Surviving the Killing Fields The long-term consequences of the Khmer Rouge Mathias Iwanowsky and Andreas Madestam Stockholm University June , Barcelona GSE Summer Forum Advances in Micro Development Economics
Surviving the Killing FieldsThe long-term consequences of the Khmer Rouge
Mathias Iwanowsky and Andreas MadestamStockholm UniversityJune 15, 2016
Barcelona GSE Summer ForumAdvances in Micro Development Economics
Motivation
• Legitimacy of and trust in government are important for statebuilding
• “Reservoir of loyalty”: increases and enhances citizens’cooperation and compliance with rules and regulations evenwithout incentives and sanctions
• In their absence, governments have to spend more resources onmonitoring and enforcement to induce compliance
• Despite agreement on importance of legitimacy and trust, weknow less of their origins
2
Motivation
• Legitimacy and trust key when rebuilding post-war societies asstate institutions are weak
• Civil war and genocide particularly damaging to trust ingovernment as state representatives often participate inconflict
• Memory of state involvement may prevent citizens fromconferring 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?⇒ 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 politicalbeliefs 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 measuresof post-conflict beliefs and behavior
• Difficult to disentangle direct and indirect experiences of violence
• Due to lack of empirical evidence, open questions are1. 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 onpolitical beliefs, behavior, and trust almost 4 decades later
• 1.5–3 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 residingin labor camps creating Killing Fields throughout country
• Allows us to study how indirect exposure to war atrocitiesaffected majority of its citizens as Killing Fields representlong-lasting trauma to nearby rural population
5
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 KillingFields 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
infrastructure7
Why and how would (indirect) experience of atrocities andmemory of Killing Fields matter today?
1. Witnessing atrocities of KR and being reminded of experiencevia Killing Fields breed mistrust in general and in the state, asrepresented 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 thestate or contribute less to public good provision
2. 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 andmemory 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 byincreased prosociality toward in- over out-group members
9
Genocide
April 17th 1975, KR win 5-year long civil war by capturing PhnomPenh
• 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 groups1. Base people: farmers and peasants in rural areas2. New people: city evacuees and those with education
• New people were targeted and eventually killed
10
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 allopposition parties to oust CPP in 2013 national election
- Its leader, Sam Rainsy, faces charges for accusing MPs of collusionwith 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”
(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 2013national election
(1) (2) (3) (4) (5) (6)Vote Share opposition Pr[CPP Win] Turnout
log(Bodies) 1.306∗∗ 1.411∗∗∗ −0.025 −0.034∗∗ 0.001 0.008∗
(0.483) (0.311) (0.016) (0.013) (0.006) (0.004)
Lat × Lon polynomial Yes Yes Yes Yes Yes YesProvince FE Yes Yes Yes Yes Yes YesPre-KR commune controls No Yes No Yes No YesDependent variable mean 40.16 40.16 0.61 0.61 0.79 0.79N 1,569 1,569 1,569 1,569 1,569 1,569
Standard errors clustered by 24 provinces in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
• Areas with more people killed have higher turnout, favoringopposition
• 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 withrice production as the main staple and use of city populationas forced labor
• Regime moved urban population to areas experiencing higher(temporal) agricultural productivity
• Use historic rainfall to generate exogenous variation in riceproductivity and, hence, variation in size of camps andsubsequent killings
14
Data
• Cambodian Genocide Project (Yale, Ben Kiernan)Geocoding of 974,734 buried bodies in Cambodia
• US Army maps series L7016, from 1969–1972Detailed maps covering Cambodia prior to the genocide
• Rainfall data at 0.25×0.25 degrees from 1951–2007Aphrodite Monsoon Asia http://www.chikyu.ac.jp/precip/
• 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-economicoutcomes, occupations, migration
• Population Census 1998/2008
15
Data
Figure 2: US Army map with commune characteristics and Killing Fields
Map accuracy16
Identification strategy
• Employ KR’s plan to use forced labor to increase riceproduction
Source: Chandler et al (1988), Pol Pot Plans the Future: ConfidentialLeadership Documents from Democratic Kampuchea, 1976-1977
17
Identification strategy
• Rice is Cambodia’s main crop. Use local rainfall shocks topredict 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 politicaloutcomes 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 µc,p and standard deviation σc,p in rain using
1951 – 2007 in each commune c and standardize rainfall duringthe KR 1975, 1976, and 1977
(2) Use within-province mean and distinguish between positiveand negative rainfall realizations
19
Distribution of rainfall
Figure 3: More and less productive communes during KR 20
Main specification
• Commune-level regressions
yc = βNeg. Production Shockc + µp + γXc + εc
• Outcomes: people killed, voting, trust, political beliefs andknowledge, community investments, socio-economic and creditmarket measures
• Neg. Production Shock: A dummy variable (= 1) if there was anegative shock to production (and, hence, fewer killings)
• µp, Xc: province FE and commune controls
• All regressions control flexibly for latitude and longitude with SEsclustered either at the province level or adjusted for spatialcorrelation using Conley at 1.5 degrees
21
Exogeneity check
• Is rainfall during KR correlated with other determinants of theoutcomes of interest?
Commune characteristics prior to KR Mean Point estimate Standard error T-statSchool in commune 0.709 −0.030 (0.040) −0.74Church in commune 0.035 0.011 (0.012) 0.90Telephone in commune 0.005 0.002 (0.004) 0.54Commune office in commune 0.396 0.026 (0.040) 0.65Post office in commune 0.017 0.003 (0.012) 0.23log(population density) 1.542 −0.227 (0.231) −0.98log(rice field area) 2.22 −0.092 (0.121) −0.76log(inundation area) 0.929 0.003 (0.109) 0.03log(plantations area) 0.128 0.099 (0.073) 1.35log(commune area) 3.818 0.224 (0.156) 1.43log(distance to Phnom Penh) 4.479 0.071 (0.089) 0.80log(distance to road) 0.531 −0.032 (0.087) −0.36log(distance to province capital) 2.588 0.147 (0.165) 0.89log(distance to border) 3.682 −0.102 (0.075) −1.36log(km of roads in commune) 1.844 −0.055 (0.097) −0.57log(km of rails in commune) 0.193 −0.057 (0.048) −1.18
22
Production shock and severity of killings
(1) (2) (3) (4) (5) (6)log(Bodies) Bodies
Neg. Production Shock during KR −0.046 −0.057 −0.057 −0.444 −0.478 −0.496(0.019)∗∗ (0.018)∗∗∗ (0.018)∗∗ (0.181)∗∗ (0.175)∗∗ (0.181)∗∗[0.018]∗∗ [0.019]∗∗∗ [0.021]∗∗ [0.140]∗∗∗ [0.150]∗∗∗ [0.170]∗∗∗
Neg. Production Shock 1972–1974 0.004 0.013(0.024) (0.135)[0.020] [0.123]
Neg. Production Shock 1978–1980 −0.012 0.277(0.036) (0.243)[0.026] [0.188]
Lat × Lon polynomial Yes Yes Yes Yes Yes YesProvince FE Yes Yes Yes Yes Yes YesCommune controls No Yes Yes No Yes YesDependent variable mean 0.141 0.141 0.141 0.621 0.621 0.621N 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 < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
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 −5.328 −4.810 −4.814 −0.036 −0.032 −0.032(1.183)∗∗∗ (0.993)∗∗∗ (1.009)∗∗∗ (0.014)∗∗ (0.011)∗∗ (0.011)∗∗[0.941]∗∗∗ [0.825]∗∗∗ [0.911]∗∗∗ [0.011]∗∗∗ [0.009]∗∗∗ [0.010]∗∗∗
Neg. Production Shock 1972–1974 −0.327 −0.010(1.289) (0.012)[1.256] [0.009]
Neg. Production Shock 1978–1980 0.881 0.012(1.447) (0.011)[1.226] [0.012]
Lat × Lon polynomial Yes Yes Yes Yes Yes YesProvince FE Yes Yes Yes Yes Yes YesCommune controls No Yes Yes No Yes YesDependent variable mean 40.160 40.160 40.160 0.791 0.791 0.791N 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 < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
24
Channels
How can we understand these differences?
• Indirect experience and memory of atrocities breed persistentmistrust toward an abusive state, as represented by today’sruling incumbent, and society in general
• People get engaged to express discontent with authority, not toconfirm status quo
Other possible explanations
• Increase in pro-social behavior in general, possibly driven bydifferential feelings toward the local over central/state
• Differential investments in public infrastructure• Change in population and social structure• Differential migration
25
Trust
(1) (2) (3) (4)Trust in neighbor Can influence government
Neg. Production Shock during KR 0.090 0.151 0.159 0.059(0.042)∗∗ (0.035)∗∗∗ (0.077)∗ (0.098)[0.031]∗∗ [0.040]∗∗∗ [0.050]∗∗ [0.051]
Lat × Lon polynomial Yes Yes Yes YesProvince FE Yes Yes Yes YesIndividual controls Yes Yes Yes YesAlive in 1978 No Yes No YesDependent variable mean 0.480 0.462 2.731 2.660N 991 450 1,849 1,199
All columns feature shock realization ∈ {0, 1}, shown in Figure (2) as main independent variable. Lat × Lonpolynomial: Latitude, Latitude2, Longitude, Longitude2, and Latitude × Longitude and province fixed effects.Individual controls are age, age2 and male. SEs clustered by 24 provinces in parentheses. ∗ p < 0.10, ∗∗ p < 0.05,∗∗∗ p < 0.01
26
Political beliefs
(1) (2) (3) (4) (5) (6) (7) (8)Has voted Government has a paternal role Supports political comp. Voting makes a difference
Neg. Production Shock during KR −0.020 −0.016 0.114 0.229 −0.049 −0.057 0.101 0.093(0.015) (0.009)∗ (0.028)∗∗∗ (0.089)∗∗ (0.022)∗∗ (0.023)∗∗ (0.048)∗∗ (0.050)∗[0.014] [0.007]∗∗ [0.028]∗∗∗ [0.084]∗∗ [0.022]∗∗ [0.020]∗∗ [0.046]∗∗ [0.043]∗∗
Lat × Lon polynomial Yes Yes Yes Yes Yes Yes Yes YesProvince FE Yes Yes Yes Yes Yes Yes Yes YesIndividual controls Yes Yes Yes Yes Yes Yes Yes YesAlive in 1978 No Yes No Yes No Yes No YesDependent variable mean 0.921 0.968 0.610 0.581 0.883 0.861 0.428 0.456N 1,963 1,294 991 450 1,963 1,294 1,484 906
All columns feature shock realization ∈ {0, 1}, shown in Figure (2) as main independent variable. Lat × Lon polynomial: Latitude, Latitude2, Longitude, Longitude2, and Latitude × Longitudeand province fixed effects. Individual controls are age, age2 and male. SEs clustered by 24 provinces in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
27
Political knowledge
(1) (2) (3) (4)Knows that voting is to elect Parliament Knows election date
Neg. Production Shock during KR −0.101 −0.100 −0.057 −0.081(0.040)∗∗ (0.037)∗∗ (0.038) (0.046)∗[0.036]∗∗ [0.040]∗∗ [0.040] [0.050]∗
Lat × Lon polynomial Yes Yes Yes YesProvince FE Yes Yes Yes YesIndividual controls Yes Yes Yes YesAlive in 1978 No Yes No YesDependent variable mean 0.682 0.579 0.477 0.489N 972 844 972 844
All columns feature shock realization ∈ {0, 1}, shown in Figure (2) as main independent variable. Lat × Lon polynomial: Latitude,Latitude2, Longitude, Longitude2, and Latitude× Longitude and province fixed effects. Individual controls are age, age2 and male.SEs clustered by 24 provinces in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
• Findings on trust, beliefs, and political knowledge consistentwith KR experience breeding general mistrust and demand forpolitical alternatives to today’s ruling incumbent
28
Public good provision, credit market behavior, andoccupational choice
To corroborate findings on trust, beliefs, and knowledge investigateindirect evidence of low general trust and lacking state legitimacy
• Low general trust• Communities less likely to solve collective action problems -
fewer community fisheries and community forest managementschemes
• Individuals less reliant on market transactions that build on trust- less informal lending from friends, family, moneylenders, andlandlords and more anonymous formal transactions
• Lacking legitimacy of public institutions• Make investments that are less dependent on the state (fewer
asset-specific investment that relies on gov’t) - less likely to workfor the government
29
Public good provision
(1) (2) (3) (4)Commune has a Commune has a community
community fishery forest management schemeNeg. Production Shock during KR 0.046 0.047 0.044 0.033
(0.019)∗∗ (0.020)∗∗ (0.019)∗∗ (0.019)[0.018]∗∗ [0.018]∗∗ [0.016]∗∗ [0.014]∗∗
Lat × Lon polynomial Yes Yes Yes YesProvince FE Yes Yes Yes YesCommune controls No Yes No YesDependent variable mean 0.059 0.059 0.079 0.079N 1,564 1,564 1,564 1,564
All columns feature shock realization ∈ {0, 1}, shown in Figure (2) as main independent variable. Lat × Lonpolynomial: Latitude, Latitude2, Longitude, Longitude2, and Latitude × Longitude and province fixed effects.SEs clustered by 24 provinces in parentheses. Conley SEs in brackets. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
30
Credit market behavior
(1) (2) (3) (4) (5) (6) (7) (8)Has any debt log(Size of debt, KHR) Informal lending Formal lending
Neg. Production Shock during KR 0.004 −0.000 0.009 0.033 0.033∗∗ 0.034∗∗ −0.039∗∗∗ −0.038∗∗∗
(0.009) (0.007) (0.055) (0.055) (0.013) (0.014) (0.012) (0.013)
Lat × Lon polynomial Yes Yes Yes Yes Yes Yes Yes YesProvince FE Yes Yes Yes Yes Yes Yes Yes YesCommune controls Yes Yes Yes Yes Yes Yes Yes YesIndividual controls Yes Yes Yes Yes Yes Yes Yes YesAlive in 1978 No Yes No Yes No Yes No YesDependent variable mean 0.277 0.247 13.225 13.154 0.552 0.564 0.443 0.427N 273,456 95,016 100,857 32,474 105,055 33,895 105,055 33,895
All columns feature shock realization ∈ {0, 1}, shown in Figure (2) as main independent variable. All columns include controls shown in Table 2 a Lat × Lonpolynomial: Latitude, Latitude2, Longitude, Longitude2, and Latitude × Longitude and province fixed effects. Individual controls are age, age2 and male. SEsclustered by 24 provinces in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
31
Occupational choice
(1) (2) (3) (4) (5) (6)State employed Private or self employed
Neg. Production Shock during KR 0.007∗∗ 0.007∗ 0.014∗∗ −0.006 −0.007 −0.032∗∗(0.003) (0.004) (0.005) (0.006) (0.007) (0.014)
Lat × Lon polynomial Yes Yes Yes Yes Yes YesProvince FE Yes Yes Yes Yes Yes YesCommune controls Yes Yes Yes Yes Yes YesIndividual controls Yes Yes Yes Yes Yes YesAlive in 1978 No Yes Yes No Yes YesNever migrated No No Yes No No YesDependent variable mean 0.060 0.092 0.044 0.834 0.853 0.894N 183,556 94,332 18,653 183,556 94,332 18,653
All columns feature shock realization ∈ {0, 1}, shown in Figure (2) as main independent variable. Lat × Lon polynomial: Latitude,Latitude2, Longitude, Longitude2, and Latitude × Longitude and province fixed effects. Commune controls are shown in table 2.Individual controls are age, age2 and male. SEs clustered by 24 provinces in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
32
Alternative explanations: migration
(1) (2) (3) (4)In-migration
All Alive in 1978Neg. Production Shock during KR 0.017 0.025 0.016 0.027
(0.022) (0.016) (0.033) (0.028)
Lat × Lon polynomial Yes Yes Yes YesProvince FE Yes Yes Yes YesCommune controls Yes Yes Yes YesIndividual controls No Yes No YesDependent variable mean 0.322 0.322 0.564 0.564N 199,501 199,501 79,931 79,931
All columns feature shock realization ∈ {0, 1}, shown in Figure (2) as main independent variable. Lat× Lon polynomial: Latitude, Latitude2, Longitude, Longitude2, and Latitude× Longitude and provincefixed effects. Commune controls are shown in table 2. Individual controls are age, age2, and male. SEsclustered by 24 provinces in parentheses. Conley standard errors at 150km in brackets.∗ p < 0.10, ∗∗p < 0.05, ∗∗∗ p < 0.01
33
Alternative explanations: population
(1) (2) (3) (4) (5) (6) (7) (8)
Age distribution: mean age in group
0–19 20–39 40-59 ≥ 60Neg. Production Shock during KR −0.021 0.022 −0.069 −0.083 −0.123 −0.119 −0.045 −0.044
(0.055) (0.048) (0.094) (0.094) (0.086) (0.094) (0.213) (0.213)
Lat × Lon polynomial Yes Yes Yes Yes Yes Yes Yes YesProvince FE Yes Yes Yes Yes Yes Yes Yes YesCommune controls No Yes No Yes No Yes No YesN 1,337 1,337 1,337 1,337 1,337 1,337 1,337 1,337
All columns feature shock realization ∈ {0, 1}, shown in Figure (2) as main independent variable. Lat× Lon polynomial: Latitude, Latitude2, Longitude, Longitude2,and Latitude × Longitude and province fixed effects. SEs clustered by 24 provinces in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
34
Alternative explanations: investment in physical and humancapital and infrastructure
(1) (2) (3) (4)log(Size farm) log(Value farm) log(p.c. consumption) log(Floor size)
Neg. Production Shock during KR 0.075 −0.021 −0.005 0.015(0.136) (0.267) (0.018) (0.016)
Individual controls Yes Yes Yes YesN 323,160 323,160 366,013 296,265Dependent variable mean 5.420 9.316 11.413 3.629
Years of education Can read Can write log(Distance to primary school)Neg. Production Shock during KR −0.103 −0.007 −0.010 −0.012
(0.064) (0.009) (0.009) (0.017)Individual controls Yes Yes Yes NoN 334,832 248,350 248,338 1,561Dependent variable mean 4.380 0.726 0.697 0.764
log(Distance to market) # Markets Share of HH with elec. Access to public elec.Neg. Production Shock during KR −0.029 −0.021 0.055 −0.013
(0.039) (0.040) (1.363) (0.022)Individual controls No No No NoN 1,561 1,564 1,136 1,019Dependent variable mean 1.555 1.211 19.951 0.354
All columns feature shock realization ∈ {0, 1}, shown in Figure (2) as main independent variable. All columns include controls shown in Table 2 a Lat × Lon polynomial:Latitude, Latitude2, Longitude, Longitude2, and Latitude × Longitude and province fixed effects. Individual controls are age, age2 and male. SEs clustered by 24 provincesin parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
35
Summary
• Communes with more people killed/more severe persecutionsduring KR
• Higher turnout in 2013 elections, favoring main opposition party• Turnout and greater political knowledge as a response to limit
current gov’t rather than evidence of pro-social behavior• Less public good provision and fewer investments relying on gov’t
or local trust
• Persistent and long-lasting impact• While some effects are stronger for individuals that experienced
KR, intergenerational memory of Killing Fields also matter
36
Additional results
Data
Figure 4: US Army map with commune characteristics and Killing Fields38
Data
Figure 5: Satellite picture 1975Back 39
Placebos
• Running placebos using rainfall in any given three-year periodduring 1951–2007
• If rainfall during KR has a causal effect, point estimate is anoutlier in placebo distribution
Back
40
Continuous variables
Table 1: Effect of production shock on severity of killings – continuousrainfall
(1) (2) (3) (4) (5) (6)log(Bodies)
Standardized within Province Raw DataNeg. Production during KR −0.026 −0.027 −0.025 −0.124 −0.145 −0.144
(0.012)∗∗ (0.011)∗∗ (0.012)∗∗ (0.034)∗∗∗ (0.046)∗∗∗ (0.051)∗∗[0.010]∗∗ [0.009]∗∗∗ [0.010]∗∗ [0.043]∗∗∗ [0.042]∗∗∗ [0.046]∗∗∗
Neg. Production 1972–1974 −0.010 −0.071(0.014) (0.050)[0.010] [0.045]
Neg. Production 1978–1980 0.006 0.013(0.017) (0.072)[0.016] [0.062]
Lat × Lon polynomial Yes Yes Yes Yes Yes YesProvince FE Yes Yes Yes Yes Yes YesCommune controls No Yes Yes No Yes YesDependent variable mean 0.141 0.141 0.141 0.141 0.141 0.141N 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 < 0.10, ∗∗ p < 0.05, ∗∗∗p < 0.01
Back41
Different dependent variable
Table 2: Effect of production shock on severity of killings – differentdependent variable
(1) (2) (3) (4) (5) (6) (7)log(Bodies) log(Bodies per capita) log(Bodies, Executed) log(Bodies per Area) Bodies per Capita Bodies per Area Mass graves
Neg. Production during KR −0.055 −0.058 −0.059 −0.129 −1.277 −0.015 −7.180(0.018)∗∗∗ (0.029)∗ (0.016)∗∗∗ (0.050)∗∗ (0.482)∗∗∗ (0.006)∗∗ (3.453)∗∗
Lat × Lon polynomial Yes Yes Yes Yes Yes Yes YesProvince FE Yes Yes Yes Yes Yes Yes YesCommune controls Yes Yes Yes Yes Yes Yes YesLeads and Lags Yes Yes Yes Yes Yes Yes YesDependent variable mean 0.141 0.233 0.086 0.405 2.393 0.025 12.079N 1,569 1,569 1,569 1,569 1,569 1,569 1,569
Standard errors clustered by 24 provinces in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Back
42