0
$\begingroup$

I have count data measuring the wildlife (animals) crossing a passage. A modification was made to the connecting passage. I want to measure whether this affected the rate of wildlife crossings.

The event data are taken daily over the course of 36 months: 18 months before the modifications until 18-months after the modifications. The species of the animal is measured as well. The frequencies are non-normal. The same animal may cross more than once. Due to the distribution of these data and the possibility for dependence, I don't think the T-test is an appropriate method to compare pre- and post- wildlife crossing rates.

  1. Am I correct in using a Wilcoxon-Signed rank test procedure, matching the number of daily crossings before and after, to compare before and after?

  2. I want to perform an aggregate analysis as well as a species stratified analysis. When stratifying by species, there are many 0-crossing days, so lots of days with zeros by species which result in a large number of ties. Does a Pratt correction fix this problem?

Improve this question
$\endgroup$

2 Answers 2

Reset to default
1
$\begingroup$

As I see it the limitations of your proposed analysis are:

  1. The data are not, in fact, paired the first day post-intervention takes place a year and a half after the first day pre-intervention.
  2. You do not account for seasonality in time series data
  3. The Wilcoxon does not actually quantify the differences in rates pre/post modification
  4. The Wilcoxon does not handle small, unmeasured sources of dependence in the data (repeat crossings by same animal).
  5. The Wilcoxon, even with Pratt correction, downweights/discards zero observations which is undesirable when such observations are very informative regarding the actual rate of crossing.

All of these issues can be solved by fitting a quasipoisson GLM. I can let an ecologist correct me if I'm wrong in my thinking here. The intervention can be handled with a pre/post indicator. You can adjust for seasonality using fixed effects.

Improve this answer
$\endgroup$
0
$\begingroup$

In order to use the signed rank test, you have to be able to logically match an observation in the one group (pre-) to an observation in the other group (post-). It sounds like you don't have a way to do this, unless, for example, you could match by calendar date from one year to three years later.

Improve this answer
$\endgroup$

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.