Preferential sampling can bias estimates of population growth rate using N-mixture models: guidance for monitoring programs

Context Species monitoring programs are essential for assessing population status and informing management decisions. N-mixture models have become a widely used tool for estimating population growth rates (λ) from passive monitoring data, such as counts from camera traps or sonar surveys, because they account for imperfect detection and are applicable to common data types in fisheries and wildlife studies. Aims Our objective was to evaluate how site selection strategies and the proportion of sites sampled influence the accuracy and precision of λ estimates in single-season and dynamic N-mixture models. We were particularly interested in whether preferential sampling – the common practice of focusing monitoring efforts on sites known or presumed to have higher densities – could bias estimates of λ when using these modeling approaches. Methods We simulated count data from populations with spatially varying growth rates that led to increasing, stable, or decreasing overall growth, under scenarios of high or low detection probability. These data were analyzed using single-season and dynamic N-mixture models, with different sampling designs, including random and preferential site selection, and varying the proportion of sites sampled. Key results Precision of λ estimates was similar between single-season and dynamic models, though dynamic models offered potential gains by borrowing information across years. However, preferential site selection consistently introduced bias into λ estimates. We also found more precise values in declining populations, likely due to reduced dispersion from the mean as population sizes contract, compared to the greater variability observed in expanding populations. Conclusions Our findings demonstrate that preferential sampling can bias population growth estimates from N-mixture models, even when using sophisticated model structures. Random site selection across all potential monitoring locations consistently provided less biased estimates of λ, reinforcing the importance of sampling design. Implications Recognizing and accounting for potential sampling biases is critical when applying N-mixture models to assess species trends. Our results offer practical guidance for designing or evaluating monitoring programs, emphasizing the need for randomized site selection to ensure accurate and unbiased population growth estimates.