[R-sig-dyn-mod] Venezuelan census models
Thomas Petzoldt
Thomas.Petzoldt at TU-Dresden.de
Sat Jan 5 13:26:29 CET 2013
Hello,
just a few thoughts:
1) Consideration of growth directly instead of (or in addition to)
population abundance seems to be a good idea. Thanks for this idea.
2) I agree with Hans, that the purpose of the model counts, i.e.
description of data, statistical testing, mechanistic systems
understanding or prediction.
3) For a statistical model comparison, the sum of squares alone is not
enough and model complexity (number of parameters) needs to be taken
into account.
Re-fitting the two-compartment models with 3 resp. 5 parameters with
nlsLM from package minpack.lm, we get:
> AIC(res3, res5)
df AIC
res3 4 385.5084
res5 6 384.0203
We see that res5, the model with 5 parameters performs (indeed and only)
a little bit better.
However, validity of all the autonomous models applied so far is rather
limited and estimated saturation levels remain pure speculation.
Therefore, non-autonomous models are required that consider external
(i.e. "environmental") factors.
Thomas P.
PS: Code can be provided in case of interest.
More information about the R-sig-dynamic-models
mailing list