Comparing Adjustment Methods for Nonresponse in a List Based Telephone Survey of Marine Anglers

The literature on nonresponse effects in surveys of recreational anglers is mixed but generally indicates that nonresponse can introduce systematic errors for certain types of estimates including totals of catch and effort.  Not all studies have found evidence of bias, however, with some reporting mixed effects: bias present for totals but not for means or proportions conditioned on an indicator of fishing (e.g., respondent being an angler or trips being reported).  This study investigates nonresponse in the Large Pelagics Telephone Survey, a list based survey that estimates effort in the permitted recreational fishery for highly migratory species along U.S. mid- and North Atlantic coasts.  Survey response rates are relatively high, typically greater than 70%, and current sample-based weight adjustments assume random nonresponse.  To evaluate this assumption, comparisons were made among the current sample-based adjustment and three alternative weight adjustment methods: response propensity scoring, ratio raking, and constrained ratio raking.  Alternative methods made use of auxiliary permit information available for all units on the list frame.  Logistic regression was used to identify correlates of response and trip reporting among auxiliaries.  Although differences were small overall, comparisons of estimates for a subset of survey variables generally supported prior findings of mixed effects with larger differences detected for totals than for conditional measures.  While alternative methods increased variability in weight adjustments, they also made limited improvements in precision for some survey estimates.  Results suggest use of alternative nonresponse adjustment methods is warranted even in angler surveys with high response rates.