3.1. Descriptive Analysis
The results of the endline descriptive analysis for group demographics are summarized in Table 2
. Comparisons of the control and intervention groups of communities were conducted. Since the start of 2016, 15 additional water sources in 11 control communities and 68 (52 by the PSM) in 30 intervention communities had been constructed. Despite this increased access over time, 98% of communities had access to improved water sources in 2016 (Table 2
d), with statistically similar functionality rates. The proportions of surface, groundwater, and piped sources were quite similar in 2019, with the vast majority being communal in nature (Table 2
c). Piped connections on plots or within the household were rare in the district, with only one control community having a distribution system providing water to each dwelling. Public piped systems had only been built in six larger communities at the baseline, mainly within the control group. The mean population was somewhat higher in the intervention group (Table 2
b). This is due to the increased number of mid-sized communities (300–1000 people) within the intervention group, as compared to the control group and district as a whole. The results should be interpreted with this selection bias in mind.
Within household samples, statistically similar proportions (p
> 0.05) of gender, household size, age range, and education were recorded. Generally, about 80% of respondents had a junior high school education or below, and about 69% of respondents that worked specified subsistence or commercial farming as their primary employment (Table S2
). For the national rural wealth index, both the mean score and quintile distributions were statistically similar within a given year (Table 2
f). When using national composite scores, which also include people from urban environments, about 78% of households were considered to be among the poorest 40% in Ghana.
3.2. Logistic Regression Analysis
presents the multinomial logistic regression for communal water sources within the district in 2019. The binomial descriptive statistics for functionality, annual reliability, and water free from fecal coliforms are shown in conjunction with the odds of a positive performance, as compared to other categorical predictor variables. An odds ratio (OR) compares the odds that an outcome will occur with and without exposure to a given variable [50
]. An odds ratio greater than one suggests that a predictor variable produces a riskier outcome, while a value less than one deems the predictor to be more protective than the standard [36
]. Each category (management and payment method) uses a standard variable of comparison, which is assigned a null odds ratio (OR = 1). Community-based management and no tariff payment were used as the standards for their respective categories, as they were the most common historical methods of water provision and service within the district. This analysis excludes unimproved surface sources that are clearly inferior.
Generally, most variables associated with the PSM showed protective odds ratios (OR < 1) with statistical significance. Improved functionality, reliability, and quality were all observed for privately-managed water points (Table 3
b). In addition, individually-managed water sources (Table 3
e) showed an improved functionality, which were predominantly self-supplied unprotected wells or standpipes from a community piped system. Community-based management was common within the district (Table 3
a), which was primarily conducted through water management committees. Elected members managed over 50% of community water sources, with less than 30% collecting regular payments. Nearly 70% had a member with a senior high school education or better, and almost 90% had a female representative. Privately-managed water sources showed an improved performance compared with community-managed sources, while other management models had a statistically similar quality and reliability (Table 3
Regular tariff payments, either monthly (Table 3
j) or pay-to-fetch (Table 3
k), also showed improved odds of a good performance in all categories. When controlling for private water sources, other pay-to-fetch sources still showed improved functionality [OR = 0.37 (0.19–0.70)], reliability [OR = 0.47 (0.24–0.89)], and biological quality [OR = 0.11 (0.06–0.21)]. For more regression analysis involving the age and type of water sources, see Table S3
Ordinal service level scores from 2019 are presented in Figure 3
, which provides a quick summary of the logistic regression analysis of household service level indicators shown in Table 4
. Households within the intervention communities were assigned to either the User or Non-User groups based on reported behaviors, practices, and usage rates of competing water sources within a given community. Households were also assigned to a management system based on their primary drinking source. Users (Figure 3
a) and households managed by the PSM (Figure 3
b) were highly correlated. These groups generally provide a better biological quality and annual reliability; equivalent quantity, collection time, and mean acceptability; and worse affordability, compared to sources managed by the community and the Control group. Non-User and Control groups were also observed to be nearly identical (Figure 3
), implying a null effect on Non-Users in intervention communities.
The results in Table 4
emphasize the generally protective or neutral effect of the intervention effort. Water quality indicators were significantly worse for Control and Non-User households in every parameter except geogenic contamination, which was not an issue in the district as a whole (Table 4
a–e). Annual reliability indicators increased for Users (Table 4
l) in conjunction with the increased continuity of the PSM water sources in Table 3
. Accessibility indicators showed little change for Users with regards to the time spent fetching and household congestion (Table 4
i–k). Although a higher proportion of Users were closer to their primary source (p
< 0.01), this may be more indicative of customer motivations than the company’s influence. Acceptability indicators showed slightly increased scores for Users (Table 4
p–s), with a significant improvement for odor and appearance (p
< 0.05). For all groups, ratings tended to fall in the “Good” (4) range, with each household using water sources they preferred. Individual sources tended to have defined traits, such as a poor taste or improved later, which created a clustering effect at the community level.
However, some trends associated with risk were also observed. Despite significantly higher evidence of residual chlorine in User sources (63%), only 5.7% had concentrations higher than 0.2 ppm (Table 4
d–e). Household water treatment was uncommon for all households (10%), and Users stored water for two to three days, on average. Considering the risk of microbial deterioration during transport to the dwelling, the low chlorine concentration was likely insufficient to maintain stored water quality under these conditions [3
]. Additionally, daily availability scores were significantly lower for Users (Table 4
< 0.01). This indicator assessed the number of hours per day that water could be collected from a given source. About a third of Users claimed they could only access water between 8 and 12 h per day. While many free sources were available at any time, the PSM sources required a vendor to be present for tariff collection. This created specific opening hours for collection each day. Therefore, private water systems were more likely to be available throughout the year, but had daily restrictions on time.
Users (37 L/p/d) collected statistically similar quantities of water as Non-User (38 L/p/d) and Control (39 L/p/d) households (Table 4
f). This amounts to about 1360 L per household per week in the dry season, on average. However, 72% of Users utilized multiple water sources to supplement 47% of their domestic needs (by volume), on average (Table 4
< 0.01). Households within the richest quintiles would use sachet water for drinking and PSM water for cooking and hygiene (cleaning, bathing, and laundry) needs (61% of total volume). Households within the other quintiles tended to supplement their hygiene needs with free alternatives, while using PSM water for drinking and cooking (41%–49% by volume). Further information on collection purposes and seasonal trends can be seen in Figures S1 and S2
With regards to affordability, 61% of Users, 87% of Non-Users, and 78% of Control households felt they could afford to pay for their domestic needs (Table 4
n). When refined by a household’s primary water source, 57% of kiosk, 88% of borehole, 93% of protected well, and 7.3% of sachet users were able to afford their domestic needs. For context, common pay-to-fetch prices for handpumps are half the kiosk price (0.10 vs. 0.20 GHS per 18 L), and sachet water costs 36 times the price per liter (0.20 GHS per 500 mL).
Service level indicators varied more widely between wealth quintiles in the Control group (Figure 4
a) than for Users (Figure 4
b). The biological quality, collection time, annual reliability, and affordability all differed significantly between Control quintiles (p
< 0.01). Within User households, only affordability varied significantly between quintiles (p
< 0.01). Interestingly, this was primarily related to an increase in sachet water usage by richer User households, who were shown to have significantly lower scores. Note that only one sample in the “Poorest” rural quintile was observed in the intervention communities (and sixteen samples in the Control), so they reflect a low sample size.
All of these comparisons with Users reflect the proportion of intervention households that choose to use the PSM water at least weekly (i.e., the penetration rate). While 37% (95% CI [30
]) of intervention households use PSM kiosks weekly, only 28% (95% CI [22
]), use them as a primary source and 9.3% (95% CI [6.1,13.8]), exclusively. This is reflected by the same households that use multiple water sources to meet their domestic water needs (Table 4
g). Penetration rates for PSM handpumps (56% weekly) tend to be higher than for kiosks, either due to their alternative tariff structure or reduced population. Handpumps are typically constructed in communities with less than 300 people, which is associated with less travel distance and water source options. Moreover, a monthly tariff is charged per household at a rate of 2 GHS per month; estimated to be thirty times less than a kiosk per liter. Interestingly, weekly (p
= 0.43), primary (p
= 0.24), and exclusive (p
= 0.50) penetration rates did not have a significant relationship with socioeconomic status, providing evidence against the study hypothesis.