I'm working on a study where I am testing whether individual species that I observed or didn't observe (presence/absence of observing species A on each survey) during drought and non-drought (presence/absence of drought). I have about 150 surveys of presence/absence for species each during drought and non-drought. I'd like to statistically test whether individual species are more likely, less likely, or equally likely to be observed during drought or non-drought.

For a couple species, I did a contingency table/chi-squared analysis. Is this appropriate, or is there a better-suited method I could use? The convenient part about contingency tables is that I was easily able to implement my data as presence/absence columns (counts of surveys in which species A was observed versus surveys when it wasn't observed) and presence/absence rows (Drought Yes/No).

Thanks for your insight.

  • 1
    $\begingroup$ The chi-squared analysis could work, but be aware that if you conduct multiple tests then you may be more likely to get at least one false positive. $\endgroup$ – Henry Feb 8 at 23:01
  • $\begingroup$ Thanks, I'm looking into that $\endgroup$ – Brian Feb 9 at 21:46

Are you trying to determine whether the proportion of sightings differs between species (e.g. species b decreased during drought more than species a) or within species (species a sightings decreased during drought) or both? How many species are you testing for?

| cite | improve this answer | |
  • $\begingroup$ I am trying to look at both. I'll probably have 35-40 species in total once all of the data is prepped. My initial ideas are as follows: 1) do contingency tables for each species, flag those in which sightings differ for drought and non-drought. 2) Possibly calculate the Jaccard Index to find out the strength of the relative associations 3) Plot the Index scores for each species to show which have the greatest response to drought. Please share any flaws in my thinking. All suggestions are welcome. Thank you! $\endgroup$ – Brian Feb 9 at 21:43
  • $\begingroup$ You could also look into a log-linear model. You're dependant variable would be the counts of drought/observed/species combos (the values from your contingency tables). The independent variables would be drought, observed, and species. You could include interacting as appropriate (based on your comment "drought x species" and "observed x species" would be a good start). The typical error distribution for this model is the Poisson, but you could use the Quasipoisson or Negative Binomial to add a dispersion parameter. $\endgroup$ – CFD Feb 10 at 2:56
  • $\begingroup$ Thank you! I will definitely consider that going forward. $\endgroup$ – Brian Feb 15 at 19:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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