Problem with covariate plot tool - conflicting results

[cbutton]

Hi all,

I am running a CJS analysis that includes individual covariates (size and tidal height) for two locations surveyed at the same time. When I use the covariate plot tool and copy the results into an excel file, the survival estimates are different from what is given if I use the "user-defined covariate" option while running the model. When I calculate the survival "by hand," I get the same answer as the "user-defined covariate" option.

Example:
The model with the most support in my analysis has survival as a function of the location (g) and the size of the animal. logit(Phi-g1)= B1 + B2*Size_s + B3 + B4*Size_s

B1 = -1.229
B2 = -0.502
B3 = -0.754
B4 = 0.442
average Size = 92.289 mm
SD Size = 24.823

I calculated (and MARK calculated with the "user-defined" option) that the survival rate of a 27 mm animal would be 0.1387. The result given when using the covariate plot tool is 0.1598.

The above calculations were done with chat set at 1.3268. If I reset the chat to 1, and use the covariate plot tool, the results match across all of the methods used.

How do I reconcile this? Is MARK giving me the wrong Beta estimates? Or, is the covariate plot tool unduly influenced by chat?

Any insight you have is greatly appreciated.

Thanks in advance,
Cynthia

[gwhite]

Cynthia:
I'm not sure what is going on. When I run the covariate plot and dump the estimates to Excel, I get exactly the same values for estimates, with only the SE and CI values changing, as they should. Likewise, the graphed line for the estimates is identical, but with wider CI for larger c values.

Gary

[cbutton]

Thanks for checking into it for me. The more I try to figure out what is happening with my version of MARK, the more confused I get.

I tried downloading and installing the latest version on my desktop (windows XP) to see if that would change anything... it didn't. Then, I downloaded the latest MARK onto my laptop (windows vista) and ran the analysis. On the laptop, the covariate plot tool gives the answer I calculated "by hand", including the correct standard errors and confidence intervals (good). The "user-defined covariate" method gives the same standard errors for the beta estimates in the "output" window no matter how I adjust chat (bad). If I ask MARK specifically for the beta estimates or the variance-covariance matrix, then the standard errors increase as chat is adjusted higher (good). So, I feel like I am probably getting the appropriate estimates using that method.

So, in short, I think the issue is with my installation of MARK somehow. But, I'm completely bamboozled as to what the root cause would be. And, I'm still not sure how to proceed with using the beta estimates given by MARK when there seems to be a bug acting on both of the versions (5.1) I have now.

Thanks again for replying to my post so quickly.

Cynthia