Hello,
I'm interested in using linear mixed models to explore the relationship between several demographic variables and changes in a measurement score over time.
I examined variables such as age, gender, and education across three time points.
Firstly, I ran a crude LMM analysis and assessed the interaction term for each variable. The model structure was defined as: score = time + age + time * age.
Secondly, I ran an adjusted model where I tested the interaction terms while controlling for other covariates.
Results: age and gender, the interaction terms were found to be statistically significant.
For education, the main effect was significant, the interaction term was not.
I have a couple of questions:
1. Concerning education, does the significant main effect have any clinical significance?
2. Given that the interaction term for education was not significant but the main effect was, should I include education as a covariate in the adjusted model?
3. For both age and gender, considering their significant interaction with time, would it be appropriate to proceed with testing the three-term interaction?
4. The linear mixed model assumes that the outcome is normally distributed, while some argue that the normal distribution applies to the residuals instead. I'm confused about which one is correct, and when I should consider invoking the central limit theorem (I have a sample size of around 1500.)
I appreciate any insights you may have regarding these queries. Thank you.
