Search

Attrition may bias estimates of the effectiveness of child nutritional supplementation programs, but modern approaches to characterizing the pattern of data missingness are not widely used. We analyzed data from the 2392 participants in the INCAP Longitudinal Study (1969-77) to examine patterns of missingness by attrition category at follow-up in 2002-4 (interviewed (n=1571), dead (n=274), untraced (n=101), living outside Guatemala (n=162); traced but not interviewed (n=284)). Determinants of attrition included male sex, younger age, village of birth, father's schooling, and parental socioeconomic status. Height (mean±SD 1.63±0.06 m (males) and 1.51±0.06 m (females)) was measured in 2002-4 for 1329 persons and at age>18y in an earlier follow-up for an additional 186. We assessed listwise deletion, mean substitution, regression imputation, and multiple-imputation-chain-estimation. All 4 approaches yielded consistent sex-specific estimates; mean substitution resulted in narrower SD. Imputed heights did not differ across attrition categories (analysis of variance p=0.6 (males), p=0.7 (females)), or between those measured and all non-measured considered as a single group (t-test p=0.6 (males), p=0.20 (females)). In this study, even non-random attrition appears not to bias estimates of attained height and is unlikely to bias estimates of effectiveness of interventions. [ABSTRACT FROM AUTHOR]