I am doing a repeated measures analysis on global longitudinal strain data from echocardiography. I am using a LMM to analyse the variance components, as there was some missing data on patients for some factors, so rmANOVA didn't really fit. My original design was using the patient as a random factor , and the factors such as observer, timepoint (or the repeat), vendor , as fixed effects. The results of the this as the only random effect give me within patient and between patient variation and residual error. originally suggested by a college, my question is, can I used these fixed factors instead as random factors in the model , which then gives me percentages of the total variance for each factor? I am aware ofcourse they are not truly random variables as in not being samples of a greater population of observers etc, and would not interpret them as such in the report. but just using the percent of variance for say observer which may be larger , is this possible?