Development and Conflict Debraj Ray, New York University
Two parts: 1. Endemic conflict with private and public goods. (Debraj Ray, Remarks on the Initiation of Costly Conflict, preliminary but on my webpage.) 2. Using economics to identify aggressors in conflict. (Anirban Mitra and Debraj Ray, Identifying the Aggressor in Hindu- Muslim Conflict, very preliminary, so not even on my webpage.)
Internal Conflict is Endemic 1945 1999: battle deaths in 25 interstate wars approx. 3.33m 127 civil wars in 73 states (25 ongoing in 1999). (over 1/3 of all countries) 16m+ dead as a direct result (not counting deaths from displacement and disease). Economic costs: 8% of world GDP (Hess (2003)) Does not count ongoing ethnic violence, such as Hindu-Muslim violence in India.
Noneconomic Markers In much of Asia and Africa, it is only modest hyperbole to assert that the Marxian prophecy has had an ethnic fulfillment. Horowitz (1985) Brubaker and Laitin (1998) on eclipse of the left-right ideological axis Fearon (2006), 1945 1998, approx. 700 ethnic groups known, over 100 of which participated in rebellions against the state. Montalvo and Reynal-Querol (2005) apply the 1994 Esteban- Ray measure to show that ethnic and religious polarization correlated with conflict. Esteban and Ray (2008) study ethnic salience.
A Variety of Markers Some groupings that have confronted the State in India: Fundamentalist religious groups (both Hindu and Muslim) Landlords and sharecroppers (sometimes on the same side!) Maoist groups (West Bengal, Andhra) Sikhs (Punjab) The northeast states High caste groups Scheduled castes Agricultural labor Protectionist trade unions Industrial lobbies
Yet Much Conflict Is Indeed Economic...... but the connections are complex:
Yet Much Conflict Is Indeed Economic...... but the connections are complex: Appropriable Resources Ross (2004, 2006) on effect of natural resources on civil war. Dube and Vargas (2009) on oil shocks in Colombia. Angrist and Kugler (2008) on coca in Colombia. Versi (1994) and André and Platteau (1998) on Rwanda [land] Mamdani (2009) and O Fahey (2009) on Darfur [land]
Uneven Economic Growth Spilerman (1979) on 1960s race riots in the US [resentment at minority progress] Thakore (1993) and Das (2000) on 1992 3 Bombay and Calcutta riots [real estate] Rajgopal (1987) and Khan (1992) on Bhiwandi and Meerut riots [competition in textile sector] Engineer (1994) and Khan (1991) on Jabbalpur, Kanpur, Moradabad [competition in bidis, brassware] Wilkinson (2004) on Varanasi [wholesale silk trade] Bagchi (1990), Engineer (1984), Wilkinson (2004) on economic aspects of Hindu-Muslim conflict. Sarkar (2007) and Gang of Nine (2007) on Singur and Nandigram [land transfers]
Poverty and the Supply of Conflict. Murshed and Gates (2005) and Do and Iyer (2007) on poverty in Nepal. Honaker (2008) on unemployment in N. Ireland. Dube and Vargas (2009) on coffee shocks in Colombia. Kapferer (1998) and Senenayake (2004) on poverty in Sri Lanka. Gandhi (2003) on Dalit participation in Gujarat. Humphreys and Weinstein (2008) on poor conflict participants in Sierra Leone.
Some of this is roughly mirrored in cross-country studies. Negative correlation between development (growth, per-capita income) and conflict. Some evidence that natural resources positively correlated with conflict. Fearon and Laitin (2003), Collier and Hoeffler (1998, 2004), Miguel et al. (2004), Bruckner and Ciccone (2007), Besley and Persson (2008), Bazzi and Blattman (2008). However, much of this too blunt to capture the nuances of within-country experience.
Why Conflict? Equilibrium conflict is Pareto inefficient Why not arrange a transfer scheme to avoid costly conflict? James Fearon (1995) refers to this as the central puzzle.
1. Incomplete Information. Myerson-Satterthwaite (1983), Fearon (1995), Esteban and Ray (2001), Bester and Warneryd (2006), Sánchez-Pagés (2008). 2. The Absence of Transfers. Kirshner (2000). 3. Limited Commitment Sneak attacks: Slantchev (2007). Fearon (1995), Slantchev (2003), Leventoglu and Dynamics and credibility: (2000), Powell (2007). Fearon (1995), Garfinkel and Skaperdas 4. Uninternalized Costs Fearon (1995), Jackson and Morelli (2007). 5. The Multiplicity of Threats; studied here.
Endemic Conflict Society has resources at stake, can be appropriated. Produces variety of goods, but all Coasian transfers available.
Endemic Conflict Society has resources at stake, can be appropriated. Produces variety of goods, but all Coasian transfers available. A marker (ethnicity, class, religion... ) is a subset M of individuals. Conflict: M versus M.
A Conflict Subgame Simplified version of Esteban and Ray (1999).
A Conflict Subgame Simplified version of Esteban and Ray (1999). M induces conflict against complement M. Winner allocates resources the way they want. (Again, all transfers available.)
A Conflict Subgame Simplified version of Esteban and Ray (1999). M induces conflict against complement M. Winner allocates resources the way they want. (Again, all transfers available.) Conflict costly and inefficient. Each group contributes resources. Per capita convex cost with constant elasticity α > 1.
A Conflict Subgame Simplified version of Esteban and Ray (1999). M induces conflict against complement M. Winner allocates resources the way they want. (Again, all transfers available.) Conflict costly and inefficient. Each group contributes resources. Per capita convex cost with constant elasticity α > 1. Win probability proportional to relative group contributions.
If π = win value to M and π = win value to M (per-capita), then γ (r/r) = (π/π) 1/α
If π = win value to M and π = win value to M (per-capita), then γ (r/r) = (π/π) 1/α and win probability is p = mγ mγ + (1 m). where m = relative size of M.
If π = win value to M and π = win value to M (per-capita), then γ (r/r) = (π/π) 1/α and win probability is p = mγ mγ + (1 m). where m = relative size of M. Expected payoff from conflict π[kp + (1 k)p 2 ], where k (α 1)/α, which lies in (0, 1).
Private Goods
Private Goods Observation 1. Privately divisible resources; payoffs equally divided under peace. Then there is m (0, 1 2 ) so that marker m < m wants conflict.
Private Goods Observation 1. Privately divisible resources; payoffs equally divided under peace. Then there is m (0, 1 2 ) so that marker m < m wants conflict. Outline of Argument. Resources = v; βv survives conflict. Then π = βv/m and π = βv/(1 m) and so γ = ( 1 m m ) 1/α
Private Goods Observation 1. Privately divisible resources; payoffs equally divided under peace. Then there is m (0, 1 2 ) so that marker m < m wants conflict. Outline of Argument. Resources = v; βv survives conflict. Then π = βv/m and π = βv/(1 m) and so γ = ( 1 m m ) 1/α Therefore p(m) = mγ mγ + (1 m) = m k m k + (1 m) k, where k = (α 1)/α. [Pareto-Olson.]
Need kp(m) + (1 k)p(m) 2 > m/β. 1 p, p 2 1/2 0 1/2 1 m
Need kp(m) + (1 k)p(m) 2 > m/β. 1 p, p 2 1/2 0 1/2 1 m
Need kp(m) + (1 k)p(m) 2 > m/β. 1 p, p 2 1/2 0 1/2 1 m
Need kp(m) + (1 k)p(m) 2 > m/β. p, p 2 1 m/! 1/2 0 m* 1/2 1 m
Of course, there is some allocation that will appease the marker (conflict is inefficient). But that allocation will need to vary with the marker in question.
Of course, there is some allocation that will appease the marker (conflict is inefficient). But that allocation will need to vary with the marker in question. A finite collection C of markers is balanced if there are weights λ(m) in [0, 1], one for each marker, such that M C,i M λ(m) = 1 for every i in society
Of course, there is some allocation that will appease the marker (conflict is inefficient). But that allocation will need to vary with the marker in question. A finite collection C of markers is balanced if there are weights λ(m) in [0, 1], one for each marker, such that M C,i M λ(m) = 1 for every i in society For instance, any partition is balanced. More generally, rules out common intersection.
Proposition 1. Suppose there exists a balanced collection of markers, each with m < m. Then there is no peaceful allocation for society that is immune to conflict.
Proposition 1. Suppose there exists a balanced collection of markers, each with m < m. Then there is no peaceful allocation for society that is immune to conflict. If society can be partitioned into markers, enough for endemic conflict.
Proposition 1. Suppose there exists a balanced collection of markers, each with m < m. Then there is no peaceful allocation for society that is immune to conflict. If society can be partitioned into markers, enough for endemic conflict. Even weaker conditions possible: E.g. quadratic costs and β = 1. For instability, enough to have six pairwise disjoint groups of size 10% (don t need partition).
Public Goods
Public Goods Social resources only produce public goods. Public goods are marker-specific (e.g., religious support, health or education, export processing zones).
Public Goods Social resources only produce public goods. Public goods are marker-specific (e.g., religious support, health or education, export processing zones). M-person gets Ψ per unit from M-public good; 0 otherwise. Production linear and symmetric.
Public Goods Social resources only produce public goods. Public goods are marker-specific (e.g., religious support, health or education, export processing zones). M-person gets Ψ per unit from M-public good; 0 otherwise. Production linear and symmetric. Efficiency: with transfers produce for largest marker m ; compensate rest Private endowments for sidepay- (Note Coasian assumption! ments.)
Existence of Peaceful Agreements
Existence of Peaceful Agreements Proposition 2. Assume public goods. Suppose that the complement of every marker is also a marker. Then there exists a peaceful allocation that is immune to conflict.
Existence of Peaceful Agreements Proposition 2. Assume public goods. Suppose that the complement of every marker is also a marker. Then there exists a peaceful allocation that is immune to conflict. Outline of Argument. If M wins, it produces M-public goods. If M wins, produces for largest marker (of relative size µ) in M. So π = βψ and π = βψµ, and therefore γ = (r/r) = µ 1/α 1.
p(m) = mγ/(mγ + [1 m]). Want Ψm βψ [ kp(m) + (1 k)p(m) 2].
p(m) = mγ/(mγ + [1 m]). Suffices: m/β [ kp(m) + (1 k)p(m) 2].
p(m) = mγ/(mγ + [1 m]). Suffices: m/β [ kp(m) + (1 k)p(m) 2]. 1 p, p 2 0 1 m
p(m) = mγ/(mγ + [1 m]). Suffices: m/β [ kp(m) + (1 k)p(m) 2]. 1 p, p 2 0 1 m
p(m) = mγ/(mγ + [1 m]). Suffices: m/β [ kp(m) + (1 k)p(m) 2]. 1 p, p 2 0 1 m
p(m) = mγ/(mγ + [1 m]). Suffices: m/β [ kp(m) + (1 k)p(m) 2]. 1 p, p 2 0 1 m
The Contrast Proposition 3. Suppose that every subset of individuals can be a marker. Then, under private goods, there is no peaceful allocation for society that is immune to conflict. Under public goods, there is always such an allocation. Conflicts over private, economic goods appear to be focal, even if the markers are not economic.
Extensions Beyond Bilateral Contests. Anticipation of further deviations. Private and Public Goods. Hybrid uses of the budget, or pure public goods without private transfers. Dynamics. Institutional sluggishness versus marker formation Inequality within Groups. conflict. An alternative approach to ethnic Multiple Identities. Sen s argument.
Ethnicity and Economics Premise so far: conflict. ethnicity, religion, just proxies for economic
Ethnicity and Economics Premise so far: conflict. ethnicity, religion, just proxies for economic Suggests a method to identifying the aggressor using economic data. Motivated by ongoing Hindu-Muslim violence in India (most recently Gujarat). Exercise that follows based on Mitra and Ray (2009).
Hindu-Muslim conflict in post-independence India.
Hindu-Muslim conflict in post-independence India. Conflict Data. Varshney-Wilkinson (TOI 1950 1995). 2000+ Hindu-Muslim riots, 10,000+ deaths, 30,000+ injuries.
Hindu-Muslim conflict in post-independence India. Conflict Data. Varshney-Wilkinson (TOI 1950 1995). 2000+ Hindu-Muslim riots, 10,000+ deaths, 30,000+ injuries. Income Data. National Sample Survey expenditure data. Rounds 38 (1983) and 43 (1987-8).
Hindu-Muslim conflict in post-independence India. Conflict Data. Varshney-Wilkinson (TOI 1950 1995). 2000+ Hindu-Muslim riots, 10,000+ deaths, 30,000+ injuries. Income Data. National Sample Survey expenditure data. Rounds 38 (1983) and 43 (1987-8). Combine: Panel at the regional level: 14 states, 55 regions. Cross-section at district level, round 43, 322 districts.
Informal Description of Model One side presumed to be the aggressor. Aggressor elites fund conflict infrastructure. Decentralized matching of potential aggressor-victim pairs. Infrastructure determines frequency of provocations. Some provocations end in conflict.
Theoretical Predictions An increase in the mean income of the victim group increases conflict. Unambiguously more to grab. True that victim can better defend himself. In simultaneous move attack-defence game, former effect dominates latter.
Theoretical Predictions An increase in the mean income of the victim group increases conflict. Unambiguously more to grab. True that victim can better defend himself. In simultaneous move attack-defence game, former effect dominates latter. An increase in mean income of aggressor group has an ambiguous effect on conflict. Lowers physical participation in confrontations. Increases elite funding.
Empirical Specification Interested in counts: outbreaks, deaths, injuries, etc. View as Poisson process with some parameter λ = f(x, ɛ).
Empirical Specification Interested in counts: outbreaks, deaths, injuries, etc. View as Poisson process with some parameter λ = f(x, ɛ). λ = Prob(crossmatch) Prob(confrontation) Prob(violence) Prob(crossmatch) related to π(1 π), where π = % of Muslims. Prob(confront) related to conflict investment plus local noise. Prob(violence) = P = f(µ a, µ v, infrastructure).
Empirical Specification Interested in counts: outbreaks, deaths, injuries, etc. View as Poisson process with some parameter λ = f(x, ɛ). λ = Prob(crossmatch) Prob(confrontation) Prob(violence) Prob(crossmatch) related to π(1 π), where π = % of Muslims. Prob(confront) related to conflict investment plus local noise. Prob(violence) = P = f(µ a, µ v, infrastructure). µ v implies conflict. Direct effect of µ a negative but positive via infrastructure.
Motivates the negative binomial specification E(Count i,t+1 X it, γ i ) = exp(x it β + γ i + τ t ) where X includes expenditures, both for Hindu and Muslim H-M polarization other time-varying controls such as population, lagged counts and the γ s and τ s are region and time effects (for the panel).
Panel: Casualties (4-year average) starting three years later [1] [2] [3] [4] Muslim pce 1.94** 1.92** 1.96** 1.95** (2.00) (2.00) (2.03) (2.03) Hindu pce 0.05-0.01-0.06-0.14 (0.05) (0.00) (0.53) (0.06) Time -0.63-0.70-0.75-0.83 (1.15) (1.26) (1.30) (1.43) Pop 0.55*** 0.59*** 0.53*** 0.58*** (5.11) (4.83) (4.94) (4.70) Muslim % 0.06** 0.06** (2.38) (2.38) RelPol 2.33** 2.33** (2.43) (2.44) CurrCasualties -0.00-0.00 (0.67) (0.67) Pop 10% Cas 10.5%; Mus exp 10% Cas 20%.
Panel: Outbreak (4-year average) starting three years later [1] [2] [3] [4] Muslim pce 1.75** 1.55* 1.45* 1.56* (2.19) (1.94) (1.84) (1.95) Hindu pce 0.93 0.54 0.40 0.62 (0.88) (0.65) (0.53) (0.74) Time 0.48 0.04 0.08 0.09 (1.18) (0.07) (0.18) (0.18) Pop 0.53*** 0.41*** 0.51*** (4.87) (3.62) (4.60) Muslim % 0.03 0.04 0.03 (1.15) (1.54) (1.15) CurrOutbreak 0.05** (2.69) CurrCasualties 0.00 (0.50) Pop 10% Out 10%; Mus exp 10% Out 16.1%.
Panel: Riot Years starting three years later [1] [2] Muslim pce 2.18** 2.16** (2.35) (2.32) Hindu pce 0.62 0.64 (0.68) (0.69) Time 0.49 0.48 (0.92) (0.89) Pop 0.66*** 0.66*** (4.52) (4.50) Muslim % 0.05* 0.05* (1.82) (1.75) Eff # Parties 0.10 (0.55)
Casualties (1990 1993), NSS 43R 1987-88 on RHS [1] [2] [3] [4] Muslim pce 2.16*** 2.11*** 2.10*** 2.04*** (3.39) (3.20) (3.13) (3.04) Hindu pce 1.77 1.72 1.79 1.74 (1.59) (1.54) (1.54) (1.49) Pop 1.93** 1.84** 1.96** 1.86** (2.34) (2.22) (2.37) (2.23) Muslim % 1.18*** 1.14*** (4.13) (4.09) RelPol 1.25*** 1.21*** (3.79) (3.76) Cas86 89 0.01 0.01 (1.14) (1.14) Pop 10% Cas 19%; Mus exp 10% Cas 22%.
Also ran casualties on H-M income ratios. Casualties (1990 1993), NSS 43R 1987-88 on RHS [1] [2] [3] [4] H-M expratio -1.84*** -1.75** -1.78** -1.68** (2.68) (2.47) (2.52) (2.32) Average pce 3.98*** 3.93*** 3.88*** 3.83*** (3.31) (3.21) (3.03) (2.94) Pop 1.87** 1.90** 1.78** 1.80** (2.27) (2.30) (2.15) (2.17) Muslim % 1.19*** 1.15** (4.29) (4.25) RelPol 1.25*** 1.21*** (3.92) (3.90) CurrCasualties 0.01 0.01 (1.00) (1.14)
examples of other specs: casualties + injuries fixed effects vs random effects (results stronger for outbreaks, same for casualties)
examples of other specs: casualties + injuries fixed effects vs random effects (results stronger for outbreaks, same for casualties) An important concern is endogeneity. Effect of conflict on Muslim incomes Effect of conflict on Hindu incomes
Anecdotal evidence suggests these concerns should bias the results against us: 1985 1987 526 Hindu-Muslim incidents in 10 states. Muslims were 12% of the population, but suffered 60% of the 443 deaths 45% of the 2667 injuries 73% of the estimated property damage from Wilkinson (2004), who quotes the 9th and 10th Annual Reports of the Minorities Commission (1988 and 1989).
Muslim expenditure NSS 38R and 43R CurrCasualties -0.0005*** [1] [2] [3] (4.00) CurrOutbreak -0.011*** (3.67) CurrRiotYears -0.046** (2.30) Pop -0.042-0.044 0.023 (0.34) (0.35) (0.18) Muslim % 0.007 0.008 0.008 (1.17) (1.6) (1.33) Time Dummy yes yes yes Cas 100 MusExp 5% (small effects). Effect of lagged conflict on Muslim expenditure is also 0.
Hindu expenditure NSS 38R and 43R CurrCasualties -0.0001 [1] [2] [3] (1.14) CurrOutbreak -0.0022 (1.18) CurrRiotYears -0.197 (1.32) Pop -0.061-0.062 0.030 (0.54) (0.56) (0.24) Muslim % -0.006 0.006 0.005 (1.58) (1.50) (1.47) Time Dummy yes yes yes Effect of lagged conflict on Hindu expenditure is also 0.
Another concern (thanks Abhijit!): this is a proxy for relative Hindu malaise which manifests itself in undirected group violence, no offence meant to Muslims in particular. GOI Crime, All Riots (1990 1993), NSS 43R 1987-88 on RHS [1] [2] [3] [4] Muslim pce 0.31 0.29 0.19 0.17 (1.22) (1.15) (1.37) (1.25) Hindu pce 0.09 0.06 0.12 0.12 (0.26) (0.18) (0.48) (0.50) Pop 0.62*** 0.63*** 0.22* 0.22* (3.44) (3.50) (1.85) (1.86) Muslim % 0.32*** 0.17** (3.30) (2.20) RelPol 0.32*** 0.16*** (3.40) (2.83) Riots 86 89 0.007*** 0.007*** (7.00) (7.00)
Remarks on the Lag Structure
Remarks on the Lag Structure I ve reported on conflict in 1986 1989 using 1983 expenditure data (and 1990 1993 using 1987 expenditure data). Other extreme is contemporaneous conflict. Results very similar to regressions of Muslim and Hindu expenditures on conflict (reported here). Other lags generate an inverted-u Muslim effect peaking at our lag structure.
Summary I study economic explanations for endemic conflict. Main argument: ethnicity can be used as a marker to corner economic resources (money, business opportunities, property). Multiplicity of ethnic markers can create inefficient conflict. Threats to privately divisible resources may be more likely to result in conflict than differences over public goods (though in particular situations, either is compatible with conflict, or peace).
Summary I study economic explanations for endemic conflict. Main argument: ethnicity can be used as a marker to corner economic resources (money, business opportunities, property). Multiplicity of ethnic markers can create inefficient conflict. Threats to privately divisible resources may be more likely to result in conflict than differences over public goods (though in particular situations, either is compatible with conflict, or peace). The economic approach can form the basis of a test to identify the aggressor. I conduct such a test for Hindu-Muslim conflict in India. The test suggests that Hindus have been the aggressors.