Page 1 of 1

Correlations with time-varying covariates

PostPosted: Wed May 17, 2017 10:02 pm
by jCeradini
Analyzers,

Goal: Assess correlations (pairwise, such as Pearson, or multicollinearity, such as VIF) between predictors before modeling in order to identify "highly" correlated predictors that shouldn't be in the same model, like is commonly done in regression analysis. But, the twist (at least in my mind, currently) is that I'm wondering what the appropriate approach is for assessing correlations when one or more predictors are time-varying, such as a survey specific covariate?

Example:
Single season occupancy model with 3 occasions and 3 detection predictors. 2 predictors are survey specific, lets say, temperature and jdate, and one is site level, cover.

If the data are in "vertical" format, do I ignore the time dimension and look at the correlation between temp, jdate and cover, ignoring survey. Or, does the correlation need to be grouped-by survey?

Some made up data to help show the format (sorry for the dots, wasn't sure how else to maintain the spacing):
Site Survey Temp Jdate Cover
1.......1......10....100....30
1.......2......11....105....30
1.......3......20....120....30
2....

Thanks for any guidance!
Joe