1

I have a yes/no dependent variable; independent variables are also a combination of yes/no variables and one 3-level sexual identity variable. Also hoping to control for for 2 of the yes/no variables and one categorical variable. What analysis is appropriate? Should I be looking at binary?

Cindy
  • 31
  • 2
    You are sparing with the details, but it does sound like logistic regression would be an appropriate model for your study. – dipetkov Aug 31 '22 at 15:22
  • Thank you. What is the best way to control for the particular variables when doing binary logistic? – Cindy Aug 31 '22 at 16:04
  • 2
    "Control for" X means that we include X as a predictor. A better term is "adjust for". We don't control someone's age for example. I suggest that you read up on logistic regression before applying it to your data. – dipetkov Aug 31 '22 at 17:02
  • It sounds like your goal is to infer [tag:causality]. If so, be aware that adjusting/controlling for a variable does not necessarily improve your inferences - it can even make them worse. This often happens in a situation called 'conditioning on a collinder' and I don't find a comprehensive answer on this site but there are several relevant threads, for e.g. https://stats.stackexchange.com/questions/431427/what-is-the-interpretation-of-a-collider-within-graphical-causal-inference-in and https://stats.stackexchange.com/questions/525369/which-ols-assumptions-are-colliders-violating – mkt Aug 31 '22 at 17:18
  • @dipetkov I certainly will. I just wanted to be sure I was in the right analysis before jumping in. Thank you! – Cindy Aug 31 '22 at 17:49
  • 1
    @mkt Thank you! I will read more into this. – Cindy Aug 31 '22 at 17:50
  • 1
    @mkt The subtle implication that there are causal relationships involved is the primary reason to avoid the term "control for". Or perhaps the term is popular precisely because of that. I think that "adjust for" doesn't have the same implication, but maybe that's not so. – dipetkov Aug 31 '22 at 17:55
  • 1
    @dipetkov Yes, agree completely. But it's not hard to imagine a world in which we switch the term we use but keep the underlying implication of causality. So we also need to keep banging on about the underlying problem too. – mkt Aug 31 '22 at 18:00
  • Oof, that should be 'conditioning on a collider, not collinder. – mkt Aug 31 '22 at 18:01

0 Answers0