2

I have matrices of genetic distances for x number of individuals within a population and their corresponding point coordinates -one genetic distance matrix per point coordinates.

I was imagining using the Mantel test, i.e. evaluate the significance of correlations between genetic and geographic distances. However, what I did not think of was that these two types of matrices will have different dimensions (e.g. a genetic matrix of 15 ind = 15x15, sampled at three locations = 3x3), hence the Mantel test will not work. I am now looking into Moran's I, testing for spatial autocorrelation, but it will not work with my data as it currently is. I would need one measure of variation per point coordinates. Would it be legit (never seen anyone do this) to derive a measure of variation (e.g. var, sd, cv) from a single distance matrix? Since the matrix is normalized, ranging from 0-1, maybe var or sd would work?

I am kinda new to analyzing spatial data, so I would appreciate any pointers or alternative tests I might be able to do.

Thanks,

jO.
  • 143
  • 8
  • I am sure that you find useful information in this question (it also contain links to useful articles) http://stats.stackexchange.com/questions/72685/why-is-mantels-test-preferred-over-morans-i – Ladislav Naďo Jan 31 '14 at 11:36
  • @LadislavNado Cheers for commenting. Yeah, I have read through this already. It's a good one. – jO. Jan 31 '14 at 11:57
  • 1
    Hi. I am sorry but with just 3 locations you can forget about spatial statistics - you won't be able to calculate Moran's I or Mantel properly (believe me such approach will be hard to defend in peer-review), and i think you should do some simple "ANOVA-like" analysis instead (try to explain variability within and between groups). – Ladislav Naďo Jan 31 '14 at 14:43
  • @LadislavNado Ah, yeah you're probably right. Cheers – jO. Jan 31 '14 at 15:03
  • 1
    ...Im thinking of **adonis** {vegan} analysis which is a non-parametric version of **MANOVA** – jO. Jan 31 '14 at 15:30
  • I will look on it {adonis} - (may be it will solve not only your problem be my problem also). Thanks you for it! I contact you later. – Ladislav Naďo Jan 31 '14 at 16:20
  • 1
    Now, after reading and trying to understand various tests, I think **Betadisper** {vegan} would work for me. It tests if groups differ in their variance compared to **adonis** which tests the difference in centroid position. It is basically **Levene's test for homogeneity of variances/homoscedasticity** which assesses the equality of variances for a variable *(genetic distance)* calculated for two or more groups *(locations)*. – jO. Jan 31 '14 at 17:31

0 Answers0