I am working with ecological count data where the raw data is divided into multiple sites among sampled years. So you have site 1,2,3...etc for year 1999, 2000, 2001, etc. Each year is within either Group 1 or 2 depending on their temperature grouping. So I am more interested in measuring differences between Group 1 and 2, which is what I am using PERMANOVA and other multivariate tests for.

Testing the sites, there is little to no difference among them, so they were averaged into just years for easier visualization-especially regarding NMDS. However, when testing PERMANOVA, if I use the the averaged data, p is not significant. When using raw data, it does show significance.

I understand that raw data is usually always better to use, but I am unsure as to why its changing so significantly with the two versions of the data. I also want to make sure NMDS and PERMANOVA match so there is not conflicting results