Document: JOH-3-20-5

Quantifying bias in estimates from wildlife surveys.

SAUER, J.R.* and W.A.LINK

USGS Patuxent Wildlife Research Center, Laurel, MD 20708 USA ¹

Abstract:
Most wildlife surveys are based on counts, rather than censuses, of animals. These counts represent an unknown proportion of animals present at sample sites. Counts complicate analysis by introducing irrelevant variation; this variation often introduces patterns in counts that do not reflect patterns in the underlying populations. Although the potential for bias in estimation from counts is well known, controversy exists regarding appropriate approaches for mitigating the effects of the bias in estimation. Some investigators ignore the potential for bias, while others require estimation of the proportion of animals missed during counting. Occasionally, bias in estimation can be mitigated in analyses by using covariates that control for the effects of changes in the proportion of animals counted. We discuss several classes of covariates that can be used to accommodate differences in detection of animals in wildlife surveys, and describe generalized linear models that can be used for estimation of population change from count data. Example applications of these models include estimation of population change and spatial patterns of relative abundance from the North American Breeding Bird Survey when quality of observers varies over space and time, and estimation of population change in Christmas Bird Counts when counting effort varies. Unfortunately, many factors that influence proportion of animals counted cannot be accommodated through use of covariates; hence, care must be used in design of surveys to minimize the limitations of count data.

Keywords: bias, count surveys, estimation, modeling

This abstract is being presented at: 10:45 AM in session:
Symposium # 15: Measurement Error in Ecological Data.