{"id": "e7ab8054b7a19735fc3d44e808d347755e5bf04872c72cca6eea7773cad28b66", "pubkey": "fd208ee8c8f283780a9552896e4823cc9dc6bfd442063889577106940fd927c1", "created_at": 1752071125, "kind": 30041, "tags": [["d", "less-partnering-less-children-or-both-3-data-and-methods-6-by-julia-hellstrand-v-1"], ["title", "3 Data and Methods"], ["author", "Julia Hellstrand"], ["m", "text/asciidoc"], ["M", "article/publication-content/replaceable"]], "content": "3.1 DatannIn this study, we used Finnish national longitudinal population register data compiled at Statistics Finland (permission no. TK-52-1119-17). The register data were linked to different register sources such as information on childbirths, housing and educational attainment through personal identification numbers, offering full-coverage of the entire Finnish population. The study population consists of all childless men and women aged 15u201345-years-old permanently living in Finland on the last day of each year from 2000 through 2018. Individuals were followed until they had a first biological child or until they reached the age of 45. In total, the study population consists of 2 532 375 individuals and 23 847 070 person-years. Less than 0.06% of all first births were linked to two biological mothers/fathers. Consequently, the true parent for these children remained unknown. We excluded from our study 388 individuals linked to such a first birth.nnFor each individual, data include personal information on family status (single, cohabiting, or married) at the end of each calendar year. Statistics Finland defines cohabitation as a union of two unmarried adults of the opposite sex aged 18 or older who have been living in the same dwelling for at least three months, who are not siblings or differ in age by 16 or more years (Official Statistics of Finland (OSF), 2021). An individual is considered single if s/he is not living in a cohabiting or married union. Among the study population, 2.1% of men and 1.5% of women (446 787 observations) had missing information for family status (institutionalised population and/or otherwise unclassified) and were thus excluded from the study.nnWe formed yearly transitions for all individuals in the study population for whom personal information was available for two consecutive years. Information for two consecutive years was missing for all first entries into the study population (2 517 735 observations) and for individuals absent from the Finnish population during some period from 2000 through 2018 (130 467 observations). Furthermore, to avoid challenges related to incomplete educational data and an unknown number of unregistered first births to non-native Finns, we excluded individuals born abroad 1 (1 280 473 observations for 229 670 individuals). In total, we identified 19 468 815 yearly transitions between states (single, cohabitating, married, and first birth) for 2 125 172 individuals beginning in 2000. Among these, 740 537 were transitions to first births2, and 2 911 543 were transitions between partnership states. Appendix Table 1 provides descriptive information about first births and partnership transitions in more detail.nnWe also estimated the transition probabilities based on SES. We considered four categories of educational attainmentu2014primary, secondary, lower tertiary, and higher tertiary. Primary-level includes those who completed at most a lower secondary level of education (ISCED 0u20132), while secondary level refers to those who completed upper secondary and post-secondary non-tertiary levels of education (ISCED 3u20134). Lower tertiary includes short-cycle tertiary education and a Bacheloru2019s degree or the equivalent level (ISCED 5u20136), while higher tertiary refers to those who completed a Masteru2019s degree, doctoral degree or the equivalent level of education (ISCED 7u20138). We used income as a complement to education as a robustness check to overcome the limitations related to using educational attainment as an explanatory variable in the period analysisu2014that is, currently, less educated groups include those who will later attain more advanced degrees. The income variable refers to each individualu2019s annual income subject to state taxation and includes both earnings and social-security benefits. Four income groups were formed based on income quartiles stratified by age, year, and gender. Because those enrolled in educational programmes in particular are known to exhibit lower birth risks (e.g. Kravdal, 1994), we performed a sensitivity analysis which excluded students (shown in the appendix).nn3.2 MethodsnnWe used a Markov chain multistate approach, which describes the transitions between a given set of states using transition probabilities (Briggs & Sculpher, 1998). A Markov chain evolves in discrete time and moves step-by-step from state i to state j , with the property of being memoryless. That is, the probability of each transition depends only on the state attained in the previous step and not on the history of events (Kemeny & Snell, 1971). The transition probabilities from state i to state j at a specific age and time are defined asnn.State transition diagram for the Markov chainnimage::https://i.nostr.build/sFlRaApoHqolB7hS.png[Figure 2, 300]nn_Formula 1 as LaTeX_n[stem]++++np_{ij}(age, t) = pr\left( State_t = j \mid State_{t-1} = i;\, age_{t-1} \right)n++++nnimage::https://i.nostr.build/rRkC2LZrPEOjWZqJ.png[Formula 1, 300]nnThe step size in our analyses is one year.footnote:[When we refer to a specific year, we refer to the end of that specific year.] Our state space includes the states of u2018singleu2019, u2018cohabitatingu2019, u2018marriedu2019, and u2018first birth. An illustration of the state space and the transitions between these states appears in Fig. 2. In our analysis, we distinguish between the transitions from u2018singleu2019 to u2018first birth and singleu2019 and u2018first birth and unionu2019 in order to distinguish single parents from couples who begin cohabitating closer to the first birth event. The first birth event represents an absorbing state, meaning that once entered it cannot be left. All other states are nonabsorbing (transient) states. We estimated the yearly age-specific transition probabilities for each of the given set of states between the ages of 15 and 45 from 2000 through 2018 asnn_Formula 2 as LaTeX_n[stem]++++np_{ij}(x, t) = frac{n \#\text{individuals in state } j \text{ in year } t \text{ aged } x \text{ and in state } i \text{ in year } t-1n}{n \#\text{individuals aged } x \text{ in state } i \text{ in year } t-1n}n++++nnimage::https://i.nostr.build/CngNXtCcNthAaO62.png[Formula 2, 300]nnusing simple cross tabulations.footnote:[Fitting multinomial logistic regression models to the data would be an alternative, but the resultsnwould not differ.] The probabilities were estimated separately for men and women, as well as for educational and income groups, respectively.nnWe used the estimated transition probabilities and counterfactual simulation[Related methodological approaches have been applied only in few prior studies: hazard ratios werenimplemented in microsimulation models to link fertility to marital behaviours in Canada (Bu00e9lange et al.,n2010), and a counterfactual approach was employed to examine the impact of union dissolution on fertil-nity in Uruguay (Fernu00e1ndez Soto & Laplante, 2020).] to calculate what proportion of the decline in first births was attributable to changes in union dynamics versus the decline in fertility within unions.footnote:[Our counterfactual design does not necessarily take into account (changes in) the order of events. Forninstance, if we were to observe a first birth decline attributed to changes in declining marriage rates, itncould in part reflect a tendency to marry increasingly after the first birth.] (For specific details, see the Technical appendix.) First, we calculated the age-specific first birth rates that would have been observed if the population in 2010 would have experienced the 2010 transition rates in the period from 2010 through 2018. We labelled this scenario u2018constant probability birthsu2019. Using the age-specific first birth rates, we calculated the proportion ever having a first birth according to a life-table approach. Second, we calculated the age-specific first birth rate and the proportion ever having a first birth that we expect to have observed if the population in 2010 would have experienced the observed changes in transition rates in the period from 2010 through 2018. We labelled this scenario u2018natural course birthsu2019. We decompose the difference between these two scenarios by changing the transition probabilities one at a time. For education groups, we adjusted the procedure to take into account that the study population progresses to higher education levels over time. Additional details appear in the Technical appendix.", "sig": "a577b3609aa4773b2468343b1643b3fac379bf2b5386aaa0375e166cacb8e8b78603068b36eca0bb192faf0524565c6cffcba055d96187702e3d0798bc901bb8"}