Proxy problems: Why a calibration is essential for interpreting quantified changes in energy expenditure from biologging data

Abstract

Abstract We do not yet have the means to directly measure the energy expenditure of wild animals, particularly at a high temporal resolution. However, we can instead measure factors that correlate with energy expenditure, such as heart rate or accelerometry, that is, energy expenditure proxies. To estimate the magnitude of differences in energy expenditure between contexts (such as day and night, or travelling singly vs. in a group), our energy expenditure proxy must be calibrated. However, perhaps because of the logistical challenges of conducting such calibrations and the observation that most calibrations are linear, researchers sometimes interpret the degree of change in energy expenditure between contexts based on uncalibrated proxy data. When doing so, they implicitly assume that the ratio of the proxy values between contexts represents the ratio of energy expenditure values between those contexts. This approach, however, is usually spurious because unless the relationship between energy expenditure and proxy passes through the origin (i.e. y‐intercept = 0), the ratio of change in the proxy between contexts does not represent the ratio of change in energy expenditure. We model the error induced when data from an uncalibrated proxy are used to interpret changes in energy expenditure. We also estimate the errors made in published articles that employ this approach, by comparing the claimed change in energy expenditure between two contexts with the change in energy expenditure estimated by applying the best available calibrations now available. As predicted by the models we developed, the apparent size of the error in published articles is related to the relative size of the calibration y‐intercept. Species‐ and activity‐specific empirical calibrations are the best option for interpreting proxies of energy expenditure. However, for mammals and birds, enough data have been amassed such that energy expenditure can be predicted from measures of heart rate by knowing animal heart mass or even body mass; perhaps a somewhat analogous predictive tool will be available in the future for accelerometry. A free Plain Language Summary can be found within the Supporting Information of this article.