부산대학교 도서관

This paper studies correlation coefficient estimation of food frequency questionnaires (FFQs) and the underlying average of repeated food records. In some clinical trials, fat intake from FFQ is used as a screening criteria. As a result, repeated food records are available only among subjects with FFQ-measured dietary intake of percent energy from fat above a prespecified quantity. Wang et al. (1999) proposed methods for the situation when the average of 4-day food records is used. In this paper, we extend their methods to the situation when repeated food records, rather than a single average, are available for analysis under the restricted sampling due to a screening FFQ value. Intensive simulation studies have been conducted to compare an estimating equation method and a simple imputation approach. Finite sample performance is presented. We show that the naive estimator may either underestimate or overestimate the correlation coefficient.