|Home | About | Journals | Submit | Contact Us | Français|
This article explores socioeconomic differences in the effect of family allowances on fertility. Although several studies have examined the relationship between cash benefits and fertility, few studies have addressed the possible differential effects of cash benefits on families of different income or education levels. I reconstructed the birth histories of women in the past two Israeli censuses of 1983 and 1995 to study socioeconomic differences in the effect of family allowances up to the seventh parity. The results indicate that family allowances have a significant effect at every parity. Using female education as an indicator of socioeconomic status, I find that socioeconomic status is a significant modifier of the effect of family allowances. Family allowances seem to have a relatively large impact on more-educated women.
With fertility levels lower than ever in the developed world, some governments are looking for ways to raise fertility. Recent attempts date from September 2005, when the French government pledged more money for families with three children; and from March 2006, when the Russian President ordered parliament to more than double monthly child support payments (Population and Development Review 2006; Wyatt 2005). Although microeconomic theory predicts that cash benefits will increase fertility, empirical studies that use aggregate time-series data generally find a weak, although positive, relationship between cash benefits and fertility (e.g., Ekert 1986; Gauthier and Hatzius 1997; McNown and Ridao-Cano 2004; Zhang, Quan, and van Meersbergen 1994). The United States has not implemented such explicit policies as family allowances. However, the U.S. federal income tax code provides an exemption for each dependent child, and that code may implicitly affect the decision to have a child. Using aggregate time-series data, Whittington, Alm, and Peters (1990) found that an increase in the tax value of the personal exemption leads to an increase in the demand for children, although the elasticity of the birth rate with respect to the exemption is small. In a related article using individual-level data, Whittington (1992) substantiated the aggregate finding.
Few studies have addressed the possible differential effects of cash benefits on families of different income or education levels. Assuming that the cost of children is lower for low-income families than for high-income ones because low-income families tend to invest less in the “quality” (e.g., education) of their children and/or because the opportunity cost of mothers’ time is lower for low-income families, and assuming that family benefits are paid independently of income, rational choice models would predict family benefits to have a larger impact on the fertility of low-income families (Gauthier and Hatzius 1997:304).
Education—another component of socioeconomic status—may also moderate the effects of cash benefits. Well-educated individuals tend to have higher incomes. Hence, rational choice models would predict family benefits to have a smaller impact on the fertility of higher-educated individuals. Family benefits, however, may also have a larger impact on the fertility of higher-educated individuals. It has been widely accepted that education is conducive to low lifetime fertility. Kravdal (1992), however, reported that American and Norwegian women who received more than 12 years of schooling were more likely to give birth to a third child than were other women. A similar finding has also been reported for Sweden and Britain (Hoem 1993; Ní Bhrolcháin 1993). Freedman, Baumert, and Bolte (1959) suggested that the greater family income of more-educated individuals makes a larger family more affordable. Heiland, Prskawetz, and Anderson (2005), however, showed that more-educated individuals have a greater preference for three or more children even when household income is accounted for. Another explanation is that educated women have lower costs. Education may lower child-rearing costs, for example, because educated mothers are more efficient in child-rearing (Michael 1973:S138). Of course, the efficiency effect of education needs to be strong to compensate for the effect of education on opportunity costs. Educated women also may find it easier to combine paid work and family responsibilities because of more flexible work hours (Kravdal 2001).
In the economic theory of fertility, the time-labor cost of child-rearing looms large; and the market wage-rate of women, who are assumed to be the usual care providers, is taken as a proxy for the cost of children (Robinson 1997:68). The wage rate of women, however, may be a poor proxy for the cost of children when working hours and participation rates vary by educational level, thus making predictions about the correlation between education and opportunity costs difficult. Although higher levels of education may be associated with greater forgone earnings, they tend to be associated with fewer lost hours and years over the lifetime (Calhoun and Espenshade 1988:26). When lost time is taken into account, the opportunity costs of children may be lower for higher-educated women than for lower-educated ones. Studies of the opportunity cost of mothers’ time in Britain and the Netherlands seem to indicate that opportunity costs for higher-educated mothers are much lower than those of lower-educated ones; the loss of lifetime earnings of higher-educated mothers is much lower because of fewer lost hours and years over the lifetime (Dankmeyer 1996; Davies, Joshi, and Peronaci 2000; Joshi 2002). In Australia, women with higher levels of education lose less of their earnings because of motherhood than women at lower levels of education in proportional terms, although their dollar amounts of forgone earnings are higher (Breusch and Gray 2004). If the opportunity cost of mothers’ time is lower for higher-educated women than for lower-educated ones, then rational choice models would predict family benefits to have a larger impact on the fertility of higher-educated women, all other things being equal.
There has been little research on socioeconomic differences in the effect of family allowances. Andersson, Hoem, and Duvander (2006) found no important educational differences in the response to a premium given to Swedish women for shortening birth intervals, but their results may be due to the computation of the premium as a percentage of previous income. Using birth histories of women in the 20% samples of the 1983 and 1995 censuses of Israel, this article shows that (1) family allowances have a significant effect at every parity; and (2) female education is a significant modifier of the effect of family allowances. Family allowances seem to have a relatively large impact on more-educated women.
Population policy in Israel grew from a concern for the socioeconomic conditions of disadvantaged social groups and for the low fertility rates of the Jewish majority compared with the Arab minority (Friedlander 1974). In 1975, Israel instituted a generous non-income-tested family allowance program, replacing a complex system of benefits for children that included tax credits, small mandatory child payments by private employers, minor allowances by the National Institute of Insurance to large families, and a more substantial allowance to families of army veterans that was enacted in 1971. A notable feature of the program that started in 1975 is that the size of the allowance varies with birth order, with parents receiving minimal sums for the first two children and large sums for each fourth-born and later child; the sum received for the third child is relatively small compared with that for fourth-born and later children, but not as small as the sum received for the first two children (Manski and Mayshar 2003). Otherwise, cash benefits in Israel do not vary according to socioeconomic characteristics. They did vary, however, according to veteran status. Families received enlarged benefits if at least one of the parents, grandparents, or siblings of the child had served in the Israeli army or other security forces. In practice, this meant that all Jewish families received enlarged benefits, but large sections of the Arab minority did not (Rosenhek 1999). Beginning in 1993, the dependence of the allowance on veteran status was gradually phased out over four years. Since 1997, all children receive the full allowance (Manski and Mayshar 2003:187).
Figure 1 shows how family allowances evolved over time in New Israeli Shekels (NIS) at constant 1995 prices ($1.00 = NIS 3). The erosion of the new family allowance program, which set in immediately upon its institution in 1975 and continued until 1985, is due to inflation. In 1983, the allowance for families with four or more children was raised by approximately 50%. In 1987, the value of the family allowance was linked to the consumer price index (CPI; Mayshar and Manski 2000). After 1994, the value of family allowances did not change until the large cuts of 2003. To give a sense of the magnitude of child allowances, consider a family with six children under age 18, in which at least one of the parents, grandparents, or siblings of the children had served in the Israeli army or other security forces. In 1994, such a family would have received an annual allowance of about NIS 22,000, which is equivalent to more than $7,300.
The use of Israeli data has the advantage of family allowances not being means-tested. Thus, the allowance of each family in the sample can be determined without knowing its income. Part of the Arab minority—mostly Druze and Bedouin—serve in the army or other security forces and, hence, received enlarged benefits, but I cannot identify all of these. Hence, the Arab minority has been omitted from the analysis. The first two parities have been omitted from the analysis because the first two children received minimal sums.
Using the “own-children” method in the 20% samples of the 1983 and 1995 censuses, I reconstruct birth histories going back to 1972. A woman and her own children represent a partial birth history. Omitted are deceased children and children living elsewhere (Cho, Retherford, and Choe 1986). Infant and child mortality, however, are very low in Israel. The survival of Jewish children from birth until age 12.5 increased from 97.5% in 1970–1974 to 99% in 1990–1994 (State of Israel 1989:143; 1995:152). By limiting the reconstruction to 12 years before each census, only a very small number of children are omitted because of their living elsewhere. After age 12, some children start to attend boarding schools. Because children in boarding schools are not listed at home, fertility histories beyond 12 years would be biased. Often, it is not possible to identify the father of children of remarried and divorced women. Hence, only women who were still in their first marriage at the time of the census are included. Israel is a country of immigration. Therefore, it is important to exclude women-years spent abroad.
A comparison of the total marital fertility rate (TMFR) at age 20, obtained from the reconstruction of fertility histories using the own-children method with the TMFR computed from current statistics, illustrates the extent to which the reconstructed birth histories are biased.1 Using the proportions who are currently married by age, the current TMFR is obtained by converting the published series of age-specific fertility rates into age-specific marital fertility rates after correction for extramarital births. Unfortunately, there is no published series of the number of currently married women broken down by age before 1989 with which to convert the published series of age-specific fertility rates into age-specific marital fertility rates. Thus, it is not possible to compute TMFR from current statistics before 1989. Hence, I also present a biased estimate of TMFR that uses the proportions of ever-married women instead. I refer to the former TMFR computed from current statistics as the unbiased estimate, and to the latter as the biased estimate from current statistics. Figure 2 shows that my reconstructed series is below the unbiased TMFR computed from current statistics after 1989. The impression of a growing bias in the post-1989 period is entirely due to one year. The unbiased estimate of TMFR in 1989 should have been higher than the biased estimate. At least one possible source for the discrepancies between my reconstructed series of TMFR and that based on current statistics is the refusal of many ultra-Orthodox couples to participate in the census. This population group is known for its high fertility (Manski and Mayshar 2003). Figure 2 shows that trends in my reconstructed series and the biased series based on current statistics are very similar. Because I am mostly interested in the fertility response to family allowances rather than in estimating absolute levels of fertility, the downward bias in the level of the estimates should not influence the results to any large extent.
Figure 2 shows that marital fertility declined in the second half of the 1970s, regaining much of its initial level in the 1980s. Although it is not the aim of this article to explain long-term trends in marital fertility, the extent to which the statistical model can account for such trends is an important tool in assessing the validity of the results, as will be shown later in the article.
Family allowances are more likely to create period than cohort effects in fertility. Hence, I use only period measures of fertility (Ní Bhrolcháin 1992). Thus, the dependent variable is a variable indicating whether an i-th birth occurred in year t to a woman with i – 1 children in year t – 1. I also use aggregate measures, such as parity-specific birth rates. Infant and child mortality being very low, I estimate the birth rate for parity p in year t as the number of children born in year t and still alive at the time of the census, per 1,000 women in year t who were in their first marriage at the time of the census and who had p – 1 children born before year t and still alive at the time of the census. Figure 3 shows that all birth rates, but particularly sixth and seventh birth rates, declined in the second half of the 1970s and recovered in the 1980s.
The benefit levels shown in Figure 1 suggest that the greatest increases applied to the fourth parity, followed by the fifth to seventh, while the third parity experienced very little change. Yet Figure 3 shows that the fourth and fifth parities experienced virtually no change in birth rates over the period following the change in benefits, while the sixth and seventh parities showed substantial increases. The multivariate model presented below, however, shows that benefits also had an impact on fourth and fifth birth rates and that these birth rates would have continued to decline if not for the increase in benefits.
I measure family allowances as the addition to family income in 1995 NIS that a couple may expect in case of the birth of an i-th child in year t. Because it takes at least about one year between the decision to have another child and the actual birth, this variable is measured in year t – 1. The response to family allowances may not be constant over time. Couples may be slow to respond to a change in family allowances if information about these changes is not widely publicized in the media. To model the pace of response, I include a set of variables that measure the part of the addition to family income in the case of the birth of an i-th child in year t that is due to a change in the allowances between two consecutive years. I allow for a lagged response of up to five years. Hence, I include five variables in the analysis that measure changes in the allowance that occurred between year t – 2 and year t – 1, between year t – 3 and year t – 2, between year t – 4 and year t – 3, between year t – 5 and year t – 4, and between year t – 6 and year t – 5.2 All the allowance variables are measured in thousands of NIS.
Maternity-leave benefits that include the duration of the leave and the pay received during this period may also influence fertility. I am unable to measure these variables. There are also tax exemptions for working mothers. These cannot be taken into account, either, because they depend on the employment status and income level of the woman, which are known only for the year of the census. Government housing loans may also influence fertility. However, to the extent that these variables do not correlate strongly with the level of family allowances, omitting measures of maternity benefits, tax exemptions to family income, or housing loans from the analysis should not affect the coefficients of the allowance variables to any large extent. The duration of maternity leave was constant at 12 weeks from 1954 until 2007, when two weeks were added. Pay received by a working mother during maternity leave is 75% of her salary. There is also a one-time lump-sum payment to all mothers that amounts to 20% of the average monthly salary. Because the value of tax credits is adjusted whenever the cost-of-living allowance changes, there were no simultaneous changes in family benefits and the value of tax credits that are likely to change the interpretation of the results substantially. Neither were there simultaneous changes in family benefits and housing loans. Later in the article, I will try to show that the inclusion of a measure of maternity benefits, the value of tax exemptions, or housing loans is unlikely to improve the goodness of fit of the statistical models to any large extent.
The analysis includes three independent demographic variables: the age of the woman, marital duration, and the number of years since the last birth. Age and duration squared do not improve the goodness of fit of the models; neither do they attenuate the effects of any of the other variables in the equation (results not shown). Hence, age and duration effects were included with linear representations.
Given assortative mating, it may be difficult to estimate the separate effects of the woman’s education and that of her husband because of the correlation between the educational attainment of the woman and that of her husband. Hence, I use only female education as a measure of socioeconomic status.3 Educational attainment of the woman is entered into the analysis as two categorical dummy variables: one dummy variable indicating less than 9 years of education and the other indicating 13 or more years (postsecondary education). The reference category is 9 to 12 years of education. The educational dummy variables should be representative of the educational history because almost all Israelis with postsecondary education would have started their postsecondary education before the birth of their second child. I do not use income reported in the census because current income may not be representative of the income histories.
Jews of Oriental and North African origin tend to have higher fertility levels (Friedlander, Eisenbach, and Goldscheider 1980; Friedlander and Feldmann 1993). To take this into account, I add to the model variables for the origin of the woman and of her husband, indicating whether they were immigrants from Asia or North Africa.
In the late 1970s and early 1980s, hyperinflation may have affected reproductive behavior by causing uncertainty about the immediate future. To control for this effect, I added the natural logarithm of change in the level of the CPI that occurred in the previous year (State of Israel 1996:248).
The reconstructed birth histories are in the form of event histories. A discrete-time hazard model is used to assess the effects of the independent variables on the probability of giving birth. Because the month of birth is missing in the census, I assume that the hazard for a birth is constant within annual intervals. Following Allison (1982), I estimate discrete-time event-history models using logistic regression. This kind of analysis can accommodate two common features of event histories: censored data and time-varying variables, such as age and marital duration.
Observations in a time-series are likely to be temporally dependent. Ignoring this may produce misleading results. Following Beck, Katz, and Tucker (1998), I add the number of years since the previous event (length of the birth interval). To correct for nonlinearity in the number of years that elapsed since the previous birth, I also add a lagged dependent variable. Many women contribute more than one year to the analysis. Hence, in preliminary analyses, I added random effects to control for unobserved heterogeneity between women. I used MIXNO, a computer program for mixed-effects logistic regression, to estimate the random-effect variance (Hedeker 1999). The random-effect variance, however, could not be reliably estimated as being different from zero in any of the regression models. In this case, using a model without random effects may be warranted (Yamaguchi 1986).
One assumption made in most applications of survival analysis is that the event of interest would eventually occur if there was no censoring. This is true for death, but in the case of births, this assumption is untenable. Those cases that would never experience the event are termed long-term survivors. Ignoring the possibility of long-term survivors may yield biased parameter estimates (McDonald and Rosina 2001). To prevent women who finished their childbearing from biasing the results too much, I censor birth intervals exceeding approximately five years. The dependent variable in the statistical model is the annual log odds of giving birth. The unit of analysis is the woman-year: that is, each woman contributes as many units to the analysis as the number for which she is observed. The number of cases thus obtained for the analyses is quite large, amounting to more than one-quarter of a million woman-years in the third birth interval, more than 200,000 in the fourth interval, and almost 100,000 in the fifth interval (see Table 1). With more than 42% and 34%, respectively, the third and fourth birth intervals make up the bulk of woman-years in the analyses, but the sixth and seventh intervals with less than 6% and 3%, respectively, make only a marginal contribution.
When the sample size is large, significance values (α) much lower than the conventional ones can be appropriate. This will balance significance and power of the tests. However, a precise way of doing this is lacking (Raftery 1995).
Table 1 provides some descriptive statistics by birth interval. Note the changing composition of the population when progressing from lower parities to higher ones. The percentage of women with a postsecondary education declines with parity, but the percentage of immigrants from Asia or North Africa increases with parity. The mean ages at giving birth are compressed for the last three parities—less than nine months apart in some cases. This is most likely because of the changing religious composition of the sample; the higher the parity, the higher the percentage of Orthodox Jews.4 Orthodox couples tend to marry relatively early.
The descriptive data in Table 1 show that highly educated women in Israel have fewer higher-order children than others. This is partly due to censoring: younger and more-educated cohorts have not had a chance to reach the highest parities. The following multiple regression analyses show that in all birth intervals, women in the highest educational category are actually more likely to continue childbearing than those in the reference category.
For each birth interval, Table 2 presents the log odds (eb) in a model that does not include any interaction effects. Coefficients are presented as odds ratios. Hence, I will report a negative relationship whenever an odds ratio shows an effect of less than unity. A more realistic model of the effect of family allowances ought to include a measure of the pace at which couples respond to a change in family allowances. Hence, I also include differenced series of the allowance variable. The addition of differenced series influences the interpretation of the main allowance variable: it measures the effect of family allowances in the absence of a change in the value of allowances over the years. The effect of the family allowance variable is significant and in the expected direction at every parity. If allowances were constant over time, then my model predicts that an allowance of NIS 1,000 per child will raise the odds of giving birth by 3% to 14%, depending on the parity. Family allowances seem to have their maximum effect in the first or second year after an increase in family allowances. These results should be treated with care, however, because many of the lagged effects are not even significant at 5%, although a stronger initial impact of family policies is not unheard of. The family policies adopted in the German Democratic Republic in 1976, for example, affected the timing of births more than completed fertility (Monnier 1990). Relatively large short-term effects of cash benefits have also been observed for other demographic phenomena, such as marriage (Prioux 1993).
Assuming the existence of economies of scale in a family, the cost of a higher-order child should be lower than that of a lower-order child. Hence, rational choice models would predict that child allowances will have a larger impact on the birth of higher-order children (Gauthier and Hatzius 1997:295). If, on the other hand, the cost of children increases with parity as Bloch and Glaude (1983) argued, or if the marginal “utility” of children declines at higher parities, then I expect benefits to have a smaller effect on higher-order births. The results show that the effect of family allowances increases with parity until the fifth interval, but it declines after the fifth birth interval. Notice, however, that the increase in the effect of family allowances with parity until the fifth interval is not large enough to offset the decline in birth rates with parity (see Figure 2). Notice also that birth intervals are not strictly comparable because of a change in religious composition, for which I cannot control. Hence, one should be careful with interpreting this finding. One possible explanation for the declining effect of family allowances after the fifth birth is that the reproductive behavior of religious couples is less responsive to financial incentives and more motivated by ideology.
Before I proceed to the full model, a few remarks are in order on the effects of the other variables. If market wage is a function of education and if children are more intensive in the value of the mothers’ time, then a negative relationship can be expected between female education and fertility (Ben-Porath 1973). Indeed, women in the lowest educational category in the third to fifth birth intervals are more likely to continue childbearing than those in the reference group (women with 9 to 12 years of education). Women in the highest educational category, however, are also more likely to continue childbearing, in all birth intervals. Thus, the correlation between female education and fertility tends to be U-shaped in the third to fifth interval. A positive relationship between female postsecondary education and third-birth rates has been reported before in several countries (Berinde 1999; Callens and Croux 2005; Kravdal 1992). I also report a positive correlation in fourth- to seventh-birth rates. Berinde (1999) suspected that women with a university education who already have two children are a select group with a higher preference for children than other women at the same educational level. Hence, Kravdal (2001) argued that the positive effects of high education on fertility should disappear when pooling birth intervals. However, the elevated higher-order fertility that is observed for highly educated women was not produced by the selection of women into ever higher parities because the effect does not disappear when all the birth intervals are pooled, even when the first and second intervals are included (result not shown).5
An Asian or North African origin of the woman and her husband raises the likelihood of giving birth. The effect of origin is not significant after the fifth-birth interval. Finally, the effect of the CPI suggests that hyperinflation had an inhibiting effect on reproductive behavior.
To study socioeconomic differences, the second model adds interaction effects between family allowances and education. Preliminary results showed no consistent educational differences across birth intervals in the timing of the response to an increase in family allowances. Perhaps this inconsistency is due to the statistical insignificance of many of the main effects of the differenced series. Hence, I do not include interactions between the differenced series of the allowance variable and education in the final analysis.
Table 3 shows that woman’s educational level is a significant modifier of the effect of family allowances. To the extent that education is positively correlated with income, rational-choice models would predict family benefits to have a larger impact on less-educated women. To the extent that opportunity costs are lower for women with higher education, however, rational-choice models would predict family benefits to have a larger impact on more-educated women. Table 3 seems to indicate that family benefits have a larger impact on more-educated women. In the third-birth interval, family allowances do not have a smaller impact on women in the lowest educational category, but they do have a larger impact on women with postsecondary education. After the third-birth interval, family allowances have a smaller impact on women in the lowest educational category, but they do not have a larger impact on women in the highest educational category.
The inclusion of interaction effects attenuates the main effect of a woman’s education. After controlling for the differential effect of family allowances, the positive relationship between postsecondary education and fertility that was observed in the third interval disappears. Thus, although in the first model, women with postsecondary education have higher third-birth rates than women in the reference category, the full model suggests that women with postsecondary education have lower third-birth rates in the absence of family allowances.
Education is only one component of socioeconomic status, and there are other indicators, such as occupation and income level. I experimented with running the models using current income levels for recent time periods immediately preceding the censuses to see what kind of results would emerge, but I found no significant effects, probably because of the small number of person-years in these models (results not shown).
The analysis included two period-based indicators: one for family allowances and one for inflationary tendencies. How can one be sure whether these indicators—and no others—matter most for fertility change, after controlling for structural change? One way to investigate this possible problem of underspecification is to evaluate the ability of the full model to explain the temporal component of the variance. If the variables included in the analyses account for much of the temporal component of the variation, then the effect of these omitted variables on the coefficients in the statistical models is unlikely to be large.
Figures 4–8 present observed and predicted parity-specific birth rates. The predicted third-birth rates would seem to be the least satisfactory. Although the full model predicts the long-term decline in third births, it also predicts a dip in the third-birth rate in the mid-1980s that never happened. If the effect of family allowances increases after the third birth, the inclusion of family allowance variables should improve prediction after the third birth, all other things being equal. Indeed, for higher parities, the predictive power of the full model is more satisfactory, suggesting that this model does not omit any major determinant of temporal trends of higher-order birth rates, including any unmeasured government policies.
Although unmeasured government policies do not seem to explain major trends to any large extent, family allowances seem to have had a visible effect. To illustrate the effect of family allowances, Figures 4–8 also present counterfactual-predicted, parity-specific birth rates in the absence of any family allowances. Without family allowances, parity-specific birth rates would have been lower, especially in the second half of the 1980s and early 1990s, when allowances peaked. More specifically, the results seem to indicate that third- to fifth-birth rates would not have increased in the 1980s, if not for the increase in family allowances, but sixth- and seventh-birth rates would have increased even in the absence of the increase in family allowances. I can only speculate on these differences between parities. Religious couples contribute the majority of sixth and seventh births. Hence, I suspect that ideological factors partly explain the increase of sixth- and seventh-birth rates.
The empirical results of this article seem to indicate that family allowances have a significant effect at every parity. Although several studies have reported a positive relationship between cash benefits and fertility, few studies have addressed the possible differential effects of cash benefits on families of different income or education levels. The results of this article seem to indicate that family allowances have a relatively large impact on more-educated women.
Previous research has shown that the impact of cash benefits on fertility is relatively small. Thus, a large database is required to investigate socioeconomic differences in the impact of cash benefits. Because the nature of cash benefits in Israel does not vary according to certain social characteristics, Israeli census data would seem to be an appropriate data set for the study of socioeconomic differences in the impact of cash benefits. However, the use of census data also has its drawbacks because of the limited number of variables available, possibly affecting the internal validity of the results. Important variables may have been omitted from the statistical models. This raises the question whether the policy effects that I claimed are indeed policy effects. They could be due to any other macro-level influences that could not be measured using census data. I used two period-based indicators—one for family allowances and one for inflationary tendencies—but I cannot be entirely sure that it is these indicators and no others that indeed matter for fertility change, such as maternity benefits, tax exemptions, or housing loans. My approach to this possible problem of misspecification was to study the extent to which the full model is able to account for observed time trends. In most birth intervals, the full model seems to provide an adequate prediction of major trends over time. Thus, the omission of a measure of maternity benefits, the value of tax exemptions, or housing loans probably does not affect the results to any large extent.
Several limitations imposed on the analysis by the data set may affect the external validity of the results. For technical reasons, my analyses are conditioned on women remaining in their first marriage, creating a study population whose behavior may not be representative. For example, to the extent that union commitment is a major determinant of fertility in higher-order unions, higher-order unions may be less responsive to family allowances. Thus, educational differences in higher-order unions may be smaller. Including only women in their first marriage also raises the issue of the responsiveness of marriage and divorce to family allowances. Because family allowances are independent of the mother’s marital status, however, marriage and divorce in Israel should not be responsive to family allowances. My analyses also excluded Israeli Arabs, most of whom are Muslims. Their fertility started to decline in the 1970s (Schellekens and Eisenbach 2002), and the causes of the decline are not known. When variables responsible for the decline are omitted from the analysis, family allowances variables are likely to pick up the effect of these omitted variables. Hence, the inclusion of Arab couples probably would have biased the results. Another issue that may affect the external validity of the results is the religious composition of the population after the fourth-birth interval, which may be peculiar to Israel. Note, however, that most of the results are quite similar for all five birth intervals, suggesting that the inclusion of a religiosity variable would not have affected the results to any large extent.
This article focuses on third and higher-order births. Nevertheless, the findings of this study may be relevant to the policy discussion in low-fertility contexts. Compared with other developed countries, fertility in Israel is relatively high. In 1994, the last year of the analysis, the total fertility rate was 2.6 births per woman among Israeli Jews (State of Israel 1996:119), compared with 2.0 births per woman in the United States (Hamilton 2004). Given this difference, the kind of dilemmas faced by Israeli couples after a second birth are likely to be similar to those faced by many couples in other developed countries after a first birth. Thus, the response of Israeli couples to incentives in the third interval is likely to resemble the response in the second interval elsewhere. Moreover, third births in other developed countries are not as rare as might be suggested by their total fertility rate. In Canada, for example, 13% of all births in 2004 were third births (Statistics Canada 2006:22), compared with 18% among Israeli Jews in the same year (State of Israel 1996:121; 2007:198). In any case, third-birth rates in several developed countries were high enough to have received scholarly attention in the past (e.g., Callens and Croux 2005; Hoem 1993; Kravdal 1992; Ní Bhrolcháin 1993).
My empirical results may also contribute to the discussion of the relationship between college education and fertility. Several studies reported that college education among women has a positive impact on birth rates (Berinde 1999; Callens and Croux 2005; Kravdal 1992). I also found evidence for such a correlation. The reason for this correlation, however, is not clear, and some believe it to be an artifact (Kravdal 2001). This does not seem to be the case in Israel, though, because the effect does not disappear in an analysis that pools all intervals. The results seem to indicate that in the case of Israel, this correlation is in part due to the effect of family allowances. This analysis seems to indicate that without family allowances, women’s postsecondary education would have had a much smaller positive impact on birth rates.
I would like to thank Moshe Ophir and Anat Ziv for research assistance; Zvi Eisenbach, David Gliksberg, and Guy Stecklov for their advice; four anonymous reviewers for their comments; and Joram Mayshar for sharing his family allowance data file with me. The census data used in the analysis were provided by the Israel Central Bureau of Statistics through the Israel Social Science Data Center.
A previous version of this article was presented at the annual meeting of the Population Association of America, New York, March 2007.
The research reported in this article is an initiative of the Public Council for Demography and is supported by a grant from the National Insurance Institute of Israel.
1.The TMFR is defined as the sum of the age-specific marital fertility rates per 1,000 women, multiplied by five and divided by 1,000, when age is given in five-year age groups.
2.The inclusion of lagged responses of up to 10 years does not influence the results of the analysis to any large extent (result not shown).
3.The inclusion in the analysis of the husband’s education and its interaction with incentives does not alter the effects of wife’s education to any large extent (results available from the author upon request).
4.The first Social Survey of 2004 shows that a small majority of those aged 40–54 with at least five births were Orthodox (62% of 261 women). Although a large majority in the survey aged 40–54 with at least six births were Orthodox (82% of 138 women), almost all women at that age with at least seven births were Orthodox (96% of 98 women).
5.An n-shaped association may occur when the mother’s education also serves as a proxy for another omitted variable. Poor knowledge and less-efficient use of birth control, for example, may explain the relatively high fertility of the least-educated women (Kravdal 1992:459).