Estimates of gestational age

To the Editor. — The article by Kramer et al1reaffirms the poor quality of estimates of gestational age afforded by dating of the last normal menstrual period (LNMP). The authors express surprise that the magnitude of the bias toward the mean of such estimates increases the further the LNMP estimate differs from the mean of the estimates. This should not come as a surprise. In fact, it can be anticipated from probabilistic considerations without collecting any data. To illustrate why bias toward the mean increasing as distance from the mean increases could be anticipated, consider the following hypothetical simplification. Suppose that all gestational ages at delivery are 39 weeks, but that the LNMP estimate has a distribution ranging from 36 to 42 weeks. Then consider a plot, like the figure on page 3307 of the article by Kramer et al, of the difference between the LNMP estimate