The Relationship between Fertility and Female Participation in the Labour Force in OECD Countries 2000–2020: It Is (Again) Negative

OECD countries, at least most of them, have undergone a considerable rise in female labour force participation (Thévenon, 2013). Precisely, at the beginning of the 21st century, the female labour force participation (FLFP) rates (Figure 1) varied from just 46.3% in Italy and 50.5% in Greece to 83.3% in Norway (OECD, 2022a). A decade later, the same pattern was perceived; namely, the lowest rates of female participation have been recorded in Italy with only 51.1 % but now instead of Greece, we have South Korea in second place with 54.5%, while the highest recorded rates remained in Iceland with 81.1% (OECD, 2022a).

Figure 1.

In 2020, the countries with the lowest participation rates were Italy and Greece with 54.7% and 59.3%, while Iceland remained on the top with 80.7% (OECD, 2022a).

The reasons for the observed differences are various. The earlier findings suggest a U-shaped relationship between FLFP and economic development (Goldin, 1995). Arguably, while the early stages of economic growth are accompanied by lower participation of women in the labour force, the share eventually increases as income rises further (Goldin, 1995). In addition, researchers suggest the influence of cultural patterns, e.g., gender roles and position in the household are also important factors in curbing the observed level of female participation in the labour force (Clark, Ramsey & Adler, 1991; Uberti & Douarin, 2023). According to the report regarding policy and action implemented until 2020 in G20 countries, the persistent gap in female participation rates between, for example, Mediterranean countries (e.g., Italy and Greece) and Nordic countries (e.g., Iceland and Sweden), could be, to some extent, rooted in culture and social norms, but at the same time may reflect economic realities in the particular country (ILO and OECD, 2020).

Regarding the trends in fertility rates recorded at the beginning of the time frame that this study explores (2000–2020), the Czech Republic and Spain recorded the fewest children per woman at 1.14 and 1.23, respectively, in 2000 (Figure 2). On the other hand, the United States and Israel at 2.23 and 2.95, respectively, recorded the highest number of children per woman in 2020 of all OECD countries (OECD, 2022b).

Figure 2.

In the middle of the observed time interval, we find Hungary and South Korea with only 1.26 and 1.23 children per woman, respectively, while Iceland with 2.2 and Israel with more than 3 children per woman occupied the first two places (OECD, 2022b). Finally, in 2020, the lowest level of total fertility rates (TFR) has been recorded in South Korea, with just 0.86 children per woman, while Israel stayed on the top with 2.93 children per woman (OECD, 2022b).

To account for the recent trends experienced across OECD nations regarding fertility rates, authors of these reports suggest that institutional variation and labour market insecurity may play a role (e.g., Adserà, 2004), accompanied by well-elaborated and documented inverse associations between age at first marriage, education, female labour force participation, and so forth. (Kohler, Billari & Ortega, 2002). Arguably, every exit from the labour market at childbirth hints at a potential loss of income, a lower wage after returning to the job market, and a higher risk of unemployment (Adserà, 2004). Consequently, the wide range of labour market arrangements across OECD countries sets the childbearing and participation decisions of women. Government answers usually focus on the compatibility of work and family for women. But, with an increase in female labour force participation, family formation becomes irreconcilable with career. Furthermore, the dual responsibility for women induces a trade-off between the size of the family and employment. As an illustration, across OECD countries in 2011, the mean personal ideal number of children for women was around 2.3 (OECD Family Data Base, 2022). Furthermore, most adults prefer to have two children: more than 50% of 15- to 64-year-old women consider two children to be an appropriate ideal family size in most OECD countries (OECD Family Data Base, 2022). Eventually, the mean actual number of children was just around 1.3 children per woman (OECD Family Data Base, 2022). It is clear that women adopted plans intentionally to integrate the two spheres of life through the timing of motherhood and return to the labour force (Ni Bhrolchain, 1986).

Given that both FLFP rate and fertility could have a decisive role in determining women's position in society and family, there has been huge interest in the relationship between them (Behrman & Gonalons-Pons, 2020). Arguably, the cross-sectional correlation between FLFP rates and fertility in developed countries shifted from negative to positive during the 1980s (Esping-Andersen, 1999; Brewster & Rindfuss, 2000), and the general view is that the positive link is false due to “the country-specific heterogeneity”, while the time-series association remained negative (Kögel, 2004).

Therefore, Kogel's assertion that the time-series association between TFR and FLFP is negative motivated me to explore whether this is still valid for the latest data. In particular, it may not hold because both welfare support for the reconciliation between paid work and childcare and the involvement of men in childcare have expanded since 2000, which should facilitate the positive relationship between TFR and FLFP. Furthermore, Oshio, after utilizing annual data from 1970 to 2017, concluded that “the more recent the data set used, the more likely it is that the time-series association will be positive between FLFP and TFR” (Oshio, 2019, p. 1284). In addition, he contended that by using a more recent data set, the inconsistency between cross-country and time series should disappear; that is, both would indicate a positive association between FLFP and TFR (Oshio, 2019, p. 1284). On the other hand, many advanced economies experienced declines in fertility rates after the global financial crisis in 2008, so the trend for the TFR–FLFP relationship may have changed.

In order to examine the above assumptions, this study will investigate the relationship between TFR and FLFP using a macro approach, with an ambition to enrich the body of evidence about the time-series relationship between the total fertility rates and female labour force participation rates in advanced economies since the year 2000.

For this purpose, I have compiled a dataset with variables obtained from the OECD database. All examinations were performed at the country level and explore aggregated connections between employment and fertility. The benefit of employing aggregated data is “that the experience of living in a country where many women are employed may have an important spill over effect even among unemployed women” (Behrman & Gonalons-Pons, 2020, p. 40). Arguably, the increase in women's engagement could advance the onset of broader transformations in norms on gender and fertility even among women who are not engaged but just uncovered new modes of behaviour (Behrman & Gonalons-Pons, 2020, p. 40).

In the next section, the theoretical underpinnings of the relationship between fertility and employment are presented. After that, it presents the data, followed by the employed methodology and results. The paper closes with a short discussion of the advantages of the applied method and conclusions.

The prominent economist Becker (1965) proposed the microeconomic theory of fertility and women's labour supply emphasising the role of specialization within the family. He stressed that women are merely additional income providers. Importantly, according to the author, women's employment and fertility decisions depend on the price effect. Finally, the suggested model expects the price effect to exceed the income effect. Consequently, it predicts an inverse correlation between female labour supply and fertility.

The opposite sign of the correlation is assumed by Oppenheimer's most famous pieces, published in 1977 and 1997, which criticized the role specialization assumption (Oppenheimer, 1977; 1997). Namely, women's participation in the labour force increased significantly around Europe, thus reducing the duration of family-related employment gaps. Therefore, she has argued that in contemporary societies, women's earnings contribute more than ever to the home budget, especially in highly developed countries of the global north where female employment is more promoted and socially accepted (e.g., Sweden, Norway, Denmark). Consequently, the income effect of women's employment should exceed the price effect, thus making the correlation between female participation and fertility positive.

Similarly, Esping-Andersen and Billari (2015) argued that the gender revolution and the departure from the traditional woman-homemaker family model brought about a decline in fertility. With time, according to the gender equilibrium theory, new and more egalitarian gender arrangements will be established because societies begin to settle into the new balance regarding family forms, i.e., more gender-egalitarian family arrangements (Esping-Andersen and Billari, 2015). By adopting this nontraditional shift in gender role, a positive relationship between female labour participation and the fertility rate is to be expected. Moreover, Nordic countries, characterized by high values for the human development index (HDI) and relatively high fertility, served as role models for the supporters of this theory. For example, Myrskylä, Kohler, and Billari (2009) argued that in developed countries with a higher human development index (HDI), the association between female participation in the labour force and fertility would also turn from negative to positive once the level of HDI reaches the higher values. Namely, in countries such as Norway, the US, the UK, and Denmark, among others, the TFR reversed and became positive between 1975 and 2005. Unfortunately, the problem was the design of the study, which was based on measurements in a cross-section. According to Lesthaeghe (2020), the study by Myrskylä et al. is a case of what Arland Thornton (2005) calls “reading history sideways”. Arguably, the future evolution of societies in this area was based on a single and static cross-sectional pattern, that is, without including time as a factor (Thornton, 2005).

Similarly to the gender equilibrium theory, Goldscheider, Bernhardt, and Lappegård (2015) argue that there has been a weakening and reversal of entrenched traditional gender roles since the employment of women became a common practice. They refer to these shifts in gender relationships as the gender revolution. Namely, the higher female participation in the labour force should be perceived just as the initial phase of the gender revolution. Subsequently, with the higher involvement of men in the private sphere of the family, according to the gender revolution theory, the stability of the family will increase (Goldscheider et al., 2015). Arguably, the negative aspects of family change (e.g., low fertility and increase in divorce rates) induced by the movement of women into the labour market, will be lessened by the higher involvement of men in the home (Goldscheider et al., 2015). Eventually, a positive relationship between female labour supply and fertility rates is expected, with an increase in welfare support for combining paid work, childcare, and men's involvement in childcare.

Assertion of how the cross-sectional relationship between FLFP and fertility rates across developed countries has changed from negative to positive, has also further fuelled a debate about the association between work and childbearing (Oshio, 2019). Arguably, the shift in the connection between fertility and female employment occurred because of the reduced collision in women's function as mother and worker (Rindfuss, Guzzo & Morgan, 2003), followed by the boost in womens’ salaries (Ahn & Mira, 2002).

By using panel data on fertility and FLFP rates, Ahn and Mira (2002) showed how the correlation shifted from negative to positive around 1985. Namely, during the 1970s and up to 1985, the correlation was negative and significant while from 1985 until 1995 it become positive and significant (Ahn & Mira, 2002). A similar change in the positivity of correlation was confirmed by Rindfuss and colleagues (2003). By employing data from 1960 until 1997 for 22 low-fertility countries, they noted a shift in the association of fertility levels to women's levels of labour force participation. Moreover, the study documented the same positive association between the country-level TFR and the proportion of births to unmarried women. Arguably, these shifts in the associations are the result of the adopted policies that have reduced both the incompatibility between women's roles as mother and worker and decreased the link between marriage and childbearing (Rindfuss, Guzzo, & Morgan, 2003).

Contrary, Kögel (2004) did not detect a positive association between fertility and female participation when using time-series data for OECD countries. Arguably, the reversal in the sign of the cross-country correlation is most likely due to the existence of unmeasured country-specific elements, and due to “the country-heterogeneity in the magnitude of the negative time-series association between fertility and female employment” (Kögel, 2004, p. 46). Namely, the existence of country-specific effects suggests that differences between countries regarding public policies or labour market practices may be the main cause for why some countries have high fertility and high female employment while others have low levels of both (Kögel, 2004). Therefore, he applied FE models to country-level time-series data from countries that are members of the OECD by dividing the analysis period into two timespans: 1960–1985 and 1985–2000. Subsequently, he confirmed that the time-series connection between female participation and fertility is inverse and that the negative sign of this relationship has weakened since 1985.

Kögel (2004) illustrated that the relationship between TFR and FLP remains negative and significant once we include country dummy variables in pooled time series and cross-country data. Accordingly, Engelhardt and Prskavetz (2004) have tried to find omitted variables that may be behind dummy country variables. Therefore, they clustered countries according to the dynamics exerted regarding FLFP. Afterwards, they employed several economic and social variables that have changed following a transition in female labour supply by using a descriptive approach. Although descriptive pieces of evidence cannot strongly prove that a shift in cross-country correlation between fertility and female employment occurred because of the country-specific characteristics, they concluded how both the trend in the labour market and demographic indicators are consistent with new home economics theory; therefore, a negative association between TFR and FLFP is expected (Engelhardt & Prskavetz, 2004).

Following a path similar to that of Kögel (2004), Oshio (2019) estimated the time-series association between TFR and FLFP by fixed-effects regression models, which can be controlled for country-specific heterogeneity, and found that more recent the data set we use, the more likely it is that the time-series association will be positive (Oshio, 2019). The FLFP-TFR correlation diverted from negative to positive when FLFP surpassed a value of 0.6 for data on 24 countries belonging to the OECD from 1970 to 2017 (Oshio, 2019).

Although the current study is on the macro level, it is important to highlight the significant contributions of the papers that dealt with the relationship between fertility and labour participation on the individual level in developed countries. For most Western economies, studies on the micro level support the existence of a negative correlation between female employment and childbearing, indicating a prevalence of the “price” effect over the “income” effect (Matysiak & Vignoli, 2008). Furthermore, the delay of motherhood and avoidance of subsequent childbearing occurs in countries where the labour market institutions are not following the need for female work, i.e., in Mediterranean countries (Adserà, 2005).

In contrast to the above, in countries that were once members of the Eastern Bloc, studies have offered evidence that the probability of having a first child did not depend on a woman's employment status (Matysiak, 2009). Having in mind that these countries experienced the eradication of public support for working parents after the fall of state socialism, the results of the studies are even more puzzling. In addition, cultural norms regarding work–family balance that prevail in formerly socialist countries are similar to Mediterranean countries (Thévenon, 2011), therefore making this issue even more interesting from the researcher's point of view. Arguably, since the female participation rates were always high in Eastern Europe, probably because women have a long history of being income providers in comparison to their counterparts in the West (Matysiak & Vignoli, 2013).

Overall, in light of the premises cited above, in this study I expected to find strong friction between fertility and women's participation rates across the OECD member countries, originating from predominantly unfavourable circumstances for work and family reconciliation. Nevertheless, I anticipated time series that is, within-country differences between the fertility and participation rates, in order to dig deeper into the relationship beyond simple association. Moreover, by clustering countries according to their historical and cultural experience, the occurrence of a positive correlation between the total fertility rates and female labour participation on the OECD level will be further investigated.

The study uses annual data from 32 countries out of the 38 in the OECD. The analysis excluded five countries (Chile, Colombia, Mexico, Turkey, and Costa Rica), because the full data set for 2000–2020 was not available. In addition, Israel is excluded due to the much higher values of TFR.

Two essential variables in this study are FLFP and TFR, the data for which was downloaded from the databases of the OECD website (https://stats.oecd.org). Also, the model incorporates the gender wage gap, the share of part-time employed women, public expenditure on family (in cash and in-kind) as a share of GDP per capita using purchasing power parity rates, the share of women (15–64) with tertiary education, mean age at birth, and the categorical variable that grouping OECD countries in five clusters.

The study employed the FE model and linear random intercept model (maximum likelihood), a so-called “unified” model (Bartels, 2009). The critical value chosen was 0.05. The particular advantages and shortcomings of the models used will be elaborated on in the Methods section. The significance codes used are: < 0.05**; <0.001***.

All estimations were conducted using the STATA 16 statistical software package.

By applying two different models to the same data set, we expect to explore the possibilities of each in detecting the association between female employment and the level of fertility. By employing distinct models with their specificities, on the latest data set available, we expect to scrutinize more comprehensively the relationship between employment and childbearing and portray a clear picture of the nature of this relation. Because the model's unobserved heterogeneity in clustered data (in our case countries) is the subject of debate in the statistical literature (see Wooldridge, 2002), this section will highlight some differences between the modelling approaches usually employed by previous studies.

In Table 7, the study juxtaposes the estimation of the correlation coefficients between the TFR and explanatory variables for both models that were utilized.

First, by introducing the lag variable for the change in TFR, the study confirmed the conditionality of the current level of TFR on the past values. Consequently, life in a country with a higher level of TFR could have “spillover effects” even among childless women (Behrman & Gonalons-Pons, 2020). Eventually, a positive correlation across countries, captured by unified models, explains higher than average values of TFR in Scandinavian countries until the 2008 GFC and the long-lasting trend of higher values of TFR could explain the smallest negative time-series association that Kögel (2004) stressed for these countries.

Further, one can notice that the positive effect of FLFP on TFR was captured at 0.05 alpha level only by the unified approach because FE makes the effects of independent variables exclusively within-cluster. Eventually, the cross-sectional association captured by the unified approach, as Oshio (2019) already emphasised, ignores country-specific effects, thus these estimations are biased.

Regarding time-series association, based on the results of both previous studies (Kogel, 2004: Oshio, 2019) and the current one, the FE and unified models are shown as the most promising approaches to detecting the true nature of the relationship between TFR and FLFP, and both have suggested an inverse relationship between our variables of interest. Even though neither FE nor unified approaches captured the association between TFR and FLFP within a 95% confidence interval, further dividing the sample from 2000 to 2010 and from 2011 to 2020 reveals the real nature of the relationship. Importantly, it is not that a unified approach is superior to the FE, moreover, the within-effects in the unified model show basically the same results as the FE model, but it has the possibility, by decomposing the effect of FLFP on TFR, to detect the opposite signs in correlation across and within countries.

Besides the negative effect of the GFC in 2008 on TFR, results in Table 6 depict how TFR dropped significantly under the level observed before the GFC in many advanced economies, suggesting the negative influence of other factors beyond the GFC. Therefore, a purely economic logic that focuses on financial insecurity cannot solely explain the negative correlation between fertility development and FLFP that the study finds after the 2010s. Consequently, we need to adopt a more inclusive theory that combines additional views beyond the exclusively economical ones (Comolli et al., 2020). Having said that, Vignoli and his colleagues (2022) suggested the way in which people build their narratives of the future as a fruitful avenue for understanding the decrease of TFR in the years following the GFC (or Great Recession) in 2008. Arguably, individual decisions regarding fertility are not just the function of one's past but also of one's future with all uncertainties based on upcoming imagined scenarios (Vignoli et al., 2022). Hence, the decision to have a child is more complex because it involves not only economic facts but also one's perception of the future in a national and global context. Eventually, in the context of Nordic countries, cuts in social and family policies during and after the crisis could have a more pronounced effect on fertility reduction because individuals are used to relying on it (Comolli et al., 2020).

The next variables are the share of women with tertiary education, mean age at birth, and expenditure on family in cash, effects that were captured by both the models. Again, only a unified model enables an unambiguous interpretation of results, by separating between- and within-country effects.

Regarding the gender wage gap, not only was the positive time association captured by the unified model, but also the FE model captured its effect. Importantly, the former suggested a statistically significant difference between cross-section and time series associations (see Table 3, 3rd column). Consequently, the clarity of the unified model provides more information regarding the effect of the gap in wages between men and women on TFR by offering a fresh avenue for the research by indicating the importance of national context in the relationship, since the within-country coefficient is statistically significant and has the opposite sign from the coefficient across countries.

Eventually, the estimated coefficient for the effect of expenditure on family in-kind fully demonstrated the advantage of the unified model. While the FE approach offered a one-sided interpretation of the relationship between TFR and in-kind payments for families, the unified model delivered a more holistic perspective of the nature of the association, subsequently increasing the accuracy of interpretation.

In the end, for region-specific effects, the unified model didn’t manage to detect statistically different values of TFR, for the 2000–2020 period, between regions.

The present paper, besides the inclusion of a wider set of explanatory variables, has also included more recent data (2000–2020) and expanded the sample to 32 OECD countries, in comparison to the previous studies. In line with Kögel (2004), the results confirmed a negative time-series association between female employment and fertility, although not at the conventional 0.05 alpha level. However, the paper verified a negative and statistically significant correlation for the latest period, i.e., 2011–2020. Moreover, it endorsed findings from previous studies regarding the negative effects of the 2008 GFC on the total fertility rate (Sobotka et al., 2011; Camolli, 2017; Camolli et al., 2020). In contrast to this view, Oshio (2019) suggested that if a more recent data set is used, the time series correlation between FLFP and TFR will be positive. If we look back to the past, Rindfuss et al. (2000) discovered that the cross-country association between the TFR and the FLFP in OECD countries (which was negative until 1985) has changed to positive. Correspondingly, by employing the cross-sectional association, the paper confirmed a positive cross-sectional association between FLFP and TFR, for the 2000–2020 period, but further offered reasons why positive cross-country association is biased.

With the intention of explaining some limitations of the FE method, which was suggested by researchers who are interested in the relationship between TFR and FLFP, the study proposed a unified approach (Bartels, 2009), by retaining the best characteristics of the former method, while proposing a strategy that offers a better interpretation of results. Notably, it provides researchers with the opportunity to be more accurate when interpreting the results of longitudinal data. Eventually, it could have an impact on the policymakers who are considering whether a particular policy should be implemented.

In the end, it is important to emphasise some limitations of the study that should be addressed in the future. Primarily, researchers are constrained by data availability, thus using TFR, as Oshio (2019) nicely emphasised, was the second or third best solution due to the lack of data regarding cohort TFR. Also, it was limited only to the OECD countries, so widening our sample beyond the limits of developed countries should be sine qua nonfor subsequent researchers. Importantly, the current study has examined the association within the latest data set, because there is already enough evidence on both the sign and size of the correlation between TFR and FLFP from 1960 until 2000 (Kogel, 2004) and from 1980 to 2017 (Oshio, 2019). Eventually, given the bi-directional relationship between TFR and FLFP (Engelhardt & Prskawetz, 2004; Mishra & Smyth, 2010; Subramaniam et al., 2018), it is inappropriate to draw strong causal inferences at this point. Nonetheless, it is clear that shifting FLFP and change in TFR are closely interrelated.