Public health insurance expansions and labour supply of married women: the state children's health insurance programme

Abstract

While the relationships between health insurance and the labour supply of women have been explored in the literature, little is known about the effects of offering public health insurance on the labour supply decisions of married women. This article examines the labour supply decisions of married women using the State Children's Health Insurance Programme. Our empirical analysis implies that certain groups of married women may be leaving the labour force in order to provide public health insurance for their children. We conclude that the programme causes unexpected efficiency losses through distorted labour supply decisions.

Acknowledgements

We gratefully acknowledge suggestions by Thomas Buchmueller, Li Ming Dong, William N. Evans, Elizabeth Hendrey, Sarah Senesky and Alan Weinman, and thank Jason Pau and Molly Sherlock for their excellent research assistance. Financial support from the Economic Research Initiative on the Uninsured and the City University of New York Research Foundation is greatly acknowledged. All errors are ours.

Notes

¹Near poor families are usually defined as those with income between 100 to 200% of the federal poverty level (FPL). In 2003, 200% of the FPL was approximately $30520 for a family of three.

²‘Medicaid eligibility for nondisabled children was originally limited to single-parent families receiving cash assistance under the Aid to families with Dependent Children (AFDC) (Card and Shore-Sheppard, 2002, p. 4).’

³Several works study the effects of Medicaid expansions on the substitution of public health insurance for private health insurance. The literature indicates the possibility of crowd-out during the period of the late 1980s into the early 1990s (Cutler and Gruber, 1996; Dubay and Kenny, 1996, 1997; Thorpe and Florence, 1998/1999; Blumberg et al., 2000; Shore-Sheppard, 2000; Yazici and Kaestner, 2000; Ham and Shore-Sheppard, 2001; Card and Shore-Sheppard, 2002). Kronick and Gilmer (2001) and LoSasso and Buchmueller (2002) find similar results by considering public health insurance expansions during the more recent period of the late 1990s into the early 2000s.

⁴The authors’ tabulation shows that more than half of the beneficiaries of SCHIP are children of married women (See appendix for a detailed discussion). Yelowitz (2003) highlights the important role married women play in public health insurance expansions.

⁵There are advantages and disadvantages associated with using the CPS vs. other data sets. Longitudinal data sets such as the SIPP and NLSY are preferable when the objective is to capture dynamic changes in labour supply. However, the sample sizes in the SIPP and NLSY are much smaller than in cross-sectional data sets such as the CPS. Blumberg et al . (2000) admit, ‘The small sample size and complex sampling design of the SIPP reduce our statistical power…’ Also, the SIPP suffers from ‘seam bias’ (Young, 1989; Marquis and Moore, 1990), and does not have the information concerning the residents of nine low population states. The CPS does not suffer from these particular problems. One advantage of the SIPP is the accuracy of the information. Data collected is close to the time of interview (cf. the CPS asks the last year's information in its interview), as Ham and Shore-Sheppard (2001) points out. The NLSY also has some problems. ‘Since the NLSY is composed of one cohort of mothers who are aging over the time period of the expansions, trends in insurance coverage for children in the NLSY are different from the trends in the general population…. Consequently, estimated effects of the expansions from the NLSY may not be generalized to the entire population of children’ (Ham and Shore-Sheppard, 2001, p. 10). The CPS data are nationally representative and are less likely to suffer from such attrition. We use the CPS that most previous works use so that we are able to contrast our results to those in the literature. In fact, Ham and Shore-Sheppard (2001) estimate the same model by using both the SIPP and the CPS. They conclude that differences in estimates come from the different nature of each data set, i.e. monthly data or annual data. They get similar estimates in either data once they correct for the differences.

⁶We stratify the sample using each state's income eligibility threshold after accounting the number of children.

⁷We exclude self-employed women from the sample as is common in the literature. Chou and Staiger (2001) point out that ‘their labour force status is often ambiguous (e.g. there is less distinction between housekeeper, employed, and self-employed).’ Self-employed husbands are included in the sample as private sector employees.

⁸Our identification strategy implicitly assumes that married women take husbands’ labour supply decisions as given when making their labour supply decisions. The assumption is similar to most of the literature on married women's labour decisions (e.g. Triest, 1990). One possible interpretation is that married women are secondary earners so that their labour supply is more flexible than their husband's. This assumption is not unreasonable given the findings of Buchmueller and Valletta (1999). They find that husbands’ work hours have insignificant effects on wives’ work hours.

⁹Bertrand et al . (2004) warn that analysis without correcting for serial correlation may understate the SD of the treatment effects and, thus, leads to serious overestimation of t-statistics. We may conclude that (policy) changes affect a dependent variable, when in fact this is not the case.

¹⁰Nelson and Mills (2002) point out the limitations of the CPS data. The survey does not ask questions about SCHIP coverage if all children in households are covered by Medicaid. The problem is that some SCHIP participants confuse SCHIP with Medicaid and did not answer questions about SCHIP. Nelson and Mills tabulate the number of children on SCHIP by using the CPS and show that the number is substantially below the administrative enrollment figure. Since the gap is almost equivalent to the number of M-SCHIP enrollment, they conclude that M-SCHIP participants confuse their health insurance coverage with Medicaid. If the CPS data is mainly composed of S-SCHIP, this may cause sample bias. However, we do not know if, or how, the bias affects the characteristics of parents whose children are on SCHIP.