Offset or not: guidance on accounting for sampling effort when modelling abundance data

Ecological data are often dependent on sampling effort, such as counts measured per unit area. Such data come from monitoring programs across ecology and fisheries (e.g. point counts, camera and trap surveys, fishery sampling), and underpin abundance indices, species distribution models, and often management decisions. A common tool to account for effort differences is the 'offset term' in generalized linear models, enforcing fixed proportionality between effort and the response. However, detailed guidance is limited on applying offsets and transformations, or when an effort covariate is preferable. This article reviews approaches for modelling sampling effort in regression analyses, provides best-practice recommendations, and uses simulation to examine performance of alternative parameterisations across scenarios. Offsets of log-transformed effort are preferable when proportionality is strongly supported, guaranteeing proportionality and avoiding misspecification bias. When deviation from proportionality is possible (which may be common), effort should be included as a log-transformed covariate, ideally a constrained smoother, to allow estimation of non-linear or saturated relationships. The choice between offset and covariate also depends on modelling aims: offsets support standardisation, whereas covariates enable inference on effort effects. In delta or hurdle models, effort covariates are useful because offsets have different interpretations across model components. Additional considerations include multiple effort variables, collinearity, endogeneity (when effort responds to abundance), and species-specific responses in multi-species models. Rather than modelling effort routinely, researchers should explore the effort-response relationship (proportional, otherwise linear on the link or original scales, or non-linear), drawing on prior knowledge, practical experience, data exploration, and statistical tests.