Random year intercepts in mixed models help to assess uncertainties in insect population trends

Research output: Journal contributionsComments / Debate / ReportsResearch


An increasing number of studies is investigating insect population trends based on time series data. However, the available data is often subject to temporal pseudoreplication. Inter‐annual variability of environmental conditions and strong fluctuations in insect abundances can impede reliable trend estimation. Temporal random effect structures in regression models have been proposed as solution for this issue, but remain controversial. We investigated trends in ground beetle abundance across 24 years using generalised linear mixed models. We fitted four models: A base model, a model featuring a random year intercept, a model featuring basic weather parameters, and a model featuring both random year intercept and weather parameters. We then performed a simple sensitivity analysis to assess the robustness of the four models with respect to influential years, also testing for possible spurious baseline and snapshot effects. The model structure had a significant impact on the overall magnitude of the estimated trends. However, we found almost no difference among the models in how the removal of single years (sensitivity analysis) relatively affected trend coefficients. The two models with a random year intercept yielded significantly larger confidence intervals and their p‐values were more sensitive during sensitivity analysis. Significant differences of the model with random year intercept and weather parameters to all other models suggest that the random year effects and specific weather effects are rather additive than interchangeable. We conclude that random year intercepts help to produce more reliable and cautious uncertainty measures for insect population trends. Moreover, they might help to identify influential years in sensitivity analyses more easily. We recommend random year intercepts in addition to any variables representing temporally variable environmental conditions, such as weather variables. We tested how random year intercepts in generalised linear mixed models (GLMMs) affect the sensitivity of insect abundance trend estimates towards single years. Random year intercepts had significant effects on the overall magnitude of the estimated trends, trend uncertainties and p‐values, but almost no effect on trend sensitivity towards single years. Random year intercepts help to account for year effects and temporal pseudoreplication in insect time series. They allow estimating more conservative trend uncertainties; i.e. larger confidence intervals and p‐values.
Original languageEnglish
JournalInsect Conservation and Diversity
Issue number4
Pages (from-to)531-537
Number of pages7
Publication statusPublished - 01.07.2023

Bibliographical note

Funding Information:
We would like to acknowledge that the data were collected in the course of teaching activities at Eberswalde University for Sustainable Development. We therefore thank more than 1000 students who helped collecting the data between 1999 and 2022. We would also like to thank Gergana Daskalova for the constructive exchange about random year effects and Tobias Kirchenbaur and Jakob Visse for their comments on various statistical issues. Xavier Pochon of Cawthron Institute kindly provided office space. This research was funded through the Biosphere Reserves Institute and the Innovation and Career Center “ProBio‐LaB” by the Ministry of Science, Research and Culture of the federal state of Brandenburg (MWFK). Open Access funding enabled and organized by Projekt DEAL.

Funding Information:
Fabio Weiss was funded through the Biosphere Reserves Institute and the Innovation and Career Center “ProBio‐LaB” by the Ministry of Science, Research and Culture of the federal state of Brandenburg (MWFK).

Publisher Copyright:
© 2023 The Authors. Insect Conservation and Diversity published by John Wiley & Sons Ltd on behalf of Royal Entomological Society.

    Research areas

  • Biology - arthropod decline, false baseline effect, insect decline, mixed models, robustness, sensibility analysis, snapshot, temporal pseudoreplication, time series, uncertainty, year effects