Bernard Phiri writes:

I am relatively new to glm models, anyhow, I am currently using your book “Data analysis using regression and multilevel/hierarchical models” (pages 109-115). I am using a Poisson GLM model to analyse an aerial census dataset of wild herbivores on a ranch in Kenya. In my analysis I have the following variables:

1. Outcome variable: count of wild herbivores sighted at a given location

2. Explanatory variable1: vegetation type i.e. type of vegetation of the grid in which animals were sighted (the ranch is divided into 1x1km grids)

3. Explanatory variable2: animal species e.g. eland, elephant, zebra etc

4. Exposure: proximity to water i.e. distance (km) to the nearest water point

My questions are as follows:

1. Am I correct to include proximity to water point as an offset? I notice that in the example in your book the offset is a count, does this matter?

2. By including proximity to water in the model as an exposure am I correct to interpret this as “the chances of sighting an animal is the same for every kilometre away from the water point”?

My reply:

I would use proximity to water as a predictor, not an offset, just as, in the wells example in chapter 5 of our book, we use proximity to the nearest well as a predictor.