# When Do Single‐Species Occupancy Models Outperform Multispecies Models?

## TL;DR

Single-species models are better than multi-species models for studying rare species when data is limited.

## Contribution

The study identifies conditions under which single-species occupancy models outperform multi-species models in estimating species-specific effects.

## Key findings

- At low sampling intensities, multispecies models produce biased estimates of community-level treatment effects.
- Single-species models provide more accurate species-specific effect estimates when there is high variance in treatment effects.
- Single-species models are more robust to outliers for rarely observed species.

## Abstract

Occupancy models have become increasingly popular for species monitoring and assessment, in part, because detection/non‐detection data are readily obtained using a variety of methods. Multispecies occupancy models (MSOMs) can yield more accurate parameter estimates than single‐species models (SSOMs) with less data through their hierarchical structure, making MSOMs an attractive option when species are hard to detect or when data collection is constrained, leading to sparse datasets. Such constraints may arise from limited sampling resources, but also occur in rare species monitoring or where preliminary results are desired to inform adaptive management. Further, experimental habitat treatments often impose spatial constraints on sampling based on the scale of their implementation. Whether a MSOM outperforms SSOMs depends on the volume of data, characteristics of the ecological community, research goals of a study and how these factors align with modeling assumptions. We performed a simulation study of hypothetical pollinator communities under varying sampling intensities for scenarios in which experimental habitat treatments produced different community‐level effects. We fit occupancy models to simulated datasets and assessed model performance. At lower sampling intensities (< 20 spatial replicates and < 4 temporal replicates), MSOM community‐level treatment effect estimates were biased. Even at twice this sampling intensity, SSOMs yielded more accurate species‐specific effect estimates in treatment effect scenarios with high variance. In some cases, MSOMs can pull species in the tails of distributions too far toward the community mean effect, which risks incorrect conclusions concerning whether treatments help or harm individual species. When quantifying species‐specific effects is the main objective, particularly for rarely observed species, SSOMs are more robust to outliers across a range of community response scenarios. Researchers can use this information to inform study design, guide simulation studies and decide whether the higher precision of MSOMs outweighs risks of improperly estimated effects for some species.

Plains prickly pear (
Opuntia polyacantha
) flower being pollinated by multiple members of the sweat bee family (Halictidae) in eastern Montana, USA. When species‐specific effects are the main research question of interest, particularly for rare species, single species occupancy models are more robust than multispecies models to outliers across a range of community response scenarios. Photograph by Erica Gustilo, U.S. Geological Survey, 2024.