I'd like to check whether women and men differ on those five variables, that can be correlated with each other. The groups are also not equal with 589 females and 293 males.
The solutions that come to me are:
-
(a) five t-tests with Welch correction and 'sex' being a grouping variable, but I think it's not so good solution, 'cause I will not take into account that these five variables are inter-correlated.
-
(b) extension of t-test: Manova with five dependent variables and sex as a grouping variable. This would show how the groups differ across these five variables. It seems convincing to me, but I am wondering if the next proposal better.
-
(c) logistic regression with sex as a dependent variable with five independent variables (those moral codes), entered simultaneously. I think that this approach will show me which moral code 'predicts' the probability that the respondent is a male/female controlling for other moral codes.
Being aware of the fact, that 'numbers does not know where they come from', I'd like to consult with stackers, if what I figured is sensible and best examines the differences between both sexes.
Context: This question is going to be quite relevant to current political situations all over the world.
My aim is to compare males and females in the context of five dependent variables that represent moral codes. (I'll skip how to measure these codes but as a sidenote I do have to say that it's quite interesting why it's better to give a heated discussion with someone who exhibits different pattern of moral codes).
Best Answer
Option (c) makes the least assumptions and is the most robust. It works because you can't predict sex if the distributions of the other variables are identical for the two sexes. This approach was developed by O'Brien: https://www.tandfonline.com/doi/abs/10.1080/01621459.1988.10478564