Women Rule: Preferences and Fertility in Australian Households

While it is true that around 25% of couples in our estimating sample are in a de facto relationship, we use the generic terms – husbands and wives – to denote the male and the female partners, respectively.

One might wonder as to why the results using the Australian sample of couples are so different to those from other countries. There are a number of reasons for this difference. First, the datasets are from different countries; hence, it is not inconceivable that the outcomes will be quite different. Second, the way the preferences of the husband and the wife have been defined in the different papers are quite different. Thomson and Hoem (1998) use as reference category – both said that they definitely wanted to have another child. While the Thomson (1997) approach is closer to what we adopt in this paper, ultimately all preferences are aggregated to define four categories: both wanting children, husband only, wife only, and neither. We look at differences in fertility, defined by wife’s desire minus husband’s desire. Finally, the Hener’s (2010) approach is quite different and in terms of methodology cannot really be compared to what we do. Hener (2010) estimates a bargaining model of fertility: treatment group is couples with heterogeneous preferences for additional children; control group is couples with homogeneous preferences. Then, he uses the within-couple preference heterogeneity to test the predictions of the unitary and bargaining models. The approach is therefore very different from that used in this paper.

Empirical evidence often suggests that public income incentives enhance fertility (see Rosenzweig 1999; Milligan 2005; McDonald 2006; Cohen, Dehejia, and Romanov forthcoming; Laroque and Salanie 2008). From a broader perspective, it is often argued that cash transfers targeted toward women is associated with a change in household consumption patterns (Duflo 2003; Ashraf, Field, and Lee 2009) and in fertility outcomes (Eswaran 2002; Seebens 2005), due to an increase of women’s bargaining power over household decisions.

The true effects of the Baby Bonus program are open to debate. Gans and Leigh (2009) and Australian Government (2010, p. 41) find that the program has an influence on the timing of births. Parr and Guest (2011) argue that the effects of the Baby Bonus program have been quite small and that the increase in fertility rate in Australia has essentially been driven by other socioeconomic and demographic factors. On the other hand, Lain et al. (2009) find an increase in birth rates in New South Wales for the first 2 years after the introduction of the Baby Bonus, and Drago et al. (2009) show that fertility intentions rose after the announcement of the program and the birth rate is estimated to have risen moderately as a result of the program. Similarly, Langridge et at. (2010) find that, in Western Australia, the Baby Bonus has been associated with an increase in fertility rate especially among women residing in the highest socioeconomic areas. Using 19 years of birth and macroeconomic data, Sinclair, Boymal, and de Silva (2010) find a significant increase in birth numbers 10 months following the announcement of the Baby Bonus. While these studies might detect a postprogram increase in fertility, the causal effect of the program is less clearly identified due to lack of a proper control group.

We examine the robustness of our results to both these sample restrictions (i.e. dropping the couple from the estimation sample if they separate or divorce and once the wife turns 41). We come back to these issues in.Section 4.2.1

There is also attrition due to refusal to interview or other reasons. But the attrition rate is relative low for HILDA − only 6.7% per year.

Effectively, the Baby Bonus scheme replaced two existing payments: the Maternity Allowance (MA) and a “baby bonus” administered through the Australian Tax Office. The MA was a relatively modest payment (a maximum of AU$842.64 per child at the time the scheme came to an end) restricted to women who were eligible for Family Tax Benefit, and thus by extension lived in households with at most modest incomes. The existing baby bonus was, on the other hand, administered through the tax system. While being potentially much more generous (with a maximum sum of up to AU$12,500 per child available over a 5-year period following a birth) than the MA, the bonus seems to have not been widely utilized. Low utilization rates were probably due to the program functioning as a complicated and delayed tax rebate system. The most substantial payments were reserved for women with relatively high employment income in the year prior to birth, who subsequently remained out of the workforce for a total of 5 years.

As in many other developed countries, after the post World War II boom, the total fertility rate declined in Australia, from a peak of 3.5 children per woman in 1961 to 1.75 in 2003 (http://www.abs.gov.au). These conditions triggered a signicant public debate in Australia about the causes of this decline and appropriate policy responses to reverse this trend (Gray, Qu, and Weston 2008).

Along with the first announcement in May 2004, it was announced that the amount of the baby bonus would go up in the future; but the jump of $834 that happened in 2006 was unexpected in terms of the amount of the increase. Recently (in October 2012), this amount was reduced to AU $3000 for the second child and beyond, as a part of the government’s drive to reduce spending and bring the budget back to surplus.

While the age of all existing children is not included as additional covariates in the regressions, we do examine the robustness of the results to the inclusion of the age of the youngest child in the set of regressors. The parameter estimates remain qualitatively and quantitatively similar.

We deal with the right-censoring in the model by assuming that any censoring occurs randomly and is unrelated to the reason of failure. Tied failures are addressed using the method of Breslow approximation.

An alternative to using the hazard model (as in eq. [1]) would be to use a Probit model to estimate the likelihood of having an additional birth. However, one of the disadvantages of using a Probit model isthat it does not allow updating of preferences following the birth of a child without causing endogeneity problems. The solution is to use preferences at the initial year, which can be assumed as pre-determined and unaffected by subsequent birth outcomes. The predicted power of these initial preferences are however likely to be limited to the first birth. After having one birth, the couples’ fertility preferences can change and this breaks the link between initial preferences and later births. Thus, unlike the hazard model, the Probit model only allows us to estimate the average effect of initial preferences on the first birth that happens.

An indirect way to examine whether our hazard model is plagued by omitted variables bias is to check whether our main regression results are robust to various sets of covariates that feature individual charac-teristics. In unreported regressions, we conduct a number of specification tests and find that the coefficients of interest (that is, the coefficients of initial desires) are highly robust to controlling for various sets of characteristics. These results are available on request.

Our empirical results are highly robust to an alternative categorizations of the female wanting more, the female wanting less and no conflict groups based on a higher and a lower degree of Disparity in fertility desire. See discussion in Section 4.2.1 below.

To be more precise, exp(θ₁) gives the hazard ratio associated with the risk of having an additional birth for the F_more group relative to the risk for the benchmark group.

By legislation, it is the primary carer who receives the payment. In the vast majority of cases, it is the mother of the child who is the eligible person. Using the infant cohort of the LSAC dataset for Australia, Harrison et al. (2010) find that out of a sample of 5,107 infants aged between 3 and 19 months, 98% had mother as these primary carer. This proportion goes down to 97.3% for children aged between 51 and 67 months (sample size is 4,983). If it is the mother who is making the claim, she does not need to provide her spouse’s details. More information is available at: http://www.ato.gov.au/taxprofessionals/content.asp?doc=/content/38285.htm.

Note that we are testing θ₁ = −θ₂, which is equivalent to testing exp(θ₁) × exp(θ₂) = 1 in terms of hazard ratio.

In column (1), we reproduce the original results from column (1) of Table 3.

We also re-estimated eq. [1] using this restricted sample of couple (i.e. excluding couples who were separated or divorced during the period of 2002–2007. The coefficient estimates of FD^f and FD^m are very similar to those presented in column (3) of Table 2. These results are available on request.

We examine the robustness of the results presented in column (1) of Table 6 by interacting the Y04 dummy with the entire set of explanatory variables. The results (which are not presented but are available on request) remain unaffected.

For the sub-sample, the initial conditions are defined as of 2003, not 2001. In this case, instead of using eq. [5], we estimate the following regression: