子女对英国女性劳动力供给和收入的影响:使用双胞胎作为工具变量的实证研究
Mary A. Silles 赫尔大学商学院经济学系
摘要:本文的目的是调查子女对英国女性劳动力供给和收入的影响。很难去估计家庭规模与经济状况之间的因果关系,因为增加家庭规模的因素同样可能影响女性的劳动力市场结果。我们把双胞胎作为工具来解决这个问题。在有13岁以下子女的女性中,IV估计表明,双胞胎出生导致的家庭规模增加对女性的劳动力供应和收入产生不利影响。尽管OLS是显著的,但在有年龄较大子女的女性中,IV结果不显著。最后,我们将这些结果与使用双胞胎性别构成工具得出的估计值进行比较。使用这一工具的估计值与使用双胞胎出生的估计值非常接近,并意味着生育的劳动力市场惩罚随着时间的推移而消失。
本文的目的是研究家庭规模对女性劳动力市场真正的因果影响。OLS估计可能无法用来解释,因为生育率对于工作偏好而言可能是内生的。在本研究中,双胞胎自然实验被用来测量家庭规模。保证双胞胎出生作为工具变量具备有效性有两个条件。第一,要求工具变量与家庭中的子女数量相关联。过去的研究和我们的描述性统计表明,情况确实如此。第二,双胞胎工具是随机的,而且随着子女数量的增加而增加,超过了原本可以实现的期望家庭规模(罗森茨威格和沃尔宾,1980)。双胞胎出生这一事件并不一定与期望的家庭规模无关;想要更大家庭规模的女性更有可能生双胞胎。因此,按出生顺序简单地合并双胞胎数会倾向于对生过几次孩子的女性进行过采样,这通常会导致对希望拥有更大家庭的女性进行过采样。
我们的方法是在第一次、第二次和第三次怀孕时分别检查是否是双胞胎,以消除双胞胎的发生和怀孕数量之间的任何联系对劳动力市场结果的影响。因此,分析集中在第一次、第二次和第三次出生时有双胞胎的女性的三个子集,以探究未预料到的孩子的影响如何增加不同地区的家庭规模。因此,我们估计至少有k个孩子的妇女样本的IV模型如下:
(1)
(2)
等式(2)表示IV估计的第一阶段,其中等式(1)是第二阶段。因变量是衡量劳动力市场活动的一个指标。控制变量包括女性生育时的年龄、接受调查时的年龄、受教育年限和调查年份。女性的生育年龄是一个重要的控制因素,因为双胞胎出生的可能性随着年龄的增长而增加,而且还因为女性首次生育时的年龄与劳动力市场的参与有关。表示在至少有k个子女的女性的子样本中,样本i所生子女的总数(kge;1;kge;2;kge;3)。是工具变量,这是一个虚拟变量,用于确定样本i第j次生育的是否是双胞胎。对于至少有一个孩子的女性样本,双胞胎工具是,如果一个女性在第一次生育时是双胞胎,设置为1,如果是独生子女,则设置为0。这样,运用工具变量,我们就能够识别从一个孩子到两个或更多孩子的边际效应。对于至少有两个孩子的女性的子样本,该工具被重新定义为,如果家庭中生育的第二胎是双胞胎,则取值为1,否则为0。运用工具变量,我们可以获得从两个孩子的家庭到三个或更多孩子的家庭的边际效应。通过将样本限制在至少有两个孩子的家庭,我们可以确保双胞胎或独生子女家庭对家庭规模的偏好是相同的。对于至少有三个孩子的女性比例,该工具被定义,如果第三胎是双胞胎,则取值为1,如果是独生子女,则取值为0。运用工具变量可以让我们测量第四个孩子因双胞胎出生所产生的影响。同样,通过关注至少有三个孩子的女性这一子样本,我们可以避免这样一个可能出现的问题,即选择多胎妊娠的女性更有可能至少生一个双胞胎。
将双胞胎出生作为一种手段,还有其他一些可能的问题。一个担忧是,生异卵双胞胎的倾向存在于家庭中。排卵时释放两个而不是一个卵子的超排卵是遗传性的。由于女性不太可能知道自己是否携带这种基因,这一因素不太可能对双胞胎出生工具产生影响。
使用双胞胎出生作为工具的第二个可能的反对意见是,众所周知,非洲黑人妇女生双胞胎的比率高于其他种族的妇女(米里扬索普洛斯,1970)。在我们的数据中,非洲黑人女性的比例太小,无法按种族对数据进行有意义的分类。然而,我们将样本仅限制于白人女性作为对我们主要结果稳健性的检验,在未报告的分析中,我们发现并没有实质性地改变以下的回归结果。
将双胞胎出生作为生育工具变量的第三个问题是辅助生殖技术的使用,特别是体外受精技术,众所周知,这种技术会增加双胞胎出生的可能性。与美国不同的是,在英国,体外受精技术和其他提高生育能力的药物由国家卫生服务机构(NHS)为所有女性提供。因此,双胞胎出生产生的生育率变量在英国的随机比例可能要比美国大得多。然而,生育治疗的使用越来越多,特别是在两国社会经济地位较高的老年夫妇中,这可能对近年来确定人口估计值的工具变量产生重大影响。尽管从定义上来说,工具性外生性无法直接检验,但为了评估这一问题对我们结果的内部有效性的威胁,通过利用对母亲在第一次生育时的年龄和调查年份的控制来估计在生育顺序中的每个点处的双胞胎出生的可能性的线性概率模型,我们可以检查双胞胎在第一次、第二次和第三次生育时的概率是否与可观察的特征相关,诸如母亲的教育。在这些回归中,母亲教育系数可以忽略不计,在任何常规显著性水平上都没有统计显著性。虽然这一点令人放心,因为社会经济地位应与受教育年限高度相关,但在解释我们的结果时,务必记住寻求生育治疗的决定可能与对劳动力供应的意愿有关。由于关于生育治疗的没有全球统一的标准,这个问题无法直接解决。
最后一个关于使用双胞胎来测量家庭规模的问题是,近年来双胞胎数量的增加与使用牛生长激素(bST)有关(斯泰因曼,2006)。这种激素在美国主要用于增加牛的奶和肉产量。食用含有这种激素的乳制品,尤其是牛奶,会增加多次排卵的几率。这种激素在英国于1990年被禁止使用,尽管我们的数据延伸到了这项立法之前出生的孩子,但由于几乎每个人都食用乳制品,因此这一因素不应导致基于双胞胎IV估计的估计出现偏差。
除了IV估计,我们还提供了OLS对女性劳动力市场结果和家庭规模之间关系的估计。一般来说,OLS回归只在方程(1)中的误差项与家庭规模之间没有相关性的情况下,对家庭规模的增量产生一致的估计,这取决于集合控制变量。如果家庭规模与被排除在控制变量之外的一些特征(例如,职业抱负)之间存在相关性,而这些特征也影响劳动力市场的结果,则OLS的估计将有偏差。如果这一遗漏的变量与劳动力市场的正相关,而与子女数量负相关,则将其排除在OLS回归之外将使估计值向上(以绝对值计)倾斜,因为估计的子女负面影响的一部分事实上并非由于家庭规模。如果能够成功地证明双胞胎出生是误差项的外生因素,并且与家庭规模相关,IV估计法应该可以解决这个问题。
结论:
表2至表4分别列出了我们针对至少有一个孩子、至少有两个孩子和至少有三个孩子的女性得出的主要结果。第一栏和第二栏报告了对有一个13岁以下孩子的有伴侣和单身女性的估计。最后两栏显示了子女年龄大于13岁的女性样本的相应估计。为了节省空间,表中仅呈现等式(1)中关于孩子数量的系数以及等式(2)中关于双胞胎工具的第一阶段系数。
表2第一行所列的估计数显示了双胞胎在出生时对所生子女数量的影响。对于第一次怀孕产生双胞胎的有伴侣的女性,系数(标准误差)为0.697(0.039),t统计为17.9。对于单身女性,相应的估计值略小于0.613(0.088),t统计值为6.9。t统计值小于5表明工具变量存在不足,因此我们的IV估计模型可以有力地检测生育对劳动力市场的影响。对于长子大于13岁的女性来说,双胞胎在第一次出生时的影响会大大降低,大约会增加0.5个孩子。这种效应对于有伴侣和单身的女性来说几乎是相同的,但值得注意的是,单身女性样本的标准误差大约是有伴侣的女性的样本的两倍。从长期来看,双胞胎在第一次怀孕时出生对家庭规模的影响要小得多,这一发现与家庭及时调整其生育率以应对双胞胎出生的影响有关。虽然这些第一阶段的结果不具有直接可比性,但在雅各布森等人(1999)为美国提供的估计范围内。这些学者发现,双胞胎在第一次出生时的影响随着第一个孩子的年龄稳步下降,但稳定在0.6左右。在英国,双胞胎首次出生时对家庭规模的影响下降得更为明显,这可能反映了本研究使用的最新数据,以及生育率随着时间的推移而普遍下降的事实。
Table1.
所有妇女的汇总统计数据可变平均值(标准差)
所有孩子 |
第一个孩子的年龄lt;=13岁 |
第一个孩子的年龄gt;13岁 |
||||
没有双胞胎孩子的母亲 |
有双胞胎孩子的母亲 |
没有双胞胎孩子的母亲 |
有双胞胎孩子的母亲 |
没有双胞胎孩子的母亲 |
有双胞胎孩子的母亲 |
|
子女数量 |
2.224 |
3.181 |
1.839 |
2.850 |
2.457 |
3.375 |
(0.982) |
(0.921) |
(0.807) |
(0.787) |
(1.005) |
(0.939) |
|
母亲第一次生育时的年龄 |
23.780 |
24.408 |
25.253 |
26.389 |
22.887 |
23.245 |
(4.702) |
(4.985) |
(5.185) |
(5.534) |
(4.135) |
(4.222) |
|
母亲在接受调查时的年龄 |
41.724 |
42.461 |
32.122 |
33.804 |
47.540 |
47.543 |
(10.032) |
(9.445) |
(5.919) |
(5.915) |
(7.121) |
(7.157) |
|
有偿工作 |
0.615 |
0.590 |
0.539 |
0.469 |
0.662 |
0.661 |
(0.487) |
(0.492) |
(0.498) |
(0.499) |
(0.473) |
(0.473) |
|
每周工作小时 |
16.745 |
15.518 |
13.362 |
10.776 |
18.794 |
18.301 |
(16.158) |
(15.586) |
(15.054) |
(13.554) |
(16.456) |
(16.028) |
|
兼职状况 |
0.764 |
0.795 |
0.838 |
0.895 |
0.719 |
0.735 |
(0.424) |
(0.404) |
(0.368) |
(0.306) |
(0.449) |
(0.441) |
|
每周总收入 |
130.189 |
127.222 <!--剩余内容已隐藏,支付完成后下载完整资料
The impact of children on womenrsquo;s labour supply and earnings in the United Kingdom: evidence using twin births Mary A. Silles Department of Economics, Hull University Business School
Abstract:The objective of this paper is to investigate the effect of children on womenrsquo;s labour supply and earnings for the United Kingdom. Estimating the causal relationship between family size and economic status is complicated because the same factors that increase family size may also affect the labour market outcomes of women. The birth of twins is used as an instrument to address this problem. Among women with children under age 13, the IV estimates indicate that a larger family induced by a twin birth adversely affects womenrsquo;s labour supply and earnings. Among women with older children, the IV results show no evidence of a causal effect, despite significant OLS relationships. Finally, we compare these results to estimates produced using a twinsrsquo; sex composition instrument. Estimates using this instrument are very close to the estimates using twin births and imply that the labour market consequences of childbearing disappear over time. The objective of this paper is to distinguish the true causal effect of family size on womenrsquo;s labour market outcomes. OLS estimates may be difficult to interpret since fertility is likely to be endogenous with respect to tastes for work. In the present study the twins natural experiment is used to instrument for family size. There are two conditions for the validity of the twin birth instrument. The first requires that the instrument is correlated with the number of children in a household. Past research and our descriptive statistics indicate that this is indeed the case. The second states that the twins instrument is virtually random and increases with the number of children beyond the desired family size that would have otherwise been achieved (Rosenzweig and Wolpin 1980). The event of a twin birth is not necessarily independent of desired family size; women who desire larger families would be more likely to have twins. Therefore, simply pooling twins across birth order would tend to oversample women who had several births which in general would result in oversampling women who desired larger families. Our approach to purging the effect of a twin birth on labour market outcomes from any link between the occurrence of twins and the number of pregnancies is to examine twin births at first, second and third pregnancies separately. Thus, the analysis focuses on three subsets of women who had twins at first, second and third birth to provide a sense of how the effects of an unanticipated child may increase family size at different parities. We therefore estimate the following IV model for the sample of women with at least k children: (1) (2) Equation (2) represents the first-stage of the IV estimation, where equation (1) is the second stage. The dependent variable is a measure of labour market activity. The vector of control variables includes motherrsquo;s age at the time of birth, age at survey year, years of schooling, and survey year dummies. Maternal age at birth is an important control because the probability of a twin birth increases with age but also because maternal age at first birth is related to labour market participation. The denotes the total number of children given birth to by individual i in the subsample of women with at least k children (kge;1; kge;2; kge;3). The instrumental variable is which is a dummy variable for whether the birth was a twin birth for person i. For the sample of women with at least one child, the twins instrument is twins first, which is set to one if a woman had twins at first birth and zero if a singleton. In this way, we are able to identify the marginal effect of moving from one to two or more children. For the subsample of women with at least two children, the instrument is redefined to denote twins second which takes a value of one if the second birth in the family is a twin birth and zero otherwise. The twin-second instrument allows us to examine the marginal effect of going from a family of two to three or more children. By restricting the sample to families with at least two children, we are making sure that preferences over family size on average are the same in families with twins or singletons at the second birth. For the fraction of women who have had at least three children, the instrument is defined as twins- third and takes a value of one if the third birth is a twin birth and zero if it is a singleton. The twins-third instrument allows us to measure the effect of a fourth child who is born as a consequence of twinning. Once again by focusing on the subsample of women with at least three children we can avoid the possible problem that women who choose to have more pregnancies are more likely to have at least one twin birth. There are a number of other possible problems with the use of twin births as an instrument. One concern is that the propensity to give birth to fraternal twins runs in families. Hyperovulation, where two eggs instead of one are released at the time of ovulation, is hereditary. Since women are unlikely to know whether or not they carry this gene, this factor is unlikely to have implications for the twin birth instrument. A second possible objection to the use of twin births as an instrument is that the incidence of having twins is known to be higher in black African women than among women of other races (Myrianthopoulos 1970). In our data, the proportion of black African women is too small to meaningfully disaggregate the data by race. However, as a check on the robustness of our main results, in analysis not reported, we found that limiting our sample to white women did not materially alter our regression results presented below. A third 剩余内容已隐藏,支付完成后下载完整资料 资料编号:[270971],资料为PDF文档或Word文档,PDF文档可免费转换为Word |
课题毕业论文、外文翻译、任务书、文献综述、开题报告、程序设计、图纸设计等资料可联系客服协助查找。