PARENT SESSION
Oral Session # 13: Statistical Ecology.
Presiding: R Stevens
Monday, August 4. 8:00 AM to 11:30 AM, SITCC Meeting Room 205.
Using auxiliary data on measurement error to improve estimates in population time series.
Ferrari, Matthew *,1, Taper, Mark2, 1 IGDP in Ecology, Universtiy Park, PA, USA2 Department of Ecology, Bozeman, MT, 59717
ABSTRACT- Few, if any, populations are monitored perfectly. Population monitoring often involves a trade-off between precision of the counts and the expense of the monitoring program. As a result, time series of population counts incorporate two sources of variability: the process variance inherent in the population dynamics, and variance due to stochastic measurement error. If ignored, the variance component due to measurement error will result in over-estimates of the biological process variance (a quantity of interest in population viability analyses) and poor precision in estimates of the population trend. Often auxiliary data, in the form of double counts, can be collected which provides information on the measurement error, independent of the time series and independent of the assumed population model. These auxiliary data can be combined with the time series observations to form a joint likelihood which can be maximized to obtain simultaneous estimates of population trend and the process and measurement variance components and associated standard errors. We present a justification for the joint likelihood estimation using a simple example. We also evaluate the performance of the joint likelihood estimation relative to the mixed-model estimators for geometric population growth in the presence of measurement error developed by Staples et al. (2003). The incorporation of auxiliary data through the joint likelihood resulted in a reduction of the expected confidence interval for population trend as well as the variance components without sacrificing the coverage rate. Thus, in future monitoring programs, less expensive, imperfect counts could be improved by collecting coincident data on the measurement process itself and calculating simultaneous estimates of measurement and biological parameters.
Key words: process variation, restricted maximum likelihood, measurement error, trend estimation