1

How do I conduct a power analysis for a multinomial logistic regression analysis? I have 1 independent variable (with 3 levels) and 1 dependent variable (with 2 levels). All variables are continuous. I cannot do ordinal.

Thanks!

Matt Mullens
  • 11
  • 4
  • If you have 1 dependent variable w/ 2 levels, you have *binomial* logistic regression, not multinomial. Do you actually have $\ge 3$ unordered response categories? If not, there is a great deal of information about power analysis in my answer here: [Simulation of logistic regression power analysis - designed experiments](http://stats.stackexchange.com/a/35994/7290). Even if it isn't exactly what you're looking for, it may be worth your time to read that & get a clearer sense of the issues involved, as I don't think there is enough information in your question statement for it to be answerable. – gung - Reinstate Monica Nov 27 '15 at 21:53
  • "All variables are continuous" is inconsistent with each variable having either 2 or 3 levels. Moreover, with just 2 non-ordinal variables, instead of using regression, you might be better off looking at chi-square and its related measures of association such as Cramer's V. – rolando2 Nov 28 '15 at 01:35
  • Thank you very much for all of your input. I appreciate both of you taking the time to respond to this. Based on your feedback I realized I needed to completely revise my statistical model based on the variables and study design. It turns out that a 3x3 mixed ANOVA is what winds up being best suited for the study. I feel like you both helped get me on the right foot, not sure what to do with the bounty but @gung if you want to paste that in as an answer I can select it. – Matt Mullens Nov 30 '15 at 00:53
  • Thanks. I think @rolando2's comments are more on point here, given what your actual situation is. I wonder if it might be more appropriate for him to paste in his comment as an answer? If not, I can add mine in a day or so. – gung - Reinstate Monica Nov 30 '15 at 01:00
  • Ok that sounds great. Yes @rolando2 please feel free to do that. Thanks again! – Matt Mullens Nov 30 '15 at 01:23
  • bounty expires in 3 days! Someone claim it :) – Matt Mullens Dec 03 '15 at 01:15
  • @gung You are always a class act. – rolando2 Dec 04 '15 at 03:37
  • Should this question be deleted or just reworded? The title is now very misleading and this might cause confusion. – Peter Flom Mar 21 '17 at 12:39

1 Answers1

2

"All variables are continuous" is inconsistent with each variable having either 2 or 3 levels. Moreover, with just 2 non-ordinal variables, instead of using regression, you might be better off looking at chi-square and its related measures of association such as Cramer's V.

rolando2
  • 11,645
  • 1
  • 39
  • 60