Targeting performance of community-based development interventions: an econometric analysis of a women-focused and women-managed non-governmental organisation in Rural Pakistan.

Khan, Hidayat Ullah ; Kurosaki, Takashi

This paper investigates whether the community-based development (CBD) approach effectively reaches out to the poor. The CBD approach is expected to improve targeting performance by reducing leakage to the non-poor, elite capture, and programme placement costs. However, the existing literature lacks comprehensive and innovative ways to assess the targeting performance involving women. We thus examine the targeting performance of CBD interventions adopted by a women-focused and women-managed non-governmental organisation (NGO) in north-western Pakistan. The NGO intervenes through female organisations called Community Organisations (COs), which is rather unusual for a male-dominated society like Pakistan. To assess the targeting performance, we employ rich village- and household-level survey data and compare villages with and without COs on the one hand and member and non-member households on the other hand. The comparison is in terms of poverty and vulnerability. The study shows that the NGO, with proactive involvement of women, has been able to successfully target poorer and environmentally vulnerable villages as well as households.

JEL Classification: 01, 02, 013, 015

Keywords: Community-based Development, Targeting Performance, WomenFocused and Women-Managed NGO, Pakistan

1. INTRODUCTION

The approach of community-based development (CBD) is expected to improve targeting and reduce programme costs of poverty reduction policies, besides other positive contributions [Mansuri and Rao (2004)]. (1) Furthermore, the use of local knowledge is expected to bear greater relevance in a situation where credible monetary data for potential use in targeting activities are not available. According to Alatas, et al. (2012), in developing countries--where the majority of potential target group is employed in the informal sector--the availability of verifiable income records is always an issue. Therefore, it is difficult to identify target groups by employing conventional targeting techniques such as means tests. For these reasons, identification through the CBD approach is expected to improve targeting.

However, the absence of institutional support and/or homogeneity within a community may diminish the usefulness of local information. In the absence of local governance institutions, it is difficult to ensure accountability in the course of implementing CBD initiatives in decentralised settings. For instance, according to Conning and Kevane (2002), within-community heterogeneity may result in a variety of perceptions vis-a-vis poverty, and this may adversely impact targeting performance. The situation becomes worse when the perceptions of donors (i.e., governments, non-government organisations (NGOs), multilateral donors, and philanthropists) with regard to poverty differ from those of the local community. These conditions may create an environment conducive to elite capture.

In addition, even when the CBD approach is able to target poorer villages, it may fail in reaching out to the poor households within each village [Mansuri and Rao (2004)], which can be termed as "poor targeting or mistargeting."For instance, the study of Galasso and Ravallion (2005)--whose motivation closely resembles that of this paper--investigates the targeting performance of the Food-for-Education (FFE) Programme in Bangladesh. The targeting mechanism adopted for the programme comprises two stages: selection of the participating communities by the central government and the identification of eligible households by the communities concerned. By employing both household and commurtity-level data, Galasso and Ravallion (2005) show that the larger a programme is, the lower the levels of land inequality and remoteness therein are, the lower the number of shocks is, and also the lower private redistribution of transfers is, the more improved the within-village targeting becomes. Furthermore, the decision-making ability of the community has a strong influence on the programme outcomes and the centre's programme placement did not take into account village attributes that may potentially help in reaching out to the poor.

Given these findings within the literature, this paper attempts to gamer a better understanding of targeting performance for the case of Pakistan. (2) First, we employ village and household-level data that contain an array of geographic, socioeconomic, demographic, and vulnerability-related measures, to analyse the targeting performance. The list of variables therein is more comprehensive than any adopted in the existing literature. Second, some of the parameters--like networking with the local elite and environmental vulnerability--are used here for the first time, to analyse the targeting performance of CBD interventions. In assessing the performance of targeting, we define "good" targeting as the success of an implementing NGO in placing its programmes in poor villages (in terms of lower adult literacy, poor access to basics amenities, higher level of susceptibility to the natural disasters, etc.) and reaching out to the poor households (poorer access to basic civic services and environmental vulnerability). This is because the aim of the NGO is to improve the livelihood of poor and vulnerable households.

The rest of this paper is organised in the following manner. Section 2 describes the study area while Section 3 elaborates the data used in the empirical analysis. Section 4 proposes the empirical strategy, followed by Section5 that shows quantitative results. Section 6 concludes the paper.

2. STUDY AREA AND THE NGO

Pakistan is an under developed country in terms of both economic and human development. As per the United Nations Development Programme [UNDP (2013)], Pakistan is ranked the 146th of 185 countries on Human Development Index. Moreover, the country has very low mean years of schooling, i.e.4.9 years and per-capita gross national income, i.e. USD 2,566 (in purchasing power parity dollars of 2005). Meanwhile, over 60 percent of the Pakistan's population dwells in rural areas. The rural population of the country has generally poor access to basic amenities and is highly vulnerable to various shocks.

Given the public sector's failure to deliver basic public services to the nation--and especially to the rural poor--NGOs have been actively intervening and providing such services. Several of them have adopted CBD approaches since the 1990s. To analyse the targeting performance of such NGOs and success or failure to outreach the rural poor, in 2010, we began a study on an NGO called the Pakistani Hoslamand Khawateen Network (PHKN), which has its headquarters in District Haripur of Khyber Pakhtunkhwa (KP).

PHKN intervenes in areas of microfinance, human resource development (HRD) training, microinfrastructure projects, and the like. In providing these services, PHKN adopts a CBD approach, under which dwellers of a village or rural community are outreached and organised into community-based organisations. In the case of PHKN, such organisations are called "Community Organisations" (COs). Owing to socio-cultural norms, PHKN has separate COs for males and females. On average, the COs have 16-40 members. PHKN is a woman-led and a women-focused NGO. Its current president is a woman, all members of its board of directors are women, almost three-quarters of the COs are managed by women, and most of its activities are focused on women. This characteristic distinguishes PHKN from other NGOs in the region. Such NGOs are rare in the context of the male-dominated society of Pakistan [Khan (2013)].

The formation of a CO involves a number of steps (Khan et al, 201 l). (3) Under the CO formation process, some of the contacted villages may refuse to form a CO in their village(s). Similarly, some of the non-CO villages can eventually become CO villages, although this did not happen frequently after we began our survey in 2010. Once a CO is formed in a village, PHKN's interventions become active and routine. (4)

3. DATA

During September-December 2010, we implemented a benchmark survey comprising three tiers; the three tiers are villages, COs, and households. Khan, et al. (2011) describe the survey in detail. In this paper, we employ village- and household-level data.

The village survey was designed as a census survey to cover all villages that were (potential) target areas of PHKN. We gathered 105 observations of villages, of which 99 are located in District Haripur. COs of PHKN existed in 40 out of 105, all in District Haripur. We call them CO villages. The rest, 65 villages, are called nonCO villages.

Table 1 lists variables taken from the village survey and analysed in this paper. The variables include village population, the occupation-based distribution of the populations of the villages, the literacy rate,5 connectivity with canal-irrigation system, access to amenities, health and education institutions, local-governance institutions (we call them dispute settlement forums, or DSFs below), and susceptibility to shocks including damages due to the July-August 2010 floods.

In the household data, three types of households were randomly chosen: (i) those who have been members of PHKN activities (henceforth referred to as T-group households), (ii) non-member households (henceforth called as Q-group households) living in CO villages, and (iii) households living in non-CO villages (henceforth labelled as C2-group households). The total size of the sample is 583, divided into 249 T-group households, 234 Ci-group households, and 100 [C.sub.2]-group households. (6) The sample represents predominantly rural households living in Haripur District that are potential targets of PHKN.

Table 3 lists variables taken from the household survey and analysed in this paper. The variables include demographic characteristics, education, housing conditions, access to amenities, assets holding, susceptibility to shocks, and social status of the sample households and their networking with the local elite (native and social status, and relationship with local elite). The statistics suggest household-level disparity in education between male and female members, which is a reflection of male domination in the study area. The housing conditions and asset holding reveal that most of the sample households are poor. We consider housing conditions and land ownership exogenous to PHKN's outreach, while livestock ownership and access to amenities as potentially endogenous to PHKN interventions.

4. EMPIRICAL STRATEGY

To assess the targeting performance of the CBD approach, we test the two following hypotheses. First, we test Hi: Whether CO villages are systematically poorer and more vulnerable than non-CO villages. As a statistical test, we employ the null hypothesis that observable characteristics of CO villages and non-CO villages are the same. Second, we test [H.sub.2]: Whether CO members (7-group) are systematically poorer and more vulnerable than non-members ([C.sub.1]-group) within CO villages. As a statistical test, we employ the null hypothesis that observable characteristics of 7-group and [C.sub.1]-group households in CO villages are the same.

To focus on targeting--rather than on impact--throughout this paper, we mainly analyse the predetermined and exogenous factors that reflect the targeting performance of the PHKN, which makes non-CO villages a valid counterfactual.7We conduct both bivariate and multivariate regression analyses to obtain robust results. The reason we conduct regression analyses is that many of variables are correlated so that partial correlation controlling for other variables may be more meaningful. The multiple regressions exactly controls for other variables.

4.1. Inter-Village Comparison

Testing [H.sub.1] is an inter-village targeting analysis. If, in the course of testing H" we find that CO villages are poorer than non-CO villages, say the CO villages have lower adult literacy, access to basic amenities, and higher susceptibility to the natural disasters, etc., we will conclude that the PHKN targets poorer villages. This finding would reflect the net effect of two mechanisms: that the PHKN endogenously approaches poorer villages, and that poorer villages elect themselves in approaching the PHKN.

Hypothesis [H.sub.1] is tested both using village-level characteristics and household-level characteristics. We compare (i) CO villages and non-CO villages, and (ii) households living in CO villages and households living in non-CO villages. Considering PHKN's community mobilisation process described in Section 2, we test [H.sub.1] by altering the definitions of "CO villages" and "non-CO villages." As the results are qualitatively similar, we report only the results based on the default definition in this paper due to the space limit (Khan, 2013). To implement (ii), we compare the weighted sum of 7-and [C.sub.1]-group households (those living in CO villages) and that of [C.sub.2]-group households (those living in non-CO villages). As the sampling probability is different across villages and across the three groups of sample households (7, [C.sub.1], and [C.sub.2]), we employ the weighted average when we use household-level observations to test Hi.

4.2. Intra-Village Comparison

Hypothesis [H.sub.2] is tested using household-level characteristics. It is a comparison between the 7-group (member households in CO villages) and the [C.sub.1]-group (nonmember households in CO villages).In other words, this is an a-village targeting analysis. If we test [H.sub.2] and we find that member households have worse access to amenities and are more vulnerable to natural disasters than non-member households, we infer that the member households are poorer than the non-member households. This would reflect the self-selection of households, as we analyse [H.sub.2] only using households in CO villages.

In the bivariate analysis (the comparison of means between T and [C.sub.1] households), we employ the weighted average to control for the difference in sampling probability. In the multivariate analysis (regression analysis), we also add village fixed effects to the list of explanatory variables, to cleanly identify the difference.

5. EMPIRICAL RESULTS

5.1. Comparison of CO and Non-CO Villages using Village Characteristics

Table 1 shows empirical results comparing CO and non-CO villages using village-level variables in a bivariate way. It reports statistical tests of equality of means.

CO villages are characterised by a literacy rate lower than that of non-CO villages by 8 percentage points. Both village types are similar in their population size. Non-CO villages have a higher level of occupational diversification, which is an indication of their higher standard of living. The two sets of villages are similar in their access to basic amenities like clean drinking water and market access roads, whereas they are noticeably different in accessibility to natural gas, cable TV, and internet. Non-CO villages have better access to the aforementioned amenities, which are generally associated with economically better-off areas. Non-CO villages tend to have more grocery shops called Karyana shops and hence a better village market place.

We find no difference between the two sets of villages in access to formal health facilities, whereas CO villages have better access to informal health services, e.g., trained TBAs, than CO villages. Similarly, the villages are similar in the availability of formal educational facilities, whereas CO villages have better access to informal education facilities, e.g., community-based schools, than non-CO villages. The strong presence of informal institutions and facilities in CO villages suggests minimal presence and/or effectiveness of government at the grass-root level in the study area and PHKN's provision of these informal services.

DSFs provide a basis for local governance. No difference is found between CO and non-CO villages in terms of the presence of a traditional DSF (e.g. jirga)--a characteristic exogenous to PHKN interventions and is evenly spread across all the villages. However, the number of nontraditional DSFs in CO villages is significantly larger than that in non-CO villages; (8) this reflects the strong presence in the CO villages of local-governance institutions essential to the effective use of local information, the presence of accountability, and hence better targeting performance [Mansuri and Rao (2004)].

Regarding the incidence of damage due to the 2010 floods, the damages were higher in CO than in non-CO villages. This suggests that CO villages tend to be more vulnerable to natural disasters.

Table 2 shows the regression results using the dummy for CO villages as the dependent variable and variables analysed in Table 1 as the explanatory variables. As the multivariate analysis is meant to be used solely for descriptive purposes, we employ a linear probability model. (9) Owing to the small sample size and inherent multicollinearity issues, we opt for a reduced-form regression model. (10) In Model 1, we employ as explanatory variables only those time-invariant variables that are clearly determined prior to PHKN interventions, with the objective of analysing only the targeting result. We include some potentially endogenous variables in Models 2-5, but only as robustness checks. The aforementioned endogenous variables are non-traditional DSFs (dsf), availability of CBS (cbsch), and availability of TBAs (tba).

The results of the multivariate analysis agree with those of the bivariate analysis, with varying levels of statistical significance. Once we control for other factors, the literacy rate is no longer associated with the presence of a CO in a village. The pattern of pro-poor targeting persists, regarding the access to natural gas, internet, and grocery shops, and susceptibility to disasters. These results provide slightly weaker evidence than that suggested through the bivariate analysis but the direction of targeting remains robust.

Unexpectedly, the coefficient on market road access (rd length) becomes significantly negative in multivariate regressions. This suggests that CO villages are more likely to beat shorter distances from a major market than non-COvillages, when controlling for other factors. Although this is against our expectation of pro-poor targeting, we interpret this as a reflection of a cost-minimisation strategy on the part of PHKN--especially in the wake of rising transportation costs.

When we add the potentially endogenous variables (dsf, cbsch, and tba) to Models 2-5, positive and significant correlations are derived; this accords with the results of the bivariate analysis. What is important here is that the inclusion of the potentially endogenous variables does not qualitatively alter coefficients on the more predetermined variables. (11)

To summarise the village-level analysis using village characteristics, we found that a village that is closer to a major market, lacks amenities, and is prone to natural disasters is more likely to be targeted by PHKN and hence form a CO. This suggests that the overall targeting by PHKN is pro-poor. The results of both bivariate and multivariate analysis support this.

5.2. Comparison of CO and Non-CO Villages using Household Characteristics

Table 3 shows empirical results comparing households in CO villages and households in non-CO villages using household-level variables in a bivariate way.

The two sets of households are similar in demography, whereas the education level is higher in non-CO villages than in CO villages. We also find a sharp contrast regarding household assets. Except for the livestock assets, the T and [C.sub.1] group households are poorer than those in the [C.sub.1] group in terms of housing conditions (i.e., house flooring and access to drainage) and access to amenities (i.e., gas, internet, and cable TV). The livestock asset level is higher among the T and [C.sub.1] group households than those in the [C.sub.1] group, probably reflecting the PHKN's facilitating role for the poor households to accumulate livestock. Overall, the bivariate analysis shows a tendency that the T and [C.sub.1] group households are poorer than [C.sub.1] group households in various aspects. Moreover, the T and [C.sub.1] group households are highly vulnerable to shocks (e.g., wild boar attacks), compared to the [C.sub.2] group; this result reflects village-level PHKN placement and supports our earlier claim of pro-poor targeting by the PHKN, that is, the PHKN can successfully outreach environmentally vulnerable segments of society. A larger number of the T and [C.sub.1]group households are native, compared to the [C.sub.2]group households; however, among the former, there is a lower proportion of households with higher social status. Both of these characteristics suggest that CO villages are homogenous and the least socially empowered, which once again confirms that PHKN targets the marginalised segments of Pakistani society. We find an interesting difference between the CO village households and non-CO village households, based on their networking with the local elite. The T and [C.sub.1] group has better networking with the local elite than the [C.sub.2]group households. coefficients on most of the explanatory variables bear signs that are similar to the one seen in the bivariate analysis. A significantly small proportion of the T and [C.sub.1] group households use natural gas for cooking, while a significantly larger proportion of the same exhibit radio ownership and usage, compared to the [C.sub.2] group households. The use of radio could be interpreted as the sign of relative poverty. The [C.sub.2] group households have larger land holdings than the T and [C.sub.1]group households. On the other hand, the T and [C.sub.1] group households have stronger networking with the local elite than the C2group households.

To summarise the findings of village-level analysis using household characteristics, we found that villages whose households have poor access to basics amenities (e.g. natural gas), less land assets, and strong networking with the local elite are more likely to be served by PHKN.

5.3. Intra-Village Analysis Comparing Member and Non-Member Households

Within CO villages, what kinds of households are more likely to be a member? To address this issue, the results of bivariate comparison between member and nonmember households in CO villages are reported in Table 5. Mostly, the two groups are highly similar. At the 5 percent level, only two variables show a statistically significant difference: Member households are more likely to be affected by the 2010 floods than non-member households; member households are less connected with the local elite than non-member households. Although significant only at the 10 percent level, member households are more likely to be affected by wild boar attacks than non-member households. In contrast, the two groups of households have similar characteristics in demography, education, and assets. We interpret these patterns as an outcome of self-selection, that is, the households prone to natural disasters and have less network connections, even within the same village, are more likely to join a CO. These findings thus support the pro-poor targeting of PHKN interventions within CO villages.

Table 6 shows multiple regression results to predict the probability for a household living in a CO village to participate in a CO. We regress a dummy that represents the T group households on a set of household-level variables from Table 5, as well as all village dummies as explanatory variables, for a subsample of CO villages. The results confirm that the two groups are highly similar. There are two variables whose coefficients are statistically different from zero at the 1 percent level: A significantly smaller proportion of the T group households use natural gas for cooking; the T group households have better access to cable TV than the [C.sub.1] group households. Although the sign is the same, these two variables were associated with insignificant differences in the bivariate analysis. The negative correlation with the gas access is a sign of pro-poor targeting. On the other hand, we interpret the positive correlation with cable TV as more aware and socially sensitised are more likely to become CO members owing to their access to independent and vibrant electronic media on cable TV than the state-run terrestrial TV network. The higher probability for households prone to natural disasters to be a member is confirmed from the regression analysis as well, statistically significant at the 5 percent level.

To summarise the findings of household-level analysis within CO villages, we found that member households and non-member households are somewhat similar in their characteristics. If something, the tendency for the poor and less-connected to become a member was found. Regarding vulnerability to natural disasters, we found that more vulnerable households were more likely to join a CO.

6. CONCLUSION

In this paper, we quantitatively investigated the targeting performance of the CBD approach using detailed primary data at the village and household levels. The village-level data was collected through a census survey, whereas the household-level data was collected from a random sampling survey that covered both member and non-member households of a woman-led and women-focused NGO in rural Pakistan.

We found that villages whose households are poorer in terms of access to amenities and more susceptible to natural disasters are more likely to have a CO of the NGO. The correlation involving the networking showed an interesting contrast: Villages where networking with the local elite is strong are more likely to form a CO, while within such villages with a CO, households whose networking with the local elite is weak are more likely to become a member. In contrast to the sharp contrast between CO villages and non-CO villages, the difference between member and non-member households within CO villages was not highly significant. In other words, the NGO's pro-poor targeting functioned well at the selection of recipient villages, whereas we found no evidence of anti-poor targeting within CO villages.

To conclude, the women-focused NGO has been able to target villages and households that are poor and vulnerable to natural disasters. The results suggest that the CBD approach through woman-led and women-focused NGOs is able to improve targeting performance of a poverty reduction policy. The higher likelihood of more socially endowed villages joining the NGO may raise concerns about potential elite capture. The results for within-village analysis presented here and our preliminary analysis using the same dataset and later rounds of primary data (see Chap. 4, Khan, 2013, for details) do not support these concerns, however.

In the current paper, we were not able to separately identify the endogenous placement effect and the self-selection effect. In future research, we intend to overcome this shortcoming by having further rounds of surveys and through collection of recall data.

REFERENCES

Alatas, V., A. Banerjee, R. Hanna, B. A. Olken, and J. Tobias (2012) Targeting the Poor: Evidence from a Field Experiment in Indonesia .American Economic Review 102:4, 1206-1240.

Conning, J. and M. Kevane (2002) Community-based Targeting Mechanisms for Social Safety Nets: a Critical Review. World Development 30:3, 375-394.

Galasso, E. and M. Ravallion (2005) Decentralised Targeting of an Antipoverty Programme. Journal of Public Economics 89, 705-727.

Khan, H. U. (2013) Economic Analysis of Community-based Development Interventions in Rural Pakistan. Graduate School of Economics, Hitotsubashi University, March 2013. (Unpublished Ph.D. dissertation).

Khan, H. U., T. Kurosaki, and K. Miura (2011) The Effectiveness of Community-Based Development in Poverty Reduction: A Descriptive Analysis of a Women-Managed NGO in Rural Pakistan. Centre for Economic Institutions (CEI). Hitotsubashi University, Tokyo. (Discussion Paper No.2011-4). http://cei.ier.hit-u.ac.jp/English/ index.html.

Khawaja, A. I. (2009) Can Good Projects Succeed in Bad Communities? Journal of Public Economics 93:7-8, 899-916.

Khwaja, A. I. (2004) Is Increasing Community Participation Always a Good Thing? Journal of the European Economic Association 2:2-3, 427--36.

Kurosaki, T. (2005) Determinants of Collective Action under Devolution Initiatives: The Case of Citizen Community Boards in Pakistan. The Pakistan Development Review 44:3, 253-270.

Kurosaki, T. and H. U. Khan (2012) Vulnerability of Microfinance to Strategic Default and Covariate Shocks: Evidence from Pakistan. The Developing Economies 50:2, 81-115.

Mansuri, G. and V. Rao (2004) Community-based and Driven Development: A Critical Review. World Bank Research Observer 19:1, 1-39.

UNDP (2013) Human Development Report 2013: The Rise of the South--Human Progress in a Diverse World. New York: UNDP.

Hidayat Ullah Khan <masmaleo@yahoo.com> is affiliated with the Department of Economics, Kohat University of Science and Technology, Kohat. Takashi Kurosaki <kurosaki@ier.hit-u.ac.jp> is affiliated with the Institute of Economic Research, Hitotsubashi University, Tokyo, Japan.

Author's Note: The authors are grateful to Xavier Gine, Katsuo Kogure, Yutaka Arimoto, and participants of the 30th Conference of the Pakistan Society of Development Economists (PSDE), the P'National Conference on Emerging Trends and Challenges in Social Sciences (ETCSS' 2014), and the 5th South Asian International Conference (SAICON)for their useful comments on the earlier versions of this paper. All remaining errors are ours.

(1) The CBD approach is also expected to contribute to the decentralisation of power; the creation of high-quality, low-cost public goods; and empowerment. These, however, are not the focus of this paper.

2 There is not much quantitative evidence regarding the CBD approach in Pakistan using a microeconometric approach. Notable exceptions include Khwaja (2004), Kurosaki (2005), Khwaja (2009), and Kurosaki and Khan (2012).

(3) The CO formation process involves the following steps. First, PHKN contacts a village through a meeting with peer leaders (e.g., village elders, school teachers, local elected members, and religious leaders). At the first contact, initial assessment of the area is undertaken, covering general information on the village society and on its development needs. The introduction of PHKN to a village can be made through PHKN staff members who find potential villages from available secondary information, the concerned local administration (e.g., social welfare, agriculture, health, education, and livestock departments) or local politicians, and the peer leaders of a village. The first route i.e. contact through the PHKN staff is employed most frequently. After the initial contact, PHKN holds a series of meetings with peer leaders, local communities, and stakeholders. This stage is called the 1st Dialogue, and it is recorded in the PHKN log books. Subject fo satisfying the minimum criteria qualification and eliciting the willingness of a considerable number of villagers, a CO is formed. This stage is called the 2nd Dialogue. During the 2nd Dialogue, community development tools such as participatory rapid appraisal and village resource mapping are employed to identify developmental needs and priorities, and CO office bearers (the president, secretary, and activists) are elected and trained on how to run a CO (i.e., record-keeping, accounting, and savings management). All interventions undertaken by PHKN are categorised as the 3rd Dialogue.

(4) Usually, COs have a monthly meeting called the general body meeting, where CO members discuss PHKN activities, prevailing issues in the village, and future plans to address issues. CO members also deposit savings during these meetings. CO savings are recorded in individual savings accounts. All COs are provided with HRD training, the emphasis of which is on the development of income-earning skills and microenterprise management; the exact training differs from CO to CO, reflecting each community's unique needs. In villages with deficits in educational institutions, PHKN sometimes provides assistance to community-based schools. Similarly, in villages with poor health facilities, PHKN may train and mobilise informal health workers, such as traditional birth attendants (TBAs). PHKN staff members regularly visit each CO, with the average visit frequency being once every two months. During these visits, PHKN personnel discuss various issues with CO members while also checking CO records.

(5) Both the occupational distribution of population and the literacy rate figures are consistent with that at the national level.

(6) See Khan, et al. (2011) for the detail of sampling procedures. Regarding T-group households, in the first stage of sampling, 50 sample COs were chosen, and in the second stage of sampling, we collected information on five-member households, randomly chosen from the member list. To collect information on C]-group households, we surveyed non-member households living in the CO village where T households were surveyed. The sample for [C.sub.1] households was randomly selected from the electoral list of the villagers, at the rate of one per one T household. Regarding [C.sub.2]-group households, we randomly selected five households from 20 non-CO villages; these 20 villages were randomly selected from the village list.

(7) As robustness check, we also investigate factors that are potentially endogenous to PHKN interventions, particularly in the village-level multivariate analysis.

(8) This illustrates PHKN's facilitation in bringing about a local-governance system that is more inclusive than traditional institutions. Analysis in this vein is left to future research.

(9) The Probit results are qualitatively the same as the results reported in this paper.

(10) A number of variables have a potential association with some other variables, or do not show variation in the bivariate comparison; they are not included as explanatory variables in multivariate analysis.

(11) See Khan (2013) for a quantitative analysis of the causal impact of PHKN's interventions.

Table 1
Comparison of CO villages and non-CO villages (bivariate analysis)

Variable        Definition

Demography
lit_rate        Adult literacy rate (%)
vil_pop         Village Population
agri_prof_~c    %age of total population in agriculture
services        %age of total population in services
self_emp        %age of total population in self employment
lab_nform       %age of total population in non-farm labour
other_prof      %age of total population in others

Basic amenities, infrastructure, and shops
irrigated_~e    Connection to canal irrigation (dummy variable)
rd_length       Length of the road (in km) connecting the village
                  with a major market
cln_drnk_wat    Clean drinking water availability (%age of total
                  village population)
gas             Availability of gas connection in the village
                  (dummy var.)
c_tv            Avail, cable TV connection (dummy var.)
i_net           Avail, internet connection (dummy variable)
kar_shop        Avail. Karyana (grocery) shop (dummy var.)
veg_shop        Avail, vegetable shop (dummy var.)
frt_shop        Avail, fruit shop (dummy var.)

Existence of medical facilities in the village (dummy variables)
bhu             Basic Health Unit (Govt)
rhu             Rural Health Center (Govt)
dr_bhu_rhu      Doctor's presence in BHU or RHC
tba             Avail, traditional birth attendant (TBA)

Existence of education institutions in the village (dummy variables)
prim_school     Primary school (1st to 5th grades)
mid_sch         Middle school (6th to 8th grades)
hi_scho         High school (9th to 10th grades)
cbsch           Community based school
d_madra         Deni Madrassah (religious school)

Dispute settlement forums (DSF) (dummy variables)
jirga           Avail. Jirga--traditional DSF
dsf             Avail, non-traditional DSF
ler             Locally elected representative is from the village

Susceptiblity to natural disasters
dis_prone_~l    Village is prone to disaster (dummy var.)

                Mean for each group      Difference (A)-(B)

                  (A) CO     (B) Non-       Mean        (S.E.)
                 villages       CO
                  (n=40)     villages
Variable                      (n=65)

Demography
lit_rate          49.13        57.54       -8.41 *      (3.86)
vil_pop            2252        2612         -360        (369)
agri_prof_~c      55.28        52.06        3.21        (4.36)
services          16.80        22.11      -5.31 (+)     (2.97)
self_emp           5.60        9.14        -3.54 *      (1.47)
lab_nform         15.10        11.58        3.52        (2.29)
other_prof         7.23        5.11         2.12        (1.65)

Basic amenities, infrastructure, and shops
irrigated_~e      0.250        0.292       -0.042      (0.090)
rd_length         14.13        15.51        -1.38       (2.22)
cln_drnk_wat      71.38        76.52        -5.15       (6.64)
gas               0.025        0.323      -0.298 **    (0.064)
c_tv              0.175        0.323     -0.148 (+)    (0.084)
i_net             0.100        0.354      -0.254 **    (0.077)
kar_shop          0.725        0.877     -0.152 (+)    (0.082)
veg_shop          0.625        0.492        0.133      (0.100)
frt_shop          0.325        0.431       -0.106      (0.097)

Existence of medical facilities in the village (dummy variables)
bhu               0.125        0.185       -0.060      (0.072)
rhu               0.025        0.062       -0.037      (0.039)
dr_bhu_rhu        0.125        0.215       -0.090      (0.074)
tba               0.825        0.646       0.179 *     (0.085)

Existence of education institutions in the village (dummy variables)
prim_school       0.850        0.877       -0.027      (0.070)
mid_sch           0.325        0.369       -0.044      (0.096)
hi_scho           0.250        0.200        0.050      (0.085)
cbsch             0.250        0.092       0.158 *     (0.078)
d_madra           0.475        0.446        0.029      (0.101)

Dispute settlement forums (DSF) (dummy variables)
jirga             0.850        0.769        0.081      (0.078)
dsf               0.925        0.769       0.156 *     (0.067)
ler               0.650        0.738       -0.088      (0.094)

Susceptiblity to natural disasters
dis_prone_~l      0.975        0.831      0.144 **     (0.053)

Notes: 1. The standard errors (SE) are reported in parenthesis,
estimated under the assumption that allow unequal variance of two
groups. 2. The definition of a CO village is the default definition
(listed as having a CO or similar activities in the PHKN village
list). 3. ** p < 0.01, * p < 0.05, (+) p < 0.1. 4. The table is
prepared by the authors.

Table 2
Correlates of Village-level Participation (Multiple Regression
Results)

                                         Dependent variable: CO
                                         village - dummy (d_col)

                                      Model 1     Model 2     Model 3

Village-level variables
  lit_rate/100                        -0.1812     -0.0471     -0.1163
                                      (0.300)     (0.329)     (0.329)
  vil_pop/l000000                     -0.0069     0.0008      -0.0038
                                      (0.039)     (0.038)     (0.034)
  agri_Prof_Prc                       -0.0021     -0.0030     -0.0040
                                      (0.003)     (0.003)     (0.003)

Basic amenities, infrastructure,
and shops
  irrigated_village                   -0.0420     -0.0440     -0.0700
                                      (0.139)     (0.137)     (0.136)
  rd_length                          -0.013 **   -0.014 **   -0.013 **
                                      (0.004)     (0.004)     (0.004)
  cln_drnk_wat                        0.0001      -0.0007     -0.0003
                                      (0.002)     (0.002)     (0.002)
  gas                                -0,436 *    -0.380 *    -0.436 *
                                      (0.195)     (0.179)     (0.193)
  i_net                               -0.2180     -0.2140     -0.1980
                                      (0.182)     (0.167)     (0.172)
  kar_shop                            -0.1600     -0.1580     -0.1500
                                      (0.151)     (0.157)     (0.147)

Access to education and medical
facilities
  prim_school                         -0.0490     -0.0310     -0.0520
                                      (0.144)     (0.146)     (0.144)
  mid_sch                             -0.0730     -0.0740     -0.0750
                                      (0.111)     (0.109)     (0.110)
  hi_scho                             0.0950      0.0590      0.0860
                                      (0.154)     (0.157)     (0.152)
  d_madra                             0.1520      0.1600      0.1030
                                      (0.116)     (0.116)     (0.112)
  bhu                                 0.0960      0.0350      0.0650
                                      (0.164)     (0.165)     (0 164)

Susceptibility to natural
disasters
  dis_prone_vil                       0.2550      0.2830      0.1970
                                      (0.156)     (0.155)     (0.159)

Potentially endogenous variables
  dsf                                             0.246 *
                                                  (0.118)
  cbsch                                                       0.289 *
                                                              (0.138)
  tba

Intercept                            0.852 **     0.5630     0.926 **
                                      (0.284)     (0.310)     (0.290)
R-squared                              0.291       0.321       0.327
F-statistics for zero slopes           6.045       4.503       6.985
Level of Significance                  0.000       0.000       0.000

                                      Dependent variable:
                                      CO village - dummy
                                            (d col)

                                      Model 4     Model 5

Village-level variables
  lit_rate/100                        -0.1657     -0.0208
                                      (0.300)     (0.294)
  vil_pop/l000000                     0.0082      0.0131
                                      (0.038)     (0.034)
  agri_Prof_Prc                       -0.0022     -0.0034
                                      (0.002)     (0.002)

Basic amenities, infrastructure,
and shops
  irrigated_village                   -0.0980     -0.1150
                                      (0.139)     (0.134)
  rd_length                          -0.012 **   -0.012 **
                                      (0.004)     (0.004)
  cln_drnk_wat                        -0.0010     -0.0010
                                      (0.002)     (0.002)
  gas                                -0.354 *    -0.419 *
                                      (0.177)     (0.175)
  i_net                               -0.2380     -0.2140
                                      (0.172)     (0.162)
  kar_shop                            -0.1930     -0.1770
                                      (0.134)     (0.140)

Access to education and medical
facilities
  prim_school                         -0.0600     -0.0480
                                      (0.137)     (0.139)
  mid_sch                             -0.1110     -0.1060
                                      (0.113)     (0.111)
  hi_scho                             -0.0060     -0.0190
                                      (0.155)     (0.155)
  d_madra                             0.1590      0.1190
                                      (0.116)     (0.113)
  bhu                                 0.0900      0.0230
                                      (0.158)     (0.156)

Susceptibility to natural
disasters
  dis_prone_vil                       0.2980      0.2570
                                      (0.152)     (0.155)

Potentially endogenous variables
  dsf                                             0.1640
                                                  (0.130)
  cbsch                                           0.260 *
                                                  (0.128)
  tba                                0.312 **     0.252 *
                                      (0.097)     (0.104)
Intercept                             0.679 *     0.5860
                                      (0.299)     (0.329)
R-squared                              0.352       0.393
F-statistics for zero slopes           5.901       8.110
Level of Significance                  0.000       0.000

Notes: 1. In addition to those explanatory variables listed above,
intercept, Mansehra dummy, and Abbottabad dummy are also included. 2.
Estimated by OLS (linear probability model), with robust standard
errors (reported in brackets). 3. The number of observations is 105.
4. * p<0.1, ** p<0.05, *** p<0.01 . 5. Number of observations is 105.
6. The table is prepared by the authors.

Table 3
Household-level Comparison of CO Villages and non-CO Villages
(Bivariate Analysis)

Variable                         Definition

Demography
  hhsize             Number of household members
  fem_rate           Ratio of female over male members
  fem_hh             Dummy for a female-headed household
  hh_edu             Years of education of the household head
  hh_lite            Literacy dummy of the household head
  hh_age             Age of the household head

Education
  educ_yrs           Average years of education of adult
                       household members
  fem_edu            Av. yrs of education of female members
  mal_edu            Av. yrs of education of male members
  d_lit              Adult literacy rate
  fem_lite           Female literacy rate
  mal_lite           Male literacy rate

Household asset indicators
  h_floor            The flooring of the house is paved (dummy var.)
  drainge            The house has drainage (dummy var.)
  gas                The house is connected with gas for cooking
                       (dummy var.)
  land_val           Value of land owned (Rs. 1,000,000)
  livestock_val      Value of livestock owned (Rs. 1,000,000)
  radio              The household has and uses a radio (dummy)
  internet           The household uses internet (dummy)
  cab_tv             The house is connected with cable TV
                       (dummy)

Susceptibility to natural disasters
  fldaffected_hh     Affected by 2010 floods
  wildboar_attack    Suffered damages due to attacks by wild
                       boars

Social status
  native             Native household
  sol_status         Social status is high
  networking         Blood or non-blood relation with local elite

                         Weighted Mean for           Difference:
                             Each Group           (Tand [C.sub.1])-
                                                     ([C.sub.2])
                       (T and
                     [C.sub.1])    ([C.sub.2])       Mean      (S.E.)
                     Households     Households
                        in CO       in non-CO
                      villages       villages
Variable               (n=483)       (w=100)

Demography
  hhsize                6.088         6.681         -0.593     (0.561)
  fem_rate              1.127         1.042         0.085      (0.125)
  fem_hh                0.097         0.050         0.046      (0.031)
  hh_edu                5.847         6.846         -0.999     (0.840)
  hh_lite               0.701         0.744         -0.043     (0.077)
  hh_age               50.164         50.518        -0.354     (1.953)

Education
  educ_yrs              5.603         7.018       -1.415 **    (0.538)
  fem_edu               2.170         2.912       -0.742 **    (0.251)
  mal_edu               3.594         4.623        -1.030 *    (0.469)
  d_lit                 0.746         0.775         -0.029     (0.035)
  fem_lite              0.318         0.321         -0.003     (0.029)
  mal_lite              0.428         0.454         -0.026     (0.034)

Household asset indicators
  h_floor               0.083         0.461       -0.378 **    (0.098)
  drainge               0.424         0.819       -0.394 **    (0.057)
  gas                   0.001         0.822       -0.821 **    (0.045)
  land_val              0.491         0.670         -0.179     (0.374)
  livestock_val         0.015         0.003        0.012 **    (0.002)
  radio                 0.334         0.290         0.044      (0.097)
  internet              0.000         0.167        -0.167 *    (0.085)
  cab_tv                0.003         0.341       -0.338 **    (0.102)

Susceptibility to natural disasters
  fldaffected_hh        0.329         0.289         0.040      (0.084)
  wildboar_attack       0.333         0.066        0.268 **    (0.034)

Social status
  native                0.961         0.768        0.194 *     (0.086)
  sol_status            0.927         1.000       -0.073 **    (0.018)
  networking            0.408         0.058        0.350 **    (0.063)

Notes: 1. Means are weighted to reflect differences in sampling
probability. 2. The standard errors are reported in parenthesis. 3.
** p < 0.01, * p < 0.05, + p < 0.1. The table is prepared by the
authors.

Table 4
Correlates of Village-level Participation (Household-level Multiple
Regression Results)

                                 Model 1      Model 2      Model 3

                                          Dependent Variable:
                                   Dummy representing T or [C.sub.1]
                                  household with [C.sub.2] household
                                           as the reference
Explanatory Vars                               (d t cl)

Village-level variables
  lit_rate/100                     -0.300       -0.300       -0.298
                                  (0.304)      (0.305)      (0.307)
  vil_pop/1000000                  0.051        0.051        0.052
                                  (0.026)      (0.026)      (0.027)
  agri_prof_~c                     -0.002       -0.002       -0.002
                                  (0.001)      (0.001)      (0.001)
  rd length/100                    0.015        0.017        0.007
                                  (0.768)      (0.769)      (0.767)
  cln_drnk_wat                    -0.001-       -0.001      -0.0013
                                  (0.001)      (0.001)      (0.001)

Household education
  d_lit                            0.052
                                  (0.055)
  fem_lite                                      0.059
                                               (0.060)
  mal_lite                                      0.044
                                               (0.065)
  fem_edu/10                                                 -0.003
                                                            (0.054)
  mal_edu/10                                                 -0.013
                                                            (0.067)

Household asset indicators
  h_floor                          0.041        0.041        0.043
                                  (0.040)      (0.040)      (0.041)
  drainge                          -0.054       -0.054       -0.052
                                  (0.037)      (0.037)      (0.038)
  gas                            -0.690 ***   -0.691 ***   -0.691 ***
                                  (0.132)      (0.132)      (0.133)
  land_val                       -0.027 **    -0.027 **    -0.027 **
                                  (0.008)      (0.008)      (0.009)
  radio                           0.051 *      0.051 *      0.052 *
                                  (0.023)      (0.023)      (0.024)
  internet                         -0.151       -0.147       -0.141
                                  (0.141)      (0.141)      (0.142)
  cab_tv                           -0.04        -0.04        -0.038
                                  (0.090)      (0.090)      (0.090)

Household level susceptibility
to natural disasters
  fldaffecte~h                     -0.025       -0.025       -0.025
                                  (0.025)      (0.025)      (0.025)
  wildboar_a~k                     0.037        0.038        0.039
                                  (0.028)      (0.029)      (0.028)

Household level social status
and networking
  native                          0.296 **     0.296 **     0.299 **
                                  (0.104)      (0.104)      (0.104)
  sol_status                       -0.081       -0.08        -0.081
                                  (0.047)      (0.046)      (0.047)
  networking                      0.137 *      0.137 *      0.138 *
                                  (0.054)      (0.054)      (0.054)
Intercept                        0.809 ***    0.812 ***    0.840 ***
                                  (0.214)      (0.214)      (0.210)
R-squared                          0.578        0.579        0.578
F-statistics for zero slopes       71.067       68.376       68.953
Level of Sig.                      0.000        0.000        0.000

Notes: 1. Standard errors in parentheses. 2. * p < 0.05, ** p < 0.01,
*** p < 0.001.3. The number of observations is 583. The table is
prepared by the authors.

Table 5
Comparison of Member and Non-member Households within CO Villages
(Bivariate Analysis)

                                    Weighted Mean for
                                        Each Group

                                                  ([C.sub.1])
                                (7) Member         Non-member
                             households in CO   households in CO
                             villages (n=249)   villages (n=234)

Demography
  hhsize                          6.403              5.899
  fem_rate                        1.123              1.130
  fem_hh                          0.088              0.102
  hh_edu                          6.098              5.697
  hh_lite                         0.738              0.680
  hh_age                          50.046             50.235

Education
  educ_yrs                        5.767              5.505
  fem_edu                         2.157              2.178
  mal_edu                         3.773              3.486
  d_lit                           0.763              0.735
  fem_lite                        0.317              0.318
  mal_lite                        0.447              0.417

Household asset indicators
  h_floor                         0.115              0.063
  drainge                         0.456              0.406
  gas                             0.000              0.002
  land_val                        0.553              0.454
  livestock_val                   0.016              0.014
  radio                           0.319              0.343
  internet                        0.000              0.000
  cab_tv                          0.008              0.000

Susceptibility to natural
disasters
  fldaffected_hh                  0.405              0.284
  wildboar_attack                 0.397              0.296

Social status
  native                          0.977              0.952
  sol_status                      0.936              0.922
  networking                      0.322              0.460

                               Difference: (T) -
                                  ([C.sub.1])

                                Mean       (S.E.)

Demography
  hhsize                     0.504 (+)    (0.280)
  fem_rate                     -0.007     (0.095)
  fem_hh                       -0.014     (0.033)
  hh_edu                       0.401      (0.533)
  hh_lite                      0.058      (0.054)
  hh_age                       -0.189     (1.598)

Education
  educ_yrs                     0.262      (0.262)
  fem_edu                      -0.021     (0.209)
  mal_edu                      0.287      (0.216)
  d_lit                        0.028      (0.027)
  fem_lite                     -0.002     (0.023)
  mal_lite                     0.030      (0.023)

Household asset indicators
  h_floor                    0.052 (+)    (0.029)
  drainge                      0.051      (0.058)
  gas                          -0.002     (0.002)
  land_val                     0.100      (0.133)
  livestock_val                0.002      (0.003)
  radio                        -0.023     (0.055)
  internet                     0.000      (0.000)
  cab_tv                       0.008      (0.006)

Susceptibility to natural
disasters
  fldaffected_hh              0.121 *     (0.053)
  wildboar_attack            0.101 (+)    (0.054)

Social status
  native                       0.025      (0.021)
  sol_status                   0.014      (0.031)
  networking                  -0.138 *    (0.055)

Notes: 1. Means are weighted to reflect differences in sampling
probability. 2. The standard errors are reported in parenthesis. 3.
** p < 0.01, * p < 0.05, (+) p < 0.1. The table is prepared by the
authors.

Table 6
Correlates of Household-level Participation Within CO Villages
(Multiple Regression Results)

                                   Model 1      Model 2      Model 3

                                           Dependent Variable:
                                   Dummy representing T household with
                                       [C.sub.1] household as the
Explanatory Vars                             reference (d_t)

Household Education
  d_lit                              0.025
                                    (0.058)
  fem_lite                                        -0.018
                                                 (0.074)
  ma_Mite                                         0.080
                                                 (0.100)
  fern_edu                                                     0.001
                                                              (0.009)
  mal_edu                                                     0.020 *
                                                              (0.009)

Household level susceptibility
to natural disasters
  h_floor                            0.097        0.094        0.089
                                    (0.066)      (0.063)      (0.064)
  drainge                            0.031        0.030        0.038
                                    (0.046)      (0.047)      (0.046)
  gas                              -0.380 ***   -0.373 ***   -0.386 ***
                                    (0.037)      (0.039)      (0.038)
  land_val                           -0.002       -0.003       -0.003
                                    (0.013)      (0.015)      (0.015)
  radio                              0.007        0.007        0.003
                                    (0.041)      (0.042)      (0.043)
  cab_tv                           0.479 ***    0.487 ***    0.472 ***
                                    (0.057)      (0.058)      (0.076)

Household level susceptibility
to natural disasters
  fldaffecte~h                      0.107 *      0.108 *      0.110 *
                                    (0.046)      (0.047)      (0.043)
  wildboar_a~k                      0.094 *      0.091 *       0.087
                                    (0.042)      (0.041)      (0.042)

Household level social status
and networking
  native                             0.272        0.184        0.255
                                    (0.133)      (0.119)      (0.131)
  sol_status                         -0.054       -0.056       0.015
                                    (0.083)      (0.084)      (0.040)
  networking                         -0.104       -0.112       -0.11
                                    (0.074)      (0.075)      (0.087)
Village fixed affect                  Yes          Yes          Yes
Intercept                            0.118        0.113        0.088
                                    (0.151)      (0.152)      (0.137)
R-squared                            0.075        0.076        0.079
F-stat for zero slopes               122.43       88.81        26.20
Level of Sig.                        0.000        0.000        0.000

Notes: 1. The number of observations is 483 (only a subsample of
households belonging to CO villages is used). 2. Standard errors in
parentheses. 3. * p < 0.05, ** p < 0.01, *** p < 0.00l. 4. The table
is prepared by the authors. 5. "F-stat for zero slopes shows the
F-statistics for the null hypothesis that all slopes are zero except
for the intercept and village fixed effects.