It is common in animal behavior and cognition research to evaluate consistent individual differences (repeatability) in performance measured as the time to accomplish some event (latency to do a behavior or the number of trials to meet a pre-set criterion). Some individuals may never do the target behavior or meet the criterion and failing to account for the right-censored nature of these data is potentially problematic. Survival analyses explicitly account for censored data but there is currently no method for using this technique to quantify repeatability. Through a collaboration funded by the SQuID fellowship, we address this gap in our ability to statistically quantify individual differences in animal behavior.
In our manuscript: "Repeatability and intra-class correlations from time-to-event data: towards a standardized approach" we describe this issue in more detail and present a solution.
We compiled supplementary materials that consist of R code guiding the user through several worked examples from openly accessible data sets. In addition, we present results from a simulation study demonstrating the robustness of our proposed solution in relation to similar methods and varying data inputs. We provide all files from the simulated datasets, which make up of a large R object with a list of 12,000 datasets, at OSF.
We wrote a function in R that estimates the repeatability, as well as a confidence interval and p-value, from time-to-event data using survival analysis and the residual variance estimator we present in the manuscript.