Zabel: Imputed education histories and fertility analysis in the western German context
degree,’ 13% stems from the ‘vocational training’ category in the original histories
(Table A3b). This is substantially more than in either of the first two imputations,
where 6% and less than 0.5% (rounded to 0%), respectively, of the imputed exposure
time for the category ‘vocational degree’ stems from the ‘vocational training’ category
in the original histories (Tables A1b and A2b). Since risks of first birth during
vocational training are generally very low, a disproportionately small number of births
is allocated to the ‘vocational degree’ category from the ‘vocational training’ category
along with the additional exposure time (Table A3c). This results in an underestimation
of risks of first birth for the ‘vocational degree’ category, which, as we have seen, is
reflected in the low baseline estimates in the last model shown in Table 2.
Likewise, a substantial amount - 21% - of the exposure time allocated to the
category ‘university degree’ in the last imputation was in reality spent in university
education (Table A3b). Here, too, the corresponding proportion of events allocated to
the ‘university degree’ category stemming from the ‘in university education’ category is
much smaller (Table A3c). Thus, the risk of first birth for those with university degrees
is underestimated as well when using data from the last imputation (Table 2).
Assuming that respondents took longer to obtain their highest degrees than the
standard durations used for the last imputation might improve the estimates to some
extent. Kravdal (2004) found that assuming more realistic, longer durations up until
respondents obtained their degrees rather than normative durations for obtaining each
type of degree gave improved, though still not satisfactory, estimates in the case of
Norway. However, using longer durations of time could on the other hand have the
effect of mis-imputing time that people actually spent holding a degree as time spent in
education or vocational training. This would then likely result in an overestimation of
first birth risks for the ‘in vocational training’ or ‘in university education’ categories in
the imputed histories. If there is a lot of individual variation in study durations, any
assumed standard duration is likely to lead to extensive mis-imputation.
In any case, it seems that knowledge of the date respondents obtained their highest
degrees clearly improves imputations. The first four imputations all give estimates of
first birth risks that are closer to the original estimates than the fifth imputation, for
which no information on the date respondents obtained their highest degree is used.
Thus, when constructing questionnaires, even if time constraints do not allow surveying
complete education histories, it appears it would still be worthwhile to include a
question on the date the highest degree was obtained.
160
http://www.demographic-research.org