Overcoming Pitfalls when Estimating Variation in Parental Care: A Simulation Study

Abstract 1. Background In many animal species, parents allocate resources to enhance offspring survival and thereby reproductive success (‘parental investment’). Parental care behaviors are a crucial part of parental investment, and one critical question in behavioral ecology is to which degree variation in parental care affects offspring development and survival. Crucially, parental care behaviors vary with offspring age, which creates non-trivial problems for statistical analyses, specifically when sampling is incomplete or biased. 2. Methods We conducted a simulation study to illustrate the problem when analyzing average parental care behaviors while not correcting for offspring age (‘conventional approach’), and propose a modeling approach to correct for offspring age and sampling biases. We simulated a set of different scenarios, varying with regard to sample size, sampling scheme, and variation in parental behavior. To model parent behavior as a function of offspring age, we fitted a Generalized Linear Mixed Model to each simulated data set. We included offspring age as a fixed effect and extracted Best Linear Unbiased Predictors (BLUPs) for the random intercept of parent identity. These BLUPs represent the parental behavior corrected for offspring age. Finally, we assessed how our proposed modeling approach compares to the conventional approach. 3. Results The proposed modeling approach was superior to the conventional approach. More specifically, the conventional approach clearly overestimated the variation between parents, and even diagnosed appreciable variation when there was none in the simulated data. This bias was exacerbated in incomplete or unbalanced sampling schemes. 4. Conclusion. Our simulation study clearly demonstrates the necessity of correcting for offspring age in analyses of parental behavior. We strongly suggest including random slopes of offspring age within parent-offspring dyads, because doing so can reveal parent-specific trajectories of the modeled behavior against offspring age. Our approach allows for the inclusion of further categorical and continuous predictor variables that may affect parental behavior, such as parity or ecological conditions. In summary, the proposed modeling approach has several advantages over conventional methods.