The validity of retrospective breast-feeding-duration data: an illustrative analysis of data quality in the Malaysian Family Life Survey

The effects of breast-feeding on infant health and mortality, particularly in the developing nations, are a matter of controversy and importance. The Malaysian Family Life Survey (MFLS) of over 1200 women has recently been the source of a great deal of valuable information on the influence of breast-feeding and interacting social variables on the incidence of infant mortality. Accuracy of reporting of breast-feeding duration is a key issue in the validity of studies of breast-feeding and infant mortality. This paper presents an illustrative analysis of the quality of breast-feeding data from the Malaysian Family Life Survey, using logit model schedules. Lesthaeghe and Page derived a logit model schedule of breast-feeding, summarizing empirical experience. This family of model breast-feeding duration curves is similar to the logit model life tables developed by Brass, and was intended for similar applications. To verify the MFLS retrospective breast-feeding reports, the observed median duration and variability were calculated for ethnic group/cohort subsets, and expected duration distribution curves were generated from the model using these observed parameter values. The expected curve generated from the model fit the observed curve of breast-feeding discontinuation extremely closely. Thus it is unlikely that any significant distortion of the pattern of discontinuation of breast-feeding occurred in data collection. Extensions of this method of data quality checking to other duration distributions are suggested.