1
$\begingroup$

I'm new here and have some doubts. I'm working with a large data set of water quality and want to test for differences between the hydrological stations and between stretches of the river. There is a large internal variability into class factors. For this, I intend to do the following analyses: Exploratory PCA: to select variables that explain the variability in dataset. DA: to sort and visualize the a priori groups. Manova (transformed data) or ANOSIM: to detect statistical significance between a priori groups. In this analysis lies my main doubt, because I frequently use ANOSIM or PERMANOVA for species abundance data. Could I use these analyses for abiotic data? Another doubt: When I use PERMANOVA or ANOSIM, I do SIMPER analysis with a complement, to track what variables (species) that caused the difference detected. Could I use this analysis (SIMPER) too? Or should I test each variable separately by univariate analyses as Kruskal-Wallis or ANOVA?

$\endgroup$
2
  • $\begingroup$ What's "AD" stand for? $\endgroup$
    – Glen_b
    Dec 14, 2013 at 23:37
  • $\begingroup$ Sorry, DA = Discriminant analysis... $\endgroup$
    – Jorge
    Dec 15, 2013 at 13:27

1 Answer 1

3
$\begingroup$

As the data in question sounds like variables such as Temp, Conductivity, pH, etc., I would find it reasonable to normalize the variables and to place all variables on even scales first (mean$=0, SD=1)$. From there, I would use Euclidean distance as your resemblance matrix and move on with testing (PCA forces this anyway).

With Euclidean distance you are able to use PCA (or NMDS) to visualize things to see if anything of interest arises. Then use PERMANOVA or ANOSIM to test for differences (PERMANOVA is much more robust to correlations and heterogeneous variances; Anderson & Walsh, 2013). SIMPER would still be able to identify those variables that account for a large portion of the differences between groups of interest. It won't really tell you if specific variables are significantly different between locations, but it can give you a good idea of what drives sample placement in NMDS or PCA space.

$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.