Regression or Mixed Model?


I have a data set with 80 patients. At baseline we measured all with a symptom severity measure (t1) and three other variables of interest (v1, v2 and v3). I have followed upp all patients one time at 12-18 month after baseline assessment. In the follow-up we repeated the symptom severity measure (t2) employed at baseline. I want to explore if v1, v2 and v3 effect symptom change.

If I run a linear regression with t2 as the dependent variable and t1, v1, vt2 and v3 as independent variables, I get non-significant results.

If I run a linear regression with a change score variable (i.e. t1-t2) as dependent and v1, v2 and v3 as independents, I get non-significant results.

However, if I perform a repeated ANOVA or Mixed Model with t1 and t2 as repeated measures and v1, v2 and v3 as covariates both v2 and v3 becomes significant in the model affecting symptom change.

Why is this? Which statistical analyses are best suited to answer the research question: Are v1, v2 and/or v3 predictors of symptom change in this group of patients?

All Best,


TS Contributor
In the third model, it is not the main effect of v1, v2, v3 you are interested in, but their interactions
with the repeated-measures factor..
The interaction e.g. between v1 and the repeated-measures factor tells you whether the magnitude
of change over time is affected by the level of v1.

By the way, "significant" could mean 0.049 or 0.00000...1 , so it would be better if you describe
actual p-values.

With kind regards

Thank you Karabiner!

So it's actually two different hypothesis I am testing with regression and a repeated measures model?

I am interested in if v1, v2 and v3 at baseline could be important predictors of the curse of the disorder, i.e. how it goes for patients (as measured with the symptom severity measure). Which statistical analysis would in your opinion be best suited to answer this question given the variables at hand? All patients differ in their initial symptom severity level and the initial symptom severity level is also correlated with v1 and v2 at baseline.

Here are some p-values.

Linear regression with t2 as dependent and t1, v1, v2, and v3 as independent variables gives this: t1 (p=0.46); v1 (p=0.82); v2 (p=0.11); v3 (p=0.23).

A mixed model with t1 and t2 as repeated measures and v1, v2 and v3 as covariates gives these p-values: v1 (p=0.09); v2 (p=0.002); v3 (p=0.17).

I am very thankful for your help!