Journal Article

Avoiding misleading estimates of among‐individual variance caused by non‐random sampling of individuals in a changeable environment

Araya‐Ajoy, Yimen G., Réale, Denis, Pick, Joel L., Allegue, Hassen, Westneat, David F., Dochtermann, Ned A., Dingemanse, Niels J., Schielzeth, Holger, Nakagawa, Shinichi

2026

Overview

Animal ecologists frequently quantify variance in hierarchically structured traits in wild populations. Importantly, phenotypic plasticity within the period of measurement can modify the trait of interest in response to various unmeasured, temporally or spatially changeable, environmental conditions. Non‐random sampling among units of the random effect (e.g. individuals) regarding the environment at issue may lead to estimates of the variance among (σ̂I2) or within (σ̂W2) such units that conflate several types of processes. This mixing of underlying biology can affect interpretations of the random effect variance. Here, we explore the conditions leading to this situation and assess potential solutions when relevant information is missing. We simulated a trait's phenotypic values that depended on the environmental variable, and individuals that differed in their deviation to the mean population phenotype (random intercepts). We also simulated different types of variation in an environmental variable that was either shared or specific to each individual. We then varied the repeatability in the timing of sampling (RIS2) and analysed simulated datasets using linear mixed‐effect models with different fixed‐ and random‐effect structures. In the presence of unmeasured environmental factors, the estimated among‐individual variance (σ̂I2) contained a larger signature of the current environment as the strength of the temporal autocorrelation and the repeatability in the timing of sampling (RIS2) increased. For low to moderate values of RIS2 (e.g. <60% of the total variance in our simulations) the risk of pre‐study and within‐study effects conflating estimates of variance components was low and could easily be corrected with a model including period or individual‐period combination as random effects. Higher RIS2 led to an increase in conflating effects that were difficult to correct. Our study shows the importance of limiting the variance among individuals in the timing structure of sampling (RIS2). We recommend researchers estimate RIS2 and report it in papers. Finally, RIS2 can be limited by sampling all individuals in the same period, or sensitivity analyses could be conducted by removing extreme sampling dates at the analysis stage to reduce RIS2.

• Publisher: John Wiley & Sons, Inc,Wiley
• Subject: biased sampling / conflation of effects / Environmental conditions / Environmental factors / Estimates / heritability / linear mixed‐effect models / Phenotypes / Phenotypic plasticity / Random sampling / repeatability / Reproducibility / Sensitivity analysis / Statistical sampling / temporal autocorrelation / temporal trends / unmeasured environmental effects / Variance