Poor classifications or classifications of the poor?
Elite capture in community-based targeting methods
Master Thesis for the Double Degree Program of University of Göttingen, Master of Arts: International Economics
University of Groningen, Master of Science: International Economics and Business
Submitted by: Submission date:
Rebecca Lohmann June 19th 2018
Oostersingel 70A
9711XE Groningen Supervisor:
Dr. Anna Minasyan, University of Groningen Dr. Friederike Lenel, University of Göttingen Rebecca.lohmann@gmail.com
Abstract
This thesis examines the occurrence of elite capture in community based targeting methods. Using a data set of household and network characteristics from 21 villages in the Identification of Poor Households Program in rural Cambodia, error rates of the poverty classifications are calculated. Using a Linear Probability Model, it is tested to which extend social connections to local elites increase one’s probability of becoming a beneficent of the program or subject to an inclusion or exclusion error. Evidence from qualitative field interviews provide additional insights about the underlying mechanisms. The results suggest that elite capture cannot unambiguously explain the error rates.
Keywords: Decentralization, Elite capture, Community based targeting
Acknowledgment
Table of Contents
1 Introduction ... 1
2 Literature Review ... 2
3 Background ... 7
3.1 Decentralization in Cambodia ... 7
3.2 The Identification of Poor Households Program ... 9
3.3 Potential for Elite Capture during the classification process ... 11
1
Introduction
Targeting the poor is a core challenge of most development programs. As a consequence, community based targeting methods have gained popularity during the last decades. In community based targeting methods, the central government allows the community to decide which members are eligible candidates for an aid program. Such an approach has several advantages. Firstly, exploitation of the information advantage of the community members about each other increases efficiency. Secondly, ownership of the program is increased as people become active participants instead of passive receivers. Thirdly, social capital within the community is strengthened as the group works together for a purpose.
However, handing more power to communities may in fact mean handing more power to the local elites, which could lead to elite capture. Dasgupta and Beard (2007) define elite capture as the process by which certain individuals within a community, who have more social, political or economic power than the rest, “dominate and corrupt community-level planning and governance” (ibid., p. 230). This represents the exact opposite of the initial goal of empowering the poor. Additionally, it can lead to severe mistargeting, misallocation of resources and reinforce local power imbalances. Thus, scientifically investigating the underlying mechanisms of elite capture can have important political impact.
Scholars, studying elite capture have come to ambiguous results (Alatas et al., 2012; Persha and Andersson, 2014). The main identification problem remains the lack of a purposeful baseline scenario to compare the targeting outcome with. As the main reason to use community based targeting is that conventional identification methods do not work and specific identification criteria of the communities are not known to the researchers, this is an inherent problem.
circumstances and wealth. The households receive a score based on the interview responses, which then results in poverty classifications. This structure transfers a considerable amount of control to the selection committee, whose members represent the local elite in this framework.
To study whether elite capture is a cause for misclassifications a mixed methods approach is applied, combining quantitative analysis with results from qualitative field interviews. First, a unique household data set on 21 villages in two northern provinces of Cambodia, where the IDPoor program is conducted, is used. The data set does not only contain information of who was officially identified as poor or non-poor in 2016 but also on the household characteristics that are elements of the official questionnaire. Thus, it is possible to replicate the score calculations and compare them to the official classifications. The score replication of the IDPoor score has shown that there is a considerable degree of misclassification in the villages. Whether elite capture was a cause for the resulting errors, is tested in a Linear Probability Model regression analysis.
Second, qualitative field interviews are conducted in four of the villages from the quantitative sample. The interviews are evaluated according to the following three main questions: In how far was the official procedure of the identification process indeed applied by the villages? How is power distributed among the participants of the program? And, to what extent do the participants “own” the program?
The main findings from the quantitative analysis suggest that the number of members of the local elites to one’s social network did not increase the probability to be (wrongly) identified as poor. The results from the qualitative interviews suggest that there is considerable scope for the local elites to capture the program benefits but it could not confirm the presence of elite capture unambiguously. Therefore, I conclude that elite capture seems not to be the main reason for mistargeting in the villages studied.
The thesis proceeds as follows: In section 2, the relevant scientific literature is discussed. In section 3, the IDPoor Program will be described and weak spots open for elite capture will be pointed out. The data set and summary statistics are presented in section 4. Section 5 presents the applied methodology of the quantitative analysis and the qualitative interviews. Section 6 reports and discusses the results. Section 7 concludes.
2
Literature Review
income records and eligible households have incentives to hide assets (for example during household surveys) in order to appear poorer than they actually are. Therefore, decentralized, community-based targeting methods have become popular since the 1990s (Blunt and Turner, 2005; Mansuri and Rao, 2004). Such approaches intend to involve the communities of the poor in a more active way in the design, implementation and monitoring of the provision of social safety measures.
Conning and Kevane (2002, p. 378) define community-targeting as a mechanism which can be applied to different targeting methods. Examples for targeting methods are individual assessment of each case by a program agent, categorical targeting (or “tagging”), where members of a certain group, e.g. single women with children, are eligible, or self-targeting, where the program is designed in a way that non-targeted households will not enroll (Conning and Kevane, 2002). The targeting mechanism specifies the “choice of intermediary agents, beneficiary selection criteria, and the longer-term funding formula based in part on pre-defined evaluation methodology” (Conning and Kevane, 2002, p. 378).
The accuracy of decentralized targeting methods has been evaluated by several scholars, who have come to mixed results. Coady, Grosh and Hoddinott (2004) compare in a meta-regression analysis of 122 targeted anti-poverty interventions between 1985 and 2003 the efficacy of different targeting interventions. They compare the outcomes of three different targeting methods (individual assessment, categorical and self-selection) with the outcomes of “neutral targeting”. Under neutral targeting, each decile of the population receives ten percent of the transfer budget or each decile accounts for ten percent of the program beneficiaries. Besides simple means tests and proxy means tests, community-based targeting is one of the tested individual targeting approaches. The authors conclude that community assessment generally does not yield more accurate targeting outcomes than self-targeting, based on consumption. Yet, the sample Coady, Grosh and Hoddinott (2004) use includes only six programs using community-based targeting methods, which is quite little in order to produce globally reliable results. Additionally, the results show a large variance, making unambiguous interpretations difficult. Finally, the authors point out that there may be some sort of sample selection bias caused by insufficient documentation of unsuccessful interventions, this is especially prevalent in the case of community targeting.
program benefits among the households. However, targeting accuracy is computed based on comparison to household expenditure estimates which bear potential for inaccuracies.
In a randomized controlled trial, Robertson et al. (2014) compare the accuracy of community-based targeting with asset based targeting in a cash transfer program for orphaned and other vulnerable children (OVC) in eastern Zimbabwe. The community participation treatment involves local leaders in creating a wealth ranking of the households. The authors find that agreement between the two approaches was relatively low. Forty percent of the households were assigned to the lowest two wealth quintiles by the community-based method, including a considerable number of households which were categorized as non-poor based on the asset analysis. An additional qualitative analysis among the program participants revealed that there was more “ownership” and appreciation, more transparency, less conflict and increased feelings of unity and group identification in the community treatment. Still, favoritism, nepotism and lying were mentioned as challenges by the participants. Therefore, the authors conclude the community-based approach was more in line with the actual or perceived needs of the community than the asset-based wealth quintiles. Of course, interpretations based on qualitative data have only very limited potential to be transferred to other cases. Further, the study suffers from data limitations. First, there is no information on which additional factors may have been influencing the participatory wealth ranking. Second, some households were excluded by the participants during the community-based methods and the reasons for this are not well known.
Another comparison of targeting accuracy is conducted by Alatas et al. (2012). The authors compare three targeting approaches: Proxy means test (PMT), community targeting, where villagers jointly develop a ranking of all village members from richest to poorest and a hybrid form, where the potentially eligible members are identified by the ranking and then evaluated based on the PMT. In their field experiment on 640 Indonesian villages, that are part of a cash transfer program for poor households, they find that the PMT leads to more accurate targeting than the community based or hybrid method, when compared to per capita expenditure data. However, the differences in the targeting accuracy between the approaches are small. A second finding is that methods which actively involved community participation yielded higher levels of satisfaction and legitimacy among the participants. One shortcoming of this approach is that it relies on per capita expenditure reported by the households them-selves which is typically vulnerable to inaccuracies.
Robertson et al. (2014), Alatas et al. (2012) and Coady, Grosh and Hoddinott (2004) all conclude that community-based methods show considerable deviations from their baseline scenarios (self-targeting, means testing and per capita expenditure). Elite capture may be one explanation for the limited targeting accuracy of community-based targeting methods considered by the literature.
Conning and Kevane (2002) conduct an interpretative review of several case studies on community involvement. They argue, while the inclusive approach can provide legitimacy to poverty reduction programs and build political support, this bottom-up approach can lead to conflict and division within the community on the other hand. Local elites may exploit their power, perpetuate hierarchies and monopolize benefits that are meant to serve the poor.
Platteau and Gaspart (2003) support this view in their analysis of an illustrative case study. They explain that elite capture emerges from missing identification of the target group with community-based methods. They argue, that participatory approaches are seldom demanded by the local community but rather a manifestation of western values implemented by donor organizations. Hence, this would not lead to the local people demanding “what they would have wanted to do anyway”. Rather, the whole system remains supply driven instead of oriented on the actual needs and demands of the communities. In order to be successful, such a system required a change of historically grown local processes which can only be achieved by comprehensive training. Yet, to overcome the lack of identification of the local community with the participatory approach, democratic elections of local leaders who would facilitate the distribution of funds and benefits of the program were implemented in many cases. However, given the lack of ownership by the local community and the missing awareness of its members´ rights, this led to contrary effect. Rather, the empowerment of local elites increases the distance between the donor and the grassroots community and enables local elites to manipulate participatory methods by subtly presenting their own interests as community concerns. Moreover, this can lead to corruption of the traditional role of local elites by opportunism (Platteau and Gaspart, 2003). While these two studies apply a case-study approach, quantitative studies in the existence of elite capture is limited.
that in this setting elites and their relatives are less likely to be selected as beneficiaries regardless their income level. To compute the error rate, Alatas et al. (2012) compare the results of the community targeting approach to a baseline scenario based on a local PPP$2 per-day consumption threshold. This is a major limitation to the study, as it does not fully reflect local views on poverty. Local preferences might well be deviating from this poverty line. Another shortcoming is that the study does only consider family bonds (by blood or marriage) between the households and elites. This may lead to an underestimation of the effect as it does not consider any other type of relation (e.g. friendship).
Empirical evidence for the existence of elite capture comes from studies applying a more macro point of view. For instance, Hodler and Raschky (2014) show that political leaders favor their birth region when it comes to transfers, tax reliefs, public good provision or other benefits. Possible reasons for this may be an intrinsic urge to support their families or clans, ethnic preferences or the intent to secure political loyalty in their stronghold. While better political institutions and more education seem to reduce this behavior, higher aid inflows tend to enforce it. Öhler and Nunnenkamp (2014) come to the same results when studying the distribution of multilateral aid within 27 recipient countries.
Persha and Andersson (2014) provide further evidence for elite capture from decentralized forest governance programs on 56 sites in four countries (India, Kenya, Nepal, Uganda). They find that a small number of user group members received a disproportionately larger share of the harvest benefits from the forest. Additionally, they find that this disproportionate distribution intensifies the longer the decentralization is in place. Furthermore, involvement of an independent entity in the organization of the user group improves the distribution. However, they have only limited data on characteristics of the beneficiary group, which weakens the internal validity of the study, as it cannot be fully ruled out that there is some other reason for the group to benefit over proportionally.
Two central conclusions can be drawn from the existing literature. Firstly, targeting performance of community based methods is not better than that of other targeting methods. However, it yields in general higher levels of identification and satisfaction among the target groups. Secondly, evidence for the existence of elite capture comes mainly from case studies and studies focused on the macro perspective. Among the micro economic studies, evidence for elite capture is scarce. Only Persha and Andersson (2014) could confirm the existence of elite capture.
1. Households with connections to local elites are more likely to be identified as poor in a community-based targeting program.
2. Having connections to the local elites increases the probability of non-poor
households to be falsely identified as poor in a community-based targeting program. 3. Having connections to the local elites decreases the probability of poor households to
be excluded from the identification process in a community-based targeting program. The Identification of Poor Households Program in Cambodia represents a rich environment to test these hypotheses empirically. The main limitation all the cited studies face, is when estimating the error rate, they have to rely on a constructed baseline scenario, e.g. based on income or expenditure data or universal targeting, that may not reflect actual weights and preferences of the communities. As a consequence, error rates calculated based on these comparisons may not reflect actual errors. The IDPoor program in Cambodia provides the ideal environment to overcome this limitation, because the underlying criteria and weights are publicly accessible. Moreover, this program has not been evaluated regarding elite capture before. The setting of the empirical study will be described in the following section.
3
Background
3.1 Decentralization in Cambodia
After the overthrow of the Khmer Rouge, there was practically nothing left of the old state´s institutions in Cambodia. Army, police, hospitals, religious structures and the legal system, all was comprehensively destroyed. Moreover, grown social networks and community structures were eradicated due to large-scale displacement of the urban and rural population. The Vietnamese occupants stayed in Cambodia to oversee the building of a new and stable government, but with little democratic features, before they finally withdrew in 1989 (Blunt and Turner, 2005).
In the 1990s, with the UN´s Paris Peace contract, foreign aid and economic growth came into the country but were mostly concentrated on the capital, Phnom Penh. Among the plenty donor-funded initiatives, decentralization was one element, which became more and more popular as a means for good governance in that era. In 2002, the first elections of local commune councils took place (Blunt and Turner, 2005).
identify four central actors in local decision-making. The commune councils and commune chiefs are directly elected every five years by proportional representation and party lists. The commune council consists of 5-11 members and has two main functions: On the one hand, its members are representatives of the local population, on the other hand, they are agents of the central states. The commune councils are the central entity for local decision-making and focal point of contact for other development actors. The commune councils are supported by a number of commune council committees. These provide general representation of all villages in the commune and are headed by the commune chief. Among the central committees is the Planning and Budgeting Committee (PBC) which consists of commune council representatives, village representatives selected by the commune and 2-4 ordinary citizens.
The village chiefs and their deputies (vice chief) are appointed by the elected commune council, since 2006. Before, they were appointed by the Ministry of Interior. They facilitate the council´s communication with the broader village population and are member of PBC. However, they cannot be members of the commune council.
Finally, there are administrators at village, commune, district or provincial level. They are in general appointed by the Ministry of Interior. One exception is the district administrator in charge of the IDPoor program, who is appointed by the Ministry of Planning. The administrators are involved in the planning and implementation of almost all activities, as their duties are financial management, procurement and civil registration.
Several studies have shown that dominant traits of the Cambodian culture in general are in conflict with the basic idea of decentralization. For instance, Blunt and Turner (2005) identify key values that build the necessary foundation for a successful decentralization process. These are “commitment to popular participation, acknowledgement of local autonomy, support for bottom-up decision-making, special consideration of the most vulnerable” (ibid., p. 77). Analysing the cultural context in Cambodia based on the framework of cultural dimensions of Hofstede (1980), they rather conclude that high preferences for power-distance, uncertainty avoidance and collectivism and medium preference for masculinity exist. Features that disagree fundamentally with the idea of decentralization.
Öjendal and Sedara (2006) argue that democratic decentralization in Cambodia would require profound changes in the attitude of the people. However, they do see Cambodia on the right track towards this goal. They observe changing conditions by which the former authoritarian political culture is challenged. Further, they describe a rural society which “is becoming more plural, more complex, more open and at the same time less predictable, less docile, and less easily subdued” (p. 526). Furthermore, they describe new avenues for exploitation and repression, for example in the form of local elites who are now competing for wider political space.
3.2 The Identification of Poor Households Program
The “Identification of Poor Households Program” (IDPoor program) is an element within the national decentralization effort in Cambodia. It is designed to standardize the identification of poor households and provide a data base to facilitate the distribution of benefits for governmental and non-governmental providers. The identified households receive an Equity Card, which grants access to free health care services to its owner, as well as other benefits such as scholarships for poor students, provision of social concession land, provision of agricultural services and partial or total exemption for poor households from the payment of local contributions. Additionally, based on the data of identified households, the central government is enabled to select priority villages for development (Cambodian Ministry of Planning, no date, p. 3).
The Cambodian Ministry of Planning established the IDPoor program in 2006, in cooperation with the German development agency (GIZ), who will continue the support at least until the year 2019. Initially, the program was fully funded by the German Ministry of Cooperation and Development (BMZ). In 2009, Australia´s Department of Foreign Affairs and Trade took over a large share of the costs. The implementation process is structured in different “rounds”. The first round took place in 2007 in 17 provinces, since then all provinces have been covered by the program and updates of the data are carried out every three years. In 2011, the Cambodian government confirmed the IDPoor program as the new standard method for the identification of poor households.
as poor, as well as exclusion errors, i.e. classifying poor households as non-poor) (e.g. Licht, 2015), however a detailed analysis of the program´s targeting mechanism has not yet been conducted.
Following, the identification procedure is outlined as it is described in the Implementation Manual of the Procedure for Identification of Poor Households (Cambodian Ministry of Planning, no date) published by the Ministry of Planning. The procedure is divided in seven steps on provincial, commune and village-level. After outlining the procedure, I will draw attention to several “weak spots” that provide potential for elite capture.
The procedure starts at commune level. In a first step, two to four individuals will be selected for the Planning and Budgeting Committee Representative Group (PBCRG) on commune level. The PBCRG is the central coordinating and monitoring organ on commune level. Subsequently, there is a four-day training on district level for the PBCRG members, the Commune Chief, trainers for the Village Representative Group and NGO representatives, to inform them about the purposes, procedures and relevant details of the program.
In the second step, the central operating entity on village level, the Village Representative Group (VRG), is selected. Suitable candidates, who have former experience from other committees, are identified by the PBCRG and then elected by all villagers in an open election at a village meeting. Depending on the number of households in the village, the VRG will consist of five to ten members.
Step three is the central step of the identification process. After preparing a list of all households, excluding only those that are certainly not poor, the VRG members will conduct interviews with all households on the list. Those interviews are based on standardized questionnaires provided by a District Facilitation Team and contain question about the living conditions, income sources and assets owned by the households. 1 The answers receive points, which are finally used to calculate a poverty score by the VRG. Based on this score, the households are classified into Non-Poor (0-44 points), Poor Level 2 (45-58 points) or Poor Level 1 (59-69 points). Subsequently, the VRG comes together and discusses whether some of the results should be recategorized, e.g. if the households experienced a negative shock and the results from the survey do not reflect the actual living conditions of the households. Finally, the VRG creates a Draft Household Poverty Categorization List. After this list is reviewed by the PBCRG on commune level, it is publicly displayed on a central place in the village next to some explanations about the procedure for complaints.
Step four consists of a Village Consultation Meeting a few days after the publication of the Draft Categorization List. During this meeting, the identification process and the meaning of the poverty categories are explained again. Villagers have the opportunity to hand in complaints or objections about the First Draft List within seven days after this meeting. After that period the VRG meets again to discuss these objections and prepare the Final Draft List of Poor Households. In a commune council meeting (step five), this Final Draft List and the objections are reviewed as well as the solutions the VRG proposed and the “Final List of Poor Households” is endorsed.
Steps six and seven consist of entering the data from the identification process into the data base and sending it to the Ministry of Planning, which will then print the Equity Cards accordingly and the individual cards will be distributed within the villages by the Village Chief and VRG. Moreover, the Ministry of Planning will file several reports for relevant stakeholders. In areas where providers of services demand photographs for the Equity Cards, a local NGO will be instructed to take photographs of the poor households.
3.3 Potential for Elite Capture during the classification process
During the process described above, there are several “weak spots” that have potential to be exploited by opportunistic elites. For example, right in the beginning, the requirements for PBCRG members are former experience with leadership in committees as well as reading and writing skills and basic calculus. This favors village chiefs and deputies as well as other privileged villagers for the important monitoring and coordination tasks of the PBCRG.
Further, the selection of VRG candidates in step two can be severely distorted, if the PBCRG already consists of the village chief or other elites. They will have the power to select candidates for the VRG, that suit their own purposes, such as family members or close friends. Additionally, the open election process will prevent villagers to vote for candidates that are not favored by the village elites. Even more alarmingly, there is evidence that such village meetings are not taking place at all or without informing all villagers beforehand (Licht, 2015). Moreover, the training of the VRG is crucial to guarantee a smooth operation of the process. However, it is not confirmed that those trainings are done in every village.
interviews give the VRG members a range of opportunities to manipulate the scoring. They could for example give information about the scoring to the interviewees, turn a blind eye at asset possession or use some inaccuracies in the wording of the questions in favor of the interviewees. Of course, the basic idea of community-based targeting is that the whole community functions as a monitoring unit and uncovers such manipulations. However, this mechanism will not work if the complaint procedure is flawed. Licht (2015) is presenting evidence that this is exactly what is happening: In the village, where she conducts her qualitative ethnographic study, the complaint procedure is not explained to the villagers or the villagers do not dare to question the decisions of the chief of village or other elites involved in the process or plainly do not trust that their objections will be taken seriously.
Similarly, the Village Consultation Meeting (step 4) may become useless if its importance and purpose are not communicated correctly to the villagers and those that attend do not dare to express their objections or the objections are not treated accordingly.
Moreover, at the commune level (step 5), PBCRG members are in charge of preparing the meeting. Thus, they prepare an overview of the objections in the villages and prepare the Poverty Rate Comparison Table. This gives the PBCRG members quite some power over the monitoring process. If objections are not reviewed thoroughly and if the comparison of the actual classifications with the Poverty Rate Comparison Table shows significant deviations, PBCRG members have the power to influence the classification outcomes.
Anecdotal evidence for problems in the identification process due to the local power structures is further delivered by Licht (2015). In her ethnographic master thesis, she qualitatively assesses the interrelations between pre-existing local social structures and the IDPoor program in one village in the province Takéo in southern Cambodia. The author finds that the village community expects the leader to take over the management of development interventions, such as the IDPoor program and accepts this as the “normal course of events” (p.52) and a good relation to the chief is seen as the basic prerequisite to be considered a beneficiary of the program. Therefore, villagers decide to subordinate rather than making use of their right to complain out of strategical considerations. Thus, the power position of the chief is strengthened and pre-existing power imbalances are reproduced. However, these findings are not quantitatively backed up and given the qualitative nature of the data, they are not necessarily transferable to other villages.
4
Data
4.1 Data sources
The data for the analysis are three-fold: A rich household and network survey data set, official program documents, and qualitative interviews.
Household survey data: The data on household and network characteristics stems from a set of unique household survey data, collected from March to October 2015 in 21 villages in the provinces Banteay Meanchey and Siem Reap in North-western Cambodia. The villages were selected based on their size, level of migration and remoteness. The survey data set contains comprehensive data on characteristics of the 1,262 surveyed households as well as information on their social networks and the positions they hold in the village administration. Additionally, a list of all households officially classified as poor in 2016 per village was collected.
The 21 surveyed villages have between 84 and 234 households. In total, 1,262 Households were interviewed (between 53 and 83 per village). Among the total 2,732 households in the 21 villages, 398 households (14.6 percent) were categorized as poor level 1 or 2 in 2016.2 Figure 1 displays the total number of households per village. The lower and darker bars represent the share of surveyed households per village. The filled areas represent the share of households identified as poor, the cross hatched parts represent those identified as non-poor. There is considerable variation between the villages. For example, in village 9 and 18 only 1 percent of the population received the poverty status, while in villages 1 and 13, 20 percent were identified.3
2 The villages started the implementation of the IDP programme in different years. Thus, 2016 was already the
third round for some, while for other villages it was the second and for one village it was even the first round. Further, there are differences in the time in-between the rounds. While in most villages it was three years, some have repeated the classifications by a two-year rhythm.
3 The data set was kindly made available to me by Dr. Friederike Lenel (University Göttingen) and Dr. Susan
Data from public program documents: Further information was derived from the official documents of the IDPoor program published by the Cambodian Ministry of Planning. Especially relevant were the Household Questionnaire for Identification of Poor Households (appendix A 1) and the Implementation Manual on the Procedures of Identification of Poor Households.4
Qualitative interviews: The third data source consists of qualitative interviews with relevant stakeholders, such as households, chiefs of villages, members of the village representative groups as well as commune and district level representatives. I conducted the interviews in April 2018. The main goal of the interviews was to gain insights into the degree of compliance with (or deviation from) the official steps of implementation of the IDPoor Program and by that derive additional evidence for the existence of elite capture.
4 The Implementation Manual in English can be found online:
http://www.idpoor.gov.kh/Data/En/Reference/IDPoor_Procedures_Manual-2012-05-29-Eng-FINAL.pdf 0 20 40 60 80 100 120 140 160 180 200 220 240 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 N u mb er of H H Village
Surveyed & non-poor HH Surveyed & HH officially classified as poor Not surveyed & HH officially classified as poor Not surveyed & non-poor HH
4.2 Summary Statistics
4.2.1 Poverty classifications
The household survey data contains all the household characteristics needed to replicate the items from the official questionnaire used in the identification process, this makes it possible to reproduce the score for each household.5 Figure 2 shows the distribution of the reproduced score over the population. About 13 percent received a score of zero. The median score is 26, meaning that half of the sample received a score above 26. Only 5 percent received a score above 45 and thus qualified to be identified as poor.
Matching the reproduced score with the information about which households were officially classified as poor, allows to calculate corresponding error rates. Table 1 shows the results of these calculations.
5 I would like to thank Dorothee Buehler (University Hannover) at this point for her support with the score
calculations.
Figure 2: Reproduced score
Table 1: Households officially and technically identified as poor
(1)
Number of officially identified as poor 200
Number of technically identified as poor 81
Number of inclusion error 162
Share of inclusion error of total officially identified as poor. 0.81
Number of exclusion error 43
Share of exclusion error of total technically identified as poor 0.53
The number of those identified as poor based on the reproduced score (“technically identified”) is with only 81 substantially lower than that of the officially identified as poor. The inclusion error consists of those households which are officially classified as poor but based on the reproduced score are not. This is true for 162 households in the sample. The share of inclusion error is a substantial 81 percent out of all households officially identified as poor. The large inclusion error might be caused by elite capture: Local elites use their influence on the classification outcome to favor community members they have a good relation to. Such that close friends of the local elites receive the social benefits that come with being classified as poor, even though they are non-poor.
On the other hand, an exclusion error exists when households are technically poor but were not actually identified as poor. 53 percent of all households, categorized as poor by the reproduced score, did not receive the actual identification as poor. In this case, poor households are excluded from the social benefits they are entitled to. If this can be associated to whether the household has social relations to the VRG member, this is also a form of elite capture. Underlying reasons may be that the local elites simply do not know the poor households and therefore forget to consider them, or that the elites actively decide to exclude certain households from the classifications, for example as some sort of punishment.
source is the activity where more than half of the household members engages in. To exclude the probability that results are driven by these inaccuracies an alternative score was calculated excluding the main income source as a factor.6
Further, many questions are phrased in a way that allows the interviewer to apply own judgement, such as “Little equipment and in fair condition” (question 6c) where the meaning of “fair condition” lies in the eye of the beholder. This issue is further discussed in the results section 6.2.7
4.2.2 Social networks
The household survey data set contains a number of questions regarding the social networks of the respondents. In the context of the IDPoor program the VRG is the central selection committee and therefore the central group of elites. Figure 3 shows the number of VRG members per village in 2015. The size of the VRG ranges from only two members in villages 12 and 13 to five members in villages 4, 8, 9 and 18. In many villages the chief or vice chief of village are also members of the VRG. Not all VRG members are part of the household survey. The lower and darker section of the bars represents the VRG members who have been surveyed.
A “connection” to a VRG member exists when a household (a) visits the VRG member in their free time or is visited, (b) shares labor with the VRG member, (c) would ask the VRG member for financial assistance (borrow 50,000 Riel8), (d) would assist the VRG member financially (lend Riel 50,000), (e) exchanges in-kind goods with the VRG member on a regular
6 Appendix A 3 shows the summary statistics of these calculations.
7 Section A 2 of the appendix explains the construction of Q6 in more detail. 8 50,000 Riel ≈ 10.50 EUR 0 1 2 3 4 5 6 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Village
basis, (f) ask members of the VRG member for advice on important decisions, (g) received regular financial assistance in the last 12 months from the VRG member or (h) provided regular financial assistance in the last 12 months to the VRG member. If at least one of these is true, the household has a connection to a VRG member. If the same VRG member was named several times by the same household it is not counted again.
The share of households which reported at least one connection to a VRG member is displayed in the three left bars of Figure 4. 27 percent of the full sample reported at least one connection to a VRG member. The share of the households officially identified as poor who have at least one connection to a VRG member is with 23 percent only slightly lower.
This definition includes several distinct types of relationships. Therefore, different variables (A-C) have been constructed for tree different types of relationships. If the household members named the VRG member in question (a) at least once, a type A relationship exists. This is true for 12 percent of the sample, regardless the poverty status. If the household named the VRG member in any of the questions (c), (d), (g) or (h), a type B relationship, based on financial support exists. This is true for 15 percent of the total sample. The share is with 16 percent slightly higher among the non-poor households, while it is with 14 percent slightly lower for the households identified as poor. The group of connections based on other support such as exchanging in-kind goods, sharing labor or asking for advice (questions (b), (e) and (f)) is the most occurring type. Out of the total sample 24 percent report such a connection, while it is
0,00 0,05 0,10 0,15 0,20 0,25 0,30 Connection to VRG
member Spend free timetogether (A) Financial support (B) Other support (C)
Sh are of su rv eyed H H
Officially non-poor Full sample Officially poor Figure 4: Connections to VRG members
Note: Total of officially non-poor: 1,062 households; Full sample: 1,262 households; Total of officially poor:
only 20 percent of those identified as poor. Overall, the differences in the numbers of connections are quite small between the sample of officially poor and non-poor households.
4.2.3 Household characteristics
Table 2 displays characteristics of the surveyed households in the sample. On average, the head of household received 2.6 years of education, with non-poor households having on average about 3 more years than poor-households. On average, a household consist of 6 members. The income is on average 471 USD per (good) month, however there is substantial variation within the sample. Half of the sample has an income below 245 USD, whereas the highest reported income is above 25,000 USD. Also, non-poor households earn on average almost 200 USD more than poor households.
As the rural population depends strongly on agriculture, land size is an important determinant of wealth. This is reflected in the large difference in land size between poor and non-poor households (320 and 175 Ar on average, respectively). Also, poor households own less cows and have smaller houses than non-poor households.
Table 2: Selected household characteristics of poor and non-poor households in the sample
Full sample
Officially non- poor
Officially poor
VARIABLE N Mean Median
Standard
deviation Min Max N Mean N Mean Years head of household
went to school 1,262 2.60 2 2.92 0 16 1,062 5.57 200 2.49 Number of household
members 1,262 5.63 5 2.28 1 18 1,062 5.65 200 5.53 Maximum earnings of the
household in a good month ($US)
1,249 470.81 245 1,039.45 0 25,150 1,053 496.75 196 331.46 Land size in Ar 1,262 297.60 215 304.35 0 5,000 1,062 320.95 200 175.41 Number of cows the
household owns 1,262 1.23 0 2.15 0 40 1,062 1.34 200 0.64 House size in square meters 1261 52.8 49 30.2 6 448 1062 55.4 200 39.1
Note: A number of additional household characteristics have been examined. Table A 4 in the appendix
Figure 5 demonstrates additional household characteristics by which poor and non-poor households differ significantly. In the male dominated Cambodian society (s. Blunt and Turner (2005)), a larger share of the households officially classified as poor has a female head of household than among the non-poor. Further, among the non-poor households more than 60 percent own a motor-cycle, while it is only about 40 percent among the poor households. In rural Cambodia, a motor cycle is the single most important means of transportation. Overall, very few households in the sample own a tractor, which shows the general level of poverty in the agriculture-focused area. Yet, the difference between poor and non-poor households can be observed here as well. Finally, the housing situation is significantly worse for poor households, who live 20 percent more often in dilapidated houses.
5
Methodology
In order to find out whether elite capture is a cause for misclassifications in community-based targeting methods a mixed methods identification approach is applied. The core element of the analysis is a quantitative regression analysis, using a unique survey data set of household
0,00 0,10 0,20 0,30 0,40 0,50 0,60 0,70 Head of household is female Household owns a motorcycle Household owns a tractor In general, the house is in average condition In general, the house is in a dilapidated condition In general, the house is in a good condition Sh are of su rv eyed H H Officially Non-poor
Full sample Officially Poor
and network characteristics of participants of the Identification of Poor Households (IDPoor) Program in rural Cambodia. The findings of this quantitative analysis will be discussed in the context of the results of qualitative, semi-structured interviews with participating households in the same area. This explanatory sequential design has the advantage that the findings from the quantitative analysis can be interpreted in the context of the insights from the qualitative field interviews and therefore the analysis gains depth and additional internal validity.
5.1 Regression analysis
For the data analysis, I follow the paper of Alatas et al. (2012) who examine the existence of elite capture by analyzing the impact of interpersonal relations of villagers with the local elites. To test the first hypothesis, the following equation is estimated using a Linear Probability Model:
(1) 𝐼𝐷𝐸𝑁𝑇𝐼𝐹𝐼𝐸𝐷𝑖𝑣 = 𝛽0+ 𝛽1𝐶𝑂𝑁𝑁𝑉𝑅𝐺𝑖𝑣+ 𝑋𝑖𝑣𝛽2 + 𝜀𝑖𝑣
where i represents a household, v the village, X is the vector of additional control variables. ε are standard errors clustered at village level, as the village is the main decision-making level in this context. The dependent variable is the binary variable “IDENTIFIED”, taking on the value one if the household was officially identified as poor in 2016 (either level 1 or 2) and zero otherwise.
The variable of interest is the dummy variable “CONNVRG”, indicating whether a household has at least one connection to a member of the VRG. The variable equals one if the household reported at least one connection to a member of the VRG and zero otherwise.
The vector of control variables, X, consist of several variables. The score is the main determinant for the classifications, therefore the continuous variable “SCORE” representing the reproduced score of each household is included in the regression. Households with larger social networks are per se more likely to also count VRG members to their network links. Therefore, the overall size of a household’s network needs to be controlled for by “NWSIZE”. “NWSIZE” is the sum of all unique households the household named as connections. Moreover, households with a female head may be less likely to have a connection to a VRG member. Therefore, the binary variable “FEMALEHEAD” is included in the regression. It takes the value one if the household has a female head and zero otherwise.
(2) 𝐼𝑁𝐶𝐿𝐸𝑅𝑅𝑂𝑅𝑖𝑣 = 𝛽0+ 𝛽1𝐶𝑂𝑁𝑁𝑉𝑅𝐺𝑖𝑣+ 𝛽2𝑋𝑖𝑣 + 𝜀𝑖𝑣
where i represents a household, v the village, X is the same vector of additional control variables. ε are errors clustered at village level. The dependent variable “INCLERROR” takes the value one if a household was identified as poor but was non-poor according to the reproduced score. It is zero if it was neither officially classified as poor nor based on the reproduced score.
In the presence of elite capture the coefficient of “CONNVRG” would be positive, indicating that having a connection to at least one member of the VRG increases the probability of being wrongly identified as poor.
The third hypothesis is tested by the following regression equation: (3) 𝐸𝑋𝐶𝐿𝐸𝑅𝑅𝑂𝑅𝑖𝑣 = 𝛽0+ 𝛽1𝐶𝑂𝑁𝑁𝑉𝑅𝐺𝑖𝑣+ 𝛽2𝑋𝑖𝑣 + 𝜀𝑖𝑣
where i represents a household, v the village, X is the same vector of additional control variables. ε are errors clustered on village level. The dependent variable here is the exclusion error (“EXCLERROR”). The binary variable takes the value one if a household was not officially identified as poor even though the reproduced score suggests that the household is poor. A negative coefficient for “CONNVRG”, suggests that having at least one connection to members of the VRG decreases the probability of being wrongly excluded from the classification. This is what is expected under elite capture.
5.2 Field interviews
Qualitative field interviews were conducted in April 2018 in four villages, all of which are located in Beanty Meanchey Province in northern Cambodia. The selected households were also part of the survey from 2015. Hence, for most interview partners, there is rich information about their socio-economic background available.
The villages for the qualitative interviews were chosen mainly based on the share of interviewed households who had actually received the IDPoor status in 2016. All villages with a share below 18 percent were dropped. Additionally, there must have been inclusion error in the village. The final selection criterium was the accessibility of the villages by motor cycle. That way, four villages were chosen for the qualitative interviews.
while village 3 has the largest amount with 14. The number of exclusion errors remains quite low in absolute numbers and ranges between 1 or 2 households per village. However, compared to the number of technically poor the share is large. For example, in village 2, 100 percent of those households identified as poor based on the reproduced score are not officially identified as poor.
Table 3: Village characteristics for qualitative interviews
In each village two households were interviewed, with the exception of village 1, where an additional four households were interviewed. The households were selected carefully, based on their poverty status. Most interview partners at the households were women. It was ensured that there were households of all types represented: Those officially classified as poor, those identified as non-poor who had the card before and those who did not. Additional factors were the number of (dependent) household members and the maximum monthly income. Thus, a broad overview of the perspective of the villagers can be included into the analysis. Further, the chief or vice chief of each village was interviewed and at least one additional member of the
VRG. Thus, this allows a broad impression of the actual procedure of identification in the villages.9
The field interviews with individuals involved in the IDPoor program were guided by three key questions:
1. Did the actual identification process deviate from the steps described in the official implementation manual?
2. How is power distributed among the different actors involved in the program? 3. What is the attitude of the participants (especially the households) towards the
program?
Two experienced Cambodian university graduates accompanied me to the villages and translated the interviews in Khmer and English. They are familiar with the region and the IDP program.
In each village we first presented our-selves and the purpose of our visit (“Doing research about the IDP program”) to the chief of village. The chief then guided us to the households for the interviews. Almost all interviewees were at home due to the Khmer New Year festivities that took place in April 2018. All interviewees were willing to participate spontaneously in the interviews. Despite the invaluable support by the local research assistants, it cannot be ruled out that my presence as a foreign, German woman per se influenced the behavior of the interview partners. Cilliers, Dube and Siddiqi (2015) point out, that participants of their field experiment in Sierra Leone were influenced by the presence of a person perceived as a white foreigner. They observe experimenter demand effects that go into two directions: Some of the participants of the dictator game decided to give more, in order to please the white foreigner. Others decided to give less in order to appear poorer. To minimize this effect, I emphasized during each interview that I am not involved with the program administration. However, it cannot be ruled out that the interview partners did not believe me or still (subconsciously) altered their behavior.
9 The names of the villages, communes and districts are not published in order to keep the privacy and anonymity
6
Results
6.1 Regression analysis
Table 4 shows the results of regression (1). The dependent variable in the Linear Probability Model is the dummy variable IDENTIFIED.
Table 4: Was household officially identified as poor?
VARIABLES (1) IDENTIFIED (2) IDENTIFIED (3) IDENTIFIED (4) IDENTIFIED Connection to VRG member (CONNVRG) -0.0253 (-0.773) -0.00498 (-0.152) Number of connections HH has to VRG members 0.0108 (0.404) Connection to VRG member:
Spending free time (A)
0.0168 (0.452) Connection to VRG member: Financial support (B) 0.0124 (0.283) Connection to VRG member: Other support (C) -0.0218 (-0.496) ID Poor score (SCORE) 0.00697*** (7.367) 0.00699*** (7.338) 0.00698*** (7.312) Number of links in the social
network (NWSIZE) -0.00548* (-1.992) -0.00631** (-2.120) -0.00571** (-2.091) Head of household is female
(FEMALEHEAD) 0.0915*** (3.107) 0.0915*** (3.127) 0.0908*** (3.068)
Village FE Yes Yes Yes
Constant 0.165*** (7.596) 0.118*** (3.711) 0.116*** (3.658) 0.117*** (3.655) Observations 1,262 1,262 1,262 1,262 Adjusted R-squared 0.000 0.128 0.128 0.127
Note: Robust t-statistics in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Dependent variable: Was HH officially identified as poor? (Y=1, N=0)
fit of the model, according to the adjusted R-squared but does not change this result.10 The score a household received (SCORE) is statistically highly significant and positive indicating that a higher score makes it more likely for a household to be identified as poor. This is in line with the expectations as the score is supposed to be the main deciding factor, yet, the effect is economically very small. The negative coefficient of NWSIZES indicates that households with larger social networks in general are less likely to be classified as poor. Finally, whether the head of household is female (FEMALEHEAD) seems to increase the probability of being identified as poor. The coefficient is positive and statistically significant at the 1 percent level. Thus, households having a female head are about 9 percent more likely to be identified as poor.
To exclude the possibility that the results are driven by the construction of CONVRG alternative definitions of the connection to VRG members are tested. In column (3) of Table 4, the total number of unique connections to VRG members of the household are measured, rather than the dummy variable. This takes into account that one household could have connections to several VRG members, which could in turn increase the number of advocates for the household in the selection committee. The coefficient is positive, indicating that one additional connection increased the likelihood of being identified as poor by about one percent. Yet, the result stays statistically insignificant.
Given the diverse types of relationships included in CONNVRG, it is further possible, that the effects cancel each other out. Therefore, different dummy variables according to the types of relationship are included in the regression in column (4). Three different types are identified: Spending free time together (A), Financial support (B) and other support (C), such as sharing labor or exchanging in-kind goods. It is noteworthy, that the coefficient of spending free time together is negative, indicating that this type of relationship makes an erroneous identification less likely. On the other hand, relationships based on financial or other support have a positive influence. Village fixed effects were included to control for village level differences. Nevertheless, none of the coefficients is statistically significant.
Regression (2) tests hypothesis 2. In particular, it tests whether having at laest one connection to a member of the VRG increases the probability for a household of being identified as poor even when it is non-poor based on the reproduced score. Table 5 displays the results. The dependent variable in column (1) indicates whether an inclusion error exists (INCLERROR).
10 Whether the head of household was born in the village could influence the relationship it has with any local
Table 5: Inclusion Error VARIABLES (1) INCLERROR (2) INCLERROR (3) INCLERROR (alternative score) (4) INCLERROR (5) INCLERROR Connection to VRG member (CONNVRG) -0.00118 (-0.415) 0.00938 (0.335) 0.0232 (0.820) Number of connections HH has to VRG members 0.0208 (0.860) Connection to VRG member: Financial support (B) 0.0223 (0.521) Connection to VRG member:
Spending free time (A)
-0.00356 (-0.0857) Connection to VRG member: Other support (C) 0.00385 (0.0899) ID Poor score (SCORE) 0.00567*** (6.194) 0.00568*** (6.218) 0.00567*** (6.170)
Alternative IDPoor score 0.00545***
(5.987) Number of links in the social
network (NWSIZE) -0.00606** (-2.146) -0.00624** (-2.468) -0.00694** (-2.272) -0.00621** (-2.124) Head of household is female
(FEMALEHEAD) 0.0792** (2.533) 0.0767** (2.428) 0.0793** (2.531) 0.0791** (2.527)
Village FE Yes Yes Yes Yes
Constant 0.140*** (7.303) 0.123*** (4.403) 0.123*** (5.327) 0.121*** (4.368) 0.123*** (4.372) Observations 1,181 1,181 1,103 1,181 1,181 Adjusted R-squared -0.001 0.094 0.088 0.095 0.093
Note: Robust t-statistics in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Errors clustered at village level.
Dependent variable INCLERROR: Is the household officially identified as poor but not according to the reproduced score? (Y=1, N=0).
less likely. FEMALEHEAD has a positive coefficient and is significant at the 5 percent level, suggesting that households with a female head are more likely to be wrongly identified as poor. In order to control for village specific differences, village fixed effects are included. The outcome of interest (CONNVRG) is positive and economically small but stays insignificant at any conventional level.
As discussed in section 4.2.1, it is possible that some misclassifications are caused by inaccuracies in the score calculation. Therefore, an alternative score is calculated leaving out the questions about the main income source. Column (3) of table 5 shows the corresponding results. The results stay robust. There is no statistically significant effect of CONNVRG reported. Similar to Table 4, two alternative definitions of connections to the VRG member have been tested. The results are displayed in columns (4) and (5) but do not reveal any statistically significant results. However, it is noteworthy that the coefficient for spending free time with a VRG member (column (5)) is negative, while financial and other support have a positive coefficient.
In regression (3) hypothesis (3) is tested. With the exclusion error as the dependent variable, it is estimated in how far a connection to a member of the VRG reduces the probability of not being identified as poor even though the household is poor. Estimating the probability of exclusion error completes the analysis of the different types of elite capture. However, none of the results is statistically significant at a conventional level. Also, the number of observations used for the regression estimation is reduced drastically due to the small subsample of households identified as poor based on the reproduced score. Thus, the results from regression (3) do not have a high explanatory value. The corresponding regression table can be found in the appendix (A 6). It shows signs of the coefficients which are in line with the expectations: At least one connection to a member of the VRG decreases the probability of poor households not being identified as poor.
Given the non-significant results for the variable of interest in all three regression specifications, none of the three hypotheses can be supported empirically. Thus, an unambiguous conclusion regarding the effect of having a connection to a VRG member cannot be drawn at this point.
6.2 Field interviews
identification process were described in section 3.3. Elite capture can only emerge if the local elites have the power to capture the benefits of the program.
The field interviews revealed that there exist large differences in the understanding of and knowledge about the IDPoor program on the different institutional levels. For example, the interviewees on village level almost entirely referred to the practical use of the equity card, i.e. free health care and in some cases also the free transportation to the hospital. Moreover, they only commented those activities they saw happening, such as the interviews when they participated in it. They generally had a good idea about the criteria tested in the interview. The commune representatives showed a little more distance to the subject and named “improving the lives of the poor” and “gather information of the level of poverty in the villages” as the main purposes. Finally, the district representative, mentioned “poverty reduction” and “identification of households most in need in case of disaster” as main purposes, demonstrating an ability to bring the IDPoor program into the broader context of national development.
One reason for the low level of understanding of the program by the households could be their overall very low level of education. This makes it difficult for the households to search for information by them-selves (e.g. on the internet – access to information sources is of course an additional problem). Moreover, the interviewees did not show any particular interest or drive to find out who the actual decision-maker was or how exactly the program worked. They seemed to just assume that the chief or some other authority made the decisions and accepted the outcome without further questions. Such a behavior was also described by Öjendal and Sedara (2006, p. 519). The authors explain that Cambodian civil society had often been described as “docile, (…), cowed into apathy by authoritarian politics and belittled by a patronizing donor community”, attributes which they argue contradict the idea of community participation.
One of the concerns described above was, that the VRG members might not receive the trainings they are supposed to. Against the expectations, the trainings seemed to have served very well to teach the VRG the relevant practical points for the interviews and the calculation of the scores. During the interviews some VRG members proudly explained how they observed some of the indicators rather than asking for them, demonstrating how well they were trained in these matters. Only in village 3, the VRG members were not trained but informed individually by the chief of village.
In conclusion, the members of the VRG, in general, take their duty of the interviews seriously and have an idea of the importance of their role in the identification process. However, this is subject to comprehensive trainings and there was no evidence that their engagement extended the boundaries of conducting the interviews and tallying the scores. The example of village 3 shows that the procedure depends on the engagement of the chief of village as the highest local authority. These findings indicate that elite capture could be concentrated on certain villages with specific characteristics (e.g. a chief with certain personality traits). Thus, a separate estimation of only those villages with above 80 percent inclusion error rates was conducted. However, this did not yield different results, regarding elite capture. The corresponding table can be found in appendix A 7.
The powerful position of the local authorities becomes more apparent when participatory practices during the implementation process are analyzed. Plummber and Tritt (2012, p. 32) observed several types of participatory practices in the communities they examined. Some of them were also described during the field interviews. For example, the village meetings, which are supposed to be interactive events, were in facto focussed on “information”, where “authorities provide information but apply no procedures for altering decisions”. Other examples for this type are the assignment of the VRG members by the chiefs alone, or the informal communication of the poverty status.
maker. At the same time the field interviews revealed that this seems not only to be accepted but rather to be expected by all involved actors.
Given the dominant positions of the chief and vice chief of village, it makes sense to additionally test whether having at least one connection to the chief or vice chief of the village could lead to elite capture rather than connections to the VRG member. The survey data set allows to identify connections to the chief in 12 villages and to the vice chief in 13 villages.
Table 6 shows that 28 percent of the households in the subsample have at least one connection
to the chief of village (CONNCHIEF), while only 8 percent named the vice chief as a connection (CONNVICE). The correlation analysis reveals a weak positive relation of having a connection to the chief or vice chief of village with being erroneously identified as poor. Yet, given the low number of observations this relation needs to be explored further in future research.11
Table 6: Network characteristics
Variable Total Mean
Correlation with Inclusion Error Connection to chief* (CONNCHIEF) 201 0.28 0.02 Connection to vice** (CONNVICE) 32 0.08 0.01
*Only for 12 villages in the sample; **Only for 13 villages in the sample
Further, an important participatory element is the option for the households to file complaints regarding the final classifications. The field interviews revealed that the official way of filing complaints during the Village Consultation Meeting, was not used by the households. Yet, this does not necessarily mean that they did not complain. While none of the interview partners admitted complaining them-selves, many households had heard about others complaining to the chief or the commune council about the classification. Thus, the actual complaint procedure can be described as informal. This is in line with the notion that poverty classifications are not considered a matter that should be publicly discussed. Only few interviewees reported talking with relatives about the practical use of the card. Non-surprisingly, none of the households was able to tell how many other households of the village owned the card.
11 The corresponding regression table shows no significant results for these types of connections. It can be found
These observations show that households see the card as something desirable but also very private. They do get engaged and actively advocate for their rights and at the same time the decision makers consider their demands, at least to some extent. However, this does not happen in the transparent and public way intended by the program but rather on a private and personal level. Thus, good personal relations to the authorities (VRG members, chief or commune representatives) are certainly a necessary prerequisite in this case. Also, the actual number of complaints and share of reconsiderations cannot be tracked.
This obvious deviation from the official identification procedure may signal limited accordance of the procedure with the local value system in the sense of Platteau and Gaspart (2003). They argue that programs imposed by foreign donors are often driven by western values rather than local preferences and needs. As the IDPoor program was mainly initiated and developed by the German GIZ this might well be the case. A possible consequence could be low ownership and identification with the program and therefore increased risk of elite capture.
