The impact of lockdowns during the COVID-19 pandemic on fertility intentions

Abstract

Lockdown edicts during the COVID-19 pandemic have led to concerns about consequences for childbirth plans and decisions. Robust empirical research to either refute or confirm these concerns, however, is lacking. To evaluate the causal impact of lockdowns on fertility, we exploited a large sample of Australians (aged 18–45) from a nationally representative household panel survey and leveraged variation from a unique natural experiment that occurred in Australia in 2020: a lockdown imposed in the state of Victoria, but not elsewhere in Australia. Difference-in-differences models were estimated comparing changes in fertility intentions of persons who resided in Victoria during lockdown, or within four weeks of the lockdown being lifted, and those living elsewhere in Australia. Results revealed a significantly larger decline in reported intentions of having another child among women who lived through the protracted lockdown. The average effect was small, with fertility intentions estimated to fall by between 2.8% and 4.3% of the pre-pandemic mean. This negative effect was, however, more pronounced among those aged over 35 years, the less educated, and those employed on fixed-term contracts. Impacts on men’s fertility intentions were generally negligible, but with a notable exception being Indigenous Australians.

1. Introduction

The COVID-19 pandemic has affected nearly every aspect of human life. There are the obvious health impacts, reflected in more than 400 million persons infected and around 6 million deaths world-wide over the first two years following the initial outbreak. However, the impact of the pandemic extends well beyond mortality and morbidity, with a large body of research documenting impacts on, among other things, mental health (Robinson et al., 2022), physical exercise (Stockwell et al., 2021), body weight (Bakaloudi et al., 2021), educational achievement (König and Frey, 2022), employment and job security (Adams-Prassl et al., 2020, Forsythe et al., 2020; OECD, 2020), income and poverty (Cantó et al., 2022, Crossley et al., 2021), loneliness (Bu et al., 2020, Dahlberg, 2021), and domestic violence (Piquero et al., 2021). Another outcome that may have been affected by the pandemic is fertility (Aassve et al., 2020, Ullah et al., 2020), with findings from survey-based research reporting evidence that many fertile people delayed or ceased trying to have children because of the pandemic (Emery and Koops, 2022, Kahn et al., 2021, Lin et al., 2021, Lindberg et al., 2020, Lindberg et al., 2021, Luppi et al., 2020, Micelli et al., 2020, Naya et al., 2021, Zhu et al., 2020). This is potentially of large significance given predictions that health pandemics will become more frequent in the future (Marani et al., 2021) and given the ongoing decline in fertility rates, which some are forecasting could ultimately mean global net population growth reaching zero by the end of the century (Aitken, 2022).

The existing evidence on the relationship between the COVID-19 and fertility or fertility plans is, however, far from convincing. Most of the studies cited above involved small convenience samples, and all employed cross-sectional designs where respondents were asked how fertility plans had changed because of the pandemic. Responses may thus be affected both by recall bias and priming. Very differently, while there is evidence of an acceleration in the long-term decline in fertility rates in high-income countries beginning late 2020 (Aassve et al., 2021), this pattern did not apply to all countries and the decline was only statistically significant in a minority.

A complication in existing research is that much of the documented impacts may not be the result of the virus per se, but of measures introduced by governments to contain the spread of the virus and to limit its effect on population health. Among the suite of policies implemented, most contentious has been the use of lockdowns – compulsory business closures, stay-at-home orders and other interventions designed to restrict population movements (Haug et al., 2020). Despite this, their use has been widespread: According to one source, stay-at-home requirements had been adopted in just over 80% of countries at some point during 2020, and workplace closure policies in around 95% (Hale et al., 2021).

While the channels through which the pandemic might influence fertility are numerous and varied (Voicu and Bădoi, 2021), the role of social distancing, and especially lockdown-type measures, looms large. Lockdowns placed constraints on people’s social and intimate lives, with marriages postponed and couples not living together having fewer opportunities to meet and plan for the future. Likely even more important is how couples coped with and responded to the many stressors associated with lockdowns, and the pandemic more generally (Pietromonaco and Overall, 2021). Increased economic uncertainty (especially when accompanied by job loss), school closures and reduced access to child care (both formal and informal), for example, are all factors that would have placed families under greater strain and potentially adversely affected couple functioning. For some families, and especially those with young children, lockdowns increased the amount of household labor and care work, with typically more of the burden falling on the women in those families (Carlson et al., 2022, Craig and Churchill, 2021, Sevilla and Smith, 2020). Working in the other direction, lockdowns, and more specifically stay-at-home orders, by leading couples to spend more time together, could have strengthened relationships (Neff et al., 2021).

Despite these arguments, there is little empirical evidence disentangling changes in fertility intentions due to the imposition of lockdowns from other channels associated with the pandemic. Indeed, the closest we could identify was a study where the outcome was not a measure of births or fertility intentions, but of Google searches for words related to fertility (and other demographic variables) (Berger et al., 2021). This study, however, is notable for its quasi-experimental design, which, by exploiting cross-country and (in the case of the USA) cross-state differences in the timing of lockdowns, was able to derive estimates of the impact of lockdown that could be defended as causal.

Our study also involves a quasi-experimental design, but where the outcome is a more direct indicator of fertility plans – the expected likelihood of having another child. We exploit household panel survey data collected in Australia in the second half of 2020 when residents of one state of Australia – Victoria – were subject to a highly restrictive lockdown involving curfews and compulsory stay-at-home orders for all but essential workers. We estimate difference-in-differences (DiD) models to test whether and how much the lockdown changed fertility intentions. A feature of our experimental setting is that there was an extremely low incidence of COVID-19 infection in Australia during 2020. The nexus between the introduction of lockdowns and very high community rates of COVID-19 cases was therefore weaker in Australia than in most other countries, and thus we can be more confident that our estimates reflect the impact of lockdowns rather than some other channel, such as the health effects of the virus.

We also test whether the impact of lockdown on fertility intentions varies with selected individual characteristics. As already discussed, it is generally believed that women, and especially mothers, bore much of the increased care burden associated with lockdown and thus impacts on fertility plans might differ between men and women and, among women, between mothers and the childless. Slightly differently, we also examine whether impacts vary with household income, education, employment status, age, race, and immigrant status, given the economic hardship associated with lockdowns are often thought to have had greater impacts on the most economically vulnerable (e.g., Crossley et al., 2021; Perry et al., 2021).

This study contributes to the emerging pandemic literature in two major ways. First, it provides new empirical evidence about whether and how much policy responses to the pandemic (as distinct from fears about contracting the virus) dampened intentions to have children. Second, by examining the differential fertility impact of lockdowns on different socioeconomic and demographic groups, our study uncovers whether lockdowns have generated a change in ‘who’ may have children in the future, and in particular whether such a shift is more pronounced among disadvantaged populations.

2. The quasi-experimental setting

In Australia, the pandemic was met with international border closures (with arrivals largely restricted to returning Australians who were required to enter an official quarantine facility or hotel upon arrival), the imposition of restrictions on movement across state borders by some state governments, and a nationwide partial lockdown commencing late March 2020, which, among other things, involved the closure of many non-essential businesses and most schools (which moved to online learning) and advice to work from home where possible (Stobart and Duckett, 2022). While government responses were initially coordinated through a ‘national cabinet’ comprising the Prime Minister and all state and territory leaders, the response from the various state governments was far from uniform. Most notably, New South Wales, Victoria, and Queensland introduced additional home confinement measures that effectively made stay-at-home a directive rather than advice, with residents in these states liable for financial penalties if caught not observing the conditions imposed by these orders.

Reflecting the success of both the lockdown and border closures in suppressing the spread of the virus, the national lockdown was lifted in early May, though the rate with which restrictions were eased varied across states. Primary responsibility for policies intended to contain the virus was now largely left to the different state and territory governments, thus providing the potential for the type of exogenous variation needed for a quasi-experimental approach. Such variation emerged when, in late June 2020, Melbourne, the largest city in Victoria, became the epicenter of a local outbreak (linked to breaches in the hotel quarantine of returning international travelers) that saw an increase in the number of COVID-19 cases. This prompted the Victorian State Government to impose a range of lockdown measures, including business closures, stay-at-home orders, remote schooling, bans on all public gatherings, restricted access to family members living in aged-care facilities, and an evening curfew. The restriction on personal freedom was particularly severe, with people only permitted to leave home for four reasons: (i) shopping for essentials (which was limited to one person per household); (ii) outdoor exercise and recreation (which was restricted to one hour a day and only alone or with other members of the household); (iii) to attend medical appointments or to provide care to others; or (iv) to work or study if unable to work or study from home. Additionally, the wearing of face masks was mandatory whenever outside the home. In Greater Melbourne, these restrictions commenced on 9 July and remained in place until 27 October, making it the second longest continuous lockdown in the world (111 days). In regional Victoria, they commenced on 4 August and were lifted on 17 September.

Elsewhere in Australia, life seemed to be slowly returning to some degree of normalcy. Most importantly, there were no stay-at-home orders in place in any other state during the period of the Victorian lockdown. Other forms of restrictions, such as capacity limits on public venues, mask wearing mandates in certain settings, and restrictions on inter-state travel, remained in place in most states, but the extent of these restrictions and the degree to which they were enforced were highly variable across jurisdictions. This is reflected in scores on a measure of the stringency of the COVID-19 containment measures (with values that range from 0 to 100), which in mid-September varied from 25 in the Northern Territory to 51.85 in Queensland (Australian Bureau of Statistics, 2021a).1 In contrast, Victoria’s score rose quickly from 52.3 at the start of July to 92.59 by early August, remaining at that level until mid-September, and remaining relatively high until stay-at-home orders were completely lifted.

A consequence of the complex patchwork of restrictions, together with the highly stringent restrictions on international inflows and outflows, was that rates of community transmission of COVID-19 in Australia in 2020 were extremely low. Indeed, even in Victoria the rate of infection was very low by international standards – fewer than 17,000 cases were reported during the lockdown period (out of a population of around 6.7 million).

3. Data

The data used come from the Household, Income and Labour Dynamics in Australia (HILDA) Survey, a longitudinal study following members of a nationally representative sample of Australian households on an annual basis since 2001 (Watson and Wooden, 2021). Critical to our study, data collection is concentrated each year in the period August to November, and thus overlaps closely with the timing of the Victorian lockdown.2

Response rates are high, especially the annual re-interview rate, which over the period covered by this study averaged 95.2%. Thus, whereas non-response means the sample does not precisely match the wider Australian population, differences are mostly small. The exception here is recent immigrant arrivals. The nature of the panel design means that without constant refreshment samples (and one was added in 2011), the study cannot adequately represent migrants entering Australia after the panel commenced.

While the HILDA Survey commenced in 2001, for this analysis we restricted our pre-treatment period to the years since 2012, thus after the refreshment sample was added. Given the focus on fertility, the sample was then further restricted to persons aged 18–45. This provided an initial sample comprising 75,800 observations from 14,429 individuals.

The main outcome variable, fertility intentions, comes from an interviewer administered question that reads: ‘How likely are you to have a child / more children in the future?’ Responses are provided on an 11-point scale where only the end points are labeled (0 = very unlikely, 10 = very likely). A similar 11-point scale was also employed in recent research undertaken in Norway (Lappegård et al., 2022), with the only notable difference being the use of different end-point labels (“definitely not” and “definitely yes”). Most importantly, previous research (Drago et al., 2011, Fan and Maitra, 2013) has found that this variable is strongly predictive of realized fertility, with the analysis of Fan and Maitra (2013) suggesting this association is especially strong among women.3

While this question is administered in every year of the survey, differences in the sequencing and filtering of questions in survey years when additional questions about fertility behaviors were collected – 2015 and 2019 – means the fertility intentions data from those years are not strictly comparable with that collected in other years. The analysis was therefore restricted to observations collected in 2012, 2013, 2014, 2016, 2017, 2018 and 2020. Again, this is in line with the approach adopted in earlier studies of the HILDA Survey fertility intentions data. After removing cases with missing data on the outcome variable and with a zero-population weight (mostly persons who had moved into non-private dwellings), the final sample for analysis comprised a maximum of 57,184 observations from 13,798 persons. A detailed step-by-step summary of the sample selection process is provided in Fig. 1.

Fig. 1.

Sample selection process.

Two analytical samples were used. In the first sample, the treatment group was respondents living in Victoria and interviewed during the time of the second lockdown in 2020 (July 9 to October 27 in Melbourne and August 4 to September 16 in regional Victoria). This provided 56,967 observations from 13,756 unique individuals. In the second sample, the treatment group was identified using the lockdown end dates extended by 4 weeks (July 9 to November 24 in Melbourne and August 4 to October 14 in regional Victoria). This provided a slightly larger sample (57,066 person-year observations). In both samples, respondents living in Victoria who were interviewed after the treatment period ended were removed. This group represents approximately 10% of all Victorians interviewed in survey year 2020 when the treatment period is defined using exact lockdown dates, and 6% when the treatment period is extended by four weeks.

4. Methodology

Quasi-experimental DiD models are used to estimate the impact of lockdown on fertility intentions. The general approach is very similar to that used in an earlier study using these same data, but where the outcome of interest was a measure of mental health (Butterworth et al., 2022).

Changes in fertility intentions of those in the treatment group (Victorians interviewed during lockdown, or within 4 weeks of the lockdown being lifted) are compared with changes in the fertility intentions of those in the control group (persons living in the rest of Australia). Specifically, the following equation is estimated:

$${\mathit{FI}}_{i,t}={{\beta }_0+\beta }_1({\mathit{Victoria}}_i\times \mathit{lockdown})+{{\beta }_2{\mathit{Year}}_t+\mu }_i+{\delta }_s+{\epsilon }_{i,t}$$ (1)

${FI}_{i,t}$ denotes the fertility intentions of individual i in year t, and ${\mathit{Victoria}}_i\times \mathit{lockdown}$ equals 1 if respondent i lived in Victoria during the treatment period, and 0 if otherwise. The longitudinal nature of the data is exploited through the use of person-specific fixed effects (${\mu }_i)$, which control for all time-invariant factors (both observed and unobserved). ${\mathit{Year}}_t$ represents year fixed effects, with 2012 used as the base category. The DiD estimation in Eq. (1) also includes state fixed effects (${\delta }_s$). The standard errors were robust and clustered at the state level. Probability weights are used to ensure the samples were representative of the population in each year. This model is also used to test for heterogeneity in the effects of lockdown on fertility intentions.

The coefficient of interest, ${\beta }_1$, summarizes the effects of lockdown on fertility intentions under the assumption that fertility intentions in Victoria and the rest of Australia would have followed the same trend in the absence of lockdown. This parallel trends assumption is central to the DiD research design. As a test of the validity of this assumption, an event-study specification is estimated in which the single treatment indicator of Eq. (1) was replaced by a set of year dummies (${\gamma }_{\tau }$) interacted with the treatment state indicator:

$${\mathit{FI}}_{i,t}={\alpha }_0+\sum _{\tau =2012}^{2020}{\mathit{Victoria}}_i\times {\gamma }_{\tau }+{\mu }_i+{\delta }_s+{\gamma }_t+{\epsilon }_{i,t}$$ (2)

Additionally, we provide support for the causal interpretation of our results in different ways. First, we check whether our results were robust to the inclusion of time-varying controls. Second, we add state-specific time trends to control for potential bias stemming from time-varying variables measured at the state level. Third, we conduct a placebo test to assess the validity of our analysis. More specifically, we remove Victorian observations from the analysis and make residents of New South Wales (Australia’s most populous state) the treatment group. Since Victoria was the only state that imposed extended lockdowns during the second half of 2020, we should not find any systematic differences across states in the change in fertility intentions in 2020.

The assumption that β ₁ captures the effect of lockdown also requires that levels of anxiety and fear surrounding the contraction of the virus are no higher in the treatment group than in the control group. As previously emphasized, levels of prior exposure to COVID-19 in both samples were extremely low, and indeed were actually lower in the Victorian sample (0.3% vs 0.6%). More significantly, Victorian sample members were less likely to report a fear of getting seriously ill in the event of becoming infected (28% vs 32%). On the other hand, Victorians were more likely to report that they expected to get infected in the 12 months ahead (26% vs 22%).

Following previous analyses of the determinants of the fertility intentions that have used these data (e.g., Atalay et al., 2021; Bassford and Fisher, 2020; Li, 2019), fertility intentions is treated here as a linear continuous variable. Strictly speaking, however, the outcome variable only provides a limited set of discrete ordered choices and so, by definition, is not continuous. There are estimators designed specifically for this type of limited dependent variable – ordered probit and ordered logit – which have been used in some studies of fertility intentions (e.g., Billari et al., 2009; Busetta et al., 2019). In our case though, we need an estimator that can also accommodate person-specific fixed effects. Baetschmann et al. (2015) have developed such an estimator for the conditional ordered logit case, but we have opted not to report results from models using that estimator here.4 This decision was a function of both the inability of this estimator to incorporate weights and the much greater difficulty interpreting the incremental effect of the DiD coefficient within non-linear models (Athey and Imbens, 2006; Blundell and Costa Dias, 2009).

Summary statistics for the variables used in our analysis are provided in three tables provided in an (online) Appendix. Table S1 provides means and standard deviations for all variables disaggregated by gender. Mean fertility intentions were slightly higher among men (5.15) than women (4.68), possibly a function of the much higher rate of childlessness among the men in the sample – 60.6% compared with 50.2% among the women. As expected, more men than women were employed in permanent jobs (50.7%) or were self-employed (10.4%). In contrast, women were much more likely to be not working (22.6%), reflecting the fact that Australian mothers still bear the brunt of child care responsibilities, and when in work, were relatively more likely to be found working in casual jobs (17.4%), much of which involves part-time hours. We also distinguish between non-indigenous Australians, Indigenous persons and those born overseas, with 3.2% of the sample reporting to be of Indigenous origin, and 27.9% of men and 30.1% of women reporting being born overseas. The distribution of men and women across different household income quartiles was about the same. Finally, more women (37.1%) than men (29%) in the sample had obtained a university qualification, which reflects the marked growth in female participation in post-school education since the 1970 s in Australia (Wyn et al., 2017).

Table S2 compares mean fertility intentions of treatment and control groups during the pre-COVID period. This table reveals that mean fertility intentions of the control group during the pre-COVID period were much higher than for the control group (5,236 in Victoria vs. 4,826 in the rest of Australia). This is a key reason for the inclusion of individual fixed effects in Eq. (1). Comparable statistics for the treatment period (2020) are reported in Table S3.

5. Results

Table 1 reports the DiD estimates for the effect of lockdown on fertility intentions, with columns 1–3 using the exact treatment dates and columns 4–6 using the slightly wider treatment window (plus 4 weeks). All DiD models performed well, explaining a little over three quarters of the variation in fertility intentions (R ² = 0.76–0.78).

Table 1.

Effect of Victoria’s lockdown on fertility intentions.

	Panel A Exact lockdown dates			Panel B Lockdown + 4 weeks
	Persons	Men	Women	Persons	Men	Women
	(1)	(2)	(3)	(4)	(5)	(6)
Lockdown effect	-0.090	-0.032	-0.143*	-0.155**	-0.087	-0.219**
	(0.050)	(0.048)	(0.069)	(0.051)	(0.048)	(0.069)
Year fixed effects (baseline = 2012)
2013	-0.302**	-0.306*	-0.298***	-0.302**	-0.306*	-0.298***
	(0.091)	(0.144)	(0.051)	(0.092)	(0.144)	(0.051)
2014	-0.652***	-0.605***	-0.698***	-0.651***	-0.604***	-0.698***
	(0.038)	(0.082)	(0.033)	(0.038)	(0.082)	(0.033)
2016	-1.265***	-1.203***	-1.324***	-1.265***	-1.203***	-1.325***
	(0.054)	(0.104)	(0.054)	(0.053)	(0.104)	(0.053)
2017	-1.625***	-1.519***	-1.726***	-1.624***	-1.518***	-1.727***
	(0.038)	(0.068)	(0.037)	(0.039)	(0.069)	(0.037)
2018	-1.912***	-1.774***	-2.047***	-1.914***	-1.775***	-2.049***
	(0.036)	(0.068)	(0.040)	(0.036)	(0.068)	(0.040)
2020	-2.475***	-2.272***	-2.669***	-2.476***	-2.273***	-2.672***
	(0.080)	(0.102)	(0.066)	(0.080)	(0.102)	(0.065)
Equality test (F-statistic)		85.93***			117.18***
Pre-COVID mean – treatment	5.236	5.418	5.057	5.236	5.418	5.057
Pre-COVID mean – control	4.826	5.067	4.583	4.826	5.067	4.583
R-squared	0.770	0.761	0.777	0.769	0.761	0.776
Observations	56,967	27,498	29,469	57,066	27,546	29,520

Notes: Robust standard errors (clustered at the state level) in parentheses. All regressions also include state and person-specific fixed effects, and are weighted, adjusting for both complex survey design and non-response. *** p < 0.01, ** p < 0.05, * p < 0.10.

Prior research and theory led us to expect more pronounced changes in the fertility intentions of those in the treatment group (Victorians interviewed during lockdown, or within 4 weeks of the lockdown being lifted) compared with the fertility intentions of those in the control group (persons living in the rest of Australia). The first three columns provide strong support for this hypothesis, but only for women. That is, all else equal, fertility intentions declined among Victorian women in 2020 (b = −0.143, p < 0.1), but not among Victorian men. This gender difference in lockdown effects was statistically significant (F = 85.93). The estimated impact on women translates to a reduction in fertility intentions of 2.8% of the pre-lockdown mean. This resonates with accounts that the unequal distribution of paid and unpaid work during lockdowns became a source of relationship stress, particularly for women (e.g., Craig and Churchill, 2021). A slightly larger negative impact was observed when we allowed for a longer treatment window (columns 4–6, Table 1). This negative impact was only significant among women, with the estimated coefficient of − 0.219 (p < 0.05) equivalent to a 4.3% reduction in fertility intentions. Lengthening the treatment window further (by up to 12 weeks) produced estimates (but not reported here) that were little different from these.

We also draw attention to the point estimates on the year dummies. These all take negative values and become progressively larger with time, which is to be expected given the decline in fertility rates in Australia over this period (Australian Bureau of Statistics, 2021b). More importantly, there is no evidence here of any acceleration in this decline, with the magnitude of decline between 2018 and 2020 being no larger than in any of the three preceding two-year periods.

As previously mentioned, a critical assumption underpinning our identification strategy is that differences in the outcome between Victoria and the rest of Australia are not associated with differential trends in the absence of the lockdown, an assumption that we test by estimating an event-study specification. Fig. 2 depicts the results of this specification separately for men and women. The reference period here is 2018. As can be seen, for both men and women the pre-COVID period coefficients (and 95% confidence intervals) all hover around zero, providing support for the parallel trends assumption and adding credibility to our claim that trends in fertility intentions would have continued to be the same in Victoria and the control states in the absence of the lockdown.

Fig. 2.

Event-study of Victoria’s lockdown effect – test of parallel trends assumption. Notes: Reported are estimates and their 95% confidence intervals from separate DiD regressions of fertility intentions where the treatment group indicator (Victoria) is interacted with year fixed effects (the reference period is 2018). The graph shows differences between fertility intentions in Victoria and elsewhere in Australia relative to 2018, both before and during the 2020 lockdown. All analyses are weighted, adjusting for both complex survey design and non-response. Robust standard errors (clustered at the state level) are used for confidence interval calculation. The treatment group is identified using lockdown dates plus 4 weeks.

We next examined whether the effect of lockdowns on fertility intentions varied across different socio-economic groups. Drawing on previous findings that job losses and economic hardship disproportionately affected women, the less educated and those in lower income strata employed in industries that were hardest hit by the pandemic (e.g., Perry et al., 2021), we expected the impact of lockdown on fertility intentions to be larger for persons from these groups. Parameter estimates from a series of separate DiD models for different demographic and socio-economic groups are plotted in Fig. 3. The treatment period used here is the date of lockdown plus 4 weeks, and thus will typically be at the upper end (in absolute terms) of the range of possible estimates.

Fig. 3.

Estimated lockdown effect on fertility intentions by selected socio-demographic characteristics.Notes: Reported are treatment effect estimates and their 95% confidence intervals for men and women. Each estimate comes from a separate DiD regression of fertility intentions on a treatment group indicator and state, and year and person-specific fixed effects. All analyses are weighted, adjusting for both complex survey design and non-response. Robust standard errors (clustered at the state level) are used for confidence interval calculation. The treatment group is identified using lockdown dates plus 4 weeks.

As can be seen, estimated impacts do vary with demographic and socio-economic characteristics, and it is women where this variation is most pronounced. Most obviously, older women (36–45 years) were more affected (b = −0.447, p < 0.01) than younger women (the negative coefficients for the two younger age groups were both small and statistically insignificant). There is also a slightly larger impact on mothers (b = −0.387, p < 0.01) than on the childless (b = −0.162, p < 0.1), but the difference between these two coefficients was not statistically significant. These results resonate with previous accounts that lockdowns affected reproductive behaviors of different groups differently, possibly because the effect of lockdowns on feelings of uncertainty about the future also varies across groups. In the case of older women, lockdowns have been found to be associated with greater economic (Perry et al., 2021) and health anxieties (Butterworth et al., 2022), which could explain the more pronounced reductions in fertility intentions among this group. The pattern of results also provides some tentative evidence that it is women in more disadvantaged and economically vulnerable groups whose fertility plans were most likely to have been disrupted by lockdown. Most obviously, the negative impacts were largest for women with the lowest levels of education (that is, those who did not complete high school) (b = −0.899, p < 0.01). There were also larger negative impacts for both Indigenous women (b = −0.449, p = ns) and migrant women from non-English speaking countries (b = −1.295, p < 0.01). Only the latter, however, was significantly different from the smaller coefficient for non-indigenous Australian-born women (b = −0.162, p < 0.1). These findings suggest that temporarily shutting down many sectors of the economy may have widened pre-existing socioeconomic inequalities. On the other hand, the negative impact for women living in the lowest income households (i.e., the bottom household income quartile) (b = −0.398, p < 0.01) was little different from those in the second and third quartiles (b = −0.223, p = ns; b = −0.426, p = ns, respectively). Clearly there is no straightforward association between economic disadvantage and the impact of lockdowns on fertility plans, with the negative impacts concentrated on specific sub-groups.

There is also an unexpected finding with respect to employment status. Given widespread evidence that both unemployment and more insecure, time-limited forms of employment are associated with lower fertility rates (Alderotti et al., 2021), we expected to find negative lockdown effects to also be most pronounced among the unemployed and workers employed on fixed-term and casual contracts.5 While a relatively large negative impact was found for women employed on fixed-term contracts (b = −1.135, p < 0.05), the estimated impact on the unemployed was close to zero (b = −0.198, p = ns) while the impact on casual workers was actually positive (b = 0.636, p < 0.01). One possible explanation for this lies in the relatively generous, but temporary, increases in the level of out-of-work income support provided by the Federal Government during the pandemic.6 This suggests large effects might be observed if current income was controlled for. We thus re-estimated our models after including a measure of inflation-adjusted individual weekly income (comprising wages and salaries plus government benefit income). This made very little difference to our results: All DiD estimates were of broadly the same magnitude. Another possible explanation, at least for the finding of a positive impact on casual employees, lies in the secondary earner status of female casual employees in most households (71%, which compares with 46% among women employed in non-casual jobs). Job and income insecurity likely matter far less where workers are responsible for a relatively small proportion of household income, and thus fertility decisions of the women in those households will also be less impacted.

Among men, effect sizes are obviously much smaller and mostly insignificant. Nevertheless, there is a very large negative impact on Indigenous men (b = −1.519, p < 0.01). Very differently, and in contrast to women, low educated and unemployed men both experienced a heightened desire to have children (b = 0.265, p < 0.05; b = 0.655, p < 0.05 respectively). These generally small, and mostly insignificant, effects among men most likely reflect the fact that men have less control over fertility outcomes and hence effects on fertility intentions are smaller than among women.

Finally, we conducted a range of sensitivity checks to test whether our estimates are robust (reported in Table 2). We focus here on the results using the date of lockdown plus 4 weeks (panel B in Table 1), but our results are qualitatively similar when using the exact lockdown date (panel A). We also restrict our attention to women given the lack of significance of any treatment effect for men.

Table 2.

Robustness checks (women sample).

	Adding individual controls	Allowing for state-specific time trends	Placebo test: NSW as the treatment	Allowing for LGA-specific time trends
	(1)	(2)	(3)	(4)
Panel A: Exact lockdown dates
Lockdown effect	-0.151**	-0.127***	-0.115	-0.237**
	(0.051)	(0.004)	(0.068)	(0.096)
R-squared	0.791	0.777	0.777	0.794
Observations	29,429	29,469	21,572	29,467
Panel B: Lockdown + 4 weeks
Lockdown effect	-0.215***	-0.201***	-0.105	-0.303**
	(0.049)	(0.007)	(0.068)	(0.098)
R-squared	0.790	0.776	0.777	0.782
Observations	29,479	29,520	21,699	29,518
Pre-COVID mean – treatment	5.058	5.057	5.047	5.057
Pre-COVID mean – control	4.583	4.583	4.232	4.583

Notes: Robust standard errors (clustered at the state level) in parentheses. All analyses are weighted, adjusting for both complex survey design and non-response, and include year, and person-specific fixed effects. In column (1), individual controls are included for age group, education, marital status, employment type, having children, long-term health conditions, equivalized real household income (log), and a non-positive income indicator. *** p < 0.01, ** p < 0.05, * p < 0.1.

First, we checked whether our results were robust to the inclusion of time-varying controls. As shown in column 1 of Table 2, inclusion of these controls made almost no difference to the estimated impact (b = −0.215, p < 0.05 vs b = −0.219, p < 0.05).

Second, in column 2, we added state-specific time trends to control for potential bias stemming from time-varying variables measured at the state level. Again, this had very little impact on our coefficient estimates.

Third, we conducted a placebo test, making New South Wales the treatment group (and dropping Victorian observations), to assess the validity of our analysis. As shown in column 3, the coefficient of interest is, as hypothesized, not statistically significant. We also repeated this placebo test using each other of the alternative States and Territories as the treatment group. The results of this test are summarized in Fig. 4. With the exception of the Northern Territory (Australia’s smallest province, accounting for less than 1% of the total Australian population), results are always insignificant. The significant and positive differential in fertility intentions for the Northern Territory relative to other states possibly reflects the much more relaxed imposition of COVID-related restrictions (such as capacity limits on public venues, mask wearing mandates and restrictions on inter-state level) in that state. Indeed by the time the Victorian lockdown commenced the only significant restrictions in place in the Northern Territory related to cross-state and international travel.

Fig. 4.

Placebo test results using other states as the treatment. Notes: Reported are estimates and their 95% confidence intervals from separate DiD regressions of fertility intentions where each of the alternative States and Territories is separately treated as the treatment group (and removing Victoria) and is interacted with year fixed effects (the reference period is 2018). All analyses are weighted, adjusting for both complex survey design and non-response. Robust standard errors (clustered at the state level) are used for confidence interval calculation. The treatment group is identified using lockdown dates plus 4 weeks.

Fourth, we replaced the State fixed effects with fixed effects measured at the local government area (LGA) level, and also added interactions between these LGA-specific effects and year.7 This helps us test the extent to which our results are driven by area-based differences in exposure to the COVID-19 virus. If fertility intentions are correlated with COVID-19 infection rates at the local area level, then we would expect the inclusion of LGA fixed effects to cause our estimated lockdown effects to become smaller in absolute size. As shown by the results presented in column 4, the estimated negative lockdown effect on female fertility intentions actually becomes larger in the presence of LGA fixed effects.8 This leads us to reject the hypothesis that variations in LGA-specific circumstances (including local COVID-19 infection rates) explain part of our estimated negative lockdown effect.

6. Discussion

Around the world, lockdowns have been a commonly used tool in the battle against the COVID-19 pandemic. Their effects on fertility, however, remain insufficiently understood. Drawing on the emerging literature on the effects of the COVID-19 pandemic on fertility (e.g., Emery and Koops, 2022; Kahn et al., 2021; Lin et al., 2021; Lindberg et al., 2020, Lindberg et al., 2021; Luppi et al., 2020; Micelli et al., 2020; Naya et al., 2021; Zhu et al., 2020), we expected lockdowns to cause an acceleration in the decline in fertility intentions among men and women given the economic and health uncertainties attached to the imposition of COVID-19 restrictions. We also expected these lockdown effects to vary across different socioeconomic and ethnic groups, with families from the most marginalized groups more likely to experience greater financial and health hardships and so be at greater risk of delaying childbirth. To test these hypotheses, we combined rich nationally representative longitudinal survey data with a natural experiment that emerged in Australia during the second half of 2020 to quantify whether and how much lockdowns changed fertility intentions. We deployed fixed effects models to account for time-invariant state-specific unobserved characteristics and addressed omitted-variable bias.

Our first set of findings suggest that the Victorian lockdown caused a dampening effect on women’s fertility intentions when averaged across the population aged 18–45. We found an average effect for women ranging between − 0.143 and − 0.219, which is modest in size, and some might even say small. Despite its small size, this finding is consistent with the prediction that prolonged periods of lockdowns placed constraints on couple’s social and intimate lives, which in turn caused fertility intentions to drop (e.g., Lindberg et al., 2021). The slightly larger and stronger negative effects among women when we allowed for a longer treatment window suggest that uncertainty about the length and progression of lockdowns made it harder for women to face decisions about having children. These findings are potentially of large significance from a policy point of view given both the strong predictive effect of fertility intentions on realized fertility in Australia documented in previous research (e.g., Drago et al., 2011; Fan and Maitra, 2013) and the ongoing decline in total fertility rates in Australia (as in most other countries). The Australian Treasury, in its most recent Intergenerational Report, for example, describes Australia’s greatest demographic challenge as its ageing population, which in turn is a function of both increasing life expectancies and falling fertility rates (Australian Government Department of the Treasury, 2021). Anything that exacerbates the decline in fertility will thus be of concern to policymakers.

Second, and in line with our expectations, we found that the lockdown effects on fertility intentions varied across women of different ages, and was largest for the less educated, immigrants from non-English speaking countries, and those employed on fixed-term contracts. The burden of lockdowns was thus not experienced equally across socioeconomic and demographic groups, consistent with the emerging pandemic literature on a range of socioeconomic outcomes (for a review, see Brodeur et al., 2021). There was, however, no evidence that effects were, on average, more pronounced for those living in low-income households.

Third, the results indicated no significant acceleration in the downward trend in fertility intentions in 2020 in other states of Australia (i.e., outside of Victoria), implying that the negative impact of the pandemic on fertility intentions was entirely the result of lockdowns.

Perhaps most importantly of all, our results suggest that COVID-19 and the associated policy responses have had a much smaller effect on women’s fertility intentions than suggested by other surveys conducted in other countries. Previous research, however, has relied on cross-section surveys, mostly involving small non-probability samples, providing us reasons to believe that the results obtained may not provide an accurate reflection of the magnitude of decline in fertility intentions in wider populations. Further, and as previously noted, previous research has not been able to distinguish the effect of policy responses to the pandemic (such as lockdowns) from other channels through which COVID-19 may have affected fertility behaviors.

Whether the relatively small declines in fertility intentions reported here will result in actual declines in future birth rates remains an open question. If women delay having children during periods of lockdowns but compensate with more births at later ages as jobs become available and economies recover from the social and economic disruptions of the pandemic, total fertility rates over the longer-term may be little affected. This scenario would seem to be especially likely in the Australian context given the adverse economic impact of the pandemic was concentrated in a relatively short period (the second quarter of 2020) and employment levels had reached record highs by the end of 2021. Alternatively, if some women not only delay childbearing but forgo parenthood due to perceived income and employment uncertainties, then even small declines in the fertility intentions could accelerate the ongoing decline in fertility rates.

Our study is not without limitations. First, the outcome in this study is fertility intentions which, while predictive of, is not the same as actual births. Tracking births of women before and after the pandemic would more accurately reflect the true consequences of the pandemic and its associated lockdowns on fertility. Second, although methods were used to correct for both non-random sample attrition and non-random response, this re-weighting approach is not sufficient to address the under-representation of immigrants arriving in Australia after 2011. If recent immigrants were more likely to be employed in industries that were hit hardest by the pandemic and business closures, then our results may have underestimated the true impact of lockdowns on fertility intentions. Third, although imposed lockdowns in many parts of the worlds share many common factors, our results may be specific to Australia, suggesting the need for future research in other countries to test whether our results can be replicated in other settings with other data. Fourth, and most critically, a key claim in this study is that an estimate of the effect of lockdown on fertility intentions has been obtained, and that differences between treatment and control groups are not due to perceived differences in the risk of contracting the virus. Victorians, however, were more likely to report that they expected to contract the virus in the future than residents in other States (though were actually less likely to think that infection would result in serious illness), and thus the possibility that some of the estimated treatment effect is not solely due to lockdowns cannot be ruled out.

This uncertainty about the precise source of the decline notwithstanding, our results suggest that policymakers, when designing responses to future pandemics, should give at least some consideration to the impacts on women’s fertility plans: The immediate health and economic impacts cannot be the only considerations. Further, the importance of such considerations is only likely to grow over time as governments grapple with the consequences of declining and aging populations. (Jarzebski et al., 2021).

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

Supplementary material

.

Data availability

The authors do not have permission to share data.