Conceptual points:
- The analysis is mostly well done and replicates. The paper is readable and understandable. The results displayed in the text almost always correspond to the ones in the text.
- The paper is unclear whether it is arguing for a Null result or not. It is also questionable whether the results are a "precise" Null.
- The paper misses this important, related paper: https://doi.org/10.1017/gov.2021.40
- The Green Party result appears driven by the fact that the Green party only has ever had two district MPs
- Figure 1 is somehow broken. The text description does not match the visual
- All absences, even excused absences, are considered deviations. This is to some extent a defensible choice, but not necessarily the standard.
- Given the very low rate of deviations, a LDV model might be beneficial
- The model could consider leadership positions (presumably, leaders follow the party line more often to avoid being stripped)
- Restricting to 1983 is not directly convincing, as no evidence is shown of roll calls being more common from then on
- It is unclear why the actual DDD specification is moved to the appendix. The presentation of this is flawed (e.g. t-statistics presented as SE, some faulty rounding)
Coding points:
- The merge process leads to a lot of missing values for gender, leading to a 20% drop in sample size
- Seniority is constructed wrongly in the code and is dropped by feols. This is not commented further in the paper
- The RDD specifications are not always fitting regarding the research questions and are sometimes not proper RDDs (uniform kernels with a large cutoff and linear slope only as opposed to the advertised rdrobust)
Contact: jan.ringling@econ.uzh.ch
Conceptual points:
Coding points:
Contact: jan.ringling@econ.uzh.ch