Hello everyone,
May I know what is the major difference in examining interaction between two IVs on a DV by adding interaction term into my model instead of stratifying the data ?
I was trying to examine the interaction between plasma calcium and ethnicity on pregnancy induced hypertension (PIH) . I found that there is an interaction between plasma calcium and ethnic groups in my study . Significant interaction of plasma calcium on PIH was observed in ethnic group 1 and ethnic group 3 .
To have a closer look of the effect of plasma calcium on PIH across three groups , I went on doing a subgroup analysis by running a regression model in each ethnic group individually , but the effect of plasma calcium in ethnic group 3 ended up being no longer significant despite the same direction of effect was observed .
In this case, which analysis should I rely on? I think the cause of difference is probably due to the fact that, the subgroup analysis itself does not take the interaction effect into account?
As far as I know stratification is an alternative to examine interaction term between IVs on DV, it is just that it is less favourable esp when we have smaller sample size?
Any comments or inputs would be greatly appreciated. :)
Thank you in advance.
Em
Dear Emerald Chang
You can Experimentation by using spline regression as a nonparametric method which is more suitable with a small size of the sample.
This is a complex theme to discuss in few words. Well, an alternative to the regression inside each ethnic group is to represent this nominal variable in the regression model through dummy variables. It can be the way to get a more compreensible model, I guess...
https://www.moresteam.com/whitepapers/download/dummy-variables.pdf
I think you already gave the answer to yourself: subgroup analyses are less favourable since they lack power (although the power concept and frequentistic approach are flawed in the first place...), what could simply explain your results. Do simple slope analyses, which are pretty simple to do. If you use SPSS, it gets even simpler with the PROCESS macro by Andrew Hayes. Generally I can recommend to read some textbooks on this topic, either the classic one Cohen, Cohen, West, & Aiken (2003). Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences or more recently Darlington & Hayes (2016). Regression analysis and linear models: Concepts, applications, and implementation. Or if you want to go more into detail Hayes, A. F. (2018). Introduction to mediation, moderation, and conditional process analysis (2nd ed).
Dear Emerald Chang
You can Experimentation by using spline regression as a nonparametric method which is more suitable with a small size of the sample.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size