SLIDE 1
Detecting Audience Costs in International Disputes & - - PowerPoint PPT Presentation
Detecting Audience Costs in International Disputes & - - PowerPoint PPT Presentation
Detecting Audience Costs in International Disputes & Informational Effects of Audience Costs Shuhei Kurizaki 1 Taehee Whang 2 1 School of Political Science and Economics Waseda University 2 Division of International Studies Korea University
SLIDE 2
SLIDE 3
Introduction What explains the origin of war? One prominent explanation is the “audience costs” model:
SLIDE 4
Introduction What explains the origin of war? One prominent explanation is the “audience costs” model:
A leader who backs down in an international crisis suffers audience costs of some form. Figure: Fearon (1994)
SLIDE 5
Introduction What explains the origin of war? One prominent explanation is the “audience costs” model:
A leader who backs down in an international crisis suffers audience costs of some form. Figure: Fearon (1994) Audience costs make a decision to go to war rational.
SLIDE 6
Introduction What explains the origin of war? One prominent explanation is the “audience costs” model:
A leader who backs down in an international crisis suffers audience costs of some form. Figure: Fearon (1994) Audience costs make a decision to go to war rational. A conjecture –audience costs are higher in democracies– led to massive applications
SLIDE 7
Audience costs: Problem Very controversial because ACs are only assumed to exist.
SLIDE 8
Audience costs: Problem Very controversial because ACs are only assumed to exist.
1 Theory: No consensus yet on the microfoundaiton.
SLIDE 9
Audience costs: Problem Very controversial because ACs are only assumed to exist.
1 Theory: No consensus yet on the microfoundaiton. 2 Empirics: None has presented definitive evidence.
SLIDE 10
In this paper We establish an empirical foundation of the AC model:
1 Direct evidence of the existence of audience costs.
SLIDE 11
In this paper We establish an empirical foundation of the AC model:
1 Direct evidence of the existence of audience costs. 2 Test common conjectures about audience costs:
SLIDE 12
In this paper We establish an empirical foundation of the AC model:
1 Direct evidence of the existence of audience costs. 2 Test common conjectures about audience costs:
Audience costs ∝ democracy
SLIDE 13
In this paper We establish an empirical foundation of the AC model:
1 Direct evidence of the existence of audience costs. 2 Test common conjectures about audience costs:
Audience costs ∝ democracy Audience costs ≥ war costs
SLIDE 14
In this paper We establish an empirical foundation of the AC model:
1 Direct evidence of the existence of audience costs. 2 Test common conjectures about audience costs:
Audience costs ∝ democracy Audience costs ≥ war costs Audience costs → bargaining power
SLIDE 15
In this paper We establish an empirical foundation of the AC model:
1 Direct evidence of the existence of audience costs. 2 Test common conjectures about audience costs:
Audience costs ∝ democracy Audience costs ≥ war costs Audience costs → bargaining power
3 Test the causal mechanism of the AC model:
SLIDE 16
In this paper We establish an empirical foundation of the AC model:
1 Direct evidence of the existence of audience costs. 2 Test common conjectures about audience costs:
Audience costs ∝ democracy Audience costs ≥ war costs Audience costs → bargaining power
3 Test the causal mechanism of the AC model:
Audience costs → effective signaling/belief updating
SLIDE 17
1
Audience Cost Model
2
Inference Problems
3
Research Strategy
4
Statistical Model of Audience Costs
5
Results Existence of Audience Costs Magnitude of Audience Costs Effects of Audience Costs Causal Mechanism of Audience Costs
SLIDE 18
1
Audience Cost Model
2
Inference Problems
3
Research Strategy
4
Statistical Model of Audience Costs
5
Results Existence of Audience Costs Magnitude of Audience Costs Effects of Audience Costs Causal Mechanism of Audience Costs
SLIDE 19
A Common AC Model and Selection Effects
Resist Back Down
1 ) ( ) (
2 1 1
- BD
u a BD u
~Challenge Status Quo
1 ) ( ) (
1 1
- SQ
u SQ u
Stand Firm
2 1 1 1
) ( ) ( w SF u w SF u
- Challenge
~Resist Fight ~Fight State 1 State 2 State 1 Concession
2 2 1
) ( 1 ) ( a CD u CD u
- Definition
Audience costs for State 1 exist iff u1(BD) < u1(SQ)
SLIDE 20
A Common AC Model and Selection Effect Unique Equilibrium Pooling equilibrium if ˆ a1 ≤ a1 < ˜ a1 Semi-pooling equilibrium if ˜ a1 ≤ a1 < a1 Separating equilibrium if a1 ≥ a1 Equilibrium and AC
SLIDE 21
A Common AC Model and Selection Effect Unique Equilibrium Pooling equilibrium if ˆ a1 ≤ a1 < ˜ a1 Semi-pooling equilibrium if ˜ a1 ≤ a1 < a1 Separating equilibrium if a1 ≥ a1 Equilibrium and AC Equilibrium behavior/outcome is shaped by AC
SLIDE 22
A Common AC Model and Selection Effect Unique Equilibrium Pooling equilibrium if ˆ a1 ≤ a1 < ˜ a1 Semi-pooling equilibrium if ˜ a1 ≤ a1 < a1 Separating equilibrium if a1 ≥ a1 Equilibrium and AC Equilibrium behavior/outcome is shaped by AC Equilibrium selects type (size) of observable AC
SLIDE 23
Audience Cost Model and Selection Effects
Resist Back Down ~Challenge Status Quo Stand Firm Challenge ~Resist Fight ~Fight State 1 State 2 State 1 Concession 1 a1
1
a
1
~ a
1
ˆ a
(a) First Selection Stage: Pr(CH)
a1
1
1
~ a
1
ˆ a
1
a
(b) Second Selection Stage: Pr(RS)
a1
1
1
~ a
1
ˆ a
1
a
(c) Third Selection Stage: Pr(BD)
SLIDE 24
Inference Problems
Selection bias Distribution of observable AC is truncated on the right. Inferring audience costs
SLIDE 25
Inference Problems
Selection bias Distribution of observable AC is truncated on the right. Inferring audience costs Must observe audience costs, both incurred and not incurred.
SLIDE 26
Inference Problems
Selection bias Distribution of observable AC is truncated on the right. Inferring audience costs Must observe audience costs, both incurred and not incurred. Must observe audience costs both on and off the equilibrium path.
SLIDE 27
Inference Problems
Selection bias Distribution of observable AC is truncated on the right. Inferring audience costs Must observe audience costs, both incurred and not incurred. Must observe audience costs both on and off the equilibrium path. Detecting audience costs amounts to recovering the underlying payoffs and payoff relations in the data.
SLIDE 28
Previous Tests Avoid Selection Bias 1st generation: Reduced Form Statistical tests of “reduced-form” behavioral implications as “revealed preferences.” 2nd generation: Experiments Monte Carlo simulations: Schultz (2001) Survey Experiments: Tomz (2007) and Kohno (2013) 3rd generation: Ignoring selection bias Case study: Snyder and Borghard (2011) Truncated data set Downes and Sechser (2012)
SLIDE 29
1
Audience Cost Model
2
Inference Problems
3
Research Strategy
4
Statistical Model of Audience Costs
5
Results Existence of Audience Costs Magnitude of Audience Costs Effects of Audience Costs Causal Mechanism of Audience Costs
SLIDE 30
Research Strategy to Detect Audience Costs Our Strategy: Structural Approach Estimate underlying payoffs based on observed outcomes. ⇒ This allows us to measure AC that are not incurred. Estimate the theoretical model itself, not reduced-form tests
SLIDE 31
Research Strategy to Detect Audience Costs Our Strategy: Structural Approach Estimate underlying payoffs based on observed outcomes. ⇒ This allows us to measure AC that are not incurred. Estimate the theoretical model itself, not reduced-form tests
Key Innovation: Estimator = Statistical Equilibrium
SLIDE 32
Research Strategy to Detect Audience Costs Our Strategy: Structural Approach Estimate underlying payoffs based on observed outcomes. ⇒ This allows us to measure AC that are not incurred. Estimate the theoretical model itself, not reduced-form tests
Key Innovation: Estimator = Statistical Equilibrium AC theory and selection bias are a product of the PBE, so use the PBE to construct the estimator.
SLIDE 33
Research Strategy to Detect Audience Costs Our Strategy: Structural Approach Estimate underlying payoffs based on observed outcomes. ⇒ This allows us to measure AC that are not incurred. Estimate the theoretical model itself, not reduced-form tests
Key Innovation: Estimator = Statistical Equilibrium AC theory and selection bias are a product of the PBE, so use the PBE to construct the estimator. Recent development in political science Statistical models based on Quantal Response Equilibrium Statistical models based on Perfect Bayesian Equilibrium
SLIDE 34
Research Strategy
Structural Estimation Approach
Theory: mapping from preferences to outcomes.
Preference relations Choices & Outcomes Equilibrium Deduction Given by assumption
SLIDE 35
Research Strategy
Structural Estimation Approach
Theory: mapping from preferences to outcomes.
Preference relations Choices & Outcomes Equilibrium Deduction Given by assumption
Empirics: mapping from outcomes to preferences.
Preference relations Choices & Outcomes Statistical Equilibrium Estimation Given by data
SLIDE 36
Research Strategy
Structural Estimation Approach
Theory: mapping from preferences to outcomes.
Preference relations Choices & Outcomes Equilibrium Deduction Given by assumption
Empirics: mapping from outcomes to preferences.
Preference relations Choices & Outcomes Statistical Equilibrium Estimation Given by data
We ask: “given the observation of outcomes, what prefenreces make these observed outcomes most likely according to the PBE?”
SLIDE 37
1
Audience Cost Model
2
Inference Problems
3
Research Strategy
4
Statistical Model of Audience Costs
5
Results Existence of Audience Costs Magnitude of Audience Costs Effects of Audience Costs Causal Mechanism of Audience Costs
SLIDE 38
Statistical Model of Audience Costs
Resist
Pr(RS|CH)
Back Down
1 1 1 1
1 1
) (
BD BD BD BD
X BD BD u
- 2
2 2 2
2 2
) (
BD BD BD BD
X BD BD u
- ~Challenge
Pr(~CH)
Status Quo
1 1 1 1
1 1
) (
SQ SQ SQ SQ
X SQ SQ u
- Stand Firm
1 1 1 1
1 1
) (
SF SF SF SF
X SF SF u
- 2
2 2 2
2 2
) (
SF SF SF SF
X SF SF u
- Challenge
Pr(CH)
~Resist
Pr(~RS|CH)
Fight
Pr(F|CH)
~Fight
Pr(~F|CH)
State 1 State 2 State 1 Concession
1 1 1 1
1 1
) (
CD CD CD CD
X CD CD u
- 2
2 2 2
2 2
) (
CD CD CD CD
X CD CD u
- Observable payoffs: mean payoffs + unobservable noise
u1(SF) = SF1 + ǫSF1 = XSF 1βSF 1 + ǫSF1 where ǫSF1 ∼ N(0, Var(ǫSF1))
SLIDE 39
Statistical Model of Audience Costs
From theoretical model to statistical model
Two new assumptions Private Information All the payoffs, not just war payoffs, are uncertain Type Space Type space is not bounded anymore u1(SF) = SF1 + ǫSF1, ǫSF1 ∼ N(0, Var(ǫSF1))
SLIDE 40
Statistical Model of Audience Costs
Model Specification 1 of 4
Empirical specifications are true to those in theoretical model. War Payoff: u1(SF) = p − c1 p: Prob that State 1 wins in a war Balance of power: Capabilities ratio c1: Cost of war Material cost: Economic development Political will to incur the cost: Democracy
SLIDE 41
Statistical Model of Audience Costs
Model Specification 2 of 4
Back-down Payoff for State 1: u1(BD) = 0 − a1 Status quo payoff of 0: Constant w/out covariates Testing a1 ∝ democracy demands Democracy, w/ constant Back-down Payoff for State 2: u2(BD) = 1
SLIDE 42
Statistical Model of Audience Costs
Model Specification 2 of 4
Back-down Payoff for State 1: u1(BD) = 0 − a1 Status quo payoff of 0: Constant w/out covariates Testing a1 ∝ democracy demands Democracy, w/ constant Back-down Payoff for State 2: u2(BD) = 1 Status quo payoff of 1: Identification issue demands normalization: Constant=0
SLIDE 43
Statistical Model of Audience Costs
Model Specification 2 of 4
Back-down Payoff for State 1: u1(BD) = 0 − a1 Status quo payoff of 0: Constant w/out covariates Testing a1 ∝ democracy demands Democracy, w/ constant Back-down Payoff for State 2: u2(BD) = 1 Status quo payoff of 1: Identification issue demands normalization: Constant=0 u2(BD) not affected by State 2’s choice.
SLIDE 44
Statistical Model of Audience Costs
Model Specification 2 of 4
Back-down Payoff for State 1: u1(BD) = 0 − a1 Status quo payoff of 0: Constant w/out covariates Testing a1 ∝ democracy demands Democracy, w/ constant Back-down Payoff for State 2: u2(BD) = 1 Status quo payoff of 1: Identification issue demands normalization: Constant=0 u2(BD) not affected by State 2’s choice. Other payoffs theoretically important
u2(SF) determines the bargaining range u2(CD) reflects audience costs’ coercive power
SLIDE 45
Statistical Model of Audience Costs
Model Specification 3 of 4
Concession Payoff for State 1: u1(CD) = 1 Value of the disputed good
Theoretical payoff is normalized, so not informative = ⇒ Rely on empirical literature
Concession Payoff for State 2: u2(CD) = 0 − a2
SLIDE 46
Statistical Model of Audience Costs
Model Specification 3 of 4
Concession Payoff for State 1: u1(CD) = 1 Value of the disputed good
Theoretical payoff is normalized, so not informative = ⇒ Rely on empirical literature Strategic interest similarity: Alliance portfolio Vulnerability of target: Civil war in State 2 High risk factor: Geographical contiguity
Concession Payoff for State 2: u2(CD) = 0 − a2
SLIDE 47
Statistical Model of Audience Costs
Model Specification 3 of 4
Concession Payoff for State 1: u1(CD) = 1 Value of the disputed good
Theoretical payoff is normalized, so not informative = ⇒ Rely on empirical literature Strategic interest similarity: Alliance portfolio Vulnerability of target: Civil war in State 2 High risk factor: Geographical contiguity
Concession Payoff for State 2: u2(CD) = 0 − a2 Same as u1(CD) Audience costs for public concession: Theoretically irrelevant
SLIDE 48
Statistical Model of Audience Costs
Model Specification 4 of 4
Status Quo Payoff for State 2: u2(SQ) = 1 Dropped from estimation us(SQ) is irrelevant for State 2’s deicsions, so nothing to infer. Status Quo Payoff for State 1: u1(SQ) = 0
SLIDE 49
Statistical Model of Audience Costs
Model Specification 4 of 4
Status Quo Payoff for State 2: u2(SQ) = 1 Dropped from estimation us(SQ) is irrelevant for State 2’s deicsions, so nothing to infer. Status Quo Payoff for State 1: u1(SQ) = 0 Identification requires normalization: Constant=0
SLIDE 50
Model (Payoff) Specification: Summary
XSQ1: C, MaxAge XCD1: C, Alliance, CivilWar2, Contiguous XCD2: C, Alliance, CivilWar2, Contiguous Xa: C, Democracy1 XSF1: C, CapShare1, Democracy1, Develop1, Population1 XSF2: C, CapShare1, Democracy2, Develop2, Population2
SLIDE 51
Estimation
ln L =
N
- i=1
[YSQi ln P(SQi) + YCDi ln P(CDi) + YBDi ln P(BDi) + YSFi ln P(SFi)] ,
Yz: binary crisis outcomes, z ∈ {SQ, CD, BD, SF}. Pr(zi): PBE probabilities of the outcomes We estimate a log-likelihood function of equilibrium outcome probabilities, covariates, payoff specification Maximization of ln L yields the vector of MLE of β’s.
SLIDE 52
Data - Dependent Variable Unit of analysis: military challenge cases, plus SQ cases 93 dyadic crisis cases ranging from 1919 to 1939 Integrate both Militarized Interstate Dispute data (MID) and International Conflict Behavior data (ICB) N = 2187 with the addition of SQ cases Outcome ICB MID Total SQ 2094 CD 28 16 44 BD 5 7 12 SF 33 4 37
SLIDE 53
Data - covariates Variables Basic Stats Measures MaxAge ln maximum age of states in the dyad Alliance Tau-b score of alliances CivilWar2 D: was B involved in a civil war? Contiguous D: share a land border or separated by < 150 miles of water? Democracy1 D: was A democratic? Democracy2 D: was B democratic? CapShare1 A’s share of capabilities Develop1 ln energy consumption pc for A Develop2 ln energy consumption pc for B
SLIDE 54
1
Audience Cost Model
2
Inference Problems
3
Research Strategy
4
Statistical Model of Audience Costs
5
Results Existence of Audience Costs Magnitude of Audience Costs Effects of Audience Costs Causal Mechanism of Audience Costs
SLIDE 55
Estimation Results
Outcomes Payoffs Variables Est. (SE) Status Quo SQ1 Constant MaxAge 0.575** 0.135 Concede CD1 Constant 0.981 0.912 Alliance
- 3.507**
1.164 CivilWar2 4.460** 1.451 Contiguous 3.155** 0.902 CD2 Constant
- 1.395**
0.562 Alliance 0.479 0.325 CivilWar2 0.181 0.212 Contiguous
- 0.366
0.231 Back Down BD1 Constant
- 4.092**
0.820 Democracy1
- 0.411**
0.104 Back Down BD2 Constant Stand Firm SF1 Constant
- 4.620**
0.789 CapShare1 0.951** 0.469 Democracy1
- 0.366*
0.093 Develop1 0.086 0.049 SF2 Constant
- 2.733**
0.786 CapShare1 0.612 0.417 Democracy2 0.014 0.009 Develop2
- 0.016
0.024 Variance Var(CD1) 3.037** 0.871 Var(SF1) 1.370 0.809 Covariance Cov(CD1, BD1)
- 0.242
0.665 Cov(CD1, SF1) 0.867 0.634 Cov(BD1, SF1)
- 1.170
1.019 N 2102 Log likelihood
- 361.141
∗∗ = p < .01, ∗ = p < .05 (two-tailed)
SLIDE 56
Existence of Audience Costs Main Test Audience costs of some form exist: u1(BD) < u1(SQ).
SLIDE 57
Existence of Audience Costs Estimated payoff relation: u1(BD) < u1(SQ) −4.092 + Democracy1 ∗ −0.411 < 0 + MaxAge ∗ 0.575 Outcomes Payoffs Variables Est. (SE) Status Quo SQ1 Constant MaxAge 0.575** 0.135 Back Down BD1 Constant
- 4.092**
0.820 Democracy 1
- 0.411**
0.104 ∗∗ = p < .01, ∗ = p < .05 (two-tailed)
SLIDE 58
Estimated Audience Costs
- 9
- 7
- 5
- 3
- 1
1
- 10
- 5
5 10
The estimated audience costs, u1(BD) − u1(SQ), as a function of Democracy1 for the most difficult case (i.e., MaxAge is at its minimum).
SLIDE 59
Estimated Audience Costs Robustness check
Against alternative empirical specifications Against alternative data sets − → Upper bound of 95% confidence interval can be greater than zero if highly non-democratic and if very young.
- DemocracyofState1
EstimatedAudienceCosts
SLIDE 60
Testing Conjecture about Association with Democracy Audience costs ∝ democracy Audience costs of some form exist: u1(BD) < u1(SQ).
SLIDE 61
Testing Conjecture about Association with Democracy Outcomes Payoffs Variables Est. (SE) Status Quo SQ1 Constant MaxAge 0.575** 0.135 Back Down BD1 Constant
- 4.09**
0.820 Democracy 1
- 0.411**
0.104 ∗∗ = p < .01, ∗ = p < .05 (two-tailed) Fearon’s conjecture is confirmed
First evidence that audience costs increase with democracy score Support for existing applied work that attributes democratic uniqueness to audience costs.
SLIDE 62
Magnitude of Audience Costs Audience Costs ≥ War Costs A rationalist explanation for war requires u1(BD) < u1(SF) Can ex post inefficient war be a rational choice?
SLIDE 63
Are Audience Costs greater than War Costs?
10 5 5 10 2 4 6 8 Democracy of State 1 Average War and Audience Costs to State 1 Capshare 1 at 1st Qu. Audience Costs War Costs
(a) Military Capability 1 : 3
10 5 5 10 2 4 6 8 Democracy of State 1 Capshare 1 at 3rd Qu. Audience Costs War Costs
(b) Military Capability 3 : 1
SLIDE 64
Are Audience Costs greater than War Costs? Audience costs and war costs are statistically indistinguishable.
Audience costs and the cost of war are roughly the same Audience costs are very large! Very democratic, very overwhelming power may make war a rational choice
SLIDE 65
Effects of Audience Costs on Crisis Outcome Audience Costs → Bargaining Power Effect of Audience Costs (or Democracy) on Conflict Behavior “The Case for Democratic Credibility”
SLIDE 66
Effect of Audience Costs on Crisis Outcome
−2 −4 −6 −8 −10 0.3 0.4 0.5 0.6 0.7 Audience Costs to State 1 Estimated Probabilities of Resist
Pr(Resist) when CapShare1 at Mean Pr(Resist) when CapShare1 at 1st Qu. Pr(Resist) when CapShare1 at 3rd Qu.
(a) Effect of audience costs
−1 −2 −3 −4 −5 0.3 0.4 0.5 0.6 0.7 War Values to State 1 Estimated Probabilities of Resist
Pr(Resist) when CapShare1 at Mean Pr(Resist) when CapShare1 at 1st Qu. Pr(Resist) when CapShare1 at 3rd Qu.
(b) Effect of war costs
SLIDE 67
Effect of Audience Costs on Crisis Outcome Audience costs increase the prob the opponent capitulates
Statistically significant Substantively significant, compared to the effect of war cost.
SLIDE 68
Informational Effects of Audience Costs Audience costs − → signaling/belief-updating Audience costs improve leaders’ ability to signal their intentions and to make a commitment. Many studies assume the informational effect as the causal mechanism. But this is only assumed or shown only indirectly with problematic assumptions.
SLIDE 69
Testing Informational Effects of AC We provide empirical evidence of the informational mechanism of audience costs: Estimate signaling and learning as predicted in the theoretical model. Demonstrate the effect of estimated audience costs on signaling and learning.
SLIDE 70
Informational Effects of Audience Costs Singling and Learning (Theory) Belief updating = S2’s posterior minus prior beliefs.
a1 S1’s audience costs
1
a
1
ˆ a
Prior beliefs (45°) Posterior beliefs (q) Belief updating ()
1
1
~ a S2’s subjective probability that S1 is resolved
SLIDE 71
Measuring Signaling and Learning Measuring beliefs can be done with estimated state preferences in the underlying game. Prior Belief Ex ante probability that State 1 prefers standing firm to backing down if State 2 resists. Pr(SF) = Pr(u1(SF) ≥ u1(BD)) Posterior Belief Pr(SFpub|CH) = Pr (u1(SF) ≥ u1(BD)|E[u1(CH)] ≥ u1(SQ))
SLIDE 72
Measuring Signaling and Learning Prior Belief State 2’s initial assessment about the probability that State 1 will fight at the onset of a crisis. Pr(F) = Pr(u1(SF) ≥ u1(BD)) = Φ
- E[∆USF,BD]
- Var [∆USF,BD]
- ,
where ∆USF,BD = u1(SF) − u1(BD). Posterior Belief The conditional probability that State 1 will fight given that State 1 has issued a threat. Pr(F|CH)
SLIDE 73
Measuring Signaling and Learning Posterior Belief
Pr(F|CH) = Pr(u1(SF) ≥ u1(BD)|E[u1(CH)] ≥ u1(SQ)) = Φ
- E[∆USF,BD]
√
Var[∆USF,BD], E[∆USF,SQ]
√
Var[∆USF,SQ], Corr[∆USF,BD, ∆UCH,SQ]
- Pr(CH)
, where Pr(CH) = Pr(E[u1(CH)] ≥ u1(SQ)) = 1 − Φ2
- E[∆USQ,BD]
- Var[∆USQ,BD]
, E[∆USQ,SF]
- Var[∆USQ,SF]
, Corr(∆USQ,BD, ∆USQ,SF)
- ,
SLIDE 74
Measuring Signaling and Learning Estimate var-cov matrix to estimate belief updating correctly
Identification Seven additional parameters
Correct estimation of belief updating Previous models as special cases (Lewis and Schultz 2003; Wand 2006; Signorino and Whang 2009)
SLIDE 75
Estimated Informational Effects & Interpretations
Prior and Posterior Beliefs
−10 −5 5 10 0.2 0.4 0.6 0.8 1.0 Democracy of State 1 Prior and Posterior Probabilities
Posterior Prior
Posterior: No full separation Bootstrap analysis shows statistically different than full separation Empirical model assumes the normal dist. of types so full separation is not possible by construction
SLIDE 76
Estimated Informational Effects & Interpretations
Belief updating: statistical significance
−10 −5 5 10 0.20 0.22 0.24 0.26 0.28 0.30 0.32 Democracy of State 1 Amount of Belief Updating
Statistically different than zero, so learning occurs. Except for the least democratic regimes (Democracy1 = −10) Effect of AC is also significant, so ACs facilitate learning.
SLIDE 77
Estimated Informational Effects & Interpretations
Belief updating: substantive significance
Minimum AC Maximum AC Effect (Democracy1 = −10) (Democracy1 = 10)
- f AC
Before Threats 40% 53% 13% (Prior Belief) After Threats 60% 85% 25% (Posterior Belief) Effects of Threats +20% +32% +12% (Updated Beliefs) Is the effect of AC sizable, relative to that of fait accompli? AC: 32% − 20% = 12% increase FA: 60% − 40% = 20% increase (min AC) FA: 85% − 53% = 32% increase (max AC)
SLIDE 78
Conclusion: Finding Summary This study establishes an empirical foundation of the AC models
1 The existence of AC using observational data
SLIDE 79
Conclusion: Finding Summary This study establishes an empirical foundation of the AC models
1 The existence of AC using observational data 2 The correlation of AC with democraticness.
SLIDE 80
Conclusion: Finding Summary This study establishes an empirical foundation of the AC models
1 The existence of AC using observational data 2 The correlation of AC with democraticness. 3 The magnitude of AC is comparable to the war cost.
SLIDE 81
Conclusion: Finding Summary This study establishes an empirical foundation of the AC models
1 The existence of AC using observational data 2 The correlation of AC with democraticness. 3 The magnitude of AC is comparable to the war cost. 4 The effect of AC (on coercing an adversary to concede) is
both statistically and substantively significant.
SLIDE 82
Conclusion: Finding Summary This study establishes an empirical foundation of the AC models
1 The existence of AC using observational data 2 The correlation of AC with democraticness. 3 The magnitude of AC is comparable to the war cost. 4 The effect of AC (on coercing an adversary to concede) is
both statistically and substantively significant.
5 The effect of AC on belief-updating is statistically significant
but not sizable.
SLIDE 83
Testing a Formal Model: Conclusion Structural estimation of the theoretical AC model
1 Measure AC that are not incurred, off the eqlm path.
Estimating the payoffs Overcoming selection bias
2 Measure beliefs and updating.
SLIDE 84
Thank you!
SLIDE 85