Declining fertility in Taiwan: the deterring impact of housework imbalance

ABSTRACT

Fertility in Taiwan has been persistently low since 2003. Theorists have attributed this to gender inequity in domestic labour, yet this relationship has not been statistically tested. We assess the way in which the division of housework influences the probability of having an additional child. We assess this relationship for a sample of childbearing-aged married couples, as well as for education- and employment-specific subgroups. We find evidence of impacts for university-educated and working-mother couples, and when survey respondents are wives rather than husbands. The probability of a university-educated and working-mother couple with an equal division of housework having a child within five years is 0.73, whereas the probability of a couple with the mean division of housework having a child is 0.39. This finding is significant at the 1 per cent level.

KEYWORDS:

• Fertility
• Taiwan
• gender equity
• domestic labour

Introduction

According to the UN (UN, 2019), Taiwan has the third lowest Total Fertility Rate (TFR) in the world, with a period TFR of 1.13 in 2018. The long-term social and economic consequences of lowest-low fertility rates are potentially severe. As an increasing share of the population survives to older ages, rises in the dependency ratio are expected to erode the local human capital base, accelerating population aging and putting pressure on the health care and pension systems (Bloom, 2011). Persistent low fertility also builds negative momentum into population growth, reducing both the absolute number of women in future reproductive age ranges and future workforce (Feeney, 2003). Low fertility thus represents a major global policy concern, with governments seeking to increase fertility via cash transfers, parental leave, and subsidised childcare (Lee, 2009; Lin & Yang, 2009).

To understand low fertility levels in developed countries, a body of literature has emerged that emphasises the role of gender equity (Balbo et al., 2013; Goldscheider et al., 2015; McDonald, 2013). Conceptually, incoherence in the role of gender in social and domestic spheres is argued to result in low fertility levels. While some social institutions such as the family still prescribe traditional gender roles, contemporary social institutions such as education and employment are considered to adopt an egalitarian view of gender, and thus a large proportion of women are expected to forgo marriage or childbearing in favour of a career (Goldscheider et al., 2015; McDonald, 2000). These ideas have conceptually been used to explain low fertility in Taiwan, yet the direct impact of gender inequity on births has not been quantified (Frejka et al., 2010; Jones, 2019; McDonald, 2009; Raymo et al., 2015; Tu et al., 2017). While there is some quantitative evidence for a domestic gender equality effect on fertility desire, and qualitative evidence that unequal domestic labour divisions inhibit fertility decisions, there is no quantitative evidence for an effect on realised fertility behaviour (Cheng & Hsu, 2020; Freeman et al., 2018; Kan et al., 2019; Kan & Hertog, 2017). In particular, we do not know whether more equal distributions of housework and childcare are associated with a greater probability of having another child, and whether this association varies among subgroups with different levels of education and employment participation.

In this study, we seek to statistically assess the influence of the division of housework on the probability of future birth occurrences, drawing on data from the Taiwanese Panel Survey of Family Dynamics (PSFD), from 2010, 2011, 2012, 2014 and 2016. Specifically, we use binary logistic regressions to estimate the strength of associations between the share of household work and the probability of having an additional child.

The next section provides an exposition of gender equity theory and evaluates the evidence for gender equity theory, considers how gender equity may impact the fertility of different education and employment subgroups differently, and explores how the expansion of higher education and increasing female labour force participation have impacted gender and family in Taiwan. The following section describes our source of data, and discusses the restrictions we apply to obtain our analytical sample. We then specify the variables and models used in the analysis. The next section presents our model results and explores differences among educational and employment subgroups. The penultimate section discusses the implications of our results for policy and research, and a final section concludes.

Literature review

Gender equity theory

The effects of gender equity have been studied in the wider fertility literature through the lens of gender equity theory as developed by McDonald (2000, 2013). This theory suggests that low fertility is caused by incoherence between the models of the family assumed by different family-oriented social institutions. The male breadwinner model, in which the husband works and the wife takes care of the children, is contrasted with the gender equity model, in which gender does not determine which partner does which type of work (McDonald, 2000). Then, if social institutions in education and the labour market, which presume a gender equity model, coexist with traditional family values and expectations, which assume a breadwinner model, women are likely to postpone or forgo having a child given conflicting expectations on career and childbearing aspirations (McDonald, 2013). Moreover, there is evidence that women’s dissatisfaction with housework balance has been increasing over time, despite housework balance becoming more equal (Leopold, 2019).

Micro-level empirical evidence testing gender equity theory is mixed (Raybould & Sear, 2020). Prior work has shown that more unequal distributions of household labour reduce the likelihood of having an additional child across a number of Western societies and East Asian countries (Brinton & Lee, 2016; Matthews, 1999; Nagase & Brinton, 2017; Torr & Short, 2004). Torr and Short (2004) find that American couples in which women do less than 54 per cent of housework are 253 per cent more likely to have another birth within 5 years than couples in which women do 54–84 per cent of the housework. In Germany, Cooke (2004) finds that husband’s percentage share of housework has no effect on the likelihood of a second birth, but that husband’s percentage share of childcare increases the likelihood by one percent per percentage point of childcare share. Similarly, Cooke (2009) finds that husband’s percentage share of childcare increases the likelihood of a second birth by five per cent per percentage point of childcare share in Italy, but finds no association in Spain. In Japan, Kato et al. (2018) find that the division of childcare influences parity progression, but that there is no consistent effect of housework division. Raybould and Sear’s (2020) systematic review of gender equity theory finds 33 micro-level articles testing the effect of domestic labour balance on various measures of realised fertility. Nine of these studies report null effects, eleven find negative relationships, and fifteen find positive relationships.

Effects by educational and employment subgroups

Gender equity theory argues that low fertility in high-income societies can be explained by incoherence between the models of the family assumed by different social institutions. However, evidence suggests that different population subgroups experience differential impacts of gender equity on fertility, with highly educated and working-mother couples experiencing a greater impact of domestic labour balance on fertility.

Across Chinese societies, individuals with higher levels of education tend to hold more egalitarian beliefs about domestic gender roles. Qian and Li (2020) investigate survey data from China, in which respondents gave their attitude towards domestic gender roles on a scale from 1 to 5 (perfectly inegalitarian to perfectly egalitarian). On average, women with a senior high school education reported scores that were 0.561 (p < 0.001) points higher than women with a primary school education or lower (the reference group), and women with a university education reported scores that were 0.767 (p < 0.001) points higher. Also in China, Du et al. (2021) find that for each extra year of education, men and women are 4.6 percentage points likelier to hold egalitarian beliefs about housework. Pooling data on attitudes in both China and Taiwan, Yang (2016) finds that individuals with a higher education are 2.603 (p < 0.001) times as likely to believe that wives should work, than individuals with a primary education or below. Given that highly educated couples hold more egalitarian domestic gender-role attitudes, it seems likely that the fertility of such couples will be more sensitive to the gender division of household labour. We, therefore, expect that changes and differences in domestic labour balance will have more of an effect on the fertility of highly educated couples, than on couples with lower levels of education.

Another population subgroup whose fertility may be more sensitive to domestic labour balance is working-wife couples. In high-income societies, theorists have argued that fertility will be low when women face a ‘dual burden’ of both paid and unpaid labour obligations—for a recent overview, see Raybould and Sear (2020). At the individual level, women who are formally employed will have less time to perform housework and childcare. Interviews with childbearing-aged men and women in South Korea and Japan reveal that wives see a trade-off between full-time employment and birth intentions beyond the first child, and perceive a need to partly withdraw from the labour market in order to have more children (Brinton & Oh, 2019). In the UK, Schober (2013) uses a Cox proportional hazards model to find that women’s probability of a second birth changes by a factor of 0.982 (p < 0.001) per hour of paid work.¹ Overall, it, therefore, seems likely that the fertility of women with jobs will be more sensitive to the balance of domestic labour, compared with women who remain at home.

Combining the evidence accumulated here, it seems likely that the fertility of higher-educated and working-mother couples will be more sensitive to the domestic balance of domestic labour. We, therefore, expect that couples that are both highly educated, and in which the wife is working, to be most sensitive to the effect of domestic labour balance on fertility.

The Taiwanese context

In Taiwan, low fertility has been attributed to a tension between family life and public life. In the family, traditionally gendered childcare and housework roles have persisted, while roles in education and employment have become much more gender equal (Frejka et al., 2010; McDonald, 2009; Raymo et al., 2015; Tu et al., 2017). Female higher education has expanded rapidly since the mid-1990s, reflecting a shift in women’s opportunities for career progression (Cheng & Loichinger, 2017), and a postponement of marriage and childbearing (Chen & Chen, 2014). The share of 18–21-year-old women enrolled in tertiary education increased from 40 per cent in 1995 to over 85 per cent in 2006, and to 89 per cent in 2017 (Chen, 2016; Chen et al., 2013; Ministry of Education [MOE], 2018). Evidence shows that increasing female education represents an important factor in contemporary fertility decline by encouraging career aspirations for a greater proportion of women, and by increasing women’s childbearing decision-making power within couples (Chen, 2016; Hu & Yeung, 2019). At the same time, rates of female labour force participation have been increasing. From 1987 to 2010, labour participation rates increased from 56 per cent to 84 per cent for women aged 25–29, and from 55 per cent to 77 per cent for women aged 30–34 (Cheng & Loichinger, 2017). In contrast to increasing gender equality in education and employment, traditional family values and expectations have been slow to change (Raymo et al., 2015). Wives continue to bear most of the housework in Taiwan: estimates for the proportion of housework done by wives range from 72 per cent to 81 per cent (Hu & Kamo, 2007; Kim, 2013; Yu & Xie, 2011). Moreover, a recent study has found that years of education has no impact on the domestic labour participation of married women with children (Kolpashnikova & Koike, 2021). In a context of increasing career aspirations, the expectation of a heavy and unequal domestic workload seems to have resulted in the postponement or renouncement of having children by many Taiwanese women (Hori, 2017; Qian & Sayer, 2016).

Empirically, there are three quantitative studies that assess the effect of domestic labour balance on fertility desires and intentions in Taiwan, and one qualitative study that investigates Taiwanese parents’ perceptions of barriers to childbearing. Kan and Hertog (2017) focus on the cross-sectional impact of housework division on desired fertility and show that women whose husbands do more housework have a greater desire for more children. Cheng and Hsu (2020) evaluate how childcare and housework balance impacts fertility intentions among parity 1 + couples. Results show that childcare balance has an effect on intentions but housework balance does not, and that childcare imbalance has a larger impact on intentions for higher-educated couples. By contrast, Kan et al.’s (2019) study shows housework balance to have an impact on birth intentions, with more equal divisions being positively associated with intentions. Freeman et al. (2018) conducted 32 interviews with Taiwanese parents and found gendered childcare obligations to be a salient factor inhibiting further childbearing. Despite the evidence on the effect of domestic labour balance on fertility desires and intentions, and evidence on Taiwanese parents’ perceptions, there are no quantitative studies that assess the effect of domestic labour balance on realised fertility. Therefore there is no direct quantitative evidence that domestic labour balance has an impact on births.

In line with prior empirical analyses, we choose to focus on births of second or higher order—i.e., effects on having an additional child—for three key reasons. Firstly, second births are qualitatively different from births of first order, since two-child families are normative in modern societies, and so second births represent family building rather than family formation (Torr & Short, 2004; Yoon, 2016). Secondly, the number of second- and higher-order births have declined and this reduction is identified as a main driver of low fertility in Taiwan, and so transitions to second and higher-order births merit analysis for understanding trends in aggregate fertility (Cooke, 2009; Goldscheider et al., 2013; Nagase & Brinton, 2017). Thirdly, having a child introduces new forms of required domestic labour (e.g., childcare and child-specific housework). The additional burden means the division of household labour for those with one child is likely to have a stronger effect on the likelihood of a next birth, compared to the division of household labour for those with no children (Cooke, 2009; Nagase & Brinton, 2017).

Data

We draw on data from the Taiwanese Panel Survey of Family Dynamics² (PSFD, 2018), a nationally representative, longitudinal survey gathering data on household relationships and behaviours, including information on respondents’ and their partners’ age, housework, education, and income. We use data from the 2010, 2011, 2012, 2014 and 2016 waves of the PSFD. Respondents are selected by a stratified three-stage sampling procedure using household registration data. The observational unit in the PSFD is individual adult men and women, and for a given household only one person in that household participates in the PSFD. However, the PSFD asks detailed questions about respondents’ spouses (if the respondent has a spouse), so we do not need to restrict our sample only to female respondents. The PSFD gathers information on spouses by asking respondents, not by asking spouses directly.

Respondents in our data extract had an initial response rate of 47.33 per cent and 49.86 per cent respectively, which is typical for longitudinal surveys in developed countries (Rindfuss et al., 2015). Drawing on Rindfuss et al.’s review (2015), we assume these low response rates do not significantly bias our inferences. In that review, the authors evaluate the impact of response rates in a longitudinal survey of Japanese families in the 2000s (with a response rate just over 50 per cent). The authors find that, while response rates vary with demographic and socioeconomic variables, there is no effect of non-response on the relationship between socioeconomic or demographic predictors and fertility. Because the PSFD data is also longitudinal, has a comparable response rate to the Japanese study, is also from East Asia, and covers the 2000s and 2010s, we assume that the relationships between predictors and fertility in the PSFD data are unaffected by their response rates. Under this assumption, the response rates do not risk biasing statistical associations between housework and fertility.

We restricted our analytical sample to married³ heterosexual couples, with one or more children, and in which the wife was between the ages of 20 and 40 in 2010. In terms of participation in the waves, we only required that the respondent participated in the 2010 and 2016 waves. These restrictions yielded an analytical sample of 587 respondents (289 female-respondent couples and 298 male-respondent couples).

Methods

In this section, we first describe the variables used in the modelling, and then describe the model specifications for the main models and subgroup models.

Variables

Our key explanatory variable is the proportion of housework done by husbands. Housework is measured by the total hours of housework dedicated per week by each spouse and includes activities such as cooking, cleaning, and laundry, but not childcare. We focus on the share of housework done by each spouse, because domestic gender equity concerns whether the balance of household labour is considered fair, rather than concerning the absolute amount of housework. Changes in individual perceptions of fairness tend to operate according to changes in personal circumstances (Perales et al., 2015). Husband’s proportion of housework is calculated using the following formula: $share={\displaystyle \frac{hhw}{hhw+whw},}$where share gives the husband’s proportion of housework, hhw gives the husband’s weekly hours of housework, and whw gives the wife’s weekly hours of housework. Husband’s weekly hours of housework and wife’s weekly hours of housework were also included as variables in our models, in order to control for an absolute number of hours of housework done by each couple.

We included a range of control variables for variations in husbands’ and wives’ income, education, work, age, pregnancy, and parity (i.e., birth order). In terms of parity, our analytical sample is restricted to couples at parity 1 or above. Since two-child families are normative, we include a binary control variable to control for families at parities of two and over. This is because families with one child are likely to face an additional incentive to have another child (because of the two-child norm), relative to families who already have two or more children. We also included a binary variable indicating whether the respondent (or the respondent’s wife) was pregnant at the time of the survey.

Table 1 lists and provides summary statistics for all variables. Table 1 indicates that husbands do about 29.8 per cent of housework on average. The proportion of couples that had one or more births from 2011 to 2015 was 0.286, a total of 587*0.286 = 168 couples.

Table 1. Summary statistics for regression variables.

Variable	Mean	Standard deviation
Birth in following year (binary)	0.0671	0.250
One or more births in 2011–2015 (binary)	0.286	0.452
Youngest child younger than five	0.520	0.500
Wife or husband desires another child (binary)	0.187	0.390
Tertiary education—wife (binary)	0.478	0.500
Tertiary education—husband (binary)	0.469	0.499
Proportion of housework done by husband	0.298	0.207
Wife’s weekly hours of housework	16.5	13.4
Husband’s weekly hours of housework	7.23	8.45
Wife’s age in years	34.1	4.03
Wife works (binary)	0.737	0.441
Any parents live with couple (binary)	0.484	0.500
Income (New Taiwan Dollars)	62,800	54,800
Parity two or more (binary)	0.702	0.457
Pays for a nanny (binary)	0.151	0.359
Year of the Dragon (binary)	0.25	0.433
Wife is pregnant at the time of the survey	0.0616	0.240

Note: n = 587 couples.

The 145 births in the four years 2011, 2012, 2013, and 2015 are negatively distributed over time, with 56 births in 2011, 43 births in 2012, 34 births in 2013, and 12 births in 2015. This pattern is to be expected since all individuals in the sample have at least one child in 2010, and couples are not likely to have more than two children.

Main models

For our main analysis, three separate models are estimated to assess the impact of housework imbalance on fertility. One model uses data on all respondents, one model only uses data from female respondents, and one model only uses data from male respondents. We disaggregated our analysis by the gender of the respondent to account for potential biases in the reporting of hours of housework. This was done for two reasons. Firstly, the reporting of hours of housework over the previous week is to some extent a subjective measure, relying on what the respondent regards as housework, what housework they remember doing, and marital satisfaction at the time of the questionnaire (Kamo, 2000; Tao, 2013). Consequently, the reporting of housework may be subject to random error, which may co-vary with the gender of the respondent. Secondly, there is evidence to suggest that wives and husbands may have different perceptions of their relative contributions to housework, and so reported hours of housework may vary systematically the gender of the respondent (Kiger & Riley, 1996). In the US, Kamo (2000) finds that husbands tend to overestimate their own contribution to housework, whereas wives accurately estimate their housework contribution. By contrast, Press and Townsley (1998) find that both husbands and wives overestimate their own housework contribution, again in the US. Given the potential for respondents’ gender to impact reporting of housework contributions, we choose to disaggregate respondents by gender.

The models are estimated using a specification which does not use the panel structure of the data. This specification aims to capture the effect of housework on fertility by using independent variables measured in 2010 to explain the birth of a child between 2011 and 2015. Our outcome variable is whether or not a couple has a birth at any time over the next five years; equivalently, whether or not a couple has had a birth by the end of 2015.

We use the proportion of husband’s housework in a given year to predict the probability of a birth over subsequent years (Dommermuth et al., 2017). This measure may have the drawback of representing domestic gender equity at a single point in time. However, any alternative measure—such as the mean proportion of husbands’ housework over 2010–2015—produces problems of endogeneity. We evaluate the robustness of our approach by repeating the analysis using data from 2011 on husband’s proportion of housework, and data on fertility outcomes from 2012 to 2015. Three models (Models (1)–(3)) are estimated based on the following equation: (1) $\ln \left({\displaystyle \frac{Pr(y_{i,2011-2015}=1|{\mathit{x}}_{i,2010})}{1-Pr(y_{i,2011-2015}=1|{\mathit{x}}_{i,2010})}}\right)={\mathit{x}}_{i,2010}^{{}^{\mathrm{\prime }}}\mathit{\beta }.$(1)

In Equation (1), $y_{i,2011-2015}$ is a binary variable that equals 1 if a couple had one or more children from 2011 to 2015; ${\mathit{x}}_{i,2010}$ is the vector of independent variables from 2010; and, $\mathit{\beta }$ is its associated vector of coefficients. In vector $\mathit{\beta }$, the main coefficient of interest is the proportion of housework done by husbands. We expect our estimate for this coefficient to be positive, indicating that husbands’ proportion of housework has a positive effect on realised fertility.

Subgroup models

As discussed in the subsection ‘Effects on Educational and Employment Subgroups,’ we expect that the fertility of highly educated couples in which the wife is working will be more sensitive to the effect of domestic labour balance. Therefore, for each of the main Models (1)–(3), we evaluate six subgroup models, yielding 18 subgroup models in total. For each of the subgroup models, the specification is also given by Equation (1), but the sample is restricted in various ways. For each of the main Models (1)–(3), the six subgroups are as follows: working-mother couples; tertiary-educated couples; working-mother and tertiary-educated couples; and the complements of these three subgroups. We index each of the subgroup models with a trailing .1–.6; for example, Model (2.2) corresponds to female-respondent, tertiary-educated couples.

Results

Main models

Models (1)–(3) in Table 2 report the estimates for the effects of housework balance on fertility. Only key variables and significant variables are displayed in Table 2. Full results are displayed in Table A1 in the Appendix.

Table 2. Selected log-odds results from models predicting births, by survey respondent gender.

Variable	(1)	(2)	(3)
	All	Females	Males
Proportion of housework done by husband	0.813 (0.977)	2.228 (1.403)	−1.383 (1.481)
Wife's hours of housework	−0.0296 (0.0210)	0.00124 (0.0286)	−0.0702* (0.0274)
Husband's hours of housework	0.00860 (0.0248)	−0.0230 (0.0326)	0.0612 (0.0385)
Wife's age	−0.251*** (0.0498)	−0.320*** (0.0887)	−0.242*** (0.0653)
Parity 2 or more	−2.148*** (0.284)	−1.844*** (0.378)	−2.642*** (0.448)
Desires more children	0.655 (0.407)	0.378 (0.567)	0.699 (0.615)
Gender	−0.115 (0.249)
Wife pregnant at time of survey	0.0296 (0.390)	−0.226 (0.565)	0.147 (0.588)
Constant	6.264*** (1.749)	7.907** (2.788)	3.808 (4.279)
N	587	289	298

Notes: The dependent variable here is whether the couple went on to have one or more children in 2011–2015. Estimates of the log-odds and standard errors (in brackets) are reported to three decimal places.

+p < .10.

*p < .05.

**p < .01.

***p < .001.

The estimates for husband’s share of housework are not significant in any model, although the estimate for female respondents is larger than for males and all respondents. In terms of control variables, the wife’s age and the couple being at a parity of two or more were significant and negative across Models (1)–(3). The negative coefficient for wife’s age reflects a high median wife’s age of 32 in 2010, with half of the women in the sample being over 37 by 2015. As 37 is towards the end of most women’s childbearing years, these older women would have likely had less children over 2011–2015 than women who were younger in 2010, reflecting wife’s age as a negative predictor of fertility in our models. The coefficients for parity of two or more are negative, large, and highly statistically significant. Overall, these control variable coefficients suggest that couples with older wives and with two or more children in 2010 were less likely to have another child across 2011–2015.

Subgroup models

To investigate whether the effect of housework balance on fertility is different for different subgroups of respondents, we re-ran Models (1)–(3), but restricted the sample to couples in which the mother was working, couples in which both partners were tertiary educated, and couples in which both the mother was working and both partners were tertiary educated. We also ran models on the complements of these subgroups. The coefficient for the proportion of husbands’ housework in each of these models is displayed in Table 3. Full model results for the working-mother, tertiary educated, and working-mother and tertiary-educated models are displayed in Tables A2, A3 and A4 in the Appendix. Full model results for the complements of these subgroups are available on request from the authors.

Table 3. Estimates of ‘proportion of husband’s housework’ coefficients in models predicting births, by survey respondent gender and employment and education subgroups.

All
	(1.1)	(1.2)	(1.3)
	Working-mother couples	Tertiary-educated couples	Working-mother and tertiary-educated couples
Log-odds	1.780+ (0.975)	3.638** (1.366)	3.433* (1.516)
Increase in probability of a birth	0.069	0.136	0.121
N	419	248	196
	(1.4)	(1.5)	(1.6)
	Non-working-mother couples	Non-tertiary-educated couples	Non-(working-mother and tertiary-educated couples)
Log-odds	−2.253 (1.975)	−0.634 (1.237)	−0.432 (1.099)
N	163	339	391
Females
	(2.1)	(2.2)	(2.3)
	Working-mother couples	Tertiary-educated couples	Working-mother and tertiary-educated couples
Log-odds	3.265* (1.434)	6.450* (2.528)	8.428** (3.131)
Increase in probability of a birth	0.138	0.283	0.336
N	210	129	104
	(2.4)	(2.5)	(2.6)
	Non-working-mother couples	Non-tertiary-educated couples	Non-(working-mother and tertiary-educated couples)
Log-odds	−1.366 (2.666)	1.335 (1.716)	1.587 (1.540)
N	76	160	185
Males
	(3.1)	(3.2)	(3.3)
	Working-mother couples	Tertiary-educated couples	Working-mother and tertiary-educated couples
Log-odds	−0.199 (1.412)	−0.379 (2.338)	−0.168 (2.041)
N	209	119	92
	(3.4)	(3.5)	(3.6)
	Non-working-mother couples	Non-tertiary-educated couples	Non-(working-mother and tertiary-educated couples)
Log-odds	−23.48 (17.77)	−1.983 (2.626)	−2.680 (2.258)
N	87	179	206

Notes: The dependent variable here is whether the couple went on to have one or more children in 2011–2015. Estimates of the log-odds and standard errors (in brackets) are reported to three decimal places. The ‘Increase in probability of a birth’ rows refer to the increase in probability of having a birth for a couple with the mean division of housework (for that particular subgroup), to having an equal division of housework. The ‘Increase in probability of a birth’ is not reported for log-odds values that are not significant at any level.

+p < .10.

*p < .05.

**p < .01.

***p < .001.

The results in Table 3 reveal that the proportion of husband’s housework has a significant and positive effect on fertility for working-mother and tertiary-educated couples, and particularly when the respondent is female. The largest effect is found in Model (2.3), for female-respondent, working-mother and tertiary-educated couples. That model reports a coefficient of 8.428**. We can understand the meaning of this number by considering a typical couple in the 104 respondents in Model (2.3). The mean proportion of housework done by husbands amongst these respondents is 0.331. Were this proportion to move to an equal balance of housework—i.e., were it to increase to 0.5, an increase of 0.169—the odds of that couple having a child would increase by a factor of $\exp (0.169\ast 8.428)=\mathrm{4.16.}$ This means that the odds of a couple with an equal balance of housework having a birth is roughly 4 times as high as the odds of a couple with the mean balance of housework having a birth. Similar calculations for Models (2.1) and (2.2) give factors of 1.87 and 3.21 respectively, meaning that the odds of housework-equal couples having a birth are roughly 2 and 3 times as high as the odds of average couples having a birth (for female-respondent couples).

Table 3 also presents the increases in probabilities of couples having a birth, if they move from having the mean division of housework to having an equal division of housework. We estimate the probability of an average couple having a birth by the proportion of couples in that particular model that had a birth across 2011–2015; for example, 0.394 of couples in Model (2.3) had a birth from 2011 to 2015, and so we take 0.394 as the probability of a couple with the mean division of housework having a child from 2011 to 2015. Using the fact that the odds of a birth for gender-equal couples is 4.16 as high as average couples, we can calculate the probability of a gender-equal couple having a birth as 0.703, representing an increase of 0.336.

Despite the strong effects found in Models (1.1)–(1.3) and (2.1)–(2.3), when the analysis was repeated for data in 2011 (predicting one or more births from 2012 to 2015), no such effects were found. This suggests that analysing the effect of domestic gender equality on fertility over longer periods is likelier to find a significant result and that such effects may not be evident over time periods shorter than five years. It could also be the case that there was an insufficient proportion of couples having children at all over 2012–2015: while 28.6 per cent of couples had at least one additional child between 2011 and 2015, only 19.6 per cent of couples had at least one additional child between 2012 and 2015. In this way, the failure to replicate the 2010 results with data from 2011 could be due to the shorter time period or could be due to an insufficient proportion of the sample having children to ensure enough variability to identify an effect, or could be due to a combination of these two factors.

Discussion

In this section, we discuss our model results, explore the implications of our findings for research and future population trends in Taiwan, and identify some key implications for policies aiming to increase fertility.

Our results suggest that domestic labour balance affects fertility, providing evidence for gender equity theory in Taiwan. We have positive and significant estimates for the proportion of husbands’ housework in Models (1.1)–(1.3) and (2.1)–(2.3).

Beyond testing gender equity theory, our models also provide evidence of two additional findings on the relationship between housework division and childbearing in Taiwan. Firstly, we find that the impact of housework division on fertility is only evident for tertiary-educated or working-mother couples, and is especially strong for tertiary-educated and working-mother couples. As discussed in the Literature Review, better-educated women tend to have higher career aspirations and more bargaining power within couples. Working mothers face a dual burden of formal and domestic labour, and so any reduction in their domestic labour obligations should be expected to facilitate their childbearing decisions. In this sense, our findings are consistent with theoretical expectations. As female labour force participation and tertiary education increase rapidly in Taiwan, our results would seem to imply that aggregate fertility trends will become increasingly sensitive to the domestic labour balance. Secondly, we find that the gender of the respondent is crucial to identifying an effect. This suggests that perceptions of how much each partner contributes to housework vary significantly by gender, with wives perceiving husbands to be doing a smaller share of housework than husbands perceive themselves to be doing (Cerrato & Cifre, 2018). For each working-mother and tertiary-educated subgroup defined here, male respondents consistently report a higher share of husband’s housework than female respondents. For example, the (female) respondents in Model (2.3) report that their husbands do 33.1 per cent of the housework on average, whereas the (male) respondents in Model (3.3) report themselves as doing 38.9 per cent of the housework. Since women do more housework than men in general, we would expect the proportions reported by female respondents to be more accurate than those reported by male respondents. Furthermore, for the purposes of determining whether the couple has another child, wives’ perceptions of the balance of domestic labour are more important than husbands’ perceptions. This is because the wife is the one who ultimately decides whether or not to have a child; moreover, as women in Taiwan are becoming better educated, their decision-making power relating to childbearing decisions appears to be increasing (Hu & Yeung, 2019).

We may not be able to observe effects for male-respondent couples due to biases in husbands’ reporting of their and their wives’ housework hours. To some extent, this is an unavoidable consequence of the housework measure being used—i.e., a retrospective estimate of a respondent’s (and their spouse’s) hours of housework over the preceding week—and presents a limitation on the findings of this study. For future research, more accurate data collection instruments—such as time-use diaries—could enable more reliable evaluations of the impact of housework balance on fertility (Kan, 2008).

In policy terms, our results suggest that encouraging the equalisation of household labour could play a role in increasing fertility. The Taiwanese government first introduced pronatalist policies in 2006 and 2008, which focused on financial support and childcare for couples with young children (Frejka et al., 2010; Lee, 2009; Lee & Lin, 2016; Yip & Chen, 2016; Yu-Hua, 2012). Parental leave of up to two years was introduced slightly earlier, but only became paid (60 per cent of the parent’s salary during the first six months) from 2008 (Lee & Lin, 2016; Tsai, 2012). Fathers are entitled to three days of paternity leave on full pay and are also entitled to take parental leave (Tsai, 2012). Promoting domestic gender equity directly has not been a part of any pronatalist policy so far—for a recent overview, see Yip and Chen (2016).

In Scandinavian countries, paternity leave has long been regarded as a means of promoting domestic gender equality (Cools et al., 2015; Kotsadam & Finseraas, 2011). These countries are also characterised by stable and near-replacement level fertility; however, causal links between policies and fertility are potentially confounded by gender-equitable attitudes across these societies at large. Nonetheless, there is some evidence—both in Scandinavia and in other developed countries—that the availability and uptake of paternity leave has a positive effect both on subsequent domestic gender equality, and on second births (Boll et al., 2014; Kotsadam & Finseraas, 2011; Lappegård, 2010). For Taiwan, long periods of paid paternity leave could potentially be a viable means of promoting fertility. Providing longer periods of better-paid paternity leave could encourage fathers to perform a larger share of housework and childcare, leading to more gender-equitable outcomes at home and subsequent increases in fertility. However, having additional family policies available may not impact domestic life, unless the utilisation of those policies is seen as acceptable by employers (Kim & Parish, 2020).

Conclusion

Taiwan has the third lowest TFR in the world and has remained persistently low over the last two decades. Identifying the key factors underpinning these patterns is important to develop effective ways to promote fertility. In this paper, we sought to establish whether the division of household labour exerts a significant influence on the probability of having another child. Drawing on panel data from the PSFD, we examined the effects of housework division on births.

The evidence accumulated here shows that men doing a greater share of domestic labour results in a higher probability of having an additional child. Specifically, couples in which husbands do a greater share of housework are more likely to have an additional child in the next five years than couples in which wives take a larger responsibility for domestic work. However, this finding is only true for working-mother or tertiary-educated couples and is driven by female respondents. Our results indicate that among female-respondent, working-mother, and tertiary-educated couples, if the husband increases his share of housework from a sample mean of 0.331–0.5, the probability of having a birth over a five-year period increases from 0.394 to 0.730.

These results comprise the first empirical support for the validity of gender equity theory as an explanation for fertility behaviour in Taiwan. Our conclusions are consistent with existing evidence in the West and elsewhere in East Asia (Balbo et al., 2013; Kan & Hertog, 2017; Oláh, 2003). Yet, the magnitude of the effect found in this study is larger. Further research is needed to establish whether the effects of domestic gender equity on fertility are greater in Taiwan—and more widely across East Asia—than in the West.

Our findings suggest that governments seeking to raise fertility should focus on policies aiming at promoting greater domestic gender equality, such as increased paternity leave and childcare provision. Existing evidence points to the effective use of paternity leave policies seeking to increase fertility in Scandinavian countries. However, research is needed to assess if recent increases in parental and paternity leave in East Asia are having such an effect. At the same time, policies should also be targeted at teaching gender equality. Children could be taught at school about the importance of sharing all forms of labour equally between the sexes. Such policies are standard in many near-replacement fertility developed countries, such as in Sweden, where pre-school gender equality teaching projects have been implemented since the 1990s (Bayne, 2009).

Data availability statement

Data and other materials of the Panel Survey of Family Dynamics (PSFD) are available free of charge to researchers. Materials are available from the Survey Research Data Archive (SRDA) of the Academia Sinica, Taiwan. For details, see https://srda.sinica.edu.tw/.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The figure of 0.982 is calculated by exp(−0.081), with −0.081 being the coefficient given in the Cox proportional hazards model results.

2 https://psfd.sinica.edu.tw/web/plan_01en.htm

3 Childbearing in Taiwan still overwhelmingly occurs within marriages, which is why we restrict our analysis to married couples (Raymo et al., 2015).

Appendix

Table A1. Full results from models predicting births, by survey respondent gender.

Variable	(1)	(2)	(3)
	All	Females	Males
Youngest child younger than five	0.464 (0.343)	0.544 (0.459)	0.306 (0.547)
Wife's tertiary education	0.896** (0.335)	0.648 (0.481)	1.158* (0.507)
Husband's tertiary education	0.190 (0.314)	0.411 (0.440)	0.118 (0.503)
Proportion of housework done by husband	0.813 (0.977)	2.228 (1.403)	−1.383 (1.481)
Wife's hours of housework	−0.0296 (0.0210)	0.00124 (0.0286)	−0.0702* (0.0274)
Husband's hours of housework	0.00860 (0.0248)	−0.0230 (0.0326)	0.0612 (0.0385)
Wife's age	−0.251*** (0.0498)	−0.320*** (0.0887)	−0.242*** (0.0653)
Wife works	−0.781 (0.518)	−0.674 (0.798)	−0.633 (0.690)
Parents co-reside	0.0510 (0.286)	0.0787 (0.410)	0.00863 (0.417)
Log income	0.0705 (0.0651)	0.0429 (0.0810)	0.339 (0.375)
Parity 2 or more	−2.148*** (0.284)	−1.844*** (0.378)	−2.642*** (0.448)
Pays for nanny	0.576 (0.402)	0.284 (0.597)	0.883 (0.631)
Desires more children	0.655 (0.407)	0.378 (0.567)	0.699 (0.615)
Parents co-reside * pays for nanny	−0.120 (0.572)	0.0180 (0.764)	−0.209 (0.939)
Wife works * wife's hours of housework	0.0262 (0.0242)	0.0137 (0.0352)	0.0129 (0.0300)
Desires more children * wife's hours of housework	0.0214 (0.0182)	0.0351 (0.0282)	0.0268 (0.0251)
Gender	−0.115 (0.249)
Wife pregnant at time of survey	0.0296 (0.390)	−0.226 (0.565)	0.147 (0.588)
Constant	6.264*** (1.749)	7.907** (2.788)	3.808 (4.279)
N	587	289	298

Notes: The dependent variable here is whether the couple went on to have one or more children in 2011–2015. Estimates and standard errors (in brackets) are reported to three decimal places.

+p < .10.

*p < .05.

**p < .01.

***p < .001.

Table A2. Full results from models predicting births for all respondents, by employment and education subgroup.

Variable	(1.1)	(1.2)	(1.3)
	Working-mother couples	Tertiary-educated couples	Working-mother and tertiary-educated couples
Youngest child younger than five	0.476 (0.409)	0.478 (0.572)	0.318 (0.612)
Wife's tertiary education	0.929* (0.472)
Husband's tertiary education	0.517 (0.408)
Proportion of housework done by husband	1.780+ (0.975)	3.638** (1.366)	3.433* (1.516)
Wife's hours of housework	0.00986 (0.0223)	−0.0176 (0.0743)	0.0156 (0.0403)
Husband's hours of housework	−0.0157 (0.0262)	−0.0506 (0.0386)	−0.0643 (0.0436)
Wife's age	−0.240*** (0.0613)	−0.423*** (0.107)	−0.398*** (0.116)
Parents co-reside	−0.148 (0.342)	−0.0417 (0.461)	0.156 (0.466)
Log income	0.150 (0.102)	−0.348 (0.912)	−0.786 (0.809)
Parity 2 or more	−2.356*** (0.351)	−2.573*** (0.436)	−2.864*** (0.501)
Pays for nanny	0.295 (0.460)	0.372 (0.609)	0.500 (0.666)
Desires more children	0.678 (0.483)	0.902 (0.577)	1.089+ (0.643)
Parents co-reside × pays for nanny	0.254 (0.645)	0.0732 (0.932)	−0.200 (1.075)
Desires more children × wife's hours of housework	0.0287 (0.0232)	−0.00760 (0.0341)	0.0220 (0.0353)
Wife pregnant at time of survey	−0.131 (0.497)	−0.254 (0.519)	−0.787 (0.607)
Wife works		−0.166 (1.097)
Wife works × wife's hours of housework		0.0432 (0.0697)
Constant	3.711+ (2.027)	16.42+ (9.070)	20.70* (8.499)
N	419	248	196

Notes: The dependent variable here is whether the couple went on to have one or more children in 2011–2015. Estimates and standard errors (in brackets) are reported to three decimal places.

+p < .10.

*p < .05.

**p < .01.

***p < .001.

Table A3. Full results from models predicting births for female respondents, by employment and education subgroups.

Variable	(2.1)	(2.2)	(2.3)
	Working-mother couples	Tertiary-educated couples	Working-mother and tertiary-educated couples
Youngest child younger than five	0.956+ (0.575)	1.204 (0.789)	1.958* (0.995)
Wife's tertiary education	0.752 (0.721)
Husband's tertiary education	1.016 (0.678)
Proportion of housework done by husband	3.265* (1.434)	6.450* (2.528)	8.428** (3.131)
Wife's hours of housework	0.0381 (0.0276)	0.0426 (0.0884)	0.0784 (0.0737)
Husband's hours of housework	−0.0649+ (0.0345)	−0.134+ (0.0732)	−0.222* (0.0892)
Wife's age	−0.291* (0.123)	−0.517* (0.226)	−0.505* (0.204)
Parents co-reside	0.212 (0.527)	−0.170 (0.691)	0.498 (0.862)
Log income	0.112 (0.160)	−0.520 (1.432)	−2.172+ (1.251)
Parity 2 or more	−2.525*** (0.546)	−3.006*** (0.616)	−4.634*** (1.059)
Pays for nanny	0.312 (0.756)	0.364 (0.965)	2.034 (1.338)
Desires more children	0.545 (0.698)	−0.0175 (0.904)	0.484 (1.140)
Parents co-reside × pays for nanny	−0.0506 (0.928)	0.407 (1.239)	−1.209 (1.490)
Desires more children × wife's hours of housework	0.0514 (0.0332)	−0.0209 (0.0484)	0.0455 (0.0547)
Wife pregnant at time of survey	−0.620 (0.721)	−0.170 (0.777)	−1.172 (0.976)
Wife works		0.102 (1.716)
Wife works × wife's hours of housework		0.0278 (0.0832)
Constant	4.509 (3.362)	20.21 (12.39)	37.49** (13.15)
N	210	129	104

Notes: The dependent variable here is whether the couple went on to have one or more children in 2011–2015. Estimates and standard errors (in brackets) are reported to three decimal places.

+p < .10.

*p < .05.

**p < .01.

***p < .001.

Table A4. Full results from models predicting births for male respondents, by employment and education subgroups.

Variable	(3.1)	(3.2)	(3.3)
	Working-mother couples	Tertiary-educated couples	Working-mother and tertiary-educated couples
Youngest child younger than five	−0.165 (0.669)	−1.108 (0.866)	−1.464+ (0.874)
Wife's tertiary education	0.950 (0.632)
Husband's tertiary education	0.435 (0.596)
Proportion of housework done by husband	−0.199 (1.412)	−0.379 (2.338)	−0.168 (2.041)
Wife's hours of housework	−0.0527 (0.0353)	−0.220* (0.106)	−0.0927 (0.0799)
Husband's hours of housework	0.0514 (0.0396)	0.0786 (0.0753)	0.0687 (0.0628)
Wife's age	−0.259** (0.0813)	−0.505** (0.158)	−0.442** (0.150)
Parents co-reside	−0.473 (0.519)	−0.457 (0.690)	−0.447 (0.759)
Log income	0.655 (0.594)	0.798 (1.104)	0.808 (1.027)
Parity 2 or more	−2.394*** (0.495)	−2.677*** (0.726)	−2.500*** (0.742)
Pays for nanny	0.380 (0.651)	0.349 (1.010)	0.190 (0.979)
Desires more children	0.684 (0.688)	1.836* (0.935)	1.875+ (0.963)
Parents co-reside × pays for nanny	0.536 (0.935)	0.835 (1.852)	0.848 (1.747)
Desires more children × wife's hours of housework	0.0300 (0.0307)	0.0433 (0.0729)	0.0325 (0.0655)
Wife pregnant at time of survey	0.179 (0.781)	−0.515 (0.827)	−0.721 (0.943)
Wife works		−2.563 (1.574)
Wife works × wife's hours of housework		0.115 (0.105)
Constant	0.285 (6.011)	11.28 (9.630)	6.828 (9.495)
N	209	119	92

Notes: The dependent variable here is whether the couple went on to have one or more children in 2011–2015. Estimates and standard errors (in brackets) are reported to three decimal places.

+p < .10.

*p < .05.

**p < .01.

***p < .001.