If I want to examine sex differences in three variables, lets say academic attainment, study motivation, and a variable that is categorical. How should can I fit these variables with OLS regression? Should I run one regression analysis for each dependent variable and have sex as predictor and maybe include other demographic variables as control variables? Or should I do everything in one analysis? The point is to examine gender differences in the three dependent variables with OLS regression.
$\begingroup$
$\endgroup$
6
-
1$\begingroup$ There are ways (and reasons) to do either of those approaches. E.g. psycnet.apa.org/record/1992-97624-030 $\endgroup$– Jeremy MilesCommented Mar 28 at 17:10
-
$\begingroup$ i dont have access to that article:-( but when the dependent variable consists of three categories is it then logistic reegression or could you still use OLS regression $\endgroup$– user409571Commented Mar 28 at 17:18
-
$\begingroup$ What is the goal of the examination? Why do you want to use OLS regression and what is the problem you encounter in using OLS regression? $\endgroup$– Sextus EmpiricusCommented Mar 28 at 20:12
-
$\begingroup$ i want to examine sex differences in each of the three variables (academic attainment, study motivation, and stress). And i was suggested to use regression, but maybe another method is better? One of the dependent variables is categorical (3 categories), so i dont know how to deal with that in an regression $\endgroup$– user409571Commented Mar 28 at 20:35
-
$\begingroup$ Do you have control variables? $\endgroup$– Jeremy MilesCommented Mar 28 at 22:22
|
Show 1 more comment
1 Answer
$\begingroup$
$\endgroup$
The easiest way would be to turn the model around and predict sex as function of those other three variables instead.
Then you can use approaches like logistic regression which are easy to apply.
A good indication might already be obtained from a scatter plot for the two continuous variables and see how they differ. In addition you can make three times such a plot for the three conditions of the categorical variable. Is the distribution simple, like two Gaussian distributions or something that looks like it, then you could perform something like MANOVA to model the three variables together as function of the sex variable (if the behaviour for different levels of the categorical variable is a lot different then you could also add the categorical variable as an independent variable, and model effectively only two dependent variables). Performing three seperate regressions might work as well, but potential differences in the multivariate view may not be visible (see When is MANOVA most useful).
It is unclear what you mean by "to examine sex differences". If your goal is to prove some theory then the approach of logistic regression may already work. But if you examine the differences because you want to make predictions of the three variables, then a model that predicts sex based on the three variables is not much useful and you need to compute it the other way around.
answered Mar 29 at 9:17