I have one criterion variable and three predictor variables. I wish to conduct multiple linear regression for two groups (e.g. female and male) of data with the same DV and IVs and compare fit and structures (weights) between the two groups. In fact, I am able to conduct regression analysis for one group in SPSS by having DV in one column and IVs in three other columns, in the data window of SPSS. But, I am confused as to how regression analysis shall be conducted for two groups. Some examples are there, which are not easily understandable. I think, data need to be prepared accordingly with some coding. If there is any simple example, I will follow that and do my experiment. Also, is there any procedure to test fit and structures between two groups. If so, please help me. Thank you.
Hello Srinivas,
Here is a closely related thread from Researchgate: https://www.researchgate.net/post/How_can_I_compare_predictors_between_two_groups_in_regression_analysis
For me, the simplest approach is to use SEM (structural equation modeling) software. Another method is a simultaneous (multivariate) test of intercepts and predictor coefficients that was proposed by Richard Pothoff.
Here's a link on how to evaluate equality of regression coefficients across groups using the R library, lavaan: http://lavaan.ugent.be/tutorial/groups.html
Good luck with your work!
Srinivas -
If you want to compare the two models with regard to regression analysis, perhaps the most straightforward method would be to do a graphical residual analysis - predicted y on the x-axis, and estimated residuals on the y-axis - with both the male and female models on the same scatterplot. I think you might get a nice 'picture' of the comparison.
Just, as in any evaluation, however, don't overstate the results. Trying again with more data later to confirm results might be a good idea, or bifurcating, taking the data in two random sets, and noting any difference in results might be useful. That's just a thought.
If you do that, the estimated regression coefficients which provide results for the graphs will vary, but you can see how they, and their estimated standard errors for given sample sizes, range.
Since the two models won't have exactly the same predictions to compare estimated residuals, you would have a mix of predicted y values found on the x-axis, but I don't see that causing any problems.
Cheers - Jim
Thank you James and David for your immediate responses. I will certainly look into your suggestions. Actually, what I did was this. I have conducted MLR for female and male groups separately (with the 'same' three IVs and one DV). 'same' means statements are same for each construct/variable. I want to compare model fit and structures of both female and male groups. Hope, I am clear. Please help me.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS