Geographic variation in parity progression in Australia

Australia has moderately high fertility compared to many Western-industrialized countries. The current total fertility rate is around 1.88, but fertility levels are not uniform across the country. There is a distinct geographic pattern with the total fertility rate about 0.5 higher in remote and very remote Australia (2.33) compared to major cities (1.82). In this paper, we examine 2 explanations for this pattern: the compositional hypothesis and the contextual hypothesis. Using event-history methods with joint modelling to investigate parity progression, we find that after taking into account differences in age, country of birth, indigenous status, relationship status, education levels, and economic activity, women living in smaller towns in regional Australia are more likely to have a first, second, and third birth. Further, there is lower propensity to have a first child in inner or middle city areas that are characterized by smaller and more expensive housing than suburban or regional areas.


| INTRODUCTION
Australia presents an interesting case for examining the geographic distribution of fertility. It is one of the largest countries in the world and also one of the most urbanized with over 75% of the population living in cities of over 100,000 residents (Department of Infrastructure and Transport, 2013). Demographic characteristics, including socioeconomic levels and ethnic background, vary widely between those living in different sized cities, as well as between city dwellers and people living in rural or remote regions of the country.
Fertility rates also differ considerably across the country and are significantly higher in smaller towns and remote localities compared to cities. The national total fertility rate (TFR) is 1.88, but in remote parts of Australia, it is 2.33, and in major cities, it is just 1.82 (ABS, 2014a). 1 The highly urbanized nature of the Australian population means that the national TFR is strongly affected by the fertility levels in cities where the majority of people live. However, the higher fertility levels of rural areas have an important elevating effect on national TFR, as seen in Figure 1.
To date, few studies have examined the geographic distribution of fertility in Australia in any depth. In this paper, individual-level longitudinal (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012) data from the Household, Income and Labour Dynamics in Australia (HILDA) survey are used to examine the extent to which geography is an important determinant of childbearing. We jointly model the likelihood of having a first, second, or third child among women aged 16-44 living across Australia and investigate whether the effect of geography, if any, varies by parity. We further determine whether the spatial variation in parity progression persists even after controlling for compositional factors or differences in people's socioeconomic and demographic characteristics.
these studies have used different approaches and different levels of geographic disaggregation, they all observe considerable spatial variation in fertility and higher rates of childbearing in rural areas.
Two explanations have been suggested for this pattern. One is the compositional hypothesis that posits that people living in urban and rural areas have different socioeconomic and demographic characteristics, such as education levels or age profiles, and if these characteristics are also related to fertility that could explain why fertility levels may differ across geographic areas (Fulda, 2015). The contextual hypothesis, on the other hand, suggests that there is something more, over, and above a compositional effect, which has a positive or negative effect on fertility. Contextual factors could include structural reasons such as differences in local labour markets, environmental factors, or differences in cultures and attitudes regarding childbearing (Basten et al., 2011).

| Compositional effects
Much of the urban-rural differences in fertility are thought to be due to a compositional effect or "residential sorting" (Fiori et al., 2014). Among women aged 20-24, 20% of those living in major cities are either cohabiting or married, compared to 30% of their peers in inner or outer regional locations (ABS, 2011).
In addition to socioeconomic characteristics such as income and education, each country or region studied may also have its own unique influences that could affect the geographic pattern of fertility.
In Australia, one important compositional factor to consider is regional differences in the percentage of residents who are indigenous.
National statistics show that Aboriginal and Torres Strait Islander women have larger families, with a TFR of 2.34 (ABS, 2014a).
Aboriginal and Torres Strait Islander people are also 12 times as likely to live in remote or very remote areas as non-Indigenous people (22% compared with 2%, ABS, 2014b). The country of birth of a person is also important to consider as Australia has a large percentage of residents born overseas. Australia's overseas-born residents tend to have lower rates of childbearing-although there are some differences by country of birth. Those from English-speaking countries such as the United Kingdom and New Zealand tend to have fertility rates similar to Australian-born women, whereas those from North-East Asia (China, Hong Kong, South Korea, and Japan) tend to have lower fertility (ABS, 2015). The overseas-born population not only has a different fertility profile to the Australian-born population, they are also much more likely to live in capital cities (ABS, 2014b). Controlling for country of birth and indigenous status should therefore go some way to explaining differences in the regional variation of fertility levels.
We acknowledge that the composition of a place is not a static concept. Geographies change as different people move in and out of areas. So the composition of the population may be affected by migration to or from an area. Scholars have noted that short-term mobility is often associated with a change in life course circumstances, such as a change in family size and birth of a child (Clark & Huang, 2003;Haynes & Martinez, 2015;Kulu, 2008;Kulu & Steele, 2013;Rossi, 1980). Rural areas or peripheries of large cities may be seen as more suitable places to raise children leading to selective migration to these areas by those who are interested in starting a family. Similarly, those with less interest in childbearing may selectively migrate to cities (Hank, 2001).
This selectivity hypothesis is just one theory of how rural-urban and urban-rural migration may relate to fertility. Some other hypotheses that have been proposed include the socialization hypothesis, the adaptation hypothesis, and the disruption hypothesis (Kulu, 2005).
The socialization hypothesis suggests that migrants will have been strongly influenced by the fertility norms in their place of origin and maintain childbearing behaviour similar to the stayers of their place of origin, even after they have migrated. In contrast, the adaption hypothesis suggests that migrants will be influenced by their new environment and take on and exhibit similar childbearing behaviour to that which prevails at the destination they have migrated to. Finally, the disruption hypothesis suggests that immediately after a movement, FIGURE 1 Total fertility rate (TFR) by remoteness area (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013) the fertility of migrants may be limited by the disruption caused by the migration itself (Kulu, 2005).

| Contextual factors
If compositional factors do not explain all of the geographic variation in fertility levels, then it could be argued that contextual factors are at work.
As Lupton and Kneale (2010) put it, the idea is that living in a particular place may exert an independent influence such that people with similar socioeconomic characteristics may act differently depending on location.
For fertility behaviour, this could be because of structural factors or more abstract pathways such as differences in social norms and social networks (Fiori et al., 2014). Relevant structural factors could be economic conditions such as the local labour market, material factors such as the availability of and affordability of housing, or the supply of family-oriented services and infrastructure (Basten et al., 2011). Lack of available abortion services in regional and rural areas could also affect fertility (Evans, 2003). These structural factors may be particularly important for a country such as Australia where distances between places can be very large. For example, due to differential access and availability of health services, Australian women in urban areas are significantly more likely to have ever used emergency contraception compared to those in regional or remote areas (Smith, Rissel, Richters, Grulich, & Visser, 2003). Other contextual factors may be harder to quantify. Living in a smaller rural town, young women may not feel comfortable getting contraception from local doctors or health care providers due to confidentiality concerns. This has been suggested as a possible contributing factor to high teenage pregnancy rates in some parts of rural Australia (Quine et al., 2003;Roberts, Graham, & Barter-Godfrey, 2011).
Social norms and networks may also indirectly affect childbearing behaviour. It has been proposed that if a particular location has high fertility, this may have a "contagion" effect on others living in the area as they are influenced by this high-fertility setting and prompted to have children of their own. There has been some research that has looked into so-called neighbourhood effects on teenage pregnancy, but the results have been weak or mixed (Arai, 2007;Evans, 2001;Lupton & Kneale, 2010).
The degree to which compositional effects can explain geographic differences is still a matter for much debate. For the United Kingdom, Fiori et al. (2014) concluded that although the distribution of individual demographic and socioeconomic characteristics account for some of the variability in fertility levels, there are still other factors that are also at play. In Western Germany, Hank (2002) argues that regional variation in fertility is accounted for almost exclusively by the spatial distribution of individual characteristics. Examining small-area differentials in NSW using the 1986 Census, Wilson (1990) came to the conclusion that compositional factors "explain away" most of the differences in fertility levels in Australia and that it is inherently uninteresting to include a spatial perspective in the analysis of fertility in developed countries.
In Australia, beyond descriptive statistics, very little analysis has been conducted taking a more in-depth look at fertility differentials from a geographic perspective (Faulkner, 2005). In this paper, we examine the geographic variation of fertility in Australia in greater detail. Using individual-level survey data, we investigate to what degree differences in the characteristics of individuals explain the pattern of higher fertility in regional Australia.

| Data and definitions
We use longitudinal panel data from the HILDA; 12 waves of data, from 2001 to 2012. Using event-history analysis, we model the likelihood of having a first, second, and third birth for women and examine the effect of geography after controlling for individual-level characteristics. The sample is restricted to women aged 16 to 44 who are in the survey for at least two waves. For the transition to first births, 3,798 women are studied over 18,602 person-years. During the period of study, about one quarter of women had a first birth (1,065 first births).
For the second birth transitions, 1,828 women are studied over 6,354 person-years. During this period, nearly half progress to have their second child (889 births). For the third birth analysis, 1,894 women are studied over 8,267 person-years. During this period, 20% have a third child (382 births).
We use multilevel logistic regression to model the transition to birth. Multilevel modelling is used to take into account that we have multiple observations (waves) per individual with some variables changing within persons over time. The three parity transitions were modelled jointly using the applied MultiLevel (aML) software package to take into account unobserved heterogeneity.
where ln h is the log-hazard of the birth and superscripts (b1), (b2), and (b3) denote first, second, and third births, respectively. The log-hazard of each birth is assumed to be a function of the baseline log-hazard, yT (t), which is a piecewise-linear spline. The spline is a transformation of a continuous variable into a vector of variables, where each new variable represents the original variable on a specific segment of its range (Lillard & Panis, 2003 3. other city with a population of 20,000+, 4. smaller town or place in Inner Regional Australia, and 5. outer regional, remote, or very remote Australia. It is important to distinguish between central cities and outer suburbs of these cities as suburban fertility is often markedly higher in the periphery of large cities (Kulu & Washbrook, 2014;Kulu et al., 2009).
For Sydney, Brisbane, and Melbourne, the distinction between inner or middle and outer suburbs was based on the classification used by the Bureau of Infrastructure, Transport and Regional Economics The remaining categories, (4) smaller town or place in inner regional Australia and (5)  For example, according to the Remoteness Area classification, Toowoomba is classified as "Inner Regional Australia," but in our analysis, it is classified as "other city with a population of 20,000."

| Control variables
The sex of the existing children is included as an explanatory variable for the higher order parities. For the second birth model, we classify whether the first child is male or female. For the third birth model, we classify whether the first two children were (a) male and female, (b) both males, and (c) both females. Previous research from Australia has shown that parents who have a son or a daughter are equally likely to have a second birth (there is a two-child norm in Australia), but at higher parities, parents of two boys are more likely to progress to a third birth than parents with a son and a daughter or two daughters, indicating a daughter preference (Gray, Evans, Anderson, & Kippen, 2010).
Acknowledging that migration and housing moves are often associated with changes in life course circumstances, we also consider whether the respondent has moved since the last wave. We distinguish between those who have not moved, those who moved from another area of a similar population density, those who moved from less densely populated area, and those who moved from a more densely populated area. We also include the number of bedrooms in the current residence, as an additional child may necessitate an extra bedroom for accommodation.

| First births
In the first model of Table 2, we see the association between location and the hazard of first birth controlling only for age. A clear geographic pattern is evident with first birth transitions being higher in all locations compared to inner and middle suburbs of capital cities, particularly in inner regional small towns and outer regional or remote areas. In the next step (model 2), country of birth and relationship sta- Number of bedrooms in the current residence is associated with childbearing, in that people living in a house with one to two bedrooms are less likely to have a first birth than those living in a three-bedroom house. Conversely, those in a four-bedroom property are more likely to have a first child. The measure of movement since last wave is not significant for first births.

| Second births
Unlike for first births, for second births, as seen in Table 3, location appears to have no impact on the hazard of having a birth in the base model. After adding controls for country of birth, relationship status, and age at previous birth (model 2), we find that in comparison to women living in inner and middle suburbs of capital cities, those in outer suburbs, in inner regional smaller towns, or in remote areas are more likely to have a second birth. However, the effect is small and only significant at the 10% level. When we add education and activity status (model 3), we find that the effect of location becomes stronger.
Although the effect is not as strong as that for first births, women in small towns of inner regional areas, or those in outer regional and very remote areas, are more likely to have a second child.
The variable reflecting years since the first child is born indicates that the risk of having a second child increases sharply in the first 4 years and then declines. In terms of country of birth, we see the same pattern as observed with first births: Overseas-born women are less likely to have a second child and those of Aboriginal or Torres Strait Islander background are more likely to. Cohabiting and single women are less likely to have a second child compared to married women, and age at previous birth is also important. The sex of the first child had no effect on the risk of having a second child. Other city, 20,000+ population 18 20 21 Inner regional Australia 7 8 10 Outer regional Australia or remote/very remote 7 9 9 Country of birth NTV  Once again living in a house with one to two bedrooms is associated with a lower risk of having a second child compared to living in a three-bedroom home. Adding the movement variable makes little difference to the overall model.

| Third births
Turning to the transition from second to third births, in Table 4, we find that women living in small towns in Inner Regional Australia, as well as   in outer regional or remote areas, are significantly more likely to have a third child compared to those in capital cities. There is no difference between capital cities or other cities with a population of over 20,000 residents.
Adding controls for country of birth or ethnicity, relationship status, age at previous birth, and the sex of the first two children however significantly diminishes the effect of location. In the final model, we find that there remains some significant differences by geographic location: The likelihood of transitioning to a third birth is significantly higher for those in inner regional or outer regional Australia compared to inner and middle suburbs of capital cities.
Other important determinants of having a third child include the number of years since the previous child was born. The risk of having a third birth increases sharply in the first 4 years after the second child is born before declining. Women born overseas are less likely to have a third child, as are those who are single. Interestingly, for third births there is no significant difference between cohabiting and married women in the risk of having a third child.
Living in a four-bedroom house is associated with a higher chance of having a third birth compared to living in a three-bedroom house.
In all models, the standard deviation of the individual-level residuals is significantly different from 0, indicating that there are unobserved women-specific characteristics that affect all birth rates.
This indicates that joint modelling was appropriate to control for selectivity among the women in terms of their likelihood of progressing to a higher order child.

| DISCUSSION
The aim of this study was to examine whether place matters when it comes to fertility, even after taking into account compositional differences. We found that after taking into account differences in age, country of birth, indigenous status, relationship status, education levels, and economic activity, women living in smaller towns in Inner Regional Australia are more likely to have a first, second, and third birth. For first births, we find that in capital cities, those living in the inner or middle suburbs, where houses are traditionally smaller and more expensive, are less likely to have a first birth. For the other parities, there is little difference between the inner/middle and outer suburbs or between capital cities and other major cities-the key difference is between cities and regional areas.
Overall, in terms of the strength of the relationship between fertility and location, we find it is weakest for second births. In Australia, there is a strong two-child norm (Gray et al., 2010) One important feature of different areas is the housing stock and the size of the houses. Traditionally, houses in outer suburbs of capital cities tend to be larger than houses in the inner and middle suburbs.
Similarly, houses in regional areas are larger and have larger blocks of land. We did find that the number of bedrooms does explain some of the differences in the link between geography and having a first birth.
For second and third births, number of bedrooms is still an important determinant, but it has less of an impact on attenuating the relationship between childbearing and location.
In HILDA, the reasons why people moved are asked every wave and exploratory analysis of the reasons (not shown) does suggest that people who have moved in the last wave and who have recently experienced childbirth are more likely to state that the reason they moved was to move to a larger house. Although the data are available in HILDA, we have not looked at the sequence of events and whether the move occurs before or after childbirth as previous research has shown there may be strong joint determination and simultaneity between these sort of movements and childbearing.
Depending on people's financial capabilities and the housing market, some people may move to a larger house in anticipation of starting or growing a family, whereas others may wait until after the child is born.
The compositional hypothesis suggests that individuals with similar characteristics will have similar fertility levels regardless of where they live, whereas the contextual hypothesis suggests that individuals orient their fertility behaviour based on regional opportunities and restrictions (Fulda, 2015). In reality, these explanations are not mutually exclusive and both the composition factors and contextual factors, which are harder to measure, are likely to influence fertility.
Although we do find evidence that place is important in influencing fertility behaviour among Australian women after controlling for key demographic and socioeconomic characteristics, we note the following caution. Some variables associated with fertility decision making, such as religiosity, are not included in the models due to lack of data. If these omitted variables are also linked to geographic location then they might explain some of the differences in childbearing by location. Another limitation is that the original sampling frame of HILDA excludes census districts classified as remote or sparsely populated (Watson & Wooden, 2002); these findings underrepresents people living in remote or very remote Australia.
Furthermore, our unit of analysis was at the individual level, and due to data limitations, we were not able to directly include and examine contextual level factors such as the unemployment rates in local government areas or other factors such as the availability of childcare in this paper. As Hank (2002) points out, ideally, regional characteristics and individual childbearing decisions need to be placed into a unifying theoretical framework, while individual-level data need to be merged with aggregate information on the spatial context. We agree that such research would be beneficial.
Whether or not fertility differentials in Australia are due to purely compositional effects or not, we believe that the spatial variation in fertility is still an area that deserves more attention. From a policy perspective, it is important to have a better understanding of subnational fertility trends and behaviours for forecasting and planning purposes. This is especially important in a country such as Australia where service provision outside of major centres is costly and logistically challenging.
De Beer and Deerenberg (2007) note that because subnational fertility trends are important for projection purposes, without having an explanation for the regional differences, it is difficult to decide whether changes observed in the past are likely to continue in the future and, if so, to what extent. Although this analysis could not disentangle whether other migration theories such as socialization, adaptation, and disruption are involved in explaining regional differences, we have shown that composition factors and place are both important for explaining fertility progression.