Zabel: Imputed education histories and fertility analysis in the western German context
1. Introduction
Much of fertility research focuses on the influence of women’s level of education.
Unfortunately, complete education histories are rarely available from survey data.
When they are not, one option is to impute education histories from more rudimentary
information that is available, which might include the highest level of education
attained by the time of the interview and possibly also the date when this level was
reached. The aim of this paper is to determine whether such imputed education histories
can serve as satisfactory substitutes for complete education histories when analyzing the
effects of education on entry into motherhood. In the present paper we compare results
using imputed and complete education histories for the case of western Germany,
applying data from the German Life History Study.
A very similar question has been investigated for the case of Norway. Kravdal
(2004) addresses the implications of having no information on educational trajectories
at all other than the highest level of education at interview, and finds that each of three
methods of imputation lead to substantial deviations from the results produced when
using complete education histories. In the present study, we investigate whether a little
more information, that is, information on the date the respondent attained her highest
level of education, can improve imputation and lead to satisfactory results, at least in
the western German context.
The problem that incomplete education histories pose for the estimation of the
effect of level of education on fertility transitions has also been pointed out by Hoem
and Kreyenfeld (2006). The authors call attention to distortions caused by anticipatory
analysis, in particular when summary statistics of childlessness are presented by level of
education at interview only. They devise an alternative summary statistic based on
time-varying information on educational status and level. For this purpose, complete
education histories would be ideal. However, as they point out, complete education
histories are seldom available, as is the case with the data set they use, namely the
German Family and Fertility Survey. Like the first round of the Generations and Gender
Surveys, the German Family and Fertility Survey provides the date of attaining the
highest level of education at interview, but does not provide any information on
educational trajectories before that date. Hoem and Kreyenfeld (2006) suggest a way to
impute education histories using this basic information. The analyses in the present
study are based on their method of imputation.
The German Life History Study, from which we use data for our empirical
analyses, is exceptional in that it does include respondents’ complete education
histories. First, we use these complete education histories to model the effect of
educational level as a time-varying variable on transition rates to first birth. Next, all
information is ignored other than the highest level of education attained by the time of
136
http://www.demographic-research.org