While this underestimation is undesirable, it is important to keep in mind that this method produced estimates of summer prey acquisition rates that were, on average, 85% (number of prey) and 62% (biomass of prey) greater than estimates based on data obtained through a single GPS-collared wolf.
However, the generalized scale-up method, which we believe is more statistically appropriate than the scale-up method, produced estimates that were much higher than expected.
This method produced estimates similar (to within 1%) to those obtained by linear regression with logarithmically transformed CAC scores and exclusion of extreme outliers (falling above and below two standard deviations of studentized residuals) (data not shown).
These methods produce estimates that are superior to those of the older methods, but for many researchers, multiple imputation is the general solution to missing-data problems in statistics (Rubin, 1996; Schafer, 1997).
With high observer reliability, both methods required surveying approximately 0.26% of the study area (0.26 km2 out of 100 km2 in this simulation) to achieve an accurate estimate of population size; at or above this sampling effort both methods produced estimates within 33% of the true population size 50% of the time.
The two methods produce estimates of similar magnitude when standardising by age and sex.
The simulation results validate our two proposed methods, producing estimates close to the preassigned values.
Excluding BitSeq (MCMC and VB) and Casper, we see that many methods produced estimates close to the boundary of the parameter space, as seen in Figure 5.
Propensity score methods produce estimates that are more accurate than logistic regression estimates when there are seven or fewer events per confounder, as was the case in the present study [ 24].
While previous literature has shown that regression methods produce estimates with a higher mean square error and have less power (Kao, 2000), these results apply primarily to the case of interval mapping when the spacing between markers is wide (>1cM).
