1
$\begingroup$
  1. 8 plots (quadrants 3 x 3 meters), 4 invaded with an alien plant, 4 without the alien plant.

  2. I have measured the soil cover (as %) of the species found in these 8 plots.

  3. If I make a PCA with log (n+1) transformed values, the resulting PCA has low eigenvalues of the first two PC (corresponding to 34% and 28% of variance, respectively). Also loadings are mostly low (mostly < 0.30).

  4. There are also many variables corresponding to species that are poorly represented in the plots (they are present in one or few plots with very low soil cover). The then scatter plot results a mess with many very short vectors.

  5. I have read that this state of things could be due to multicollinearity and that it could be necessarity to exclude some of the variables.

  6. Could it be that invaded and non-invaded plots are similar and then the overall variance is low?

So, how to deal with such data?

NB: I am not using R

$\endgroup$
7
  • 1
    $\begingroup$ Its OK not to use R, but what DO you use? $\endgroup$ Aug 15, 2023 at 13:44
  • 1
    $\begingroup$ Why are you doing a PCA at all? $\endgroup$
    – jbowman
    Aug 15, 2023 at 13:45
  • 1
    $\begingroup$ What is the ecological question that you're trying to answer? For a similar example from an EDA perspective, see Jegede et at 2023. $\endgroup$
    – Galen
    Aug 15, 2023 at 13:56
  • $\begingroup$ 1) I am using PAST and PSPP $\endgroup$
    – dandelion
    Aug 15, 2023 at 14:08
  • 1
    $\begingroup$ @dandelion Note that you can put an @ symbol before someone's username when responding to them. It will notify them. Otherwise they may not remember to look back here. $\endgroup$
    – Galen
    Aug 15, 2023 at 14:36

0

Your Answer

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

Browse other questions tagged or ask your own question.