Page 135 - IJPS-11-3

International Journal of
Population Studies Drivers of reproductive delay in the UK

DCE methodology and that provides specialized software The DCE produces data that can be statistically analyzed
to administer the DCEs, secure servers on which to store to estimate the value of each level of each attribute and
and move the data, and have a UK nationally representative infer participant preferences. Here, mixed logit models
participant pool. The four DCEs were administered to which allow for preferences to vary across individuals
a sample of 1228 reproductive-age men and women, (Hess & Train, 2017) within demographic groups were
around half of whom were university-educated (n = 618) used. The model is estimated as follows (for example, using
and half who were not (n = 610). Fifty-three people who non-university women):
expressed never wanting children or having reached their U = V +ϵ
desired completed family size, and therefore not wanting n n n
more children, were screened out, leaving 688 people =β ×partnership stable + β × partnership less stable + β ×
3
2
1
who stated that they might want (more) children and 485 home large + β × home smaller + β × finances good + β × finances ok
5
4
6
who definitely do. One participant who reported being + β × family full support + β × family part support + β × baby comes 6
7
9
8
18 years old and also claimed to have already completed + β × baby comes + ϵ n
10
18
a university degree was removed leaving a final sample Where U is the participant’s “utility” associated with the
size of 1174. The decision to remove participants who said scenario presented, V is the deterministic component, and
they definitely do not want children was because it was the epsilon represents an error term. The data produced are
objective of this study to identify the barriers for those who binary with 1 denoting the scenario chosen and 0 denoting
are postponing having a child that they would like to have. the alternative. Statistical analyses were carried out using
However, the reasons those people definitely do not want Stata v.17 (StataCorp, 2021) and the plots were created
children may be because they perceive the barriers to be using ggplot2 in R.
insurmountable and may have had a child or another child,
if circumstances were different, should be acknowledged. 2.4.1. Interpretation of results
Unfortunately, these reasons were not discernible from the Mixed logit models produce odds ratios for each of the
data collected, and this potential limitation will be taken levels of the attributes compared with a reference category.
into account in future research. The screening question In this study, the least appealing level (level 3) is set as the
that asks if “they definitely do not want children” also reference category, so the expected outcome is for odds
means that the data are only representative of the national ratios higher than 1, denoting the assumption that people
population that might want to have children. prefer levels 1 and 2 over level 3. Willingness to pay results
2.4. Statistical analysis are given in months with an aggregate number for that level
of the attribute with its accompanying upper and lower
The DCE was designed using the four most important bounds of 95% confidence intervals. WTP estimates can be
attributes that emerged from the qualitative work, with each negative or positive with slightly different interpretations,
of these being split into three levels denoting (i) an ideal although substantively they mean the same. Negative
situation, (ii) a less-than-ideal situation, and (iii) a poorer WTP is the estimated additional amount of time needed
situation. No level was made to be unrealistically high or to compensate for a particular barrier to reproduction, for
low. A D-efficient design matrix was generated using the example, I would need six more months to be able to get a
DCETool R-package (Perez Troncoso, 2022). D-efficient pay rise before having a baby. Positive values indicate how
designs reduce D-error which improves the experimental much time a person would give up to have the barrier
design by producing minimal overlap of choice sets, well- removed, for example, I would wait six more months before
balanced levels, orthogonality, and utility balance (Rose having a baby in order to have a pay rise. In effect, these can
& Bliemer, 2009). All the DCE attributes and levels were both be understood in terms of how much reproductive
pre-tested for clarity of understanding and relevance. For time is lost.
example, for “partner support” for the university-educated
women DCE, the three levels were as follows: 2.5. Preliminary tests
(i) My partner is fully involved and happy to share the This study was partly a methodological test case to address
childcare 50/50 some peripheral questions to aid future data collection
(ii) My partner is supportive but not very hands-on with and statistical modeling. Before collecting the data, the
childcare expectation was that different groups would find the barriers
(iii) My partner travels often and leaves the childcare to more important at different ages. I expected that university-
me. educated people would not feel they were “delaying”
Full descriptions of all the attributes and their levels are children until they had established a career, likely well into
given in the Appendix. their 20s. For non-university folk, the barriers to having

Volume 11 Issue 3 (2025) 129 https://doi.org/10.36922/ijps.3600