Ch. 6.5.1-Problems with subset of DM models

[lyanneb]

Sorry, but I am learning MARK here so bare with me please.

I am working on Ch. 6 and replicated everything until I reached page 6-33. I can not replicate the AICc and Deviance values for all the models after the "variable definitions" stage. The order is slightly askew: phi(sex+flood)p(sex*time) and phi(.)p(sex*time) are 2nd and 3rd on page 3-36 respectively, but in my analyzes they are 3rd and 2nd respectively.

The AICc and deviance values are also lower in my analyzes. For instance, phi(sex+flood)p(sex*time) and phi(.)p(sex*time) AICc values are 585.0084 and 584.5618, respectively. I increased the number of parameters to match the previously created models as described on page 6-24. I did not check off any of the options on the bottom "Variable Definitions" window. Could that be why?

I can't figure it out, any suggestions?
Thanks for the help and patience everyone [/img]

[cooch]

I am working on Ch. 6 and replicated everything until I reached page 6-33.

Different startng point in most recent version of the chapter.

I can not replicate the AICc and Deviance values for all the models after the "variable definitions" stage.

In general, just focus on the deviances. If the deviances between two different approaches to the same model are the same, then the models are in fact equivalent.

The order is slightly askew: phi(sex+flood)p(sex*time) and phi(.)p(sex*time) are 2nd and 3rd on page 3-36 respectively, but in my analyzes they are 3rd and 2nd respectively.

No idea - I just re-ran it, and the model order is exactly as described (shown) in Chapter 6. From best -> worst (top to bottom in the browser), {flood}, {sex+flood}, {.}, {sex.flood}, and {sex}.

The AICc and deviance values are also lower in my analyzes.

Which indicates something is different. Confirm you're using the right data set, confirm that c-hat is 1.0, and confirm your general starting model (for this example) is phi(sex.flood)p(sex.time). If you confirm all three, and you follow the example exactly as indicated in Chapter 6, you should get exactly the same model rankings and deviances.

Worth grabbing a copy of the most recent version of the chapter.

http://www.phidot.org/software/mark/doc ... /chap6.pdf

[lyanneb]

Thanks for the help,

I was working with the data set, but I was working continuously though the instructions. I found the only way to replicate the results was to delete all the previously 'hand built' models and default to the (sex*flood)(sex*time) with 15 parameters. The hand built models were increased manually by one parameter through "Adjustments" e.g. (sex*flood)(sex*time) with 16 parameters.

I re-read the newer chapter 6 version and there was no mention that I could find to revert to this model (15 parameters) in order to build the subset models through the "Run\Subset of DM Models" routine.

Thanks again for your help. Slowly but surely I'll learn this

[cooch]

I was working with the data set, but I was working continuously though the instructions. I found the only way to replicate the results was to delete all the previously 'hand built' models and default to the (sex*flood)(sex*time) with 15 parameters.

The hand built models were increased manually by one parameter through "Adjustments" e.g. (sex*flood)(sex*time) with 16 parameters.

I'm not clear at all here what you're trying to do, or what you're talking about. I suspect its confusion of what model to select as a starting point. See below.

I re-read the newer chapter 6 version and there was no mention that I could find to revert to this model (15 parameters) in order to build the subset models through the "Run\Subset of DM Models" routine.

p. 34 - You simply "highlight (retrieve) the model with the design matrix that contains all the variables that you want to use in all possible subset model combinations - in this case, model phi(s.f)p(s.f)".

Thats it, basically. You'll also note that its the most general model shown in the results browser image shown on p. 37 (which makes it implicit - again - that you're building these models starting from phi(s.f)p(s.f) in this example.)