Review of Economics of the Household

, Volume 10, Issue 4, pp 541–556

The effect of the original introduction of Medicaid on welfare participation and female labor supply

Authors

    • National Center for Health StatisticsCenters for Disease Control and Prevention
  • Frederic W. Selck
    • Bloomberg School of Public HealthJohns Hopkins University
Article

DOI: 10.1007/s11150-011-9132-7

Cite this article as:
Decker, S.L. & Selck, F.W. Rev Econ Household (2012) 10: 541. doi:10.1007/s11150-011-9132-7

Abstract

This paper uses the fact that states introduced Medicaid programs at different times between 1966 and 1972 to estimate Medicaid’s effect on Aid to Families with Dependent Children (AFDC) participation. Using state-level data, we find that the introduction of Medicaid accounted for approximately 10% of growth in AFDC caseloads from 1964 to 1974, a time period during which there was thought to be significant unexplained growth in caseloads. Analysis of individual-level data indicates that Medicaid’s effect on AFDC participation occurred through its effect in increasing the number of eligibles who participated in the program, and not because of increases in eligibility or reductions in workforce participation.

Keywords

MedicaidWelfareAFDCLabor supply

1 Introduction

Two events characterize welfare in the United States during the mid to late 1960’s. First, there was a remarkable increase in the number of families receiving public assistance (see Fig. 1). Second, the Social Security Amendment of 1965 provided, for the first time, health insurance to families receiving government cash assistance.
https://static-content.springer.com/image/art%3A10.1007%2Fs11150-011-9132-7/MediaObjects/11150_2011_9132_Fig1_HTML.gif
Fig. 1

Average monthly number of AFDC/TANF families. Source HHS (2010)

The rapid expansion of the Aid to Families with Dependent Children (AFDC) program was unexpected and subsequently put pressure on budgets at both the state and federal level. Researchers devoted much effort to determining the proximal cause of the increase in the welfare rolls. Several possible contributors have been explored and rejected. State-level variation in eligibility requirements, unemployment rates, size of payments, and increases in the number of families headed by a single female all failed to adequately explain the increases in caseload (Moffitt 1986). Moreover, cash benefits were decreasing, seemingly making AFDC less attractive to potential recipients (Moffitt 1990).

One hypothesis is that the introduction of Medicaid in the 1960s served to increase AFDC caseloads. This may have happened in one or two ways. First, since those participating in AFDC were made automatically eligible for Medicaid, the introduction of Medicaid may have changed the budget set for individuals whose work alternatives did not include health coverage. As a result, single mothers faced an incentive to either reduce work hours or leave the labor force entirely and move to AFDC. An alternative hypothesis is that the increase in the AFDC caseloads was largely driven by families that were already eligible for enrollment, but did not participate because of some fixed-cost, stigma associated with receiving cash assistance. The introduction of Medicaid may have increased the value of participation in AFDC, and therefore have increased participation among eligible’s whose benefit from participation now exceeded their costs. There is evidence that participation rates among eligible families more than doubled by the mid-1970s from about 42% in 1967, suggesting that this hypothesis could be at least one of the reasons behind the increase (Ruggles and Michel 1987).

This paper evaluates the relationship between the implementation of Medicaid and its effect on AFDC caseloads and labor supply among female heads of household. We exploit the fact that states joined Medicaid at different times between early 1966 and early 1972 to estimate its effect on participation in the labor market and welfare participation. We first use state-level data on AFDC caseloads to examine Medicaid’s ability to help explain the aggregate growth in AFDC caseloads during the late 1960s. While controlling for other influences on the number of families participating in AFDC, we conclude that the introduction of the Medicaid program accounted for approximately 10% of total growth in AFDC caseloads from 1964 to 1974. This indirect cost of the Medicaid program (increased AFDC payments for new AFDC participants) was over one-third as large as the direct cost of introducing the Medicaid program.

We then use individual-level data on female heads-of-household during the 1966–1974 period to explore how this increase in AFDC participation took place. We find that the introduction of Medicaid led to a 6.9 percentage point (about 16%) increase in the probability that a female head participated in AFDC. Our findings therefore suggest that Medicaid’s effect on AFDC participation occurred through its effect on increasing the number of eligible mothers who participated in the program since we do not find that the introduction of Medicaid significantly affected the probability that a single mother participated in the labor force.

The paper proceeds as follows. The next section presents some background information on the history of the AFDC and Medicaid programs. The following section summarizes previous literature on the link between Medicaid, AFDC and female labor force participation. The fourth section presents estimates of the effect of Medicaid on state AFDC caseloads. The fifth section presents estimates of the effect of Medicaid on AFDC participation using the individual-level data on female heads-of-household. The sixth section explores the effect of Medicaid on female head’s labor supply, while the final section concludes.

2 Medicaid and AFDC eligibility

The AFDC program (originally Aid to Dependent Children) was created by the Social Security Act of 1935. Under this program, states received Federal matching funds for the provision of cash benefits to single parent households with income under a certain need level. Both benefit amounts and need standards were determined by individual states. Medical benefits specifically available to AFDC recipients were extremely limited prior to the advent of Medicaid in 1965. The 1965 Medicaid legislation (Title XIX of the Social Security Act) resulted in a large increase in the amount that states spent on medical vendor payments. Average state medical vendor payments per AFDC recipient increased from $3.34 per month in 1965 to $15.05 in 1971 in 1967 dollars (Social Security Administration 1966; National Center for Social Statistics 1972). These payments varied across states due to differences in optional services covered and provider payment rates as well as differences in use of and access to services across states (Moore and Smith 2005–2006; Klemm 2000). In this paper, we therefore seek to estimate the average effect across states of the introduction of Medicaid on AFDC caseloads.

Following Moffitt (1983), a woman maximizes utility, which is a positive function of disposable income and negative function of hours of work, subject to a budget constraint as pictured in Fig. 2. In the absence of both AFDC and Medicaid, a household with non-labor income AB faces a budget constraint of ABD, where the slope of the line BD reflects the head’s net wage opportunity, dependent on both her individual characteristics and on local labor market characteristics and tax rates. Now suppose that the AFDC program offers a maximum cash benefit to this household represented by the distance BE. In the absence of the Medicaid program, a potential AFDC household faced a budget constraint of AECD, where the slope of the line EC reflects both the head’s wage opportunity and the AFDC program’s tax on labor earnings. This tax rate depends on state-specific child care and other work-related deductions as well as a Federally-mandated statutory tax rate. Before the Social Security Amendments of 1967 (which became effective on July 1, 1969), the Federally-mandated nominal tax rate on AFDC earnings was 100%. Beginning in 1969, Federal law required the deduction of an initial $30 in monthly earnings plus one-third of remaining earnings (plus work-related deductions) when computing an AFDC household’s continued eligibility.1
https://static-content.springer.com/image/art%3A10.1007%2Fs11150-011-9132-7/MediaObjects/11150_2011_9132_Fig2_HTML.gif
Fig. 2

Labor Supply, AFDC, and the Medicaid “Notch”

Because Medicaid receipt was tied to participation in the AFDC program, the introduction of Medicaid in 1966 created a “notch” in the potential AFDC household’s budget constraint, represented in Fig. 2 by the distance EH or CG. The size of this notch is dependent both on the generosity of the state’s Medicaid program and on the availability of health insurance to workers who did not participate in AFDC. Several simple facts suggest that the size of this notch was substantial. In 1965 (one year before Medicaid became available), only about 28% of single mothers who gave birth had hospital insurance. Even among single mothers who worked at least part-time during their pregnancy, only approximately 34% had hospital insurance.2 This widespread lack of insurance among female heads during this time period suggests that many female heads considering leaving AFDC after the introduction of Medicaid faced a substantial Medicaid “notch”.

3 Previous research on the relationship between Medicaid, labor supply, and AFDC

The observed growth in AFDC caseloads during the mid 1960s to the 1970s remains largely unexplained. Moffitt (1986) and Michel (1980) examined the level of AFDC benefits, the benefit reduction rate, the unemployment rate, and individual determinants of welfare participation. Using time-series data on AFDC participation rates, they fail to find a predictor large enough to explain the upward trend in welfare enrollment. Medicaid implementation was not included in their analysis. Lacking any other explanation, Moffitt (1987) attributes the increase to a “structural shift” stemming from changing societal norms and decreased stigma from welfare receipt.

Although there are a number of studies that examined the incentive effects of AFDC (see Moffitt (2002) for a review), only a few papers have examined Medicaid’s influence on welfare participation or female labor supply. During the 1980s, the variation in Medicaid benefits at the state-level provided one possible means of evaluating its effect on female work decisions. Winkler (1991), Blank (1989) and Moffitt and Wolfe (1992) used this variation and found that Medicaid encouraged AFDC participation and reduced labor force participation, but only for those that would highly value health insurance. All three of these studies used data from one year to estimate their results. The cross-sectional designs used in these studies did not allow for the inclusion of state fixed-effects, raising the concern that estimates of Medicaid’s value is likely correlated with other state factors that influence both benefit generosity and AFDC participation. Moreover, without knowing what source of variation is driving the difference in Medicaid benefits across states or how to separate the effects of Medicaid from the effect of poor health on labor force participation, it seems difficult to obtain a conclusion about the causal effect of health insurance on welfare participation from these studies.

Several later studies used a combination of variation in benefits and pooled cross-section data from several years of the Current Population Survey to address this issue. Montgomery and Navin (2000) find that Medicaid’s negative relationship with labor supply is not robust to the inclusion of state fixed- and random-effects. Yelowitz (1995) used the phased implementation of Medicaid expansions for children from 1988 to 1991 to assess family labor supply decisions. These expansions removed the requirement that families be on AFDC to be eligible for Medicaid. Yelowitz found that the decoupling of Medicaid from AFDC resulted in higher labor force participation. In contrast to Yelowitz, neither Ham and Shore-Sheppard (2005) nor Meyer and Rosenbaum (2001) found that the Medicaid expansions had an effect on the labor supply. Instead, Ham and Shore find no detectable effect between Medicaid and the labor supply. Meyer and Rosenbaum find that increased labor force participation by single mothers can largely be explained by the expansion of the Earned Income Tax Credit, and to a lesser extent, welfare benefit cuts.

This paper revisits the relationship between AFDC and Medicaid, and for the first time uses the phased introduction of Medicaid by state to examine the link.3

4 The effect of Medicaid on state AFDC caseloads

We use the fact that states joined the Medicaid program at different times between 1966 and 1972 to estimate the effect of health insurance availability on the log of AFDC caseloads during the 1964–1974 period. Table 1 lists the date on which each state introduced a Medicaid program. Although 26 states began operating a Medicaid program in 1966, many large states joined the program significantly later. Texas and Georgia, for example, joined the program in 1967, Tennessee and Virginia in 1969, and Florida, Indiana and New Jersey in 1970.
Table 1

State implementation of Medicaid

Year

Month

Year

1966

January

Hawaii, Illinois, Minnesota, North Dakota, Oklahoma, Pennsylvania

March

California

May

New York

July

Connecticut, Idaho, Kentucky, Louisiana, Maine, Maryland, Nebraska, Ohio, Rhode Island, Utah, Vermont, Washington, West Virginia, Wisconsin

September

Massachusetts

October

Delaware, Michigan

December

New Mexico

1967

June

Kansas

July

Iowa, Montana, Nevada, New Hampshire, Oregon, South Dakota, Wyoming

September

Texas

October

Georgia, Missouri

1968

July

District of Columbia, South Carolina

1969

January

Colorado

July

Virginia

October

Tennessee

1970

January

Alabama, Arkansas, Florida, Indiana, Mississippi, New Jersey, North Carolina

1972

January

Alaska

Bold indicates that the state was available in the Current Population Survey. Arizona began a special Medicaid managed care program in the early 1980s

Figure 3 shows caseloads in several states. The first panel shows caseloads in Pennsylvania, which began a Medicaid program in January of 1966, and Indiana, which began a Medicaid program in January of 1970. The second panel shows caseloads in two southern states: Texas, which began a Medicaid program in September of 1967, and Tennessee, which began a Medicaid program in October of 1969. These states show considerable growth in AFDC caseloads overall during the 1964–1974 period, although the largest growth appears to be in periods following the introduction of state Medicaid programs.
https://static-content.springer.com/image/art%3A10.1007%2Fs11150-011-9132-7/MediaObjects/11150_2011_9132_Fig3_HTML.gif
Fig. 3

AFDC caseloads and Medicaid implementation for selected states. Source HHS (2010)

The national trend in AFDC caseloads is pictured in Fig. 1. The largest growth in AFDC caseloads occurred from 1968 to 1971. During this period, AFDC payment amounts were essentially flat, while real wages in the manufacturing sector increased. In the analysis below, we seek to separate the influence of Medicaid from the effect of other factors in inducing families to participate in the AFDC program. Specifically, we examine the log of monthly state AFDC caseloads during the 1964–1974 period as a function of real gross state product per capita and the real AFDC payment for a family of three.4 The “Great Society” period of the late 1960s saw many changes in welfare policy other than the introduction of Medicaid that undoubtedly contributed to the growth in AFDC caseloads.5 We control for the influence of these factors in two ways. First, we include year and month effects to control for national changes (such as changes in Federal policy) which were likely to have influenced AFDC caseloads. These changes would include, for example, the introduction of the Early and Periodic Screening, Diagnostic, and Treatment (EPSDT) mandatory benefit for Medicaid in 1967. We also include state effects to account for differences in state policy which affect the level of AFDC caseloads. All analyses allow for arbitrary correlation of standard errors by state (Moulton 1990; Donald and Lang 2007).

We use ordinary least squares to estimate Eq. 1, where i indexes states (1 through 51) and t indexes years (1 through 11), Pymt refers to the maximum AFDC payment for a family of three, and GSP refers to real gross state product per capita.
$$ {\text{AFDC}}_{\text{it}} = \Upphi \,\left( {\alpha + \beta \,{\text{Medicaid}}_{\text{jt}} + \gamma \,{\text{Pymt}}_{\text{jt}} + \psi \,{\text{GSP}}_{\text{jt}} + \sum {\theta_{\text{k}} {\text{Year}}\_{\text{Month}}_{\text{k}} } + \sum {\kappa_{\text{m}} {\text{State}}_{\text{m}} } } \right) $$
(1)
Results for this specification are contained in Table 2. The “Medicaid exists” variable in this table is equal to 1 if a state has a Medicaid program in place and zero otherwise. Although gross state product per capita does not appear to have an effect on caseloads, the generosity of AFDC benefits do have a significant effect on the growth of AFDC caseloads in column (1). This column also indicates that Medicaid has a strong effect in increasing AFDC caseloads. Results indicate that the introduction of Medicaid increased AFDC caseloads by nearly 11%.
Table 2

The introduction of Medicaid and AFDC caseloads (Log of annual AFDC caseloads by state: 1964–1974)

Independent variables

(1)

(2)

Medicaid exists

0.106*** (0.035)

0.050** (0.021)

Real gross state product per capita/1,000

−0.075 (0.106)

0.022 (0.067)

Real AFDC payment for family of 3/1,000

0.015** (0.006)

0.003 (0.003)

State * linear time trend

No

Yes

R-Sq

0.984

0.995

Regressions are estimated by ordinary least squares with standard errors (in parentheses) clustered by state. All regressions include state, year and month effects. The sample size is 6,732. The symbols *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels, respectively

Our second specification considers the fact that changes in welfare policy and in attitudes toward welfare may have had different effects in different states. If, for example, the welfare rights movement was stronger in some states than in others, this could have resulted in higher growth in AFDC caseloads in the former states than in the latter. If states with stronger welfare rights movements also elected to participate in Medicaid earlier than did other states, this could result in an upward-bias in the estimate of the effect of Medicaid on AFDC caseloads. To account for this possibility, we include state-specific annual time-trends in addition to state, year and month effects, as indicated in Eq. 2 below:
$$ {\text{AFDC}}_{\text{it}} = \Upphi \,\left( {{\text{X}}^{\prime } \delta + \sum {\eta_{\text{s}} {\text{State}}\, \times {\text{Trend}}} } \right) $$
(2)

This specification includes all of the variables included in Eq. 1, but acknowledges not only that states may have different levels of AFDC caseloads, but that caseloads may also grow at different rates in different states. Results from this specification are contained in the second column of Table 2 and indicate that the introduction of Medicaid did significantly increase AFDC caseloads, though the magnitude of the effect is somewhat smaller than that estimated in the first column that did not include state-specific time trends.6

Table 3 changes the specification of the Medicaid variable. (Results in this and subsequent tables continue to include state-specific linear time trends in addition to state and year fixed effects.) The first Medicaid variable in the table (“Current year”) is equal to zero before the start of Medicaid, is equal to one during the year that the Medicaid program was started, but then returns to zero in subsequent years. In addition to the contemporaneous “Current year” variable, several lagged values of the Medicaid variable are also included. The variable labeled “Between 1 and 2 years”, for example, is equal to one in the year after Medicaid started but is equal to zero in all other years. The hypothesis here is that the availability of Medicaid will gradually increase AFDC caseloads as some households who were previously to the left of the AFDC notch join AFDC as they become aware of the new program.
Table 3

The introduction of Medicaid and AFDC caseloads (Log of annual AFDC caseloads by state: 1964–1974)

Independent variables

Lagged

Leading

Current year

0.034** (0.017)

−0.035** (0.017)

Between 1 and 2 years

0.089*** (0.030)

−0.020 (0.026)

Between 2 and 3 years

0.130*** (0.042)

0.003 (0.037)

Three or more years

0.158*** (0.051)

0.068 (0.055)

Real gross state product per capita/1,000

0.014 (0.066)

0.027 (0.066)

Real AFDC payment for family of 3/1,000

0.004 (0.004)

0.004 (0.003)

R-Sq

0.996

0.995

Regressions are estimated by ordinary least squares with standard errors (in parentheses) clustered by state. Regressions include state, year, and month fixed effects, as well as and state-specific linear time trends. The sample size is 6,732. The symbols *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels, respectively

Results indicate that Medicaid has a strong effect in increasing AFDC caseloads. Results indicate that the introduction of Medicaid increased caseloads by about 3% in the first year, 9% in the second year, and 13% in the third year. After a state has had at least 3 years to adjust to the introduction of the Medicaid program, the analysis suggests that Medicaid is associated with a permanent increase in AFDC caseloads of nearly 16%. This effect is strongly estimated. An F Test on the joint significance of the Medicaid variables in the first column of the table s0uggests that the hypothesis of no effect can be rejected at the 1% level (F(4,6604) = 122.66).

Clearly, the state-level results imply that the introduction of Medicaid significantly raised AFDC caseloads. Although the introduction of the Medicaid program led to a large increase in medical costs for states in the late 1960s, the fact that the introduction of Medicaid also significantly increased AFDC caseloads represents an additional cost of the program. Holding the number of AFDC recipients constant at the 1964 level (about 4 million recipients), we find that the Medicaid program added approximately $58 million per month in Federal and state medical payments to AFDC recipients from 1964 to 1974 (in 1967 dollars). This estimate is based on the fact that real medical vendor payments per AFDC recipient per month rose from $3.08 in 1964 to $17.55 in 1974. We compare this $58 million per month cost of Medicaid to the cost of supporting the additional families who were induced to participate in AFDC after the introduction of Medicaid. The estimates above suggest that the introduction of Medicaid increased caseloads by about 16% which is about 10% of the total increase in caseloads during this period. The introduction of Medicaid added 157,760 additional AFDC families (based on a base of about 986,000 families in early 1964). The maximum AFDC payment per family of three was $157 per month in 1967, implying that Medicaid led to an additional nearly $25 million dollars per month in AFDC payments. This implies that the indirect cost of extra AFDC participation was over 40% of the total inflation-adjusted cost of introduction Medicaid (including both the indirect cost and the direct cost of medical vendor payments).

The second column of Table 3 again changes the definition of the Medicaid variable. The “Current year” variable is defined the same way as in the first column, but subsequent variables refer to years prior to the introduction of Medicaid instead of years following the introduction. Our goal here is to test for the possibility of a reverse relationship between the introduction of Medicaid and state trends in AFDC caseloads whereby the trend in caseloads influences the timing of the introduction of Medicaid instead of the other way around. Most lead variables are statistically insignificant, implying no overall relationship between levels or growth in levels of AFDC caseloads by state and subsequent time of introduction of Medicaid. There is some indication, though, that Medicaid is introduced in a year with lower than average growth in caseloads. Since Medicaid was not introduced in response to rising caseloads, the possibility that the estimated effect of Medicaid on caseloads is biased up by reverse causality can therefore be ruled out. In fact, if the lead implementation variables in Table 3 are entered in the analysis with the lag variables, the coefficients on the lag variables change very little. For example, the coefficient on the variable indicating that Medicaid was introduced three or more years ago changes from 0.158 (with a standard error of 0.051) to 0.153 (with a standard error of 0.046).

5 The effect of Medicaid on AFDC participation

We also use the fact that states joined the Medicaid program at different times between 1966 and 1972 to estimate the effect of health insurance availability on the decision to participate in AFDC using individual level data. We use individual-level data on AFDC participation among female heads-of-household. This data set consists of the 1966–1974 March Current Population Surveys (CPS), and contains information about the demographic attributes of households, as well as an indication of whether the household participated in AFDC at all during the year prior to the year in which the survey was conducted.7 We limit the sample to female heads-of-household under the age of 65 who have at least one child under the age of 18 and who live in one of the 20 states (indicated in bold in Table 1) which are independently identified in the CPS in every year from 1966 to 1972.8 (The introduction of Medicaid may have also increased female headship, but this effect expected to be small.9) Table 4 presents summary statistics for this sample from the CPS. As Table 4 indicates, approximately 44% of the household-heads participated in AFDC during the previous year, where participation is defined from the CPS as occurring if a household had positive income from AFDC. Since AFDC eligibility is calculated monthly, the AFDC participation rate can be considered to be an estimate of the number of households who “ever participated” during the year, i.e. the number of households who participated in AFDC for at least one month. As Table 4 indicates, about 53% of the female heads work at least part-time. Approximately 53% have attended school for at least 12 years, over one-third are black, and over a half have at least one child under the age of six.
Table 4

Descriptive statistics (March current population survey, 1966–1974)

 

All years

1966

1974

Participates in AFDC

0.435 (0.496)

0.282 (0.450)

0.500 (0.500)

Participates in labor force

0.525 (0.499)

0.537 (0.499)

0.521 (0.500)

Real wage earnings

1,913.026 (2,397.064)

1,651.780 (2,042.398)

1,942.148 (2,570.611)

Medicaid exists

0.763 (0.425)

0.000 (0.000)

1.000 (0.000)

Real AFDC payment/1,000

0.204 (0.061)

0.193 (0.068)

0.200 (0.062)

Real gross state product per capita/1,000

3.789 (0.522)

3.385 (0.491)

4.046 (0.435)

Age/10

3.100 (0.581)

3.225 (0.544)

3.040 (0.594)

Number of children under age 18

2.587 (1.526)

2.880 (1.644)

2.323 (1.380)

At least one child under 6

0.541 (0.498)

0.514 (0.500)

0.540 (0.499)

Black

0.369 (0.483)

0.335 (0.472)

0.364 (0.481)

Central city

0.541 (0.498)

0.537 (0.499)

0.549 (0.498)

At least 12 years of education

0.530 (0.499)

0.441 (0.497)

0.586 (0.493)

Never married

0.115 (0.319)

0.000 (0.000)

0.163 (0.370)

Widowed

0.099 (0.299)

0.133 (0.340)

0.071 (0.258)

Divorced

0.388 (0.487)

0.392 (0.489)

0.406 (0.491)

Separated

0.398 (0.490)

0.475 (0.500)

0.359 (0.480)

This table reports weighted means from the March CPS. The sample sizes are 7,606 for all years, 607 for 1966, and 1,071 for 1974 respectively. Standard deviations are in parentheses

Below, we examine the probability that a female head participated in AFDC as a function of the demographic variables summarized in Table 4, as well as data on gross state product per capita and the maximum AFDC payment for a family of three We include year and state effects and state-specific linear time trends to control for other national and state-level influences on AFDC participation in the state. We allow state and year effects and state-specific linear time trends to vary by race, because of the large racial differences that existed in AFDC participation. As in the analyses of state caseloads, we allow arbitrary correlation of standard errors by state. Table 5 contains results using a linear probability model. According to the table, black women, women who live in a central city, women with children under the age of 6, and women who have not graduated from high school are significantly more likely to participate in AFDC. The “Medicaid exists” variable in the first two columns of the table is equal to the percentage of time during the year that a Medicaid program was in place, and varies from 0 to 1. We find that the introduction of Medicaid increased the chance that a female head of household participated in AFDC by about 6.9 percentage points. Relative to the percent of female heads participating in Medicaid (44 from Table 4), this is about a 16% increase. The second column of the table changes the definition of Medicaid to be the percentage of the time that a Medicaid program has been in place in the year of introduction (“Medicaid began this year”) and 0 otherwise. Variables are also included that lag this variable by one year (“Medicaid began last year”) and more than one year (“Medicaid began 2 years ago, and Medicaid Megan 3 or more years ago”). We find that women in states where Medicaid is available are significantly more likely to participate in AFDC. Specifically, we find that the introduction of Medicaid ultimately increases a single head’s chance of participating in AFDC by about 12 percentage points, a nearly 28% increase relative to the average percent (44) of female heads participating in Medicaid.
Table 5

Medicaid and AFDC participation among female heads of household (March current population survey: 1966–1974)

Independent variables

(1)

(2)

Medicaid exists

0.069** (0.025)

 

Medicaid began this year

 

0.071** (0.020)

Medicaid began last year

 

0.075* (0.037)

Medicaid began 2 years ago

 

0.106*** (0.025)

Medicaid began 3 or more years ago

 

0.123*** (0.030)

Real AFDC payment/1,000

0.184 (0.459)

0.240 (0.438)

Real gross state product per capita/1,000

0.274* (0.146)

0.320** (0.119)

Marital status (relative to never married)

 Widowed

−0.321*** (0.030)

−0.321*** (0.030)

 Divorced

−0.104*** (0.027)

−0.104*** (0.027)

 Separated

0.094*** (0.023)

−0.095*** (0.023)

Demographic characteristics

 Age/10

−0.397** (0.122)

−0.398*** (0.123)

 (Age/10)2

0.005** (0.002)

0.005** (0.002)

 Number of children under age 18

0.064*** (0.003)

0.064*** (0.003)

 At least one child under 6

0.058*** (0.011)

0.059*** (0.011)

 Black

0.433*** (0.018)

0.430*** (0.018)

 Central city

0.043* (0.021)

0.043* (0.021)

 At least 12 years of education

−0.188*** (0.019)

−0.188*** (0.019)

 R-Squared

0.265

0.265

The table reports estimates from ordinary least squares (OLS). Standard errors are in parentheses and are clustered by state. All regressions include race-specific state and year effects as well as race and state-specific linear time trends. There are 7,606 observations. The symbols *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels, respectively

The size of the Medicaid notch is expected to be larger for larger families. In addition, the size of the notch may vary with urban versus rural location if the ability to access providers varies according to location. The size of the notch may also vary according to race if the ability to access providers or health status varies according to race. Table 6 therefore interacts the variables measuring the number of children under age 18, race, and central city status with the “Medicaid exists” variable. Although the interactions on the race and central city variables are not statistically significant, the interaction between Medicaid and number of children under age 18 is statistically significant (at the 2% level) and positive, confirming that the size of the Medicaid notch does indeed appear larger for larger families compared to smaller ones.
Table 6

Medicaid and black, central city, and number of children interactions (March current population survey: 1966–1974)

Independent variables

(1)

(2)

(3)

Medicaid exists

0.069** (0.025)

0.030 (0.027)

−0.072 (0.055)

Black

0.420*** (0.045)

0.434*** (0.018)

0.425*** (0.017)

Central city

0.043* (0.021)

−0.016 (0.039)

0.044* (0.021)

Children under 18

0.064*** (0.003)

0.064*** (0.003)

0.026* (0.015)

Black × Medicaid

0.012 (0.043)

  

Central city × Medicaid

 

0.077 (0.053)

 

Children under 18 × Medicaid

  

0.051** (0.019)

Real AFDC payment/1,000

0.184 (0.459)

0.136 (0.461)

0.163 (0.457)

Real gross state product per capita/1,000

0.273* (0.146)

0.259* (0.134)

0.289* (0.150)

R-Squared

0.265

0.266

0.269

The table reports estimates from ordinary least squares (OLS). Standard errors are in parentheses and are clustered by state. All regressions include race-specific year and state effects, as well as race and state-specific linear time trends. All of the controls used in Table 6 are included, but not reported. There are 7,606 observations. The symbols *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels, respectively

6 Medicaid’s effect on labor supply of female heads of household

We now test whether increased AFDC participation resulted from mothers decreasing their labor supply to become eligible for AFDC after the introduction of Medicaid. This may have occurred if the introduction of the Medicaid program induced some households previously to the left of the AFDC kink in Fig. 2 to reduce hours of work and thus obtain a positive AFDC benefit as well as access to the Medicaid program. Table 7 tests whether Medicaid reduces a female head’s labor force participation. Although the coefficient estimate indicates that Medicaid reduces labor force participation, the standard error is large, and the hypothesis that Medicaid is not related to labor force participation cannot be rejected. A second test in the second column of the table estimates Medicaid’s effect on real wage and salary earnings. Results indicate no statistically significant effect on real wage and salary earnings. Considering the generosity of Medicaid benefits compared to health benefits available to non-AFDC female heads, it may be surprising that the Medicaid notch did distort women’s labor-leisure tradeoff as much as might be expected. It seems, rather, that Medicaid may have obtained some success in reaching its target population (AFDC eligible’s) without creating large efficiency costs. Since the introduction of Medicaid increased AFDC participation without detectably reducing female head’s labor supply, it is apparent that participation rates among eligible’s increased. Whether this occurred as more households chose to take advantage of the combined benefits of AFDC and Medicaid after the introduction of Medicaid or whether hospitals or other providers played a role in encouraging participation in AFDC and Medicaid after the introduction of Medicaid is not clear.
Table 7

Medicaid and AFDC eligibility, labor force participation, and earnings (March current population survey: 1966–1974)

Independent variables

Labor force participation

Real wage earnings

Medicaid exists

0.020 (0.032)

−110.652 (218.931)

Real AFDC payment/1,000

−0.352 (0.442)

−4,361.440 (2,876.387)

Real gross state product per capita/1,000

0.112 (0.173)

−462.043 (1,006.895)

R-Squared

0.254

The table reports estimates from ordinary least squares (OLS) except for column 3 which is estimated using a tobit model. There are 7,606 observations. Standard errors are in parentheses. Regressors not shown include all those in Table 6. Wage earnings are expressed in 1967 dollars. The symbols *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels, respectively

7 Conclusions

This paper indicates that the introduction of Medicaid played a significant role in explaining the large increase in AFDC caseloads during the late 1960s and early 1970s. Previously, Moffitt (1987) concluded that changes in AFDC program parameters, national unemployment rates, and demographic characteristics of households in the late 1960s were not sufficient to have generated the large increase in caseloads which took place. Rather he concludes that there was a “structural shift” in the relationship between these variables and AFDC participation, and points to the possibility of changes in attitude toward welfare and, in particular, a decrease in the stigma associated with welfare. Using the state-level data on AFDC caseloads, we find that the introduction of Medicaid accounted for approximately 10% of national growth in AFDC caseloads during this period. The individual-level data used supports this conclusion.

Footnotes
1

The "thirty-and-one-third" deduction was subsequently reduced by the Omnibus Reconciliation Act of 1981.

 
2

Author's tabulations from the National Center for Health Statistics' National Natality Survey, 1965. It should be noted that these tabulations actually underestimate the relative generosity of Medicaid, since most private health insurance plans in the 1960 s did not cover uncomplicated deliveries.

 
3

Since this paper was submitted for publication, we have learned of another paper (Strumpf 2011) that also uses the phased-in introduction of Medicaid to consider the effect on female labor supply. The Strumpf paper cites the working paper version of this paper and also finds no effect on labor supply of female heads of household as we do. The Strumpf paper does not consider the effect of the introduction of Medicaid on participation in welfare or on caseloads.

 
4

We use data on gross state product per capita (expressed in 1967 dollars) from the US Department of Commerce (various years). We use data on AFDC payment and need levels (also expressed in 1967 dollars) from a variety of sources including US House of Representatives (1989) and National Center for Social Statistics (1971).

 
5

Following the passage of Title II of the Economic Opportunity Act in 1964, for example, the Federal government provided funds to establish local welfare rights services and promote litigation on the behalf of potential welfare recipients. In 1968, the Supreme Court declared "man in the house" rules unconstitutional, thus making it illegal to deny benefits to a woman living with a man who is not contributing income to the family. In 1969, the Supreme Court also declared one-year state residency requirements for AFDC unconstitutional.

 
6

Since the Social Security Amendments of 1972 introduced the Federal Supplemental Security Income (SSI) program, we reestimated all models dropping years 1973 and 1974. Since some disabled mothers may have become eligible for SSI and Medicaid after 1972, the incentive to participate in AFDC may have changed. Consistent with the fact that we would expect few mothers of children under the age of 18 to have been eligible for SSI, dropping the years 1973 and 1974 from our analysis has little effect on the results. For example, the estimated coefficients on the Medicaid variables in columns (1) and (2) of Table 2 change to 0.088 (with a standard error of 0.030) and 0.042 (with a standard error of 0.021), respectively.

 
7

We begin our analysis of CPS data in 1966 since information on participation in AFDC was not available in the CPS before 1966. Except for 1966 and 1967, state general assistance programs are included with AFDC. It is hoped that the inclusion of state and time effects in the analysis will adequately control for this difference.

 
8

The analyses presented in the main body of this paper are limited to female heads of households, and do not include female sub-family heads. Prior to 1968, subfamilies are not explicitly identified in the Current Population Survey. Instead, the CPS indentifies the presence of a subfamily with an indicator value “Primary family with subfamily present”. (After 1967, the CPS provides an explicit variable specifically designating the subfamily). Accordingly, we have re-estimated all results in this paper to include female sub-family heads. For the years 1966 and 1967, we designate a female-headed subfamily if we observe this CPS indicator, an unmarried female older than 16, and the presence of own children. We do not find that our results are dramatically affected. For example, the coefficient on “Medicaid exists” in Table 5 changes from 0.069 to 0.062 (with a standard error of 0.022) if female sub-family heads are included. The coefficient on “Medicaid began 3 or more years ago” in the second column of Table 5 changes from 0.123 to 0.121 (with a standard error of 0.029).

 
9

Evidence on this effect is mixed. Decker (2000) found the introduction of Medicaid in the 1960s had some effect on female headship. Moffitt (1994) and Hoynes (1997) find that welfare does not increase the propensity to form female-headed households once state fixed effects are accounted for. Yelowitz (1998) does find some effects of Medicaid expansions on marriage.

 

Acknowledgments

We thank Hope Corman for helpful comments on a previous draft of this paper. The findings and conclusions in this paper are those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention.

Copyright information

© Springer Science+Business Media, LLC (outside the USA)  2011