Plot how continuous covariates influence psi

[tds]

I know it is possible to plot how each continuous covariate influences psi for a single model in MARK. In the literature, people regularly show these plots for the top model.

(1) However, my top model has a relatively low weight, and multiple models have a delta AIC <2. I therefore would like to plot how each continuous covariate influences psi taking all the models into account (perhaps this would be called a model averaged prediction?). Is this possible in MARK? If it is, could someone point me to where the mechanics are explained.

(2) If I did just plot how each continuous covariate influences psi for a single model in MARK, I would want to recreate the graph in R to make it a bit better looking for publication. Any advice on how I could replicate this plot simply?

Perhaps this information is out there, and I've just been looking in the wrong places...

Much thanks.

[cooch]

I know it is possible to plot how each continuous covariate influences psi for a single model in MARK. In the literature, people regularly show these plots for the top model.

(1) However, my top model has a relatively low weight, and multiple models have a delta AIC <2. I therefore would like to plot how each continuous covariate influences psi taking all the models into account (perhaps this would be called a model averaged prediction?). Is this possible in MARK? If it is, could someone point me to where the mechanics are explained.

(2) If I did just plot how each continuous covariate influences psi for a single model in MARK, I would want to recreate the graph in R to make it a bit better looking for publication. Any advice on how I could replicate this plot simply?

Perhaps this information is out there, and I've just been looking in the wrong places...

Much thanks.

One of your fellow Cornell grad students (same department, no less) just asked the exact same question. The basic elements are explained in Chapter 11, section on model averaging individual covariates (which, in effect, is what you have -- site specific covariates are individual covariates).

Basically, you do model average. It is easier in RMark than 'classic' MARK, but is doable. Basically you

1\ imagine your covariate covers the range from 10 to 60.

2\ for a discrete number of values of the parameter over that range (say, at 10, 20, 30...,60), calculate the parameter estimate for that value of the covariate. For models in the candidate model set that don't have the covariate, that is equivalent to all individuals having the same value of the covariate (the mean).

3\ model average the parameter estimates for each discrete level of the covariate. Then, calculte the SE (and the CI) using the standard approaches, as described in Chapter 4.

4\ your 'plot', then, would have 6-7 points for the model averaged estimates, and upper lower CI's. Then, simply 'connect the dots'.

There is your plot.

The natural interest would be in model averaging the beta estimates. But, this is problematic at a lot of levels. Worth having a read thorough the following discussion: viewtopic.php?f=1&t=996&p=2620&hilit=model+average+betas#p2620

[cooch]

2\ for a discrete number of values of the parameter over that range (say, at 10, 20, 30...,60), calculate the parameter estimate for that value of the covariate. For models in the candidate model set that don't have the covariate, that is equivalent to all individuals having the same value of the covariate (the mean).

Forgot to add (although it is discussed in Chapter 11), you can have 'classic' MARK generate the estimates of a parameter for a given covariate using the 'user-specified' option that is available when you run an individual covariates model.