Anthony, you could include a dummy variable for gender (e.g., male = 0, female = 1) if you aim at testing differences between both groups in the dependent variable.

If you are interested in comparing regression weights of the two groups, you could include the dummy variable for gender and additionally interaction terms (products of the dummy variable and each independent variable) in the multiple regression (a single regression equation).

See, e.g., the following book: Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences, 3rd ed. Hillsdale: Erlbaum.

A third alternative would be to perform a multi-sample analysis and to test whether parameters are invariant  across both groups.

Does this help? Regards, Karin

Anthony, you could include a dummy variable for gender (e.g., male = 0, female = 1) if you aim at testing differences between both groups in the dependent variable.

If you are interested in comparing regression weights of the two groups, you could include the dummy variable for gender and additionally interaction terms (products of the dummy variable and each independent variable) in the multiple regression (a single regression equation).

See, e.g., the following book: Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences, 3rd ed. Hillsdale: Erlbaum.

A third alternative would be to perform a multi-sample analysis and to test whether parameters are invariant  across both groups.

Does this help? Regards, Karin

Include dummy variable to include different sex. I agree with the previous answer

What about the other predictors , are they scale , ordinary . I think you can use any statistical software such as SPSS, SAS to calculate a linear (o non-linear) regression with those variable types. Ordinal variables such as 3-point groups Like a scale 1 to 3 can be treated as continuous variable after we give each group define number .

Anthony,I agree with Karin.You can also,compare the relations between 2 sex by using Fisher Z.

Anthony, what is your research problem?

What is your hypothesis?

Thanks Karin, Arpan, Khalid, and Elaheh for the answers. They really helped me.

Eddie, my objective is to test for sex differences in the regression weights of four predictors of academic achievement.

Thus my null is: there are no significant sex differences in the regression weights.

Dear Abthony,

If you use only an intercept dummy for gender, this will not be reflected in the coefficients. Therefore you must use slope dummies and look for the significant results.

If your model ansd sample size allow for,  you can use Chow's test which, if significant, will show the significant (if any) coefficients you are looking for in the two original equations (one for each gender subsample).

Remember Hendry's advice - that the three most important rules in Econometrics are:  test, test and test..

Please feel free to continue our chat,  if you feel like it.

Regards,

Frederico

Anthony, as Frederico pointed out, the Chow test would be a good alternative if you have two groups. However, an assumption is that the residuals in both groups are homoskedastic.

If you want to test for sex differences in the regression weights using a single regression equation, the best option would be to include interaction terms formed of the dummy variable and all predictor variables. There is a nice explanation by Brad Jones which may be helpful: psfaculty.ucdavis.edu/bsjjones/dummy.pdf

Regards, Karin

Dear all, 

....or you can use CTA via ODa and analyse the effects of different types of variables in sex. The advantage is that you do not miss any information (= it happens when you use dummies, because you only compare one category with the reference one, but...do not analyse the reference one...and if you chance de choice of the reference one category your results would change,,), and also if you violate any assumption, the result would be  correct via ODA.

Helena

A different approach is to conduct two separate regression analyses, one for males, one for females, using the exact same variables and then subsequently compare the unstandardised regression coefficients (unstandardised beta values), using a Z-score to assess significance. This enables you to examine sex differences in multiple associations (and at different steps if you conduct a hierarchical regression analysis). I personally prefer this method to the dummy coded interaction approach suggested above when I am interested in examining multiple potential moderation effects. An additional benefit is that it enables you to examine whether your model accounts for an equal amount of variance in the two sexes (the R squared and adj. R squared). Unfortunately, however, I know no way of testing whether differences in (adj.) R squared are significant. Furthermore, it is a limitation to this method that it should only be used with relatively large sample sizes with a high number of both male and female participants, as you are dividing your original sample into two sub-samples. You can also conduct Z-scores to test for moderation effects in simple correlations if you are also interested in examining sex differences before controlling for other variables. If you are interested, you can see an example of the method being used in:

Christiansen, D. M., Olff, M., & Elklit, A. (2014). Parents bereaved by infant death: sex differences and moderation in PTSD, attachment, coping and social support. General Hospital Psychiatry, 36 (6), 655-661. doi:10.1016/j.genhosppsych.2014.07.012

Thank you Dorte for suggesting a new way and for sharing the article. I will certainly look at it.