I am comparing the relationship between genetic and geographic distance of individuals in a wild animal population. The hypothesis is that individuals with higher genetic relatedness establish home ranges closer to one another (shorter geographic distance between them) than to individuals that are less related (ie: geographic distance and relatedness are inversely correlated).

Because these animals have homeranges that overlap and were sampled multiple times over the course of the study, I initially calculated a center point for each individual's home range and used that as the geographic location of the individual. I then ran a Mantel test (and the modification of the Mantel test called the Tau Kr test) and compared my correlation coefficient to the results of 1000 permutations of the data. I obtained a significant result (P<0.05).

However, I wanted to see how sensitive the analysis is to where you sample the individual. Therefore, I re-ran my Mantel tests but using every possible combination of sampling locations for each individual (rather than taking the center point).

I now have a distribution of p-values and test statistics for all of these tests and in 14% of cases the tests were significant (P<0.05) for the Mantel Test and in 52% of cases using the Tau Kr Test.

Is there a way to evaluate if the initial hypothesis is true with these results? Intuitively to me, I think that if there was no relationship between geographic and genetic distance, even 14% of tests should not be significant. But I am not really sure what to compare this distribution of p-values to - what would the null model look like? It seems to be that I am detecting a real pattern even though it is highly dependent on sampling location, but I cannot figure out however how to analyze this data set now to determine whether the signal is significant.

Many thanks for your help!