1

Perhaps a basic question but I would appreciate some explanation on this topic :)

I have a dataset covering approximately 30,000 loan applications. Besides characteristics of the loan (e.g., interest rate, loan duration, etc.) and borrower (e.g., age, educational level, etc.) it includes the year that the applicant applied for the loan as well as the applicant's country.

I want to investigate if gender has an influence on the interest rate of a loan. In deciding whether to use a fixed-effects model, I was wondering whether it would be possible to include both a country and a year fixed effect since my main variable of interest (gender) is time-invariant, and what could be the reasons for including fixed-effects in the first place?

Juliuslah
  • 11
  • 2
  • 1
    Why not? Maybe tell us some more details like sample size ... – kjetil b halvorsen Aug 21 '23 at 16:07
  • 1
    If possibly gender is associated with country or year (for instance, some countries have more females applying for loans or the proportions of females changed over time), then if you don't include ("control for") country and year, you won't have a basis to attribute any difference you observe to the gender, because it could just be caused by those other factors. Thus, it's crucial that your statistical modeling accounts for country, year, and any other factor that might be associated with gender. Search our site for "control for," "mediator," "moderator," "confounding," etc. – whuber Aug 21 '23 at 16:58
  • @whuber Maybe I am misunderstanding you, but wouldn't we want to use do calculus to identify the valid adjustment sets for the estimand instead of dog piling variables into the model that might be associated with gender? – Galen Aug 21 '23 at 17:07
  • @Galen I am not advocating throwing all possible variables in: just the ones that reasonably might be associated with gender. I cannot figure out how one would "do calculus" in this context or ascertain what you might mean by "adjustment sets." – whuber Aug 21 '23 at 17:34
  • 1
    @whuber Ah, sorry I meant do-calculus not "do calculus". Even so, I think my previous comment was presuming that we were considering interventions, which doesn't seem appropriate for this post. – Galen Aug 21 '23 at 19:29

0 Answers0