Individual Covariate: Capture Frequency

[jctroxler]

Background: Data is from non-invasive hair sampling. I'm using a robust design with Huggins full heterogeneity estimator. I have 3 primary occasions (years) each containing 8 secondary occasions (weeks).
I plan to create 3 individual covariates, corresponding to the total number of times an individual was captured in each primary occasion. The idea is to use information that is wasted when, for example, 6 captures within a secondary occasion are condensed into a 1 in the capture history.

Question 1: Can I apply these covariates to estimate survival, gamma',"", p, and c?

Question 2: Is a covariate valid for estimating parameters from a previous occasion or only a subsequent occasion?
Example: Can I use both the 2010 AND 2012 covariates to estimate p for 2011?

Thank you

[abreton]

Question 1: Can I apply these covariates to estimate survival, gamma',"", p, and c?

Assuming you can provide a defensible hypothesis to associate 'total number of times an individual was caught' on primary occasions to the parameters that you listed then ... MARK provides all the necessary tools for applying these covariates to "survival, gamma ... p, and c." Note that a bird must survive to have a 'total number ...' on a primary occasion. If you integrate the covariate into the survival component of the model you'll likely get a strong 'effect' but the interpretation would be bogus? Regarding p and c, I recently did something similar ... used the individual covariate (not time-varying) maximum number of detections from all occasions (primary in your case) to predict detection probability. It worked well in my case, because (1) birds tend to use the same burrows/bask in the sunshine nearby; and (2) biologists tend to read bands from the same hides year after year. The result is max frequency predicts detectability in all years ... if you follow me. This covariate absorbed A LOT of heterogeneity in individual detection probabilities.

Question 2: Is a covariate valid for estimating parameters from a previous occasion or only a subsequent occasion?
Example: Can I use both the 2010 AND 2012 covariates to estimate p for 2011?

As above, it's valid if you can provide a legitimate/defensible hypothesis. A nice example where it would be valid to use data from a previous occasion to predict subsequent (opposite of your inquiry) is trap dependence ... a bird is trapped on occasion i-1 has a higher (trap happy) or lower (trap shy) probability of detection on occasion i. If there is information in a datum from the present that can predict the past then that covariate would be valid for predicting the past.

As always, I hope I didn't misinterpret ... or mislead!

andre

[ehileman]

Regarding p and c, I recently did something similar ... used the individual covariate (not time-varying) maximum number of detections from all occasions (primary in your case) to predict detection probability. It worked well in my case, because (1) birds tend to use the same burrows/bask in the sunshine nearby; and (2) biologists tend to read bands from the same hides year after year. The result is max frequency predicts detectability in all years ... if you follow me. This covariate absorbed A LOT of heterogeneity in individual detection probabilities.

Hi Andre,

Was your analysis using the covariate capture frequency to explain heterogeneity in p and c published? If so, can you provide me the title of the paper?

Many thanks!

Eric

[abreton]

We're awaiting the proofs and expecting that it will be in the January Issue of Ibis ... though this is uncertain of course. Send me (andre breton) an email at insightdd@gmail.com and I'll forward you a copy as soon as I can.

[ehileman]

We're awaiting the proofs and expecting that it will be in the January Issue of Ibis ... though this is uncertain of course. Send me (andre breton) an email at insightdd@gmail.com and I'll forward you a copy as soon as I can.

Will do. Thanks, Andre! -Eric