|
Résumé :
|
Spatial sampling error remains a critical yet underappreciated challenge in abundance estimation, particularly when sampling processes are not explicitly accounted for. A persistent belief holds that increasing sample size compensates for sampling bias, while overlooking how spatial sampling processes — such as preferential or convenience sampling — shape these errors. Relying on the decomposition of sampling error into components driven by sample selectivity, incompleteness, and statistical population heterogeneity, we derive analytical results linking characteristics of the realized sample to properties of the underlying sampling process. Using simulated examples, we illustrate how five sampling processes — four archetypal (equal-propensity, preferential, spatial-spread, and road-based sampling) and one composite opportunistic process combining several of these mechanisms — influence the reliability of abundance estimates. Our results show how increasing sample size alone — a common reflex in monitoring — improves precision but fails to correct for bias caused by selective sampling. This is particularly relevant for citizen science, where massive datasets may yield precise yet systematically misleading inferences if spatial sampling processes are ignored. We advocate for a shift from a reliance on sample size toward the explicit characterization of sampling processes and the prioritization of probability-based sampling designs where feasible. By treating sampling error as a process-driven phenomenon, this work provides a contribution toward more rigorous and transparent ecological inference. The resulting framework is presented from a methodological perspective, specifically targeting quantitative ecologists, biometricians, and wildlife statisticians involved in designing studies, implementing monitoring programs, or analyzing citizen science data.
|