Sir.—Brown et al,1 in the February 1987 issue of AJDC, reported on scarring retinopathy of prematurity (SROP) and permitted a classic bias to distort their observations. They say that they have such vast amounts of data and missing data (empty cells) that appropriate "data matrices are not amenable to present statistical techniques." The fault is not in statistical techniques, it is in the research structure.
The research structure here uses procedures that produce massive attrition, sometimes called "transfer" bias. In their Figs 1, 6, and 7, the attrition of the subjects (SROP and non-SROP) is clearly shown. In Fig 6, by the fourth week of the study period, there are only about 30% of the original infants still contributing data (n = 9, SROP; n = 18, non-SROP).
The data (eg, arterial partial pressure of oxygen [Po2] values in Fig 6) are collected from the "sicker" infants in each