Demographic Research: Volume 21, Article 6
A1b). In turn, the estimation results in Table 2 for the imputations that use the first date
the highest degree was attained give lower risks of first birth for the category
‘vocational degree’ than do the results using the original histories. The category
‘vocational degree’ is the reference category here, so we can compare risks of first birth
for this category just by looking at the baseline. The estimates for the baseline are
generally slightly lower in the first and the third imputation, which use the first date the
highest degree was attained, than in the original histories, except for the youngest age
group9 (Table 2).
By contrast, when the last date the respondents attained their highest degree level
is used for the imputations, this should in principle lead to an over-estimation of the risk
of first birth for the category ‘vocational degree.’ This is because, as described in
section 2 on potential sources of bias, people who do not have a child after they receive
their first vocational degree may be more likely to go on to obtain a second vocational
degree. Then, they will be selectively taken out of the category ‘vocational degree’ in
imputations using the last date a vocational degree was attained (and instead imputed to
be enrolled in education in the gap between the two spells of educational enrollment).
Comparing Tables A2a and A2c, it seems that this is reflected in the disproportionately
small number of events that are removed from the category ‘vocational degree’ as
compared to the 10% of exposure time that is removed from this category in the
imputed histories. This appears indeed to result in an overestimation of the risk of first
birth for the category ‘vocational degree’ in the model estimation results for the second
and fourth imputations in Table 2, which use the last date the highest degree was
attained. Again comparing the baseline estimates for these imputations to those for the
original histories (since ‘vocational degree’ is the reference category), we can see that
the risk of first birth is slightly overestimated.
But, altogether, deviations from the original baseline estimate (which gives the risk
of first birth for the reference category ‘vocational degree’) are not very large in any of
the first four imputations. In the section on potential biases caused by imputation, it was
suggested that although it appears to be quite common for members of this cohort to
hold multiple degrees, gaps between spells of educational enrollment may be quite
small. It is likely that the estimates would become more severely distorted for
imputations using the last date the respondents received their highest degree if more
people tended to obtain a second degree much later in their life course and gaps
between education spells grew larger. Imputations using the last date respondents
received their highest degree would then produce results that more strongly
overestimate the risk of first birth for the category ‘vocational degree.’ However, this
9 This is likely to be due to the fact that not many people have completed their first vocational degree and
already started on their second before the age of 19, and that birth risks at this age generally are very low.
http://www.demographic-research.org
151