Self-assessed Physical and Mental Health and Fertility Expectations of Men and Women Across the Life Course Open Access

Abstract

The fertility expectations of older women and men are becoming increasingly important for understanding fertility dynamics, given the increasing share of births after age 30. Because most health conditions deteriorate with age, understanding the relationship between health and fertility expectations is essential. We investigate whether changes in self-assessed general, physical, and mental health are linked to revised fertility expectations and how these associations vary over the life course. Drawing on a large longitudinal dataset for Australia, we demonstrate that across each health indicator, self-assessed poor health corresponds to lower fertility expectations and that a deterioration (or improvement) in self-assessed health coincides with a decrease (or increase) in men's and women's expectations of having a child. Individuals adapt their expectations more in response to physical health changes if they are older, and mental health conditions at younger ages appear relevant to men's fertility intentions. The results highlight that general, physical, and mental health are crucial drivers of changes in fertility plans, emphasizing the importance of integrating health considerations into future theoretical frameworks and empirical analyses of fertility.

Introduction

Studying fertility intentions provides useful insights into the factors and situations that promote or hinder childbearing, enabling researchers to identify the prerequisites for parenthood and the barriers that might impede individuals from starting a family. Indeed, a long tradition of research has explored the determinants of fertility intentions and expectations, documenting the importance of religion (Hayford and Morgan 2008), gender role attitudes (Miettinen et al. 2011), social norms (Regnier-Loilier and Vignoli 2011), and family supportiveness (Schaffnit and Sear 2017) in shaping individuals’ fertility plans. The uncertainty resulting from precarious employment and housing conditions (Begall and Mills 2011; Modena and Sabatini 2012; Vignoli et al. 2013) or major events, such as recessions or pandemics (Lazzari et al. 2024; Matysiak et al. 2021), also influences the decision to have a child.

Another influential source of uncertainty and potential determinant of family size expectations is an individual's perceived health status. Survey data from Europe show that among a sample of childless respondents of reproductive age, 40% of women and 26% of men reported their overall health as an important reason for not having children (Sobotka and Testa 2008). Moreover, as the population desiring to have children ages, instances of poor physical health become more common (Australian Institute of Health and Welfare 2022) and might increasingly prevent the achievement of fertility goals.

Despite numerous cross-sectional studies examining the relationship between health and fertility, the extent to which men's and women's fertility expectations vary in response to changes in health status and over the life course in high-income countries remains poorly understood. In addition, most research has focused on physical health, leaving the relationship between mental health and fertility largely unexplored (Liu et al. 2024). This gap is particularly concerning given the rising prevalence of mental health issues, especially among young people (Dykxhoorn et al. 2024; Krokstad et al. 2022). In this study, we look beyond the relationship between general health and fertility intentions by considering the nature of the health problem. Our research is among the first to analyze the separate impacts of general, physical, and mental health on fertility expectations while also applying a life course perspective to these issues.

To explore these topics, we focus on Australia because of the availability of a unique nationally representative dataset, the Household Income and Labour Dynamics in Australia (HILDA) survey, which provides comprehensive panel information on individuals’ health and fertility preferences. Specifically, we aim to answer three research questions. First, do individuals’ fertility expectations vary by self-assessed health status? Second, do changes in self-assessed health status correspond to changes in fertility expectations? Finally, do these associations vary over the life course? We consider three health indicators, reflecting individuals’ assessments of their general, physical, and mental health. We start with descriptive statistics to investigate the relationship between each dimension of health and fertility expectations. Then, we use fixed-effects (FE) models that control for unobserved heterogeneity to assess whether within-individual changes in health are associated with revisions in fertility expectations. Finally, we estimate interaction terms to investigate how associations between health and fertility expectations vary by age and are shaped by parenthood status and gender.

Literature Review

Prior Research and the Present Study

Extensive research has shown that fertility intentions and expectations are not fixed; they are revised in response to changing circumstances and information (Berrington 2004; Hayford 2009; Jones 2016). These revisions are commonly associated with life course domains, including education, work, living arrangements, and partnerships (Iacovou and Tavares 2011; Liefbroer 2009; Testa and Bolano 2021). However, the potential impact of health on these adjustments has received limited research attention.

Most longitudinal research in this area is concentrated in sub-Saharan Africa (Hayford and Agadjanian 2019; Kodzi et al. 2012), where the health and fertility landscapes differ significantly from those of high-income countries. Notably, health issues, such as HIV and malaria, are widespread and can substantially influence fertility trajectories (Hayford and Agadjanian 2019; Hayford et al. 2013; Kodzi et al. 2012; Yeatman 2009a, 2009b). Findings from this region indicate that women who perceive a worsening of their health are more likely to want to stop childbearing. However, the generalizability of these findings to other contexts might be limited.

By contrast, research from high-income countries has strongly relied on cross-sectional studies, showing that self-assessed general health negatively correlates with the intention to have children (Alderotti and Trappolini 2022; Dommermuth et al. 2011; Holton, Fisher et al. 2011; Holton, Rowe et al. 2011; Mynarska and Wróblewska 2017). Further, a large body of longitudinal studies has compared fertility expectations between women with specific adverse health conditions (e.g., HIV, cancer, rheumatoid arthritis, obesity, and multiple sclerosis) and women in the general population (Chen et al. 2001; Frisco and Weden 2013; Katz 2006; Schmidt et al. 2016; Shandra et al. 2014; Smeltzer 2002). Such specific health conditions are generally associated with lower fertility expectations. Finally, qualitative research indicates that women with severe mental illness often do not intend to become pregnant (Lawley et al. 2022). Adding to this evidence, a quantitative study among migrants found that individuals with mental health disorders are less likely to want a child in the long term (Alderotti and Trappolini 2022). Although some evidence suggests that fertility desires can change in the face of poor health (Gray et al. 2013), this phenomenon has not been studied further, and no research in the high-income context has established a robust relationship between perceived general, mental, or physical health and fertility expectations.

Establishing this relationship is methodologically challenging. Because health has a strong structural underpinning (Pearlin et al. 2005), some correlates of health status might also explain differences in fertility attitudes, which would result in a correlation being observed even though there is no causality or association between health and fertility expectations. Moreover, the way health is reported appears to be particularly sensitive to the understanding of health issues, which is related to possible unobserved differences between individuals (Schneider et al. 2012). Thus, the issue of unobserved heterogeneity is particularly relevant when analyzing individual preferences and self-assessed measures of health. To overcome these limitations, we use individual-level FE models in this study. This approach enables us to control for time-invariant individual-level factors, reducing the impact of unobserved variables. Although it does not establish a causal effect, this method provides more robust evidence of the association between health and fertility expectations than cross-sectional models (Collischon and Eberl 2020).

Pathways Linking Fertility Expectations and Health

Theoretical distinctions exist between fertility desires, intentions, and expectations. Desires refer to unconstrained personal motivations for having children, intentions incorporate situational factors that might prevent individuals from achieving their desires, and expectations combine intentions with external factors beyond an individual's control (Ajzen 1991; Miller and Pasta 1993). Fertility desires appear to be a less proximate component of the motivational stream leading to behavior, whereas intentions and expectations are often used interchangeably in the literature (Iacovou and Tavares 2011; Miller 2011). Given our interest in how health, which is likely perceived as an external factor, affects considerations of having children, this study focuses on expectations. However, because of the similarities between the two constructs, we also draw on the literature on fertility intentions.

Perceived behavioral control is an important component of fertility intentions and expectations (Dommermuth et al. 2011). Hence, an episode of poor health, which is out of a person's control, might lead individuals to disengage with their original fertility plans. In this context, the perception of a health problem might matter more for fertility preferences than the objective experience of health deterioration because this perception affects individuals’ perceived control over their fertility (Greil et al. 2011). Studies on economic situations have revealed this sort of pattern in showing that individuals’ perceptions of their economic situations influence their fertility choices independently of objective indicators (Bolano and Vignoli 2021; Fahlén and Oláh 2018; Vignoli et al. 2020).

Based on the available evidence and literature, the connection between health status and fertility expectations can be established through both direct and indirect pathways, as well as through increased uncertainty (Alderotti and Trappolini 2022; Mynarska and Wróblewska 2017). We particularly focus on how an improvement or deterioration in health status might be related to a change in fertility expectations—an important aspect of our research question that has not been conceptualized before.

First, health problems can directly reduce the chances of conceiving and carrying a pregnancy to term, potentially leading individuals to disengage from childbearing plans (Ezeh et al. 2016; Shreffler et al. 2016). Physical disorders, such as obesity or endocrine dysregulation, appear to lead to subfecundity in both men and women (Amiri and Tehrani 2020; Deyhoul et al. 2017). In women, cancer treatment, severe endometriosis, infection or sexually transmitted disease, miscarriage, or complications in child delivery might result in infertility onset. Other direct effects are the physical limitations associated with illness, which might lead individuals to revise their childbearing expectations (Alderotti and Trappolini 2022; Syse et al. 2020). In addition, individuals with mental disorders face a complex set of trade-offs regarding parenthood that they rarely manage to resolve (Krumm and Becker 2006). Thus, the onset of a mental disease might lead individuals to withdraw from childbearing. Research has found that psychotic disorders are the type of disease most associated with lifetime childlessness and smaller family sizes for both men and women (Bundy et al. 2011; Liu et al. 2024). On the other hand, for people with long-term mental illness, a recovery might lead to a positive attitude toward having children.

Second, a change in health status can also influence fertility expectations indirectly through its impact on other aspects of life. For example, mental and physical illnesses can be associated with a decrease in sexual desire (Lourenço et al. 2011; Nusbaum et al. 2003), which might then influence individuals’ fertility decisions. In addition, certain medical treatments require patients to take medications that are incompatible with pregnancy, and the patient might then have to pause treatment to have a child, further complicating the decision process. However, the greatest indirect constraint might arise when an individual's health status affects their labor market participation and earnings. For those in poor health, having a child might be relatively more costly because of the increased likelihood of financial strain (Bartel and Taubman 1986; Lyons and Yilmazer 2005) and might be associated with a reduced ability to work and an increased need to outsource care owing to illness. In contrast, the available evidence does not support the idea that people in poor health are more likely to be out of a relationship, which could affect their fertility expectations. In Germany, self-assessed health does not appear to influence partnership formation at reproductive ages (Rapp 2018). Furthermore, illness and separation are not clearly linked: a cancer diagnosis does not lead to an increased risk of divorce among Norwegian married couples (Syse and Kravdal 2007).

Finally, an individual's health status might be related to their certainty about the future (Trinitapoli and Yeatman 2018; Vignoli et al. 2020). In this case, it might be the perception of how the health condition will affect future reproduction and child-rearing rather than the deterioration of the health condition itself that contributes to reduced fertility expectations. The degree of uncertainty experienced is likely to depend on the nature of the health condition. For example, having a severe mental or physical disability might make an individual question their child-rearing ability (Krumm and Becker 2006; Nattabi et al. 2009). Concerns might extend to worries about being physically unable to care for a child because of illness or premature death. Additionally, individuals might be apprehensive about transmitting adverse health conditions to their offspring. For instance, cancer survivors, women with multiple sclerosis, and HIV-infected men and women express concerns about potentially passing on the same disease to their children (Nattabi et al. 2009; Prutny et al. 2008; Schmidt et al. 2016; Yeatman 2009b). The uncertainty around childbearing prospects is likely amplified by the limited access to family planning information and counseling for patients with chronic conditions (Lawley et al. 2022; Phillips et al. 2018). Infertility occupies a unique position in the sphere of uncertainty: the reproductive outcomes for those affected by this health condition are inherently uncertain. Research suggests that women who experience infertility are less likely to expect to have a child despite holding stronger desires for children (Shreffler et al. 2016). The uncertainty associated with infertility might be further intensified by the lack of affordable infertility treatments.

We suggest that an episode of poor health might reduce behavioral control through one or more of these pathways, ultimately reducing fertility expectations. Consequently, we propose that an individual's self-assessed health is positively associated with their fertility expectations (Hypothesis 1) and that they might adjust these expectations in response to a revision of their self-assessed health status (Hypothesis 2). This dynamic perspective suggests that fertility expectations can fluctuate depending on changes in an individual's health. In addition to general health, we investigate the influence of physical and mental health on fertility expectations. Although these two health dimensions might be related (Prince et al. 2007), they remain distinct experiences that could affect perceived control and expectations in different ways (Alderotti and Trappolini 2022).

Differences by Age, Parenthood Status, and Gender

The relationship between health and fertility expectations might vary by age. Fertility intentions expressed during different stages of reproductive life are based on different factors: younger individuals are more likely to have idealized images of a family and be influenced by background factors, and older people are more often influenced by situational factors (Bachrach and Morgan 2013; Berrington et al. 2015; Hayford 2009; Heiland et al. 2008; Neels et al. 2013; Ní Bhrolcháin and Beaujouan 2019; Rackin and Bachrach 2016). Women and men who encounter a health problem during their later reproductive years might more quickly adjust their childbearing plans, leading to a stronger link between health and fertility expectations with age (Hypothesis 3).

The number of children already born might be another relevant factor. Prior research has found differences in how individuals with versus without children perceive parenthood (Purewal and van Den Akker 2007), as well as variations in the factors influencing their decisions to have (additional) children (Dommermuth et al. 2011; Miller and Pasta 1993). For instance, the association between disability and fertility intentions among women differs depending on their parity, with no association observed among childless women and disabled mothers less often intending to have a child (Shandra et al. 2014). In addition, the age of the youngest child is relevant for fertility expectations. In particular, individuals are less likely to revise their childbearing plans once their youngest child is age 4 or older (Iacovou and Tavares 2011). By this time, most decisions regarding childbearing have already been made, rendering them less sensitive to changes in their situational constraints (Iacovou and Tavares 2011). On the basis of these empirical findings, we expect that the association between health and fertility expectations will vary by parenthood status (Hypothesis 4).

Finally, the literature on the relationship between health and fertility expectations has predominantly focused on women, with only a few exceptions (Alderotti and Trappolini 2022; Yeatman 2009a, 2009b). Studies of men's fertility, including their fertility expectations, remain sparse because of data limitations and the prevailing societal belief that men's role in reproduction is secondary (Almeling 2015). However, health might influence men's fertility expectations, especially because they tend to become parents later in life. Given the greater physical impact of pregnancy on women, the relationship between health and fertility expectations might be more pronounced for women than for men (Hypothesis 5).

Australia in Comparative Perspective

Fertility preferences and outcomes in contemporary Australia are generally similar to patterns observed in other high-income countries. These trends are characterized by the tendency to start childbearing later, a growing proportion of individuals expressing positive fertility desires after age 30 (Lazzari et al. 2023), a continued acceptance of the two-child family norm (Lazzari 2021a), and a growing proportion of individuals remaining childless or having fewer children overall than the previous generation (Gray and Lazzari 2023). Comparable patterns have been identified in Europe and the United States (Beaujouan 2023; Guzzo and Hayford 2023; Sobotka and Beaujouan 2014; Zeman et al. 2017).

Perceptions of deteriorating health might be related to experiences of infertility, which has increased in prevalence in recent decades as a result of delayed childbearing. Infertility's impact on reproductive expectations might be less profound if addressing these challenges is perceived as relatively feasible—for example, because the availability of assisted reproductive technology (ART) treatments is high. Australia stands out as a country where ART is affordable and utilization rates are high relative to other countries (Chambers et al. 2021). This availability, coupled with the ongoing trend toward childbearing delay, has substantially increased the contribution of ART to fertility rates over the past decade, especially among older age groups (Lazzari et al. 2021). Thus, the findings from this study about the association between health and fertility expectations might be more easily generalizable to countries where ART utilization rates are high.

More broadly, a country's welfare context and the level of trust in its institutions can influence how individuals respond to uncertainty (Fahlén and Oláh 2018; Kreyenfeld et al. 2012). In relation to health, Australia's universal health care system (Medicare) is relatively comprehensive in its coverage. By contrast, health systems in other Anglo-Saxon countries often provide lower levels of social assistance, and welfare systems in Southern and Eastern European countries are generally less supportive (Esping-Andersen 1990; Korpi 2000).

Research Design

Data and Sample

The data for this analysis were obtained from the Household Income and Labour Dynamics in Australia survey, a nationally representative panel study of Australian adults aged 15 or older. The HILDA survey has interviewed approximately 15,000 individuals annually, achieving consistently high response rates and a retention rate of 90% (Summerfield et al. 2021). In the initial wave in 2001, a sample of Australian households was interviewed, forming the basis of the panel. Subsequently, the sample has expanded to accommodate new household members as a result of changes in household composition. To maintain sample representativeness, a top-up sample of roughly 2,000 households was added in 2011 (Watson 2011). The data are collected annually through face-to-face interviews and self-completion questionnaires. Along with collecting information on a wide range of social, demographic, and health characteristics, the survey includes topics related to family formation and fertility. Hence, it represents an ideal data source to investigate the association between self-assessed health and fertility expectations in a nationally representative sample.

We constructed an unbalanced panel of women (aged 18–44) and men (aged 18–54) who were interviewed at least twice during Waves 1–21 conducted between 2001 and 2021. Using an unbalanced panel aligns with the dynamic nature of the HILDA dataset, which has expanded over time to include new respondents. Restricting the analysis to respondents from the first wave would yield an increasingly less-representative sample and result in a substantial loss of observations. After we exclude 5.6% of observations with missing values for one or more dependent variables, our main analytic sample includes 8,080 women and 9,065 men, who were observed for a total of 51,896 and 61,774 person-years over an average of 8.7 and 9.5 waves, respectively. Item nonresponse was more likely among respondents who were male, younger, low educated, single, and in worse health conditions. As a result, our findings might underestimate the impact of health on fertility expectations. To explore gender dynamics further, we investigate the relationship between health and fertility expectations for a subset of 5,215 couples in stable relationships (married or cohabiting), contributing to a total of 26,827 person-years.

Measures

Our outcome variable is fertility expectations, which were assessed through a single survey question: “How likely are you to have [a child/more children] in the future?” Responses were measured on an 11-point Likert scale ranging from 0 for “Definitely not likely to have (more) children” to 10 for “Definitely likely to have (more) children.” Participants were informed that higher numbers indicated a greater likelihood of future parenthood, whereas lower numbers reflected a lower likelihood. Because the criteria for asking this question differed in Waves 5, 8, 11, 15, and 19, we exclude observations from these waves to ensure data comparability.¹ The fertility expectation variable is treated as continuous in our analysis.²Table S1 (shown in the online appendix, along with all other tables and figures designated with an “S”) provides its statistical characteristics, including measures of central tendency, variability, and distribution; Figure S1 shows mean values of fertility expectations by age group and parity status.

Information on the health of the respondents is derived from the 36-Item Short-Form General Health questionnaire (SF-36), a widely recognized tool for assessing health-related quality of life (Ware and Kosinski 2003), which is administered annually in the HILDA survey. HILDA's SF-36 data demonstrate excellent measurement qualities, with good internal consistency and high reliability (Butterworth and Crosier 2004). As our primary independent variable, we use the first item of the SF-36, which is a single survey question measuring respondents’ general health: “In general, would you say your health is: ‘Excellent,’ ‘Very good,’ ‘Good,’ ‘Fair,’ or ‘Poor’?”

We repeat all analyses using two SF-36 subscales that assess respondents’ overall physical and mental health. The physical subscale evaluates perceptions of physical health via questions concerning susceptibility to illness, health relative to others, and expectations about future health. The mental health subscale is based on five questions evaluating the extent to which respondents feel nervous, “down in the dumps,” peaceful, sad, and calm. The raw scores derived from the survey items corresponding to each of these scales are summed and subsequently transformed into a standardized 0–100 scale. These subscale scores are readily accessible in the HILDA dataset, computed by following the SF-36 manual (Ware and Kosinski 2003). For ease of interpretation, we present our results by grouping these scores into quintiles, with higher quintiles representing better health. Parallel analyses using the corresponding continuous variable (not shown) yielded results consistent with those presented.

Correlation analyses reveal a strong association between measures of general health and physical health, with a correlation coefficient of .72. In contrast, the correlation between general health and mental health is comparatively lower, at .39. Thus, our measure of general health is more likely to capture individuals’ assessments of their physical health than their mental well-being.

To account for potential confounding factors, we include a set of control variables in our models. Relationship status is operationalized as a time-varying variable (i.e., varying across survey waves) consisting of three categories: married, cohabiting, and single. To control for labor market conditions, we consider variations in employment status, which is operationalized as a time-varying variable with three categories: employed, unemployed, and not in the labor force.³ In addition, we include a variable capturing personal satisfaction with finances via a question asking respondents to rate their wealth given their current needs and financial responsibilities. Operationalized as a time-varying variable, it consists of three categories: very satisfied, reasonably satisfied, and dissatisfied. For parenthood status, we use a variable combining information on parity (childless vs. parent) and the age of the youngest child (younger than 4 years vs. 4 years or older). A respondent is categorized as a parent if they have biological or adopted children. We also consider three time-invariant variables based on information collected when the respondent first entered the study: education, whether the respondent had any siblings in the family of origin, and the respondent's migration and ethnic status. The models also control for current age and age squared to capture potential age-specific effects. Table S2 shows the sample characteristics at the first wave and across all waves.

For the couple-level analysis, we construct a combined variable for health categorizing couples by whether both partners reported good health (measured as excellent, very good, and good) or poor health (measured as fair or poor) and separately consider cases of within-couple disparities in self-assessed health status. We use this categorization because of our interest in exploring the association between potential differences in partners’ health statuses and revised fertility expectations. Models using couples as the unit of analysis include education, satisfaction with finances, and migrant and ethnic origin of both partners, along with a combined variable for employment status that indicates whether both partners or only one partner was employed. Table S3 provides a summary of the characteristics of the couple sample.

Analytic Strategy

We start by examining how mean fertility expectations differ according to the explanatory variables listed above. Then, we compare mean expectations by age across categories of self-assessed health to examine bivariate associations between health and fertility expectations at different life course stages.

For the longitudinal analysis, we use linear FE models to estimate changes in fertility expectations associated with changes in self-assessed health.⁴ The main benefit of using FE models is that all time-invariant characteristics associated with an individual (i.e., background characteristics, personality traits, and genetic endowments) are controlled for, which can help address issues of unobserved heterogeneity and provide more robust estimates than with classical pooled ordinary least-squares (POLS) models (Allison 2009; Collischon and Eberl 2020). Results based on the Hausman specification test support the hypothesis that individual-specific effects are correlated with the regressors (not shown). Our FE model can be written as follows:

where $Ei,t$ represents fertility expectations for individual i at time t; $HEALTHi,t$ is a vector of time-varying health variables (self-assessed general, physical, or mental health); age is controlled for by $age$ and $age2$ (to account for the nonlinearity in the relationship between age and fertility expectations); $T$ captures year-specific FEs and accounts for trends and shocks, such as economic crises or the COVID-19 pandemic; $Xi,t$ measures other time-varying variables that might influence fertility expectations (in particular, partnership and employment status); and $Zi$ captures time-constant characteristics, such as ethnic origin, that do not change during the panel study. The term $εi,t$ captures unmeasured individual, time-specific heterogeneity; $ui$ is the idiosyncratic, time-constant error term. Terms $β1$⁠, $β2$⁠, $β3$⁠, $β4$⁠, and $γ$ are coefficients, and $β0$ is the constant. Given previous research showing that gender significantly shapes childbearing preferences and behaviors across the life course (Lazzari 2021b; Spéder and Kapitány 2009), we conduct separate analyses for men and women.

An important feature of our modeling approach is our exclusive focus on within-person changes. Standard cross-sectional designs, such as those available in the literature, are potentially biased by unobserved individual-level confounders, which might be associated with both self-assessed health and fertility expectations. Health perceptions are deeply rooted in various unobserved factors, such as different ways of seeing the world, personality traits, and genetic endowments (Jurges 2007; Schneider et al. 2012). Moreover, these traits might also incorporate attitudes toward fertility behaviors. FE models are more robust than other panel modeling approaches because they control for these fixed, unobserved characteristics (Collischon and Eberl 2020). However, they do not allow for causal inference because some individual time-varying characteristics might remain uncontrolled for and affect both fertility expectations and health perceptions.

FE models operate under the assumption that the effect of explanatory variables on the outcome is symmetrical. This assumption implies that a decline in health will correspond to changes in fertility expectations that can be reversed by an improvement in health of equal magnitude. To assess the plausibility of this assumption, we conducted a test using an asymmetric FE model for panel data, following Allison's (2019) methodology. Although the coefficients estimated by the asymmetric model suggest that the effect of a health decline on fertility expectations is somewhat stronger than the effect of a health improvement, this difference is not statistically significant.

While our primary emphasis is on the results from the FE models, we also present results from POLS models, which help us evaluate the characteristics and circumstances linked to fertility expectations between individuals. Unlike FE models, POLS models allow for the meaningful inclusion of time-invariant variables, such as educational attainment, number of siblings, and migrant and ethnic origin. These results can be used as a benchmark for comparison against conventional methods employed in the literature for investigating the relationship between health and fertility-related outcomes.

We further examine how the association between health and fertility expectations varies with age using interaction analyses. Interaction terms between two time-varying covariates in FE models do not yield a within-unit estimator, and the coefficients lack a straightforward interpretation (Giesselmann and Schmidt-Catran 2020; Shaver 2019). Thus, to enhance the clarity of the interaction terms, in the interaction analyses, we treat the age variable as a fixed categorical variable with two categories (age 30 or younger and older than 30). Although this strategy does not yield a strict within-unit estimator, it provides a technically consistent interpretation by measuring how the within-person effect of a change in health on fertility expectations varies between respondents of different age groups. Interaction effects are also estimated to explore how the relationship between health and fertility expectations varies by parenthood status and gender.

All models are replicated separately for women and men. To further explore gender dynamics, analyses are repeated for a subsample of couples. In these models, the explanatory variables consider both partners’ characteristics and account for the similarity and dissimilarity of their self-assessed health.

Results

Descriptive Findings

As shown in Table 1, the mean scores for fertility expectations were 4.5 for women and 3.7 for men (Table S1 displays the distribution of this variable). On average, women and men who were in better health, younger, cohabiting, childless or parents of a young child, in the labor force, and more satisfied with finances displayed higher fertility expectations. Turning to the background factors, women and men who had a Year 12 education and were without siblings also had higher fertility expectations. Additionally, men who were of Australian and not Aboriginal or Torres Strait Islander (ATSI) origin had lower fertility expectations, whereas women in this group had higher fertility expectations. As illustrated in Figure 1, the association between each dimension of health and fertility expectations was generally strongly positive for both men and women and across age groups.

Main Findings

General, Physical, and Mental Health

Table 2 reports the panel estimates from POLS and FE models of the determinants of fertility expectations among women and men using general (panel A), physical (panel B), or mental (panel C) health as the main explanatory variable. POLS models help identify the extent to which self-assessed health is associated with different levels of fertility expectations across individuals (columns 1–2 and 5–6), whereas FE models indicate how individual-level changes in self-assessed health are associated with individual-level changes in fertility expectations (columns 3–4 and 7–8). The baseline models report results obtained by fitting a model with only health as the independent variable (columns 1, 3, 5, and 7). Models with multiple covariates demonstrate the effect of health on fertility expectations, net of confounders (columns 2, 4, 6, and 8). Full models are available in Tables S4–S6. In all models, health changes are described as a deterioration relative to the reference category of excellent health status.

The results of the POLS model for general health in Table 2 indicate that, on average, women in poor health had a fertility expectation score 0.78 lower than their peers in excellent health, after we control for socioeconomic and background factors. We find a similar general health gradient among men, with those in poor health having a fertility expectation score 0.35 lower than those in excellent health, after we control for all other covariates.

When FE models are used to control for unobserved characteristics, the link between a change in general health and fertility expectations is smaller than in the POLS models but is still statistically significant at the .001 level and robust to the inclusion of time-varying socioeconomic confounders (Table 2). Among women, a deterioration in general health from excellent to good is associated with a decline in the fertility expectations score of 0.37 in the multiple covariates model. A larger deterioration in general health, from excellent to fair or poor, is associated with a drop in fertility expectation of 0.61 and 0.62, respectively. Among men, a change in general health conditions also appears to be linked with a revision of fertility expectations in the same direction. For instance, transitioning from excellent to good general health is associated with a decline in the fertility expectations score of 0.38 in the multiple covariates model. A larger deterioration in general health to fair or poor is linked with a reduction in fertility expectations of 0.52 and 0.23, respectively. The pattern of results does not substantially vary when we use physical and mental health as the main explanatory variables (panels B and C, respectively). Although the association between mental health and fertility expectations appears weak among women before controls, physical and mental health are consistently associated with a downward revision of fertility expectations among both men and women after the inclusion of age and socioeconomic confounders.

As noted in the Analytic Strategy section, the effect of health on expectations can be assumed to be symmetric. Hence, our results indicate an association between health improvements and higher fertility expectations and between health deterioration and lower fertility expectations.

In a model incorporating both general and physical health as covariates (not displayed), the estimates for general health exhibit substantial reduction, underscoring the high correlation between physical and general health. However, the inclusion of mental health as a covariate only marginally diminishes the estimates for general health, aligning with the limited correlation observed between the metrics of mental and general health. This finding also suggests that the effect of general health is mostly mediated by physical rather than mental health conditions.

Differences by Age, Parenthood Status, and Gender

We examine the association between each dimension of health and fertility expectations in separate models for different age subgroups (not shown). The results consistently demonstrate a strong association between health and fertility expectations, highlighting the importance of health as a factor regardless of age. We test whether this relationship intensified over the life course using interactions. Figure 2 displays the predicted values of the fertility expectations score, taking into account age and one of the three dimensions of self-assessed health while holding socioeconomic confounders constant at their mean (the original coefficients of the analytic models are available in Tables S7–S9).

The graphs in the top panels illustrate that for women and men older than 30, a deterioration in general health is associated with a relatively larger downward revision of fertility expectations relative to their counterparts aged 30 or younger. For instance, among women, the drop from excellent to good health is associated with a significantly larger decline in fertility expectations, and a decline in health to fair or poor additionally reduces fertility expectations compared with women experiencing a similar change at younger ages. Among men, a decline in general health from excellent to very good or good is also particularly linked with a downward revision of fertility expectations at higher ages, but differences by age are less significant when health deteriorates to fair or poor. Regarding physical health, changes at age 30 or younger are associated with relatively stable fertility expectations compared with shifts occurring later in life (Figure 2, middle panels). Differences by age in the physical health gradients are also apparent for men, with older respondents more likely to revise their fertility expectations downward in the face of poor health than their younger counterparts. By contrast, in relation to mental health, women show no statistically significant differences, and men display a stronger association at younger ages (Figure 2, bottom panels).

Also interesting are differences by parenthood status. Women and men who were childless or whose youngest child was younger than 4 years old were more likely to revise their fertility expectations in response to a change in their health status than those whose youngest child was at least age 4 (results are available in Tables S10–S12). Models using physical and mental health as main explanatory variables show a similar pattern. Modeling women and men together, we also find a significant interaction effect of general and physical health with gender (results not shown): women were more likely than men to revise their fertility expectations in response to a decline in general or physical health. By contrast, no overall gender differences are observed in relation to mental health.

Couple-Level Analyses

Because previous research has shown that partners have a different influence on fertility decisions and might differ in their subjective perceptions of health, we explore the association between health and fertility expectations among a subsample of individuals in coresident relationships (married or cohabiting). The results of these analyses, reported in Table 3, indicate that the health status of both partners significantly predicts fertility expectations (full results of this analysis are provided in Tables S13–S15). In particular, panel estimates from POLS models indicate that couples in which both partners were in poor health are associated with lower fertility expectations than couples in which both partners were in good health, regardless of the dimension of health. The FE models also indicate that a deterioration in the health status of both partners is linked to a downward revision of fertility expectations. However, interesting gender-based differences emerge for partners who reported different health statuses. A more significant reduction in fertility expectations occurred when the female partner's health status declined than when the male partner's health status declined. These results are consistent across all three dimensions of health after controls are included in the models, with the exception of physical health for men.

Discussion

Our study contributes to understanding the relationship between self-assessed health and fertility expectations. We considered three dimensions of self-assessed health: general, physical, and mental. We combined a traditional cross-sectional approach with a longitudinal approach that allowed us to account for possible unobserved factors underlying changes in both health perceptions and fertility expectations.

In support of Hypothesis 1, we found a strong association between general, physical, and mental health and fertility expectations. In addition, in line with Hypothesis 2, the results from the FE models provided new robust evidence supporting the argument that a change in each health dimension is an important driver of the revision of fertility plans. These results contribute to the literature in several ways. First, we use the best model available to establish such a relationship in that it controls for unobserved heterogeneity, which is particularly likely to be an important issue in the case of self-reported health status. Our findings are consistent with previous longitudinal analyses indicating that the experience of a specific adverse health condition, such as obesity or HIV, might lead to a downward revision of the fertility plans later in life (Frisco and Weden 2013; Yeatman 2009a). Second, previous research explored only the relationship between general health or specific health conditions and fertility plans, providing very little evidence on the influence of mental health (Alderotti and Trappolini 2022; Lawley et al. 2022). Our results confirm that like physical dimensions of health, mental health is strongly related to fertility expectations.

Results from the interaction analyses underscore the greater importance of self-assessed general and physical health among men and women of advanced reproductive ages in influencing fertility expectations, supporting Hypothesis 3. This observation suggests that deteriorating physical health might be one of the drivers of the progression from positive to negative fertility plans among women in their 30s and 40s observed in previous research (Rybińska and Morgan 2019). However, we observed that general health emerged as a nonnegligible predictor of fertility expectations also among relatively young prospective parents and that among men, shifts in mental health were more related to changes in fertility expectations at young ages. These findings suggest that health conditions might also constrain childbearing plans at earlier life stages.

We also found differences by parental status (Hypothesis 4). Fertility expectations were less stable in the face of a health change among those who were childless or parents of a child younger than 4 than among parents of older children. A plausible explanation is that many people whose youngest child is aged 4 or older have reached their desired family size, and their already low fertility expectations are unlikely to be affected by external factors. In contrast, changes in health might have a greater impact on the fertility expectations of people who have not yet transitioned to parenthood or who have a young child, given that they are still in the process of family formation (Iacovou and Tavares 2011).

In partial support of Hypothesis 5, we found that the association between general and physical health and fertility expectations was more pronounced for women than for men. This pattern might be explained by the greater involvement of women's bodies in reproduction, potentially rendering them more sensitive to the impact of alterations in their physical health on childbearing outcomes. However, no significant gender differences were observed for mental health. The couple-level analysis further supports the notion that women's health status is a more crucial determinant of fertility expectations than their partners’ health status. Indeed, revisions in fertility expectations were more sensitive to changes in the female partner’s health than to changes in the male partner’s health. This difference might be attributed to the prevailing perception that reproduction is primarily a woman's domain (Almeling 2015) and seems to indicate the centrality of a woman's health versus a man's health in the decision to have a child.

This study has some limitations. First, although the use of a nationally representative sample strengthens the validity of our findings, caution should be exercised when generalizing the results to settings with social welfare systems that differ from those in Australia. The impact of poor health episodes on fertility expectations might vary depending on the specific social welfare system, which could act as a moderating factor. Some data-related considerations are also worth mentioning. Unfortunately, we could not account for whether women were pregnant (or whether men had a pregnant partner) at the time of the survey. Individuals in such circumstances are likely to display more stable fertility expectations in response to a deterioration in health than those who are not pregnant (or whose partner is not pregnant) because the outcome of having a child is more certain for them. However, this limitation affects only a small proportion of our sample and would, if anything, lead to an underestimation of the association between health and fertility expectations.

Furthermore, caution is warranted when comparing different metrics of self-assessed health. The general health measure is not directly comparable to the physical and mental health measures because of its construction. This disparity in measurement also complicates the explanation for why certain associations identified in relation to general health are not of the same magnitude in analyses using the other two health measures. In addition, the nature of the health shocks experienced deserves further investigation. First, isolating the changes in health that are directly related to infertility would be an important step in this direction. Second, possible heterogeneity in shock intensity and duration could be better assessed by examining the lagged effects of health on current fertility plans. This analysis focuses on the effects of contemporaneous health changes on current fertility expectations, but it is also possible that health status at earlier ages has a lingering effect on current expectations, consistent with a life course perspective (Barclay and Kolk 2020). Finally, the HILDA survey does not collect information on objective measures of health. As a first step, our analysis demonstrates that individuals are likely to adjust their fertility expectations to changes in self-assessed health. Another important step will be to explore whether these self-assessments reflect objective changes in health conditions and to determine if objective health measures have an independent effect on fertility expectations.

Conclusion

Our perspective draws attention to the impact of changing health conditions on the fertility plans of men and women at different stages of their reproductive life. The finding that both physical and mental health are strongly associated with fertility expectations highlights two important aspects of the link between health and fertility among current and future generations. First, in high-income countries, childbirth is increasingly postponed to later ages, and instances of poor physical health tend to increase with age. A growing proportion of individuals may therefore be abandoning their childbearing plans because of health problems. Second, given the escalating prevalence of mental health issues among young people (Dykxhoorn et al. 2024; Krokstad et al. 2022), future generations will likely face mental health problems as an increasingly significant barrier to starting a family. Although most mental health issues emerge during youth, these stressors often persist into adulthood (George 2013; Jones 2013) and can hence affect fertility trajectories later in life.

Acknowledgments

The authors thank Sonja Spitzer and participants of the 2023 annual meeting of the Population Association of America for their valuable comments and suggestions. This research was supported by funding from the European Research Council under the European Union's Horizon 2020 research and innovation program (grant agreement 101001410).

Notes

References