This study has several limitations. For example, the distribution of the primary outcome variable (see Annex C) has clusters around tens (especially around 0, 50 and 100), which can have altered the results. This was likely due to the design of the survey (the answers to some questions were given on a continuous interval, with every tenth highlighted). This limitation was addressed by running several robustness checks (Logit, Tobit, OLS with rounded frequencies, means testing).
Another reason for the peaks at tens (for instance in the variables Age, where peaks can be identified at 30, 40, 50 and 60), may have been that the respondents gave approximative values for their age. The respondents may have feared being identified from their responses, even if the experiment was anonymous. Nevertheless, the Age-variable still is approximatively normally distributed.
Due to the lack of data on the distribution of age, career lengths in the public administration and agencies, the study was not able to confirm whether the distributions of these variables follow the population distribution, to see whether the sample is representative of the underlying population. These factors have important implications for the external validity of the results.
To have power of 80% to find an effect size of 2.9-4 percentage points, 3000-5700 approximately observations was needed (see chapter 3). Due to high attrition, i.e., respondents not finishing their survey responses, almost half of the responses were excluded due to incomplete responses. Yet, since the effect sizes found by the regression in this study were much higher (8.21-12.00 pp, comparatively 8.99-14.46 estimated by the statistical tests), the study found sufficiently high effect sizes and gathered large enough sample, to attain statistical power of 80%.