Content uploaded by D. Daniel
Author content
All content in this area was uploaded by D. Daniel on Dec 22, 2022
Content may be subject to copyright.
Geography and Sustainability 4 (2023) 29–38
Contents lists available at ScienceDirect
Geography and Sustainability
journal homepage: www.elsevier.com/locate/geosus
Factors related to the functionality of community-based rural water supply
and sanitation program in Indonesia
D. Daniel
a
, Trimo Pamudji Al Djono
b
, Widya Prihesti Iswarani
c
,
∗
a
Department of Health Behavior, Environment, and Social Medicine, Faculty of Medicine, Public Health and Nursing, Universitas Gadjah Mada, Indonesia
b
Sekolah Tinggi Teknologi Sapta Taruna, Indonesia
c
Environmental Science for Sustainable Energy and Technology (ESSET), School of Life Sciences and Environmental Technology (ATGM/ALST), Avans University of
Applied Sciences, The Netherlands
•High community participation does not
always lead to fully functioning water
systems.
•No payment or tari system leads to no
functioning water system.
•An analysis of cost-ecient investment
for rural water supply is necessary.
•Public tap connections tend to be not
functioning opposing to household con-
nections.
Article history:
Received 3 August 2022
Received in revised form 7 December 2022
Accepted 10 December 2022
Available online 13 December 2022
Keywords:
Rural water supply
PAMSIMAS
Functionality
Indonesia
Bayesian belief networks
Logistic regression
This study used multinomial logistic regression and Bayesian belief networks (BBN) to analyze factors inuenc-
ing the functionality of the community-based rural drinking water supply and sanitation program (PAMSIMAS)
in Indonesia. 28,936 PAMSIMAS projects in 33 provinces in Indonesia were analyzed. The data indicates that
85.4% of the water supply systems were fully functioning, 9.1% were partially functioning, and 5.5% were not
functioning. In the regression analysis, good management is positively associated with functionality and a high
investment per capita is negatively associated with the functionality. The latter suggests the need for comprehen-
sive economic analysis in the feasibility study in scattered housing sites and remote-undeveloped areas. We also
found that high community participation at the beginning of the project was associated with the not functioning
system, while women’s participation was positively associated with the functionality. Furthermore, the household
connection is more likely to be functioning than communal connection. BBN analysis shows if the beneciaries
do not pay for water, the probability of not functioning systems is 20 times higher than systems with fee collec-
tion. Moreover, the combination of strong management, strong nancial status, and household connection rather
than communal connection increases the probability of fully functioning to 98%. Improvement of data collection
is also necessary to monitor the current conditions of all PAMSIMAS systems in Indonesia. This study oers a
country-level perspective for better implementation of the community-based rural water supply and sanitation
program in developing countries.
1. Introduction
Despite progress toward the Sustainable Development Goal 6.1 to
“achieve universal and equitable access to safe and aordable drinking
water for all ”by 2030, it was estimated that 785 million people world-
wide still lack basic water services in their homes in 2017 ( UNICEF and
∗ Corresponding author.
E-mail address: w.prihestiiswarani@avans.nl (W.P. Iswarani) .
WHO, 2019 ). A lack of sucient quantities of water services hampers es-
sential water, sanitation, and hygiene (WASH) practices. This situation
aects human health. A previous study reported that 60% of diarrhea
deaths globally were related to the WASH problem ( Prüss-Ustün et al.,
2019 ). Moreover, the disparity in water access between urban and ru-
ral areas occurs globally, in which rural areas often have lower water
access than urban ( Local Burden of Disease WaSH Collaborators, 2020 ).
There were 61 countries reporting to have rural-urban coverage gaps
of over 20% in 2017 ( UNICEF and WHO, 2019 ). Furthermore, an insuf-
https://doi.org/10.1016/j.geosus.2022.12.002
2666-6839/© 2022 The Authors. Published by Elsevier B.V. and Beijing Normal University Press (Group) Co., LTD. on behalf of Beijing Normal University. This is
an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ )
D. Daniel, T.P. Al Djono and W.P. Iswarani Geography and Sustainability 4 (2023) 29–38
cient amount of water leads to the improper or irregular practice of
WASH-related behaviors, e.g., household water treatment or sanitation
( Daniel et al., 2021a ).
The inequalities between urban and rural areas also occur in Indone-
sia. While 95% of households in urban areas had access to basic water
services, it was only 82% in rural areas ( UNICEF and WHO, 2019 ). To
tackle the problem, the community-based rural water supply and san-
itation program, called “Program Penyediaan Air Minum dan Sanitasi
Berbasis Masyarakat (PAMSIMAS) ”in Bahasa, was established in 2006
to increase the water and sanitation access in peri-urban and rural areas
( Kasri et al., 2017 ). There have been three periods of the PAMSIMAS pro-
gram until now: PAMSIMAS I (2008–2012), PAMSIMAS II (2013–2015),
and PAMSIMAS III (2016–2021). PAMSIMAS has beneted about 21.6
million people in 32 thousand villages throughout Indonesia until May
2021 ( PAMSIMAS, 2021a ). This makes PAMSIMAS one of the biggest
water and sanitation access programs in the world. The PAMSIMAS
project is implemented at the village level and managed by a village
water board committee, called “BPSPAMS ”in Bahasa. The BPSPAMS
consists of several members from the community and is responsible for
planning, operationalizing, and maintaining the water system. System
here means the whole PAMSIMAS infrastructure from the catchment
area, treatment facility (if any), distribution pipes, reservoir, taps, etc.
Discontinue or failure of the water supply system is another chal-
lenge besides the lack of water services ( Lee and Schwab, 2005 ). The
rate of failure was quite high in some study cases in Africa, e.g., 28%
in a case study in Nigeria, 35% in Tanzania ( Cronk and Bartram, 2017 ),
and 21% in Ghana ( Fisher et al., 2015 ). Those studies have investigated
the causes of the water system failure, such as poor management, irreg-
ular payment by the beneciaries, and system age. Scholars argue that
nancial, institutional, environmental, technical, and social aspects of
the water system should be in “good conditions ”to sustain water or
WASH-related services ( Dutch WASH Alliance, 2013 ).
PAMSIMAS projects face also the unsustainability problem, but no
study investigates this topic further. Therefore, this study aims to in-
vestigate factors related to the functionality of PAMSIMAS projects
in Indonesia. Furthermore, to the best of our knowledge, there is
no country-level study that analyses the functionality or sustainabil-
ity of the community-based rural water supply and sanitation pro-
gram, especially in the Asia region. Previous studies focus more
on Africa regions ( Cronk and Bartram, 2017 ; Fisher et al., 2015 ;
Foster, 2013 ). This study, thus, contributes to the analysis to this topic.
We utilized data that is publicly available on the PAMSIMAS web-
site ( https://pamsimas.pu.go.id/data-aplikasi ). A multinomial logistic
regression and Bayesian belief networks (BBN) were mainly used to an-
alyze the data. The ndings were then discussed and could be translated
into practical recommendations to reduce the likelihood of the failure
of the rural water supply project in developing countries.
2. Materials and methods
We used PAMSIMAS data that was updated in October 2020. There
were 28,936 PAMSIMAS projects located in 33 provinces (out of 34) in
Indonesia. We selected variables that are assumed related to the func-
tionality of the PAMSIMAS system. There were 10 variables selected as
independent variables: (1) Investment per capita, (2) management, (3)
woman participation, (4) community participation, (5) proportion of
poor people, (6) tari status, (7) location, (8) supply types, (9) distribu-
tion system, and (10) period of the PAMSIMAS program. These variables
were either assumed or mentioned in previous studies signicantly as-
sociated with the functionality of water supply systems in developing
countries, or included as one of the independent variables in previous
water supply studies. The literature background of those variables can
be found in the supplementary material. The data are submitted by the
BPSPAMS to the PAMSIMAS district facilitator, which is then inputted
into the national database. The submitted data are then processed by
PAMSIMAS sta at the national oce and presented on the PAMSIMAS
website ( www.pamsimas.org ).
2.1. Creation of independent variables
There are four variables related to community participation at the
beginning of the project: (1) the number of people joining socialization,
(2) a plenary session of the community action plan ( “Pleno RKM ”in
Bahasa), (3) triggering activity, and (4) the number of people in the
community board of trustees ( “Pengurus LKM/KKM ”in Bahasa). Social-
ization is the introduction event of the PAMSIMAS project by the PAM-
SIMAS district facilitator to the community. The community is explained
about the purpose of the project, how it works, expected contribution
from the community, etc. While socialization event is more related to the
general concept of PAMSIMAS, the “Pleno RKM ”is the event where the
district facilitator explains in more detail some technical aspects of PAM-
SIMAS, e.g., the water source, design of the system, budget plan, etc. The
triggering activity is a kick-o event of WASH behavioral change activ-
ities, e.g., a stop open defecation campaign. In addition, the members
of the community board of trustees were chosen by the community to
coordinate and monitor the progress of the PAMSIMAS project in the
village.
To allow comparison between villages, we used the percentage of
those four variables compared to the total households of PAMSIMAS
beneciaries in that village. For example, there are 250 households in
village A and 100 people join the socialization. The percentage of vari-
able socialization in this example is 40%. The assumption is that one
representative per household, either the household head or mother, is
sucient to represent that household in the activity. If the percentage
is more than 100%, i.e., because many people join the socialization, the
percentage is made to 100%. We used a principal component analysis
(PCA) to create composite scores, named community participation , from
those four variables. It is important to note that these four activities are
conducted at the beginning of the PAMSIMAS project, which therefore
does not represent the community participation in any activities after
construction.
There are 10 variables related to women’s participation: (1) number
of women in the BPSPAMS and (2) LKM/KKM, (3) number of women
joining socialization, (4) Pleno RKM, (5) triggering, (6) health promo-
tion, (7) construction training, (8) nancial training, (9) health promo-
tion training, and (10) O&M training. In contrast to community par-
ticipation, we used the percentage of those 10 variables compared to
the total participants in that activity to allow comparison between vil-
lages. For example, there are 10 members of BPSPAMS in village A and
5 are women. Then the percentage of the variable number of women in
the BPSPAMS is 50%. We used again PCA to create composite scores,
named women participation , from those 10 variables. All activities, ex-
cept the number of women in the BPSPAMS, are also conducted at the
beginning of the PAMSIMAS project.
There are four variables related to management: (1) the existence of
a list of PAMSIMAS assets, (2) bookkeeping, (3) work plan, and (4) part-
nership plan by the BPSPAMS, in which 1 = “Yes, existed ”and 0 = “No
existed ”. The work plan here means the plan of the community, includ-
ing the BPSPAMS about the strategies to achieve 100% access to wa-
ter supply and sanitation services in that community. Furthermore, the
BPSPAMS or water board is also expected to collaborate with other ac-
tors to make sure that they can achieve 100% access to water supply
and sanitation services. This plan or strategy should be well documented
by the BPSPAMS. A general indicator, i.e., composite scores, of PAMSI-
MAS management was created from those four variables by summing
the values and, thus, the value of the variable management ranged from
0 to 4.
A variable investment per capita (in USD in the PAMSIMAS database)
was measured on a continuous scale. A variable proportion of poor people
was in percentage. The other ve variables were nominal. The variable
tariff status had four categories: no tari, total tari less than O&M costs,
30
D. Daniel, T.P. Al Djono and W.P. Iswarani Geography and Sustainability 4 (2023) 29–38
Fig. 1. Percentages of fully-functioning PAMSIMAS system in 33 provinces in Indonesia. Border colors indicate the zoning in the BBN model, i.e., node location .
total tari bigger than O&M costs, and total tari bigger than recovery
costs. The variable tariff status in the dataset measures whether the col-
lected water fees by beneciaries are less or bigger than O&M costs or
recovery costs of the system. For example, if the village is categorized as
“total tari bigger than recovery costs ”, it means that the total collected
water fees in the period of the lifespan of the infrastructure, i.e., usually
15 years, are bigger than the total estimated O&M costs and recovery
costs in that period.
The proportion of poor people in the PAMSIMAS project or village
is made by the community itself and the village oce. Considering
the heterogeneity of the village conditions in the whole country, there
is no standard denition of poor people in all PAMSIMAS projects or
villages. For example, households who have a TV can be considered
poor in village A, but not in other villages. The identication of poor
people among PAMSIMAS beneciaries is conducted before the project
started, following the tools named “Methodology for Participatory As-
sessments (MPA) and Participatory Hygiene and Sanitation Transforma-
tion (PHAST) (MPA/PHAST) ”( PAMSIMAS, 2021b ).
The PAMSIMAS project aims to bring water closer to the community
by building or constructing the transmission and distribution pipe from
the source to the main distribution pipes or specic point. If the house-
holds want to have a household or private connection to their house,
they need to pay themselves and this cost is not included in the cal-
culation of investment per capita. Therefore, there is no relationship
between types of household or private connections and the investment
per capita.
Additionally, the data of private or household connection in the
PAMSIMAS does not mean that PAMSIMAS build and invest/pay for
the private connection. Rather, the data of household connection indi-
cate that the community or household builds their private connection or
extends the connection on top of the distribution pipes that PAMSIMAS
builds.
33 provinces were divided into six geographical locations: Sumat-
era, Java, Borneo (Kalimantan in Bahasa), Celebes (Sulawesi in Bahasa),
Nusa Tenggara, and Eastern Indonesia ( Fig. 1 ). This variable was named
location . We divided the provinces into six groups because the charac-
teristics of provinces that are located on the same island are relatively
similar in terms of development, e.g., logistic and road access, and en-
vironmental conditions, e.g., topography or soil condition. The variable
supply types was either household or communal connections. There is
a possibility of having a mix of household and communal connections.
However, we consider this mixed connection as the communal connec-
tion because the number of private connections in this mixed system is
below the number of beneciaries who are connected to the communal
connection.
The variable distribution system was either a fully-gravity system or
not a fully-gravity system, i.e., a mix of pump and gravity system. Lastly,
the variable period of the PAMSIMAS program was based on the starting
year of the project or building of the system, and categorized into three:
PAMSIMAS I (2008–2012), PAMSIMAS II (2013–2015), and PAMSIMAS
III (2016–2021).
2.2. Outcome variable
The dependent or outcome variable functionality of PAMSIMAS sys-
tem was categorized into three: not functioning (often mentioned as
“red ” status in the PAMSIMAS document), partially functioning ( “yel-
low ” status), and fully functioning ( “green ” status). The PAMSIMAS sys-
tem is categorized as not functioning if less than 40% of the system is
functioning, partially functioning if 40%–80% of the system is function-
ing, and fully functioning if more than 80% of the system is functioning.
Those denitions of functionality are dened and used by the PAMSI-
MAS program ( Trijunianto, 2016 ).
Functioning here means two things: infrastructure and the number
of beneciaries. For example, if the total public taps in the PAMSIMAS
system are 10, and now there are only 7 taps functioning, it means that
the system is 70% functioning, i.e., partially functioning. Furthermore,
if the original beneciaries are 100 households, and now there are only
60 households that can draw water from the system, it means that the
system is 60% functioning, i.e., partially functioning. So, functioning
status covers these two aspects, even though the number of beneciaries
is often used to assess the level or status of functionality. For example,
even though the system is 85% functioning, this 15% damage causes
25% of the beneciaries cannot draw water from the system, and the
status is partially functioning.
2.3. Data analysis
To analyze the data, we combined the results from both logistic re-
gression and BBN model. Those two approaches are often combined in
the water sector, e.g., in previous water supply system studies in Africa
( Cronk and Bartram, 2017 ; Fisher et al., 2015 ) and also water-related
behavior ( Daniel et al., 2020 ).
The data were imported and analyzed statistically using IBM SPSS
25. Since the data did not meet the assumptions of normality or linear-
ity, we applied nonparametric tests to analyze the potential relation-
ship between two variables (bivariate analyses), e.g., the Chi-squared
( X
2
), the Mann-Whitney test ( U ), and Kruskal–Wallis test ( H ). The Mann-
Whitney test ( U ) was mainly used as a post hoc test after the Kruskal-
Wallis test. Lastly, we conducted a multinomial logistic regression to
31
D. Daniel, T.P. Al Djono and W.P. Iswarani Geography and Sustainability 4 (2023) 29–38
predict the functionality of the PAMSIMAS system using the ten indepen-
dent variables. The category “not functioning ”acts as the reference cat-
egory in the regression, i.e., the categories “partially functioning ”and
“fully functioning ”are compared to the “not functioning ”. The bivariate
analyses were conducted to help interpret the results of the multinomial
logistic regression.
BBN is a directed acyclic graph showing a hypothetical causal re-
lationship between “causal ”variables (called “parent nodes ”in BBN)
and an “aected ”variable (child node) ( Pearl, 1988 ). The strength of a
probabilistic relationship between parents and a child node is indicated
by the probability values in the corresponding conditional probability
tables (CPTs). BBN has some advantages compared to other common sta-
tistical methods, such as regression analysis: BBN oers a visualization
of a complex situation, allows scenario analyses, can simulate causal re-
lationships and interaction between variables, and can combine multiple
data sources. BBN can also deal conveniently with missing data using an
algorithm in the BBN software, in which missing data could be an issue
in the regression analysis, i.e., the need to remove samples with missing
data from the analysis. The expectation-maximization (EM) algorithm in
the BBN can predict the missing data using the remaining data without
removing those samples from the analysis, as in the regression.
Unlike regression, BBN only permits discrete or nominal variables
as model inputs. Therefore, continuous and ordinal variables, i.e., in-
vestment per capita, management , and PCA of community participation
and women participation , were discretized into several categories. For
the variable investment per capita , there were three categories: “≤ $35 ”,
“$36–70 ”, and “> $70 ”. We used $35 as a threshold considering that the
average annual investment per capita for rural water supply in Indonesia
was $35 ( The World Bank, 2015 ).
For the variable management , scores “0–3 ”were categorized as “bad ”
and a score “4 ”as “good ” management. For variables community par-
ticipation and women participation , we discretized the PCA scores into
three categories: low (lowest one-third of scores, e.g., low community
participation), moderate (one-third to two-thirds of the lowest scores,
e.g., moderate community participation), and high (the top 33% of the
scores).
The software Genie 3.0 Academic was used to perform the BBN anal-
yses ( Druzdzel and Sowinski, 1995 ). The software uses the expectation-
maximization (EM) algorithm to estimate probability values in the CPT
from the data ( Do and Batzoglou, 2008 ). The algorithm is also able to
deal with missing data by learning from the available data, which is one
of the benets of this approach. The ten-fold cross-validation was used
in the same software to assess the model’s performance. This validation
method randomly selected 90% of the data for learning the parameters
(CPT values) and 10% for evaluation. The model’s performance was as-
sessed by the percentage to accurately predict the outcome variable of
all samples and the Area Under the Curve (AUC) value of the Receiver
Operating Characteristics (ROC) curve. The AUC value close to one indi-
cates the perfect prediction of the outcome variable (higher sensitivity
and lower false positives) ( Greiner et al., 2000 ). Furthermore, sensitivity
analysis was conducted using an internal function in the same software
to nd the most inuential variable for the outcome node. The predic-
tive inference was also conducted to simulate the eect of a specic
category in all nodes on the outcome node functionality of the PAMSI-
MAS system . Finally, we also looked for a combination of variables, that
give the highest probability of a fully-functioning PAMSIMAS system,
i.e., an optimistic scenario, which then can be translated into a practi-
cal recommendation in the eld.
3. Results
3.1. An overview of the current status
Until October 2020, PAMSIMAS projects have been conducted in
28,936 villages from 33 provinces in Indonesia. 23.6% of them were
conducted in the period of PAMSIMAS I (2008–2012), 18.5% in the
PAMSIMAS II (2013–2015), and about 57.8% were conducted in PAM-
SIMAS III (2016–now). There is an almost equal number of projects
conducted in Sumatera and Java islands, 29.7% and 29.9% of the to-
tal national projects, respectively, followed by 15.3% of projects which
were conducted in Celebes (Sulawesi) island. The smallest proportion
was in eastern Indonesia (the Moluccas and Papua archipelago): 6.4%.
The median beneciaries in a village were 510 people and the median
household was 136 households. The highest median beneciaries in a
village were in Java (655 people) and the lowest was in Eastern Indone-
sia (370 people).
There were 1,590 (5.5%) PAMSIMAS systems that were not func-
tioning, 2,634 (9.1%) which are partially functioning, and 24,711
(85.4%) which are fully functioning. The province with the lowest fully-
functioning system was North Maluku (66%), followed by Central Bor-
neo (69%), while Bali was the highest fully-functioning system (100%),
followed by Lampung (97%) ( Fig. 1 ).
Regarding tari status, 9.2% of the projects do not collect water
fees from the beneciaries. The median and mean of the investment
per capita were $61 and $178 (SD = 462), respectively. The median and
mean proportions of poor people among the beneciaries were 11.5%
and 19.4% (SD = 20.9), respectively.
About 49.2% of the projects use a 100% gravity ow water system,
while 40.2% use a pump, and 10.6% of the projects have no data. There
were 83.9% of the projects with household connections and 16.1% with
communal connections. About 84.7% of the projects were considered to
have a good managerial system, i.e., had a list of assets, bookkeeping,
work plans, and partnership plans.
In terms of community participation, the median and mean propor-
tions of people joining the PAMSIMAS socialization before the project
started compared to the total number of household beneciaries were
28.8% and 39.9% (SD = 31.2), respectively. The median and mean pro-
portions of women in the water board (BPSPAMS) were 33.3% and
32.1% (SD = 14.08), respectively. About 32.5% of the projects had, at
least, 40% women as the BPSPAMS members, as required by the PAM-
SIMAS guidelines ( Jusita et al., 2021 ).
3.2. Assessing potential relationship between two variables
Furthermore, there was a signicant association between PAMSIMAS
system status and geographical location ( X
2
(10) = 1,108.15, p < 0.001).
The geographical location with the highest not functioning system was
Borneo (8.6%), followed by Celebes (8.5%). On the other hand, the high-
est fully-functioning system was Java (94.5%), followed by Sumatera
(85.9%). Additionally, there was a signicant association between the
status and period of the PAMSIMAS project ( X
2
(4) = 774.70, p < 0.001),
in which the older system had a higher probability of being not func-
tioning.
Communal connection tends to be not functioning (31.1%) compared
to household connection (0.6%) ( X
2
(2) = 7,888.55, p < 0.001). A 100%
gravity system had a smaller chance of not functioning (4.0%) compared
to not a 100% gravity system (5.6%) ( X
2
(2) = 40.33, p < 0.001).
We also found that there was a signicant dierence in invest-
ment per capita between all status categories: the median of not, par-
tially, and fully-functioning systems were $60, $55, and $62, respec-
tively ( H (2) = 32.55, p < 0.001). In addition, there was a signi-
cant dierence in investment per capita between geographical locations
( H (5) = 269.05, p < 0.001). The geographical location with the highest
investment per capita was in eastern Indonesia (median = $77), fol-
lowed by Java (median = $67), and Nusa Tenggara was the smallest
(median = $63). Additionally, a no tariff system has a higher signicant
chance of not functioning (52.3%) compared to other categories of tari
status ( X
2
(6) = 14,371.41, p < 0.001).
There was a signicant dierence in community participation level,
i.e., the values were obtained from PCA, between all status categories
( H (2) = 877.52, p < 0.001), in which fully-functioning status had a lower
level of community participation at the beginning of the project. On the
32
D. Daniel, T.P. Al Djono and W.P. Iswarani Geography and Sustainability 4 (2023) 29–38
Table 1
Parameter estimates of the multinomial logistic regression analysis of the status of the PAMSIMAS project.
Partially functioning to not functioning Fully functioning to not functioning
Variable B SE B p value 𝛽95% CI Variable B SE B p value 𝛽95% CI
Intercept -1.551 2.324 0.505 − Intercept -1.556 2.276 0.494 −
Investment per capita -0.001 0.000 0.003 0.999 0.998– 1 Investment per capita -0.001 0.000 0.022 0.999 0.999–1
Management 1.562 0.451 0.001 4.767 1.968 –11.55 Management 1.699 0.436 0.000 5.470 2.326– 12.86
Woman participation -0.027 0.589 0.963 0.973 0.307 –3.08 Woman participation 0.233 0.580 0.688 1.262 0.405–3.93
Community participation 0.609 0.410 0.137 1.839 0.824– 4.11 Community participation -0.084 0.405 0.835 0.919 0.416–2.03
Proportion of poor
people
-0.006 0.020 0.779 0.994 0.957–1.03 Proportion of poor
people
0.012 0.019 0.542 1.012 0.974– 1.05
Tari status ≥ recovery
cost
13.575 3026.497 0.996 7.865E + 05 b Tari status ≥ recovery
cost
15.985 3026.497 0.996 8.752E + 06 b
Tari status ≥ OM costs -0.724 1.364 0.596 0.485 0.033– 7.03 Tari status ≥ OM costs 0.925 1.341 0.490 2.522 0.182– 34.89
Tari status < OM cost -0.411 1.074 0.702 0.663 0.081– 5.44 Tari status < OM cost -0.016 1.050 0.988 0.984 0.126– 7.71
Tarif status = no tari 0
a − − Tarif status = no tari 0
a − −
Location = Sumatera -1.209 1.211 0.318 0.299 0.028– 3.2 Location = Sumatera -0.139 1.200 0.908 0.870 0.083– 9.14
Location = Java 14.470 1561.573 0.993 1.925E + 06 b Location = Java 15.975 1561.573 0.992 8.665E + 06 b
Location = Celebes -0.830 1.220 0.497 0.436 0.04– 4.77 Location = Celebes 0.042 1.208 0.972 1.043 0.098– 11.12
Location = Nusa
Tenggara
15.515 2182.412 0.994 5.472E + 06 b Location = Nusa
Tenggara
16.713 2182.412 0.994 1.813E + 07 b
Location = Eastern
Indonesia
15.350 0.295 0.000 4.639E + 06 2.6E + 06 –
8.3E + 06
Location = Eastern
Indonesia
15.374 0.000 4.752E + 06 4.7E + 06 –
4.7E + 06
Location: Borneo 0
a - Location: Borneo 0
a
Supply
connection = Household
-0.176 1.256 0.888 0.839 0.072–9.82 Supply
connection = Household
0.353 1.246 0.777 1.423 0.124– 16.35
Supply
connection = Communal
0
a − Supply
connection = Communal
0
a − −
100% gravity
system = Yes
0.864 0.758 0.254 2.372 0.537–10.48 100% gravity
system = Yes
0.873 0.749 0.244 2.393 0.552–10.38
100% gravity
system = No
0
a − − − 100% gravity
system = No
0
a − −
Pamsimas 1: 2008–2012 -1.607 1.202 0.181 0.200 0.019–2.11 Pamsimas 1: 2008–2012 -1.625 1.171 0.165 0.197 0.02– 1.95
Pamsimas 2: 2013–2015 14.500 1858.442 0.994 1.983E + 06 b Pamsimas 2: 2013–2015 14.605 1858.442 0.994 2.202E + 06 b
Pamsimas 3: 2016–2020 0
a - - - Pamsimas 3: 2016–2020 0
a - -
a This parameter is a reference and thus no value;
b
Floating point overow occurred while computing this statistic. Its value is therefore set to system missing; B
= Unstandardised regression coecient; SE B = standard error of the coecient; 𝛽= Standardised coecient; Pseudo R
2
Nagelkerke = 0.210, n = 6,090.
other hand, there was a signicant dierence in women’s participation
level, i.e., the values were obtained from PCA, between all status cate-
gories ( H (2) = 134.97, p < 0.001), in which a fully-functioning system
had a higher woman participation level.
3.3. Multinomial logistic regression
The parameter estimates reveal that, compared with the not func-
tioning system, the partially functioning system was signicantly pre-
dicted by a decreased investment per capita ( Table 1 ). A more pos-
itive managerial system distinguished between partially functioning
and not functioning systems. Furthermore, compared with Borneo, i.e.,
the highest not functioning system, Eastern Indonesia had a higher
probability of being partially functioning than not functioning. Invest-
ment per capita and managerial also signicantly distinguished fully-
functioning systems and not functioning systems with the same direc-
tion of association as in the partially functioning. The model explained
21% of variances of the functionality status of PAMSIMAS projects in
Indonesia.
3.4. Bayesian belief networks (BBN)
The complete BBN model is shown in Fig. 2 . The baseline conditions
show that 3% of the systems were not functioning, 10% were partially
functioning, and 87% were fully functioning, which is close to the ac-
tual conditions. The average model accuracy was 89.3%. The accuracy
to predict not functioning, partially functioning, and fully functioning
systems were 80.6%, 1.8%, and 99.2%, respectively. The values of area
under the ROC curve (AUC) for the three statuses were 0.98 (highly ac-
curate, according to Greiner et al. (2000) ), 0.77 (moderately accurate),
and 0.86 (moderately accurate). The average accuracy and ROC indi-
cate that the BBN model can distinguish between the three categories in
node PAMSIMAS status based on the ten nodes or variables.
The sensitivity analysis shows that location was the most sensitive
node in the model, followed by the proportion of poor people and tariff
status ( Fig. 3 ). The eect of updating individual nodes on node PAM-
SIMAS status , i.e., predictive inferences, is shown in Table 2 . We only
contrasted the updated probability of not functioning with fully func-
tioning systems. The predictive inferences show that a node that gave
the highest probability variation of the status not functioning was tar-
iff status , followed by the node supply connection . Those two nodes also
gave the highest probability variation for status fully functioning . The
category no tari had a probability of 23% not functioning compared
to other categories. The communal connection had a probability of 15%
not functioning compared to only 1% in household connection. More-
over, we found that a combination of having household connections,
good management, and having a tari status “more than recovery costs ”
results in the highest probability of being fully functioning, from 87%
in the baseline to 98% ( Fig. 4 ).
4. Discussion
4.1. Models’ performances
The multinomial regression analysis using 10 independent variables
can only explain 21% of the variances of the PAMSIMAS system’s func-
tionality. The regression analysis suers from many missing data re-
sulting in only 6,090 data (21.05%) being included in the regression
analysis, which could be the reason for only 21% of variances being
explained.
The model’s accuracy was better in BBN. However, the accuracy to
predict partially functioning systems was low (1.8%), but very accu-
33
D. Daniel, T.P. Al Djono and W.P. Iswarani Geography and Sustainability 4 (2023) 29–38
Fig. 2. The BBN model of the PAMSIMAS project in Indonesia. The bars in each node show the probability that a node is in a certain state (existing condition).
Fig. 3. Sensitivity analysis of individual nodes on the output node PAMSIMAS status .
rate for not functioning (80.6%) and fully functioning systems (99.2%).
We may reason that, based on the authors’ eld observation, the cat-
egorization of partially functioning systems is often vague in the eld.
There were some cases where BPSPAMS reported their system as par-
tially functioning while it is not functioning to avoid a negative view
of their performance. It is also possible that there are other variables
besides the ones used in the analysis that can explain better the partial
functioning status. However, our analyses were limited by the variables
collected and available in the PAMSIMAS national data. At the same
time, eld visit to monitor all systems is another challenge, considering
the geographical conditions in Indonesia. Having another approach to
conrm or validate the water system functionality, for example, by in-
34
D. Daniel, T.P. Al Djono and W.P. Iswarani Geography and Sustainability 4 (2023) 29–38
Table 2
The predictive inference that measures the eect of each state in each node on the output node PAMSIMAS status contrasting not functioning and
fully functioning .
Variables Categories
a
ΔNot
functioning
b
ΔFully
functioning
b
PAMSIMAS program period 2008–2012 2013–2015 2016–2020 0 0
3 87 3 87 3 87
Investment per capita < 35 USD 36–70 USD > 70 USD 1 2
4 86 4 86 3 88
Location Sumatera Java Borneo Celebes Nusa Tenggara Eastern Indonesia 4 12
3 87 2 92 5 80 4 85 6 82 4 80
Proportion of poor people < 10 % 10%–25% > 25% 7 14
1 92 2 89 8 78
Tari status No tari < O&M cost > O&M cost > Recovery cost 22 37
23 57 2 82 1 94 2 93
Supply connection Household Communal 14 24
1 91 15 67
100% gravity system No Yes 1 1
4 86 3 87
Woman participation Low Moderate High 1 1
4 86 3 87 3 87
Community participation Low Moderate High 1 2
3 87 4 86 4 85
Management Bad Good 5 10
8 78 3 88
a The value under each category corresponding to a node as displayed in the rst column is the updated probability of the output node is
“not functioning ”(rst cell) and “fully functioning ”(second cell) given that all households maintain this state. The baseline probabilities of “not
functioning ”and “fully functioning ” were 3% and 87%, respectively ( Fig. 1 ).
b Δis the dierence between the lowest and highest value of the updated probability of output node, HWT practice being “not functioning ”and
“fully-functioning ”in %.
Fig. 4. The scenario of having the highest fully-functioning PAMSIMAS program.
cluding the analysis of a nancial report of the BPSPAMS to determine
the functionality, is suggested. The national ocer can match the -
nancial report with the functional status of the system, assuming that a
positive nancial balance indicates a fully functioning system. However,
if there is an anomaly in the data, they can easily conrm that with the
district facilitator.
4.2. Results’ discussion and implications
Previous studies that used both logistic regression and BBN do not
include variables related to social or community aspects ( Cronk and Bar-
tram, 2017 ; Fisher et al., 2015 ). In our case, community-related vari-
ables were signicantly associated with the functionality of the PAM-
35
D. Daniel, T.P. Al Djono and W.P. Iswarani Geography and Sustainability 4 (2023) 29–38
Fig, 5. Hypothetical chain relationship be-
tween factors inuencing the functionality of
PAMSIMAS projects.
SIMAS system. We found that high community participation at the be-
ginning of the project was associated with not functioning status, which
contradicts previous studies ( Kasri et al., 2017 ; Wijayanti et al., 2021 ).
The reason for this is not completely clear, but it might be that the eu-
phoria at the beginning of the project, i.e., shown by the number of
people joining the activities, may not be an indicator of the long-term
functionality of the community-based water and sanitation projects. The
euphoria may result from a temporary feeling of having easier access to
the water supply but may not lead to a sense of ownership for the in-
frastructure, which the latter is often mentioned as critical to sustaining
the water supply system ( Kelly et al., 2017 ). However, our argument
still needs further clarication. Furthermore, there is also a possibility
that the variables used do not completely measure the community par-
ticipation, because it only measures the number of participants joining
specic events. For example, a previous study used fund contribution by
beneciaries to the project as one of the variables to measure community
participation ( Marks and Davis, 2012 ). Moreover, another study argues
that community participation is important to sustain the rural water sup-
ply project in Indonesia, but its eect is much lower than other factors,
e.g., tari status ( Al Djono and Daniel, 2022 ). Therefore, our results on
community participation should be interpreted with caution and need
more clarication.
On the other hand, the higher the women’s participation in the
project, the higher the probability of the water system functioning,
which is in line with other studies ( Al-Mahfadi, 2016 ; Yuerlita, 2017 ).
Women and girls are mainly responsible for the water and WASH-
related activities in the house ( UNICEF and WHO, 2019 ). This makes
women able to identify water utilization problems and then provide so-
lutions ( Svahn, 2011 ). However, the requirement of having, at least,
40% women as the BPSPAMS members, was only found in 32.5% of the
PAMSIMAS projects, indicating the need to increase women’s partici-
pation in the future PAMSIMAS projects. The cultural aspect that hin-
ders women’s participation in the community could be another barrier
( Yuerlita, 2017 ). Therefore, we think that the requirement of a mini-
mum of 40% women as the BPSPAMS members is a good initiative and
starting point to enhance women’s empowerment in the water sector in
Indonesia, which should also be followed by other countries.
Another nding is that investment per capita was negatively asso-
ciated with the functionality. A high investment per capita is indicated
by a small number of beneciaries and/or wide coverage area of the
water supply system. This case often occurs in scattered housing sites
and remote-undeveloped areas, mostly in Eastern Indonesia. A previ-
ous study using data from Nepal, Egypt, and Tanzania indicates that
achieving a break-even point within 5 years in rural water supplies re-
quires an increased water price of 5 to 9 times the current price, which
is unrealistic, especially for low-income households ( Otter et al., 2020 ).
We, therefore, suggest that there should be a comprehensive economic
analysis in the feasibility study before the project starts. This analysis
may help the BPSPAMS to allocate the budget to the most cost-ecient
system and warn the BPSPAMS and PAMSIMAS district facilitator that
the project has a high chance of failure, and they need to think strategi-
cally about how to prevent the system failure. In addition, there should
be an alternative investment and construction scheme for scattered
housing sites and remote-undeveloped areas to reduce the investment
costs.
The predictive inference indicates that tari status largely inuences
the functionality of the PAMSIMAS system. Systems without payment
have a 23% probability of not functioning system or 20 times larger
than if there is a payment ( Table 2 ). In PAMSIMAS, the tari system is
decided by the community itself, either without payment or a monthly
payment based on water volume or at-rate taris. The at-rate taris
usually occur in the communal connection system. There is a consensus
among WASH scholars about the importance of payment by the bene-
ciaries for the sustainability of the water system in developing countries
( Cronk and Bartram, 2017 ; Foster, 2013 ; Marks et al., 2018 ). Fund con-
tribution does not only create a sense of ownership towards the water
system but also provides enough funds for the water board (BPSPAMS)
to perform their tasks, e.g., O&M and repairs. Moreover, based on our
data, almost 40% of the communal connections (public tap or public
hydrant) had no payment system, compared to only 3.5% of the house-
hold connection. Based on our experience in the eld, it is relatively
more dicult to implement and collect water fees in the communal con-
nection systems. Therefore, future PAMSIMAS projects should consider
only making the systems that will apply the tari system.
We found that the combination of strong management, strong nan-
cial status, i.e., which is indicated by having a tari status “more than
recovery costs ”, and household connection increases the probability of
fully functioning by 11%. These variables are closely related. In the case
of PAMSIMAS, only 3.5% of the household connections do not collect
water fees from the beneciaries, while 38.9% in the communal connec-
tions. This suggests that having a private-household connection results
in a strong nancial status and then better management performance.
On the other hand, the strong management, which also indicates the
BPSPAMS performance, is potentially related to the high trust of the
beneciaries in the BPSPAMS and then positively inuences the func-
tionality of the system. This “chain reaction or relationship ”is implied
in a previous PAMSIMAS study ( Daniel et al., 2021b ) and is illustrated
in Fig. 5 .
Some geographical locations, i.e., Borneo (Kalimantan) (8.6%) and
Celebes (Sulawesi) (8.5%), are more likely to be not functioning com-
pared to other locations. However, we could not nd a reasonable reason
for this because some characteristics of PAMSIMAS projects in Borneo
(Kalimantan) and Celebes (Sulawesi) are better than Nusa Tenggara, but
Nusa Tenggara had a slightly lower not functioning system (8%). For
example, Nusa Tenggara had a lower level of women participation and
had a higher proportion of poor people than Borneo and Celebes. Future
studies need to investigate this. Moreover, the node location was also the
most inuential node in the BBN. This further suggests that the national
PAMSIMAS facilitator should then give special attention to areas with
high not functioning systems.
Many missing data in the national data center is another challenge
in this analysis. Many BPSPAMS do not update their data regularly. This
may also indicate weak supervision and monitoring by PAMSIMAS facil-
itators at the district or province level. This especially occurs at the be-
ginning of the project, since many data are collected at the beginning of
the project, e.g., the proportion of poor people among beneciaries and
the number of women participating in PAMSIMAS events. The institu-
tional aspect is said by WASH-related stakeholders as a key to sustaining
WASH services in Indonesia ( Daniel et al., 2021c ). Therefore, strength-
ening the PAMSIMAS-related institutions at all levels should be one of
36
D. Daniel, T.P. Al Djono and W.P. Iswarani Geography and Sustainability 4 (2023) 29–38
the priorities in the next period of PAMSIMAS, i.e., after 2021, and one
of them is ensuring that all new PAMSIMAS projects must update the
data completely to the national data center.
4.3. Study limitations
There are some limitations of this study. We rely on secondary data
and cannot control the data input and data management. Furthermore,
many missing data is a major limitation of this study and led to possible
bias in ndings and interpretation, especially in the statistical analysis.
We are restricted by variables that are available in the data and acknowl-
edge that there are other important factors outside the ones we used that
inuence or are associated with the functionality of the PAMSIMAS sys-
tem, e.g., the environmental conditions of the area ( Kohlitz et al., 2020 ),
the distance of the village to the main supply chain of the spare parts
( Walters and Javernick-Will, 2015 ), cash and labor contribution of the
beneciaries to the project ( Marks and Davis, 2012 ), and the existence of
external funds from third parties ( Daniel et al., 2021b ). Further analysis
should include these factors. Furthermore, our analysis only considers
the one-way inuence of independent variables on the functionality of
PAMSIMAS projects. Future studies should consider the reverse causal-
ity in the system, e.g., water fee collected from the beneciaries could
positively inuence the BPSPAMS performance, but on the other hand,
bad BPSPAMS performance could negatively inuence the regular wa-
ter fee, i.e., people do not want to pay because the BPSPAMS misuse
the budget. Furthermore, we did not include the actual time when the
system fails in the analysis. The analysis of this critical time of the PAM-
SIMAS system needs to be investigated further. The PAMSIMAS district
facilitator usually intensively assists BPSPAMS only during the rst year
of the project. By knowing the critical time of the system, it can be the
facilitator to know how long the assistance should be conducted. Finally,
we encourage future research to conrm and validate our ndings. For
example, further study needs to investigate why some geographical loca-
tions are more likely to be functioning compared to other locations, e.g.,
by analysing the social or cultural aspects of the people. Furthermore,
a case study on a smaller level, e.g., districts, may give more insights
into the long-term functionality of the community-based rural drinking
water supply and sanitation programs in developing countries.
5. Conclusions
We analyzed factors inuencing the functionality of the community-
based rural drinking water supply and sanitation program (PAMSIMAS)
in Indonesia using 28,936 PAMSIMAS data. The multinomial logistic re-
gression showed that investment per capita, management, and location
signicantly predict the functionality of PAMSIMAS. The bivariate anal-
yses found that geographical location has a signicant association with
the functionality, the communal connection has a lower chance of be-
ing not functioning compared to household connection, and having no
payment system is associated with a not functioning system. Further-
more, the high women participation was related to functionality, while
the opposite condition for community participation at the beginning