between reform events are coded as “before” or “after” depending on their closeness to the
particular reform event.53 The analysis of variance tests for significant differences in the mean of
economic and fiscal variables before, during, and after the reform event. Where the F-test
indicates significant overall differences between means, a pair wise test for differences in mean is
added54 (employing the Bonferoni correction for multiple testing). Translated into the terms of
the stylized reform cycle above the test algorithm relates to differences between periods BR, R
and SA in combination (analysis of variance) and, subsequently, for differences between BR and
R, BR and SA and R and SA (pair wise tests).
Figure 2 depicts the evolution of the means in government primary net lending. Table 3
summarizes the results of the testing, figures A-1 to A-4 in the appendix show the development of
means of important further fiscal variables over the reform cycle.
53 Single years between reform events are dropped; the same applies to the central year of a three year reform pause.
54 The analysis of variance only allows inferring whether differences in mean among included classes are significant in total without saying which
pairs of means are significantly different. Simple t-tests would lead to biased results due to the problem of multiple testing.
146