Bias and confidence interval coverage of creel survey estimators evaluated by simulation

The Wisconsin Department of Natural Resources has carried out a complete creel census on Escanaba Lake, Wisconsin for more than 40 years. We used this creel census data set as the basis for simulations of a stratified random three-stage creel survey (stages were days, shifts, and count times) in which harvest was estimated as the product of effort and harvest rate. Effort was estimated from instantaneous counts of anglers, and harvest rate was estimated from completed-trip interviews. We evaluated the bias and precision of estimators of annual angler effort and harvest with creel census data from 3 years of varying angler effort and harvest. This creel survey method resulted in excellent estimates of annual effort. There was no evidence of bias, and coefficients of variation were less than 0.10 even though the standard errors of estimates were somewhat too large, resulting in conservative 95% confidence intervals (97–99% coverage). We found no evidence of bias for a stratum estimator of harvest in which harvest rate was estimated across all interviews in a stratum before multiplying by effort to estimate harvest. Coefficients of variation were less than 0.20, and confidence interval coverage was close to the targeted 95% level. An advantage of estimating harvest rate across all interviews in a stratum is that the sample size on which harvest rate estimates are based is then relatively large. We did find evidence for bias in a daily estimator of harvest in which harvest rate estimates were based only on interviews obtained each day.