Nest survival procedure vs. Cox proportional hazards
Posted: Tue Oct 02, 2012 2:12 pmDear Colleagues:
I have been using the nest survival procedure as a known fates model for ragged telemetry data. I have an analysis that is basically an ANCOVA type design with weekly survival of radio-marked birds vs. a categorical explanatory variable and a continuous explanatory variable.
If I analyze the data in Mark with the nest survival procedure, I can get overall estimates of weekly survival from different models of the explanatory variables (cat, con, cat+con, cat*con, etc.).
I have also analyzed the data with time-to-event models in Program R and can generate Kaplan-Meier (KM) plots and hazard functions not available in Mark (survival, gss packages). The hazard functions seem particularly useful because I can model them with splines to look at seasonal patterns of mortality risk. I have used Cox Proportional Hazards (CPH) to test the effects of the explanatory variables in different combinations, and can include female id as a random factor. What is less obvious to me is how to generate plots of weekly survival vs. the two explanatory variables from the different CPH models.
Would it be acceptable to mix and match results from the two sets of analyses? My understanding from the Gentle Introduction is that the known fate and nest survival procedures are basically KM models but redefined as binomial models for MLE (16-1, 17-4). Perhaps this is the reason that one can get the plots in Mark but not with the KM or CPH models. Anybody else wrestled with this issue? If it is possible to plot weekly survival vs. explanatory factors from the results of a CPH model, I would be interested to see an example of an R script. Comments or suggestions welcome.
Thanks, Brett (bsanderc@ksu.edu)
I have been using the nest survival procedure as a known fates model for ragged telemetry data. I have an analysis that is basically an ANCOVA type design with weekly survival of radio-marked birds vs. a categorical explanatory variable and a continuous explanatory variable.
If I analyze the data in Mark with the nest survival procedure, I can get overall estimates of weekly survival from different models of the explanatory variables (cat, con, cat+con, cat*con, etc.).
I have also analyzed the data with time-to-event models in Program R and can generate Kaplan-Meier (KM) plots and hazard functions not available in Mark (survival, gss packages). The hazard functions seem particularly useful because I can model them with splines to look at seasonal patterns of mortality risk. I have used Cox Proportional Hazards (CPH) to test the effects of the explanatory variables in different combinations, and can include female id as a random factor. What is less obvious to me is how to generate plots of weekly survival vs. the two explanatory variables from the different CPH models.
Would it be acceptable to mix and match results from the two sets of analyses? My understanding from the Gentle Introduction is that the known fate and nest survival procedures are basically KM models but redefined as binomial models for MLE (16-1, 17-4). Perhaps this is the reason that one can get the plots in Mark but not with the KM or CPH models. Anybody else wrestled with this issue? If it is possible to plot weekly survival vs. explanatory factors from the results of a CPH model, I would be interested to see an example of an R script. Comments or suggestions welcome.
Thanks, Brett (bsanderc@ksu.edu)