Publications and Data
Google Scholar: https://scholar.google.de/citations?hl=de&user=GZCW0DQAAAAJ
ORCID: https://orcid.org/0000-0003-4529-0593
My research examines party group switching, political representation and party institutionalization in the European Parliament. Below are my two published journal articles as well as the data and replication code.
Originally, I did all of my analysis using Stata. The downloads include those original Stata materials. Because I myself no longer have an institutional affiliation or a Stata license, I used AI to convert the Stata .dta files into CSVs and port the analysis code to Python. The original Stata logs are retained as references for checking the results. These converted files can be opened in Python, R or other statistical software. I did initial replication checks using the converted CSVs and Python scripts to make sure that the conversion was accurate, and the results are also posted. Each Python package includes its input data, code, saved outputs and documentation, so a Stata licence is not needed to use it.
This too in moderation
Evaluating the effect of delegate dissatisfaction on ambitious party group switching in the European Parliament
Aaron R. Martin. Party Politics 29(4), 726–740 (2023; first published online in 2022).
This article examines whether the most dissatisfied members are also the most likely to change party groups. The analysis finds that moderate ideological outliers have the highest odds of an ambitious switch, while the most extreme outliers are less likely to leave.
Data and code
| Download | Contents |
|---|---|
| Python and CSV package | Archived Python port, CSV inputs, saved results, validation reports and an R script for the earlier multilevel models. |
| Original Stata archive | March 2021 datasets, analysis do-file and Stata log. |
| Original Stata code — .do | Unmodified archived analysis code; also included in the Stata ZIP. |
| Input data — CSV | 3,977 rows and 555 columns, before the archived analysis transformations. |
| Analysis data — CSV | 3,977 rows and 563 columns in the archived Stata analysis dataset. |
| Variable labels and metadata | Variable labels, data types and Stata value-label definitions. |
| Data provenance and checksums | Source files, dimensions and SHA-256 digests. |
Party group collapse and strategic switching
Party group collapse and strategic switching in the European Parliament
Aaron R. Martin. European Union Politics 22(3), 521–544 (2021).
This article distinguishes strategic party group switches from changes forced by group collapse. It examines how group institutionalization and the policy distance between members, national delegations and party groups relate to switching during 1979–2009.
Read the article Open access PDF
Data and code
| Download | Contents |
|---|---|
| Python and CSV package | Python code, CSV inputs and reference datasets, saved outputs and validation reports. |
| Original Stata replication archive | December 2020 replication datasets, do-file, log and saved Stata figure. |
| Original Stata code — .do | Unmodified archived analysis code; also included in the Stata ZIP. |
| Input data — CSV | 3,977 rows and 206 columns, before the replication transformations. |
| Analysis data — CSV | 3,977 rows and 277 columns in the Stage B Stata reference dataset. |
| Variable labels and metadata | Variable labels, data types and Stata value-label definitions. |
| Data provenance and checksums | Source files, dimensions and SHA-256 digests. |
Replication scope: the Python port includes data construction and selected statistical results. Its saved reports document comparisons with the original Stata datasets, Table 2 t-tests, Table 4 Firth models and Appendix B. Some figures, descriptive tables and robustness checks are outside the port’s scope. See the included README for details and numerical tolerances.
Using the data
Download an individual CSV to explore a dataset, or extract a complete Python package to run the archived analysis. For example:
import pandas as pd
data = pd.read_csv("PP_dataset.csv", low_memory=False)
print(data.shape)CSV files retain numeric category codes; the accompanying metadata provides label definitions. Dates are stored as text and empty fields represent missing values. CSV does not retain every Stata storage attribute or distinguish all of Stata’s extended missing-value codes. Use the original .dta files when those details matter.
The CSV downloads were checked against their corresponding Stata files for dimensions, column names and cell values, with a small tolerance for numeric precision. Saved model-validation reports are supplied from the earlier replication work; packaging the downloads did not rerun the full statistical analyses.
Anyone can use the data for whatever they want, just please cite the appropriate articles (CC BY 4.0). Download checksums
Validation reports
These are the saved reports from the earlier Python replication work. They record comparisons with the original Stata datasets and logs. They were not regenerated when preparing this page. The Party Politics reports refer to the March 2021 analysis; their table numbering and multilevel-model discussion do not describe a verified replication of the final published article.
Python port validated against the original Stata datasets (_A/_B/_C) and replication_log.log.
Data pipeline (cell-level agreement)
- Stage A: 958457/958457 cells match across 241 columns, 3977 rows -> 100.000%
- Stage B: 1081744/1081744 cells match across 272 columns, 3977 rows -> 100.000%
- Stage C: 795/795 cells match across 15 columns, 53 rows -> 100.000%
Table 4 - Firth penalized logit (vs replication_log.log)
- all_switch_full: N=3962 (Stata 3962), logLik*=-637.68894 (Stata -637.68894)
- indiv_switch_full: N=3962 (Stata 3962), logLik*=-455.84114 (Stata -455.84114)
- bloc_switch_full: N=3962 (Stata 3962), logLik*=-326.97514 (Stata -326.97514)
Table 2 - t-tests (23 tests)
-> 23 passed, 0 failed
Table 4 - Firth models (7)
OK all_switch_full N=3962 (stata 3962) logLik=-637.68894 (stata -637.68894) OK indiv_switch_full N=3962 (stata 3962) logLik=-455.84114 (stata -455.84114) OK bloc_switch_full N=3962 (stata 3962) logLik=-326.97514 (stata -326.97514) OK all_power N=3962 (stata 3962) logLik=-1288.81064 (stata -1288.8106) OK all_power_col N=3962 (stata 3962) logLik=-697.40471 (stata -697.40471) OK indiv_power N=3962 (stata 3962) logLik=-477.35726 (stata -477.35726) OK bloc_power N=3962 (stata 3962) logLik*=-361.04744 (stata -361.04744)
Appendix B - McElroy (2 models)
OK mc_logit N=545 (stata 545) logLik=-203.50908 (stata -203.50908) OK mc_firth N=579 (stata 579) logLik=-193.07426 (stata -193.07426)
TOTAL: 32 passed, 0 failed
Data pipeline (dataset_12222020 -> PP_dataset)
- 2219166/2219166 cells match across 558 columns, 3977 rows -> 100.000%
Firth session_switch models (Table 5 robustness) vs march25_2021.log
- OK as_firth_intense: N=3977 logLik*=-643.86901 (stata -643.86901)
- OK as_firth_quad: N=3977 logLik*=-568.22092 (stata -568.22092)
- OK as_firth_full: N=3977 logLik*=-556.95429 (stata -556.95429)
Firth: 3/3 match
Multilevel Tables 4/5 (xtmelogit)
- Not reproduced here: no software available that matches Stata’s adaptive quadrature; models are convergence-sensitive. R/lme4 scripts for the model structures are in make_pp_multilevel.R.
What the reported models actually are
Tables 4 and 5 are three-level nested mixed-effects logits, not two-level:
xtmelogit <dv> <covariates> || start_group_num: || bloc_id3:a random intercept for the EP party group (start_group_num, 25 groups) AND a random intercept for the national-party delegation (bloc_id3) nested within it. lme4 equivalent: (1 | start_group_num/bloc_id3).
Ground truth (from march25_2021.log), Table 5 (amb_switch)
| Column | Model | logLik | key coefficients |
|---|---|---|---|
| 4 | intensity | -563.988 | z_intense2 -0.3592, c_npd_perc -11.160, c_epg_perc -9.314, govt2 -0.7115, mep_tenure 0.3726, female -0.7780, _cons -5.6818 |
| 7 | quadratic | -513.389 | a1_std2 1.3359, a2_std2 0.6217, a1^2 -0.1782, a2^2 -0.0933, c_npd_perc -12.744 |
| 10 | full | -493.246 | a1_std2 1.6026, a2_std2 0.8869, a1^2 -0.2034, a2^2 -0.1150 |
(The published table5.txt matches the intensity and quadratic columns exactly; the “full” column differs slightly - 1.624 vs 1.6026 - suggesting the printed table was assembled from a marginally different model run than this log.)
Reproduction status
- Exact Stata outputs are preserved in the log, so the results themselves are recoverable at full precision without Stata.
- Independent reproduction was attempted in Python and did not succeed here: a two-level fit is systematically ~20-40% attenuated (wrong structure), and the correct three-level nested fit could not be optimized within the sandbox’s compute limit (nested quadrature is expensive; Stata itself hit “not concave” iterations, so the models are numerically delicate).
- Recommended path: run
make_pp_multilevel.R(corrected to the nested(1 | start_group_num/bloc_id3)structure) in RStudio and compare its output to the ground-truth values above.lme4will be close but is not guaranteed to match Stata’sxtmelogitto the decimal.