[Rstudio] Binary Logistic Regression - Interpretation

Hello all,

I am currently doing my master thesis and have difficulties fully understanding how to interpret the results of my binary logistic regression. The code and the plot of said code are attached. The plot is labelled in German, sorry for that.

Background info:

In Switzerland, almost all judges belong to a political party. I would like to analyze how this party political affiliation might influence the decisions judges take. Since my field of research is migration, I do this in looking at the decisions on asylum appeals.

My dependent variable is therefore the decision (coded 0 for rejected, 1 for accepted). My main independent variable is the party political affiliation; i.e. the party the judge belongs to. The party political affiliation is coded as categorical (i.e. actual party names). Furthermore, I have the control variables "Amtszeit" (duration in office) and "Geschlecht" (gender).

In the model, the reference party is a party called SVP. If I have understood it correctly, my results for the other parties are always compared to/in reference to the reference party.

code and model.png Resultat_SVP.png

If we take the comparison of the parties SVP and SP, the log(odds) is 0.498532. If I convert this into odds and probability, I receive 1.6463 for the odds and 0.6221 for the probability. Here's my question:

Do I interpret the data correctly when I say that a judge from the party SP is 62% more likely to accept an asylum appeal if compared to an SVP judge? Or more general, an asylum appeal is 62% more likely to be accepted if it is decided by an SP judge rather than an SVP judge?

Thanks in advance.


Less is more. Stay pure. Stay poor.
You used multilevel model, so please clarify if results are clustered in judge, sample size, etc. Also, when reporting estimates include confidence intervals.
Thanks for your reply, hlsmith.

The data is clustered in 52 groups, those being the individual judges. Sample size is 19‘950 observations. Confidence interval for the mentionned relationship between SVP an SP is 95%.


Less is more. Stay pure. Stay poor.
Given your description, my interpretation would be:

SVP has a 1.64 (95% CI: ??, ??) times greater odds of blank related to asylum appeals compare to SP when controlling multilevel model with decisions clustered within Switzerlandian? judges.