Subject: Marine Geospatial Ecology Tools (MGET) help
Text archives
| From: | Daniel Dunn <> |
|---|---|
| To: | Rowenna Gryba <>, "" <> |
| Subject: | RE: [mget-help] artifacts when predicting GAM from rasters due to categorical variable? |
| Date: | Mon, 4 Jun 2012 19:48:31 +0000 |
| Accept-language: | en-US |
|
Hi Rowenna, I’m not the developer, just another user, but what you are encountering is to be expected based on your model. As I understand it, including region as a factor
in the model allows the model to generate different coefficients for each element in the factor- as if you had simply had binary variables (0,1) for each of the regions separately. Since you have a different coefficient for each region, the prediction for
each region will be different and there will be an obvious delineation of the regions. If you are seeing a literal line and not just the model jumping when it enters a new region it might be due to edge effects produced by the use of variables that use neighborhood
statistics (e.g. BPI does this). If your regions are analyzed separately or predicted separately, you might get an edge effect from the BPI variable.
Hope this helps, Daniel C. Dunn Marine Science & Conservation Nicholas School of the Environment Duke University Marine Lab 135 Marine Lab Rd. Beaufort, N.C. 28516 t-252.504.7605 From: Rowenna Gryba [mailto:]
Hi, I've been using MGET to predict sediment size from GAMs. One of the independent variables is 'region' which I put into the GAM as a factor. My model is : MEAN ~ s(bathy)+s(currentspeed)+s(bpi)+factor(region) The issue is that when I look at the prediction there are artifacts from the region variable i.e. there are visible lines between areas defined by the factor. Have you come across this before? If you have any suggestions I would really appreciate it. Thanks, Rowenna |