–0

138**
–0.332**
–0.011
(0.050)
(0.037)
(0.064)
(0.020)
Catholic
–0.410**
–0.183**
–0.655**
–0.076**
(0.031)
(0.027)
(0.047)
(0.017)
Hoa Hao
–0.542**
–0.303**
–1.216**
–0.384**
(0.035)
(0.031)
(0.080)
(0.020)
Other/no majority
0.620**
0.198**
–0.483**
–0.035
ethnicity
(0.079)
(0.051)
(0.156)
(0.035)
Khmer
0.410**
0.242**
–0.177**
0.028
(0.035)
(0.036)
(0.052)
(0.023)
Chinese
–0.365**
0.024
–0.622**
–0.207**
(0.124)
(0.085)
(0.206)
(0.040)
Montagnard
–0.099*
–0.089*
–0.601**
–0.120**
(0.040)
(0.035)
(0.051)
(0.023)
Urban
–0.779**
–0.463**
–1.343**
–0.328**
(0.039)
(0.034)
(0.077)
(0.014)
Development index
–0.364**
–0.168**
–1.025**
–0.232**
(0.009)
(0.006)
(0.018)
(0.006)
Rough terrain
–0.002*
0.000
0.004**
0.004**
(0.001)
(0.001)
(0.001)
(0.001)
August
–0.016
0.101**
(0.030)
(0.036)
September
0.089**
–0.134**
(0.029)
(0.037)
October
–0.111**
–0.163**
(0.029)
(0.037)
November
–0.069*
–0.267**
(0.029)
(0.037)
December
–0.179**
–0.414**
(0.029)
(0.038)
Constant
3.074**
1.997**
(0.016)
(0.008)
Cut 1
–0.960
1.248
(0.028)
(0.028)
Cut 2
0.396
2.677
(0.028)
(0.031)
N
54,952
54,952
66,131
66,131
rho
0.144
0.247
Source: Hamlet Evaluation System, United States Department of Defense.
Note: GLS = generalized least squares.
* p < .05; ** p < .01.
http://pas.sagepub.com
Downloaded from
by Natalia Spychalska on November 22, 2007
© 2007 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
196
POLITICS & SOCIETY
Figure 1A.
Probability of government control by identity group (1969).
Figure 1B.
Probability of government control by identity group (1971).
least squares (GLS) random-effects models, assuming a one-period autoregressive process and treating the dependent variable as approximately continuous.
Although the additional variables are, as expected, important determinants of government versus rebel control, in general, our measures of local ethnic predominance continue to be statistically significant and substantively important.
Chinese, Cao Dai, and Khmer identity each fail to reach statistical significance in one out of four models. The rest of the coefficients are highly significant.
In Figures 1A and 1B, we report a series of simulations of the probability of government control conditional on each religious or ethnic identity (for the ordered logit models) in 1969 and 1971. In all cases, we assume a rural hamlet with mean
http://pas.sagepub.com
Downloaded from
by Natalia Spychalska on November 22, 2007
© 2007 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
STATHIS N. KALYVAS and MATTHEW ADAM KOCHER
197
development level and mean terrain.47 Note also that the groups are arrayed along the x axis in an arbitrary way so that they can be compared to each other.
First, we simulate a “majority model,” which assumes a predominantly Vietnamese Buddhist hamlet and has a 0.43 probability of government control for 1969. Hamlets of all the minority religious groups had at least a 23 percent greater probability of being in government control. The ethnolinguistic minorities were more like the majority than religious minorities. As expected, Khmer hamlets were significantly less likely to be in government control, Chinese had a higher probability, and Montagnards were statistically indistinguishable from the majority. The results are somewhat weaker for 1971, which is unsurprising given that all hamlets were much more likely to be under government control in that period.