Anticipation Hypotheses. Next, I test the hypotheses regarding anticipation, or the theory that protests and repression are more likely to occur in the month immediately prior to the beginning of the agreement. The mechanism in this case could be that protesters try to prevent a program from being implemented or try to get the government to renege on its deal with the IMF. In order to examine these hypotheses, I first create an indicator variable that takes the value of “1” when the month is immediately prior to the start of an IMF program. I then subset my data to compare these “anticipatory” months with the rest of the control months (months not under an IMF program). The results from the OLS and fixed effects regression models indicate that the month immediately prior to the beginning of a program is correlated with more protests (23 or 10, depending on the model) and more repression (10 or 6.5 more repressive events). These results are significant at the 99% confidence level. At first blush, it would appear that there is some evidence for the hypotheses that protests and repression increase in the month prior to the implementation of an IMF program, perhaps because citizens and governments anticipate the negative effects of that program. However, I also ran a matching model, as before, which produced starkly different results. There are two caveats with using matching for testing this set of hypotheses. First, the sample size is much smaller because there are only 210 anticipatory months in the dataset, meaning that there are approximately a tenth of the number of matched pairs, compared to matching to test the backlash hypotheses. The second caveat is that covariate balance is much poorer for these hypotheses – see the figure below. Notice that though the blue line (representing the absolute mean difference between the observations for the adjusted pairs) is to the left of the red line (absolute mean difference for the unadjusted pairs), both lines are still far away from the target value of 0. This means that comparison the two groups may not be appropriate and may produced biased results.
