Hospital Stay Response Error Estimates for the Health Insurance Study's Dayton Baseline Survey

The measurement quality of Baseline Survey hospitalization responses was examined by using record check and other evaluation methods. The record check was found to be a biased method of evaluating response errors in hospital admission reports; other evaluation methods suggested that respondent admission reports contained little or no average bias. The record check was useful in evaluating the reported details of hospital stays. Response bias averages were like those in similar surveys: length of stay was overestimated by about one-half day, and admission month was recalled about a week more recently than the true date. A limited analysis of out-of-pocket cost reports suggested no response bias although uncertainty remains. Response error variances were moderately low for length of stay and month of admission and uncertain for own cost. Implications for subject matter analyses, future record check evaluation designs, and best estimation strategies are discussed.