|Contribution of People's Participation: Evidence from 121 Rural Water Supply Projects (World Bank, 1995)|
1. The gender issue was also addressed in the analysis, but because of the importance and complexity of the findings, that subject will be examined in depth in a separate paper.
2. All dollar amounts are U.S. dollars.
3. Pearson-product moment correlation measures the strength of linear association two variables and ranges from -1 to +1. A zero correlation between two variables means that they have no linear relationship. A higher, positive number means that one value is likely to have a high value when the other one does, because the two are positively associated. Association does not, of course, imply causation.
4. Multivariate regression is a statistical method of determining the percentage of variance in the dependent variable that is explained by a combination of independent variables, or by any one variable, after controlling for the effects of other independent variables.
5. The variable reduction mostly involved eliminating those variables with few observations or unreliable estimates, or those that were collinear. As seen in chapter 4, inclusion of variables was, if anything, overly generous.
6. To reduce the number of performance variables to be studied, a factor analysis was run for twenty performance outcomes. Overall project effectiveness emerged as the principal factor, accounted for the greatest variance (73 percent), and had the highest factor loading (0.98) of the twenty variables. The relative importance of overall project effectiveness justified its use in this study as the main indicator of performance.
7. The many measures of participation form a cluster, as supported by the high correlations among the measures (0.7-0.85) as well as by a factor analysis performed on nine of the participation variables with fewer than ten missing data. Overall beneficiary participation emerged as the principle factor, which explained 72 percent of the variance. It had a factor loading of 0.97.
8. For cross-tabulation, the cardinal scale of 1 to 7 was converted into low: 1-3; medium: 4-5; and high: 6-7.
9. The chi-square test of independence of the two variables based on the two-way classification in figure 4.2 gives a value of 64, which easily rejects the null hypothesis of independence, since the critical value of the test at the 1 percent significance level is 13.3.
10. Although the coefficient from a linear regression is not directly comparable to a familiar correlation coefficient (rho), the results of a bivariate regression give nearly the same results as simple correlations when the dependent and independent variables have nearly the same variance. In table 4.1 the regression coefficients are reported so that the bivariate and multivariate results are comparable.
11. The value of a t-statistic reveals the likelihood that, given the precision of the estimate, an estimate of the value observed could have resulted by pure chance even if the true value of the coefficient being estimated were zero. The results of hypothesis testing are generally expressed in two ways. A significance level (most commonly 5 percent or 1 percent) and the computed value of the t-statistic can be compared to the "critical value" for that significance level; for instance, for a large number of observations the critical value for the 5 (or 1) percent level of the t-test is 1.96 (2.57). The other use of a t-statistic is to calculate the significance level at which a given t-statistic would reject the hypothesis of a zero coefficient (this is referred to as a p-level). For instance t-statistic values of 2, 3, or 5 are said by the first method to "reject the null hypothesis at the 5 percent level"; the p-level, or the significance level at which the three statistics (for 121 observations) could reject the hypothesis, reveals very different levels of rejection: 0.047, 0.003, and 0.000002.
12. A simple exposition in equations may be helpful. The basic regression is between an outcome, or performance, variable as the dependent variable (OPE), participation (OBP), a set of direct participation determinants (call them Z), and a set of direct/indirect determinants (call them W). The regression is:
The columns of table 4.1 report the partial correlation coefficient b on participation, with various sets of the Z and W variables. The major distinction between the Z and W variables is that variables in W may also affect performance through participation, or participation may affect these factors.
13. The basic result from multivariate econometrics is that the exclusion from a multivariate regression of a variable that is both positively related to the outcome (dependent) variable and an independent variable results in an upward bias on the estimated coefficient of the included variable. Intuitively, the regression attributes (falsely) to the included variable some of the outcome associated with the excluded variable.
14. The variables expressed as percentages are resealed to 1 to 7 points so that the partial correlation coefficients are comparable across variables.
15. The correlation between project effectiveness and overall water-system sustainability was high at 0.88; accordingly, results for sustainability will not be reported separately.
16. The results of the other determinants are much less strong than for project outcomes. Other variables-namely, local social and cultural conduciveness, economic context, and understanding between agencies-were more important than the direct determinants.
17. The question was specifically related to participation; hence the correlation is very high and the variable is not independent.
18. Some experimentation with using the coders' subjective assessments of the coding reliability on a project-by-project basis as weights revealed no significant differences in the results.
19. The basic result is that with random-measurement error, a regression coefficient is attenuated, that is, biased toward zero. In the simple bivariate case, the amount of the bias is the ratio of variance of the measurement error to the total observed variance of the independent variable.
20. Of course, with hindsight it is clear that having one coder code the outputs without knowing the purpose of the data would have produced even stronger protection from a halo effect.
21. It should be noted that this result is also consistent with certain mixes of measurement error (with a downward bias) and halo effect (with an upward bias) that just offset each other.
22. The issue of establishing causation from data is discussed more fully in Isham, Narayan, and Pritchett (1994), who use statistical estimation techniques appropriate to identifying structural relationships. Those econometric results, which are beyond the scope of this paper, fully support a causal relationship.
23. Since the average number of users was based on design criteria, large piped systems received high scores. These may be multicommunity systems with pumping stations or gravity-fed systems. Just as complexity contributes negatively, a negative correlation exists also between overall effectiveness and the largest water systems.
24. The last finding, a significant negative relationship between project complexity and quality of project design, is important. One of the main arguments justifying investment in improved water supply is health improvement. Studies done in the last decade establish that health impacts are maximized when water improvements are combined with sanitation facilities (toilets) and health education. Hence most water projects try to combine these three elements in integrated projects. While some projects have been able to complete all three activities successfully, the fact that in most countries different government ministries have responsibility for these activities makes successful implementation difficult. Given the difficulty in changing the fundamental approach in the sector from a supply to a demand approach, it is important to sequence activities carefully. Only after successful completion of one activity should additional activities be undertaken. In most rural communities, this would mean water first, sanitation later.
25. Based on her in-depth analysis of 480 hand-pumps in Zimbabwe, Cleaver concluded that three-tier maintenance was more myth than reality. She identified four prerequisites for effective community participation in the long-term maintenance of pumps in Zimbabwe:
· A strong felt need for a protected water supply
· Knowledge that government will not provide significant maintenance support
· A strong, well-motivated local leader
· The anticipation of some tangible reward.
26. The Swiss agency Helvetas undertook a twenty-five-year retrospective analysis of its assistance to Cameroon, during which 114 piped-water systems had been built. Findings revealed a very similar pattern. The study found a steady increase in total costs, steady decline in village contributions, and a very poor maintenance record, despite the presence of project-trained caretakers for 95 percent of the water systems.
Maintenance was supposed to be handled by caretakers chosen, employed, and paid by villagers. However, out of 105 caretakers, only 34 were regularly compensated for their work. Fifty-eight received no compensation, and 19 received compensation rarely or irregularly, sometimes in the form of a bottle of beer. Not surprisingly, the lack of compensation and the lack of community interest in routine maintenance when water was still flowing made caretakers quickly lose interest in their work. The villagers, who did not feel any responsibility for a system financed and maintained by outside agents, had no interest in organizing to undertake maintenance.
27. Similarly, studies of nineteen small piped-water systems in Andhra Pradesh, India, illustrated the difficulty in generating and collecting payment for public standposts, but the willingness of communities to pay for higher service levels. Despite regulations against private connections, the study found that 20 percent of the people had connected illegally to the pipe system, which severely hurt both revenue collection by the local councils and water flow at public standposts and at the villages at the very end of the system. The same study postulated that if house connections were allowed, not only would the revenue base of local councils be sounder but the communities themselves would want to take over the systems from the local councils, get involved in management, and raise the revenue to expand the distribution network of the pipe systems (PRED Study Team 1991; and Job and Shastry 1991).
28. In a community water project in Indonesia, a project manager reported that, at the beginning, field workers' diaries reflected no mention of women and poor people. Field workers stayed with the chief and "hung out" with village officials. After being repeatedly queried and told that they were not doing their jobs, field workers gradually started spending more time with women and the poor and reported being less embarrassed at doing so. From these early experiences the project developed simple criteria to monitor the participatory process, including the number of women who showed up, spoke up, or challenged what was said at meetings. Over one year, leadership of the water groups shifted from village leaders to others who had not previously held leadership positions.
29. Greater autonomy is also, not surprisingly, strongly associated with greater local control (see table 53)
30. A training exercise has been developed for use in workshops with task managers and sector and WID (women in development) specialists in the World Bank. It applies a participation matrix to analyze the demand orientation of eight design features of a rural water supply project supported by the Bank. The case study is hypothetical, but it is based on Bank experiences, particularly in Asia. For more detailed information on use of the material, see Workshop Facilitator Notes, "Participatory Design of Demand-Driven Water Projects," Human Resources and Social Development Division, Asia Technical Department; and UNDP-World Bank Water and Sanitation Division, Transportation, Water and Urban Development Department, World Bank, Washington, D.C., 1993.