We measured the number of epileptic crisis every 2 weeks for 42 days in patients using a placebo-controlled design. At different times of the experiment, some patients leave the clinical research protocol.
At the time of statistical analysis we did not find normality in neither Active group measurement, nor Sham ones, but we found homocedasticity.
We used Skillings-Mack test for intra-group analysis in Active group... We need a method to compare between groups.
Is there any non-parametric test like two-way repeated measures ANOVA for this case?
If your outcome is a count (e.g., episodes), you might explore the use of GEE or multilevel models. These are flexible models that allow you to select a distributional family and a link function. For example, for right-skewed data, you don't need to log-transform data. You can use Poisson or negative binomial with a log link or identity link to compare the groups and adjust for confounders. For the missing data, the GEE and multilevel models use a maximum likelihood estimation, so patients with missing data still contribute information. (You have to assume that the data are "missing at random -- MAR, and only depend on the measured covariates. You have to confirm that patients are not leaving the study due to factors related to the treatment, e.g., tolerance)
-Paul
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA