I am looking to perform a binomial regression calculating the reselection chances of parliamentarians in the European Parliament based on their activity levels. DV= reselection: 0/1. My data is clustered. Individuals are grouped in parties which are grouped in countries (no cross-memberships). Therefore I want to use a multilevel logistic regression.
This would imply a three-level model: MP (n=1071) < party (n=202) < country (n=28). BUT I hit a wall when it comes to the sample sizes. There are only 28 countries. Each country only has a handful of different political parties (some only 2). There are many different political parties (202), but on average they contain only 5 MPs. Worse, there are 73 singletons (36% of my data).
So I feel stuck between a rock and a hard place: I cannot ignore the clustering, but I also cannot perform multilevel analysis (or maybe clustered SE) correctly on this data? Can anyone help me move forward with this? (I normally use and prefer working in SPSS, but I could try figuring things out in R if needed)
Any help is much appreciated! Isabelle
P.S.: Please let me know if it would be best I provide other specific information on the data or question.
P.S.S.: In a second step I would need to perform a multinomial regression or multilevel analysis with categorical outcome (DV = ballot position value: secure/marginal/hopeless). The population for this test is all parliamentarians that were reselected (thus scoring 1 on the DV from the previous model). Any pointers for this model would be appreciated too as the data has the same structure.