Changes in g with k
Loading...
The GenEst (GENeralized ESTimator of Mortality; Dalthorp et al. 2018; Simonis et al. 2018) and Evidence of Absence (EoA; Huso et al. 2015, Dalthorp et al. 2017) fatality estimators both incorporate a parameter, k (or the detection reduction factor) that describes how searcher efficiency changes for a carcass on successive searches. k influences the overall detection probability for a carcass but can be costly to estimate, and interacts in a nonlinear way with searcher efficiency and carcass removal such that there are some conditions under which detection probability is highly sensitive to the value of k, and other conditions under which the overall detection probability is not very sensitive to k.
One common question managers ask is, 'Do I need to estimate k?' To help answer this question, this app allows users to explore the sensitivity of detection probability to k, given a set of other parameters. In some cases, it may be sufficient to make an assumption about the value of k and avoid the expense of estimating it. In other cases, the overall detection probability may depend strongly on the value of k, and users may wish to consider estimating k in the field. Click the 'Sensitivity of detection to k' tab to see how detection probability reacts to k under a variety of conditions.
k works by modifying the searcher efficiency for carcasses that have been missed one or more times. If searcher efficiency for a fresh carcass is p, then searcher efficiency for a carcass that has been missed once will be p * k, for a carcass that has been missed twice will be p * k * k, and so forth. The Huso estimator implicitly assumes that k = 0 (i.e., if a carcass is missed once it will never be detected). The Shoenfeld estimator implicitly assumes that k = 1 (i.e., searcher efficiency is identical no matter how many times a carcass has been missed). To see how k modifies searcher efficiency, click the 'What is k' tab.
GenEst is an open source R-package developed by the USGS in collaboration with WEST, Inc. and others. The package implements a flexible statistical approach to estimating mortality given bias induced by searcher efficiency, carcass persistence on the landscape, and the decline in probability of detection for a present carcass over time (a quantity known as k). The GenEst app and documentation is available as a web app on WEST’s Shiny service and as a downloadable R package at the Comprehensive R Archive Network (CRAN), or from the USGS.
EoA is an open source R package developed by USGS that is suitable for estimating mortality when mortality is a rare event (i.e. zero or few carcasses are detected). It is available as an R package from the USGS, and an intuition-building version is implemented as a web app on WEST’s Shiny service.
Dalthorp, Daniel, Huso, Manuela, and Dail, David, 2017, Evidence of absence (v2.0) software user guide: U.S. Geological Survey Data Series 1055, 109 p., https://doi.org/10.3133/ds1055.
Dalthorp, D., Madsen, L., Huso, M., Rabie, P., Wolpert, R., Studyvin, J., Simonis, J., and Mintz, J., 2018, GenEst statistical models—A generalized estimator of mortality: U.S. Geological Survey Techniques and Methods, book 7, chap. A2, 13 p., https://doi.org/10.3133/tm7A2.
Huso, M. M. P., Dalthorp, D., Dail, D. and Madsen, L. (2015), Estimating wind-turbine-caused bird and bat fatality when zero carcasses are observed. Ecological Applications, 25: 1213–1225. doi:10.1890/14-0764.1
Simonis, J., Dalthorp, D., Huso, M., Mintz, J., Madsen, L., Rabie, P., and Studyvin, J., 2018, GenEst user guide—Software for a generalized estimator of mortality: U.S. Geological Survey Techniques and Methods, book 7, chap. C19, 72 p., https://doi.org/10.3133/tm7C19.
We quantified 'sensitivity' by regressing log(g) against k for a given searcher efficiency and mean/median carcass removal time. The slopes of these regressions is what defines the color of the cell in the heatmap for sensitivity across searcher efficiency and mean/median carcass removal time.
This table is showing the estimated detection probability (g) at specified values of k, given the chosen parameters. The table serves as a quick reference for how much higher g can get by estimating k. The final column (1/g) shows the number of fatalities that would be estimated for each detected carcass, given g, k, and other parameters.
The value of 'Other adjustment factors' is applied as a linear scaling factor to the estimated detection probability. The value is meant to represent and integrate over all other correction factors that might be linearly applied, like an area correction and the sampling fraction. For example, if the area correction is 0.8 and the sampling fraction is 0.5, then your adjustment factor would be 0.8 * 0.5 = 0.4.