Appendix C
FULL STATISTICAL RESULTS
The statistical model used in chapter 3 focuses on the relationship between arms exports and importers’ human rights records. As such, rather than including a long list of independent variables that might have an effect on arms exports, it limits the control variables in the model to those that might have a confounding effect on both arms exports and human rights. Recent work in political methodology cautions against building “garbage can” or “kitchen sink” models that include a wide array of variables that may or may not influence the outcome of interest.1
Instead, control variables should be carefully selected based on the theoretical relationship between the independent variable of interest and the dependent variable as well as between the independent variables themselves. Christopher Achen argues that without a deliberately limited set of independent control variables, the effects of variables of interest on the dependent variable will be obscured and distorted by collinearity, nonlinearities, and other problems (2002:443; see also Achen 2005). Results from models featuring unnecessary variables are often highly contingent on precise specifications and make it more difficult to explore nuances of the relationship between variables of interest and the outcome (Achen 2005; Berk 2004). To determine which control variables to include, James Lee Ray (2003) and Kelly Kadera and Sara Mitchell (2005) suggest limiting models to only those control variables that threaten to have a confounding effect on the independent variable of interest. In other words, models should contain only potentially confounding variables but not include intervening or competing variables.
Appendix B describes the variable coding and selection, including the theoretical and empirical justifications for the chosen control variables. Each independent variable is lagged one year in the regression analyses to allow for information about conditions in recipient states to reach decision makers in exporting states (Blanton 2000, 2005; Meernik, Krueger, and Poe 1998). Decision makers need time to receive government, news, and NGO reports about conditions on the ground and the opportunity to adjust arms export practices accordingly. In addition, because the data are annual, relevant changes midyear might appear only minimally in the data set, if at all.
Because the variable for MCW transfers is continuous, I use ordinary least squares regression models for all regressions with MCW as the dependent variable. The variable for SALW transfers, in contrast, is binary (dichotomous) and therefore requires logit regression models where used as the dependent variable. Because the data in both cases are cross-sectional dyadic annual data, panel-corrected standard errors must be used to avoid an understatement of the errors due to the high number of error parameters involved in panel data, including panel heteroscedasticity and temporal dependence (Beck 2001; Beck and Katz 1995).
Tables C.1 to C.4 contain the full results for the regression analyses, including the regression coefficients and standard errors (SE) for each model and time period.
TABLE C.1. INFLUENCE OF HUMAN RIGHTS ON MCW TRANSFERS (ALL SUPPLIER STATES)
  1981–1991 (SE) 1992–1997 (SE) 1998–2010 (SE)
GDP per capita 5.201**
(.655) 
 3.820**
(.434) 
 3.142**
(.289) 
Democracy −.077  
(.138)
 −.018  
(.081)
 −.132  
(.075)
Low internal conflict   7.101**
(2.455) 
  3.871*
(1.700) 
 2.840
(1.593) 
High internal conflict 5.394
(3.619) 
    8.416**
(3.061) 
 7.750
(4.014) 
Oil production 1.008
(4.394) 
 7.955
(4.606) 
 −3.281  
(2.719) 
Good human rights 4.512**
(1.358)    
 3.629**
(1.326)    
  2.674**
(.859)  
Average human rights 10.005**  
(1.701)    
 4.789**
(1.487)    
 2.094**
(.798)  
Bad human rights 16.818**  
(3.267)    
 6.908**
(1.556)    
 11.125**  
(1.969)    
Very bad human rights 20.615**  
(4.793)    
 2.711   
(1.83)      
 −.044   
(1.605)    
Constant −31.198**   
(4.877)   
 −25.584**  
(3.506)  
 −23.541**  
(2.698)  
Observations       28236       19248       37134
Dyads         2905         3521         3477
Wald Chi2           256.67           114.69           232.04
Prob > Chi2 0.000    0.000    0.000   
*Significant at the 0.05 level.
**Significant at the 0.01 level.
Note: Coefficients for year dummies excluded from all tables to save space.
TABLE C.2. INFLUENCE OF HUMAN RIGHTS ON MCW TRANSFERS (ATT SUPPORTERS)
  1981–1991 (SE) 1992–1997 (SE) 1998–2010 (SE)
GDP per capita 4.148**
(.404) 
 4.258**
(.464) 
 2.935**
(.243) 
Democracy    .174**
(.060)
 −.026  
(.079)
 −.028  
(.033)
Low internal conflict    5.816**
(1.541) 
  3.686*
(1.784) 
   2.988**
(.966)
High internal conflict   2.879**
(1.041) 
    8.255**
(3.143)
   3.175**
(.902) 
Oil production 4.680
(3.487) 
  6.903
(4.948)
 −.080  
(2.360)  
Good human rights  4.090**
(1.299)   
   4.167**
(1.415) 
 2.364* 
(.922) 
Average human rights  5.710**
(1.307)   
    4.248**
(1.546)
 1.298  
(.825) 
Bad human rights 3.381**
(1.248)    
     6.120**
(1.590)
 3.502** 
(.859)   
Very bad human rights 2.024   
(1.404)    
  2.531
(1.857)
 .568    
(.891)   
Constant −26.936**  
(3.091)  
   −29.387**
    (.404)
 −22.470**  
(2.433)  
Observations       25670         17486        33766
Dyads         2641            3213          3161
Wald Chi2           226.79              119.08           250.37
Prob > Chi2 0.000       0.000 0.000   
*Significant at the 0.05 level.
**Significant at the 0.01 level.
TABLE C.3. INFLUENCE OF HUMAN RIGHTS ON SALW TRANSFERS (ALL SUPPLIER STATES)
  1981–1991 (SE) 1992–1997 (SE) 1998–2010 (SE)
GDP per capita 1.410**
(.057) 
 1.562**
(.056) 
 1.362**
(.041) 
Democracy   .084**
(.008) 
   .076**
(.010) 
    .066**
(.008) 
Low internal conflict   .366**
(.123) 
  .336*
(.127) 
   .312**
(.091) 
High internal conflict −.077   
(.161) 
  .429*
(.197)
   .605**
(.143) 
Oil production −1.467**  
(.409) 
 −1.417** 
(.383)
 −2.014** 
(.356) 
Good human rights −.443**  
(.125)   
 .230  
(.118)  
 −.023     
(.085)   
Average human rights −.343**  
(.124)   
  .657**
(.141)  
 −.193     
(.100)   
Bad human rights −.394*   
(.157)  
   .555**
(.168)  
 −.310*   
(.121)  
Very bad human rights −.610** 
(.193)  
 −.118   
(.201)  
 −.620** 
(.169)  
Constant −12.593**   
(.452)
 −12.874**   
(.454)
 −10.936**   
(.364)
Observations       28236       19248       37134
Dyads         2905         3521         3477
Wald Chi2           1223.88           1513.94           1984.55
Prob > Chi2 0.000  0.000  0.000 
*Significant at the 0.05 level.
**Significant at the 0.01 level.
TABLE C.4. INFLUENCE OF HUMAN RIGHTS ON SALW TRANSFERS (ATT SUPPORTERS)
  1981–1991 (SE) 1992–1997 (SE) 1998–2010 (SE)
GDP per capita 1.417**
(.059) 
 1.580**
(.059) 
 1.421**
(.044) 
Democracy   .083**
(.008) 
   .076**
(.011) 
   .065**
(.008) 
Low internal conflict   .417**
(.129) 
  .317*
(.133)
   .364**
(.096) 
High internal conflict −.103   
(.166) 
  .513*
(.206)
   .593**
(.153) 
Oil production −1.409**  
(.425) 
 −1.416** 
(.405)
 −1.967**  
(.380) 
Good human rights −.377**  
(.110)   
 .246* 
(.125)  
 −.047     
(.091)   
Average human rights −.276*    
(.129)   
  .665**
(.149)  
 −.211*    
(.106)   
Bad human rights −.314     
(.163)   
  .526**
(.177)  
 −.348**  
(.129)  
Very bad human rights −.531**  
(.199)   
 −.080    
(.211)  
 −.712** 
(.181)  
Constant −12.468**   
(.469) 
 −12.919**   
(.476)
 −11.378**   
(.388)
Observations       25670       17486       33766
Dyads         2641         3213         3161
Wald Chi2           1128.53           1391.09           1867.95
Prob > Chi2  0.000   0.000   0.000
*Significant at the 0.05 level.
**Significant at the 0.01 level.