It is unclear what your dependent variable is. In the figure R2.PNG you label this dependent variable as 'linear prediction'. You say that the treatment type was implemented after the first time point (ftime0). Yet, in the figure there is a huge difference in 'Linear prediction at that time 0 between AVF and AVG access type. I would suggest you use the measurements of 'Linear prediction' at time 0 as covariates in an repeated measures ANCOVA. See attachment R script

Hi Jos,

Thank you so much for your valuable reply.

I have updated the project description.

The plot above is for the marginal means at each time points across different levels of the access type. 'Linear prediction' label was automatically spitting out from R. Actually it is the estimated albumin measurement.

Dep. Var. : Albumin measurement (Continuous)

IV : access_type(2 levels), ftime(3 levels)

The code I used-

lmer(albumin ~ ftime*access_type + (1|sitelocation/serial), data)

Now, what do you think?

I think it would be good to control for the differences in albumin measurement at ftime0 (time point zero, or baseline); i.e., before the treatments were applied. This can be done by using the albumin measres at that ftime0 as covariates, as I did in the R script. This is a common statistical technique to control for preliminary (unwanted) differences between treatment groups.

I checked things using the example from Winer ("Statistical Principles in Experimental Design") on page 613. This is for a 2 x 2, between x within design, with 1 covariate (i.e., covariates not changing over time/trials). The method in which the covariate is included but NOT the interaction between trials and this covariate yields the same results as Winer on p. 614. See attached R script.

Hi Jos,

Thank you so much for the help. I need to think about whether this baseline adjustment fits well with my other research hypotheses. I will be back here after forming some other hypotheses. Also, I need to go over with your codes and references.

Thank you.