I'm trying to do a before and after analysis. The dataset consists of bat passes in 2011 before a wind farm was built and again in 2012 after the wind farm was built. The data is paired and consists of 192 observations. The dataset contains a lot of zeros and is heavily right-skewed and remains right-skewed no what type of transformation I use. Is a paired sample sign test or the Wilcoxon signed rank test appropriate in this situation or is there an alternative method?
1
-
$\begingroup$ Welcome to this site! Could you share a picture of the distribution of your outcome and indicate what kind of transformations you tried? Are there other predictors of interest, or are you just interested in the pre-post difference? $\endgroup$– chlCommented Nov 2, 2020 at 18:24
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
The Wilcoxon test doesn't make any assumptions and the distribution of your data, so you're right, it would work fine here, although it might not be the most powerful test.
In general, count data where lots of the counts are $0$ can be modelled using a zero-inflated Poisson regression model, where you separately model a) the probability of having a zero, and b) the number of cases (bats) when there are more than zero. Since you have a repeated-measures design, you'll need a mixed-effects model. You can find details on zero-inflated Poisson mixed models here.
