I am looking for an RM-ANOVA alternative that fits with non-normal data that may not satisfy sphericity assumption. My dataset: Through a mark and recapture study, I trapped turtles, and collected fecal seed samples from these turtles (number of seeds present in per fecal sample varied between 0-~1500-- with most fecal samples having no seeds). About 20% of the fecal seed samples I have are from recaptured (same turtle captured more than ones on different instances) turtles-- therefore, my fecal seed samples are not necessarily independent samples. I need to figure out a statistical tests to see (1) whether no-seed fecal samples were significantly greater than seed-containing fecal samples AND (2) identify predictor variables (contagious and discrete variables-- such as turtle body wt, sex, sampling location) that influence number of seeds present in fecal samples. Had this been a case of independent observations, for my question 1, a chi-squared test of a Wilcoxon rank sum test (M-W type) would've worked easily. for question 2, a permutation ANOVA or a variations of GLMs could have been used. But, in this case, the samples are not really independent. Tests such as McNemar's are for before-after comparisons, Friedman does not work either as not all of my samples are repeated. So, what would be the best test to use?