I am running a multiple regression analysis for 2 predictors (parental conditional positive regard, parental conditional negative regard) and 1 DV (resilience). All are continuous. I wish to investigate if responses differ significantly by gender, e.g. are females more prone to the effects of my predictors than males.
How does one go about this in SPSS? Do I enter gender as another predictor or as a control variable? Thank you!
In addition to nice suggestions of Tiziano, you can also divide your sample into two subsamples of "male" and "female", and run your models. Then, you'll be able to compare results based on a gender gap.
Hi Zw Teo,
what I think you should do here is to dummy code your gender variable. To do it you should transform those “m”s and “f”s I guess you have into numbers, namely 0 and 1, respectively, and then add this variable as one of the predictors in your regression model. You can dummy code manually or use the “Recode into different variables” function in the Transform menu. I find it much simpler to learn from videos, so I would really recommend you to check some videos out on youtube. Search for the keywords “spss dummy coding” and you will get a lot of simple tutorials.
For what concerns the second question, whether you want to control for gender or use it as a factor it is up to you. In the former case you will measure how strong the effect is regardless of the effect of gender. In this case, in a hierarchical regression you insert the gender factor first, and then add the others, so that their effects are evaluated after the one of gender.
If you want to test gender as an additional factor, which is indeed what you're looking for, you have also to add the interaction terms between gender and other factors to the model. You do this by creating a new variable for each interaction, each one being the product between the interacting factors (go to Transform->Compute Variable... and multiply the variables).
Hope it helps!
In addition to nice suggestions of Tiziano, you can also divide your sample into two subsamples of "male" and "female", and run your models. Then, you'll be able to compare results based on a gender gap.
The UCLA web-page linked below illustrates the approach Tiziano described. HTH.
https://stats.idre.ucla.edu/spss/faq/how-can-i-compare-regression-coefficients-between-two-groups/
It is worth noting that the interaction approach is superior to the separate analysis of the two groups. For example, an effect could be statistically significant in one group and not the other yet there be no significant interaction between gender and that predictor.
Dear Zw Teo,
Perhaps a moderation analysis will be helpful to present the difference of association between your IVs and DV for males and females.
If you wish to try this than I recommend the add-on called "PROCESS" by Preacher and Hays:
http://afhayes.com/spss-sas-and-mplus-macros-and-code.html
It is very easy to install and use - you can find everything you need in Hays's website.
Good luck.
Osnat
Just to clarify: a moderation analysis is just another name for adding an interaction term. The PROCESS SPSS macros will be equivalent to manually entering a product term for the interaction effect (and maybe making it easy to centre predictors) unless you want to look at more complex models such as moderated mediation.
I agree with you, Tom S. Baguley. It'll save the manual procedure and enable a better look at the pattern of results.
Another option is to use GLM or UNIANOVA rather than REGRESSION to estimate your model. GLM and UNIANOVA allow you to enter categorical variables as "fixed factors" (as oppose to "covariates", which is what scaled explanatory variables are called. Doing it this way means that SPSS takes care of dummy-coding the categorical variables internally, thus saving you the trouble of computing them yourself. And you can simply include any interactions you want on the DESIGN sub-command. E.g., if you have variables A and B, and you want their interaction: /DESIGN A B A*B.
HTH.
Thank you all for your help.
I am trying to determine the moderating effect of gender for each predictor. However, I have 2 scores for each predictor - a mother and a father score. Does Tiziano's suggestion mean that I compute an interaction term for the mother score and gender, followed by the father score and gender, and simply add them to the model? Or must I compute a mean score (mother + father) and only then do a moderation?
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS