Surviving the Killing Fields - EH.net

1 Surviving the Killing Fields The long-term consequences of the Khmer Rouge Mathias Iwanowsky and Andreas Madestam Stockholm University June 15, 2016. Barcelona GSE Summer Forum Advances in Micro Development Economics Motivation Legitimacy of and trust in government are important for state building Reservoir of loyalty : increases and enhances citizens'. cooperation and compliance with rules and regulations even without incentives and sanctions In their absence, governments have to spend more resources on monitoring and enforcement to induce compliance Despite agreement on importance of legitimacy and trust, we know less of their origins 2. Motivation Legitimacy and trust key when rebuilding post-war societies as state institutions are weak Civil war and genocide particularly damaging to trust in government as state representatives often participate in conflict Memory of state involvement may prevent citizens from conferring authority on the state in fear of an abusive authority What determines legitimacy and trust in general and in the government in war-torn societies?

2 What are the effects on the population at large beyond those directly affected by violence? 3. Empirical challenge Unclear whether experience of war causes changes in political beliefs and behavior Conflict often nationwide without credible counterfactuals Even if intensity of violence varies, (selective) targeting of specific regions based on prewar political views may confound measures of post-conflict beliefs and behavior Difficult to disentangle direct and indirect experiences of violence Due to lack of empirical evidence, open questions are 1. Does (indirect) exposure to conflict cause political change? 2. If so, what are the mechanisms? 4. Cambodia Investigate genocide in Cambodia during Khmer Rouge (KR), 1975 1978, to study causal effect of experience of war on political beliefs, behavior, and trust almost 4 decades later 3 million (20%) Cambodians killed 63% were separated from family members 30% observed torture, 22% killings At end of reign, regime killed large share of urban pop residing in labor camps creating Killing Fields throughout country Allows us to study how indirect exposure to war atrocities affected majority of its citizens as Killing Fields represent long-lasting trauma to nearby rural population 5.

3 Labor Camps and Killing Fields Figure 1: Cambodia's Killing Fields 6. What we do We first estimate whether the Killing Fields affected Political mobilization in last national election in 2013. Vote share for the long-ruling incumbent and opposition and turnout To understand our findings, we then estimate impact of Killing Fields on Measures of trust Political beliefs Knowledge of and interest in politics Community engagement Occupational choice Credit market behavior Investments in physical and human capital and public infrastructure 7. Why and how would (indirect) experience of atrocities and memory of Killing Fields matter today? 1. Witnessing atrocities of KR and being reminded of experience via Killing Fields breed mistrust in general and in the state, as represented by national gov't Direct measures of social preferences and trust Revealed-pref argument: if public institutions have low legitimacy, make investments that are less dependent on the state or contribute less to public good provision 2.

4 Change in population and social structure Systematic killings affect social and/or labor-land ratio ( Malthusian argument ). 8. Why and how would (indirect) experience of atrocities and memory of Killing Fields matter today? 3. Differential investments in public infrastructure Recent gov't provision of public goods affects legitimacy 4. Post-traumatic growth . Individual direct exposure to violence increases social cooperation and pro-social behavior, perhaps explained by increased prosociality toward in- over out-group members 9. Genocide April 17th 1975, KR win 5-year long civil war by capturing Phnom Penh Immediately after, population is evicted from urban areas - Phnom Penh: 2 million were forced to leave within two weeks - Used as labor on rice Fields Population of Cambodia classified into two groups 1. Base people: farmers and peasants in rural areas 2. New people: city evacuees and those with education New people were targeted and eventually killed 10.

5 Genocide - new' vs base' people Classification was easily observable New people - Evicted urban population, in particular educated and former government officials - Moved to compounds outside base people's villages Base people - Allowed to live in their own houses with basic rights - Limited interactions with new people but forced to watch beatings and killings - No planned extermination 11. Current political system Cambodians People's Party (CPP) in power since 1985. - CPP leader Hun Sen, a former KR, actively supported amnesty of KR cadres - Extensive cronyism and widespread corruption In response, Cambodia National Rescue Party (CNRP) unified all opposition parties to oust CPP in 2013 national election - Its leader, Sam Rainsy, faces charges for accusing MPs of collusion with KR. I not only weaken the Opposition, I'm going to make them dead .. and if anyone is strong enough to try to hold a demonstration, I will beat all those dogs and put them in a cage.

6 (Hun Sen, Jan 20, 2011 as a response to the Arab spring. Source: Human Rights Watch Report 2015). 12. Basic idea: correlation between severity of killings and 2013. national election (1) (2) (3) (4) (5) (6). Vote Share opposition Pr[CPP Win] Turnout log(Bodies) . ( ) ( ) ( ) ( ) ( ) ( ). Lat Lon polynomial Yes Yes Yes Yes Yes Yes Province FE Yes Yes Yes Yes Yes Yes Pre-KR commune controls No Yes No Yes No Yes Dependent variable mean N 1,569 1,569 1,569 1,569 1,569 1,569. Standard errors clustered by 24 provinces in parentheses. p < , p < , p < Areas with more people killed have higher turnout, favoring opposition OVB problem Positive bias: target opposition areas Negative bias: target areas supportive of regime 13. Identification strategy Rely on regime's desire to create an agricultural empire with rice production as the main staple and use of city population as forced labor Regime moved urban population to areas experiencing higher (temporal) agricultural productivity Use historic rainfall to generate exogenous variation in rice productivity and, hence, variation in size of camps and subsequent killings 14.

7 Data Cambodian Genocide Project (Yale, Ben Kiernan). Geocoding of 974,734 buried bodies in Cambodia US Army maps series L7016, from 1969 1972. Detailed maps covering Cambodia prior to the genocide Rainfall data at degrees from 1951 2007. Aphrodite Monsoon Asia Voting outcomes from the 2013 national election Individual-level survey outcomes on trust and political beliefs in 2003 and 2013, Asia Foundation Cambodian Socio-Economic Survey 1996 2014 (12 waves). Repeated cross-section with information on socio-economic outcomes, occupations, migration Population Census 1998/2008. 15. Data Figure 2: US Army map with commune characteristics and Killing Fields 16. Map accuracy Identification strategy Employ KR's plan to use forced labor to increase rice production Source: Chandler et al (1988), Pol Pot Plans the Future: Confidential Leadership Documents from Democratic Kampuchea, 1976-1977.

8 17. Identification strategy Rice is Cambodia's main crop. Use local rainfall shocks to predict variation in production across communes during KR. Yields are sensitive to excessive rain during harvest season 18. Identification strategy Assumptions Conditional on likelihood of shocks, whether a commune had a shock during harvest season 1975 1977 is orthogonal to political outcomes today Number of people killed approximates for size of site Intuition Absence of a shock increases production, the size of labor camps, and subsequent killings Standardized rainfall (1) Calculate mean c,p and standard deviation c,p in rain using 1951 2007 in each commune c and standardize rainfall during the KR 1975, 1976, and 1977. (2) Use within-province mean and distinguish between positive and negative rainfall realizations 19. Distribution of rainfall Figure 3: More and less productive communes during KR 20.

9 Main specification Commune-level regressions yc = Neg. Production Shockc + p + Xc + c Outcomes: people killed, voting, trust, political beliefs and knowledge, community investments, socio-economic and credit market measures Neg. Production Shock: A dummy variable (= 1) if there was a negative shock to production (and, hence, fewer killings). p , Xc : province FE and commune controls All regressions control flexibly for latitude and longitude with SEs clustered either at the province level or adjusted for spatial correlation using Conley at degrees 21. Exogeneity check Is rainfall during KR correlated with other determinants of the outcomes of interest? Commune characteristics prior to KR Mean Point estimate Standard error T-stat School in commune ( ) Church in commune ( ) Telephone in commune ( ) Commune office in commune ( ) Post office in commune ( ) log(population density) ( ) log(rice field area) ( ) log(inundation area) ( ) log(plantations area) ( ) log(commune area) ( ) log(distance to Phnom Penh) ( ) log(distance to road) ( ) log(distance to province capital) ( ) log(distance to border) ( ) log(km of roads in commune) ( ) log(km of rails in commune) ( ) 22.

10 Production shock and severity of killings (1) (2) (3) (4) (5) (6). log(Bodies) Bodies Neg. Production Shock during KR ( ) ( ) ( ) ( ) ( ) ( ) . [ ] [ ] [ ] [ ] [ ] [ ] . Neg. Production Shock 1972 1974 ( ) ( ). [ ] [ ]. Neg. Production Shock 1978 1980 ( ) ( ). [ ] [ ]. Lat Lon polynomial Yes Yes Yes Yes Yes Yes Province FE Yes Yes Yes Yes Yes Yes Commune controls No Yes Yes No Yes Yes Dependent variable mean N 1,569 1,569 1,569 1,569 1,569 1,569. Standard errors clustered by 24 provinces in parentheses. Conley standard errors in brackets. p < , p < , p < Continuous first stage Different dependent variable Robustness 23. Political mobilization (1) (2) (3) (4) (5) (6). Vote share CNRP Turnout Neg. Production Shock during KR ( ) ( ) ( ) ( ) ( ) ( ) . [ ] [ ] [ ] [ ] [ ] [ ] . Neg. Production Shock 1972 1974 ( ) ( ). [ ] [ ]. Neg. Production Shock 1978 1980 ( ) ( ). [ ] [ ]. Lat Lon polynomial Yes Yes Yes Yes Yes Yes Province FE Yes Yes Yes Yes Yes Yes Commune controls No Yes Yes No Yes Yes Dependent variable mean N 1,569 1,569 1,569 1,569 1,569 1,569.