the right-hand scales and studies using other methods
of causal identification on the left-hand scales. Point estimates (bars) and the 90 % confidence intervals (whiskers) indicate the margins
of error. Circles imply that no point estimate and confidence interval can be derived for the respective study. x refers to the baseline
or control group mean. Additional information provided below each bar include the authors, the study country, and the employed
technique for data analysis, ranging from [weighted] Propensity Score Matching (PSM[-w]) over Difference-in-Differences (DiD),
Regression Discontinuity (RD) and Fixed effects (FE) to Instrumental Variables (IV) estimations (see, for example, Gertler et al., 2016).
Panel B additionally mentions the type of female employment assessed. * refers to studies with both rural and urban populations –
all other studies include only rural households

It is also important to make two qualifications about the IV approach used in the two highlighted studies with
the most pronounced effect estimates. First, these estimates are not readily comparable to estimates derived
from other methods and harder to interpret, because they refer to specific sub-populations that usually
cannot be readily characterized (the so-called local average treatment effect [LATE] issue, see Marbach and
Hangartner 2020, for example). Second, this method has recently been subject to substantive methodological
criticism, also with regards to electrification (see Bensch et al., 2020 and Lee et al., 2020). One might
furthermore be concerned about publication bias, since the incentives to pursue towards publication with a
null result in this type of work are low (see Brodeur et al., 2020). In sum, we argue that for program evaluation
purposes it is very hard to make project-specific claims based on this evidence.

1. | Literature reviews 9
Evidence on gender norms and fertility is much scarcer, with the dominating studies being Jensen and Oster
(2009) and La Ferrara et al. (2012), examining the effect of television on women’s status and fertility in India
and Brazil. Both studies rely on context-specific events for identification of causal effects, namely the arrival
of cable TV in rural India and the surge in soap operas at a Brazilian TV channel around the 1990’s and 2000’s.
La Ferrara et al. (2012) find significant reductions in fertility, especially among women with lower socio-
economic status. In numbers, they are on average 0.5 percentage points less likely to give birth in any given
year. Jensen and Oster (2009) find reductions in women's tolerance of spousal violence and increases in
women's autonomy, but their results only hold true for educated women (see also Iversen and Palmer-Jones,
2019). Fertility and contraceptive usage have furthermore been investigated in Grimm et al. (2015), Peters
and Vance (2011) and Peters et al. (2014), with mixed results.
Beyond methodological concerns, it should be stressed that virtually all these studies look at countries
outside of Africa, with South Africa as an exception, arguably a very particular country in SSA. There are
important reasons to question the transferability of these findings to rural SSA. The underlying mechanism
in many labor market studies is that appliance ownership and usage in newly electrified communities
liberates women’s time that was absorbed by household chores before. In rural Africa, though, there is a
growing consensus that uptake of time-saving appliances (in particular electric stoves, washing machines and
fridges) in newly connected rural areas is virtually zero (Bensch et al., 2019; Chaplin et al., 2017;
Lee et al. 2016, 2020b; Lenz et al. 2017; Taneja, 2018). Rural dwellers do not have the financial means to
make such investments and electrification does not lead to meaningful economic development
(see Ankel-Peters et al., 2024a). Labor market effects that materialize through time savings are hence
unlikely. Regarding the likelihood of effects on gender norms and fertility, transferability is also questionable.
TV usage once households are connected is much lower in SSA than in Asia and Latin America. Moreover, the
full coverage of mobile communication in rural Africa also prior to electrification leads to a different baseline
situation in terms of access to information, making significant impacts in these regards much less likely.
Our focus in this note is on on-grid electrification, which is also what the literature studying the gender-
related effects of electricity access is mostly looking at. Our verdict that relevant effects from a project
evaluation perspective are unlikely to materialize in rural SSA, also holds for stand-alone solar and certainly
for Pico-PV (see Grimm et al., 2017). While women also benefit from these technologies, the resulting effects
are not transformative.
Improved cookstoves
Potential impacts of improved cookstoves in rural areas inherently accrue to women because cooking related
chores – the cooking process itself as well as the fuel provision – are borne by women in SSA. Virtually all
household in rural SSA use biomass for cooking, mostly firewood. The firewood is collected and depending
on biomass availability in the region this can entail collection work of 10 or more hours per week
(Krishnapriya et al., 2021). The firewood is used in open fires or very simple metal stoves with the emitted
smoke being extremely harmful for people’s health (WHO 2016). Hence, improved or clean cooking solutions
can affect women’s health and workload. These are important dimensions affecting women’s capacities and
degrees of freedom.
We distinguish between low-cost energy-efficient biomass cookstoves (EEBC) and clean fuels and clean
stoves (LPG and gasifier stoves). EEBC are low-cost technologies, between 10 and 50 EUR, for which well-
adapted versions exist for many regions in rural SSA that have also proven to be intensely used (see Jeuland
et al., 2020). Such well-adapted EEBC have found to reduce firewood consumption considerably, depending
on the baseline cooking situation and the specific EEBC type between 10 and 50 %. This, in turn leads to a
significant workload reduction for firewood collection (see Bensch and Peters, 2015; Krishnapriya et al., 2021;
Munyehirwe et al., 2022). Many EEBC types are also more convenient to use and reduce cooking duration
(see Bensch and Peters, 2015; Krishnapriya et al., 2021). In terms of health, EEBC do not meet air pollution
standards of the World Health Organization and are generally not considered clean. While there is some
suggestive evidence that even EEBC lead to improved health status (Bensch and Peters, 2015;

1. | Literature reviews 10
LaFave et al., 2021), for example due to cooking in better ventilated places (Lenz et al., 2023), it is more
advisable not to expect positive health effects of EEBC via a reduction of air pollution.
To effectively address the detrimental exposure to cooking related air pollution, clean stoves or fuels are
required. Hitherto, however, there is no proof-of-concept evidence that LPG or gasifier stoves lead to a
measurable improvement of people’s health status. Even in efficacy studies where the clean stoves (or fuels)
were delivered free of charge to households, no health improvements could be observed (Jack et al., 2021;
Mortimer et al., 2017). This is mainly due to the so-called stacking of different stoves, including the traditional
ones, and due to inappropriate applications, but perhaps also because of ambient air pollution.
Conclusion
Rural energy access improves the quality of life considerably and it particularly affects the lives of women in
rural areas. The much deeper improvements, though, come from improving cooking technologies as
compared to rural electrification. From a project evaluation perspective, effects of electrification in SSA on
female labor market participation and gender norms through information and media exposure are possible,
but probably much less pronounced than what a project can measure and what a project would consider
relevant. The reason is that electric appliance uptake is substantially lower in SSA than Asia and Latin America
and the baseline situation in terms of access to information thanks to mobile phone coverage is better today
in SSA than it used to be when countries in Asia and Latin America were electrified, making significant impacts
in this regard much less likely. The labor market effects observed in some studies mostly materialize through
time saving effects of electrification – something which cannot be expected in rural SSA.
If anything, such effects might unfold for improved cooking interventions, because here, also simple EEBC
can lead to considerable effects on women’s time use. These diagnosed improvements for the livelihoods of
women in SSA related to EEBC dissemination are noteworthy and while not of immediate relevance for
women empowerment, they create the freedom to exploit other opportunities, employment-related or
beyond.
1.3 Cost-effectiveness of rural energy access strategies4
Abstract
Quantitative benchmarks for cost-effective provision of rural energy access are difficult to obtain because
deployment costs vary across technologies, contexts, and technical assistance approaches – but crucially also
across sustainability assumptions. As an alternative, this policy perspective provides a qualitative cost-
effectiveness assessment of different energy access strategies.
That is, we discuss the different cost factors and additionally account for differences in impact potentials
across rural energy access options. We include on-grid and off-grid electrification and improved cooking
technologies. The focus is on rural sub-Saharan Africa (SSA), where energy access rates are low.
We diagnose largely disappointing impacts of high-power electrification technologies, turning stand-alone
solar into the more cost-effective electrification strategy in that setting. We conclude by emphasising the
high impact-cost ratio for energy-efficient biomass cookstoves.
4 Authors: Jörg Ankel-Peters1,2*, Gunther Bensch1, Kevin Moull3, Mascha Rauschenbach3 & Maximiliane Sievert1.
Institutions: 1 RWI – Leibniz Institute for Economic Research, 2 University of Passau, 3 DEval – German Institute for Development Evaluation.
*Corresponding author: joerg.peters@rwi-essen.de, Hohenzollernstr. 1-3, 45128 Essen, Germany.
Keywords: energy access, rural electrification, modern cooking energy, sub-Saharan Africa. JEL: H54, O21, O33. Acknowledgements: We are
grateful for valuable comments and suggestions by Gerald Leppert and Sven Harten.
Status: Under review, March 2024.

1. | Literature reviews 11
Introduction
Investment requirements to reach Sustainable Development Goal 7 – universal access to electricity and
modern cooking energy – are high. The level of investment needs to grow by at least 35 percent to reach the
goal by 2030 or even more than 100 percent if climate goals are also to be met (IEA and IFC, 2023).
While public investment flows are scarcer due to the multiple crises around the world, more public funds are
pledged to climate mitigation and adaptation agreements, such as the Loss and Damage Fund established at
the UN Climate Change Conference in 2022, COP27.
This paper reviews costs and benefits of rural energy access options to improve the effectiveness of public
resources in achieving the universal energy access goal and subsequent poverty impacts. We consider on-
and off-grid electrification and improved cooking technologies. The regional focus of our analysis is on Sub-
Saharan Africa (SSA).
Quantitative benchmarking is difficult and hence we provide a qualitative cost-effectiveness assessment,
taking into account capital costs and technical assistance costs as well as impact potentials. This assessment,
therefore, borrows from cost-benefit analysis. The discussion is informed by our experience working in
various SSA energy sectors and several impact evaluations we have conducted. It is hence a perspective
paper, supported by substantive evidence.
The different technologies under scrutiny serve different purposes. Most notably, electricity is rarely used
for cooking in SSA, even in areas where the grid is available. Households traditionally use firewood and
charcoal as cooking fuels and improved or clean cooking solutions are based on more efficient biomass
combustion technologies or Liquefied Petroleum Gas (LPG).
Intervention assessments therefore rarely compare the cost-effectiveness of electrification and improved
cooking to justify the investment. This comparison is nevertheless important since donor investments into
these two policies often come from the same portfolios.
Qualitative cost-effectiveness assessment
In Table 1 we provide an overview of costs and benefit potentials for the different energy access technologies.
First, we compile indicative figures for capital costs of different energy access technologies (see column 1).
Note that while these numbers cannot be taken at face value in any specific context, they broadly reflect the
incurred acquisition costs regardless of who pays.
Depending on the cost-sharing model, the national government, donor agencies and end users may
contribute in varying proportions. For example, the lion’s share of grid connection costs is typically borne by
the government and its utility, often supported by an international donor, while the end users contribute a
smaller share through the connection fee. In many improved stove and off-grid solar programmes, in
contrast, it is the end user who bears the entire capital costs by purchasing the appliance at a cost-covering
price.
Here, a donor agency’s contribution typically is to provide technical assistance, for example to support
institutionalising market structures. Such technical assistance costs come on top of the numbers in column
(1). This is an important caveat for the interpretation of Table 1 because technical assistance requirements
vary considerably between the different technologies as indicated in column (2), from fairly low for grid
extension to very high for the mostly nascent mini-grid sector.

1. | Literature reviews 12
Table 1 Cost-effectiveness appraisal for rural energy access technologies
(1) (2) (3) (4) (5)
Cost per connection, Technical assistance Energy service potential, Impact evidence Technical lifetime;
in US$ requirement by MTF tier* operation & maintenance
(O&M) intensity
Electricity
Pico-PV
20-50
Medium
Tier 1
convenience and improving daily routines,
2-5 years
minor monetary or time savings
mainly to establish
one spotlight
impact potential constrained by baseline
low O&M intensity
market structures
and one charging slot
technology, typically dry-cell battery driven LED
Stand-alone
100-700
Medium
Tier 1-2
convenience and improving
5+ years
Solar Home System daily routines, minor time
(SHS)
saving impacts
e.g. depending on
mainly to establish
multiple light points, phone
productive use impacts restricted to small shops
medium O&M intensity
capacity market structures charging, radio and extended working hours, mainly by limited
and potentially TV or fan
power
Mini-Grid
750-2,000
High
Tier 3-5
few impacts beyond convenience
10-20 years
and time saving impacts
e.g. depending on
because most
Tier 2 + any medium-power
impacts constrained by low electricity consumption
high O&M intensity
connection rates and countries lack appliances such as due to limited affordability
anchor customers enabling regulatory refrigerators; partly also (to buy electric appliances), lacking market access
framework high-power appliances, for enterprises, and if mini-grids do not operate all
such as mills day
On-Grid
500-1,500
rather low
Tier 4-5
few impacts beyond convenience
20+ years
and time saving impacts
due to long-standing
Tier 3 and high-power
impacts constrained mainly by low electricity
low to medium
local know-how appliances, such as mills consumption due to limited affordability O&M intensity
(to buy electric appliances) and lacking market
access for enterprises

1. | Literature reviews 13
(1)
Cost per connection,
in US$
(2)
Technical assistance
requirement
(3)
Energy service potential,
by MTF tier*
(4)
Impact evidence
(5)
Technical lifetime;
operation & maintenance
(O&M) intensity
Cooking
Energy-efficient
biomass cookstoves
5-30 medium to high
(low in urban areas)
to establish market
structures
Tier 0-2 reduced woodfuel consumption and
subsequent impact on monetary
and time savings
2-5 years
low to medium
if provided for free
higher energy efficiency;
reduction in air pollution
no low to medium
O&M intensity
Advanced
biomass cookstove
75-100 very high to establish
market structures
Tier 2-3 even stronger reduced fuel consumption and thus
on time savings but mixed results regarding air
pollution
2-5 years
medium
if provided for free
(to train users)
higher fuel efficiency
lower emissions
and impacts constrained mainly by continued use of
traditional stoves (‘stove stacking’), inappropriate
use, and limited availability/high cost of processed
woodfuels (pellets)
medium O&M intensity
Liquefied Petroleum
Gas (LPG)
20-100 very high to
establish market
structures,
particularly LPG
supply chain in rural
areas
Tier 4-5 strong reduction of traditional fuel use and thus on
time savings, but so far no evidence for reducing
health risks (mainly due to continued use of solid
fuels and ambient air pollution)
5+ years
plus fuel costs high
free
if provided for high fuel efficiency
to zero emissions
and low adoption typically constrained
fuel supply (e.g. to rural areas)
cylinder purchase
due
and
to high costs
need of bulk
of low O&M intensity

1. | Literature reviews 14
(1)
Cost per connection,
in US$
(2)
Technical assistance
requirement
(3)
Energy service potential,
by MTF tier*
(4)
Impact evidence
(5)
Technical lifetime;
operation & maintenance
(O&M) intensity
Biogas digester 500-1,500 very high Tier 4-5 similar to LPG, in addition co-benefits for
agricultural households (fertiliser) and zero
monetary fuel costs
10-20 years
e.g. depending due to need to high fuel efficiency and low virtually all programmes in Africa have low high O&M intensity
on capacity change behaviour, emissions, lighting as co-adoption rates or have failed due to high up-front
including keeping benefit and maintenance costs, and not enough cow dung
cattle in stable and water
Source: DEval, own table, Sources on costs: Lighting Global et al. 2022 (SHS); AMDA 2022, BloombergNEF 2020, ESMAP 2022 (Mini-grids); Lee et al. 2020b, BloombergNEF 2020 (on-grid),
ESMAP 2020, Jeuland et al. 2018 (cooking)
Note: *The tiers of energy access are described in the Multi-Tier Framework (MTF), developed by ESMAP. Energy access is measured on a tiered spectrum, from tier 0 (no access)
to tier 5 (the highest level of access), differentiated by household electricity and domestic cooking energy

1. | Literature reviews 15
Table 1 also features the technologies’ energy service potential (column 3) and a qualitative assessment of
impacts effectively observed in programmes across SSA (column 4). Broadly speaking, energy-efficient
biomass cookstoves have proven to deliver in terms of their expected impacts, that is, a reduction of
fuelwood consumption and hence, of monetary expenditures or firewood collection time, depending on
whether the woodfuel is purchased or collected (Jeuland et al., 2020).
These are noteworthy impacts in most settings in rural SSA, especially since the reduced workload for
firewood collection mainly accrues to women (Bensch and Peters, 2020; Berkouwer and Dean, 2022; Das et
al., 2023; Jeuland et al., 2021). The evidence on reducing household air pollution induced by woodfuel usage,
however, is more pessimistic, not only for efficient biomass cookstoves but also for LPG and clean gasifier
stoves.
While it remains true that only exclusive use of clean stoves has the potential to fully eliminate household
air pollution, clean stoves today usually fail to fully displace all dirty stoves in a household (Pope et al., 2021).
Nevertheless, the impact potentials of improved cooking are impressive relative to the low costs, in particular
for efficient biomass cookstoves. Among energy access technologies, improved cooking therefore clearly has
the best cost-benefit ratio, even under very conservative assumptions.
For electrification, the case is much more complex. Different technologies have, in theory, different impact
potentials, but empirically impacts do not differ in most cases. For higher-power technologies, technically
possible demand potentials are not exploited, and consumption remains on a very low level.
In other words, impacts of on-grid electrification and mini-grids on the household level in most of rural SSA
are not very different from most solar home systems. Some small enterprises in newly grid-connected areas
do use electric machinery (typically shops, tailors, hairdressers, welders and carpenters), but the restricting
factor for economic development is market access – which is very limited in most villages in SSA.
New and larger enterprises rarely emerge as a result of the village’s connection to the grid. The major
difference between the technologies is that grid access would allow demand growth to give way to
endogenous local growth. In contrast, solar home systems lack this possibility due to the absence of high-
power electricity. It is also important to note that if there is productive use potential in a not-yet-connected
village, electricity is already there, by means of diesel generators in most cases. It is rare that demand
potentials are not exploited and only emerge once the grid is available.
These patterns have been observed in well-crafted impact evaluations in several SSA countries (Bensch et al.,
2019, 2022; Chaplin et al., 2017; Lee et al., 2020b; Masselus et al., 2024; Lenz et al., 2017; Peters et al., 2011;
Schmidt and Moradi, 2023; Taneja, 2018). The absence of considerable economic impacts in electrification
programmes is also documented in literature reviews (Bos et al., 2018; Lee et al., 2020a; Peters and Sievert,
2016).
The effects of small-scale solar are mostly at the level of convenience and improving daily routines like
studying at home and housework (Grimm et al., 2017, 2020; Stojanowski et al., 2021). There are only minor
impacts on time savings and monetary expenses (while amortisation is not always a given), and no
discernable positive effects on productive and commercial uses.
Women certainly also benefit from the convenience and housework chore effects of small-scale solar, but
this is hardly transformative and certainly much less pronounced than the considerable time savings and
workload reductions that have been diagnosed for energy-efficient biomass cookstoves.
It is also worth emphasising that some of the positive evidence on small-scale solar stems from a baseline
situation in which costly and dirty kerosene lamps have been replaced. This, however, is no longer the
baseline situation in most settings in SSA because LED torches and non-branded solar has replaced kerosene
virtually everywhere (Bensch et al., 2017), reducing impact potentials for small-scale solar considerably.
When scaled from small-scale solar to larger solar home systems, effects change with regards to a few
appliance types that are additionally used, mostly TV sets and fans. Productive and commercial use is still

1. | Literature reviews 16
very limited (Aklin et al., 2017; Bensch et al., 2018; Kizilcec and Parikh, 2020; Lee et al., 2016; Radley and
Lehmann-Grube, 2022), for the same reasons as outlined for grid electrification above.
Beyond the classical impact categories typically scrutinised in impact evaluations, we stress that large
infrastructure like the power grid also has more subtle but potentially important effects, which are under the
radar of such impact evaluations.
For example, the availability of the grid might provide a sense of social inclusion. It might affect participation
in elections, and via television also lead to modernisation, not least with respect to gender norms (Tanner
and Johnston, 2017). Such effects are much likelier (although largely unknown) for on-grid electrification and
perhaps functioning mini-grids than for stand-alone solar and improved cookstoves.
Yet, while these are noteworthy effects, and perhaps detectable on the country level, they are probably too
subtle to decisively affect the cost-benefit analysis on the intervention level, given the high investment costs
of grid extension.
Two important additional considerations need to be taken into account when interpreting the indicative cost
numbers in Table 1: sustainability and low connection rates. Sustainability of on-grid electrification could
indeed alter the cost-benefit analysis. When looking at a very long-term perspective, say, 15 or 20 years, the
power grid is much more likely to provide sustainable electricity access than decentralised electricity sources,
which need to be maintained and replaced.
The maintenance of the grid is a decades-old fair for utilities, and they make sure the grid operates, in the
long run – on behalf of and financially supported by the government. Organising maintenance for mini-grids
and, even more so, for stand-alone solar, is a much more difficult task (Duthie et al., 2023; Peters et al., 2019;
Tenenbaum et al., 2014; Zigah et al., 2023).
In other words, the costs of sustainable provision to the services in Table 1 might well alter the relationship
between the different technologies, in favor of grid extension. Nonetheless, this will probably not change the
qualitative verdict that grid extension into rural areas is very expensive given the low demand and impact
expectations.
This verdict is further substantiated by the importance of connection rates for costs per connection: costs
per connection easily run into thousands of EUR if only a fraction of households in a village in fact connect,
as it was observed, for example, in recent impact evaluations with connection rates below 30 % in Tanzania
(Chaplin et al., 2017) and below 10 % in Kenya (Lee et al., 2020b) and Burkina Faso (Schmidt and Moradi,
2023).
Conclusion and policy implications
All things considered, from a cost-effectiveness perspective it is hard to make a case for grid extension. The
same arguments however also apply for mini-grids, especially when sustainability considerations are taken
into account (unless mini-grids are targeted at areas far away from the grid with a high-demand anchor
customer).
It is therefore likely that the most cost-effective electricity access solution in most rural areas will be stand-
alone solar. However, broadening the scope beyond electrification, energy-efficient biomass cookstoves
stand out in terms of cost-effectiveness since they clearly deliver important impacts – especially for women
–at very low costs.
Also from a sustainability standpoint, low-maintenance models of energy-efficient biomass cookstoves exist
that do not require major investments until they need to be replaced.

2. | Portfolio analysis 17
2. PORTFOLIO ANALYSIS5
2.1 Introduction
The portfolio analysis, conducted as part of the evaluation “Access to (Green) Energy in Rural Africa” gives
an overview of the German-funded energy portfolio in Africa, including its development over time. It focuses
on off-grid energy access and cooking energy in rural Africa.
It is intended to supplement the portfolio (Chapter 5) in the evaluation report. The portfolio analysis is based
on various data sources, as illustrated in Figure 2.
Figure 2 Data sources used in the portfolio analysis
Source DEval, own figure
Portfolio analysis
MeMFIS
database
CRS database
Literature reviews
Project
documents
2.2 Evaluation subject
The focus of the portfolio analysis is on German Official Development Assistance (ODA) interventions as
mandated by the Federal Ministry for Economic Cooperation and Development (BMZ) within the energy
sector (Organisation for Economic Co-operation and Development (OECD) Development Assistance
Committee (DAC) Creditor Reporting System (CRS) purpose codes 23x), as well as within the funding area of
cooking energy implemented in Africa between 2000 and 2022, in which off-grid technologies were also
implemented (see also Chapter 4.2.2 in the evaluation report).6 German ODA mandated by the BMZ is also
referred to as German Development Cooperation (DC).
The interventions identified were mostly aimed at providing first access to energy and improving access to
modern energy for rural populations in Africa using off-grid approaches or a combination of off-grid and on-
grid approaches. The nature of the interventions varied, but they frequently included elements such as
consultations with stakeholders, support for the enabling and framework conditions, financing, subsidies and
the provision of technologies, incentive and support mechanisms for households and Micro, Small and
Medium-Sized Enterprises (MSMEs) to acquire and use the technologies, the removal of market barriers, the
creation of technical and quality standards, advice and subsidies for operators, the creation of monitoring
systems and training.
5 Author: Dr Gerald Leppert
6 2000 was chosen as the start date because there was a steady increase in the financial volume in the energy sector in the German portfolio from
2000 onwards. The most recent data available at the time of analysis is for 2022. Furthermore, the data in the MeMFIS database starts from 2000.

2. | Portfolio analysis 18
2.3 Reconstruction of the portfolio
In the sampling frame of the evaluation, we included all German ODA-eligible interventions commissioned
by the BMZ in the energy and cooking energy sector between 2000 and 2022 in rural Africa. An additional
inclusion criterion was that the interventions supported off-grid technologies, though not exclusively so,
since some of these interventions were integrated with those that also targeted on-grid access.
Retrieving the relevant interventions for the sampling frame was challenging. The management, finance and
information system (MeMFIS) database of the BMZ, which is at the core of the analyses, does not provide
information on whether an intervention is conducted in rural areas or whether it is an off-grid intervention.
It was also challenging to determine whether an intervention was relevant for the African context due to the
existence of global and sectoral interventions that include Africa but also implement beyond it. Therefore,
we applied several strategies to approximate the energy and cooking energy portfolio for rural Africa.
In a first step, we used the OECD DAC CRS purpose codes as identifiers wherever possible. Then, we requested
documents for the relevant interventions from the implementing organisations Deutsche Gesellschaft für
Internationale Zusammenarbeit (GIZ) and KfW Development Bank (KfW) based on the definition of the
evaluation subject. This was an iterative process, where we discussed the interventions belonging to the
evaluation subject with the implementing organisations based on MeMFIS intervention lists that were
prefiltered by DEval.
Afterwards, we conducted a qualitative document analysis based on the documents that we received. Lastly,
we conducted manual keyword searches in the MeMFIS database to identify cooking energy interventions
as well as off-grid interventions. The reconstruction of the portfolio relied on the data sources listed
in Table 2.
Table 2 Data sources
Data source Description
BMZ MeMFIS •
•
•
The BMZ MeMFIS database serves as a data source of ODA-eligible
bilateral development interventions under the responsibility of the
This includes bilateral development cooperation.
This includes the time period 2000-2022.
BMZ.
OECD DAC CRS7 •
•
•
The Creditor Reporting System (CRS) to OECD DAC on ODA
of all DAC member countries.
This includes bilateral and multilateral development assistance, including core
contributions to multilateral organisations.
This includes contributions from other ministries besides the BMZ.
KfW and GIZ
project documents
•
•
The implementing organisations KfW and GIZ have documentation of the interventions
implemented.
68 interventions could be linked to the MeMFIS dataset. Of these, 40 could be identified
as off-grid interventions for which detailed information on the technical approaches
implemented was available.
Literature reviews
Source: DEval, own table
•Three literature reviews to assess the state of the evidence on: the cost-effectiveness
rural energy access strategies, whether rural energy access programmes are pro-poor
interventions, and rural energy access and women’s empowerment.
of
7 The CRS data was used for the portfolio (Chapter 5) of the evaluation report, not for the supplementary information in this online appendix.

2. | Portfolio analysis 19
We operationalised the key variables using the identification strategies described above to approximate the
sampling frame and identify the relevant interventions. Table 3 gives an overview of the key variables.
Table 3 Operationalisation of key variables
Variable Definition and computation
Africa •All interventions in the funding region Africa (“Förderregion Afrika”).
Africa plus (default) •
•
All interventions in the funding region Africa.
+ interventions that were not assigned to a funding region such as global or sector
interventions, providing they were relevant for energy access in Africa. On the one
hand, some of the interventions were already included in the data request to the
implementing agencies. On the other hand, we performed a manual extraction of
relevant interventions in the MeMFIS database, since not all relevant interventions
had the (German) keyword “Afrika” in the database descriptions.
Energy interventions •All interventions with purpose code 23x (1st, 2nd, 3rd or 4th purpose code).
Cooking energy •
•
•
Interventions with the purpose code 32174 (“Clean cooking appliances
manufacturing”). There were 0 interventions coded as 1st purpose code and 2
interventions coded as 2nd purpose code.
+ Keyword search in the MeMFIS database in German and English (29 keywords in
different spellings, e.g. “Koch”, “stove”, “energy-efficient”, “EEBC”, etc.), followed
by a manual exclusion of non-applicable hits.
+ Interventions in the document analysis that were coded as cooking energy
interventions.
Sampling frame “energy
and cooking energy”
•All interventions in the variables “Energy interventions“+ “Cooking energy“.
Off-grid interventions •
•
•
Interventions in the document analysis that were coded as off-grid or off-grid/on-
grid interventions.
+ Keyword search in the MeMFIS data base in German and English (38 keywords in
different spellings, e.g. “off-grid”, “netzunabhängig”, “Solarpumpe”, “PuE stand-
alone” etc.), followed by a manual exclusion of non-applicable hits.
+ Interventions with the purpose codes 23631 (“Electric power transmission and
distribution (isolated mini-grids)”) and 23231 (“Solar energy for isolated grids and
standalone systems”).
Rural development •Dummy variable based on cross-sectoral
(“Ländliche Entwicklung”).
identifier “Rural development”
Rural interventions
Source: DEval, own table
•
•
Keyword search in MeMFIS database.
Manual extraction of rural interventions.
2.3.1 Intervention identification process
The identification process of the interventions relevant for the evaluation subject was based on the
operationalisation of key variables as outlined in Table 3. Firstly, interventions in the region “Africa plus”
were selected and then narrowed down to interventions in the sampling frame “Energy and cooking energy
interventions”.
In subsequent steps, we identified “off-grid” and “cooking energy” interventions and the relevant technical
approaches of off-grid interventions. The Preferred Reporting Items for Systematic Reviews and Meta-
Analyses (PRISMA) diagram in Figure 3 summarises the selection process for determining the energy and
cooking energy portfolio for rural Africa.

2. | Portfolio analysis 20
We analyse all energy interventions implemented by the German DC in the period under review. However, a
large proportion of these interventions can be attributed to the energy access portfolio.
In operationalising energy access interventions, we rely on the CRS purpose codes that can be attributed to
energy access according to Bazilian et al. (2011). In addition, we used the purpose codes that the
interventions have been assigned to in the GIZ and KfW documentation that was provided to us.8
Figure 3 PRISMA diagram of case selection for portfolio analysis
Source: DEval, own figure
Note: These categories include eight interventions that contained both off-grid and cooking energy interventions
Measures with unique BMZ number,
year 2000 - 2022, (n = 73,302)
Identification
Measures in region "Africa plus"
with unique BMZ number,
year 2000 - 2022, (n = 24,967)
48,335 interventions without
implementations in Africa
Off-grid
interventions
(n = 45)1
Cooking energy
interventions
(n = 25)1
24,524 not in energy and
cooking energy sector
Units of analysis
Detailed analysis of technical approaches
implemented in interventions
based on project document analysis (n = 40)
Energy interventions
that were neither
off-grid nor cooking energy
(n = 381)
Energy and cooking energy interventions
in region "Africa plus"
with unique BMZ number
year 2000 - 2022, (n = 443)
2.3.2 Limitations
Since the MeMFIS and CRS data frequently lacked detailed information on the characteristics of
implementation and technical approaches, we had to rely on identifiers and markers such as purpose codes,
the Rio marker for climate change mitigation (KLM) and the marker (GG) for gender equality.
8 When operationalising energy access interventions, the evaluation is based on the purpose codes that can be attributed to energy access according
to Bazilian et al (2011). However, their publication was based on the old C RS purpose codes. The updated list of purpose codes according to their
definition include 23110 (“Energy policy and administrative management”), 23181 (“Energy education/training”), 23182 (“Energy research”), 232*
(“Energy generation, renewable sources”), 233* (“Energy generation, non-renewable sources”), 23630 (“Electric power transmission and
distribution (centralised grids)”) and 23631 (“Electric power transmission and distribution (isolated mini-grids)”), 23640 (“Retail gas distribution”).
We also added the newer code 23642 (“Electric mobility infra structures”). Compared to the paper by Bazilian et al. (2011), w hich focused exclusively
on electricity production and gas distribution, this evaluation also acknowledges other sources of energy such as fuels, including 23641 (“Retail
distribution of liquid or solid fossil fuels”). In addition, the purpose code 23183 (“Energy conservation and demand-side efficiency”) was also
included, as this was present in interventions sent by the implementing organisations for off-grid interventions in rural Africa. This purpose code
comprises two interventions in the document analysis and a total of 28 interventions in the sampling frame.

2. | Portfolio analysis 21
One challenge with this is that even though an intervention may be assigned up to four different purpose
codes, most interventions only have one (89.2 % for the subject of this evaluation) or even none
(about 1.0 %). Furthermore, we were limited by the fact that there is no identifier in MeMFIS for interventions
that were conducted in rural areas and there is no reliable identifier or purpose code delineating off-grid
interventions.
Adding to this challenge, full-text information, including data for the variable “objectives” (“Zielsetzung”) of
the interventions, was often not available. Given the challenges in identifying the interventions that ought to
be included in the portfolio analysis we partly relied on the documentation from the relevant implementing
organisations. However, this data was incomplete, since some of the interventions were still being
implemented.
The reliance on the established markers or purpose codes may lead to results that are imprecise. Purpose
codes create a clear delineation of interventions which may not correspond to reality. For instance, energy-
relevant interventions with similar approaches are also implemented in energy-adjacent sectors such as
agricultural interventions that promote energy access through solar irrigation pumps, as is confirmed by the
evaluation’s reference group. These interventions may not be labelled with a 23x purpose code despite their
energy-related nature.
Nonetheless, the evaluation relies on purpose codes for several reasons. First, it is important to apply the
benchmarks to test the evaluation questions for the sector they were intended for, namely the energy sector
and the area of cooking energy. Second, using purpose codes ensures that the analysis and methods are
transparent and replicable.
2.4 Additional findings9
2.4.1 Descriptive overview of the portfolio
As already described in the evaluation report, German DC implemented a total of 443 interventions in the
energy sector, including cooking energy, in Africa between 2000 and 2022 (see Figure 3). These figures
include interventions that were in the funding region Africa and those that were relevant for Africa
(e.g. global and sector interventions that were also implemented in Africa). Only 45 interventions (10.2 %)
could be identified as off-grid and 25 (5.6 %) as cooking energy.
Financially, the commitments have been growing for the entirety of the energy and cooking energy sector
since 2000 as well as for off-grid interventions. While the financial commitments for the whole sector were
60.2 million euros in 2002, they increased to 649.7 million euros in 2022. Commitments for off-grid
interventions exhibited more fluctuations. For example, they added up to 12.4 million euros in 2000,
3.5 million euros in 2001, increased to 103,7 million euros in 2021 and fell back to 59.2 million euros in 2022.10
Additionally, the share of commitments for off-grid interventions increased from 10.6 % in the period
2000-2002 to 13.5 % in the period 2019-2022.
2.4.2 Overview of technical off-grid approaches
There are numerous technical approaches implemented in off-grid interventions (see Table 3 in the
evaluation report), which serve different purposes and provide different tier levels of energy access.
9 All findings on the productive use of energy for the evaluation criterion relevance can be found in the evaluation report
10 If a category (e.g. off-grid interventions) only includes a few interventions, fluctuations can be expected because the commitments are associated
with the financial year of the intervention. Large, multi-annual interventions may produce a spike in the associated financial year. In this regard, it
is important to note that interventions (by unique BMZ number) may comprise several entries in the MeMFIS database, which frequently have
separate values for the commitments and financial years.

2. | Portfolio analysis 22
During the detailed qualitative content analyses of intervention documents the following technical
approaches were assessed: mini-grids, stand-alone systems, Pico-PV systems, cooking energy, and on-grid
approaches that were embedded in off-grid interventions. We excluded interventions that only focused on
improving on-grid access.
Figure 4 shows the number of technical approaches implemented in off-grid interventions. The analysis was
based on 68 interventions in the document analysis, out of which 40 interventions were identified as off-grid
interventions that provided sufficient information on the technical approaches implemented. 14
interventions (35.0 %) of the off-grid interventions sought to implement only one technical approach, 13
(32.5 %) implemented two technical approaches and 13 (32.5 %) aimed for three to five approaches.
Mini-grids were implemented most frequently, namely 27 times (67.5 % of off-grid interventions assessed in
detail in the document analysis), followed by 25 on-grid implementations that were embedded in off-grid
interventions (62.5 %). 16 of the mini-grid interventions also implemented on-grid interventions. Stand-alone
systems were implemented 20 times (50.0 %). Cooking energy interventions were implemented eight times
(20.0 %). Pico-PV systems were implemented only six times (15.0 %).
Figure 4 Implemented technical approaches in off-grid interventions
Source: DEval, own figure. Document analysis, region Africa plus, n = 68. Interventions may implement multiple technical approaches.
Figure 5 shows the share of cooking and off-grid interventions in the overall energy portfolio over time.
010 30 4020
Number of interventions
40
27
25
20
6
8
Off-grid interventions
On-grid (within off-grid interventions)
PicoPV
Mini-grid
Stand-alone
Cooking energy

2. | Portfolio analysis 23
Figure 5 Share of off-grid and cooking energy interventions in overall energy portfolio
in the years 2000-2022
Source: DEval, own figure, MeMFIS 2000-2022, region Africa plus, n = 443 Energy and cooking energy interventions, 8 interventions
are double-counted as they included off-grid and cooking energy implementations
0 %
1 %
2 %
3 %
4 %
5 %
6 %
7 %
8 %
9 %
10 %
11 %
12 %
13 %
14 %
15 %
Percentage of interventions
2000 – 2002 2003 – 2006 2015 – 2018 2019 – 2022
Four-year episodes between 2000 – 2022
Share of off-grid
Share of cooking energy
Regression line for off-grid
Regression line for cooking energy
2007 – 2010 2011 – 2014
2.4.3 Initial access to modern energy in rural areas
Concerning the relevance of the targeted tier level of a development intervention for SDG 7, in theory, initial
access could be achieved at the highest tier level of 5. Realistically, initial access to modern energy in rural
areas in Africa is more likely to be situated at a lower tier level. For that reason, the portfolio analysis assesses
the lowest-implemented tier level over time as an indicator for the first access.
Since most interventions implement more than one technology, the range of likely targeted tier levels can
be broad. Figure 6 gives an overview of the tier level targeted by the technical approaches implemented in
off-grid interventions. 20 % of interventions implement approaches that start at tier level 0, 40 % of
interventions start at tier 1, 30 % at tier 2 and 10 % at tier 3. None of the interventions sought to exclusively
implement technical approaches at tier 4 or 5.
Figure 6 Tier level targeted by off-grid interventions
Source: DEval, own figure, Off-grid interventions in document analysis, literature review, region Africa plus, n = 40
Note: The range of tier levels depends on the technical approaches implemented.
Tier 0 Tier 1 Tier 2 Tier 3 Tier 4 Tier 5
20% of interventions
5% of interventions
35% of interventions
30% of interventions
10% of interventions

2. | Portfolio analysis 24
An important finding is the trend of initial access over the past few years. Figure 7 shows the trend over time
regarding initial access based on the results of the literature reviews and the summary of approaches suitable
for initial access (according to Table 3 in the evaluation report).
As can be seen, the average lowest tier level targeted by the interventions in a given year has substantially
increased since 2001. This is because the number of implemented technical approaches which are unsuitable
for initial access are able to provide a higher minimum tier level, e.g. mini-grids. These have been growing
faster than approaches like Pico-PV, cooking energy and stand-alone systems.
Figure 7 Lowest targeted tier level of off-grid interventions between the years 2000-2022
Source: DEval, own figure. Document analysis, off-grid interventions, region Africa plus, n = 40. Interventions may implement multiple
technical approaches. The figure represents the mean of the lowest targeted tier level of off-grid interventions.
0 3
Average of lowest tier level
2000 2005 2010 2015 2020
Year
21
The availability of modern energy in rural areas plays an important role in achieving first access to modern
energy. Rural development is a cross-sectoral indicator in MeMFIS. This indicates whether an intervention is
geared towards rural development and/or food security (BMZ, 2010).
According to the guidelines of the BMZ, the rural development label for energy can be awarded for
interventions that improve the political and institutional framework conditions in the energy sector, as well
as for the provision of energy infrastructure. We use the variable “rural development” to determine whether
an intervention contributes to rural development or not. We apply it as a proxy to assess how well rural areas
are catered for regarding energy access.
As Figure 8 demonstrates, only 19.2 % of the energy and cooking energy interventions aim to contribute to
rural development, while 71.1 % of off-grid interventions and 32.0 % of cooking energy interventions aim to
contribute to rural development. However, these results are tentative, since the cross-sectoral indicator for
rural development was not specifically designed for rural energy access.
Mean of lowest tier level Regression line,
mean of lowest tier level

2. | Portfolio analysis 25
Figure 8 Share of rural development in energy and cooking energy interventions in the years 2000-
2022
Energy and
cooking energy
Off-grid
interventions
Cooking
energy
0 %
10 %
20 %
30 %
40 %
50 %
60 %
70 %
80 %
90 %
100 %
19%
71%
32%
No rural development
Rural development
Percent
Interventions
Source: DEval, own figure, MeMFIS 2000 - 2022, region Africa plus, n = 443 Energy and cooking energy interventions.
8 interventions are double-counted as they included off-grid and cooking energy implementations.
Figure 9 is based on the technical approaches in off-grid interventions which were reported on in more detail
in the document analysis. According to the documents, interventions of cooking energy (87.5 %) and Pico-PV
(83.3 %) are most often marked as contributing to rural development. However, also according to the
documents analysed, over 70 % of the other technical approaches in off-grid interventions also contribute to
rural development.
Figure 9 Share of technical approaches with the marker ´rural development´ within the off-grid
portfolio in the years 2000-2022
Technical approaches
Cooking
energy
Pico-PV Mini-grid Stand-alone On-grid
(within Off-grid)
0 %
10 %
20 %
30 %
40 %
50 %
60 %
70 %
80 %
90 %
100 %
88 % 83 % 78 % 75 % 72 %
No rural development
Rural development
Percent
Source: DEval, own figure, Document analysis 2000 - 2022, region Africa plus, n = 40 off-grid and cooking energy
interventions. Interventions may implement multiple technical approaches.

2. | Portfolio analysis 26
2.4.4 Gender equality and women’s empowerment
An important criterion in achieving gender equality and improving the livelihoods of women and girls is
whether the needs of women and girls are considered in the objectives, conception and implementation of
interventions in the energy and cooking energy sector.
This was analysed by using the cross-sectoral marker of gender equality (GG) (“Gleichberechtigung der
Geschlechter”) in the MeMFIS database. GG2 represents gender equality as the primary (main) objective of
the development intervention, and GG1 represents a secondary objective. Gender equality has been coded
reliably since 2003.11 For this reason, our analyses cover the period from 2003-2022.
Throughout the entire energy and cooking energy sector only 32.1 % of interventions target gender equality,
with only 1.2 % marking it as a main objective and 30.9 % as a secondary objective.
As Figure 10 shows, off-grid interventions exhibit a comparatively high percentage of interventions
promoting gender equality at 64.3 %, though only as a secondary objective. The share of interventions
promoting gender equality was highest among cooking energy interventions at 81.8 %, with 13.6 % as a main
objective and 68.2 % as a secondary objective.
Figure 10 Gender equality (GG) as an objective in energy and cooking energy interventions
Source: DEval, own figure, MeMFIS 2003-2022, region Africa plus, n=408. 8 interventions were double-counted as they included both
off-grid and cooking energy implementations
Interventions
Energy and
cooking energy
Off-grid
interventions
Cooking
energy
0 %
10 %
20 %
30 %
40 %
50 %
60 %
70 %
80 %
90 %
100 %
1.2 % 0.0 % 13.6 %
30.9 %
64.3 %
68.2 %
Missing coding
GG0 coding
GG1 coding
GG2 coding
Percent
None of the 72 off-grid interventions analysed in the document analysis in detail state gender equality as
their main objective. The percentage of gender equality as a secondary objective varied substantially
between the technical approaches as illustrated in Figure 11. While all cooking energy interventions (100 %)
promoted gender equality, only 69.2 % of mini-grid and 68.4 % of stand-alone systems interventions stated
gender equality as an objective. Among the off-grid interventions studied, the technical approach with the
lowest share of gender equality was Pico-PV (50.0 %).
11 The years 2000 – 2002 still contain a substantially higher number of missing values in the GG coding compared to later periods.

2. | Portfolio analysis 27
Figure 11 Gender equality (GG) objectives in off-grid interventions in the years 2011-2022
Source: DEval, own figure, MeMFIS, Document analysis 2011-2022, region Africa plus, n=35 off-grid and cooking energy interventions.
Interventions may implement multiple technical approaches
Cooking
energy
Mini-grid Stand-alone On-grid
(within Off-grid)
Pico-PV
0 %
10 %
20 %
30 %
40 %
50 %
60 %
70 %
80 %
90 %
100 %
100 %
69 % 68 % 60 % 50 %
GG0 coding
GG1 coding
GG2 coding
Percent
Technical approaches
2.4.5 Climate change mitigation
The energy and cooking energy sector is relevant for achieving climate change mitigation goals. Furthermore,
no interventions which are reliant on fossil fuels are expected to have been implemented as a result of the
Paris Agreement in 2015.
Figure 12 shows that most interventions in the energy sector contribute to climate change mitigation. The
Rio marker for climate change mitigation (KLM) indicates whether climate change mitigation is a principal
objective (KLM2) or a significant objective (KLM1) of the intervention.
As shown in the evaluation report, 87.8 % of energy interventions have climate change mitigation as an
objective, with most interventions (70.5 %) having mitigation as the principal objective. All off-grid
interventions (100 %) contribute to climate change mitigation, with 82.9 % having it as a principal objective.
Among the interventions concerning cooking energy, the share contributing to mitigation is 75 %, with
41.7 % having it as the principal objective.

2. | Portfolio analysis 28
Figure 12 Share of interventions contributing to climate change mitigation in the years 2011-2022
Energy and
cooking energy
Off-grid
interventions
Cooking
energy
0 %
10 %
20 %
30 %
40 %
50 %
60 %
70 %
80 %
90 %
100 %
71 % 83 % 42 %
17 %
17 %
33 % Missing coding
KLM0 coding
KLM1 coding
KLM2 coding
Percent
Interventions
Source: DEval, own figure, MeMFIS 2011-2022, region Africa plus, n=319 Energy and cooking energy interventions. 8 interventions
are double-counted as they included off-grid and cooking energy implementations
The more detailed analyses of technical approaches within off-grid interventions (see Figure 13) shows that
100 % of the interventions were marked as being relevant for climate change mitigation (KLM1 and KLM2).
Among these, all on-grid interventions (within off-grid interventions) and interventions that implemented
Pico-PV systems were labelled as having climate change mitigation as their principal objective as well as
90.5 % of the interventions implementing mini-grids. Among stand-alone systems and cooking energy, over
80 % were labelled as having mitigation as their principal objective.
Figure 13 Share of technical approaches in off-grid interventions contributing to climate change
mitigation in the years 2011-2022
Source: DEval, own figure. MeMFIS, Document analysis 2011-2022, region Africa plus, n=35 off-grid and cooking energy interventions.
Interventions may implement multiple technical approaches. One figure does not add up to 100 % due to rounding.
Cooking
energy
Mini-grid Stand-alone On-grid
(within off-grid)
Pico-PV
0 %
10 %
20 %
30 %
40 %
50 %
60 %
70 %
80 %
90 %
100 %
86 % 90 % 80 % 100 % 100 %
14 % 10 % 20 %
KLM1 coding
KLM2 coding
Percent
Technical approaches

2. | Portfolio analysis 29
In response to the Paris Agreement in 2015, German DC has been expected to halt its support for
interventions that include non-renewable (fossil) energy. Table 4 shows that this has been achieved. Since
2003, only 8 interventions implemented non-renewable (fossil) energy and only one after 2018.
A closer look at this sole intervention after 2018, which is called “Crowdfunding for Energy Inclusion”, reveals
that it was misclassified in MeMFIS, and is in fact coded as KLM2 (climate mitigation as a principal objective)
and supports renewable energies.
Table 4 Non-renewable (fossil) energy interventions by financial year
Financial year Number of interventions supporting non-renewable (fossil) energy
2003 1
2008 2
2009 1
2010 1
2013 1
2015 1
2019 1
Total 8
Source: DEval, own table, MeMFIS 2000-2022, region Africa plus, n=443 energy and cooking energy interventions
Note: Non-renewable (fossil) energy was defined as purpose code 233 “Energy generation, non-renewable sources” with all sub-codes,
purpose code 23640 “Gas distribution to the customer (retail)” and 23541 “Distribution of liquid or solid fossil fuels to the customer
(retail)”
Figure 14 illustrates the financial commitments for climate change mitigation over time. The commitments
for climate change mitigation for the entire energy and cooking energy sector have increased substantially.
While they amounted to 66,5 million euros in 2011, they increased to 618,8 million euros in 2022. The
contributions for climate change mitigation for off-grid interventions and cooking energy interventions have
also increased, but at a lower level.
The commitments for off-grid interventions have fluctuated significantly, and were 1,1 million euros in 2011,
jumped to 67,4 million euros in 2012, and then reached 103,7 million euros in 2021 and 51,0 million euros in
2022. The commitments to climate change mitigation through cooking energy interventions increased from
1,2 million euros in 2011 and 13,8 million euros in 2012 to 27,4 million euros in 2022.

2. | Portfolio analysis 30
Figure 14 Financial commitments for climate change mitigation (KLM-coded) in the years 2011-2022
0
100
200
300
400
500
600
Commitments in million euro
2011 2013 2015 2017 2019 2021
Year
20 %
40 %
60 %
80 %
100 %
Share of commitments
2011 2013 2015 2017 2019 2021
Year
Off-grid
interventions
All energy and
cooking energy
Cooking energy
Source: DEval, own figure. MeMFIS 2011-2022, region Africa plus, n=319 Energy and cooking energy interventions. 8 interventions
are double-counted as they included off-grid and cooking energy implementations. Commitments include reprogramming and
approved amounts. KLM2 contributions were applied 100 %, KLM1 were 50 % discounted. Euro values represent fixed amounts in
euros at the base value 2015
Note: Left panel: KLM commitments in million euros; Right panel: Share of KLM commitments in overall commitments
Considering the entirety of the financial commitments to energy, the aggregate share of KLM commitments
constitutes 73.6 % for the whole sector of energy and cooking energy for the entire period 2011-2022. Among
these interventions, off-grid interventions have a slightly higher share of 79.7 % while cooking energy
interventions have a substantially lower share of 62.5 % (over the entire period 2011-2022).
Like the absolute financial commitment, the share of objectives towards climate mitigation (KLM) for energy
and cooking energy interventions has increased since 2011, when KLM markers were properly coded for the
first time, from 64.7 % in 2011 to 84.7 % in 2022 (see Figure 14). Unlike the development of commitments in
absolute figures over time, the various subsectors of energy show similar shares of climate-related
commitments (see Figure 14).
2.5 Conclusion
The portfolio analysis provides several insights into the state of and the trends visible in the energy and
cooking energy portfolio in Africa, in which off-grid technologies were also implemented. A main finding of
the analysis is that the share of off-grid interventions in the energy sector is comparatively low and has only

2. | Portfolio analysis 31
slightly increased over the last twenty years, though the share of financial commitments for off-grid
interventions increased.
Regarding cooking energy interventions, which indicate particularly positive effects on women, they have
rarely been implemented since 2014. Overall, the share of cooking energy interventions has decreased
substantially over time while the share of commitments has slightly increased.
Within the off-grid portfolio, mini-grids and on-grid interventions (within off-grid interventions) were
implemented most frequently, while stand-alone systems, and particularly Pico-PV and cooking energy, were
implemented the least. Only 35 % of interventions sought to implement only one technology, while most
interventions (65 %) included at least two technical approaches.
Over the years, the mini-grid and on-grid approaches (within off-grid interventions) had the strongest growth,
surpassing stand-alone systems since the mid-2000s. The number of implementations of Pico-PV grew at a
low level and cooking energy stagnated at a low level.
Achieving access to modern energy by 2030 (SDG 7.1) is one of the goals of the German DC. However, the
share of interventions providing initial access to modern energy has been decreasing and the lowest tier level
implemented in interventions has been increasing over the years. This trend makes it very unlikely that the
German DC is on track towards achieving SDG 7.1.
In relation to the entirety of the energy and cooking energy sector, off-grid interventions were more than
three times more likely to contribute to access to energy in rural areas, while cooking energy interventions
were more likely to promote rural development than the average of the energy sector, but much less often
than off-grid interventions.
However, among off-grid interventions, cooking energy interventions were the most likely to promote rural
development, followed by Pico-PV systems. Moreover, about two thirds of mini-grids, stand-alone systems
and on-grid (within off-grid interventions) implementations promoted rural development.
Evidence from the literature reviews shows that the impact of modern energy access on productive use is
very limited. Technical approaches that may possibly lead to productive use were stand-alone systems
(particularly if enterprises are targeted).
However, the results of the portfolio analysis show that the German DC does not focus on either of these
approaches. The implementation levels of both approaches are low for stand-alone systems and very low for
cooking energy respectively. Furthermore, the trend over time of the share of both approaches is strongly
negative. Targets regarding the productive use of modern energy are unlikely to be reached given these
trends and the low implementation levels.
The share of interventions promoting climate change mitigation was already high in 2019-2022. It has been
steadily increasing over the years. Along with this, the financial commitments for climate change mitigation
have also been increasing among all types of interventions, as has the share of commitments for mitigation
in relation to overall commitments.
In addition, our findings confirm that there was no promotion of interventions in the German DC portfolio
that relied on fossil fuel energy after the Paris Agreement in 2015.
The share of interventions promoting gender equality has decreased substantially for the energy and cooking
energy sector and for off-grid interventions. It increased for the few cooking energy interventions. Overall,
very few interventions marked gender equality as their main objective.
In summary, the trend of the German DC’s portfolio regarding initial access to modern energy (SDG 7.1) and
gender equality (GG) does not yet point towards Germany achieving its goals by 2025 (GG) and 2030 (SDG
7.1). The promotion of climate change mitigation has increased. No fossil fuels were supported after the Paris
Agreement in 2015, and the energy and cooking energy sector exhibit high levels of financial commitment to
climate change mitigation.

3. | Data collection of focus groups 32
3. DATA COLLECTION OF FOCUS GROUPS
3.1 Benin
A total of 10 focus groups were held in Benin. They took place in June and July 2023 in the departments of
Alibori, Borgou, Couffo, Mono and Ouémé, and were meant to capture different contexts across agricultural
zones. Five focus groups were mixed-gender while five were exclusively female. The participants were a
mixture of beneficiaries of the EnDev and GBE interventions and non-beneficiaries.
Further, the groups were conducted with groups of farmers, animal breeders and shop owners. Within each
group, at least one participant was a beneficiary of the productive use components of EnDev or GBE which
provided access to solar appliances. The female focus groups took place in cooperation with the local
women’s association in the same localities.
Senegal
A total of 20 focus groups were held in Senegal. They took place in August and September 2023 across the
departments of Guinguinéo, Kébémer, Thiès and Tivaouane in the regions of Louga, Kaolack and Thiès. They
took place in August and September 2023 across the departments of Guinguinéo, Fatick, Kébémer, Thiès and
Tivaouane in the regions of Foundiougne, Louga, Kaolack and Thiès. Like in Benin, these were intended to
cover different contexts across agricultural zones via a broad selection. For the ten focus groups on solar
stand-alone appliances, the recruitment process for participants was identical to the one in Benin.
Apart from that, the team also conducted ten focus groups with entrepreneurs in villages where EnDev had
facilitated the installation of mini-grids through the components of ERSEN1 and ERSEN2. Additionally, these
components also provided access to productive-use appliances such as fridges to be used in conjunction with
the mini-grids. These focus groups were conducted in the regions of Fatick and Thiès in the departments of
Foundiougne and Tivaouane. Again, mirroring Benin, five groups were mixed-gender and five consisted
exclusively of women, which was enabled through the cooperation with local women’s organisations.
Uganda
A total of ten focus groups were held in Uganda in August 2023. Specifically, six focus groups were held in
the districts of Gulu, Kitgum and Lamwo in northern Uganda, addressing the target groups and technical
approaches of the GBE. Four focus groups were conducted in the districts of Buikwe and Wakiso in central
Uganda to encompass the target groups and technical approaches of EnDev. Overall, seven focus groups
were mixed-gender while three focus groups were exclusively female. All target groups included a mixture
of beneficiaries of the interventions and non-beneficiaries.
3.2
3.3

4. | Surveys on stand-alone solar PuE appliances 33
4. SURVEYS ON STAND-ALONE SOLAR PUE
APPLIANCES12
The data collection is described in detail in Chapter 4 of the evaluation report.
Identification strategy
To assess the effects in terms of outcomes and impacts of the use of the stand-alone appliances for
productive use of energy, we use three types of estimates: 1) before-after comparisons within the treatment
group, 2) cross-sectional analyses between treatment and control groups at the time of the survey and,
3) difference-in-differences (DiD) analyses.
Since there is no baseline data on either GIZ beneficiaries or the control groups (non-beneficiaries), the status
quo from before the intervention had to be reconstructed. This was done using recall questions about the
past for both the beneficiaries and the control groups. The reference year is the year before the intervention
began, namely 2015 in Benin and 2019 in Senegal. 13
Before-after comparisons
The before-after comparison is performed within the treatment group through a paired Student’s t-test.
Since the same units are compared to each other, there are no time-invariant factors at individual or village
level that could confound the relationship between the treatment and control groups. A potential source of
bias, however, is a secular trend that would have improved the ‘after’ even in the absence of an intervention.
This is addressed by complementing the simple before-after comparison with a DiD model that is more
immune to time trends.
The before-after comparison compares the potential outcome of the individuals and MSMEs in the treatment
group before they received the treatment to their outcome after the treatment. This comparison serves to
verify the plausibility of the results from the more complex causal analyses. It does this by testing whether
the treatment group improved over time.
The simple comparison also allows us to explore the short-term effects of the interventions, since the before-
after comparison compares individuals a year after the individual installation date of the solar appliances to
a year before the installation date.14 This is in most cases a much shorter timespan than between the time
of the survey (2023) and the reference year before any individual or MSME received the treatment (2015) in
Benin.
In Benin, the intervention was implemented from 2015 to 2022. Here, individual dates for the pre- and post-
treatment periods are used in the before-after comparison. The period before the treatment is one year prior
to the individual installation date of the solar appliance. The period after the treatment is one year after the
individual installation date of the solar appliance. According to the key assumption of this strategy, the
12 Authors: Mascha Rauschenbach, As’Ad Assani, Fredo Bankole, Gerald Leppert, Mame Mor Anta Syll
Contributors: Whitney Edwards, Yannick Gunia
13 Many MSMEs in the study are farms. The relevant period for them is the agricultural campaign, which does not entirely overlap with the calendar
year. When the surveys were tried and piloted, it turned out that MSMEs in Benin were better able to report outcomes with respect to the
agricultural campaign, while those in Senegal were able to report on the calendar year. This is why when we refer to the reference year and the
outcome year in Benin, the questionnaire asked about the agricultural campaigns for the periods 2015-2016 and 2022-2023 respectively.
14 The installation date is the date when the appliance was received by the beneficiary according to the monitoring data by the interventions studied.
This date is used in the analyses, since it is the first possible date where the appliance could potentially have been used by the beneficiary and the
treatment could have begun to unfold.
4.1
4.1.1

4. | Surveys on stand-alone solar PuE appliances 34
intervention should be the only element impacting changes in our results over time. If the intervention did
not exist, the results would have been the same for both before and after the study period.
As the intervention duration was much shorter in Senegal – from 2021 to 2023 – outcomes at individual
points in time were not collected and no before-after-comparison was performed.
4.1.2 Constructing and sampling a control group and an alternative treatment group
We construct a control group for the cross-sectional and DiD analyses. This control group is made up of
MSMEs (many of which are farms) that share important characteristics with the beneficiaries in the
respective municipality:
•they were engaged in the same activity as the beneficiaries in the reference year
(2015 in Benin, 2019 in Senegal);
•had a level of activity comparable to that of the beneficiary in the reference year;
•did not have access in any form to an appliance supplied to the beneficiary.
The control group is divided into two sub-types: a group that does not use modern energy (C1) and a group
that uses modern, non-solar energy (C2). This is necessary, because past research suggests different effects
for these two types of controls (see Chapter 3 in the evaluation report for a review).
While enterprises that use modern energy for the first time in an economic activity have a high potential of
seeing a substantial increase in productivity and sales, enterprises that switch from fossil to solar energy in
their economic activity are more likely to experience changes in their energy expenses and more moderate
changes in other outcomes.
The main comparison is between GIZ beneficiaries of solar stand-alone appliances and a control group that
is similar in several factors, with the main difference being the use of conventional, non-renewable energy
for their economic activity (C2).
We made this decision because according to our quantitative data and information from focus group
discussions, solar appliances were mainly acquired by entrepreneurs who had previously been using fossil
energy for the same activity. On the contrary, only few MSMEs that did not use modern energy for their
activity benefitted from the intervention.
The protocol for recruiting the control groups was as follows: the enumerator asked the beneficiary
interviewed to name a fixed number of other entrepreneurs in the same municipality who the beneficiary
thought would be interested in using solar appliances for their activity.
Another prerequisite was that the interviewee and the suggested respondent were similar to each other in
terms of activity in 2019 in Senegal (the year before the COVID-19 pandemic started and used as the
reference year) and in 2015 in Benin, respectively.15 The entrepreneur to be recruited for the control group
ought to have been involved in the same activity as the beneficiary in the reference year and on a similar
scale.16
15 In Senegal 2019 was used as the pre-treatment period and 2023 as the post-treatment period in all analyses. The year 2019 was used instead of
2020, as 2020 was exceptional due to the outbreak of the COVID-19 pandemic. Using 2020 as a reference year could have led to an overestimation
of positive effects as enterprises were recovering from the pandemic.
16 Original wording in French: « niveau d’activité comparable au bénéficiaire »

4. | Surveys on stand-alone solar PuE appliances 35
Moreover, the potential control respondent should not have profited directly or indirectly from the solar
appliance of the beneficiary. Those individuals suggested were then divided by the enumerator into the two
different types of control groups (C1 or C2) or the alternative treatment group.17
The alternative treatment group (T2) consisted of MSMEs which were similar to the beneficiary with regard
to the above-mentioned criteria and which also use a solar appliance for their economic activity. The
difference between the main treatment group – the GIZ beneficiaries (T1) – was that T2 had not aquired their
solar appliance through the GIZ interventions under study. There are two reasons why this alternative
treatment group was recruited.
First, the GIZ beneficiaries’ contact data was not always up to date, meaning that there was a risk of not being
able to sample enough respondents from T1 to be able to run statistical analyses (more so in Benin than in
Senegal). Secondly, recruiting T2 (called “GIZ and non-GIZ treated” in the regression tables) allowed analyzing
whether the effects were specific to the GIZ interventions or instead resulted from using solar appliances as
compared to fossil-fuel-powered appliances or no modern energy source at all.
4.1.3 Cross-sectional analyses
To verify the robustness of the results of the before-after comparison, cross-sectional analyses between the
treatment groups and the control groups were applied. The cross-sectional analyses compare the outcomes
for treatment and control MSMEs in the post-treatment period.
The assumption underlying this approach is that the enterprises who received the treatment do not differ
systematically from those who did not receive it. This was attempted through the recruitment protocol
described. The credibility of this assumption was enhanced by applying a propensity score matching (PSM)
technique.
This means that treatment and control units were matched based on observable variables (matching
variables). Matching variables were selected that are not influenced by the treatment itself but that could
plausibly confound the relationship between the treatment and outcomes under study (for details, see
Chapter 4.3).
Cross-sectional analyses can account for differences between the treatment and control groups prior to
treatment to a certain extent thanks to the inclusion of controls that were intervened prior to treatment.
The econometric specification to be used in the cross-sectional comparison is:
Yi = α + β* Treati + γ1Covi + γ2Covv + γ3Covz + εi
Yi is the outcome variable of interest.
Treati is the treatment variable, which takes the value 0 for the control group and the value 1 for MSMEs and
their owners who have bought and installed the appliance.
Covi, Covv and Covz represent the vectors of the baseline (2015 in Benin; 2019 in Senegal) characteristic of
the MSMEs used for calculating the propensity score and the related weight in the PSM approach (age of
enterprise, size of enterprise at baseline, high-quality flooring yes/no).
The index i includes baseline characteristics of the interviewed owner of the MSME (age, at least primary
education yes/no). The index v represents a baseline characteristic at the community level (rural/urban) and
the index z represents the location of the MSME in a certain agricultural zone.
17 The conclusions of the evaluation focus mainly on the comparison between GIZ beneficiaries and the control group which uses fossil energy for the
same activity as the one carried out by the beneficiary. This is the most important comparison, as most beneficiaries themselves used fossil energy
for their economic activity prior to treatment. This is also in line with an argument in the literature that the main competitor to solar pumps is diesel
pumps (Ankel-Peters et al., 2024a; Smith & Urpelainen, 2016). The evaluation contextualises these findings wherever suitable with reference to
those from the other comparisons.

4. | Surveys on stand-alone solar PuE appliances 36
β captures the treatment effect on the treated (ATT), i.e. the use of stand-alone PuE appliances, as a
coefficient of the treatment Treati.
4.1.4 Difference-in-differences (DiD) analyses
We complement the before-after comparisons and the cross-sectional analyses with DiD analyses. The
assumption underlying this approach is that there are no unobservable differences between treatment and
control that affect the changes over time.
To reduce selection bias and control for confounding factors, we apply the same PSM matching procedure as
described above for the cross-sectional comparison (for details on the PSM procedure and matching
variables, see Chapter 4.3).
The econometric specification in the DiD estimations is:
Yi = α + β1 * Ti + β2 * Treati + β3 * Ti *
Treati + γ1Covi + γ2Covv + γ3Covz + εi
Yi is the outcome variable of interest.
Ti denotes the time period which takes the value 0 for the baseline year (2015 in Benin; 2019 in Senegal) and
1 for the year 2023. The data on the baseline year T0 was obtained using recall questions in the survey at T1.
Treati is the treatment variable, which takes the value 0 for the control group and the value 1 for
individuals/MSMEs who have bought and installed the appliance.
Covi, Covv and Covz represent the vectors of the baseline (2015 in Benin; 2019 in Senegal) characteristic of
the MSMEs used for calculating the propensity score and the related weight in the PSM approach (age of
enterprise, size of enterprise at baseline, high-quality flooring yes/no).
The index i includes baseline characteristics of the owner of the MSME (age, at least primary education
yes/no). The index v represents a baseline characteristic at the community level (rural/urban) and the index
z represents the location of the MSME in a certain agricultural zone.
β captures the treatment effect on the treated (ATT), i.e. the use of stand-alone PuE appliances, as a
coefficient of the treatment Treati.18
Outcome variables
We study two main types of outcomes. Firstly, we examine MSME performance, operationalised through
production, sales, energy expenses, innovation/value added and the number of employees. According to the
theory of change, the outcomes that were investigated are the following: revenue19, sales20, processing of
products (dummy)21, energy expenses22, number of employees, having customers from outside their own
municipality (dummy for Benin, numerical as percentage of customers for Senegal)23, planted area24, quantity
18 The study did not intend to capture the intention to treatment effect (ITT), since it is not plausible that MSMEs were offered an appliance, but did
not obtain it.
19 Original wording in French: « Dans quel intervalle se situent les revenus totaux de l’entreprise? »
20 Original wording in French: « A combien estimez-vous les ventes totales de l’entreprise? »
21 Original wording in French: « Procédez-vous à la transformation de vos produits avant de les vendre? »
22 Original wording in French: « Combien dépensez-vous en termes d’énergie (FCFA/Mois) pour votre activité économique principale au cours des
différentes périodes suivantes? »
23 Original wording in French: « Quelle es/était la structure de votre clientèle en termes de localisation géographique? »
24 Original wording in French: « Quelle est la superficie emblavée pour [culture principale]? »
4.2

4. | Surveys on stand-alone solar PuE appliances 37
of product sold25, quantity of product sold during dry season26, cultivation during dry season (dummy,
variable only used in Senegal sample)27, food security (dummy)28 and assets29.
Secondly, we also study the economic wellbeing of the owner of an enterprise (and their family),
operationalised through questions on food security, assets and housing quality. A third category of outcomes
concerns the non-material wellbeing of MSME owners and some gender-specific outcomes. These outcomes
are assessed in separate descriptive analyses that focus on self-reported effects. These analyses are only
performed on the subsample of GIZ beneficiaries.
Causal pathways and operationalisation
Chapter 3 of the evaluation report identifies the causal pathways from access to solar appliances to the
outcomes and impacts including wellbeing of entrepreneurs and their families and economic impacts.
However, there are some factors that may lead to a bias in the estimations of the treatment effect on
different outcomes and impacts if unaccounted for. These factors are potential confounders, which
simultaneously predict the selection into treatment as well as the outcome and impact under study
(Rosenbaum and Rubin, 1983). In the absence of a random allocation to the treatment group, we need to
correct for this selection bias in the analysis.
One potential confounder between access to a solar appliance and the likelihood of switching to more
valuable crops or processing products before selling them is the age of the owner. Akudugu et al. (2012) find
a U-shaped relationship between age and adoption of a new technology, with middle-aged entrepreneurs
being more likely to adopt a new technology (Akudugu et al., 2012). Interestingly, the results from Benin
contradict this finding, and show that older enterprise owners are more likely to acquire a solar appliance
but also those whose enterprise is younger. Adding to that, larger firms also appear more likely to purchase
a solar appliance (Closas and Rap, 2017) and owners of larger enterprises seem more likely to adopt a new
technology (Akudugu et al., 2012).
Another potential group of confounders that we account for are regional factors. First, the agricultural zone
in which a respondent is located may be a proxy for regional differences in weather, temperatures and
climate in general. This is important, since weather drives both the need for irrigation or the provision of
water for animals, and also directly affects the production level and the health of animals (Burney et al.,
2010). Secondly, depending on infrastructure and the institutional setting, the degree of urbanisation of an
area is generally associated with higher income and productivity, and also with effects on labour markets and
knowledge spillovers (Dudwick et al., 2011). These factors may also be correlated with the accessibility of
solar appliances and at the same time post-treatment outcomes such as revenues.
The pre-treatment size of an enterprise is also expected to affect its capacity and, consequently, its likelihood
of acquiring a costly solar appliance. At the same time, enterprise size may also correlate with indicators of
post-treatment enterprise performance.
25 Original wording in French: « Quelle quantité de votre culture principale vendez-vous pendant la saison des pluies au cours les différentes périodes
suivantes? »
26 Original wording in French: « Quelle quantité de votre culture principale vendez-vous pendant la saison sèche au cours des différentes périodes
suivantes? »
27 Original wording in French: « Cultivez -vous pendant la saison sèche? »
28 Original wording in French: « Au cours des périodes suivantes, avez-vous jamais été affamé ou n’avoir pas mangé du fait de l’insuffisance de la
nourriture ou d’argent? »
29 Original wording in French: « Quels sont les biens matériels dont dispose le propriétaire/ Président ou les membres de son ménage pendant les
périodes? »
4.3

4. | Surveys on stand-alone solar PuE appliances 38
Moderating factors are another type of variable which are relevant for the analysis and identification of the
causal pathways. As for moderating factors, they influence the strength of the relationship between
treatment and outcome.
For example, a MSME that has access to markets outside the community may benefit more from the
treatment compared to other MSMEs. Or in other words, their treatment effect may be larger if market
access exists. Generally, there may be an overlap between moderating factors and confounders if the
moderating factor also influences the selection into the treatment, but a moderating factor can also only
influence the relationship between treatment and outcome, or it may influence the relationship of treatment
and outcome as well as the outcome itself.
In our study, we assume that the effect of having access to a solar appliance on sales and income is amplified
by an enterprise having better access to customers and markets (Ankel-Peters et al., 2024b). This is because
local demand typically does not increase economic welfare in the community, since the money spent by a
local customer for one product will be a substitute for buying a different local product. Overall, the customer
will not spend any more in the municipality. It is assumed that an increase in production will only lead to an
increase in sales (and productivity) if the enterprise can expand beyond local markets. Additionally, we also
perform subgroup analyses for female entrepreneurs to account for the possibility that effects may differ
between men and women.
Selection bias and confounding factors can be addressed through matching on them. A requirement for their
selection is that they must remain unaffected by the treatment to avoid the introduction of endogeneity
(Angrist and Pischke, 2009, p. 52–59). In order to control for selection bias and potential confounding factors,
PSM and specifically kernel matching was applied. See Garrido et al. (2014) for more general information on
the kernel method.
As described above, the control group was selected to have similar characteristics to the treatment group.
The control group was relatively large. Kernel matching, which uses relatively more information, is expected
in such cases to lead to a reasonable estimation, performing better than, for example, the frequently used
nearest-neighbour matching (Caliendo and Kopeinig, 2008; Frölich, 2004).
Specifically, kernel matching was applied with replacement using the Gaussian kernel and a bandwidth of
0.05. For the diagnostics of the models, several characteristics were assessed, such as the balancing and
reduction of differences after matching, the reduced and remaining bias, the variance ratio, Rubin’s B and
Rubin’s R.
With slight variations between Benin and Senegal, matching variables were applied that fall into three
categories: characteristics of the area (agricultural zone, rural environment), characteristics of the MSME
(age of enterprise, size of enterprise at baseline, high-quality flooring yes/no), and characteristics of the
owner (age, at least primary education yes/no).
Using PSM, treatment and control groups were matched on several characteristics that fulfil these conditions.
The matching and subsequent analyses were performed separately for each outcome pair (e.g. revenue 2015
and revenue 2023 for Benin) and for each subgroup on variables from the baseline year. This approach was
selected to preserve case numbers, considering the variation in data availability between groups and
outcomes. Country-specific differences in the matching procedure are discussed in the respective sections.
Moreover, apart from analyses performed on the full sample, subgroup analyses differentiate further
between GIZ beneficiaries only (T1), farmers and women. To minimise data loss, outliers were trimmed using
a threshold set at three times the interquartile range (3 IQR), a more lenient approach than the suggested
1.5 IQR by Heumann et al. (2016). This means that outliers are not removed and are instead replaced with
the last value in the dataset, still falling into the 3 IQR.

4. | Surveys on stand-alone solar PuE appliances 39
4.4 Results for Benin
4.4.1 Descriptive results
As a first step, outcome variables in 2015/2016 (pre-treatment) and 2022/2023 (post-treatment) were
compared using a mean comparison test (Student’s t-test) on the unmatched samples. As a general rule, the
number of observations in 2015/2016 is somewhat smaller than for 2022/2023 due to missing values, given
that recall questions were used. According to the t-tests, GIZ beneficiaries (T1) (see Table 5) only significantly
statistically differed from the control group (C2) at baseline (2015) concerning two outcome variables: the
number of employees and the servicing of customers from outside their own municipality. On average,
enterprises in the treatment group employ more than twice as many employees compared to the control
group. Furthermore, enterprises in the GIZ treatment group are significantly more likely to have customers
from outside their own municipality. Regarding their performance, GIZ beneficiaries tend to have higher
revenues, be more likely to process products and have higher monthly energy expenses but lower sales,
though the differences between the means in these outcomes is not statistically significant. GIZ beneficiaries
also tend to have fewer personal assets on average and are more food secure, but these findings are also not
statistically significant.
Table 6 shows the results for the outcome variables in the post-treatment period in 2022/2023 respectively.
The two significant outcomes from the pre-treatment period persist in the post-treatment period in
2022/2023: beneficiaries still employ significantly more staff than the control group, and they are also still
more likely to service customers from outside their own municipality. Additionally, they have higher revenues
than the control group, are more food secure and farmers cultivate a larger area. It should be noted,
however, that the outcome of the area planted was measured only once (in 2022/2023), and it remains
unclear whether there was already a significant difference pre-treatment in 2015/2016. The differences in
all other outcomes remain insignificant.
The mean comparison of the variables to be matched on is presented in the section on matching for Benin
(see Table 8 and Table 9).
Table 5 Mean comparison tests of outcomes in 2015/2016 (pre-treatment),
GIZ beneficiaries vs. control group, Benin
Obs Mean Std.dev t-test
Control Treatment Control Treatment Control Treatment Control/Treatment
Processing 186 111 0.06 0.09 0.25 0.29 -0.03
Customers
from out of
municipality
198 114 0.43 0.59 0.50 0.49 -0.16***
Sales 75 20 82,479.96 5,400.45 336,133.52 22,337.96 77,079.51
Assets 198 114 0.19 0.18 0.09 0.07 0.01
Energy expenses 198 114 16,122.47 43,452.50 56,114.00 263,042.67 -27,330.03
Revenue 158 101 315,981.01 3,926,250 1.17e+06 2.99e+07 -3.61e+06
Number of
employees
99 64 2.10 5.30 3.26 10.17 -3.20***
Food security 197 112 0.79 0.86 0.41 0.35 -0.07
Source: DEval, own table
Note: Mean comparison between the GIZ treatment group and control group (T1 vs. C2) in Benin in 2015, equal variances,
not trimmed. * p < 0.10, ** p < 0.05, *** p < 0.01
outliers

4. | Surveys on stand-alone solar PuE appliances 40
Table 6 Mean comparison tests of outcomes in 2022/2023 (post-treatment),
GIZ beneficiaries vs. control group, Benin
Obs Mean Std.dev t-test
Control Treatment Control Treatment Control Treatment Control/Treatment
Processing 198 114 0.12 0.11 0.33 0.32 0.01
Customers
from outside
the
municipality
198 114 0.47 0.66 0.50 0.48 -0.19***
Assets 198 114 0.44 0.44 0.13 0.17 0.00
Sales 75 20 313,080 65,800.05 1.13e+06 75,674.88 247,279.95
Energy
expenses
198 114 60,674.24 55,822.81 357,414.64 255,346.13 4,851.44
Revenue 158 101 689,493.67 6.42e+06 1.02e+06 3.20e+07 -5.73e+06**
Number of
employees
192 113 2.05 5.05 3.05 9.17 -3.01***
Food security 197 114 0.81 0.89 0.39 0.31 -0.08*
Cultivation
during
dry season
77 72 0.97 0.99 0.16 0.12 -0.01
Area planted 77 72 1.95 4.89 2.67 11.95 -2.95**
Source: DEval, own table. Mean comparison between the GIZ treatment group and control group (T1 vs. C2) in Benin in 2023, equal
variances, outliers not trimmed
Note: Cultivation during the dry season and area planted were only measured in 2023, not in 2015.
* p < 0.10, ** p < 0.05, *** p < 0.01
4.4.2 Before-after comparison
Table 7 shows the results of the before-after comparison for the treatment group (GIZ beneficiaries only).
The date of treatment (i.e. installation of the appliance) varies between the MSMEs. Therefore, GIZ
beneficiaries were asked about their circumstances 12 months prior to installing their appliances and 12
months afterwards. The results show that the MSMEs had more staff and higher revenues a year after the
installation of the solar appliance promoted by the GIZ than a year before they bought them, whereas other
outcomes did not differ significantly from the baseline period. However, the simple before-after comparison
does not allow an interpretation of these as positive short-term consequences of the interventions with
certainty. The before-after comparison needs to be triangulated with the more sophisticated cross-sectional
and difference-in-difference analyses that follow below.

4. | Surveys on stand-alone solar PuE appliances 41
Table 7 Paired Student’s t-test after vs. before within the GIZ treatment group, Benin
Obs Mean Std.dev t-test
Before After Before After Before After Before/After
Customers
from outside
municipality
114 114 0.62 0.66 0.48 0.47 -1.15
Sales 71 71 107.38 149.52 429.1452 447.3894 -0.96
Energy
expenses
114 114 32,142.54 37,953.95 187,805.1 192,958.4 -0.66
Revenue 99 99 4,934,205 6,074,551 3.13e+07 3.18e+07 -1.81**
Number of
employees
92 92 4.67 5.36 8.54 9.28 -3.01**
Food security 113 113 0.88 0.89 0.32 0.30 -0.08
Source: DEval, own table, Paired t-test with equal variance within the GIZ treatment group (T1), after vs. before installation of the
appliance in Benin
Note: Comparison between one year after vs. one year before the individual installation date for each beneficiary. Group, outliers not
trimmed.* p < 0.10, ** p < 0.05, *** p < 0.01
4.4.3 Matching
As mentioned above, the weights that result from the PSM were included in the cross-sectional as well as
the DiD analyses. Given the variability in missing values across outcomes and diverse populations within
subgroups, separate matchings for each outcome, subgroup and treatment type were conducted. The aim
was to conserve as many observations as possible.
Table 8 and Table 9 further illustrate that the amount of available data points differed widely between
outcomes and times. In a trade-off between the goodness-of-fit of the matching and comparability across
countries, analyses and subgroups, the decision was made to include the same matching variables in every
matching, with adjustments only being made in terms of operationalisation (e.g. inclusion of quartiles vs.
tertiles) and reference categories within variables. The only exception from this is the inclusion of
geographical factors in Benin but not in Senegal, specifically agricultural zone and rural classification. It was
determined that the distribution of those variables in Senegal did not allow for a balanced sample, resulting
to inferior matching results where they were included. Consequently, it was concluded that while these
geographical factors may act as confounding factors in Benin, their influence appears to be less pronounced
in Senegal. This circumstance might be related to country-specific differences in geography and established
infrastructure.
Table 8 and Table 9 display the matching variables that were used for Benin, including the GIZ sample (T1)
(see Table 8), as well as the GIZ and non-GIZ treated (T1 and T2) (see Table 9), respectively, and descriptive
information on them. The variable agricultural zone was coded based on Abdul-Jalil et al. (2023), with two
comparable zones being combined due to low case numbers and with Atacora Ouest being the reference
category. Agricultural zone was included in the tables below for the sake of completeness, despite not
displaying any values due to it being a nominal variable. High-quality floor is a dummy variable, which is
coded 1 if the flooring was reported to be PVC flooring/asphalt, tiles or cement and 0 if flooring was dirt or
cow dung/droppings. This variable was included on the recommendation of local experts. The variable age
of the enterprise is coded as tertiles, with the middle category being the reference category, and the age of
the owner is a numerical variable. The size of the enterprise is a composite variable that captures differences

4. | Surveys on stand-alone solar PuE appliances 42
between enterprises. It has a three-step coding (small, medium and large), and is composed of the size of
cultivated land, tertiles of revenue in 2015 and differences in livestock.
Lastly, the primary education variable is a dummy variable that indicates whether the owner of the enterprise
has completed primary education or higher; and, in accordance with the World Bank definition, rural is a
binary variable that is coded 1 if the density of inhabitants in the respondent’s municipality is less than 300
per km2 and is coded 0 otherwise (World Bank, 2020).
Table 8 Mean comparison tests for the matching variables
Benin
2015, GIZ beneficiaries vs. control group,
Obs Mean Std.dev t-test
Control Treatment Control Treatment Control Treatment Control/Treatment
Agricultural zone
(4 categories)
198 158 - - - - -
Alibori Sud -
Borgou Nord -
2KP - Vallé
198 158 0.20 0.18 0.40 0.38 0.02
Atacora Ouest 198 158 0.04 0.05 0.20 0.22 -0.01
Borgou Sud -
Donga - Collines
198 158 0.22 0.34 0.34 0.42 -0.09**
Ouémé -
Atlantique - Mono
198 158 0.63 0.55 0.49 0.50 0.08
High-Quality Floor 198 158 0.86 0.87 0.34 0.34 0.00
Age of enterprise
(in years)
198 158 10.48 10.49 9.59 10.00 -0.01
Age of owner
(in years)
198 158 42.21 45.65 10.96 12.23 -3.44***
Size of enterprise* 198 158 1.83 2.20 0.79 0.76 -0.37***
At least
primary education
198 158 0.78 0.81 0.42 0.39 -0.03
Rural 198 158 0.42 0.52 0.50 0.50 -0.09*
Source: DEval, own table, Equal variances. GIZ treatment group. * p < 0.10, ** p < 0.05, *** p < 0.01.
Note: *Size of enterprise is an ordinal variable with three levels of size: 1 = small, 2 = medium, 3 = large

4. | Surveys on stand-alone solar PuE appliances 43
Table 9 Mean comparison tests for the matching variables
GIZ and non-GIZ treated vs. control group, Benin
2015,
Obs Mean Std.dev t-test
Control Treatment Control Treatment Control Treatment Control/Treatment
Agricultural zone
(4 categories)
198 158 - - - - -
Alibori Sud -
Borgou Nord
-Vallé
– 2KP
198 158 0.20 0.18 0.40 0.38 0.02
Atacora Ouest 198 158 0.04 0.05 0.20 0.22 -0.01
Borgou Sud -
Donga - Collines
198 158 0.22 0.34 0.34 0.42 -0.09**
Ouémé -
Atlantique - Mono
198 158 0.63 0.55 0.49 0.50 0.08
High-quality floor 198 158 0.86 0.87 0.34 0.34 0.00
Age of enterprise
(in years)
198 158 10.48 10.49 9.59 10.00 -0.01
Age of owner
(in years)
198 158 42.21 45.65 10.96 12.23 -3.44***
Size of enterprise* 198 158 1.83 2.20 0.79 0.76 -0.37***
At least
primary education
198 158 0.78 0.81 0.42 0.39 -0.03
Rural 198 158 0.42 0.52 0.50 0.50 -0.09*
Source: DEval, own table, Equal variances, GIZ
Note: *Size of enterprise is an ordinal variable
treatment group, * p < 0.10, ** p < 0.05, *** p < 0.01
with three levels of size: 1 = small, 2 = medium, 3 = large.
We applied several matching diagnostics to ascertain the quality of the matching results. First, Rubin’s B is
reported in Table 10 for the matching results for the GIZ sample (T1 vs. control) and in Table 11 for the full
sample (GIZ and non-GIZ treated and control group, T1, T2, C2). This statistic measures the balance between
the treated and control samples by computing the absolute standardised difference of the means of the
linear index of the propensity score in the treated and (matched) non-treated group (Rubin, 2001).
Additionally, we also report Rubin’s R, which measures the ratio of treated to (matched) non-treated
variances of the propensity scores (Rubin, 2001). Both measures are considered important diagnostic tools
for assessing the reduction in bias and the quality of a propensity matching. Rubin (2001) recommends that
B should be less than 25 and R remains in a range between 0.5 and 2 as an indication of a sufficient balancing
of the samples. As both tables show, the B-values as well as the R-values stay well within the recommended
range, which marks the performed matching as successful in the sense of achieving a greater comparability
between treatment and control groups.
Figure 15 and Figure 16 illustrate the how the control and treatment groups became more similar in statistical
terms after the matching for the outcome variable of energy expenses.

4. | Surveys on stand-alone solar PuE appliances 44
Table 10 Results of the matching per outcome in 2015 and 2023,
GIZ beneficiaries vs. control group, Benin
Outcome Obs Common Support Rubin’s B Rubin’s R
2015 2023
Treated Untreated Treated Untreated Treated Untreated Matched Matched
Revenue 78 136 99 157 75 136 13.6 1.14
Processing 30 109 154 112 155 103 154 20.6 1.15
Energy expenses 112 197 112 197 101 197 11.6 1.33
Number of
employees
65 98 111 191 61 98 17.5 1.43
Customers
from outside
the municipality
112 197 112 197 101 197 11.6 1.33
Assets 112 197 112 197 101 197 11.6 1.33
Food security 110 196 112 196 99 195 12.4 1.31
Source: DEval, own table, Outliers trimmed
It is notable that even though the number of observations for the combined sample of GIZ and non-GIZ
treated is higher (Table 11), which is generally associated with an improvement in the balancing of the
samples, Rubin’s B indicates a better balance for the outcomes of energy expenses, customers from outside
the municipality, the assets and food security in the sample where only GIZ beneficiaries are matched with
the control group (Table 10). This might be an indication that the GIZ sample is more homogenous or has
different baseline characteristics from the sample, including all solar appliance users for these outcomes.
Furthermore, regarding the quality of matching of the single matching variables, we tried to ensure that the
percentage of bias remained low at well under 10 and that the variance ratio was between 0.5 and 2.
Table 11 Results of the matching per outcome in 2015 and 2023, GIZ and non-GIZ treated
vs. control group, Benin
Outcome Obs Common Support Rubin’s B Rubin’s R
2015 2023
Treated Untreated Treated Untreated Treated Untreated Matched Matched
Revenue 104 136 135 157 98 136 9.2 1.28
Processing31 144 154 155 155 138 154 24.9 1.19
Energy expenses 155 197 155 197 150 197 10.7 1.16
Number of
employees
82 98 154 191 78 98 14.1 1.22
30 Numbers reported are for the comparison GIZ treatment sample vs. the control group that uses no modern energy, since it will be used in the
analyses.
31 Numbers reported are for the comparison of GIZ and non-GIZ treated vs. the control group that uses no modern energy, since it will be used in the
analyses.

4. | Surveys on stand-alone solar PuE appliances 45
Customers from
outside the
municipality
156 197 156 197 151 197 10.8 1.16
Asset index 156 197 156 197 151 197 10.8 1.16
Food security 153 196 155 196 147 195 10.4 1.12
Source: DEval, own table. Outliers trimmed
Regarding the matching for the subgroups, there is a stark difference in the overall quality of the matching
between farmers and women. The matching performed the same or even better for farmers on some
outcomes compared to the sample including GIZ and non-GIZ treated. Despite lower case numbers, farmers
appear to be a somewhat homogenous group.
For women on the other hand, the quality of the matching was worse than for the sample of all entrepreneurs
and, in some cases, the strict thresholds for Rubin’s B and R could not be upheld. The case numbers for
women were lower than for farmers, and it can be concluded that the women in the sample are a
heterogenous group. However, the matching for the subgroup women still leads to an overall improvement
in the balance of the samples when compared to the unmatched sample. Figure 15 and 16 show how the
matching helped make control and treatment group statistically more similar to each other, as illustrated for
the outcome of energy expenses.
Figure 15 Density plot for pre-treatment outcome energy expenses for before and after matching
state, GIZ and non-GIZ treated versus control group, Benin
Source: DEval, own figure
0 1 2 3
Kernel density
0 .2 .4 .6 .8 1
Propensity scores BEFORE matching
0 1 2 3
Kernel density
0 .2 .4 .6 .8
Propensity scores AFTER matching
Treated Control

4. | Surveys on stand-alone solar PuE appliances 46
Figure 16 Density plot for outcome energy expenses for before and after matching state,
GIZ treatment sample, Benin
Source: DEval, own figure
0 1 2 3
Kernel density
0 .8
Propensity scores BEFORE matching
0 1 2 3
Kernel density
0 .8
Propensity scores AFTER matching
Treated Control
.2 .4 .6 .2 .4 .6
4.4.4 Cross-sectional analysis
Table 12 to Table 14 show the results of the cross-sectional analyses performed on the matched samples. In
terms of economic performance, GIZ beneficiaries and non-GIZ treated alike report higher revenues, lower
energy expenses and the employment of more staff (see Table 12 and Table 13) than the respondents who
use fossil fuels for the same economic activity (control group).Additionally, GIZ beneficiaries (T1) seem more
likely to be food secure than the control group (see Table 14), and all treated MSMEs appear more likely to
service customers from outside their municipality (see Table 13) than the control group.
Furthermore, the results show that apart from the effects of the treatment mentioned, there are no
statistically detectable differences between the treated and the control group regarding their likelihood of
processing their products before selling them or their sales, personal assets of the owners of the enterprises
(see Table 12, Table 13 and Table 14) or the size of the planted area that farmers work on (see Table 17 and
Table 18).

4. | Surveys on stand-alone solar PuE appliances 47
Table 12 Cross-sectional treatment effect on revenue, processing, energy expenses,
number of employees, customer origin, GIZ beneficiaries vs. control group, Benin
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of
employees
(5)
Having customers
from outside
the municipality
GIZ treatment 301,751.3** -17,734.2*** 0.996* 0.190
(133,824.9) (3721.8) (0.561) (0.164)
GIZ treatment
(vs. no modern energy)
0.152
(0.252)
Constant 742,792.8*** -1.573*** 34,614.6*** 2.562*** 0.135
(65,770.2) (0.174) (2,420.3) (0.298) (0.104)
Observations 251 257 299 295 299
R2 0.026 0.089 0.014
Pseudo R2 0.003 0.004
Source: DEval, own table. T = GIZ treatment sample vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed. All outcome
variables measured in 2023. Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01
Table 13 Cross-sectional treatment effect on revenue, processing, energy expenses, number of
employees, customer origin, GIZ and non-GIZ treated vs. control group, Benin
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of
employees
(5)
Having
customers
from outside
the municipality
GIZ and
non-GIZ treatment
266,513.8** -16,532.5*** 1.078** 0.275*
(114,105.5) (3,555.6) (0.491) (0.147)
GIZ and
non-GIZ treatment
(vs. no modern energy)
0.0788
(0.269)
Constant 721,797.9*** -1.394*** 35,239.8*** 2.604*** 0.0433
(63005.8) (0.224) (2545.6) (0.305) (0.104)
Observations 289 292 348 342 349
R2 0.022 0.072 0.017
Pseudo R2 0.001 0.009
Source: DEval, own table, T = GIZ and non-GIZ treated vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed
Note: All outcome variables measured in 2023. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

4. | Surveys on stand-alone solar PuE appliances 48
Table 14 Cross-sectional treatment effect on sales, assets and food security,
GIZ beneficiaries vs. control group, Benin
(1)
Sales
(2)
Assets
(3)
Food security
GIZ treatment -34,344.2 -0.0108 0.346*
(38,182.6) (0.0205) (0.204)
Constant 117,050.1*** 0.427*** 0.853***
(33,291.5) (0.0117) (0.125)
Observations 67 299 300
R2 0.024 0.001
Pseudo R2 0.015
Source: DEval, own table. T = GIZ treatment sample vs. C = non-renewable energy for all outcomes. OLS regression if
regression if pseudo R2 is reported. Outliers trimmed
Note: All outcome variables measured in 2023. Standard errors in parentheses.* p < 0.10, ** p < 0.05, *** p < 0.01
R2 and probit
Cross-sectional subgroup analysis: farmers
We ran the same models on sub-samples of farmers (see Table 15 to Table 18) and female entrepreneurs
only (see Table 19 to Table 22). The findings from the sample of GIZ and non-GIZ treated are largely robust
in restricting the analysis to farmers. Accordingly, among farmers, the treated have higher revenues, more
employees, lower energy expenses and are more likely to be food secure (GIZ beneficiaries only) than the
control group.

4. | Surveys on stand-alone solar PuE appliances 49
Table 15 Cross-sectional treatment effect on revenue, processing, energy expenses,
number of employees, GIZ and non-GIZ treated vs. control group, Benin, farmers only
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of employees
GIZ and
non-GIZ treatment
313,110.1** -17,429.1*** 1.035*
(136,240.8) (4,335.3) (0.606)
GIZ and
non-GIZ treatment
(vs. no modern energy)
0.251
(0.323)
Constant 791,065.7*** -1.532*** 35,494.7*** 2.840***
(75,020.8) (0.273) (3258.4) (0.364)
Observations 170 235 206 205
R2 0.032 0.086 0.016
Pseudo R2 0.008
Source: DEval, own table. T = GIZ and non-GIZ treated vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed
Note: All outcome variables measured in 2022/2023. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01
Table 16 Cross-sectional treatment effect on revenue, processing, energy expenses,
number of employees, GIZ beneficiaries vs. control group, Benin, farmers only
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of employees
GIZ treatment 341,922.3** -20,130.4*** 1.057
(155,806.6) (5,019.4) (0.705)
GIZ treatment
(vs. no modern energy)
0.454
(0.310)
Constant 803,249.0*** -1.689*** 37,187.0*** 2.891***
(79,304.3) (0.250) (3,860.1) (0.409)
Observations 149 218 186 185
R2 0.036 0.106 0.015
Pseudo R2 0.026
Source: DEval, own table. T = GIZ treatment sample vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed.
Note: All outcome variables measured in 2022/2023. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

4. | Surveys on stand-alone solar PuE appliances 50
Table 17 Cross-sectional treatment effect on planted area, assets and food security,
GIZ and non-GIZ treated vs. control group, Benin, farmers only
(1)
Planted area
(2)
Assets
(3)
Food security
GIZ and non-GIZ treatment 0.0891 -0.00139 0.295
(0.498) (0.0214) (0.221)
Constant 2.368*** 0.409*** 0.806***
(0.362) (0.0143) (0.152)
Observations 145 207 205
R2 0.000 0.000
Pseudo R2 0.011
Source: DEval, own table. T = GIZ and non-GIZ treated vs. C = non-renewable energy for all outcomes. OLS regressions. Outliers
trimmed. All outcome variables measured in 2022/2023. Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01
Table 18 Cross-sectional treatment effect on planted area, assets and food security,
GIZ beneficiaries vs. control group, Benin, farmers only
(1)
Planted area
(2)
Assets
(3)
Food security
GIZ treatment 0.0476 -0.0199 0.473*
(0.492) (0.0232) (0.251)
Constant 2.250*** 0.410*** 0.764***
(0.342) (0.0150) (0.159)
Observations 133 186 183
R2 0.000 0.005
Pseudo R2 0.027
Source: DEval, own table. T = GIZ treatment sample vs. C = non-renewable energy for all outcomes. OLS regression if R2 and probit
regression if pseudo R2 is reported. Outliers trimmed.
Note: All outcome variables measured in 2022/2023. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01
Cross-sectional subgroup analysis: women
Among women entrepreneurs, the effect of the treatment on reduced energy expenses remains for GIZ
beneficiaries only (see Table 19). However, the effects on higher revenues and the number of employees
disappears (the latter only for GIZ beneficiaries versus control group). In addition, the results suggest a
negative treatment effect among female entrepreneurs on their food security.

4. | Surveys on stand-alone solar PuE appliances 51
Table 19 Cross-sectional treatment effect on revenue, processing, energy expenses and
number of employees, GIZ beneficiaries vs. control group, Benin, women only
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of employees
GIZ treatment 108,122.7 -16,325.0** 1.070
(298,141.0) (7,118.2) (0.995)
GIZ treatment
(vs. no modern energy)
-0.434
(0.507)
Constant 500,968.2*** -0.926*** 28,658.3*** 1.055***
(62,265.0) (0.341) (2,829.2) (0.244)
Observations 61 43 75 74
R2 0.005 0.125 0.031
Pseudo R2 0.023
Source: DEval, own table. T = GIZ treatment sample vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed.
Note: All outcome variables measured in 2022/2023. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01
Table 20 Cross-sectional treatment effect on revenue, processing, energy expenses and
number of employees, GIZ and non-GIZ treated vs. control group, Benin, women only
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of employees
GIZ and
non-GIZ treatment
3,881.1 -8,236.1 1.593*
(190,906.6) (6,016.3) (0.829)
GIZ and
non-GIZ treatment
(vs. no modern energy)
-0.330
(0.463)
Constant 514,539.9*** -0.839** 24,200.4*** 1.074***
(67,967.9) (0.365) (2,460.0) (0.267)
Observations 69 53 88 85
R2 0.000 0.030 0.060
Pseudo R2 0.013
Source: DEval, own table. T = GIZ and non-GIZ treated vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed.
Note: All outcome variables measured in 2022/2023. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

4. | Surveys on stand-alone solar PuE appliances 52
Table 21 Cross-sectional treatment effect on sales, assets and food security,
GIZ beneficiaries vs. control group, Benin, women only
(1)
Sales
(2)
Assets
(3)
Food security
GIZ treatment -60,217.0 -0.0616 -0.854**
(70,055.2) (0.0461) (0.423)
Constant 133,359.9** 0.465*** 1.285***
(63,864.5) (0.0185) (0.256)
Observations 32 75 75
R2 0.071 0.043
Pseudo R2 0.081
Source: DEval, own table. T = GIZ treatment sample vs. C = non-renewable energy for all outcomes. OLS regression if
regression if pseudo R2 is reported. Outliers trimmed
Note: All outcome variables measured in 2022/2023. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p <
R2 and probit
0.01
Table 22 Cross-sectional treatment effect on sales, assets and food security,
GIZ and non-GIZ treated vs. control group, Benin, women only
(1)
Sales
(2)
Assets
(3)
Food security
GIZ and non-GIZ treatment -38,272.9 -0.00507 -0.887***
(34,414.3) (0.0394) (0.338)
Constant 97,091.1*** 0.449*** 1.351***
(28,334.1) (0.0244) (0.230)
Observations 51 88 88
R2 0.045 0.000
Pseudo R2 0.086
Source: DEval, own table. T = GIZ and non-GIZ treated vs. C = non-renewable energy for all outcomes. OLS regressions.
Note: Outliers trimmed. All outcome variables measured in 2022/2023. Standard errors in parentheses.
* p < 0.10, ** p < 0.05, *** p < 0.01
4.4.5 Difference-in-differences (DiD) analysis
Table 23 and Table 24 show the results of the DiD analysis for the GIZ beneficiaries and GIZ and non-GIZ
treated vs. the control group across different outcomes. They further confirm the treatment effects from the
cross-sectional analyses regarding the revenues for the GIZ sample and the energy expenses for all treatment
groups. However, the effect on the number of employees disappears.

4. | Surveys on stand-alone solar PuE appliances 53
Table 23 DiD treatment effect on revenue, processing, energy expenses, number of employees, Benin
(1)
Revenue
(2)
Revenue
(3)
Processing
(4)
Processing
(5)
Energy expenses
(6)
Energy expenses
(7)
Number of employees
(8)
Number of employees
Post 371,718.4*** 365,638.3*** -3.07e-16 -1.80e-16 23,717.6*** 22,555.8*** 0.373 0.334
(78,337.5) (81,089.6) (0.317) (0.246) (2,856.5) (2,729.6) (0.621) (0.594)
GIZ and
non-GIZ treatment
3,304.8 -3,968.6** 1.294*
(52,899.2) (1,616.9) (0.708)
GIZ and non-GIZ DiD 175,389.8 -12,563.8*** 0.230
(130,566.1) (3,906.0) (1.015)
GIZ treatment -7,977.0 -4,244.2** 1.640**
(58,916.7) (1,689.7) (0.792)
GIZ DiD 273,962.3* -13,490.0*** 0.634
(154,495.0) (4,087.4) (1.138)
GIZ and
non-GIZ treatment
(vs. no modern energy)
0.0788
(0.269)
GIZ and non-GIZ DiD
(vs. no modern energy)
2.21e-16
(0.380)
GIZ treatment
(vs. no modern energy)
0.152
(0.251)

4. | Surveys on stand-alone solar PuE appliances 54
(1)
Revenue
(2)
Revenue
(3)
Processing
(4)
Processing
(5)
Energy expenses
(6)
Energy expenses
(7)
Number of employees
(8)
Number of employees
GIZ DiD
(vs. no modern energy)
8.44e-16
(0.355)
Constant 374,591.8*** 414,486.7*** -1.394*** -1.573*** 11,522.2*** 12,058.8*** 2.680*** 2.426***
(38,083.5) (39,966.6) (0.224) (0.174) (1,296.0) (1,262.1) (0.423) (0.408)
Observations32 480 426 584 514 696 598 352 318
R2 0.121 0.135 0.176 0.176 0.029 0.053
Pseudo R2 0.001 0.003
Sample Full GIZ Full GIZ Full GIZ Full GIZ
Source: DEval, own table. T = GIZ and non-GIZ treated; GIZ beneficiaries sample, C = users of non-renewable energy, except for processing, for this outcome C = no modern energy. Robust standard
errors in parentheses, * p<0.10, ** p<0.05, *** p<0.01. OLS regressions if R2 is reported, probit regressions otherwise. All outcomes measured in 2015 and 2023. Outliers trimmed
32 All DiD tables presented report the total number of statistical observations, wherein two data points are included per respondent: one observation from the baseline and another from the post-treatment period.
Therefore, when seeking to retrieve the number of respondents included in the analysis, it is necessary to divide the number of observations from the table by two to account for the number of data points per
respondent.

4. | Surveys on stand-alone solar PuE appliances 55
Table 24 DiD treatment effect on sales, assets, customer origin and food security, Benin
(1)
Sales
(2)
Sales
(3)
Assets
(4)
Assets
(5)
Having customers
from outside
the municipality
(6)
Having customers
from outside
the municipality
(7)
Food security
(8)
Food security
Post 128,375.4*** 117,049.4*** 0.235*** 0.232*** 0.0914 0.0909 0.0338 0.0234
(26,034.1) (33,291.5) (0.0130) (0.0150) (0.147) (0.147) (0.166) (0.175)
GIZ and
non-GIZ treatment
0.598 -0.00993 0.0316 0.0328
(0.630) (0.00986) (0.146) (0.166)
GIZ and non-GIZ DiD -46,843.4 0.0139 0.244 0.157
(31,055.1) (0.0200) (0.207) (0.241)
GIZ treatment 0.225 -0.0191* 0.129 0.169
(0.732) (0.0115) (0.163) (0.194)
GIZ DiD -34,344.5 0.00835 0.0613 0.263
(38,182.6) (0.0235) (0.231) (0.287)
Constant 0.736** 0.716* 0.193*** 0.196*** -0.0481 0.0439 0.848*** 0.832***
(0.307) (0.365) (0.00741) (0.00930) (0.104) (0.104) (0.116) (0.122)
Observations 202 134 698 598 698 598 686 592
R2 0.304 0.310 0.490 0.471
Pseudo R2 0.009 0.005 0.004 0.015
Sample Full GIZ Full GIZ Full GIZ Full GIZ
Source: DEval, own table. T = GIZ and non-GIZ treated; GIZ treatment sample, C = users of non-renewable energy. Robust standard errors in parentheses,
* p < 0.10, ** p < 0.05, *** p < 0.01. OLS regressions if R2 is reported, probit regressions otherwise. All outcomes measured in 2015 and 2022/2023. Outliers trimmed

4. | Surveys on stand-alone solar PuE appliances 56
DiD analysis: farmers
Table 25 and Table 26 show the results when the sample is restricted to farmers. The only treatment effect that is robust is
appliances on energy expenses. Apart from that, none of the treatment effects from prior analyses persists for farmers.
the negative impact of solar
Table 25 DiD treatment effect on revenue, processing, energy expenses, number of employees, Benin, farmers only
(1)
Revenue
(2)
Revenue
(3)
Processing
(4)
Processing
(5)
Energy expenses
(6)
Energy expenses
(7)
Number of employees
(8)
Number of employees
Post 403762.4*** 414904.7*** 5.00e-16 -9.93e-17 21609.9*** 22412.5*** 0.338 0.134
(91,787.8) (96,701.9) (0.388) (0.323) (3,462.7) (3,668.4) (0.688) (0.672)
GIZ and
non-GIZ treatment
34,198.5 -3,010.0 0.682
(62,164.5) (2,016.0) (0.751)
GIZ and
non-GIZ DiD
211,812.3 -13,495.4*** 0.333
(151,382.2) (4,642.3) (1.071)
GIZ treatment 2,956.4 -3,596.5* 0.844
(65,246.5) (2,094.1) (0.868)
GIZ DiD 270,328.9 -14,630.5*** 0.786
(168,291.0) (5,035.3) (1.233)
GIZ and
non-GIZ treatment
(vs. no modern energy)
0.274
(0.323)
GIZ and
non-GIZ DiD
(vs. no modern energy)
-5.12e-16
(0.457)

4. | Surveys on stand-alone solar PuE appliances 57
(1)
Revenue
(2)
Revenue
(3)
Processing
(4)
Processing
(5)
Energy expenses
(6)
Energy expenses
(7)
Number of employees
(8)
Number of employees
GIZ treatment
(vs. no modern energy)
0.538*
(0.294)
GIZ DiD
(vs. no modern energy)
-4.49e-16
(0.415)
Constant 394,389.1*** 397,883.8*** -1.555*** -1.766*** 12,648.6*** 12,714.5*** 3.034*** 2.916***
(46,170.9) (46,005.0) (0.274) (0.229) (1,549.1) (1,594.1) (0.480) (0.488)
Observations 300 268 470 434 416 374 266 232
R2 0.156 0.172 0.160 0.173 0.015 0.026
Pseudo R2 0.010 0.036
Sample Full GIZ Full GIZ Full GIZ Full GIZ
Source: DEval, own table. T = GIZ and non-GIZ treated; GIZ treatment sample, C = users of non-renewable energy, except for processing, for this
errors in parentheses, * p<0.10, ** p<0.05, *** p<0.01. OLS regressions if R2 is reported, probit regressions otherwise.
Note: All outcomes measured in 2015 and 2022/2023. Outliers trimmed
outcome C = no modern energy. Robust standard

4. | Surveys on stand-alone solar PuE appliances 58
Table 26 DiD treatment effect on assets, customer origin and food security, Benin, farmers only
(1)
Assets
(2)
Assets
(3)
Having customers
from outside
the municipality
(4) Having customers
from outside
the municipality
(5)
Food security
(6)
Food security
Post 0.237*** 0.236*** 0.175 0.167 0.0880 0.0869
(0.0160) (0.0166) (0.196) (0.199) (0.209) (0.219)
GIZ and
non-GIZ treatment
-0.00964 -0.0815 0.111
(0.0109) (0.187) (0.206)
GIZ and non-GIZ DiD 0.00868 0.343 0.162
(0.0235) (0.272) (0.301)
GIZ treatment -0.0193* 0.00322 0.227
(0.0108) (0.202) (0.229)
GIZ DiD 0.00142 0.210 0.291
(0.0252) (0.293) (0.343)
Constant 0.187*** 0.188*** 0.335** 0.346** 0.753*** 0.723***
(0.00704) (0.00723) (0.137) (0.140) (0.145) (0.153)
Observations 420 374 420 374 414 370
R2 0.529 0.530
Pseudo R2 0.018 0.011 0.008 0.023
Sample Full GIZ Full GIZ Full GIZ
Source: DEval, own table. T = GIZ and non-GIZ treated; GIZ treatment sample, C = users of non-renewable energy. Robust standard errors in parentheses, * p
regressions if R2 is reported, probit regressions otherwise. All outcomes measured in 2015 and 2022/2023. Outliers trimmed
< 0.10, ** p < 0.05, *** p < 0.01. OLS

4. | Surveys on stand-alone solar PuE appliances 59
DiD analysis: women
These results are also further confirmed by the subgroup analysis for female entrepreneurs, reported in Table 27 and Table 28. It should be mentioned that
the subgroup analysis for women has a low number of case numbers, resulting in the omission of certain outcomes. As such, the results for this subgroup
should ideally be regarded as supplementary for the sake of corroborating the robustness of other analyses.
Table 27 Treatment effect on revenue, processing, energy expenses, number of employees, Benin, women only
(1)
Revenue
(2)
Revenue
(3)
Processing
(4)
Processing
(5)
Energy expenses
(6)
Energy expenses
(7)
Number of employees
(8)
Number of employees
Post 265,276.2*** 261,180.9*** -1.57e-16 -3.79e-16 17,796.8*** 21,406.7*** 0.221 0.273
(96,814.6) (81,694.7) (0.508) (0.463) (2,974.2) (3,143.7) (0.510) (0.530)
GIZ and
non-GIZ treatment
43,243.0 1,115.6 1.335
(102,227.8) (2,839.0) (1.276)
GIZ and non-GIZ DiD -29,165.1 -9,368.2 0.000862
(228,082.3) (6,701.6) (1.552)
GIZ treatment 62,810.8 2,338.2 -0.815***
(114,031.8) (3,655.8) (0.302)
GIZ DiD 88,819.1 -17,473.4** 1.227
(320,550.0) (7,965.5) (1.218)
GIZ and
non-GIZ treatment
(vs. no modern energy)
-0.371
(0.458)
GIZ and non-GIZ DiD
(vs. no modern energy)
-6.06e-16
(0.647)

4. | Surveys on stand-alone solar PuE appliances 60
(1)
Revenue
(2)
Revenue
(3)
Processing
(4)
Processing
(5)
Energy expenses
(6)
Energy expenses
(7)
Number of employees
(8)
Number of employees
GIZ treatment
(vs. no modern energy)
-0.459
(0.494)
GIZ DiD
(vs. no modern energy)
4.92e-16
(0.699)
Constant 253,979.2*** 219,007.3*** -0.798** -0.924*** 6,420.1*** 6,061.8*** 0.776** 0.815***
(59,482.5) (47,656.6) (0.359) (0.327) (1,462.3) (1,569.3) (0.362) (0.302)
Observations 124 110 106 88 176 150 76 62
R2 0.055 0.076 0.128 0.188 0.067 0.125
Pseudo R2 0.017 0.026
Sample Full GIZ Full GIZ Full GIZ Full GIZ
Source: DEval, own table.
errors in parentheses, * p
T = GIZ and non-GIZ treated; GIZ treatment sample, C = users of non-renewable energy, except for processing, for this outcome C = no modern energy. Robust standard
< 0.10, ** p < 0.05, *** p < 0.01. OLS regressions if R2 is reported, probit regressions otherwise. All outcomes measured in 2015 and 2022/2023. Outliers trimmed

4. | Surveys on stand-alone solar PuE appliances 61
Table 28 Treatment effect on sales, assets, customer origin and food security, Benin, women only
(1)
Sales
(2)
Sales
(3)
Assets
(4)
Assets
(5)
Having customers from
outside
the municipality
(6)
Having customers
from outside
the municipality
(7) Food security (8) Food security
Post 97,264.9*** 138,861.1** 0.272*** 0.282*** 0.0893 0.0855 0.107 0.140
(29,161.9) (68,268.6) (0.0264) (0.0226) (0.291) (0.304) (0.338) (0.350)
GIZ and
non-GIZ treatment
-0.462 -0.0103 0.230 -0.977***
(1.223) (0.0185) (0.319) (0.332)
GIZ and
non-GIZ DiD
-38,448.2 0.00472 0.00497 0.0852
(35,098.9) (0.0437) (0.449) (0.483)
GIZ treatment -1.078 -0.0172 0.591 -0.780*
(1.006) (0.0209) (0.389) (0.404)
GIZ DiD -65,718.3 -0.0419 -0.255 -0.140
(74,092.2) (0.0515) (0.553) (0.582)
Constant 1.916** 1.078 0.194*** 0.197*** -0.596*** -0.675*** 1.249*** 1.268***
(0.772) (1.006) (0.00950) (0.00887) (0.206) (0.214) (0.228) (0.235)
Observations 102 64 176 150 176 150 174 150
R2 0.303 0.348 0.562 0.572
Pseudo R2 0.007 0.027 0.098 0.080
Sample Full GIZ Full GIZ Full GIZ Full GIZ
Source: DEval, own table. T = GIZ and non-GIZ treated; GIZ treatment sample, C = users of non-renewable energy. Robust standard errors in parentheses,
* p < 0.10, ** p < 0.05, *** p < 0.01. OLS regressions if R2 is reported, probit regressions otherwise. All outcomes measured in 2015 and 2022/2023. Outliers trimmed.

4. | Surveys on stand-alone solar PuE appliances 62
4.4.6 Perception of outcomes and impacts
The most common effects of using solar appliances, as self-reported by the respondents in the surveys, are
increases in production volume and profits, increases in customer numbers and increased satisfaction with
their working conditions (see Figure 17 and Figure 18). The perceived changes of more women-specific
effects and impacts, such as those illustrated in Figure 19, are discussed in the evaluation report in Chapter
6.3.
Figure 17 Perceived changes in enterprise-related outcomes for farmers, Benin
Source: DEval, own figure. Some figures do not add up to 100 % due to rounding
1%
1%
23%
60%
1%
3%
62%
75%
86%
27%
21%
73%
86%
27%
23%
12%
49%
19%
26%
11%
11%
Less More Same
Production volume
Pro

t
Number of employees
Working hours
Number of customers
Investments
0 10 20 30 40 50 60 70 80 90 100
Satisfaction with
conditions
working
Figure 18 Perceived changes in enterprise-related outcomes for trade and other, Benin
Source: DEval, own figure. Some figures do not add up to 100 % due to rounding
24%
33%
43%
71%
86%
24%
33%
76%
76%
43%
29%
14%
52%
33%
24%
24%
14%
Less More Same
Production volume
Pro

t
Number of employees
Working hours
Number of customers
Investments
0 10 20 30 40 50 60 70 80 90 100
Satisfaction with
conditions
working
The focus group participants in Benin primarily reported lower energy expenses (FOKG 31-36), accompanied
by an increase in turnover (FOKG 39), the ability to grow crops in the dry season (FOKG 34) and a
diversification of their production and livestock farming (FOKG 31, 35). Similarly, participants in Senegal
reported decreases in energy expenses (FOKG 11-13, 18) and production costs (FOKG 11-12, 20), as well as
increased production (FOKG 13, 18). Lastly, participants in the focus groups in Uganda also reported
increased crop yields and a boost in income. As a result, they are now able to pay medical bills and school
fees, build up assets and secure food for themselves and their families (FOKG 1-10).

4. | Surveys on stand-alone solar PuE appliances 63
Figure 19 Perceived changes in gender-related outcomes and impacts, Benin
Source: DEval, own figure, Some figures do not add up to 100 % due to rounding
35%
27%
25%
25%
19%
21%
26%
27%
31%
33%
23%
42%
37%
47%
43%
42%
58%
38%
Totally agree Partially agree Don't agree
Male Perception: women spend less
time fetching water
Female Perception: women spend
less time fetching water
Male Perception: women spend less
time for housework
Female Perception: women spend
less time for housework
Male Perception: women’s decision-
making power is increased
Female Perception: women’s
decision-making power is increased
0 10 20 30 40 50 60 70 80 90 100
4.5 Results for Senegal
The analyses performed on the survey data from Senegal follow the same logic as those performed on the
Benin data using the same types of treatment and control groups.
4.5.1 Descriptive results
Table 29 reports descriptive results for the pre-treatment period in 2019 before the matching. The results
suggest that even before the matching, the GIZ beneficiaries and the control group are similar pre-treatment
on a number of outcomes. Their means are statistically indistinguishable.
Table 29 Mean comparison tests of outcomes in 2019 (pre-treatment),
GIZ beneficiaries vs. control group, Senegal
Obs Mean Std.dev t-test
Control Treatment Control Treatment Control Treatment Control/Treatment
Quantity sold
(dry season)
44 112 14,590.23 6,763.69 75,058.00 35,264.23 7,826.54
Percentage of
customers
from outside
the
municipality
19 72 93.26 87.93 22.41 19.81 5.33
Quantity sold 52 153 14,097.65 6,757.10 69,243.24 39,881.31 7,340.55
Energy
expenses
45 134 339,444.44 257,242.54 435,760.28 604,298.51 82,201.91

4. | Surveys on stand-alone solar PuE appliances 64
Obs Mean Std.dev t-test
Control Treatment Control Treatment Control Treatment Control/Treatment
Revenue 50 138 3.30 3.04 1.57 1.52 0.26
Number of
employees
57 168 5.30 4.77 13.41 10.87 0.52
Food security 57 168 0.56 0.47 0.50 0.50 0.09
Source: DEval, own table. Mean comparison between the control (non-renewable energy) and treatment group (GIZ) in Senegal in
2019, outcome processing omitted due to a low number of cases, equal variances, outliers not trimmed.
Note: The variable revenue was used as an ordinal variable with five categories: category 1: 0-100,000,
category 2: >100,000 – 200,000, category 3: > 200,000 – 500,000, category 4: > 500,000 – 1,000,000, category 5: > 1,000,000.
* p < 0.10, ** p < 0.05, *** p < 0.01
Table 30 presents the mean comparison tests for the outcomes of interest post-treatment for GIZ
beneficiaries compared to the control group. Overall, there is a positive time trend within the treatment
group between 2019 and 2023. The treatment group is better off in 2023 on most outcome indicators (with
the exception of percentage of customers serviced outside the municipality and the number of employees)
and their energy expenses are substantially lower than in 2019. Energy expenses is the only outcome
according to which the treatment group is statistically significantly better off than the control group in 2023.
Table 30 Mean comparison tests of outcomes in 2023 (post-treatment),
GIZ beneficiaries vs. control group, Senegal
Obs Mean Std.dev t-test
Control Treatment Control Treatment Control Treatment Control/Treatment
Quantity sold
(dry season)
44 112 29,438.16 7,813.49 150,524.97 38,566.65 21,624.67
Processing 57 168 0.09 0.15 0.29 0.36 -0.06
Percentage
of customers
from outside
the municipality
22 106 84.50 82.93 31.68 26.58 1.57
Quantity sold 57 168 24,739.37 7,305.17 132,353.51 41,988.07 17,434.20
Energy expenses 57 168 659,129.82 126,502.98 2.51e+06 313,908.55 532,626.85***
Revenue 55 168 3.51 3.67 1.56 1.27 -0.16
Number of
employees
57 168 5.30 4.77 13.41 10.87 0.52
Food security 57 168 0.54 0.48 0.50 0.50 0.07
Source: DEval, own table, Mean comparison between the control (non-renewable energy) and treatment groups (GIZ) in Senegal in
2023, paired, equal variances, outliers not trimmed
Note: The variable revenue was used as an ordinal variable with five categories: category 1: 0 – 100,000,
category 2: > 100,000 – 200,000, category 3: > 200,000 – 500,000, category 4: > 500,000 – 1,000,000, category 5: > 1,000,000.
* p < 0.10, ** p < 0.05, *** p < 0.01
Whether any of these differences pre- and post-treatment can be interpreted as causal will be analysed in
more sophisticated analyses on the matched sample and in the triangulation with the findings from the focus
groups.

4. | Surveys on stand-alone solar PuE appliances 65
4.5.2 Matching
To establish a foundation for comparing the results, the matching method in Senegal is identical to the one
performed in Benin. The only differences are firstly the exclusion of geographical variables, namely
agricultural zones and rural, and secondly some slight deviations in the operationalisation of matching
variables. The exclusion of agricultural zones and rural is owed to the fact that their inclusion led to a notable
decline in matching quality for Senegal. As mentioned above, this is partly due to the distribution of those
variables in Senegal and perhaps due to country-specific differences in geography and established
infrastructure. Generally, the coding for the matching variables is the same as in Benin.
Concerning the differences in variable operationalisation, quartiles were used for the age of the enterprise
owner, and the variable size of enterprise. The computation of quartiles of the age of the owner resulted in
the following ranges: quartile 1: 18-32 years, quartile 2: 33-41 years, quartile 3: 42-51 years, and quartile 4:
52-78 years. As in the Benin sample, the size of the enterprise is a composite variable that captures
differences between enterprises, which is composed of the size of cultivated land, tertiles of revenue in 2015
and differences in livestock. For the Senegal sample, the variable only distinguishes between small (which
includes enterprises coded as medium in Benin) and large, due to it leading to an improvement in matching
quality. Table 31 and Table 32 show the matching variables that were used for Senegal and include descriptive
information about them.
Table 31 Mean comparison tests for
Senegal
the matching variables 2019, GIZ beneficiaries vs. control group,
Obs Mean Std.dev t-test
Control Treatment Control Treatment Control Treatment Control/Treatment
High-quality
floor
57 168 0.74 0.47 0.44 0.50 0.27***
Age of owner*
(ordinal,
4 categories)*
57 168 2.86 2.51 1.08 1.06 0.35**
Age of
enterprise
(ordinal,
3 categories)*
57 168 2.07 1.76 0.82 0.81 0.31**
Size of
enterprise
(small/medium
v. large)
57 168 0.53 0.40 0.50 0.49 0.12
At least
primary
education
57 168 0.25 0.32 0.43 0.47 -0.07
Source: DEval, own table. Equal variances. GIZ treatment group and control group that uses non-renewable energy.
Note: * p < 0.10, ** p < 0.05, *** p < 0.01. As for the ordinal variables, the lowest category (youngest) is 1 and the highest category
(oldest) is 4 (age of owner) or 3 (age of enterprise) respectively

4. | Surveys on stand-alone solar PuE appliances 66
Table 32 Mean comparison tests for matching variables,
GIZ and non-GIZ treated vs. control group, Senegal
Obs Mean Std.dev t-test
Control Treatment Control Treatment Control Treatment Control/Treatment
High-quality floor 57 262 0.74 0.53 0.44 0.50 0.21***
Age of owner
(ordinal,
4 categories)*
57 262 2.86 2.47 1.08 1.09 0.39**
Age of enterprise
(ordinal,
3 categories)*
57 262 2.07 1.91 0.82 0.81 0.16
Size of enterprise
(D, small/medium
v. large)
57 262 0.53 0.45 0.50 0.50 0.07
At least
primary education
57 262 0.25 0.29 0.43 0.45 -0.04
Source: DEval, own table. Equal variances. GIZ and non-GIZ treated (GIZ & non-GIZ) and control group that uses non-renewable
energy * p < 0.10, ** p < 0.05, *** p < 0.01.
Note: As for the ordinal variables, the lowest category (youngest) is 1 and the highest category (oldest) is 4 (age of owner)
or 3 (age of enterprise) respectively
Mirroring Benin, Table 33 and Table 34 demonstrate that the number of available cases differs widely
between outcomes and time. We once again applied the same matching diagnostics, Rubin’s B and R, to
estimate the quality of the matching. The results, reported in both tables, confirms that the threshold of 25
for B and the range of 0.5 to 2 for R can be upheld. This marks the performed matching for Senegal as an
overall success in terms of achieving a balance between the samples. Due to a comparatively low number of
cases for the control group, there are more off-support cases within the treatment group than in Benin. This
is simply because there is a smaller pool of respondents to draw from when trying to find a match in terms
of characteristics for the treatment group. Figure 20 and Figure 21 show how the matching helped make
control and treatment group statistically more similar to each other, as illustrated for the outcome of energy
expenses.
Overall, however, the matching for the full sample group performs worse in Senegal than in Benin, which
may be an indication that the respondents in Senegal are more heterogenous. However, the opposite trend
can be identified for the GIZ sample. While a certain homogeneity can already be identified for GIZ
beneficiaries in Benin, this trend is even more pronounced in Senegal. Lastly, it is important to note that the
remarks made regarding the matching quality for the subgroups in Benin also apply to Senegal.

4. | Surveys on stand-alone solar PuE appliances 67
Table 33 Results of the matching per outcome in 2019 and 2023,
GIZ beneficiaries vs. control group, Senegal
Outcome Obs Common Support Rubin’s B Rubin’s R
2019 2023
Treated Untreated Treated Untreated Treated Untreated Matched Matched
Revenue 119 48 168 55 106 47 12.7 1.40
Energy expenses 134 45 168 57 117 45 11.2 1.19
Number of
employees
168 57 168 57 145 57 12.5 1.16
Customers
from out of
municipality
72 19 106 22 60 18 21.6 1.24
Asset index 168 57 168 57 145 57 12.5 1.16
Food security 168 57 168 57 145 57 12.5 1.16
Quantity sold 153 52 168 57 135 52 12.4 1.12
Quantity sold
(dry season)
112 44 112 44 96 44 14.5 1.42
Source: DEval, own table. T = GIZ treatment, C = non-renewable energy
Note: Outcome processing omitted due to the variable having too few cases for 2019. Outliers trimmed
Table 34 Results of the matching per outcome in 2019 and 2023,
GIZ and non-GIZ treated vs. control group, Senegal
Outcome Obs Common Support Rubin’s B Rubin’s R
2019 2023
Treated Untreated Treated Untreated Treated Untreated Matched Matched
Revenue 197 48 260 55 182 47 18.0 1.42
Energy expenses 220 45 262 57 201 45 17.8 1.62
Number of
employees
262 57 262 57 243 57 22.4 1.51
Customers
from out of
municipality
105 19 147 22 83 18 13.4 1.17
Asset index 262 57 262 57 243 57 22.4 1.51
Food security 262 57 262 57 243 57 22.4 1.51
Quantity sold 238 52 262 57 221 52 22.5 1.53
Quantity sold
(dry season)
195 44 195 44 152 44 8.6 1.34
Source: DEval, own table. T = GIZ and non-GIZ treated, C = non-renewable energy
Note: Outcome processing omitted due to the variable having too few cases for 2019. Outliers trimmed

4. | Surveys on stand-alone solar PuE appliances 68
Figure 20 Density plot for outcome energy expenses for before and after matching state,
GIZ and non-GIZ treated, Senegal
Source: DEval, own figure
1 2 3 4 5
Kernel density
.6 1
Propensity scores BEFORE matching
1 2 3 4 5
Kernel density
.6 1
Propensity scores AFTER matching
Treated Control
.7 .8 .9 .7 .8 .9
Figure 21 Density plot for outcome energy expenses for before and after matching state,
GIZ treatment sample, Senegal
Source: DEval, own figure
0123 4
Kernel density
.4 1
01234
Kernel density
.4 .9
Treated Control
.6.8 .5 .6 .7 .8
Propensity scores BEFORE matching Pr

4. | Surveys on stand-alone solar PuE appliances 69
4.5.3 Cross-sectional analysis
The results of the cross-sectional analyses performed on the matched sample are largely in line with the
simple comparison using t-tests reported in Table 29 and Table 30. The treatment groups have lower energy
expenses than the control group (see Table 37 and Table 38). Furthermore, as Table 35 and Table 36 show,
farmers among both GIZ beneficiaries and non-GIZ treated are more likely to cultivate during the dry season
than those who do not use modern energy.33 Apart from that, the treatment group appears to have no
significant advantage over the control group regarding the size of the personal assets of the owners or the
enterprises, their food security, the quantity in tons that they sell during a period (dry season or otherwise),
their revenues, their number of employees, the share of customers who they serve from outside their
municipality or their likelihood of processing products before selling them.
Table 35 Cross-sectional treatment
and cultivation during dry
effect on assets, food security, quantity sold
season, GIZ beneficiaries vs. control group, Senegal
(1)
Assets
(2)
Food security
(3)
Quantity sold
(4)
Quantity sold
(dry season)
(5)
Cultivation
during dry season
GIZ treatment -0.176 -0.243 -1,798.1 -1,853.2
(0.227) (0.213) (1,604.7) (1,486.8)
GIZ treatment
(vs. no modern energy)
2.464***
(0.454)
Constant 3.334*** 0.165 6,399.1*** 6,296.0*** -0.824**
(0.195) (0.186) (1,507.5) (1,387.2) (0.402)
Observations 202 202 202 140 113
R2 0.004 0.011 0.019
Pseudo R2 0.007 0.481
Source: DEval, own table. T = GIZ treatment sample vs. C = non-renewable energy for all outcomes except for cultivation during dry
season, C for this outcome = no use of modern energy. All outcomes measured in 2023. Robust standard errors in parentheses,
Note: * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed
Table 36 Cross-sectional treatment effect on assets, food security, quantity sold and cultivation
during dry season, GIZ and non-GIZ treated vs. control group, Senegal
(1)
Assets
(2)
Food security
(3)
Quantity sold
(4)
Quantity sold
(dry season)
(5)
Cultivation
during dry season
GIZ and
non-GIZ treatment
-0.222 -0.0221 -941.3 -159.2
(0.200) (0.196) (1,591.8) (1,260.3)
33 We were unable to run the comparison only between GIZ beneficiaries and those in the control group that use fossil energy, because the sample
was too small for the model to converge.

4. | Surveys on stand-alone solar PuE appliances 70
(1)
Assets
(2)
Food security
(3)
Quantity sold
(4)
Quantity sold
(dry season)
(5)
Cultivation
during dry season
GIZ and non-GIZ
treatment
(vs. no modern energy)
2.665***
(0.418)
Constant 3.296*** 0.151 6,719.9*** 5,730.2*** -0.827**
(0.176) (0.178) (1,493.8) (1,125.1) (0.377)
Observations 300 300 300 197 196
R2 0.006 0.003 0.000
Pseudo R2 0.000 0.520
Source: DEval, own table. T = GIZ and non-GIZ treated vs. C = non-renewable energy for all outcomes except for cultivation during dry
season, C for this outcome = no use of modern energy. All outcomes measured in 2023. Robust standard errors in parentheses
Note: * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed
Table 37 Cross-sectional treatment effect on revenue, processing, energy expenses,
number of employees, customer origin, GIZ beneficiaries vs. control group, Senegal
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of employees
(5)
Share of customers
from outside
municipality
GIZ treatment 0.145 -157,319.0*** 0.0347 4.159
(0.234) (35,695.5) (0.694) (9.887)
GIZ treatment
(vs. no modern energy)
0.285
(0.180)
Constant 3.566*** -1.327*** 257,167.3*** 3.752*** 77.73***
(0.209) (0.135) (32,945.1) (0.603) (9.462)
Observations 204 397 202 202 109
R2 0.003 0.142 0.000 0.005
Pseudo R2 0.010
Source: DEval, own table. T = GIZ treatment sample vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy
Note: All outcomes measured in 2023. Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01. OLS-Regression
if R2 and probit regression if pseudo R2 is reported. Unlike in Benin, the revenue variable is ordinal (1 to 5) and the share of customers
variable is numerical. Outliers trimmed

4. | Surveys on stand-alone solar PuE appliances 71
Table 38 Cross-sectional treatment effect on revenue, processing, energy expenses, number of
employees, customer origin, GIZ and non-GIZ treated vs. control group, Senegal
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of employees
(5) Share of customers
from outside
municipality
GIZ and
non-GIZ treatment
0.0400 -190,662.5*** -0.0538 1.820
(0.218) (33,262.5) (0.642) (8.571)
GIZ and
non-GIZ treatment
(vs. no modern energy)
0.194
(0.165)
Constant 3.607*** -1.349*** 271043.2*** 3.811*** 79.94***
(0.197) (0.131) (31733.9) (0.583) (8.232)
Observations 296 487 300 300 141
R2 0.000 0.203 0.000 0.001
Pseudo R2 0.005
Source: DEval, own table. T = GIZ and non-GIZ treated vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy. All outcomes measured in 2023. Robust standard errors in parentheses, * p < 0.10, ** p < 0.05,
*** p < 0.01. OLS regression if R2 and probit regression if pseudo R2 is reported. Unlike in Benin, the revenue variable is ordinal (1 to
5) and the share of customers variable is numerical. Outliers trimmed
Cross-sectional subgroup analysis: farmers
Table 39 to Table 42 show the results after restricting the sample to farmers, where the findings are similar
to the sample including all MSME types. As Table 39 and Table 40 show, farmers in the treatment group are
more likely to cultivate during the dry season than the control group, but the intervention appears to be
associated with a smaller planted area for GIZ beneficiaries (see Table 39). Lastly, for farmers the treatment
also appears to be associated with a decrease in energy expenses (see Table 41and Table 42).
Table 39 Cross-sectional treatment effect on planted area, food security, quantity sold and
cultivation during dry season, GIZ beneficiaries vs. control group, Senegal, farmers only
(1)
Planted area
(2)
Assets
(3)
Food security
(4)
Quantity sold
(5)
Quantity sold
(dry season)
(6)
Cultivation during
dry season
GIZ treatment -0.434* 0.0246 -0.0507 -3,081.1 -2,151.0
(0.224) (0.246) (0.249) (2,035.1) (1,482.0)
GIZ treatment
(vs. no modern energy)
2.463***
(0.457)
Constant 1.338*** 3.104*** 0.0879 9147.0*** 6390.1*** -0.823**
(0.190) (0.200) (0.215) (1,926.4) (1,395.5) (0.405)
Observations 145 145 145 145 139 113
R2 0.035 0.000 0.029 0.027
Pseudo R2 0.000 0.480
Source: DEval, own table. T = GIZ treatment sample vs. C = non-renewable energy for all outcomes except for cultivation during dry
season, C for this outcome = no use of modern energy. All outcomes measured in 2023. Robust standard errors in parentheses
Note: * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed

4. | Surveys on stand-alone solar PuE appliances 72
Table 40 Cross-sectional treatment effect on planted area, food security,
quantity sold and cultivation during dry season,
GIZ and non-GIZ treated vs. control group, Senegal, farmers only
(1)
Planted area
(2)
Assets
(3)
Food security
(4)
Quantity sold
(5)
Quantity sold
(dry season)
(6)
Cultivation
during
dry season
GIZ and non-GIZ
treatment
-0.222 -0.00603 0.235 -586.7 -179.6
(0.203) (0.226) (0.226) (1,764.9) (1,276.6)
GIZ and
non-GIZ treatment
(vs. no modern energy)
2.679***
(0.417)
Constant 1.322*** 3.133*** 0.00548 8,482.7*** 5,781.4*** -0.835**
(0.175) (0.191) (0.202) (1,588.6) (1,141.8) (0.377)
Observations 202 202 202 202 196 198
R2 0.009 0.000 0.001 0.000
Pseudo R2 0.006 0.524
Source: DEval, own table. T = GIZ and non-GIZ treated vs. C = non-renewable energy for all outcomes except for cultivation during dry
season, C for this outcome = no use of modern energy. All outcomes measured in 2023. Robust standard errors in parentheses
Note: * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed
Table 41 Cross-sectional treatment effect on revenue, processing, energy expenses, number of
employees, customer origin, GIZ beneficiaries vs. control group, Senegal, farmers only
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of employees
(5)
Share of customers
from outside
municipality
GIZ treatment 0.204 -168,434.9*** -1.254 8.284
(0.255) (42,662.1) (0.817) (10.20)
GIZ treatment
(vs. no modern energy)
-0.247
(0.300)
Constant 3.598*** -1.461*** 295,029.0*** 4.551*** 78.99***
(0.228) (0.217) (38,617.4) (0.748) (9.829)
Observations 144 259 145 145 78
R² 0.006 0.150 0.026 0.023
Pseudo R² 0.008
Source: DEval, own table. T = GIZ treatment sample vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy. All outcomes measured in 2023. Robust standard errors in parentheses,
Note: * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed

4. | Surveys on stand-alone solar PuE appliances 73
Table 42 Cross-sectional treatment effect on revenue, processing, energy expenses, number of
employees, customer origin, GIZ and non-GIZ treated vs. control group, Senegal, farmers
only
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of employees
(5)
Share of customers
from outside
municipality
GIZ and non-GIZ
treatment
0.0675 -225,149.2*** -0.949 1.256
(0.259) (37,767.3) (0.754) (8.359)
GIZ and
non-GIZ treatment
(vs. no modern energy)
-0.288
(0.260)
Constant 3.550*** -1.463*** 307,497.3*** 4.480*** 83.26***
(0.232) (0.204) (35,807.7) (0.695) (7.998)
Observations 200 345 202 202 104
R2 0.001 0.268 0.014 0.001
Pseudo R2 0.011
Source: DEval, own table. T = GIZ and non-GIZ treated vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy. All outcomes measured in 2023. Robust standard errors in parentheses, * p < 0.10, ** p < 0.05,
*** p < 0.01. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers trimmed
Cross-sectional subgroup analysis: women
In order to improve the matching quality for the subgroup of women, the reference category was adjusted
for the variables of high-quality flooring and age of owner. Moreover, the subgroup of women contained low
case numbers for certain outcomes. Therefore, some outcomes had to be omitted since convergence could
not be achieved in the regression models.
Table 43 to Table 46 present the results of the subgroup analysis for women. It should be noted that the case
numbers for the analyses, especially for the GIZ treatment group, are very low and should therefore be
interpreted with caution. Nevertheless, the negative treatment effect for energy expenses persists, though
it is only statistically significant when both treatment groups are taken together (including T1 and T2, Table
46), which is likely due to the low number of cases for the GIZ treatment group.
Interestingly, the use of solar appliances appears to be related to additional effects for women. It appears to
negatively impact their assets (see Table 43 and Table 44), the number of employees (see Table 45, GIZ
beneficiaries only) and the likelihood of processing products before selling them compared to those not using
modern energy (see Table 46, when GIZ and non-GIZ treated are taken together).

4. | Surveys on stand-alone solar PuE appliances 74
Table 43 Cross-sectional treatment effect on planted area, assets, food security and quantity sold,
GIZ beneficiaries vs. control group, Senegal, women only
(1)
Assets
(2)
Food security
(3)
Quantity sold
GIZ treatment -1.397*** 0.705 973.2**
(0.353) (0.599) (447.4)
Constant 4.818*** 0.298 45.79
(0.146) (0.484) (46.54)
Observations 28 28 28
R2 0.301 0.119
Pseudo R2 0.056
Source: DEval, own table.
errors in parentheses, * p
trimmed
T = GIZ treatment sample vs. C = non-renewable energy. All outcomes measured in 2023. Robust standard
< 0.10, ** p < 0.05, *** p < 0.01. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers
Table 44 Cross-sectional treatment effect on assets, food security and quantity
GIZ treated vs. control group, Senegal, women only
sold, GIZ and non-
(1)
Assets
(2)
Food security
(3)
Quantity sold
GIZ and
non-GIZ treatment
-1.568*** 0.269 2,217.2***
(0.369) (0.480) (720.0)
Constant 4.568*** 0.128 35.20
(0.300) (0.447) (35.44)
Observations 64 64 64
R2 0.265 0.082
Pseudo R2 0.008
Source: DEval, own table, T = GIZ and non-GIZ treated vs. C = non-renewable energy. All outcomes measured in 2023. Robust standard
errors in parentheses, * p<0.10, ** p<0.05, *** p<0.01. OLS regression if R2 and probit regression if pseudo R2 is reported. Outliers
trimmed. The sample size was too small to estimate effects on the planted area in the subsample of women

4. | Surveys on stand-alone solar PuE appliances 75
Table 45 Cross-sectional treatment effect on revenues, processing, energy expenses and number of
employees, GIZ beneficiaries vs. control group, Senegal, women only
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of employees
GIZ treatment -0.00248 -57,185.8 4.337**
(0.633) (41,750.8) (1.606)
GIZ treatment
(vs. no modern energy)
0.464
(0.284)
Constant 3.687*** -0.959*** 106,501.6*** 1.347*
(0.557) (0.225) (32655.3) (0.705)
Observations 28 150 28 28
R2 0.000 0.079 0.169
Pseudo R2 0.025
Source: DEval, own table. T = GIZ treatment sample vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy
Note: All outcomes measured in 2023. Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regression
if R2 and probit regression if pseudo R2 is reported. Outliers trimmed
Table 46 Cross-sectional treatment effect on revenues, processing, energy expenses and number of
employees, GIZ and non-GIZ treated vs. control group, Senegal, women only
(1)
Revenue
(2)
Processing
(3)
Energy expenses
(4)
Number of employees
GIZ and
non-GIZ treatment
-0.0737 -70,296.5** 2.621
(0.611) (31,606.5) (1.637)
GIZ and
non-GIZ treatment
(vs. no modern energy)
0.503**
(0.221)
Constant 3.565*** -1.144*** 104987.4*** 2.342
(0.571) (0.170) (29,498.6) (1.420)
Observations 64 184 64 64
R2 0.001 0.148 0.058
Pseudo R2 0.030
Source: DEval, own table. T = GIZ and non-GIZ treated vs. C = non-renewable energy for all outcomes except for processing, C for this
outcome = no use of modern energy,
Note: All outcomes measured in 2023. Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regression if
R2 and probit regression if pseudo R2 is reported. Outliers trimmed

4. | Surveys on stand-alone solar PuE appliances 76
4.5.4 Difference-in-differences (DiD) analysis
The DiD analyses on the survey data from Senegal are run on the matched samples, just like with the data
from Benin. The results for both types of treatment groups are presented in Table 47and Table 48. Overall,
most of the effects observed in the cross-sectional comparison are not robust to using a DiD estimator.
However, there are two interesting effects. The DiD analyses confirm the finding that energy expenses have
decreased as a result of the interventions, even though this effect is only statistically significant for the
subgroups of farmers (see Table 49) and female entrepreneurs (see Table 51), and not for the sample of all
MSME types (see Table 47). In addition, the results suggest a negative effect on the amount of personal assets
by the owners of the enterprises, which appears robust across the MSME types (Table 48, Table 50 and Table
52).

4. | Surveys on stand-alone solar PuE appliances 77
Table 47 DiD treatment effect on revenue, energy expenses, number of employees and food security, Senegal
(1)
Revenue
(2)
Revenue
(3)
Energy expenses
(4)
Energy expenses
(5)
Number of employees
(6)
Number of employees
(7)
Food security
(8)
Food security
Post 0.292 0.373 1,617.8 4,191.2 1.064 1.022 -0.0321 -0.0218
(0.320) (0.338) (65,215.4) (72,742.8) (0.789) (0.829) (0.252) (0.263)
GIZ and
non-GIZ treatment
0.0899 -99,762.0* 0.0224 -0.0438
(0.260) (58,291.6) (0.586) (0.196)
GIZ and non-GIZ DiD 0.0976 -109,652.6 -0.0761 0.0217
(0.355) (69,953.0) (0.869) (0.277)
GIZ treatment 0.170 -54,557.0 -0.267 -0.213
(0.288) (68,324.9) (0.639) (0.214)
GIZ DiD 0.0982 -128,152.8 0.302 -0.0301
(0.385) (81,238.8) (0.944) (0.302)
Constant 3.333*** 3.226*** 287,553.0*** 275,347.6*** 2.747*** 2.729*** 0.183 0.187
(0.234) (0.253) (53,678.1) (59,972.5) (0.532) (0.569) (0.178) (0.186)
Observations34 458 306 492 324 600 404 600 404
R2 0.015 0.028 0.094 0.065 0.016 0.023
Pseudo R2 0.000 0.006
Sample Full GIZ Full GIZ Full GIZ Full GIZ
Source: DEval, own table. T = GIZ and non-GIZ treated; GIZ treatment sample, C = users of non-renewable energy.
34 All difference-in-differences tables presented report the total number of statistical observations, wherein two data points are included per respondent: one observation from 2019 and another from 2023 for Senegal.
Therefore, when seeking to retrieve the number of respondents included in the analysis, it is necessary to divide the number of observations from the table by two to account for the number of data points per
respondent.

4. | Surveys on stand-alone solar PuE appliances 78
Note: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regressions if R2 is reported, probit regressions otherwise. Unlike in Benin, the share of customers from
outside the municipality is numerical. All outcomes measured in 2019 and 2023. Outliers trimmed
Table 48 DiD treatment effect on assets, share of customers from outside the municipality and quantity sold, Senegal
(1)
Assets
(2)
Assets
(3)
Share of customers
from outside
municipality
(4)
Share of customers
from outside
municipality
(5)
Quantity sold
(dry season)
(6)
Quantity sold
(dry season)
(7)
Quantity sold
(8)
Quantity sold
Post 0.554** 0.528** -6.313 -6.942 2,689.6** 3,050.1* 2,170.0 1,932.3
(0.246) (0.262) (15.82) (17.41) (1,266.0) (1,589.6) (1,822.2) (1,811.4)
GIZ and
non-GIZ treatment
0.476** -1.603 291.5 -164.0
(0.193) (11.52) (712.2) (1,104.8)
GIZ and non-GIZ DiD -0.698** 7.145 -212.7 -406.2
(0.278) (16.19) (1,440.3) (1,966.1)
GIZ treatment 0.401* 0.516 -297.6 -391.7
(0.210) (12.77) (879.6) (1,163.4)
GIZ DiD -0.576* 6.392 -1,555.5 -979.5
(0.309) (17.85) (1,727.5) (1,976.6)
Constant 2.742*** 2.806*** 86.87*** 85.72*** 2,995.1*** 3,245.9*** 4,186.5*** 3,936.4***
(0.172) (0.175) (11.24) (12.47) (611.9) (776.2) (1,006.4) (1,026.9)
Observations 600 404 202 156 392 280 546 374
R2 0.024 0.021 0.008 0.012 0.047 0.053 0.015 0.014
Sample Full GIZ Full GIZ Full GIZ Full GIZ
Source: DEval, own table, T = GIZ and non-GIZ treated; GIZ treatment sample, C = users of non-renewable energy.
Note: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regressions. Unlike in Benin, the share
outcomes measured in 2019 and 2023. Outliers trimmed
of customers from outside the municipality is numerical. All

4. | Surveys on stand-alone solar PuE appliances 79
DiD analysis: farmers
Table 49 DiD treatment effect on revenue, energy expenses, number of employees and food security, Senegal, farmers only
(1)
Revenue
(2)
Revenue
(3)
Energy expenses
(4)
Energy expenses
(5)
Number of employees
(6)
Number of employees
(7)
Food security
(8)
Food security
Post 0.275 0.449 12,547.2 22,182.1 1.625* 1.542 -0.0459 -0.0291
(0.373) (0.382) (61,813.2) (66,435.8) (0.909) (1.026) (0.285) (0.304)
GIZ and
non-GIZ treatment
-0.0351 -80,400.2 0.0562 0.205
(0.293) (54,978.5) (0.648) (0.226)
GIZ and non-GIZ DiD 0.244 -147,367.0** -1.005 0.0295
(0.414) (68,741.5) (0.994) (0.319)
GIZ treatment 0.238 3,408.3 -0.544 -0.00510
(0.320) (64,968.2) (0.763) (0.249)
GIZ DiD 0.144 -181,810.0** -0.710 -0.0456
(0.428) (80,016.5) (1.118) (0.352)
Constant 3.302*** 3.111*** 294,893.4*** 276,940.5*** 2.855*** 3.009*** 0.0513 0.117
(0.263) (0.285) (47,791.3) (51,138.6) (0.586) (0.702) (0.202) (0.215)
Observations 334 244 366 244 404 290 404 290
R2 0.020 0.048 0.112 0.065 0.028 0.041
Pseudo R2 0.006 0.000
Sample Full GIZ Full GIZ Full GIZ Full GIZ
Source: DEval, own table. T = GIZ and non-GIZ treated; GIZ treatment sample, C = users of non-renewable energy.
Note: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regressions. Unlike in Benin, the revenue variable is
2023. Outliers trimmed
ordinal (1 to 5). All outcomes measured in 2019 and

4. | Surveys on stand-alone solar PuE appliances 80
Table 50 DiD treatment effect on assets, customer origin and quantity sold, Senegal, farmers only
(1)
Assets
(2)
Assets
(3)
Share of customers
from outside
municipality
(4)
Share of customers
from outside
municipality
(5)
Quantity sold
(dry season)
(6)
Quantity sold
(dry season)
(7)
Quantity sold
(8)
Quantity sold
Post 0.507* 0.429 -8.222 -6.610 2,752.3** 3,122.4* 2,833.3 2,908.4
(0.283) (0.288) (11.86) (22.78) (1,292.4) (1,601.6) (2,046.7) (2,419.0)
GIZ and
non-GIZ treatment
0.760*** -8.198 174.1 -280.1
(0.234) (8.097) (719.4) (1,298.1)
GIZ and non-GIZ DiD -0.766** 9.955 -261.7 -180.8
(0.325) (12.58) (1,460.1) (2,287.8)
GIZ treatment 0.542** 5.959 -445.2 -830.6
(0.248) (16.94) (880.9) (1,528.5)
GIZ DiD -0.518 6.193 -1,705.8 -1,652.0
(0.350) (23.21) (1,724.0) (2,612.4)
Constant 2.626*** 2.675*** 91.56*** 80.74*** 3,009.6*** 3,267.7*** 5,538.4*** 5,607.6***
(0.209) (0.208) (7.462) (16.64) (622.9) (785.8) (1,137.4) (1,355.8)
Observations 404 290 152 104 388 278 362 262
R2 0.041 0.025 0.018 0.029 0.048 0.060 0.025 0.029
Sample Full GIZ Full GIZ Full GIZ Full GIZ
Source: DEval, own table. T = GIZ and non-GIZ treated; GIZ treatment sample, C = users of non-renewable energy
Note: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regressions. Unlike in Benin, the share
and 2023. Outliers trimmed
of customers variable is numerical. All outcomes measured in 2019

4. | Surveys on stand-alone solar PuE appliances 81
DiD analysis: women
Table 51 DiD treatment effect on revenue, energy expenses and number of employees, Senegal, women only
(1)
Revenue
(2)
Revenue
(3)
Energy expenses
(4)
Energy expenses
(5)
Number of employees
(6)
Number of employees
Post 0.201 0.179 96,852.0** 88,022.8* 1.647 1.320
(0.920) (0.588) (47,307.4) (48,845.3) (1.443) (1.185)
GIZ and non-GIZ treatment -0.609 -3,505.3 2.305***
(0.781) (16,660.1) (0.770)
GIZ and non-GIZ DiD 0.370 -92,076.2* 0.533
(1.068) (49,053.0) (1.809)
GIZ treatment -1.252* -27,971.8 1.142
(0.700) (25,742.4) (1.023)
GIZ DiD 1.266 -47,772.8 3.055
(0.886) (56,929.1) (2.285)
Constant 3.276*** 4.030*** 17,315.6 29,721.8 0.555 0.545
(0.684) (0.492) (13,878.0) (25,682.9) (0.361) (0.351)
Observations 56 32 70 24 118 50
R2 0.031 0.201 0.330 0.308 0.116 0.194
Sample Full GIZ Full GIZ Full GIZ
Source: DEval, own table. T = GIZ and non-GIZ treated; GIZ treatment sample, C = users of non-renewable energy. Robust standard errors in parentheses, * p
regressions. Unlike in Benin, the revenue variable is ordinal (1 to 5). All outcomes measured in 2019 and 2023. Outliers trimmed
< 0.10, ** p < 0.05, *** p < 0.01. OLS

4. | Surveys on stand-alone solar PuE appliances 82
Table 52 DiD treatment effect on assets, quantity sold and food security, Senegal, women only
(1)
Assets
(2)
Assets
(3)
Quantity sold
(4)
Quantity sold
(5)
Food security
(6)
Food security
Post 1.902*** 1.999*** 26.00 33.45 -1.36e-16 2.71e-16
(0.365) (0.362) (26.81) (34.84) (0.646) (0.726)
GIZ and
non-GIZ treatment
0.761** 1,505.5** 0.230
(0.293) (640.4) (0.493)
GIZ and
non-GIZ DiD
-2.282*** 524.2 7.37e-17
(0.467) (1,005.6) (0.697)
GIZ treatment 0.854* 280.0* 0.461
(0.426) (159.5) (0.631)
GIZ DiD -1.812*** 496.5 -1.40e-16
(0.589) (478.7) (0.892)
Constant 2.699*** 2.709*** 1.02e-12 5.68e-14 0.237 0.426
(0.231) (0.283) (0.457) (0.513)
Observations 118 50 114 48 118 50
R2 0.265 0.306 0.065 0.118
Pseudo R2 0.006 0.025
Sample Full GIZ Full GIZ Full GIZ
Source: DEval, own table. T = GIZ and non-GIZ treated; GIZ treatment sample, C = users of non-renewable energy. Robust standard
errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01. OLS regressions if R2 is reported, otherwise probit regression. All outcomes
measured in 2019 and 2023. Outliers trimmed
4.5.5 Perception of outcomes and impacts
The most common effects of using solar appliances, as self-reported by the respondents in the survey in
Senegal, are increases in production volume and profits, increases in customer numbers and increased
satisfaction with their working conditions (Figure 22 and Figure 23), and are therefore similar to the findings
from Benin. More effects, including the perceived changes on gender-related outcomes and impacts (Figure
24), are discussed in the evaluation report in Chapter 6.3.

4. | Surveys on stand-alone solar PuE appliances 83
Figure 22 Perceived changes in enterprise related outcomes for farmers, Senegal
Source: DEval, own figure. Some figures do not add up to 100 % due to rounding
9%
6%
25%
21%
3%
5%
28%
66%
91%
34%
69%
89%
92%
64%
24%
3%
41%
10%
8%
3%
8%
Less More Same
Production volume
Pro

t
Number of employees
Working hours
Number of customers
Investments
0 10 20 30 40 50 60 70 80 90 100
Satisfaction with
conditions
working
Figure 23 Perceived changes in enterprise related outcomes for trade and other, Senegal
Source: DEval, own figure. Some figures do not add up to 100 % due to rounding
9%
4%
9%
22%
4%
9%
9%
48%
87%
48%
52%
83%
78%
70%
43%
9%
43%
26%
13%
13%
22%
Less More Same
Production volume
Pro

t
Number of employees
Working hours
Number of customers
Investments
0 10 20 30 40 50 60 70 80 90 100
Satisfaction with
conditions
working

5. | Mini-grid survey 84
Figure 24 Perceived changes in gender-related outcomes and impacts, Senegal
Source: DEval, own figure. Some figures do not add up to 100 % due to rounding
26%
30%
26%
36%
20%
36%
16%
30%
20%
29%
24%
27%
58%
40%
54%
36%
56%
38%
Totally agree Partially agree Don't agree
Male Perception: women spend less
time fetching water
Female Perception: women spend
less time fetching water
Male Perception: women spend less
time for housework
Female Perception: women spend
less time for housework
Male Perception: women’s decision-
making power is increased
Female Perception: women’s
decision-making power is increased
0 10 20 30 40 50 60 70 80 90 100
5. MINI-GRID SURVEY35
The ERSEN 1 and ERSEN 2 interventions implemented by EnDev facilitated the installation of mini-grids in 90
villages in Senegal between 2016 and 2021. In addition to providing the mini-grids, EnDev also facilitated
access to appliances such as fridges, mills and sewing machines to be used in conjunction with the grids.
Moreover, they set up boutiques to aid in the payment system for the tariffs to use the mini-grid’s energy.
Usually, the same end users who had a boutique also obtained a fridge to facilitate their enterprise. Such
appliances were provided in nine municipalities. The aim of this component was to foster the use of energy
from the mini-grid for productive purposes.
We interviewed the village head and the person responsible for managing the mini-grid at village level in 82
of those villages. The objective of this descriptive survey was firstly to find out how many mini-grids were
functioning up to seven years after their installation, and secondly to assess the potential for the productive
use of the energy from those grids within the villages. The survey contained a range of questions addressed
to the village head or their deputy and to the local manager of the mini-grid. The design of the questionnaire
was closely modelled on a questionnaire administered in the same localities in 2019 by researchers at the
RWI – Leibniz-Institut für Wirtschaftsforschung, Essen (RWI Essen), Julian Rose and Jörg Ankel-Peters.
Our survey was carried out by phone in September and October 2023. Out of a list of 90 villages where mini-
grids were installed, 82 consented to participate in the survey.
5.1 Relevance for SDG 7.1 – energy for all by 2030
The mini-grids examined in Senegal are not very effective at creating initial access. The survey in the mini-
grid villages showed that before the mini-grids were installed, 38 percent of villages (31 villages) used car
batteries, 32 percent (26 villages) used generators and 83 percent (68 villages) used solar panels. This means
35 Contributors: Mame Mor Anta Syll, Whitney Edwards.

5. | Mini-grid survey 85
that a large proportion of the villages already had access to modern electricity, even though the mini-grids
presumably increased the tier level (Figure 25 and Figure 26).
Figure 25 Share of villages by type of energy source before the mini-grids were installed
Source: DEval, own figure
333888%%%
333222%%%
888333%%%
111%%%
Share of villages
Car battery
Generator
Solar panel
Anterior mini-grid
0 10 20 30 40 50 60 70 80 90 100
Figure 26 Energy used before the mini-grids were installed by area
Source: DEval, own figure
31%
29%
21%
26%
24%
19%
43%
47%
50%
60%
100%
50%
Car battery Generator Solar panel Central grid
Anterior mini-grid
Fatick
Kaolack
Kolda
Sédhiou
Thiès
0 10 20 30 40 50 60 70 80 90 100
5.2 Functionality of the mini-grids
While our data does not allow for a causal estimation of the economic impact of the mini-grids on rural
enterprises and populations in these villages, it can be used to assess the potential for economic impacts.
The main prerequisite for the productive use of energy from the grids is that the grids are functional and
provide sufficient energy at the right times.
Out of the 82 villages surveyed, only nine reported still having functioning mini-grids. The villages with their
mini-grids still working are the following: Yacine Mandina (1), Yacine Tambana (2), Néma Diaour (3) in
Sédhiou, Bangalère (4) and Keur Babou Ndity (5) in Foundiougne, Lamel (6) and Sina (7) in Goudomp, Saré

5. | Mini-grid survey 86
Koubé (8) in Kolda and Souaki (9) in Bounkling. Villages with mini-grids which were reported to have never
been functional are Bantanto and Mansang in Bounkiling and Keur Mandiaye Fatim in Foundiougne.
Out of the 82 villages interviewed, 79 of them have at least enjoyed the use of their mini-grids some of the
time. The time of operation was 24 hours a day for 39 villages (or 49 % of the sample) and less than 24 hours
for 40 villages (or 51 % of the sample) as shown in Figure 30.
Figure 27 demonstrates that even mini-grids which started operating quite recently (2018, 2019 and 2020)
already no longer work. Interestingly, some of the mini-grids which were installed comparatively early on
(2013, 2014 and 2016) were reported to still be functional at the time of the survey.
Figure 27 Year of installation and functionality in 2023
Source: DEval, own figure
Year of installation
Amount of units
21
23 22
13
4
212 2
22
3
1
1
1
Mini-grid never functioned Mini-grid originally functioned
Mini-grid still functional
2013 2014 2015 2016 2017 2018 2019 2020 2021
0
5
10
15
20
25
During the dry season, the mini-grids experience power outages frequently (Figure 28). In 35 % of the villages
in the study, the shutdown during the dry season occurs every day. Only few villages had experienced weekly
or monthly shutdowns (7 % and 9 % of villages respectively).

5. | Mini-grid survey 87
Figure 28 Frequency of power outages of the mini-grids during the dry season
Source: DEval, own figure. Figure does not add up to 100 % due to rounding
Every day Every week Every month Sometimes
48%
35%
7%
9%
Entrepreneurs who are not using the mini-grids as their main source of energy use different sources of energy
for their economic activities. The data shows that they primarily use (other) solar energy in 67 % of the
villages. For 12 % of villages, the primary source of energy for income-generating activities which are not
connected to the mini-grids is the central electricity grid. However, in 13 % of villages respondents reported
that they had no source of electricity for the economic activities within their villages (see Figure 29).
Figure 29 Sources of alternative electrical energy used by enterprises and households
(when not connected to mini-grids)
Source: DEval, own figure. Figure does not add up to 100 % due to rounding
11%%
48%
35%
7%
9%
None Solar energy Generator
Electricity from neighbour Car battery Central grid
Firewood
2%
67%
1% 12% 13%
2%

5. | Mini-grid survey 88
Figure 30 Share of villages by mini-grid operating hours (left)
and by reasons for not operating all day (right)
Source: DEval, own figure
777111%%%
5
558
88%
%%
1
114
44%
%%
4
440
00%
%%
Share of Villages
Not enough capacity
Lack of sunlight
Generators
adequately
Mini-grid broke down
time ago
0 20 40 60 80 100
24 hours/day < 24 hours/day
51%
49%
not functioning
a long
Mini-grids have a variety of operational challenges that inhibit their continuous operation (see Figure 30).
The analysis of the provided data indicates that, on average, 71 % of villages face issues with round-the-clock
operation. In 40 % of the villages the mini-grid had completely broken down and stopped working. In the
majority of the remaining villages, the capacity of the mini-grid was not sufficient to operate all day. In 58 %
of the villages mini-grids did not operate all day due to the lack of sunlight. Normally, the mini-grids are
supposed to be powered by a generator that uses fossil fuel, mostly diesel, in the absence of sunlight.
But 14 % of the villages reported that those generators did not work as intended, either because they were
not functional and not repaired or simply because the fuel was not available. It is the operator of the mini-
grid who would have been responsible to supply the generators with fuel.
Overall, these findings underscore the widespread challenges faced by mini-grids, emphasising the need for
addressing technical issues, ensuring operator competency and exploring solutions for consistent energy
availability across different locations. The survey data highlights disparities in household connections to the
mini-grids across the villages. Figure 31 offers insights into the reasons behind households remaining
unconnected. In 22 % percent of the villages, households were on waiting lists, in 13 % of the villages more
households could not get connected due to the capacity of the grid already being full. Other reasons were
households’ inability to afford the connection or trust issues.

5. | Mini-grid survey 89
Figure 31 Share of villages by reasons for not being connected to the mini-grids
Source: DEval, own figure
999%%%
555%%%
222222%%%
888777%%%
Share of villages
Not able to pay
Not convinced
by the system
On the waiting list
Mini-grid capacity is full
0 20 40 60 80 100
5.3 Economic activities in mini-grid villages
Table 53 provides an overview of the different income-generating activities in the villages where mini-grids
were installed as part of the GIZ interventions under study. The mean values represent the percentage of
villages in which the activities were practiced. The main activity in the mini-grid villages is farming and there
are also small enterprises across the board. Related to the predominance of farming, a wide-spread economic
activity is processing of agricultural produce. Also sewing is widespread.
Table 53 Share of villages by income-generating activities
N Mean Min Max Median Std. dev. t-value
Woodwork 82 .378 0.000 1 0 .488 7.017
Hairstyling 82 .195 0.000 1 0 .399 4.431
Sewing 82 .707 0.000 1 1 .458 13.991
Woodwork 82 .512 0.000 1 1 .503 9.222
Small enterprise 82 .841 0.000 1 1 .367 20.735
Catering 82 .378 0.000 1 0 .488 7.017
Bar 82 .012 0.000 1 0 .11 1
Processing (mill) 82 .646 0.000 1 1 .481 12.167
Farming 82 .927 0.000 1 1 .262 32.031
Selling fresh produce 82 .573 0.000 1 1 .498 10.429
Seller 82 .061 0.000 1 0 .241 2.293
Source DEval, own table

5. | Mini-grid survey 90
Table 54 reports the prevalence of various appliances used in the mini-grid villages, to get an overview over
the potential to power appliances with the mini-grid for income generating activities. Like above, the means
represent the percentage of villages in which an appliance was used. The most frequently used appliances in
the villages are electric ovens, freezers, electric irons, computers, electric mills and black and white TVs.
Conversely, fans, oil or fuel mills, and dryers are less commonly used on average. The data, furthermore,
suggest only little variability between villages in the patterns in which the appliances are used.
Table 54 Share of villages by appliances used
N Mean Min Max Median Std. dev. t-value
Charcoal iron 82 .171 0.000 1 0 .379 4.084
Electric iron 82 .305 0.000 1 0 .463 5.96
Generator
fridge
82 .037 0.000 1 0 .189 1.754
Electric fridge 82 .012 0.000 1 0 .11 1
Freezer 82 .293 0.000 1 0 .458 5.789
Electric oven 82 .293 0.000 1 0 .458 5.789
Fan 82 0 0.000 0 0 0 .
Black and
white TV
82 .244 0.000 1 0 .432 5.112
Color TV 82 .073 0.000 1 0 .262 2.529
Computer 82 .244 0.000 1 0 .432 5.112
Oil/fuel mill 82 .061 0.000 1 0 .241 2.293
Electric mill 82 .28 0.000 1 0 .452 5.619
Mechanical
sewing
machine
82 .146 0.000 1 0 .356 3.726
Electric sewing
machine
82 .146 0.000 1 0 .356 3.726
Grinding wheel 82 .024 0.000 1 0 .155 1.423
Welding
machine
82 .049 0.000 1 0 .217 2.038
Dryer 82 .012 0.000 1 0 .11 1
Lighting 82 .317 0.000 1 0 .468 6.132
Other (specify) 82 .012 0.000 1 0 .11 1
Source: DEval, own table
Table 55 shows the income-generating activities for which the mini-grids were used. This overview suggests
that on average, a substantial proportion of enterprises that use the electricity of the mini-grid are engaged
in small enterprise activities or selling fresh produce. Conversely, there appears to be less usage of the mini-
grid energy for woodwork, for running a bar and for farming.

5. | Mini-grid survey 91
Table 55 Main income-generating activities for enterprises connected to the mini-grids
N Mean Min Max Median Std. dev. t-value
Woodwork 82 0 0.000 0 0 0 .
Hairstyling 82 .037 0.000 1 0 .189 1.754
Sewing 82 .122 0.000 1 0 .329 3.354
Small enterprise 82 .768 0.000 1 1 .425 16.388
Catering 82 .024 0.000 1 0 .155 1.423
Bar 82 0 0.000 0 0 0 .
Processing (mill) 82 .061 0.000 1 0 .241 2.293
Farming 82 .012 0.000 1 0 .11 1
Selling fresh produce 82 .524 0.000 1 1 .502 9.45
Shop seller 82 .159 0.000 1 0 .367 3.907
Source: DEval, own table
Lamps, freezers and fridges are the main electric appliances used in conjunction with the mini-grids for
income-generating activities.
The survey also explored the potential of the mini-grids to foster economic activities in general and among
women in particular. There are an average of 20 enterprises per village, nine of which are directed by women.
On average, four of these enterprises source energy from the mini-grids. Of these four enterprises that are
supplied by the grids, one – i.e. one-quarter – is led by women (Table 56). According to information from the
village heads, less than one of these 20 enterprises began its economic activity as a result of the arrival of the
mini-grids.
Table 56 Share of villages by number of enterprises
N Mean Min Max Median Std. dev. t-value
# of enterprises 82 19.683 3.000 70 15 12.954 13.759
# of enterprises
led by women
82 8.768 1.000 50 7 8.351 9.508
# of enterprises
connected
to the mini-grid
82 4.305 0.000 32 3 5.562 7.009
# of enterprises
connected
to the mini-grid
led by women
68 1.088 0.000 7 0 1.751 5.124
# of new enterprises
or economic activities
in the village
thanks to the mini-grid
82 .683 0.000 1 1 .468 13.208
Source: DEval, own table

5. | Mini-grid survey 92
Figure 32 Types of road connecting the villages
Source: DEval, own figure
7% 59%
50%
33%
27%
21%
34%
50%
67%
73%
79%
Tarmac road Laterite Track
Sédhiou
Kolda
Thiès
Kaolack
Fatick
0 10 20 30 40 50 60 70 80 90 100
Another important factor when considering the accessibility of a village are the weather conditions during
the rainy season. Figure 33 shows that in 22 % of the villages in the sample, accessibility remains good during
the rainy season. Accessibility is moderate in 26 % of villages. However, for the majority of the sample
(51 %), accessing the villages during the rainy season is very difficult, though not impossible. One village is
only accessible during this period in the event of an emergency.
5.4 Productive use potential of mini-grids
However, some of the village heads do report that enterprises were created as a result of the arrival of the
mini-grids in their localities. These enterprises include hairdressers, tailors, carpenters, shops, processing
units (mills) and enterprises that sell ice cubes and cool drinks (likely users of freezers). Small enterprises in
terms of small trade and small kiosks were the most important category of businesses created thanks to the
mini-grids.
Assuming that the mini-grids are functional and their energy is being used productively, there is another
important prerequisite for the productive use of energy being able to contribute to economic development.
Producers and enterprises need to be able to market their produce profitably, which is more likely to be
achieved if they export their goods outside their localities.
An important factor in this is the accessibility of villages, which is, among other things, influenced by the type
of road leading into it. Most villages can only be reached via laterite and track roads (Figure 32). It is worth
noting that tracks can include roads that are just sand tracks but which are supposed to eventually be
converted to main roads. Only few villages can be reached via paved roads.

5. | Mini-grid survey 93
Figure 33 Share of villages by level of accessibility during the rainy season
Source: Deval, own figure
Good Moderate Possible with di culties
Only
i
case of emergencies
51%
22%
26%
1%
ff
in
5.5 Sustainability and maintenance of the mini-grids
As previously reported in Figure 27, only nine (11 %) of the mini-grids were functional at the time of the
survey. The average lifespan of the mini-grids was 4.8 years. As Figure 34 shows, 17 % of these villages have
since been electrified by the central grid.
Figure 34 Share of villages entirely or partially electrified by central grid
Source: DEval, own figure
Yes No
83%
17%

5. | Mini-grid survey 94
In terms of maintenance, nearly every mini-grid has an assigned manager. Figure 35 shows that only 4 % of
the villages, i.e. three villages, reported they do not have a manager for their mini-grid. The tasks of the
managers are varied and include recharging the battery fluid (in 89 % of villages), cleaning the panels
(in 89 % of villages), recharging the generator (in 79 % of villages) and switching the control unit on and off
(in 94 % of villages).
Figure 35 Share of villages by tasks of the mini-grid managers
Source: DEval, own figure
444%%%
888999%%%
888999%%%
777999%%%
999444%%%
Share of villages
There is no manager
Recharge battery uid
Clean panels
Recharge generator
Switch the control unit
on/o
0 20 40 60 80 100
Repairs to the mini-grids are carried out periodically, whereby the frequency can differ, as is shown by Figure
36. In 50 % of villages repairs are carried out every month, in 9 % every two months, in 7 % every year and in
5 % every week. Interestingly, 16 % of villages report that their units have never been repaired.
Figure 36 Share of villages by frequency of maintenance of the mini-grid
Source DEval, own figure
Every week Monthly Every two months Every year
Never Other
9%
50%
7%
16%
13% 5%
Along those lines, the survey data also shows that 94 % of the mini-grids are not supplied with fuel by their
operators on a regular basis. These findings are also illustrated in Figure 37.

5. | Mini-grid survey 95
Figure 37 Share of villages supplied with fuel on a regular basis by the mini-grid operator
Source: DEval, own figure
Yes No
94%
6%
Figure 38 shows that when the unit breaks down, most villages (94 %) call the company responsible for the
installation, 18 % repair it themselves and 34 % call someone else to repair it.
Figure 38 Share of villages by repair entity of the mini-grid
Source: DEval, own figure
999444%%%
111888%%%
333444%%%
Share of villages
Mini-grid operator
Village inhabitants
Technician
0 10 20 30 40 50 60 70 80 90 100
In the event of a breakdown, operators are often called to carry out repairs. As Figure 39 shows, for most
villages these repairs take an average of one day (as is the case in 46 % of villages). For 29 % of mini-grids,
the repairs take two or three days. For 7 % of villages, the repairs take a week. For 2 % of villages, a month
was needed to carry out the repairs, and for 6 % it took even longer. However, 9 % of villages also reported
that their units were never repaired.

5. | Mini-grid survey 96
Figure 39 Time it takes the operator to repair the mini-grid
Source: Deval, own figure
Note: Figure does not add up to 100 % due to rounding
1 day 2 - 3 days 1 week 1 month Longer
Operator never repairs the grid
29%
46%
7%
2%
6% 9%

6. | References 97
6. REFERENCES
Abdul-Jalil, S., A. T. Kinkpe, Z. Elnour, H. Grethe, C. S. Aballo, C. C. Acakpo, B. O. K. Gado, F. Hounnou, J.
Luckmann, F. Thom, M. D. Toure, J. Volkholz and F. T. V. Yorou (2023), “Policy Brief: Climate
Change Impacts Across Agroecological Zones on Agriculture in Benin: An Economy-wide Analysis”,
https://www.agrar.hu-berlin.de/de/institut/departments/daoe/ihe/Veroeff/1-policy-brief_cc-
impacts-benin_eng.pdf (accessed 24.07.2024).
Aklin, M., P. Bayer, S.P. Harish, and J. Urpelainen (2017), “Does Basic Energy Access Generate
Socioeconomic Benefits? A Field Experiment with Off-grid Solar Power in India”. Science advances,
3(5), e1602153, doi.org/10.1126/sciadv.1602153.
Akudugu, M. A, E. Guo and S. K. Dadzie (2012), “Adoption of Modern Agricultural Production Technologies
by Farm Households in Ghana: What Factors”, Journal of Biology, Agriculture and Healthcare, 2, 1–
13.
AMDA (2022), Benchmarking Africa’s Minigrid Report 2022, Africa Minigrid Developers Association,
Nairobi.
Angrist, J. D., & J.-S. Pischke (2009), Mostly Harmless Econometrics—An Empiricist’s Companion, Princeton
University Press, New Jersey.
Ankel-Peters, J. and C. M. Schmidt (2023), “Rural Electrification, the Credibility Revolution, and the Limits
of Evidence-based Policy”, USAEE Working Paper Series No. 23-608.
Ankel-Peters, J., G. Bensch, M. Sievert, A. Köngeter and M. Rauschenbach (2024a), “Are Rural Energy
Access Programs Pro-poor Interventions?”, under review.
Ankel-Peters, J., G. Bensch, M. Sievert, K. Moull and M. Rauschenbach (2024b), “Cost-effectiveness of
Rural Energy Access Strategies”, under review.
Ankel-Peters, J., G. Bensch, M. Sievert, A. Warnholz and M. Rauschenbach (2024c), “Rural Energy Access
and Women’s Empowerment”, under review.
Ayana, O. U., J. and Degaga (2022), “Effects of Rural Electrification on Household Welfare: A Meta-
regression Analysis”, International Review of Economics, 69(2), pp.209-261,
doi.org/10.1007/s12232-022-00391-7.
Banerjee, S. G., A. Singh and H. A. Samad (2011), Power and People: The Benefits of Renewable Energy in
Nepal. World Bank Publications, Washington.
Barron, M. and M. Torero (2017), “Household Electrification and Indoor Air Pollution”, Journal of
Environmental Economics and Management, 86, pp.81-92, doi.org/10.1016/j.jeem.2017.07.007.
Barry, M.S. and Creti, A. (2020), “Pay-as-you-go Contracts for Electricity Access: Bridging the “Last Mile”
Gap? A case study in Benin”, Energy economics, 90, 104843, doi.org/10.1016/j.eneco.2020.104843.
Bazilian, M. et al. (2011), “Considering the Energy, Water and Food Nexus: Towards an Integrated
Modelling Approach”, Energy Policy, Vol. 39, No. 12, pp. 7896–7906,
doi.org/10.1016/j.enpol.2011.09.039.
Beltramo, T., G. Blalock, D. I. Levine and A. M. Simons, (2015), “The Effect of Marketing Messages and
Payment Over Time on Willingness to Pay for Fuel-efficient Cookstoves”, Journal of Economic
Behavior & Organization, 118, pp.333-345, doi.org/10.1016/j.jebo.2015.04.025.
Bensch, G. and J. Peters (2015), “The Intensive Margin of Technology Adoption – Experimental Evidence on
Improved Cooking Stoves in Rural Senegal”, Journal of Health Economics, 42, pp. 44–63,
doi.org/10.1016/j.jhealeco.2015.03.006.

6. | References 98
Bensch, G. and J. Peters, J. (2020), “One-Off Subsidies and Long-Run Adoption—Experimental Evidence on
Improved Cooking Stoves in Senegal”. American Journal of Agricultural Economics, 102(1), 72–90,
doi.org/10.1093/ajae/aaz023.
Bensch, G., C. Steinmetz and H. Teklewold (2022), Productive Use of Grid Electricity in Rural Ethiopia.
Current Status and Prospects in Two Case-study Areas. Applied Research Programme on Energy and
Economic Growth, Oxford.
Bensch, G., W. Cornelissen, J. Peters, N. Wagner, J. Reichert and V. Stepanikova (2019), Electrifying Rural
Tanzania. A Grid Extension and Reliability Improvement Intervention, Impact Report commissioned
by the Netherlands Enterprise Agency, Ministry of Foreign Affairs of the Netherlands, The Hague.
Bensch, G., G. Gotz and J. Peters (2020), “Effects of Rural Electrification on Employment: A Comment on
Dinkelman (2011)”. USAEE Working Paper 20-436.
Bensch, G., M. Grimm, M. Huppertz, J. Langbein and J. Peters (2018), “Are Promotion Programs Needed to
Establish Off-grid Solar Energy Markets? Evidence from Rural Burkina Faso”, Renewable and
Sustainable Energy Reviews, 90, pp.1060-1068, doi.org/10.1016/j.rser.2017.11.003.
Bensch, G., M. Jeuland, L. Lenz and O. Ndiaye (2023), “A Bridge to Clean Cooking? The Cost-Effectiveness
of Energy-Efficient Biomass Stoves in Rural Senegal”. Ruhr Economic Papers, No. 1057,
doi.org/10.4419/96973227.
Bensch, G., J. Peters and M. Sievert (2017), “The Lighting Transition in Rural Africa—From Kerosene to
Battery-powered LED and the Emerging Disposal Problem”, Energy for Sustainable Development,
39, pp.13-20, doi.org/10.1016/j.esd.2017.03.004.
Berkouwer, S. B. and J. T. Dean (2022), “Credit, Attention, and Externalities in the Adoption of Energy
Efficient Technologies by Low-income Households”, American Economic Review, 112(10), pp.3291-
3330, doi: 10.1257/aer.20210766.
Bernard, T., and M. Torero (2009), Welfare Impact of Rural Electrification: ‘Short-Term’ Evidence from
Ethiopia: A Report for the Ethiopian Electric Power Corporation and the World Bank, International
Food Policy Research Institute, Washington.
BloombergNEF (2020), State of the Global Mini-grids Market Report 2020, BloombergNEF, New York.
BMZ (2010), Handbuch der bilateralen EZ, Handreichung Nr. HR025, Leitfaden zur Beurteilung
entwicklungspolitischer Ziele Ländliche Entwicklung und Ernährungssicherung, Bundesministerium
für wirtschaftliche Zusammenarbeit und Entwicklung (BMZ), Bonn/Berlin.
Bos, K., D. Chaplin and A. Mamun (2018), “Benefits and Challenges of Expanding Grid Electricity in Africa:
A Review of Rigorous Evidence on Household Impacts in Developing Countries”, Energy for
Sustainable Development, 44, pp.64-77, doi.org/10.1016/j.esd.2018.02.007.
Brodeur, A. et al. (2020), “Methods Matter: P-Hacking and Publication Bias in Causal Analysis in
Economics”, American Economic Review, Vol. 110, No. 11, pp. 3634–3660,
doi.org/10.1257/aer.20190687.
Burney, J., L. Woltering, M. Burke, R. Naylor and D. Pasternak (2010), “Solar-powered Drip Irrigation
Enhances Food Security in the Sudano–Sahel”. Proceedings of the National Academy of Sciences,
107(5), 1848–1853, doi.org/10.1073/pnas.0909678107.
Caliendo, M. and S. Kopeinig (2008), “Some Practical Guidance for the Implementation of Propensity Score
Matching”, Journal of Economic Surveys, 22(1), 31–72, doi.org/10.1111/j.1467-6419.2007.00527.x.
Carrión, D., R. Prah, T. Tawiah, O. Agyei, M. Twumasi, M. Mujtaba, D. Jack and K. P. Asante (2021),
“Enhancing LPG Adoption in Ghana (ELAG): A Trial Testing Policy-Relevant Interventions to Increase
Sustained Use of Clean Fuels”. Sustainability, 13(4), 2213, doi.org/10.3390/su13042213.

6. | References 99
Chaplin, D., A. Mamun, A. Protik, J. Schurrer, D. Vohra, K. Bos, H. Burak, L. Meyer, A. Dumitrescu, C. Ksoll
and T. Cook (2017), Grid Electricity Expansion in Tanzania by MCC: Findings from a Rigorous Impact
Evaluation. Report submitted to the Millennium Challenge Corporation, Mathematica Policy
Research, Washington.
Chokkakula, S. and M. Giordano (2013), “Do Policy and Institutional Factors Explain the Low Levels of
Smallholder Groundwater Use in Sub-Saharan Africa?” Water international, 38(6), pp.790-808,
doi.org/10.1080/02508060.2013.843842.
Closas, A. and E. Rap (2017), “Solar-based Groundwater Pumping for Irrigation: Sustainability, Policies, and
Limitations”, Energy Policy, 104, 33–37, doi.org/10.1016/j.enpol.2017.01.035.
Das, I., T. Klug, P. P. Krishnapriya, V. Plutshack, R. Saparapa, S. Scott, E. Sills, M. Jeuland, N. Kara, and S. K.
Pattanayak (2020), “A Virtuous cycle. Reviewing the Evidence on Women’s Empowerment and
Energy Access, Frameworks, Metrics and Methods”, https://efficiencyforaccess.org/wp-
content/uploads/White-paper-on-gender-and-energy-access-1.pdf (accessed 24.07.2024).
Das, I., T. Klug, P. P. Krishnapriya, V. Plutshack, R. Saparapa, S. Scott, E. Sills, N. Kara, S. K. Pattanayak and
M. Jeuland (2023), “Frameworks, Methods and Evidence Connecting Modern Domestic Energy
Services and Gender Empowerment”, Nature Energy, 8(5), pp.435-449, doi.org/10.1038/s41560-
023-01234-7.
Day, R., G. Walker and N. Simcock (2016), “Conceptualising Energy Use and Energy Poverty Using a
Capabilities Framework”, Energy Policy, 93, pp.255-264, doi.org/10.1016/j.enpol.2016.03.019.
Dinkelman, T. (2011), “The Effects of Rural Electrification on Employment: New Evidence from South
Africa”. American Economic Review, 101(7), pp.3078-3108, doi.org/10.1257/aer.101.7.3078.
Dudwick, N., K. Hull, R. Katayama, F. Shilpi and K. Simler (2011), From Farm to Firm:
Rural-Urban Transition in developing countries, World Bank, Washington,
https://documents.worldbank.org/en/publication/documents-
reports/documentdetail/547481468330918163/From-farm-to-firm-rural-urban-transition-in-
developing-countries (accessed 24.07.2024).
Duthie, M., J. Ankel-Peters, C. Mphasa, and R. Bhat, (2023), “The elusive quest for sustainable off-grid
electrification: New evidence from Indonesia”, USAEE Working Paper Series, No. 23-609.
ESMAP (2020), The State of Access to Modern Energy Cooking Services, Energy Sector Management
Assistance Program (ESMAP), World Bank, Washington.
ESMAP (2022), Mini Grids for Half a Billion People: Market Outlook and Handbook for Decision Makers,
Energy Sector Management Assistance Program (ESMAP), World Bank, Washington.
Fetter, T. R., and F. Usmani (2024), “Fracking, Farmers, and Rural Electrification in India”, Journal of
Development Economics, 170, 103308, doi.org/10.1016/j.jdeveco.2024.103308.
Frölich, M. (2004), “Finite-Sample Properties of Propensity-Score Matching and Weighting Estimators”,
Review of Economics and Statistics, 86(1), 77–90, doi.org/10.1162/003465304323023697.
Garrido, M. M., A. S. Kelley, J. Paris, K. Roza, D. E. Meier, R. S. Morrison and M. D. Aldridge (2014),
“Methods for Constructing and Assessing Propensity Scores”, Health Services Research, 49(5),
1701–1720. doi.org/10.1111/1475-6773.12182.
Gertler, P. J., S. Martinez, P. Premand, L. B. Rawlings and C. M. Vermeersch (2016), Impact Evaluation in
Practice, World Bank Publications, Washington.
Golumbeanu, R. and D. F. Barnes (2013), Connection Charges and Electricity Access in Sub-Saharan Africa,
Policy Research Working Paper 6511, World Bank, Washington.
Grantham, K. (2022), Mapping the Intersection of Women’s Economic Empowerment, Care Work and Clean
Energy, International Development Research Centre, Ottawa.

6. | References 100
Greenwood, J., A. Seshadri and M. Yorukoglu (2005), “Engines of Liberation”, The Review of Economic
Studies, 72(1), pp.109-133, doi.org/10.1111/0034-6527.00326.
Grimm, M. and J. Peters (2016), “Solar Off-Grid Markets in Africa. Recent Dynamics and the Role of
Branded Products”, The Journal of Field Actions, Special Issue 15, pp.160-163.
Grimm, M., and J. Peters, (2015), “Beer, Wood, and Welfare ‒ The Impact of Improved Stove Use Among
Dolo-Beer Breweries”, PLOS ONE, 10(8), e0132603, doi.org/10.1371/journal.pone.0132603.
Grimm, M., L. Lenz, J. Peters, and M. Sievert (2020), “Demand for Off-grid Solar Electricity: Experimental
Evidence from Rwanda”, Journal of the Association of Environmental and Resource Economists,
7(3), pp.417-454.
Grimm, M., A. Munyehirwe, J. Peters and M. Sievert (2017), “A First Step up the Energy Ladder? Low-cost
Solar Kits and Household’s Welfare in Rural Rwanda”, The World Bank Economic Review, 31(3),
pp.631-649, doi.org/10.4419/86788635.
Grimm, M., R. Sparrow, and L. Tasciotti (2015), “Does Electrification Spur the Fertility Transition? Evidence
from Indonesia”, Demography, 52(5), pp.1773-1796, doi.org/10.1007/s13524-015-0420-3.
Groenewoudt, A. C., H. A. Romijn and F. Alkemade (2020), “From Fake Solar to Full Service: An Empirical
Analysis of the Solar Home Systems Market in Uganda”, Energy for Sustainable Development, 58,
pp.100-111, doi.org/10.1016/j.esd.2020.07.004.
Grogan, L. and A. Sadanand (2013), “Rural Electrification and Employment in Poor Countries: Evidence
from Nicaragua”, World Development, 43, pp.252-265, doi.org/10.1016/j.worlddev.2012.09.002.
Grogan, L. (2016), “Household Electrification, Fertility, and Employment: Evidence from Hydroelectric Dam
Construction in Colombia”, Journal of Human Capital, 10(1), pp.109-158, doi.org/10.1086/684580.
Hamburger, D., J. Jaeger, P. Bayer, R. Kennedy, J. Yang and J. Urpelainen (2019), “Shades of Darkness or
Light? A Systematic Review of Geographic Bias in Impact Evaluations of Electricity Access”, Energy
Research & Social Science, 58, 101236, doi.org/10.1016/j.erss.2019.101236.
Heumann, C., M. Schomaker and Shalabh (2016), Introduction to Statistics and Data Analysis: With
Exercises, Solutions and Applications in R (1st Edition), Springer International Publishing,
doi.org/10.1007/978-3-319-46162-5.
IEA and IFC (2023), Scaling Up Private Finance for Clean Energy in Emerging and Developing Economies,
International Energy Agency (IEA) & International Finance Corporation (IFC), Washington,
doi.org/10.1787/054f472d-en.
IEG (2008), The Welfare Impacts of Rural Electrification: A Reassessment of the Costs and Benefits,
Independent Evaluation Group (IEG) Impact Evaluation, World Bank, Washington.
Iversen, V. and R. Palmer-Jones (2019), “All you need is cable TV?” The Journal of Development Studies,
55(5), pp.946-966, doi.org/10.1080/00220388.2018.1506581.
Jack, D. W., K. Ayuurebobi, C. F. Gould, E. Boamah-Kaali, A. G. Lee, M. N. Mujtaba, S. Chillrud, S. Kaali, A.
K. Quinn, S. Gyaase, and F. B. Oppong (2021), “A Cluster Randomised Trial of Cookstove
Interventions to Improve Infant Health in Ghana”. BMJ Global Health, 6(8), p.e005599,
doi.org/10.1136/bmjgh-2021-005599.
Jenkins, K., D. McCauley, R. Heffron, H. Stephan and R. Rehner (2016), “Energy Justice: A Conceptual
Review”, Energy Research & Social Science, 11, 174-182, doi.org/10.1016/j.erss.2015.10.004.
Jensen, R. and E. Oster (2009), “The Power of TV: Cable Television and Women’s Status in India”, The
Quarterly Journal of Economics, 124(3), pp.1057-1094, doi.org/10.1162/qjec.2009.124.3.1057.
Jeuland, M., T. R. Fetter, Y. Li, S. K. Pattanayak, F. Usmani, R. A. Bluffstone, C. Chávez, H. Girardeau, S.
Hassen, P. Jagger, and M. M. Jaime, (2021), “Is Energy the Golden Thread? A Systematic Review of

6. | References 101
the Impacts of Modern and Traditional Energy Use in Low-And Middle-Income Countries”,
Renewable and Sustainable Energy Reviews, 135, 110406, doi.org/10.1016/j.rser.2020.110406.
Jeuland, M., S. Pattanayak and J. Peters (2020), Do Improved Cooking Stoves Inevitably go up in Smoke?
Evidence from India and Senegal, VoxDev, https://voxdev.org/topic/energy-environment/do-
improved-cooking-stoves-inevitably-go-smoke-evidence-india-and-senegal (accessed 24.07.2024).
Jeuland, M., J. S. T. Soo and D. Shindell (2018), “The Need for Policies to Reduce the Costs of Cleaner
Cooking in Low-Income Settings: Implications from Systematic Analysis of Costs and Benefits”,
Energy Policy, 121, pp.275-285, doi.org/10.1016/j.enpol.2018.06.031.
Kakwani, N. and E. M. Pernia (2000), “What is Pro-poor Growth?”, Asian Development Review, 18(01),
pp.1-16.
Kizilcec, V. and P. Parikh (2020), “Solar Home Systems: A Comprehensive Literature Review for Sub-
Saharan Africa”. Energy for Sustainable Development, 58, pp.78-89,
doi.org/10.1016/j.esd.2020.07.010.
Krishnapriya, P.P., M. Chandrasekaran, M. Jeuland, and S. K. Pattanayak (2021), “Do Improved
Cookstoves Save Time and Improve Gender Outcomes? Evidence from Six Developing Countries”,
Energy Economics, 102, 105456, doi.org/10.1016/j.eneco.2021.105456.
La Ferrara, E., A. Chong, and S. Duryea (2012), “Soap Operas and Fertility: Evidence from Brazil”, American
Economic Journal: Applied Economics, 4(4), pp.1-31, doi.org/10.1257/app.4.4.1.
LaFave, D., A. D. Beyene, R. Bluffstone, S. T. Dissanayake, Z. Gebreegziabher, A. Mekonnen and M.
Toman (2021), “Impacts of Improved Biomass Cookstoves on Child and Adult Health: Experimental
Evidence from Rural Ethiopia”, World Development, 140, p. 105332,
doi.org/10.1016/j.worlddev.2020.105332.
Lee, K., E. Miguel, and C. Wolfram (2016), “Appliance Ownership and Aspirations among Electric Grid and
Home Solar Households in Rural Kenya”, American Economic Review, 106(5), pp.89-94,
doi.org/10.3386/w21949.
Lee, K., E. Miguel and C. Wolfram (2020a), “Experimental Evidence on the Economics of Rural
Electrification”, Journal of Political Economy, 128(4), pp. 1523-1565.
Lee, K., E. Miguel and C. Wolfram (2020b), “Does Household Electrification Supercharge Economic
Development?”, Journal of Economic Perspectives, 34(1), pp. 122-144,
doi.org/10.1257/jep.34.1.122.
Lenz, L., G. Bensch, R. Chartier, M. Kane, J. Ankel-Peters and M. Jeuland (2023), “Releasing the Killer from
The Kitchen? Ventilation and Air Pollution from Biomass Cooking”, Development Engineering, 8,
p.100108, doi.org/10.1016/j.deveng.2023.100108.
Lenz, L., A. Munyehirwe, J. Peters and M. Sievert (2017), “Does Large-Scale Infrastructure Investment
Alleviate Poverty? Impacts of Rwanda’s Electricity Access Roll-Out Programme”. World
Development, 89, pp. 88-110, doi.org/10.1016/j.worlddev.2016.08.003.
Lewis, J. (2018), “Infant Health, Women's Fertility, and Rural Electrification in the United States, 1930–
1960”, Journal of Economic History, 78(1), pp. 118-154, doi.org/10.1017/S0022050718000050.
Lighting Global et al. (2022), Off-Grid Solar Market Trends Report 2022: State of the Sector. World Bank,
Washington.
Lukuyu, J. and J. Taneja (2023), “Powering Up in Africa: Electricity Demand Stimulation to Build Inclusive
Economies, Reduce Inequality, and Improve Power System Sustainability”, Current
Sustainable/Renewable Energy Reports, 10(1), pp. 1-11, doi.org/10.1007/s40518-023-00207-4.

6. | References 102
Lukuyu, J., G. Bensch, T. Conlon, A. Patel, V. Modi and J. Taneja (2022), “Diesel GenSat: Using Satellite
Data to Detect Diesel-Powered Irrigation for Guiding Electrification in Ethiopia”, In Proceedings of
the Thirteenth ACM International Conference on Future Energy Systems (pp. 325-337).
Marbach, M. and D. Hangartner (2020), “Profiling Compliers and Noncompliers for Instrumental-Variable
Analysis”, Political Analysis, 28(3), pp. 435-444, doi.org/10.1017/pan.2019.48.
Masselus, L., J. Ankel-Peters, V. Modi, A. Munyehirwe, N. Williams, and M. Sievert (2024), “Ten Years
After: Long Term Adoption of Electricity in Rural Rwanda”, Mimeo.
Mekonnen, A., A. Beyene, R. Bluffstone, Z. Gebreegziabher, P. Martinsson, M. Toman and F. Vieider
(2022), “Do Improved Biomass Cookstoves Reduce Fuelwood Consumption and Carbon Emissions?
Evidence From a Field Experiment in Rural Ethiopia”, Ecological Economics, 198, 107467,
doi.org/10.1016/j.ecolecon.2022.107467.
Meriggi, N. F., E. Bulte and A. M. Mobarak (2021), “Subsidies for Technology Adoption: Experimental
Evidence from Rural Cameroon”, Journal of Development Economics, 153, 102710.
doi.org/10.1016/j.jdeveco.2021.102710.
Moner-Girona, M., G. Kakoulaki, G. Falchetta, D. J. Weiss and N. Taylor (2021), “Achieving Universal
Electrification of Rural Healthcare Facilities in Sub-Saharan Africa with Decentralized Renewable
Energy Technologies”, Joule, 5(10), pp. 2687-2714, doi.org/10.1016/j.joule.2021.09.010.
Mortimer, K., C. B. Ndamala, A. W. Naunje, J. Malava, C. Katundu, W. Weston, D. Havens, D. Pope, N. G.
Bruce, M. Nyirenda and D. Wang (2017), “A Cleaner Burning Biomass-Fuelled Cookstove
Intervention to Prevent Pneumonia in Children under 5 Years Old in Rural Malawi (The Cooking and
Pneumonia Study): A Cluster Randomised Controlled Trial”, The Lancet, 389(10065), pp.167-175,
doi.org/10.1016/S0140-6736(16)32507-7.
Mukoro, V., M. Sharmina and A. Gallego-Schmid (2022), “A Review of Business Models for Access to
Affordable and Clean Energy in Africa: Do They Deliver Social, Economic, and Environmental
Value?”, Energy Research & Social Science, 88, 102530, doi.org/10.1016/j.erss.2022.102530.
Munro, P., G. Van Der Horst and S. Healy (2017), “Energy Justice for All? Rethinking Sustainable
Development Goal 7 Through Struggles over Traditional Energy Practices in Sierra Leone”, Energy
Policy, 105, pp.635-641, doi.org/10.1016/j.enpol.2017.01.038.
Munyehirwe, A., J. Peters, M. Sievert, E. H. Bulte and N. Fiala (2022), “Energy Efficiency and Local
Rebound Effects: Theory and Experimental Evidence from Rwanda”, Ruhr Economic Papers, 934,
doi.org/10.4419/96973094.
Pattanayak, S. K., M. Jeuland, J. J. Lewis, F. Usmani, N. Brooks, V. Bhojvaid, A. Kar, L. Lipinski, L. Morrison,
O. Patange, N. Ramanathan, I. H. Rehman, R. Thadani, M. Vora and V. Ramanathan (2019),
“Experimental Evidence on Promotion of Electric and Improved Biomass Cookstoves”, Proceedings
of the National Academy of Sciences, 116(27), pp.13282–13287,
doi.org/10.1073/pnas.1808827116.
Peters, J. and M. Sievert (2016), “Impacts of Rural Electrification Revisited – the African Context”, Journal
of Development Effectiveness, 8(3), pp.327-345, doi.org/10.4419/86788637.
Peters, J. and C. Vance (2011), “Rural Electrification and Fertility–Evidence from Côte d'Ivoire”, The Journal
of Development Studies, 47(5), pp.753-766, doi.org/10.1080/00220388.2010.527954.
Peters, J., M. Sievert and M. A. Toman (2019), “Rural Electrification Through Mini-Grids: Challenges
Ahead”, Energy Policy, 132, pp.27-31, doi.org/10.4419/86788909.
Peters, J., C. Strupat and C. Vance (2014), “Television and Contraceptive Use–A Weak Signal?”, The Journal
of Development Studies, 50(11), pp. 1538-1549, doi.org/10.1080/00220388.2014.931939.

6. | References 103
Peters, J., C. Vance and M. Harsdorff (2011), “Grid Extension in Rural Benin: Micro-Manufacturers and the
Electrification Trap”, World Development, 39(5), pp.773-783,
doi.org/10.1016/j.worlddev.2010.09.015.
Pope, D., M. Johnson, N. Fleeman, K. Jagoe, R. Duarte, M. Maden, R. Ludolph, N. Bruce, M. Shupler, H.
Adair-Rohani and J. Lewis (2021), “Are Cleaner Cooking Solutions Clean Enough? A Systematic
Review and Meta-Analysis of Particulate and Carbon Monoxide Concentrations and Exposures”,
Environmental Research Letters, 16(8), 083002, doi.org/10.1088/1748-9326/ac13ec.
Pueyo, A. and M. Maestre (2019), “Linking Energy Access, Gender and Poverty: A Review of The Literature
on Productive Uses of Energy”, Energy Research & Social Science, 53, pp.170-181,
doi.org/10.1016/j.erss.2019.02.019.
Puzzolo, E., D. Pope, D. Stanistreet, E. A. Rehfuess and N. G. Bruce (2016), “Clean Fuels for Resource-Poor
Settings: A Systematic Review of Barriers and Enablers to Adoption and Sustained Use”,
Environmental Research, 146, pp.218-234, doi.org/10.1016/j.envres.2016.01.002.
Radley, B. and P. Lehmann-Grube (2022), “Off-Grid Solar Expansion and Economic Development in the
Global South: A Critical Review and Research Agenda”, Energy Research & Social Science, 89,
102673, doi.org/10.1016/j.erss.2022.102673.
Raitzer, D. A., N. Blöndal and J. Sibal (2019), Impact Evaluation of Energy Interventions: A Review of the
Evidence, Asian Development Bank, Manila.
Ravallion, M. and S. Chen (2003), “Measuring Pro-poor Growth”, Economics Letters, 78(1), pp.93-99,
doi.org/10.1016/S0165-1765(02)00205-7.
Rewald, R. (2017), Energy and Women and Girls: Analyzing the Needs, Uses, and Impacts of Energy on
Women and Girls in the Developing World, Oxfam Research Backgrounder series, Oxford.
Rosenbaum, P. R. and D. B. Rubin (1983), “The Central Role of the Propensity Score in Observational
Studies for Causal Effects”, Biometrika, 70(1), 41–55, doi.org/10.2307/2335942.
Rubin, D. B. (2001), “Using Propensity Scores to Help Design Observational Studies: Application to the
Tobacco Litigation”, Health Services and Outcomes Research Methodology, 2(3), 169–188,
doi.org/10.1023/A:1020363010465.
Rupf, G. V., P. A. Bahri, K. De Boer and M. P. McHenry (2015), “Barriers and Opportunities of Biogas
Dissemination in Sub-Saharan Africa and Lessons Learned from Rwanda, Tanzania, China, India, and
Nepal”, Renewable and Sustainable Energy Reviews, 52, pp.468–476,
doi.org/10.1016/j.rser.2015.07.107.
Schmidt, M. and M. Moradi (2023), “Community Effects of Electrification: Evidence from Burkina Faso's
Grid Extension”, doi.org/10.2139/ssrn.4523234.
Smith, M. G., and J. Urpelainen (2016), “Rural Electrification and Groundwater Pumps in India: Evidence
from the 1982–1999 period”, Resource and Energy Economics, 45, 31–45,
doi.org/10.1016/j.reseneeco.2016.05.004.
Sovacool, B. K. and M. H. Dworkin (2014), Global Energy Justice, Cambridge University Press, Cambridge.
Stojanovski, O., M. C. Thurber, F. A. Wolak, G. Muwowo and K. Harrison (2021), “Assessing Opportunities
for Solar Lanterns to Improve Educational Outcomes in Off-Grid Rural Areas: Results from a
Randomized Controlled Trial”, The World Bank Economic Review, 35(4), pp.999-1018,
doi.org/10.1093/wber/lhab002.
Taneja, J. (2018), If You Build It, Will They Consume? Key Challenges for Universal, Reliable, and Low-Cost
Electricity Delivery in Kenya. Center for Global Development, Washington.
Tanner, A. and A. Johnston (2017), “The Impact of Rural Electric Access on Deforestation Rates”, World
Development, 94, pp.174–185, doi.org/10.1016/j.worlddev.2016.12.046.

6. | References 104
Tenenbaum, B., C. Greacen, T. Siyambalapitiya and J. Knuckles (2014), From the Bottom up: How Small
Power Producers and Mini-grids can Deliver Electrification and Renewable Energy in Africa. World
Bank, Washington, doi.org/10.1596/978-1-4648-0093-1.
Terrapon-Pfaff, J. et al. (2018), “Productive Use of Energy – Pathway to Development? Reviewing the
Outcomes and Impacts of Small-Scale Energy Projects in the Global South”, Renewable and
Sustainable Energy Reviews, Vol. 96, pp. 198–209, doi.org/10.1016/j.rser.2018.07.016.
Usmani, F., J. Steele and M. Jeuland (2017), “Can Economic Incentives Enhance Adoption and Use of a
Household Energy Technology? Evidence from a Pilot Study in Cambodia”, Environmental Research
Letters, 12(3), 035009, doi.org/10.1088/1748-9326/aa6008.
Van de Walle, D., M. Ravallion, V. Mendiratta and G. Koolwal (2017), “Long-term Gains from
Electrification in Rural India”, The World Bank Economic Review, 31(2), pp.385-411,
doi.org/10.1093/wber/lhv057.
Wamalwa, F., L. Maqelepo and N. Williams (2023), “Unlocking the Nexus Potential: A Techno-economic
Analysis of Joint Deployment of Minigrids with Smallholder Irrigation”, Energy for Sustainable
Development, 77, 101345, doi.org/10.1016/j.esd.2023.101345.
WHO (2016), Burning Opportunity: Clean Household Energy for Health, Sustainable Development, and
Wellbeing of Women and Children. World Health Organization, Geneva.
Wilhite, H. (2017), Gender Implications of Energy Use and Energy Access. EEG State-of-Knowledge Paper
Series.
World Bank (2020), How Do We Define Cities, Towns, and Rural Areas?, World Bank Blogs,
https://blogs.worldbank.org/en/sustainablecities/how-do-we-define-cities-towns-and-rural-areas
(accessed 27.05.2024).
Xie, H., C. Ringler and M. A. H. Mondal (2021), “Solar or Diesel: A Comparison of Costs for Groundwater-
fed Irrigation in sub-Saharan Africa Under Two Energy Solutions”, Earth's Future, 9(4),
e2020EF001611, doi.org/10.1029/2020EF001611.
Zigah, E., M. Barry and A. Creti (2023), “Are Mini-Grid Projects in Tanzania Financially Sustainable?”
Electricity Access, Decarbonization, and Integration of Renewables, pp.233-26,
doi.org/10.1007/978-3-658-38215-5_10.