PT  - JOURNAL ARTICLE
AU  - Olivier Gimenez
AU  - Christophe Barbraud
TI  - Dealing with many correlated covariates in capture-recapture models
AID  - 10.1101/097006
DP  - 2016 Jan 01
TA  - bioRxiv
PG  - 097006
4099  - http://biorxiv.org/content/early/2016/12/27/097006.short
4100  - http://biorxiv.org/content/early/2016/12/27/097006.full
AB  - Capture-recapture models for estimating demographic parameters allow covariates to be incorporated to better understand population dynamics. However, high-dimensionality and multicollinearity can hamper estimation and inference. We propose a modeling framework to account for these two issues. Principal component analysis is used to reduce the number of predictors into uncorrelated synthetic new variables. Principal components are selected by sequentially assessing their statistical significance. We provide an example on seabird survival to illustrate our approach. Our method requires standard statistical tools, which permits an efficient and easy implementation using standard software.HighlightsHigh-dimensionality and multicollinearity hamper model inference capture-recaptureThese issues are addressed with principal component capture-recapture (P2CR) modelsWe provide an example on seabird survival to illustrate the P2CR methodP2CR requires standard statistical tools and is implemented with standard software