The probability of being criticized can be analyzed more systematically with an ordered logistic model using as a dependent variable the most critical summary assessment by any organization present. Such analysis need only include monitored elections, so the potentially endogenous relationship between monitoring and criticism is eliminated. Because it is concerned with criticism of the election, however, this analysis needs to control for the quality of the election. To avoid bias from the monitoring organizations themselves, the two overall measures from the U.S. State Department dataset (“overall assessment” and “level of problems”) are used as a third-party evaluation of the acceptability of the election and the level of problems in the election.1
Table B.1 shows that solitary monitoring organizations are less likely to be critical. Indeed, simulations2 based on the model in Table B.1 suggest that when an organization is present alone, then the likelihood that an election is criticized drops from about 13.2 to about 4.1 percent, or about a two-thirds drop.3 The overall percentages differ from Table 3.2 in Chapter 3 because the other quality variables are included in the analysis, but the relative decrease in the likelihood is remarkably similar.
TABLE B.1
Determinants of most critical overall assessment by monitors
Only one organization |
0.276*** (0.095) |
Election quality# |
8.200*** (3.143) |
Level of problems# |
2.504*** (0.618) |
Observations |
328 |
Log likelihood |
−230.3 |
Pseudo r-squared |
0.270 |
Number of clusters |
114 |
Notes: Ordered logit, robust standard errors clustered on countries (in parentheses). For simplicity of presentation, the two cut-values are not reported. For discussion of variables, see the Appendix A.
#These are the overall assessments variables from Table A.1 in Appendix A.
***p < 0.01, ** p < 0.05, * p < 0.1.