Content uploaded by Daniel M. Kammen
Author content
All content in this area was uploaded by Daniel M. Kammen on Jan 19, 2019
Content may be subject to copyright.
AnAlysis
https://doi.org/10.1038/s41893-018-0204-z
1Department of Mechanical Engineering, Tufts University, Medford, MA, USA. 2Berkeley Institute for Data Science, University of California, Berkeley, CA,
USA. 3Energy and Resources Group, University of California, Berkeley, CA, USA. 4Renewable and Appropriate Energy Laboratory, University of California,
Berkeley, CA, USA. 5California Institute for Energy and Environment, University of California, Berkeley, CA, USA. 6Escuela de Gobierno y Transformación
Pública, Instituto Tecnológico y de Estudios Superiores de Monterrey, San Pedro Garza García, Mexico. 7Goldman School of Public Policy, University of
California, Berkeley, CA, USA. *e-mail: deborah.sunter@tufts.edu; sergioc@berkeley.edu
As prices of solar photovoltaics (PV) continue to decline1,
accelerated adoption of solar PV is expected among utili-
ties, businesses and communities2. In fact, techno-economic
analyses project that PV total annual installed capacity in the United
States will amount to 16 GW within the next 5 years given the attrac-
tive economic value proposition3.
Growth to date can be attributed in part to top-down approaches,
such as enacted public policies and alternative financing mecha-
nisms, that have gradually led to customers understanding the ben-
efits of solar PV4. In a similar vein, bottom-up approaches, such as
the social diffusion effect, have been identified as significant drivers
in catalysing solar PV adoption5. An example of the diffusion effect
takes place when a ‘seed’ customer installs rooftop PV and, by con-
sequence, influences their neighbours to also install solar, creating
an adoption chain within a radius of influence6,7.
However, this expected growth contrasts with current decel-
eration reports by many distributed solar PV companies across
the United States8, despite historically low PV installation prices1.
Studies suggest that this can be explained by multiple factors9,
including a potential saturation of medium-to-high-income cus-
tomers having already adopted rooftop PV3, and in some instances,
a wide disparity in willingness to acquire PV given electric grid price
competitiveness10. Although reports have elucidated the income
distribution of owners11, sample sizes have been limited, and details
on the customer demographics are not reported.
In response, there have been federal and state efforts to encour-
age low-income participation in rooftop PV. The Renew300
Initiative aims to install 300 MW of solar PV (enough to power
50,000 homes) on federally assisted housing in programmes such
as the US Department of Housing and Urban Development’s rental
housing portfolio, US Department of Agriculture’s Office of Rural
Development Multi-Family Housing Programs, and rental hous-
ing supported by the Low-Income Housing Tax Credit12. The US
Department of Housing and Urban Development also broadened
the applicability of Section 108 Community Development Block
Grants to support renewable energy13. Several states have devel-
oped policies to further include low-income individuals. California
has the Solar on Multifamily Affordable Housing Program14 and
New Solar Homes Partnership15. Massachusetts’ Solar Carve-Out
II programme and the Solar Massachusetts Renewable Target pro-
gramme provide tiered benefits based on income16. New York offers
Affordable Solar Initiatives and Affordable Solar Predevelopment
and Technical Assistance13. California, Colorado, New York and
Oregon have incorporated low-income carve-outs into their com-
munity solar policies17. Many states have integrated rooftop solar
into their low-income weatherization assistance programmes13.
Despite the efforts in the United States to encourage participation
from low-income communities, those specifically targeting racial
and ethnic minorities are still missing.
Distributional energy justice considers both the physically
unequal allocation of energy access and associated environment
benefits and burdens, as well as the uneven distribution of their
associated financial and economic responsibilities. In an interna-
tional context, distributional energy justice concerns, such as the
siting of energy infrastructure and access to low-cost energy ser-
vices, have been raised. Large-scale, centralized renewable energy
projects have been documented in some instances to displace popu-
lations or alter ecosystems18–22. On the other side of the spectrum,
policies aimed at increasing small-scale distributed energy access,
such as the ones in Germany through their Energiewende, have
resulted in financial burden on lower-income communities, where
these are reported to have been paying higher relative shares of their
Disparities in rooftop photovoltaics deployment in
the United States by race and ethnicity
DeborahA.Sunter 1,2,3,4*, SergioCastellanos 3,4,5,6* and DanielM.Kammen 3,4,7
The rooftop solar industry in the United States has experienced dramatic growth—roughly 50% per year since 2012, along with
steadily falling prices. Although the opportunities this affords for clean, reliable power are transformative, the benefits might
not accrue to all individuals and communities. Combining the location of existing and potential sites for rooftop photovoltaics
(PV) from Google’s Project Sunroof and demographic information from the American Community Survey, the relative adoption
of rooftop PV is compared across census tracts grouped by racial and ethnic majority. Black- and Hispanic-majority census
tracts show on average significantly less rooftop PV installed. This disparity is often attributed to racial and ethnic differences
in household income and home ownership. In this study, significant racial disparity remains even after we account for these
differences. For the same median household income, black- and Hispanic-majority census tracts have installed less rooftop PV
compared with no majority tracts by 69 and 30%, respectively, while white-majority census tracts have installed 21% more.
When correcting for home ownership, black- and Hispanic-majority census tracts have installed less rooftop PV compared with
no majority tracts by 61 and 45%, respectively, while white-majority census tracts have installed 37% more. The social disper-
sion effect is also considered. This Analysis reveals the racial and ethnic injustice in rooftop solar participation.
NATURE SUSTAINABILITY | VOL 2 | JANUARY 2019 | 71–76 | www.nature.com/natsustain 71
AnAlysis NATUre SUSTAINABIlITy
total income for energy costs20. Similar examples of solar rooftop PV
economic benefits disproportionately advantaging higher-income
communities can be found in several locations around the world20,23.
Furthermore, in instances where societal sectors perceive cli-
mate change threats and recognize the importance of low carbon
approaches in everyday life activities (for example, clean energy
sources, as we posit), the lack of economic resources and prop-
erty ownership have been stated as main contributors for inac-
tion24. These factors therefore constitute an uneven equity scenario
for some segments of the population, commonly only grouped
by income.
The aim of our study is to understand the energy justice land-
scape from a distributional perspective (that is, the distribution of
access to benefits, such as access to lower-cost electricity, income
from feed-in tariffs and avoided costs from tax credits) in small-
scale distributed renewable energy systems by evaluating the
installation of solar rooftop PV. We hypothesize that PV adoption
is not hindered by economic resources nor property ownership
only. To test this hypothesis, we analyse solar rooftop PV deploy-
ment, correcting for both median household income and property
ownership, to elucidate the role of racial and ethnic composi-
tions in detail—a variable that gains relevance in a multi-racial
and multi-ethnic society that aims to aggressively deploy clean
energy technologies.
To gain insight into the disparity in solar rooftop PV adoption,
we combined high-resolution PV rooftop georeferenced maps with
census demographics data. We used information on the existence
and potential of rooftop PV on more than 60 million buildings
across all 50 US states from Google’s Project Sunroof (https://www.
google.com/get/sunroof/data-explorer/) to quantify the relative
rooftop PV deployment. Variations across states, such as available
solar resources25, incentive programmes and policies (http://www.
dsireusa.org/), electricity prices26 and state racial compositions27,
were mitigated by normalizing the rooftop PV adoption by the aver-
age solar adoption of all census tracts in each state. To evaluate the
social demographic characteristics at the census tract level, median
household income and racial composition from the 2009–2013
5-year American Community Survey (ACS)27 were merged with the
Project Sunroof data. Figure 1 shows the geographic coverage of this
analysis. We categorized census tracts as majority and strong major-
ity, corresponding to any census tract in which more than 50 or
75%, respectively, of the population self-identified as the same race
or ethnicity. Tracts where no single racial or ethnic group comprises
more than 50 or 75% of the population are categorized as no major-
ity and no strong majority, respectively. To investigate the role of
race and ethnicity, we used the locally weighted scatterplot smooth-
ing (LOWESS) method to fit local relationships between household
income and home ownership to rooftop PV adoption for each racial
and ethnic majority group.
Of all the challenges in terawatt-scale PV2, a critical and largely
understudied one is that of equity and inclusivity. We posit that
additional demographic variables, such as racial composition, can
provide social insights into adoption patterns for rooftop PV, and
can be used to better target top-down approaches to increase solar
deployment and improve energy justice conditions.
Evaluation of racial bias in rooftop PV installations
Household income. The differences in the fitted LOWESS curves
denote disparity in the deployment of rooftop PV based on racial
composition across different income levels (Fig. 2). Overall, black-
and Hispanic-majority census tracts have deployed less rooftop
solar than the other census tracts in their state (Fig. 2b), and are
disadvantaged on average 69 and 30%, respectively, compared with
no majority tracts (Fig. 2c). In contrast, white-majority census tracts
show an advantage over no majority census tracts with an increase
in rooftop PV adoption of 21% on average. While on average Asian-
majority census tracts show a disadvantage of 2%, it is interesting to
note that low-income Asian-majority census tracts exhibit a relative
disadvantage in rooftop PV adoption, whereas high-income Asian-
majority census tracts show a relative advantage compared with no
majority tracts. Similar results were found for strong majority com-
munities (Supplementary Fig. 1).
The value of one’s income is related to the local cost of living.
Using county-level cost-of-living estimates from the Living Wage
Calculator28, we subtracted the local cost of living from the cen-
sus tract median household income to calculate the local surplus
500 kmN
Fig. 1 | Census tracts analysed in the United States for solar rooftop
adoption, median household income, home ownership and racial
composition. The analysed region (yellow) contains 58% of the national
technical potential for rooftop PV annual energy generation.
State-normalized
solar deployment
Solar deployment relative to
no-majority census tracts
Median household income
(2013 US$)
a
b
No majority
4,000
2,000
0
3.0
2.5
2.0
1.5
1.0
0.5
0
150
100
50
0
–50
–100
–150
25,000
50,000
75,000
100,000
150,000
200,000
250,000
Asian
Black
Hispanic
White
Tracts
c
Fig. 2 | Relationship between household income and rooftop PV
installation by race and ethnicity. a, Histogram of the distribution
of census tracts analysed at intervals of US$5,000. b,c, Rooftop PV
installations relative to the available rooftop PV potential and normalized
by state as a function of the median household income for majority census
tracts in absolute values (b), and normalized relative to the rooftop PV
adoption of no majority census tracts (c). Each colour represents a majority
race or ethnicity in the census tract. Dark continuous curves represent the
results of the LOWESS method applied to all data in each racial and ethnic
majority group. Lighter shading represents the 90% CIs based on 1,000
bootstrap replications of each racial and ethnic majority group. Note that
the x axes are plotted on a base 10 logarithmic scale.
NATURE SUSTAINABILITY | VOL 2 | JANUARY 2019 | 71–76 | www.nature.com/natsustain
72
AnAlysis
NATUre SUSTAINABIlITy
income. The analysis was repeated using the surplus income, and
comparable results were found (Supplementary Fig. 2). While
this analysis cannot address one’s willingness to pay to install
rooftop PV, it provides a proxy for one’s ability to pay to install
rooftop PV.
Home ownership. People who identify as belonging to a racial
or ethnic minority group are disproportionately more likely to
rent their home. In 2016, 58% of black and 54% of Hispanic
household heads rented their home, compared with only 28% of
white household heads29. The split-incentive problem for roof-
top PV occurs in landlord–tenant relationships30. The landlord
accepts the risk and up-front cost of rooftop solar, yet the benefits
of energy cost savings are reaped by the tenants, often hinder-
ing adoption. To determine whether the racial bias seen in Fig. 2
was the result of racial biases in home ownership, we repeated
the analysis with the median household income replaced by the
percentage of renter-occupied households. Figure 3b shows the
expected trend of decreased solar deployment as the percentage
of renter-occupied households increases. However, when we con-
sidered the solar deployment of each racial and ethnic majority
group relative to no majority census tracts, we found uniform
racial bias across all percentages of renter occupancy, except
for Asian-majority census tracts, as seen in Fig. 3c. Once again,
black- and Hispanic-majority census tracts have deployed less
rooftop PV than the other census tracts in their state, and are dis-
advantaged on average 61 and 45%, respectively, compared with
no majority tracts (Fig. 3c). White-majority census tracts show
an average advantage over no majority census tracts of 37% on
average (Fig. 3c).
Social diffusion effect. Communities that lack any rooftop PV
installations (also known as ‘seed’ rooftop PV customers) are
prone to a delayed future solar adoption7. We found that 47% of
black-majority census tracts do not have any existing solar instal-
lations, representing in some cases more than double that for the
corresponding white-, Asian- and Hispanic- majority census tracts
(Fig. 4). The trend was consistent when disaggregated by income
decile for both majority and strong majority black census tracts
(Supplementary Figs. 3 and 4).
After excluding census tracts without existing rooftop PV instal-
lations, we repeated the analysis and found that the rooftop PV
deployment for black-majority census tracts increased substantially
for those tracts with a median household income below the 2013
national median (US$52,250; ref. 27). In fact, the 90% confidence
interval (CI) for the black-majority census tracts shows greater
installation of rooftop PV than the 90% CI for the no majority
communities for median household incomes below the national
average (Fig. 5c). Within a small portion of the household income
range, the 90% CI for the black-majority census tracts shows greater
installation of rooftop PV compared with the 90% CI for the white-
majority census tracts. In contrast, the Hispanic-majority census
tracts showed disparity comparable to that in Fig. 2. Negligible dif-
ference can be seen in the results for the Hispanic-majority cen-
sus tracts regardless of whether the analysis included (Fig. 2b,c)
or excluded (Fig. 5b,c) census tracts without existing rooftop PV
installations. The trend was similar for strong majority census tracts
(Supplementary Fig. 5).
Conclusions
We found racial/ethnic differences in the adoption of rooftop PV,
even after accounting for median household income and household
ownership. When correcting for median household income, major-
ity black, Hispanic and Asian census tracts showed on average sig-
nificantly less rooftop PV installation relative to no majority census
tracts by 69, 30 and 2%, respectively. In contrast, white-majority
census tracts showed on average 21% more rooftop PV deployment
across all income levels compared with no majority census tracts.
When correcting for household ownership, black- and Hispanic-
majority census tracts have installed less rooftop PV compared with
no majority tracts by 61 and 45%, respectively, while white-majority
census tracts have installed 37% more.
State-normalized
solar deployment
Solar deployment relative to
no-majority census tracts
Percentage of households occupied by renters
020406080
100
a
b
No majority
Asian
Black
Hispanic
White
Tracts
c
–150
–100
–50
50
0.2
0.4
0.6
0.8
1.0
0
2,000
4,000
0
Fig. 3 | Relationship between home ownership and rooftop PV installation
by race and ethnicity. a, Histogram of the distribution of census tracts
analysed at intervals of 5%. b,c, Rooftop PV installations relative to the
available rooftop PV potential and normalized by state as a function of
renter-occupied households for majority census tracts in absolute values
(b), and normalized relative to the rooftop PV adoption of no majority
census tracts (c). Each colour represents a majority race or ethnicity in the
census tract. Dark continuous curves represent the results of the LOWESS
method applied to all data in each racial and ethnic majority group. Lighter
shading represents the 90% CIs based on 1,000 bootstrap replications of
each racial and ethnic majority group.
79
83
76
53
02040
Percentage
60 80 10
0
White
Asian
Hispanic
Black
Existing installations
No installations
47
24
17
21
Fig. 4 | Percentages of each census tract with and without existing rooftop
photovoltaic installations. In the census tracts listed, at least 50% of the
population self-identified as a single race or ethnicity.
NATURE SUSTAINABILITY | VOL 2 | JANUARY 2019 | 71–76 | www.nature.com/natsustain 73
AnAlysis NATUre SUSTAINABIlITy
Additionally, black-majority communities suffer from a dis-
proportional lack of initial deployment, or ‘seeding’. In contrast,
Hispanic-majority census tracts have more similar seeding patterns
to white- and Asian-majority census tracts (Fig. 4), yet deploy sig-
nificantly less rooftop PV than those census tracts (Figs. 2 and 5).
Since rooftop PV adoption is significantly influenced by spatial
neighbouring effects, we hypothesize that the Hispanic-majority
census tracts may have been undergoing a delayed seeding process,
presumably resulting in their observed lower state-normalized roof-
top PV deployment levels. Time, social interactions and population
group similarities have been found to be intrinsically related in epi-
demiology studies31, and propagation behaviours from initial ‘seed’
groups could similarly apply to rooftop PV propagation. Ultimately,
extended time-series rooftop PV adoption data could strengthen an
analysis to elucidate the evolution of adoption rates.
In addition, potential low diversity in the renewable energy
workforce in terms of race32 could be hindering proper PV tech-
nology diffusion to black and Hispanic communities. The lack of
racial diversity is particularly pronounced in management and
senior executive positions in solar firms, where in the United States
over 80% of these positions are held by white people33. While this
paper focuses on distributional injustices, the cause for this uneven
deployment might be more complex and point to procedural
(inclusion of citizens in the decision-making process of accessing
energy) injustices, too23.
The root causes of the differences between black- and Hispanic-
majority census tracts (Figs. 2 and 5) are difficult to predict and
fully explain, and can also have social-psychological attributions34
that require further validation. Interestingly, when communities
of colour are initially seeded—or have first-hand access to rooftop
PV technologies—the deployment significantly increases compared
with other racial/ethnic groups for median household income below
the national average. These results suggest that appropriately ‘seed-
ing’ racial and ethnic minority communities may mitigate energy
injustice in rooftop PV adoption.
As the rooftop PV industry grows, and states discuss next steps
for their energy policies35, it is important for this development to
be inclusive to maximize its potential, and provide equal and just
access to the economic benefits of rooftop PV. While the benefits of
rooftop PV vary regionally, examples of these benefits include lower
cost of electricity, tax credits, feed-in tariffs and rebates. Delayed
participation by a community can exacerbate disparity gaps rela-
tive to other communities that may increase with time. While this
paper provides evidence of the already apparent racial disparity in
rooftop PV adoption, without intervention, the disparity gap would
probably increase. Our results highlight a more profound adoption
characteristic that might shift the focus to more specialized govern-
ment interventions and adaptive business models to fully achieve
the national rooftop PV potential. How well we understand and
address the barriers to participation in rooftop PV will determine
whether or not the solar industry can achieve racial inclusivity and
maximize adoption.
Methods
To gain insight into the disparity in solar rooftop PV adoption, we merged the
Project Sunroof data (https://www.google.com/get/sunroof/data-explorer/) and
the 2009–2013 5-year ACS27 by matching census tracts between the two datasets.
We used the highly spatially resolved dataset from Project Sunroof (https://www.
google.com/get/sunroof/data-explorer/), which contains more than 60 million
buildings across all 50 US states and over a range of approximately 4 years starting
in 2012, to quantify the number of buildings with existing rooftop PV systems
relative to the total number of buildings that could support rooftop PV, according
to Project Sunroof’s methodology36, in each census tract. To evaluate the social
demographic characteristics, we used tract-level data on the median household
income and the percentage of the population that self-identifies as: (1) Asian
(no Hispanic origin); (2) black (no Hispanic origin); (3) Hispanic; and (4) white
(no Hispanic origin). Other races and ethnicities included in the 2009–2013 ACS
were excluded from this analysis given their low percentages. While there is both
uncertainty in the reported tract-level values in the 2009–2013 5-year ACS data
and variation within the census tract37, national high-resolution information at
the individual household level is not currently available.
Census tracts where (1) Project Sunroof data do not cover at least 95% of the
buildings, (2) there are invalid data entries or (3) the median annual household
income is below the 2013 poverty threshold of $23,834 for a 4-person household38
were excluded, leading to a total of 34,156 census tracts used (Fig. 1). Project
Sunroof estimates the annual energy generation potential for rooftop PV in
these census tracts to be 829 TWh yr−1 (https://www.google.com/get/sunroof/
data-explorer/). The National Renewable Energy Laboratory estimates the total
nationwide technical potential for rooftop PV to be 1,432 TWh yr−1 (ref. 39).
Therefore, the region considered in this analysis contains 58% of the national
technical potential for rooftop PV.
Census tracts (CTs) were categorized by how well they reach their rooftop PV
potential. The number of buildings with installed PV systems in each census tract
(NExistingRooftopPV) was divided by the total number of buildings in that tract (NCT),
as shown in equation (1):
=
N
N
SolarDeployment(1
)
CT
ExistingRooftopPV
CT
where both the numerator and denominator entries were obtained from the Project
Sunroof dataset (https://www.google.com/get/sunroof/data-explorer/), following
their detection algorithm and criteria to identify appropriate potential rooftop
space for PV deployment36.
Variations across states, such as available solar resources25, incentive
programmes and policies (http://www.dsireusa.org/), electricity prices26 and
state racial compositions27, were mitigated by normalizing the census tract solar
deployment performance by the population (P)-weighted census tract solar
deployment performance average in each state, as shown in equation (2). Hence,
State-normalized
solar deployment
Solar deployment relative to
no-majority census tracts
Median household income
(2013 US$)
a
b
Tracts
No majority
Asian
Black
Hispanic
White
c
4,000
2,000
0
3.0
2.5
2.0
1.5
1.0
0.5
0
150
100
50
0
–50
–100
–150
25,000
50,000
75,000
100,000
150,000
200,000
250,000
Fig. 5 | Relationship between household income and rooftop PV
installation after excluding census tracts without existing rooftop PV
installations by race and ethnicity. a, Histogram of the distribution
of census tracts analysed at intervals of US$5,000. b,c, Rooftop PV
installations relative to the available rooftop PV potential and normalized
by state as a function of the median household income for majority census
tracts with existing rooftop PV in absolute values (b), and normalized
relative to the rooftop adoption of no majority census tracts (c). Each
colour represents a majority race or ethnicity in the census tract. Dark
continuous curves represent the results of the LOWESS method applied to
all data in each racial and ethnic majority group. Lighter shading represents
the 90% CIs based on 1,000 bootstrap replications of each racial and
ethnic majority group. Note that the x axes are plotted on a base 10
logarithmic scale.
NATURE SUSTAINABILITY | VOL 2 | JANUARY 2019 | 71–76 | www.nature.com/natsustain
74
AnAlysis
NATUre SUSTAINABIlITy
any value greater than 1 indicates that the census tract has installed more rooftop
PV relative to the state average installation, and the opposite is the case for values
less than one:
=∑
StateNormalizedSolarDeployment
SolarDeployment
SolarDeployment(2
)
P
P
CT CT
CT CT
State
CT
State
To investigate the role of race and ethnicity, we categorized census tracts as
majority and strong majority, corresponding to any census tract in which more
than 50 or 75%, respectively, of the population self-identified as the same race
or ethnicity. Each census tract was grouped by the race or ethnicity that the
population most self-identified as.
To correct for variations due to income, the median annual household income
was plotted against the state-normalized solar deployment for all majority and
strong majority census tracts. High variability and a large number of outliers
made it difficult to directly observe and compare a relationship between income
and solar adoption. To more easily compare the results across different groups,
we applied the LOWESS (locally-weighted scatterplot smoothing) method to fit
local linear relationships between household income and rooftop PV adoption.
The primary advantage of the LOWESS method is that it does not require a
specification of a global function that would fit all of the data. The LOWESS
method was implemented using the Python package statsmodels40. The smoothing
parameter, f, was varied between 0.2 and 0.8 and then chosen based on the value of
f that minimized the sum of the residuals squared. The selected values of f can be
found in Supplementary Tables 1–3. The bootstrap method was applied with 1,000
bootstrap replications for each racial and ethnic group, to establish 90% CIs of
the LOWESS method41. At increments of US$50 on the median annual household
income, the bootstrap replications in both the 5th and 95th percentile were selected
and plotted.
Variations due to home ownership and the social diffusion effect were analysed
following a similar method. To evaluate the influence of home ownership, we
applied the bootstrapped LOWESS method to the fraction of households occupied
by renters27 plotted against the state-normalized census tract solar deployment
for each racial and ethnic group (Fig. 3). To explore the influence of the social
diffusion effect, the fraction of census tracts with no existing rooftop PV
installations was calculated for each racial and ethnic group, both overall (Fig. 4)
and by income decile (Supplementary Figs. 3 and 4). We repeated the bootstrapped
LOWESS method excluding census tracts without existing rooftop PV installations
(Fig. 5 and Supplementary Fig. 5).
Data availability
The data that support the findings of this study are available from Google Project
Sunroof (https://www.google.com/get/sunroof/data-explorer/) and the 2009–2013
5-year ACS27. The computer codes used for this study are available online at https://
github.com/DeborahSunter/Rooftop-PV-Deployment-Disparities.
Received: 4 April 2018; Accepted: 30 November 2018;
Published online: 10 January 2019
References
1. Fu, R., Feldman, D., Margolis, R., Woodhouse, M. & Ardani, K. US Solar
Photovoltaic System Cost Benchmark: Q1 2017 Technical Report
(National Renewable Energy Laboratory, 2017).
2. Haegel, N. M. et al. Terawatt-scale photovoltaics: trajectories and challenges.
Science 356, 141–143 (2017).
3. Solar Market Insight Report 2017 Q3 Technical Report (Solar Energy
Industries Association, 2017).
4. Feldman, D. & Bolinger, M. On the Path to SunShot: Emerging Opportunities
and Challenges in Financing Solar (National Renewable Energy Laboratory,
2016).
5. Richter, L.-L. Social Eects in the Diusion of Solar Photovoltaic Technology
in the UK Working Paper (Univ. Cambridge, 2013).
6. Bollinger, B. & Gillingham, K. Peer eects in the diusion of solar
photovoltaic panels. Market. Sci. 31, 900–912 (2012).
7. Graziano, M. & Gillingham, K. Spatial patterns of solar photovoltaic system
adoption: the inuence of neighbors and the built environment. J. Econ.
Geog r. 15, 815–839 (2015).
8. Pyper, J. New US residential solar capacity down 17% year-over-year for Q1.
Green Tech Media https://www.greentechmedia.com/articles/read/residential-
solar-capacity-down-17-year-over-year-for-q1#gs.a= WhwhI (2017).
9. Rai, V., Reeves, D. C. & Margolis, R. Overcoming barriers and uncertainties
in the adoption of residential solar PV. Renew. Energy 89, 498–505 (2016).
10. Ghaith, A. F., Epplin, F. M. & Scott Frazier, R. Economics of grid-tied
household solar panel systems versus grid-only electricity. Renew. Sust. Energy
Rev. 76, 407–424 (2017).
11. Kann, S. & Toth, A. How Wealthy are Residential Solar Customers? Household
Income and SolarAdoption in the US (GTM Research, 2017).
12. Renew300 Federal Renewable Energy Target (US Department of Housing and
Urban Development, 2016).
13. Report on Alternative Approaches to Providing Low and Moderate Income
(LMI) Clean Energy Services Technical report (Clean Energy Advisory
Council and LMI Clean Energy Initiatives Working Group, 2017).
14. California Public Utilities Commission. CPUC creates $100 million solar for
multifamily aordable housing program. EE Online https://
electricenergyonline.com/article/energy/category/solar/142/673274/
cpuc-creates-100-million-solar-for-multifamily-aordable-housing-program.
html (2017).
15. Dodson, G., Hutchison, E. & Omoletski, J. New Solar Homes Partnership
Guidebook 11th edn CEC-300-2017-054-ED11CMF (California Energy
Commission, 2017).
16. Baker, C. D., Polito, K. E., Beaton, M. A. & Judson, J. Solar Massachusetts
Renewable Target (SMART) Final Program Design Technical Report
(Massachusetts Department of Energy Resources, 2017).
17. Low- and Moderate-Income Solar Policy Basics (National Renewable Energy
Laboratory, 2017).
18. Cowell, R. Wind power, landscape and strategic, spatial planning—the
construction of ‘acceptable locations’ in Wales. Land Use Policy 27,
222–232 (2010).
19. Gross, C. Community perspectives of wind energy in Australia: the
application of a justice and community fairness framework to increase social
acceptance. Energy Policy 35, 2727–2736 (2007).
20. Jenkins, K., McCauley, D., Heron, R., Stephan, H. & Rehner, R. Energy
justice: a conceptual review. Energy Res. Soc. Sci. 11, 174–182 (2016).
21. Sovacool, B. K., Heron, R. J., McCauley, D. & Goldthau, A. Energy decisions
reframed as justice and ethical concerns. Nat. Energy 1, 16024 (2016).
22. Yenneti, K. & Day, R. Procedural (in)justice in the implementation of solar
energy: the case of Charanaka solar park, Gujarat, India. Energy Policy 86,
664–673 (2015).
23. Simpson, G. & Clion, J. Subsidies for residential solar photovoltaic energy
systems in Western Australia: distributional, procedural and outcome justice.
Renew. Sust. Energy Rev. 65, 262–273 (2016).
24. Catney, P. et al. Big society, little justice? Community renewable energy and
the politics of localism. Local Environ. 19, 715–730 (2014).
25. Sengupta, M. et al. e National Solar Radiation Data Base (NSRDB). Renew.
Sust. Energy Rev. 89, 51–60 (2018).
26. Electric Power Monthly with Data for May 2018 Technical Report
(US Energy Information Administration, July 2018); https://www.eia.gov/
electricity/monthly/
27. 2015 Planning Database (US Census Bureau, 2015); https://www.census.gov/
research/data/planning_database/2015/
28. Glasmeier, A. K. Living wage calculator (Massachusetts Institute of
Technology, 2018); http://livingwage.mit.edu/
29. Cilluo, A., Geiger, A. & Fry, R. More US households are renting than at any
point in 50 years. Pew Research Center http://www.pewresearch.org/
fact-tank/2017/07/19/more-u-s-households-are-renting-than-at-any-point-in-
50-years/ (2017).
30. Inskeep, B., Daniel, K. & Proudlove, A. Solar on Multi-unit Buildings: Policy
and Financing Options to Address Split Incentives Technical Report
(Solar Outreach Partnership, 2015).
31. Colgate, S. A., Stanley, E. A., Hyman, J. M., Layne, S. P. & Qualls, C. Risk
behavior-based model of the cubic growth of acquired immunodeciency
syndrome in the United States. Proc. Natl Acad. Sci. USA 86,
4793–4797 (1989).
32. Taylor, D. E. Green jobs and the potential to diversify the environmental
workforce. Utah Environ. Law Rev. 31, 47–77 (2011).
33. US Solar Industry Diversity Study: Current Trends, Best Practices, and
Recommendations Technical Report (e Solar Foundation, 2017).
34. Taylor, D. E. Blacks and the environment: toward an explanation of the
concern and action gap between blacks and whites. Environ. Behav. 21,
175–205 (1989).
35. CA SB-338 Integrated Resource Plan: Peak Demand (LegiScan, 2017);
https://legiscan.com/CA/bill/SB338/2017
36. Project Sunroof Data Explorer: a Description of Methodology and Inputs
Technical Report (Google Project Sunroof, 2017).
37. Folch, D. C., Arribas-Bel, D., Koschinsky, J. & Spielman, S. E. Spatial
variation in the quality of American Community Survey estimates.
Demography 53, 1535–1554 (2016).
38. Poverty resholds (US Census Bureau, 2013); https://www.census.gov/data/
tables/time-series/demo/income-poverty/historical-poverty-thresholds.html
39. Gagnon, P., Margolis, R., Melius, J., Phillips, C. & Elmore, R. Rooop Solar
Photovoltaic Technical Potential in the United States: a Detailed Assessment
Technical Report (National Renewable Energy Laboratory, 2016).
40. Seabold, S. & Perktold, J. Statsmodels: economic and statistical modeling with
Python. In Proc. 9th Python Sci. Conf. 57–61 (2010).
41. Efron, B. & Tibshirani, R. Bootstrap methods for standard errors, condence
intervals, and other measures of statistical accuracy. Stat. Sci. 1, 54–75 (1986).
NATURE SUSTAINABILITY | VOL 2 | JANUARY 2019 | 71–76 | www.nature.com/natsustain 75
AnAlysis NATUre SUSTAINABIlITy
Acknowledgements
The authors thank Google’s Project Sunroof team for providing valuable data, and
S. Hsiang for insightful discussions. D.A.S. gratefully acknowledges support from
the Energy Efficiency and Renewable Energy Postdoctoral Research Award from the
US Department of Energy and Berkeley Institute for Data Science. S.C. gratefully
acknowledges support from the Berkeley Energy and Climate Institute–Instituto
Tecnológico de Estudios Superiores de Monterrey Energy Fellowship. D.M.K.
acknowledges support from the Karsten Family Foundation and Zaffaroni
Family Foundation.
Author contributions
D.A.S. and S.C. designed and performed the research, analysed the data and wrote
the paper. D.M.K. supervised the research, guided the study and edited the paper.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/
s41893-018-0204-z.
Reprints and permissions information is available at www.nature.com/reprints.
Correspondence and requests for materials should be addressed to D.A.S. or S.C.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
© The Author(s), under exclusive licence to Springer Nature Limited 2019
NATURE SUSTAINABILITY | VOL 2 | JANUARY 2019 | 71–76 | www.nature.com/natsustain
76
A preview of this full-text is provided by Springer Nature.
Content available from Nature Sustainability
This content is subject to copyright. Terms and conditions apply.