In my research, I conducted multivariate regression using SPSS, involving a single continuous predictor and multiple continuous dependent variables. While I have interpreted the results, I am uncertain about how to effectively report them in a research article. Could you recommend any exemplary articles that I could refer to for guidance? Additionally, I would appreciate insights into why multivariate regression is not as commonly utilized in research compared to other methods like MANOVA (when the predictor is categorical) or multivariate logistic regression. I am concerned that I may have made an error in my approach or that multivariate regression is not the appropriate method for my research question.
Multivariate regression is pretty standard today, it comes down to the nature of your variables. In your case, as your predictor is continuous and you have multiple continuous DVs, multivariate multiple regression is appropriate as it lets you assess shared variance and protects against Type I error (assuming the assumptions are ok).
How to report - Just use google scholar and search for "multivariate regression" and whatever field is related to yours, a typical phrasing would be something like "A multivariate multiple regression was conducted to assess whether [predictor] significantly predicted [DV1], [DV2], .... The overall multivariate test was significant, with [Wilks = .., F = ..., ...], indicating that ...
Might also be a good idea to follow up with univariate regressions
Marius Ole Johansen Thank you for your insight. How do you think about using Bootstrapping for violation of error of variance for Multivariate regression?@
If you are referring to a violation of heteroscedasticity, then no - bootstrapping won't fix this at all, it will only change how you estimate inference (confidence intervalls).
There are a few ways to deal with heteroscedasticity issues though,
1) Use robust standard errors (HC3 estimators, don't know if this is in SPSS but you can use it in R)
2) transform the DVs (log-transform)
3) If you use SEM instead you can model residual covariances, which might be the best option
- Badges
- Science topic
- Similar topics
- Computer Science
- Research Topics