An Illustration of the Problems Caused by Incomplete Education Histories in Fertility Analyses

When assessing the importance of education for fertility, one should ideally use complete education histories. Unfortunately, such data are often not available. It is illustrated here, using register data for Norwegian women born in 1969, that inclusion of educational level at the latest age observed (28), rather than at the current age, can give substantially biased education effect estimates. It is also illustrated that imputation of education for earlier ages may lead to wrong conclusions. A simple imputation of educational level and enrolment based on the assumption that everyone passes through the educational system with the officially stipulated progress gives particularly misleading results. Somewhat better estimates are obtained when a slower progress more in accordance with reality is assumed, or when educational level and enrolment are imputed stochastically on the basis of distributions calculated from real data. Obviously, one should be very careful when faced with incomplete education histories, and try to make use of relevant information from other sources about the actual educational careers.