search

mikprov / Indonesia_fish_WWF

Collaboration from SAMSI with Andrew Tredennick.
0 stars 0 forks source link

Variogram results for initial model #3

Open atredennick opened 9 years ago

atredennick commented 9 years ago

I fixed the call to the coordinate() function (it requires a dataframe), so now it works. See my commit on the master branch of preliminary_fitting.R for the result. When I run the variogram model on the residuals, we get a result like this:

rplot

By no means is this a "final" result or anything we should base confidence in, since we still need to fit the full model to test for spatial effects, but, this figure suggests to me that there is no spatial dependence, so we can ignore space. Philipp, is that the correct interpretation here? If we got something that looked like this after fitting the full model and looking at residuals, could we conclude that there is no "important" spatial autocorrelation to explain?

pboesu commented 9 years ago

The variogram looks alright (i.e. it does not suggest strong spatial autocorrelation) on the scale that it is plotted, and yes, for model checking a plot like this is a good first step. The variogram, however, is a model itself, so you have to make sure that the underlying assumptions are not violated. In particular it might be worth to have a look at a histogram of inter-site distances, to be sure that the distance bins in the variogram are representative of the data, and you should also check whether it is safe to assume isotropy or not.

Sorry for the brevity, and for not coding what I outline above. I have code to that effect somewhere and I'll have a proper look at it when I'm not between flights.