www.phidot.org

Hi all,

I know that normally when you fit two models identical except for one variable, like for instance:
S(envcov1)
S(envcov1+envcov2)
and you find a slightly lower AIC for the second model but a very similar Deviance you can suspect that the envcov2 is a pretending variable.
In these cases it is recommended to have a look at the beta value of the suspected pretending variable and see if the CIs are significant or not (including zero or not).

OK, now I have a case that is quite different with respect to the above scenario but I am still in doubt about how to go on, should I treat it as a pretending variable issue and remove that model from the results or not?

I have two models:
S(envcov1) Deviance:16144.1 QAICc:16269.6
S(envcov1+envcov2) Deviance 16138.8 QAICc:16266.45

The difference in the number of parameters is of course just one, but the differences in both Deviances and QAICc values are not that trivial at all.
Anyway, when I see the beta value of the envcov2 I find this:

Intercept: value(-1.36) CI-(-1.72) CI+(-0.99)
envcov1: value(0.52) CI-(0.27) CI+(0.77)
envcov2: value(-0.05) CI-(-0.36) CI+(0.26)

Any opinion on this?

The first thing I would do is make sure that the simpler model {S(envcov1)} converged correctly by starting it with the initial estimates from the more complex model {S(envcov1+envcov2)}.

Gary

The first thing I would do is make sure that the simpler model {S(envcov1)} converged correctly by starting it with the initial estimates from the more complex model {S(envcov1+envcov2)}.

Gary

Thanks for answering.
Yes, I have tried it and also I have run both the models several times with different initial values to avoid problems with local minima: it seemed there are no problems with that.

Perhaps it might be useful adding some details over the analysis I am doing. The data set cover 13 sessions (wintering seasons) and I am modelling transition rate. The covariates are:
envcov1= drought (there were just two years of drought, 2 levels: 0 wet year, 1 dry year)
envcov2= population size (13 values, they are counts, the population size show a pronounced increase over time)

Both covariates have been standardized.
Any other suggestion/commentary?

Simone

Perhaps you should read this article:

Grosbois V, Gimenez O, Gaillard J-M, Pradel R, Barbraud C, Clobert J, Møller AP, Weimerskirch H (2008)
Assessing the impact of climate variation on survival in vertebrate populations, Biological Reviews, vol. 83 pp.357-399