Competition and Conflict: Evidence from a Community Driven Development Program in Indonesia. Preliminary draft

Transcription

1 Competition and Conflict: Evidence from a Community Driven Development Program in Indonesia Preliminary draft Teevrat Garg, Caterina Gennaioli, Stefania Lovo, Gregor Singer October 8, 2017 Abstract In this paper, we investigate the impact of increased local competition for public funds on conflict in Indonesia. We use a triple differences approach that exploits an exogenous source of variation in competition given by the number of villages that compete for sub-district level funds. We find a small but significant impact of competition on conflict. We find larger effects on local (within-village) and ethnic-type conflicts. We argue that competition within a community-driven development program helps increase local participation and group interaction and results in the strengthening of local conflict management capacity. 1 Introduction A growing body of work examines the role of local economic windfalls on institutional quality and conflict with the twin channels of political resource curse and dutch disease reinforcing the negative effects of such windfalls on conflict. We examine the world s largest community driven development program to examine the effects of introducing local competition in fiscal transfers and find that increased competition between groups reduces intra-group conflict but has no effect on inter-group conflict. The empirical strategy exploits the programme design that assigns a block grant to all participant sub-districts; villages then compete for funding against other villages in the sub-district. The number of villages in each sub-district, therefore, represents an exogenous variation in the intensity of competition. There is little variation in the We thank Sam Bazzi, Larry Chavis and Ben Olken for sharing Indonesian data. We thank John Voss for sharing helpful and detailed information about the Kecamantan Development Program. We thank seminar participants at LSE, UCSD, University of Reading, East Carolina University and Lafayette College and conference participants at the 2016 AAEA meetings and the 2016 AERE meetings. We acknowledge funding from ESRC Centre for Climate Change Economics and Policy and the Grantham Foundation for the Protection of the Environment. 1

2 size of the block grants, but considerable variation in the number of villages in a subdistrict. The identification strategy is based on comparing pre- and post-programme patterns of conflicts, conditional on the number of villages, between participant and non-participant sub-districts. The inspection of a long series of pre-reform conflicts shows that the common trends assumption underlying this triple differences approach holds in the data. Consistent with a model where inter-group competition increases intra-group cooperation, we find that competition increases local participation and the entire decline in conflict is explained by a reduction in ethnic conflicts. We find no evidence of negative spillovers in participation in existing local institutions. Our results suggest that competition in allocation of fiscal transfers not only improves efficiency but can also have positive unintended benefits. 2 The Kecamatan Development Program The Kecamatan Development Program (KDP, also Program Pengembangan Kecamatan, PPK) was a Government of Indonesia program aimed at reducing poverty in rural communities and improving local governance. This community driven program was not designed as a conflict reduction or management program. Facilitators were not trained nor were willing to engage in conflict resolution among villagers (Barron et al., 2006). It was one of the largest World Bank-financed community-driven development projects. The program provided a block grant to sub-districts (kecamantan) depending upon their population size. Sub-districts with a population below 25,000 received 750,000 rupiah (39,000 USD in 1998), those above the threshold received one billion rupiah (111,000 USD in 1998) 1. Each sub-district participates for three years and then rotates out of the program. This paper focuses on the first round of KDP that began in Sub-districts entering the first round of the program were the poorest as determined by the Ministry of Home Affairs (2002). By 2002, the end of the first round, 986 sub-districts (25% of all sub-districts) in 22 provinces received funding and more than 15,000 villages had participated in the programme. All sub-districts participating in the first round received the entire block grant while in subsequent rounds with the participation of relatively better off sub-districts the block grant was often reduced. While participation was non-random, the fact that all sub-districts received the maximum grant ensures that the number of villages within a sub-district is the only feature of the programme that constitute a source of variation in competition and in the probability of a village receiving funding across sub-districts. All villages in a participating sub-district can compete for the allocation of funding. Each village can submit up to 2 proposals for small-scale infrastructure, social and economic activities. If there are 2 proposals one needs to come from women. Each proposal needed to be matched by a contribution in labour and/or materials by the villagers. Village proposals can range between a minimum of Rps 35 million (4,375 USD) to a maximum of Rps 150 million (18,750 USD). Most of the proposals submitted to the first round of KDP had a focus on infrastructure, in particular roads and bridges while only few proposals focused on education and health (Ministry of Home Affairs, 2002). The KDP had a well structured and monitored activities cycle (up to 14 months) that started with socialisation and information dissemination, proceeded through plan- 1 In the more populous provinces of Java, the threshold was set to 50,000 people. 2

3 ning, proposal preparation and verification and ended with funding decisions and the actual implementation of the project. Village meetings were a crucial element of this process (see diagram in Figure 8 of the Appendix. In particular, three village-level meetings (musbangdes I, II an III) and one hamlet (group) level meeting (musbangdus) were at the core of the programme. The first village meeting was convened to publicize KDP and select village facilitators. This was followed by facilitated meetings at the hamlet level and of women s groups to develop proposal suggestions. Proposals were then taken to the second village meeting for a collective decision. Finally, at the third meeting villages find out whether their proposal was successful and discuss how to move forward. Often, the participation in the planning and decision making forums was the first occasion in which villagers from different identity groups had ever congregated with the purpose of engaging in decision making and take a collective action (Gibson and Woolcock, 2005). More importantly, whether people genuinely participate in the formulation of proposal ideas is a necessary condition for a project to pass the screening (Verification) stage and be allowed to be put forward at the sub-district level for the final project selection. Another crucial aspect of KDP is that not all villages submitting a proposal is guaranteed funding. However, as shown in Chavis (2010), when there are only a few villages in a sub-district, villages tend to simply divide the funds among themselves with little regards for the merits of the proposed projects. There is, in fact, some anecdotal evidence that the programme was sometimes perceived as proyek bagi-bagi duit 2 (cash hand out project) from the government to the poor (Nugraha, 2010) or that villagers were having only a passive role while village heads were the main decisionmakers (Evers, 2000). We argue that this is less likely to happen when a higher number of villages compete for funding. As the number of villages increases it becomes more difficult to achieve a mutual agreement and avoid competition, the equal share gets smaller, and the block grant becomes insufficient to finance all village projects (all funding requests need to be within the lower and upper bound requirements). Table 1 summaries data on participation and funding allocation by number of villages. Overall about 60% of all the villages in KDP sub-subdistricts participated by submitting a request for funding. The information disaggregated by number of villages at the sub-district level shows that while almost all villages in districts with a low number of villages participated in the programmed, only 37% of the villages participated in sub-districts that are at the other hand of the distribution. Similarly, the changes of winning are much higher in low-competition villages (91%) while are about 20% lower in sub-district with a very large number of villages. As the number of villages increases, the greater degree of competition is, therefore, reflected in both the lower percentage of participants (low are discouraged from participating) and the lower changes of receiving funding. A number of studies have investigated the impact of KDP on various local outcomes including conflicts. Barron et al. (2006), for example, examine how the KDP interacts with social tensions and local conflict, that is at the sub-district level and below. Using a mix of quantitative and qualitative assessment they find some but small positive direct impacts on conflict with larger effects in places where local conflict management institutions did not exist.they also find notable beneficial effects on the local institutions that have helped improve inter-group relations ( ethnic, religious and 2 Bagi-bagi means sharing or distribute and can assume different connotations. It can refer to the traditional Indonesian culture of sharing among people (bagi-bagi rejeki, sharing the good fortune) but also to the tendency during the Suharto s regime of extracting money from projects and investments and spreading the wealth among those with power (Gellert, 2015) 3

4 Table 1: Participation and allocation of funding in KDP Order Number of Share of village Share of Average amount Average amount Number of villages participating winners requested awarded sub-districts Overall Authors calculations based on data from Chavis (2010). Amount requested and awarded is in millions of Rupiahs. The share of winners is computed over the number of villages participating. class relations), and the relationship between citizens and the state that make conflict less likely to arise and/or to escalate. Similarly, Gibson and Woolcock (2005) also use a mixed methods approach and find that while KDP might induces conflict through the competitive bidding process, when well implemented, it can enable otherwise unequal groups to engage peacefully trough the promotion of a set of collaborative routines of conflict management that villagers can use to interface with more organized and influential actors. While these studies provide localized qualitative and quantitative evidence on the beneficial indirect effects of KDP on conflict resolution, there is a lack of a systematic programme-wide quantitative evaluation of its impact on conflicts. Chavis (2010) finds that greater competition among villages in KDP reduced cost of road-construction and increased community participation. Two possible reasons. Best projects get selected. Local governance improved. 3 Conceptual framework Development projects and conflict go hand in hand as development is an inherently political and contested process (Barron et al., 2006) Negative impacts on local conflict and conflict management capacity Resources into poor areas inter-group tensions New decision making rules impact on local power balances and structures Positive impacts Conflict management (direct effect) Collective decision-making process improved inter-group relations Inclusive decision-making process marginalized groups Change norms, attitudes, and expectations on how disputes should be resolved 4 Research Design In this section, we first describe our strategy to estimate the average treatment effect of KDP on conflict using a difference-in-differences approach. Second, we detail our triple-differences approach to estimate the effect of competition in KDP on conflict. 4

5 4.1 Difference-in-Differences We first estimate the overall effect of the KDP program on conflict. As discussed above, the KDP program was rolled out non-randomly at the sub-district level. To uncover the causal effect of the program, we first estimate a difference-in-differences regression comparing sub-districts treated with the KDP program with those excluded from the program before and after the program was introduced. Conflict st = β 1 KDP st + γ s + η t + ɛ st (1) Conflict st is the number of conflicts in subdistrict s in year t, KDP st is a binary indicator equal to 1 if sub-district s was treated with the KDP program in year t, β 1 is the coefficient of interest - the difference-in-differences estimate of the impact of KDP on conflict, and ɛ st is the error term. Standard errors are clustered at the district level. Conditional on sub-district fixed effects (γ s ) and time fixed effects (η t ), β 1 can be interpreted as causal if absent KDP, conflict in treated and untreated subdistricts would have followed similar trends. While this assumption is inherently untestable, we test for pre-program trends in conflict. Therefore interpreting β 1 as the causal effect of the KDP program relies on the weaker assumption that these pre-trends are a reasonable proxy for the counterfactual trends between treated and untreated subdistricts. In extensions, we include subdistrict specific linear and quadratic time trends, as well as province-year fixed effects to control for underlying trends within subdistricts and province-level year to year shocks respectively. 4.2 Triple Differences We then estimate the effect of KDP program conditional on then number of villages in a sub-district, a proxy for the level of competition. Conflict st = β 1 KDP st #numvillages + β 1 P ost#numvillages + γ s + η t + ɛ st (2) 5 Data KDP data are obtained from Chavis (2010). Regarding KDP I ( ) we have information on the name of the 986 districts that participated, their population and the total block grant received. We also have information at the village level on the amount requested, amount allocated and number and type of projects. We do not have information on meeting attendance. We recover the population of the non-participants using PODES Regarding KDP II we have information at the village level on meeting participation, amount requested, amount allocated and number and type of projects. Information on meeting participation is available only for a subsample of villages (about 6,000). At the time of the program Indonesia was a country in the midst of an ongoing and uneven democratic transition, a process that has, at times, been accompanied by violence. In addition to outbreaks of large-scale and violent communal conflict in a number of locations, and secessionist conflict in two provinces, widespread (and often violent) local conflict has occurred across the country. 5

6 Data on conflicts are obtained from the United Nations Support Facility for Indonesian Recovery (UNSFIR). The dataset covers 14 out of 28 provinces 3 and was compiled by a team of 14 researchers. It focuses only on collective violence and not individual violence or crime. It also does not include secessionist violence due to the inability to collect information in areas where a war of insurgency was in place (Varshney et al., 2004). Of the 3,608 incidents recorded 1,100 could not be attributed to a specific sub-districts and where therefore excluded. Table 3 shows a sharp increase in conflict starting the year preceding the fall of the Suharto regime in While about 3 percent of the sub-districts in the sample experienced at least one conflict, the incidence of conflict increased to almost 10 percent in the post-1998 period. This is true in both KDP and non-kdp subdistricts although the former are, on average, less affected by conflicts. Table 4 shows the number of conflict by type and year. Conflicts are categorized as across villages when the name of the village is not specified or when more than one village is indicated as the location of the conflicts. They also include conflicts where the specific cause of the violence is inter-village brawls. Another distinction by type of conflict (right panel) shows that the majority of incidents are categorized as other. These are composed mainly of conflict resulting from popular justice, inter-village and intra-groups fighting, santet (the killing of people allegedly practicing black magic) and political factions. Figure 1: Number of conflicts over time. Comparison between subdistricts with a low and high number of villages, in and out of KDP Conflicts / Conflicts in Less than 16 villages In KDP Out of KDP Conflicts / Conflicts in More than 16 villages In KDP Out of KDP Author s calculation based on the UNSFIR database. The figure plots the raw data on conflict for subdistrict in and out of KDP and with low or high number of villages (> 16) Figure 5 plots the raw data on conflict. It compares trends in conflict of polluting by number of villages in a sub-district by reporting the total number of conflicts in more or less than 16 villages normalized by 1990 values. The figure shows a fairly similar trend in the pre-kdp period between high and low competition sub-districts. On the other hand, for sub-districts participating in KDP, the pattern diverges after the 3 Riau, DKI Jakarta, Central Java, West Java, East Java, Banten, Central Kalimantan, West Kalimantan, South Sulawesi, Central Sulawesi, East Nusatenggara, West Nusatenggara, Maluku, and North Maluku. 6

7 implementation of the programme. In particular, while conflicts in low-competition sub-districts followed on an upward trend, we observe a relative slowdown in conflicts in high-competition sub-districts. 6 Results 6.1 The Effect of KDP on Conflict We show the results of estimating of equation (1) in the form on an event study in figure 2. Related regressions tables are reported in the Appendix and show both linear and poisson estimates. Figure 2: Difference in differences esimates: impact of partecipation in KDP on conflict Author s calculation based on the UNSFIR database. The plots are created by regressing the number of conflicts by sub-district on a full set of event time indicators interacted with a dummy indicating the participation in KDP controlling for sub-district and year fixed effects. The dashed line indicates 90% confidence interval. Results refer to 14 out of 28 provinces. 6.2 The impact of competition on participation In this section we first investigate whether higher competition is correlated with greater participation of villagers to the KDP programme. Table 2 shows a significant positive correlation between the number of villages and meeting attendance in the three villagelevel meetings organised at different stages of the KDP process. This finding is crucial to support our hypothesis that greater participation was key in reducing the chances of conflict. While KDP was not designed as a conflict reduction or management program, by implementing the principles of participation and local choice in a competitive selection process, helped villagers to acquire civic skills and decision-making opportunities that were important for mediating conflicts (Barron et al., 2006). Figure 3 shows non-linear effects of village numbers on participation. At low levels, an increase in the number of villages has a much larger effect on participation. The impact achieves a maximum at about 50 villages and then diminishes. This is in line 7

8 Table 2: Impact of competition (number of villages) on participation to KDP meetings Dependent var: Attendance (%) Musbangdes I Musbangdus (Hamlets) Musbangdes II (1) (2) (3) (4) (5) (6) Num villages (log) 3.225*** 1.035** 0.451** 0.572* 4.332*** 3.453** (0.417) (0.469) (0.190) (0.307) (0.977) (1.431) Population *** *** *** *** *** *** (0.256) (0.428) (0.217) (0.285) (0.660) (1.082) District FE No Yes No Yes No Yes Observations Villages Average The dependent variable is the log of attendance at the village level (meeting participants over population). Standard errors clustered at subdistrict level in parenthesis. with our hypothesis that when there are only a few villages in a sub-district, villages tend to simply divide the funds among themselves with little regards for the merits of the proposed projects, as described above. This result shows that this is less likely to happen when a higher number of villages compete for funding. As the number of villages increases it becomes more difficult to achieve a mutual agreement and avoid competition, the equal share gets smaller, and the block grant becomes insufficient to finance all village projects. Figure 3: Number of villages and meeting attendance Attendance Number of villages The figure plots the results of a regression of number of villages on participation with cubic splines. The graph refers to the first village meeting (musbangdes I). Similar results are obtained with other types of village meetings 6.3 The impact of competition in KDP on conflict In this section we report the results of estimating equation 2 in the form of an event study with heterogenous effects (regression results are shown in the Appendix). Figure 4 is derived by allowing the triple interaction term to vary by year. The graph plots the coefficients of the triple interaction terms and confirm the validity of the empirical design showing no evidence of differential effects of the number of villages prior to 8

9 1998. It also shows that after the implementation of KDP a higher number of villages has led to a decrease in conflict for participating sub-districts. Figure 4: KDP versus non-kdp sub-districts: impact of the number of villages on conflicts over time Author s calculation based on the UNSFIR database. The plots are created by regressing the number of conflicts by sub-district on a full set of event time indicators interacted with the number of villages (in log) and a dummy indicating the participation in KDP controlling for sub-district and year fixed effects. The dashed line indicates 95% confidence interval. Results refer to 14 out of 28 provinces. Figure 5 show the effects separately for within-village conflicts and cross-village conflicts (conflicts involving more than one village). It shows that competition has a positive effect in reducing villages internal conflict. The KDP s model of decisionmaking is based upon the principle of open competition. Such competition is a potential source of conflict. Different groups within villages put forward proposals. Groups use the traditional tools of coalition building (the powers of rhetoric, appeals to reason, bargaining and deal-making) to win support for their proposals (Barron et al., 2006). The KDP, therefore, explicitly aims to create non-violent conflict through stimulating competition. This potential source of conflict, however, does not seem to be related to the level of competition we investigate. Since the proposals are limited to 2 and are capped at 18,750 USD competition between groups within a village will exist independently of the level of competition among villages. If, instead, KDP did indeed promote conflicts between villages in areas where tensions were already high, this would anyway bias our estimates downwards. The could also partially explain while the effect on cross-village conflicts is much lower although it can be more likely attributed to the fact that KDP was aimed mainly at building civic skills and participation within a village rather than promoting intra-village cooperation. We should also not exclude the possibility that KDP could lead to an increase in conflict. According to Barron et al. (2006) competition-led conflicts are generally more pronounced between villages rather than within a village. This is because there is more pressure on village heads to deliver resources and services than there is for groups within villages to have their proposals forwarded on to the sub-district level. Our results, however show no increase in inter-village conflicts at higher levels of competition. We then explore the effects of competition by type of conflicts. Findings are 9

10 Figure 5: Event study by extent of conflict (a) Conflicts within village (b) Conflicts across villages Author s calculation based on the UNSFIR database. The plots are created by regressing the number of conflicts by sub-district on a full set of event time indicators interacted with the number of villages (in log) and a dummy indicating the participation in KDP controlling for sub-district and year fixed effects. The dashed line indicates 95% confidence interval. Results refer to 14 out of 28 provinces. presented in the form of an event study in Figure 6, regression results are reported in the Appendix. Figure 6 shows that the effect of competition is more pronounced for ethnically motivated conflicts. This is in line with Barron et al. (2006) who find that the KDP has helped improve ethnic and religious relations since KDP provides a space for different groups to come together to collectively discuss their needs and priorities, an opportunity that is rarely afforded them elsewhere. KDP also facilitates group interaction by improving transportation networks. 7 Robustness We conduct a set of robustness checks including controlling for the interaction of a post-kdp dummy with initial sub-district level characteristics that could be potentially correlated with the number of villages such as population and ethnic diversity. We then construct a placebo test treatment by comparing sub-districts that experienced at least one conflict before 1998 (20%, the omitted baseline) with other sub-districts.the results show that the number of villages did not produce a relative decrease in conflict after 1998 when sub-districts with no conflict are compared to those with at least on conflict. We, finally, also run our regressions at the village level, here we are considering only within village conflict only, and find similar results. 10

11 Figure 6: Event study by type of conflict (a) Economy (b) Ethnic (c) State (d) Other Author s calculation based on the UNSFIR database. The plots are created by regressing the number of conflicts by sub-district on a full set of event time indicators interacted with the number of villages (in log) and a dummy indicating the participation in KDP controlling for sub-district and year fixed effects. The dashed line indicates 95% confidence interval. Results refer to 14 out of 28 provinces. 11

12 References Barron, P., R. Diprose, and M. J. Woolcock (2006). Local conflict and community development in Indonesia: assessing the impact of the Kecamatan development program. Number 10. Decentralization Support Facility. Chavis, L. (2010). Decentralizing development: Allocating public goods via competition. Journal of Development Economics 93 (2), Evers, P. J. (2000). Resourceful villagers, powerless communities: Rural village government in indonesia. A World Bank Bappenas Research Project. Gellert, P. K. (2015). Optimism and education: The new ideology of development in indonesia. Journal of Contemporary Asia 45 (3), Gibson, C. and M. Woolcock (2005). Empowerment and local level conflict mediation in indonesia: a comparative analysis of concepts, measures, and project efficacy. Ministry of Home Affairs, C. D. A. (2002). Kecamatan development program. Phase one: final report. Technical report, KDP National Secretariat adn National Management Consultant., Jakarta. Nugraha, A. (2010). Community-driven development, collective action and empowerment: Effective communication matters (a lesson from the kdp and the upp in indonesia). The Indonesian Journal Of Communication Studies (IJCS) 3 (2). Varshney, A., M. Z. Tadjoeddin, and R. Panggabean (2004). Patterns of Collective Violence in Indonesia, UNSFIR. 8 Appendix 12

13 Figure 7: KDP activity cycle Based on Ministry of Home Affairs (2002). A kecamatan is a sub-district. 13

14 Table 3: Number and shares of conflicts by year and participation in KDP year % of sub-districts with conflict Total number of conflicts All Out of KDP In KDP All Out of KDP In KDP Total Before After Authors calculations based on UNSFIR dataset. Table 4: Number of conflicts by type year Within Across Economy Ethnic State Other village villages Total Before After Authors calculations based on UNSFIR dataset. 14

15 Table 5: Correlation between conflict and participation Dep. var: Total Economy Ethnic State Other Post # Attendance ** *** (0.014) (0.005) (0.007) (0.003) (0.010) Number of new districts 0.046** *** (0.019) (0.003) (0.017) (0.002) (0.007) Year FE Yes Yes Yes Yes Yes Sub-district FE Yes Yes Yes Yes Yes Observations Sub-district Standard errors clustered at district level in parenthesis. Includes only KDP sub-districts that could be matched to meetings data. Column header indicates type of conflicts. Data refers to 14 out of 28 provinces. Table 6: DiD regressions: ols (1) (2) (3) (4) (5) (6) Dep. var: Total Economy Ethnic State Other Deaths Post # in KDP *** ** * (0.016) (0.003) (0.012) (0.003) (0.008) (0.166) Year FE Yes Yes Yes Yes Yes Yes Sub-district FE Yes Yes Yes Yes Yes Yes Observations Sub-districts Standard errors clustered at district level in parenthesis. Column header indicates type of conflicts. Data refers to 14 out of 28 provinces. All specifications controls for the number of new districts (at district level) due to district splitting. Figure 8: Difference in differences esimates: impact of partecipation in KDP on conflict Author s calculation based on the UNSFIR database. The plots are created by regressing the number of conflicts by sub-district on a full set of event time indicators interacted with a dummy indicating the participation in KDP controlling for sub-district and year fixed effects. The dashed line indicates 90% confidence interval. Results refer to 14 out of 28 provinces. 15

16 Table 7: DiD regressions: poisson (1) (2) (3) (4) (5) (6) Dep. var: Total Economy Ethnic State Other Deaths Post # in KDP (0.147) (0.313) (0.584) (0.391) (0.151) (0.603) Year FE Yes Yes Yes Yes Yes Yes Sub-district FE Yes Yes Yes Yes Yes Yes Observations Sub-districts Standard errors clustered at district level in parenthesis. Column header indicates type of conflicts. Data refers to 14 out of 28 provinces. All specifications controls for the number of new districts (at district level) due to district splitting. Table 8: Impact of competition (number of villages) before and after KDP: ols Dep. var: Total Economy Ethnic State Other Deaths Total Economy Ethnic State Other Deaths Post # in KDP 0.160* * ** (0.084) (0.020) (0.060) (0.017) (0.044) (1.058) Post # number of villages (log) 0.100*** *** 0.007* 0.032*** 0.973*** (0.028) (0.004) (0.024) (0.004) (0.009) (0.373) Post # in KDP # number of villages (log) ** * ** (0.036) (0.007) (0.027) (0.006) (0.016) (0.466) Sub-district FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Observations Sub-districts Standard errors clustered at district level in parenthesis. Number of villages is in log. Column header indicates type of conflicts. Data refers to 14 out of 28 provinces. Table 9: Impact of competition (number of villages) before and after KDP: Poisson Dep. var: Total Economy Ethnic State Other Deaths Total Economy Ethnic State Other Deaths Post # in KDP 1.669* ** *** (0.865) (2.007) (2.216) (1.996) (0.967) (3.441) Post # number of villages (log) 0.571*** *** 0.664* 0.346** 3.061*** (0.174) (0.320) (0.546) (0.378) (0.156) (1.169) Post # in KDP # number of villages (log) ** ** *** (0.315) (0.728) (0.749) (0.733) (0.352) (1.190) Sub-district FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Observations Sub-districts Robust standard errors in parenthesis. Number of villages is in log. Column header indicates type of conflicts. Last column refers to total number of deaths. Data refers to 14 out of 28 provinces. 16

17 Table 10: Impact of competition (number of villages) before and after KDP: Poisson Dep. var: Total Economy Ethnic State Other Post 1998 # in KDP ** * (0.981) (1.874) (2.244) (2.035) (1.003) Post 1998 # number of villages 0.538** *** 0.608* (0.226) (0.371) (0.538) (0.359) (0.203) Post 1998 # in KDP # number of villages * ** * (0.362) (0.680) (0.755) (0.747) (0.370) Number of district split (0.032) (0.133) (0.113) (0.147) (0.057) Sub-district FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Observations Sub-districts Standard errors clustered at district level in parenthesis. Column header indicates type of conflicts. Data refers to 14 out of 28 provinces. 17

18 Table 11: Additional controls with village level ethnic diversity (1) (2) (3) (4) (5) (6) Post 1998 # in KDP 0.196** 0.235*** 0.333*** 0.350*** 0.340** 0.585** (0.090) (0.087) (0.118) (0.124) (0.145) (0.231) Post 1998 # number of villages (log) 0.090*** 0.088*** 0.115*** 0.109*** 0.129*** 0.043** (0.027) (0.026) (0.035) (0.033) (0.048) (0.020) Post 1998 # in KDP # number of villages (log) ** *** *** *** *** *** (0.035) (0.033) (0.041) (0.039) (0.052) (0.088) Post 1998 # population 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Post 1998 # in KDP # population (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Number of new districts (0.031) (0.035) (0.035) (0.032) (0.045) Post 1998 # new districts (0.036) (0.040) (0.040) (0.043) (0.052) Post 1998 # in KDP #new districts (0.026) (0.027) (0.025) (0.033) (0.034) Post 1998 # Ethnic diversity (district) 0.211** (0.084) Post 1998 # in KDP # Ethnic diversity (district) * (0.092) Post 1998 # Ethnic groups (district) 0.020*** (0.008) (0.005) Post 1998 # in KDP # Ethnic groups (district) (0.009) (0.011) Post 1998 # Ethnic diversity (village) (0.184) Post 1998 # in KDP # Ethnic diversity (village) (0.198) Subdistricts FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Year-province FE No No No No No Yes Observations Subdistricts Standard errors clustered at district level in parenthesis. r Data refers to 14 out of 28 provinces. 18