Medicaid mysteries: transitional benefits, Medicaid coverage, and welfare exits.

The links between Medicaid and welfare exits are examined using longitudinal Medicaid program data. Few people who leave welfare get any sort of ongoing or transitional Medicaid protection. Moreover, it appears that many who are eligible for transitional benefits are not getting them. Finally, people with high expected medical costs appear to be less likely to leave welfare. The loss of Medicaid associated with leaving welfare probably does have an important deterrent effect on welfare exists.

Introduction

Both logic and anecdote suggest that Medicaid ought to have an important influence on welfare dynamics and movements to self-support. For every dollar spent in fiscal year 1987 providing cash support to families enrolled in the Aid to Families with Dependent Children (AFDC) program, another 69 cents was spent providing AFDC-related Medicaid protection (Committee on Ways and Means, 1989). And unlike cash benefits which phase out as income rises, Medicaid benefits are typically provided in full or not at all. Economic theory would certainly predict some important impacts of a benefit program of this magnitude. Moreover, welfare clients and administrators assert that the loss of Medicaid benefits is one of the major obstacles to leaving welfare and seeking market employment.

Largely to overcome the likely disincentives that loss of Medicaid coverage might have for those thinking about leaving welfare, Congress has enacted several types of transitional Medicaid protection. For a number of years, when someone left welfare for work, they were, in principle, eligible for at least 4 months of transitional Medicaid-only coverage. The Family Support Act of 1988 (Public Law 100–485) expanded transitional benefits further.

Yet there has been relatively little systematic empirical work that looks descriptively at the dynamics of Medicaid or the use of transition benefits. The only article documenting Medicaid dynamics in detail is Short, Cantor, and Monheit (1988). It does not examine transitional benefits at all. Only a couple of major papers empirically examine whether the loss of Medicaid actually deters exists from welfare. Blank (1989) found little evidence that the State level of Medicaid benefits or the presence of a medically needy program had any significant influence on welfare participation. But Blank's sample allowed her to separately identify only four States, and so she had little cross-state variation with which to identify Medicaid effects.

The only paper that used longitudinal data to examine the links between the loss of Medicaid and welfare exits is Moffitt and Wolfe (1989). They used a sophisticated model with individual data drawn from the Survey of Income and Program Participation (SIPP) to examine the impacts of Medicaid. They found very strong effects of private insurance (or the lack thereof) on AFDC participation. They estimated that AFDC caseloads would drop by 10 percent if all working female heads of households were guaranteed private insurance. And if all working female household heads were given coverage equivalent to that of Medicaid, the caseload was projected to drop by 16 percent. They also found that families with higher expected Medicaid benefits (presumably because of greater medical need) are more likely to participate in AFDC. The incentive effects of Medicaid appear to operate chiefly with people expecting very high medical costs.

In contrast to previous efforts, this article, drawn from a larger report Medicaid Mysteries: Medicaid and Welfare Dynamics (Ellwood et al., 1990), examines Medicaid dynamics and behavioral impacts using actual program data from Medicaid in two States, California and Georgia. We begin with a brief review of Medicaid eligibility provisions and consider whether or not welfare recipients can be expected to understand the transitional provisions. Next we discuss our unique data set. We then explore two main questions: First, what happens to Medicaid coverage once people leave cash assistance, and, in particular, how many people get some type of transitional or ongoing Medicaid-only coverage? Second, are families with high expected future medical costs less likely to leave cash assistance, and if so, is lack of medical coverage the reason such families are hesitant to leave?

Medicaid eligibility

Considerable confusion surrounds Medicaid eligibility. The program is often described as a system of medical care for the poor, implicitly suggesting that all poor persons might be eligible as they are under the food stamp program. In fact, Medicaid is extended to only a subset of the poor.

Medicaid eligibility for families has always been closely linked to eligibility for AFDC cash assistance. Welfare enrollees are automatically given Medicaid coverage in all States. But some families who do not qualify for cash assistance still get Medicaid-only protection. Yet even in these programs, for whole families to be covered, they must meet the so called “categorical” requirements for cash assistance. Only single-parent families, two-parent families where one is disabled or incapacitated, or, in States that choose to have an AFDC-Unemployed Parent (AFDC-UP) program, two-parent families where one has had significant work experience but is now unemployed, are potentially eligible.

For children, though, the links between cash assistance rules and Medicaid have been reduced considerably in recent years. Poor children are sometimes eligible even when their parents are not. States can now cover some children in families with incomes below the poverty level even if the family does not meet either the categorical or financial standards for cash assistance.

This array of coverage provisions causes confusion among even the most knowledgeable observers. Often policy analysts and even welfare mothers talk as though anyone leaving welfare immediately loses Medicaid coverage. That depends. Adults who leave because they are no longer single parents with children (because they marry or their children have grown) do lose coverage immediately. But those who leave because their earnings rose almost always could qualify for some transitional Medicaid coverage. And in many States they may qualify for Medicaid-only coverage under a medically needy program. The pure “Medicaid notch” where someone earns an extra dollar and instantly loses $2,000 in medical benefits, in principle, exists nowhere.

As of the time this article was written and for much of our sample period, families or children qualify under the following programs. (The Family Support Act of 1988 changed the transitional programs as of April 1, 1990):

AFDC

Medicaid is given to all those receiving benefits under the cash assistance program for single-parent families (or two-parent families with at least one incapacitated parent) in all States.

AFDC-UP

Medicaid is automatically granted to those receiving cash assistance under the program for two-parent families with at least one unemployed parent. UP coverage is available in 28 States.

4-month transitional

Four months of extended Medicaid benefits are offered to people who previously collected AFDC or AFDC-UP, but who lost benefits when their earnings rose. Mandatory for all States.

9-month transitional

Nine months of extended Medicaid benefits are offered to people with earnings who previously collected AFDC or AFDC-UP, and who still would have been eligible for cash assistance except for a rules change regarding the treatment of earned income after 4 months of work (starting in 1981 “$30 and 1/3” was eliminated after 4 months). Mandatory for all States.

Medically needy (MN)

Medicaid-only protection is extended to families who otherwise would have qualified for AFDC or AFDC-UP, whose income was above the AFDC income limit, but below the medically needy income limit. Available in 36 States.

Spend down

As part of a medically needy program, families who otherwise would have qualified for medically needy coverage except for excessive income can “spend down” their excess income on medical care (so that their income net of medical expenses is below the medically needy income limit). It is a mandatory part of a medically needy program.

Child only

Sometimes children can be covered for Medicaid even if their parents are not. States are now required to cover children under the age of 6 whose families meet the AFDC income and resource standards, even if their parents do not meet the categorical family criteria. States have the option of granting such coverage to all children under 21. States also have the option of covering infants and young children in families with income above the AFDC level but below the poverty level. In addition, States can provide coverage for foster and adoptive children, among others.

Other

A variety of other ways exist to qualify for Medicaid. Disabled and aged persons who qualify for Supplemental Security Income (SSI) get Medicaid automatically in most States. Pregnant women can qualify for Medicaid-only coverage. And there are a few other special coverage groups such as refugees.

These are the primary classifications, but there are more detailed categorizations possible. California has at least 50 different Medicaid eligibility codes. The categorical and financial rules for qualifying for each of these programs differ, sometimes rather considerably. Thus, just who is eligible for Medicaid and when can often be quite confusing, especially when people have earnings. Detailed examples of how the provisions work can be found in the full report.

Eligibility following Family Support Act

The situation has been changed somewhat by the Family Support Act of 1988. The 4 months (or 9 months) of transitional coverage for people leaving welfare because of excess earnings has been replaced with 12 months of transitional coverage. During the first 6 months, coverage at least as complete as Medicaid must be provided either through continued Medicaid eligibility or through so called “wraparound” protection whereby the State would pay any premiums, coinsurance, and deductibles for a family's private health insurance coverage offered by its employer.

During the second 6 months of coverage, States have a number of options, including limiting protection to acute care only and charging a partial premium of no more than 3 percent of gross monthly earnings for families with incomes from 100 percent to 185 percent of the poverty level. These provisions took effect in April 1990 and will expire in October 1998, at which time the 4-month transitional coverage would be reinstated.

In principle then, this 12-month transitional protection coverage is longer, more complete, and potentially more integrated with private insurance coverage than the 4-month or 9-month transitional coverage now offered. But the multiple State options after 6 months and the “wraparound” provisions will potentially increase the confusion regarding extended Medicaid coverage.

Welfare enrollees' understanding of Medicaid

Given Medicaid's complexity, virtually no one is likely to understand what they are eligible for a priori. Moreover, the medically needy and transitional programs almost always require separate, new applications. Enrollees (and many welfare workers) seem unlikely to know about them. When a woman finds a job, she often leaves welfare without explaining why. Her case is closed for administrative reasons when she fails to complete some form or report for an interview. In this case, no one in the welfare department would even know she has excess earnings and that she would still qualify for some coverage.

As is discussed in the full report (Ellwood et al., 1990), we observed considerable dynamics of families, with family members (even very young children) apparently leaving the unit for a few months, returning for a period, and sometimes leaving again. Such dynamics are likely to confuse the eligibility process still further. It is only in the case of earnings exits that transitional benefits are offered. But if most exits are for other reasons, former enrollees would discover that most of the time when they went off of welfare (because they remarried or reconciled, because there were no children at home, because they failed to comply with some administrative procedure, etc.), they immediately lost Medicaid. They might logically infer that losing welfare for any reason leads to the loss of Medicaid.

Thus, one is led to wonder what, if any, role the various Medicaid transitional coverage provisions play to help women move from welfare to work. It is possible that people who really need such coverage find it somehow. On the other hand, women may not know about such protections and thus hesitate to leave welfare. We do not have information on how many people know about various Medicaid options, but we can look to see how many people actually use them.

Tape-to-Tape data

For our analysis, we used the Health Care Financing Administration's “Tape to Tape” data which contains a 100-percent sample of all enrollees and all claims paid by Medicaid in several study States for each year starting in 1980. We used data for Georgia and California for the period 1980-86. These two States were chosen because they represent two extremes for coverage and benefits. Throughout this period, California has had the most generous Medicaid financial eligibility criteria in the 48 contiguous States. It has elected to extend welfare and/or Medicaid protection to many groups not eligible for AFDC cash benefits, including two-parent families with an unemployed parent (AFDC-UP), the medically needy, and poor children in a variety of settings, even when their parents are not eligible for Medicaid.

In contrast, Georgia has always had very low benefit levels and has extended coverage to relatively few optional AFDC-related groups, though there has been some expansion over time. In 1980, the only optional groups covered in Georgia were children in AFDC families who were age 18-20 and who were in school. Even this coverage was eliminated in 1982. Some expansion began in 1983, with Medicaid being provided only to prospective single women who were pregnant. In 1984, the State implemented the federally mandated coverage of infants in families satisfying AFDC financial criteria (including infants in two-parent families). Further expansion occurred in 1985, as the State moved to extend Medicaid-only coverage to all pregnant women meeting AFDC financial requirements, regardless of family structure. It also instituted a limited medically needy (Medicaid-only) program for children under age 18 (so-called Ribicoff children) and pregnant women.

Tape-to-Tape contains information on every person enrolled in Medicaid including all persons in AFDC, AFDC-UP, SSI, medically needy, or any Medicaid-only coverage group. For each person enrolled in a given month, the raw data contains all claims and associated diagnosis codes, and the person's age, sex, race (not available in California), and eligibility category (such as AFDC cash, 4-month transitional, etc.). Eligibility codes are quite detailed in California, but far less so in Georgia.

We created a multiyear longitudinal case file by linking individual's records over time and combining information on all the persons who were in the case for each month. Our main unit of observation was the mother. We selected all mothers in cases receiving AFDC or AFDC-UP cash aid at some time during the sample period. We then generated a longitudinal file covering the entire study period. For each month in each year, we knew whether each mother was enrolled in Medicaid, the eligibility category she fell into, the number and ages of all members of her family, and information on the family's Medicaid use during the month. In Georgia, we also knew the race of the mother. Ultimately, we had a data set of some 1,106,592 mothers in California and 183,140 in Georgia.

Unfortunately, although we had eligibility information for Georgia for 1980-86, we were able to process Georgia claims data for only 1980-84 in time for our analysis. Thus, when we describe patterns of eligibility dynamics, we can use Georgia and California through 1986. But when we estimated models examining the links between expected Medicaid expenditures and welfare dynamics, we used data through 1986 for California and through 1984 for Georgia.

From welfare to what?

We begin by examining what happens to the Medicaid coverage of people who leave welfare. We were particularly interested in seeing whether recipients got into either medically needy programs or into some sort of transitional aid. The obvious way to do so is to follow up on the Medicaid coverage of people after they leave welfare.

With the Tape-to-Tape data, we could take all the people who left cash assistance in a given month and examine their Medicaid status in subsequent months. Here, we will concentrate on people who left cash assistance in August 1985—that is, all the people who were getting cash aid in July 1985 who were not receiving it in August. Sample sizes are large enough that a single month gives very reliable information, and the choice of month does not influence the results much.

There are three obvious groups to look at in our data: AFDC enrollees in California, AFDC enrollees in Georgia, and AFDC-UP enrollees in California. (Georgia has no program for two-parent families with an unemployed parent.)

California AFDC

California has one unique temporary Medicaid-only classification called “Edwards versus Meyers” (EVM). As a result of a court case for which judgment was implemented in May 1982, which was in place in all subsequent years of our data, certain enrollees whose AFDC benefits were terminated continued to receive Medicaid coverage until the State could determine whether they were eligible for some other Medicaid-only program. In principle, this program applied chiefly to people who left for reasons other than increased earnings, because the earners would automatically have gotten 4 or 9 months of transitional assistance anyway. By 1986, most people stayed in EVM for only a month or 2 (the mean stay was only 1.3 months), although in earlier periods, EVM could last much longer.

What happened to people who left AFDC in California is shown in Table 1. Some 32.3 percent of those who left AFDC cash assistance in August 1985 returned to it within a year. In that State, 26.6 percent of AFDC enrollees who left were back within 3 months, and 30.7 percent were back within 5 months.

Table 1

Medicaid status of cases that left AFDC in California in August 1985

Medicaid status	August 1985 (1st month)	October 1985 (3rd month)	December 1985 (5th month)	July 1986 (12th month)	December 1986 (18th month)
All cases	Percent distribution
Returned to AFDC	0.0	26.6	30.7	32.3	32.1
Enrolled in AFDC-UP	0.0	0.8	1.1	1.3	1.5
Not receiving AFDC or AFDC-UP	100.0	72.6	68.2	66.4	66.4
Cases not receiving AFDC or AFDC-UP
Total	100.0	100.0	100.0	100.0	100.0
Medicaid only	36.6	14.2	9.3	8.7	7.4
Medically needy	1.5	4.7	6.3	6.0	5.4
Spend down	0.1	0.3	0.3	0.3	0.2
4-month transitional	5.4	5.9	0.6	0.9	0.6
9-month transitional	0.1	0.2	0.2	0.2	0.0
Edwards versus Meyers1	29.5	3.2	1.9	1.4	1.2
Other2	1.9	3.2	3.4	3.6	3.5
Not enrolled in Medicaid	61.4	82.8	87.4	87.7	89.3

Edwards versus Meyers is a temporary holding status created by court order in California only for selected cases losing cash assistance.

Other includes a variety of other coverage groups including a large refugee demonstration project.

NOTES: AFDC is Aid to Families with Dependent Children; UP is Unemployed Parent. All cases shown here were enrolled in AFDC cash in July 1985 and not enrolled in AFDC nor in AFDC or AFDC-UP in August 1985. Percents may not sum to totals because of rounding.

SOURCE: Health Care Financing Administration, Office of Research and Demonstrations: Data from the Medicaid Tape-to-Tape project.

People who did not return to welfare usually lost Medicaid quite quickly. In the first month after leaving AFDC, 61.4 percent of mothers in California had no Medicaid coverage. Moreover, virtually all of those who got Medicaid were in the Edwards versus Meyers temporary holding status. By October 1985 (the third month after leaving AFDC cash assistance), 82.8 percent of those who had not returned to welfare had no Medicaid coverage, and 87.7 percent had no coverage after a year.

The data reveal that transitional programs reach a very small minority of those leaving AFDC. On average, from 5 to 6 percent of those leaving had 4-month protection, and only 0.2 percent of those leaving AFDC in August 1985 were covered by the 9-month program (derived from Table 1). An equally small proportion of people leaving AFDC went into the medically needy program—no more than 6.3 percent in any month. Spend-down enrollment was minuscule—about the same size as for the 9-month transitional program.

Clearly, for the overwhelming majority of people, leaving AFDC in California meant leaving Medicaid. Still, these data must not necessarily be interpreted as meaning that most who leave welfare for work get no Medicaid-only protection. On the contrary, because previous research such as Ellwood (1986a) has shown that the vast majority of women leaving welfare do not leave exclusively because of an increase in earnings, and because most other reasons for leaving welfare will not give people access to any extended Medicaid coverage, one would expect only a minority of people leaving welfare to get any Medicaid-only protection. To interpret these results fully, one really needs to know how many people left AFDC because of earnings increases and were thus potentially eligible for transitional coverage.

Using annual data from the Panel Study of Income Dynamics, Ellwood (1986b) reports that only 20 percent of all exists were directly attributable to increased earnings, and in another 15 percent of exits, significant increases in earnings were coincident with another event, such as marriage. These figures could overstate the situation in California, though, because they are based on national data. One finds that earnings exits are less common in high benefit States such as California (Bane and Ellwood, 1983; Ellwood, 1986a). On the other hand, annual data is likely to show fewer earnings exits than monthly data as is used here. Thus, a 20- to 40-percent figure for the fraction of exits due to increased earnings may be reasonable. If so, only 1/4 to 1/2 of all women earning their way off of AFDC in California had some Medicaid-only coverage (after EVM placements) (derived from Table 1).

All women who left because of increased earnings theoretically qualified for 4 months or 9 months of transitional Medicaid or for ongoing coverage under the medically needy provisions. Why did they all not get or take advantage of these benefits? Some may have had coverage with their employers and did not need the protection. Still, we suspect that the most likely explanation is that very few recipients realized they could be eligible for coverage. They got it only if the welfare department knew they left welfare for work.

The State reports information based on administrative records about why people left welfare. Those reports show that 6 percent of people leaving AFDC during this time period did so because of excess earnings during the relevant months (Ku, 1989). This figure corresponds closely with the number getting transitional aid. But, the State also reports that 55 percent of exits were at “recipient initiative”—a catchall for persons who left for unknown reasons (Ku, 1989). This grouping surely includes many people who left welfare for work. Rather than report higher earnings to the welfare department, many clients simply stopped complying with administrative rules and were thus cut off of welfare. We expect that people would have chosen to get the transitional benefits if they had known about them because Medicaid service coverage is more comprehensive than all but the most generous employer plans. Obviously, if people do not even know about the existence of transitional coverage, then longer or more comprehensive transitional protection by itself cannot be expected to have much influence on welfare exits.

Note also that relatively few people went from welfare onto the medically needy and spend-down programs, either initially or in later months. The medically needy program is sometimes conceived of as a type of transitional program, one that allows people to move gradually from welfare to Medicaid-only protection and finally to full independence as their income rises. The spend-down provision is designed to help protect those with even higher incomes. The data in Table 1 suggest that medically needy program coverage helps only a small minority of those leaving welfare.

Georgia AFDC

In Georgia, State eligibility codes were far less detailed, allowing only a distinction between the AFDC population that was receiving cash assistance and those who were on some sort of Medicaid-only program. Those found in the latter category, in Table 2, include persons receiving 4- or 9-month transitional assistance along with a small group of other optional and mandatory Medicaid-only groups. Fortunately, when we follow cases longitudinally, it is possible to make some inferences about what type of coverage people have.

Table 2

Medicaid status of cases that left AFDC in Georgia in August 1985

Medicaid status	August 1985 (1 st month)	October 1985 (3rd month)	December 1985 (5th month)	July 1986 (12th month)	December 1986 (18th month)
All cases	Percent distribution
Returned to AFDC	0.0	16.1	22.2	27.7	28.3
Not receiving AFDC	100.0	83.9	77.8	72.3	71.7
Cases not receiving AFDC
Total	100.0	100.0	100.0	100.0	100.0
Medicaid only1	33.3	28.2	15.9	5.8	4.9
Other2	1.6	1.2	0.5	0.8	0.6
Not enrolled in Medicaid	65.1	70.6	83.6	93.4	94.6

Detailed breakdowns of Medicaid-only groups are not available in Georgia.

Other includes a variety of other coverage groups.

NOTES: AFDC is Aid to Families with Dependent Children. All cases shown here were enrolled in AFDC cash in July 1985 and not enrolled in AFDC in August 1985. Percents may not sum to totals because of rounding.

SOURCE: Health Care Financing Administration, Office of Research and Demonstrations: Data from the Medicaid Tape-to-Tape project.

Recidivism was slightly less common in Georgia than in California (Table 2). Roughly 16 percent of those leaving AFDC in August 1985 were back within 3 months; 27.7 percent within a year. More importantly, just as in California, the vast majority of people who left AFDC and did not return got no Medicaid-only protection. After 1 month, 65.1 percent had no coverage; after 3 months, 70.6 percent were without it; after 5 months, 83.6 percent had none; and within a year, 93.4 percent had no Medicaid coverage.

Still, there are striking differences between the two States. A considerably larger proportion of the cases leaving AFDC in Georgia got some form of Medicaid-only protection for the family. Though roughly the same proportions in the two States had protection in the first month (because so many cases initially were under Edwards versus Meyers in California), by the third month, 28.2 percent of cases leaving AFDC in Georgia had some Medicaid-only coverage in comparison to only 14.2 percent in California. In spite of the far less extensive array of optional coverage in Georgia, it appears that people leaving welfare are far more likely to have transitional support.

It would be helpful to know what Medicaid-only programs people were moving into in Georgia. Unfortunately, the lack of detailed eligibility codes makes coverage groups impossible to determine with certainty. Based on the timing of when coverage ends, one can make very rough estimates of the percentage getting 4-and 9-month transitional coverage. We estimate that approximately 12 percent of those leaving AFDC got 4-month coverage in Georgia—twice as many as in California. Nine-month coverage also appears to play a significant role in Georgia, with an estimated 10 percent of those leaving AFDC getting coverage.

Why did so many more people get transitional aid in Georgia? There is reason to believe earnings exits were more common in Georgia, because national evidence shows them to be more common in low benefit States. We doubt that earnings exits would be twice as common in Georgia. Alternatively, private coverage could have been less common in Georgia, leading more of those leaving welfare for work to seek transitional coverage. Still, we suspect that more of those leaving welfare for work in Georgia than those in California were learning about transitional aid.

Interestingly, just as in California, the number of people getting transitional aid corresponds closely with the State's reports of how many people are leaving because of excess earnings. Whereas roughly 6 percent of exits in California were reported by administrators as being a result of excessive earnings, 22 percent of exits were so reported in Georgia (Ku, 1990). This corresponds exactly with our estimate that 22 percent of those leaving AFDC got transitional assistance. Some evidence that Georgia finds more of those who leave welfare for work is based on the fact that only 39 percent of Georgia's exits are reported as being due to “recipient initiative,” while California classified 55 percent of exits that way. Thus, differences in administrative practices or information dissemination might explain the observed differences between the two States.

California AFDC-UP

We would anticipate a very different pattern of transitional receipt for AFDC-UP. Although there are no existing estimates on the number of exits from AFDC-UP that are earnings related, we would expect a much larger number than for AFDC. To qualify for the program, at least one adult with previous work experience must be unemployed. AFDC-UP cases tend to stay on cash aid for a much shorter time, and the numbers are very sensitive to economic conditions. Moreover, whereas family changes (marriage and reconciliation) are the most common reasons for leaving regular AFDC, AFDC-UP families start with two parents, so family changes are unlikely to move them off of welfare. If we focus on families that left AFDC-UP and did not go into regular AFDC, we would expect the majority (probably the large majority) of the exits to be earnings related.

The exit patterns and Medicaid status for AFDC-UP cases enrolled in July 1985 but off of welfare in August 1985 are shown in Table 3. Roughly 41 percent of cases leaving AFDC-UP in California continued on some Medicaid-only program in the first month. The level fell to 27.2 percent after 3 months and to less than 9 percent after 5 months. Initially, about one-half (21.9 percent) of the Medicaid-only coverage was provided by 4-month transitional benefits. The bulk of the remainder was Edwards versus Meyers coverage. By the third month virtually all of those still having coverage were enrolled in the 4-month transitional group (21.4 percent). The medically needy program covered only 3 to 5 percent of cases that left AFDC-UP.

Table 3

Medicaid status of cases that left AFDC-UP in California in August 1985

Medicaid status	August 1985 (1st month)	October 1985 (3rd month)	December 1985 (5th month)	July 1986 (12th month)	December 1986 (18th month)
All cases	Percent distribution
Returned to AFDC-UP	0.0	14.3	19.8	17.6	19.0
Enrolled in AFDC	0.0	5.1	7.1	10.6	11.7
Not receiving AFDC or AFDC-UP	100.0	80.6	73.1	71.8	69.3
Cases not receiving AFDC or AFDC-UP	Percent distribution
Total	100.0	100.0	100.0	100.0	100.0
Medicaid only	41.2	27.2	8.8	9.0	6.3
Medically needy	1.2	2.6	4.8	5.0	3.9
Spend down	0.0	0.1	0.0	0.0	0.1
4-month transitional	21.9	21.4	1.8	2.9	1.7
9-month transitional	0.1	0.0	0.0	0.0	0.0
Edwards versus Meyers1	18.0	3.1	2.2	1.0	0.6
Other2	9.4	12.4	13.8	13.7	9.7
Not enrolled in Medicaid	49.3	60.4	77.4	77.3	84.0

Edwards versus Meyers is a temporary holding status created by court order in California only for selected cases losing cash assistance.

Other includes a variety of other coverage groups including a large refugee demonstration project.

NOTES: AFDC is Aid to Families with Dependent Children; UP is Unemployed Parent. All cases shown here were enrolled in AFDC-UP cash in July 1985 and not enrolled in AFDC nor in AFDC or AFDC-UP in August 1985. Percents may not sum to totals because of rounding.

SOURCE: Health Care Financing Administration, Office of Research and Demonstrations: Data from the Medicaid Tape-to-Tape project.

That only 21.9 percent of those leaving AFDC-UP got 4-month coverage is rather surprising. As noted previously, one would expect the bulk of exits to be earnings related, thus making the family eligible for transitional coverage. Note that this figure does, again, correspond with administrative records which show that roughly 20 percent of exits from AFDC-UP were a result of earnings increases (Ku, 1989). If one-half of all exits from AFDC-UP were for earnings—and that seems very conservative—then at least half of those who were potentially eligible for temporary aid were not getting it. Again, we cannot judge how many had private coverage and thus may not have needed 4-month coverage, or how many simply didn't know about it or get it.

In summary, then, a small minority of all adults who left welfare got any ongoing Medicaid protection at all, in part because most exits were for reasons other than earnings, and extended coverage is not generally available in those cases. Nonetheless, a large number of those eligible for transitional benefits probably were not getting them. Some of these people probably had some private health insurance coverage, though we have no direct evidence on that. We suggest that a major reason people were not getting benefits that they were automatically entitled to is that welfare administrators often do not have information on whether people really left welfare because of increased earnings. There is no active search for such people, and recipients may be unaware that transitional Medicaid is available.

Of course, even those who got transitional coverage lost it after a few months. Our data cannot be used to examine how many people who left welfare and Medicaid had inadequate private coverage to replace it. But given the likely loss of Medicaid coverage and the low quality of private health coverage they were likely to find in their first job after leaving welfare, the potential loss of Medicaid should have been a prime consideration of recipients in deciding whether to seek or accept employment.

Transitional benefits were instituted in large part to encourage movements from welfare to work by reassuring recipients that they would continue to have some Medicaid protection. Yet it is hard to see how these benefits can be having any positive effect on welfare exits when so few people get them. We suspect that many of those recipients who are most fearful about losing Medicaid know little about benefits for which they actually qualify.

The Family Support Act of 1988 mandates a more generous set of transitional benefits than the ones in place when the data used in this article were generated. Nonetheless, these data strongly suggest that barring a change in administrative practice, few people will get transitional benefits when they leave welfare. We think even fewer will realize the transitional benefits have been improved, because we wonder how many people could possibly understand what they are eligible for now. Therefore, barring a major new information and outreach campaign, we expect that the transitional benefit provisions of the Act will be relatively ineffectual.

Does losing Medicaid deter exits?

The incentive effects of Medicaid are extremely hard to determine without experimental data. Nonetheless, our longitudinal Medicaid Tape-to-Tape data could be used to examine whether families that could anticipate unusually high medical expenses in the future are less likely to leave welfare.

Our basic methodology was straightforward. First, for a group of people on welfare, we estimated expected medical expenses for the subsequent 3 months (as if they had stayed on welfare). We then estimated a model of welfare exits where we included expected medical expenses as an independent variable. In other words, we modeled the odds that a person leaves welfare in a given month, say August 1985, as a function of demographic characteristics and predicted future medical expenses. The expectation was that people with higher expected medical costs would be less likely to leave welfare.

Our method for projecting future medical expenses is discussed in detail in the full report and will not be described in detail here. After considerable experimentation, we used expenses from the previous 6 months in predicting expenses for the subsequent 3 months. Our methodology allowed us to separately project adult and children's expenses so that both can be used in our models of welfare exits. We found we were able to predict a very low proportion (6 percent) of the variance in medical costs using past medical expenditures and demographic information for this relatively healthy group.

This limited ability to predict future expenses is typical. The results are quite consistent with other studies that have sought to predict future medical costs, though most previous work has focussed on projections for the elderly. Epstein and Cumella's (1988) review of the literature note that studies using past expenditures among the elderly to predict future expenditures typically have predictive power of about 6 percent, the same level we found. Newhouse et al. (1989) found that in a sample of nonelderly individuals, a combination of health measures and prior utilization could explain only 9 percent of future utilization. Indeed those authors claim that 85 percent of utilization is truly random and inherently unpredictable. If the vast majority of the variance in medical expenses is unpredictable, any recipient can rightly fear that medical catastrophe could strike.

Nonetheless, there is considerable variation in predicted future costs, so if the loss of Medicaid is a deterrent, we should see it most strongly for those with high expected medical costs. Our procedure also required that we have past Medicaid information on people for 6 continuous months in order to project future costs; thus, we were required to limit our sample to persons with at least 6 months continuous Medicaid enrollment.

In our exit model, we took persons enrolled in one month (say September 1986), and used a binary dependent variable indicating whether the person left welfare and stayed off for 3 continuous months. We chose not to count very brief moves off of welfare as a true exit because many such exits are caused by administrative churning. We used a logit model because we had a binary dependent variable and because it was computationally straightforward.,

Unfortunately, Tape-to-Tape data have only limited socioeconomic information. We were able to include the number of adults and children in the case, several measures of children's ages, variables capturing the number of months that the person was enrolled, and the log of expected medical expenses. The log was used because predicted medical costs are highly skewed, and because the logarithmic specification proved to be the most stable.

Thus, we estimated the following model:

where

Shown in Table 4 are the results for three samples of enrollees: those enrolled in AFDC in California in September 1986, those enrolled in AFDC-UP in California in September 1986, and those enrolled in AFDC in Georgia in September 1984. Means and standard deviations are shown in Table 5. The month of September was selected because it was a time when school age children would be returning to school, so that work might be more feasible for a parent. The latest year for which expenditure data were available in California was 1986, and 1984 was the latest usable year in Georgia.

Table 4

Coefficients and standard errors for logistic exit models of AFDC and AFDC-UP in California in September 1986 and AFDC in Georgia in September 1984

Variable	California AFDC	California AFDC-UP	GeorgiaAFDC
	Coefficients, (Standard errors)
Intercept	−1.224	0.225	−0.563
	(0.148)	(0.329)	(0.361)
LPRED.EXP3	−0.094	−0.097	−0.173
	(0.017)	(0.042)	(0.050)
NUM.ADULTS	0.126	−0.271	0.463
	(0.091)	(0.082)	(0.216)
NUM.CHILDREN	−0.141	−0.199	−0.127
	(0.017)	(0.032)	(0.030)
CHILD<13	-0.335	−0.030	−0.250
	(0.046)	(0.136)	(0.095)
CHILD<6	−0.176	−0.128	−0.228
	(0.037)	(0.086)	(0.067)
MOTHERS.AGE	−0.004	−0.008	-0.015
	(0.002)	(0.005)	(0.004)
MON.ENR.7-12	−0.093	−0.176	−0.059
	(0.012)	(0.025)	(0.022)
MON.ENR.13-24	−0.075	-0.101	−0.066
	(0.004)	(0.009)	(0.007)
NONWHITE	NA	NA	−0.523 (0.064)
Number of observations	148,896	28,831	39,986
−2 × Log likelihood	39,832	−3,555	12,480

NOTES: AFDC is Aid to Families with Dependent children; UP is Unemployed Parent; NA is not applicable.

SOURCE: Health Care Financing Administration, Office of Research and Demonstrations: Data from the Medicaid Tape-to-Tape project.

Table 5

Means and standard deviations of variables in logistic exit models of AFDC and AFDC-UP in California in September 1986 and AFDC in Georgia in September 1984

Variable	California AFDC	California AFDC-UP	Georgia AFDC
	Means,(Standard deviations)
PROB.EXIT	0.031	0.030	0.038
	(0.000)	(0.001)	(0.001)
LPRED.EXP3	4.58	4.96	4.77
	(0.84)	(0.66)	(0.57)
NUM.ADULTS	1.03	1.80	1.00
	(0.17)	(0.40)	(0.10)
NUM.CHILDREN	1.96	2.92	2.00
	(1.10)	(1.56)	(1.15)
CHILD<13	0.84	0.91	0.88
	(0.37)	(0.29)	(0.32)
CHILD<6	0.45	0.59	0.57
	(0.50)	(0.49)	(0.50)
MOTHERS.AGE	32.33	33.91	29.54
	(8.09)	(8.06)	(8.47)
MON.ENR.7-12	5.67	5.73	5.68
	(1.10)	(1.01)	(1.03)
MON.ENR.13-24	10.05	10.31	9.48
	(3.76)	(3.60)	(4.16)
NONWHITE	NA	NA	0.82 (0.39)

NOTES: AFDC is Aid to Families with Dependent Children; UP is Unemployed Parent; NA is not applicable.

SOURCE: Health Care Financing Administration, Office of Research and Demonstrations: Data from the Medicaid Tape-to-Tape project.

The results displayed are the coefficients in a logit model for each of the samples. The coefficients themselves are not readily interpretable in their current form, but their sign and rough statistical significance are. Increases in variables with negative coefficients are associated with reduced exits and vice versa. And coefficients with standard errors less than one-half of their absolute value are statistically significant at the .05 level.

The magnitude of effects can be seen more easily in Table 6 where the effect of changes in key variables on the probability of exit is shown. Shown is what happens to the probability of exit when only one variable is changed based on the logit results. One can see the marginal effects of particular variables while holding all others constant at their mean in much the same way that a regression coefficient does. For example, the average exit probability was 0.031, 0.030, and 0.038 for the California AFDC, California AFDC-UP, and Georgia AFDC populations, respectively. In families with predicted medical expenditures of $25 (well below the average), those figures rise to 0.035, 0.035, and 0.049.

Table 6

Marginal effects of individual variables on the monthly probability of exit for cases enrolled in AFDC or AFDC-UP in California and Georgia: Based on results of logistic exit models

Variable	California AFDC	California AFDC-UP	Georgia AFDC
	Monthly exit probability
Overage average	0.031	0.030	0.038
Expected medical costs for case
$25	0.035	0.035	0.049
$100	0.031	0.031	0.039
$200	0.029	0.029	0.035
$500	0.027	0.027	0.030
Number of children
1	0.035	0.043	0.043
2	0.031	0.036	0.038
3	0.027	0.030	0.034
4	0.023	0.024	0.030
Age of youngest child
Under 6 years	0.027	0.028	0.034
6-12 years	0.032	0.032	0.042
Over 12 years	0.044	0.033	0.053
Age of mother
20 years	0.032	0.033	0.044
30 years	0.031	0.031	0.038
40 years	0.030	0.029	0.033
50 years	0.029	0.027	0.028
Months previously enrolled in past 2 years
6	0.103	0.194	0.094
12	0.062	0.077	0.068
18	0.040	0.044	0.047
24	0.026	0.024	0.032
Race
White	NA	NA	0.057
All other	NA	NA	0.035

NOTES: AFDC is Aid to Families with Dependent Children; UP is Unemployed Parent; NA is not applicable. In generating this table, all variables were held at their means except the one(s) whose impact is being projected. In addition, the intercept of each of the models was adjusted slightly so that the grand mean was the same as the sample mean. (As logitistic models are nonlinear, plugging in mean values for all independent variables will not necessarily yield the mean value of the dependent variable.) This table applies only to people continuously enrolled for 6 months or more prior to the current month.

SOURCE: Health Care Financing Administration, Office of Research and Demonstrations: Data from the Medicaid Tape-to-Tape project.

We begin by considering the nonmedical variables. Having more children, having younger children, and being of a race other than white all significantly depress exit rates in all or most of the models. Holding all else constant, older mothers are somewhat less likely to leave welfare than younger ones.

These results are generally consistent with other work on welfare dynamics, such as Bane and Ellwood (1983), O'Neill, Bassi, and Wolf (1987), and Ellwood (1986a). The age of the youngest child is more significant here than in some other work, but most also control for marital status which we were unable to do. Other research has shown that never-married mothers are significantly less likely to leave welfare. Because mothers with very young children are presumably much more likely to have never married, the age of the youngest child may be picking up some of the effects of marital status. Similarly, the negative effect of mother's age is a bit surprising, but older welfare recipients tend to be less well educated, and we are unable to control for education in this model.

Very strong duration effects exist in our models. Families who were enrolled for only 6 months out of the previous 24 were three or four times more likely to leave regular AFDC in both Georgia and California than were families that had been enrolled for the entire 24 months. In the California AFDC-UP program, those who had been enrolled for only 6 months were nearly eight times more likely (0.194 versus 0.024) to leave welfare than families that had been enrolled continuously for 2 years.

These duration effects are very large by almost any standard. If we had been able to include observations for people enrolled for less than 6 months, we would have observed a U-shaped duration effect. When we run the model on all persons (omitting, by necessity, the medical variables), we find very low exit rates in the first few months after starting welfare (few people come on for just a month or two), followed by much higher rates in the period around 6 months, followed by falling exit rates as time increases beyond 6 months.

Previous research, notably Blank (1986), Ellwood (1986a), and O'Neill, Bassi, and Wolf (1987), found less of a duration effect. We have no clear explanation for these differences. Ours was a very different sample and we were considering monthly rather than annual exits over a far shorter time period than is usually used. Bane and Ellwood (1983) did find a strong decline in exit rates after 2 years. Another possibility is that these results come from our inability to control for many individual characteristics such as education, work experience, and marital status.

Our main interest, however, is in the effects of predicted medical costs on welfare exits. We find that predicted medical expenditures have significant negative effects on welfare exits in all three samples. In California, families in either AFDC or AFDC-UP with expected monthly expenditures of $200 are roughly 20 percent less likely to leave welfare than similar families expecting only $25 in expenses (.029 versus .035). In Georgia, the effects are much larger with the $25 group being 40 percent more likely to leave than those with higher expected costs.

But are these effects large or small? Having $25 rather than $200 in expected medical costs has about the same impact on exit probabilities as having one rather than three children. The difference is almost as great as the one caused by having a youngest child over age 12 versus a youngest child under age 6. And if the effect of high medical costs is chiefly to depress earnings exits (which are perhaps 20 to 30 percent of exits), then a reduction in the overall exit probability of 20 to 40 percent could represent a very large change in the probability of leaving via earnings.

Nonetheless, our results in general are not quite as strong or stable as those shown previously seem to imply. We tried running our models for a variety of years. Though the effects are almost always negative and usually significant, the impacts do differ from year to year in a nonsystematic fashion. The coefficients on medical care are not very stable. Moreover, using alternative functional forms (such as a linear rather than logarithmic one) for medical costs can often yield insignificant results, though the results are not much affected by the use of logit versus OLS methods.

Still, we do find significant effects both in the statistical and the policy sense in most of our models. That such effects were found at all is quite reassuring considering the limitations of this model. The large standard error of the estimates is surely the result of several features of the model. First, the variation in predicted medical expenditures is quite skewed, with a few cases having vastly larger expected expenditures than the others. This will tend to lead to a much larger standard error. Second, predicted expenditures is a very imprecise variable already. Less than 6 percent of the variance in future costs could be explained in our expenditure models discussed previously. We are, undoubtedly, doing a relatively poor job of measuring people's true expectations about the future in these models.

Accepting these results at face value, though, we cannot necessarily assert that the reason high expected costs depress exit rates is that people fear losing Medicaid. Indeed, there is an equally plausible, arguably, even more likely explanation. One would expect sick and disabled adults to be less likely to leave welfare simply because they are less able to work. As the sick and disabled are also likely to have higher than average medical expenses, any association between expected medical expenses and welfare exits might simply be capturing the incapacity of sicker people to work.

We sought to overcome this confounding effect by including projected costs for both children and adults separately. Disabled adults may be unable to work, but children are never expected to work, whether or not they are sick. An association between adult expenditures and welfare dynamics may be picking up the joint effect of both a fear of lost Medicaid and the impact of illness on the capacity to work. But, if we control for expected adult medical expenditures, any association between children's expenses and welfare exits seems more likely to represent a Medicaid effect.

Of course, it is possible that parents will be deterred from working because they want to stay at home with a sick child, not because they fear losing Medicaid for that child. So, even looking at the effect of expected expenses of children on exits may not give a true picture of the impact of Medicaid. But, presumably such an effect should be much stronger in single-parent homes than in two-parent ones, because one parent can work and the other look after the children. Thus, by comparing the results for AFDC-UP and AFDC families, one can get a further indication of whether the loss of Medicaid is the real reason for an association.

The results from models where adult and children's expected expenditures were both included are shown in Table 7. (Coefficients and standard errors are shown in Table 8.) High expected medical costs for children appear to have at least as large an effect on exit patterns from AFDC as high expected costs for adults. This children's effect cannot be a reflection of the fact that adult illness or disability might directly impede welfare exits because adult expenditures are controlled for separately. In all three samples the children's effects are significant and sizable.

Table 7

Marginal effects of individual variables on the monthly probability of exit for cases enrolled in AFDC or AFDC-UP in California and Georgia: Based on results of logistic exist models

Variable	California AFDC	California AFDC-UP	Georgia AFDC
	Monthly exit probability
Overall average	0.031	0.030	0.038
Expected medical costs for adults
$25	0.034	0.033	0.040
$100	0.029	0.029	0.037
$200	0.027	0.028	0.036
$500	0.025	0.026	0.034
Expected medical costs for children
$25	0.032	0.035	0.041
$100	0.029	0.028	0.032
$200	0.027	0.025	0.028
$500	0.026	0.021	0.024

NOTES: AFDC is Aid to Families with Dependent Children; UP is Unemployed Parent. In generating this table, all variables were held at their means except the one(s) whose impact is being projected. In addition, the intercept of each of the models was adjusted slightly so that the grand mean was the same as the sample mean. (As logitistic models are nonlinear, plugging in mean values for all independent viariables will not necessarily yield the mean value of the dependent variable.) This table applies only to people who have been continuously enrolled for 6 months or more prior to the current month.

SOURCE: Health Care Financing Administration, Office of Research and Demonstrations: Data from the Medicaid Tape-to-Tape project.

Table 8

Coefficients and standard errors for logistic exit models of AFDC and AFDC-UP in California in September 1986 and AFDC in Georgia for September 1984: Models with separate expenditures for children and adults

Variable	California AFDC	California AFDC-UP	Georgia AFDC
	Coefficients, (Standard errors)
Intercept	-1.011	0.830	-0.566
	(0.161)	(0.406)	(0.357)
LPRED.EXP3.ADULT	-0.103	-0.087	-0.055
	(0.022)	(0.063)	(0.047)
LPRED.EXP3.CHILD	-0.075	-0.168	-0.194
	(0.019)	(0.057)	(0.044)
NUM.ADULTS	0.179	-0.347	0.442
	(0.096)	(0.089)	(0.219)
NUM.CHILDREN	-0.122	-0.178	-0.071
	(0.020)	(0.038)	(0.034)
CHILD<13	-0.369	0.003	-0.254
	(0.047)	(0.157)	(0.095)
CHILD<6	-0.193	-0.062	-0.172
	(0.039)	(0.094)	(0.069)
MOTHERS.AGE	-0.004	-0.008	-0.013
	(0.002)	(0.006)	(0.004)
MON.ENR.7-12	-0.093	-0.188	-0.059
	(0.013)	(0.027)	(0.023)
MON.ENR.13-24	-0.075	-0.100	-0.067
	(0.004)	(0.010)	(0.007)
NONWHITE	NA	NA	-0.572 (0.064)
Number of observations	138,698	26,957	39,984
−2 × Log likelihood	36,454	6,375	12,468

NOTES: AFDC is Aid to Families with Dependent Children; UP is Unemployed Parent; NA is not applicable.

SOURCE: Health Care Financing Administration, Office of Research and Demonstrations: Data from the Medicaid Tape-to-Tape project.

Increased expected medical costs for children appear to be as large or larger for AFDC-UP families as for AFDC families in California. Because such families could send one person to work and still have a second adult available to care for the children, these results suggest that the association between medical costs of children and welfare dynamics may not reflect primarily a need to have an adult remain at home as a caregiver for a sick child.

Taken together, these results do provide significant evidence for the proposition that the lack of Medicaid protection is an important deterrent to leaving welfare for families anticipating high medical expenses. The critical policy question is how much higher would exit rates be if people faced no worries about medical expenses? Our data do not really allow us to say with any reliability at all what the effects would be. But, if we supposed the effect would be comparable to moving the expected medical expenses from the roughly $100 average that people face to a $25 average, then our models suggest that exit rates would increase by 10 to 20 percent in the various States and programs. This would generally be expected to translate into an equivalent long run reduction in the caseload. Note this estimate is remarkably close to the predicted 16 percent reductions in caseloads if all private employers provided coverage as thorough as Medicaid, which was predicted by Moffitt and Wolfe (1989) using a completely different methodology.

There are reasons to believe the 10- to 20-percent estimate is too high because some of the effect of higher expected expenses is a result of links between disability and work. Also, there are reasons to believe it is too low, because we cannot determine what role uncertainty plays in deterring exits and that is not incorporated in these models. Realistically, a model that looks only at program data can never tell us the overall effects of Medicaid. We need far more information on private health protection, job opportunities, and the like. The 10- to 20-percent figure is within the plausible realm given our results, but the results are somewhat tenuous.

Conclusion

Our examination of detailed longitudinal Medicaid Tape-to-Tape data suggest several key results. Very few people who leave welfare get any sort of ongoing or transitional Medicaid protection. Moreover, it appears that many who are eligible for transitional benefits are not getting them. Finally, we do find evidence that the loss of Medicaid has an important deterrent effect on leaving welfare. Taken together, these findings imply that both policies and administration of nonwelfare Medicaid programs deserve far more attention if the goal is to encourage independence from welfare. In drawing policy conclusions, one must recognize that we looked at only two States and that we were unable to look at any information other than that available on State Medicaid files. Still, a few observations seem important.

The present eligibility system of ongoing support is a morass. It seems extremely unlikely that enrollees or eligibility workers really understand what is available. As a result, many, perhaps most enrollees probably do not know that they are eligible for some transitional Medicaid if they go to work, and they certainly do not know what benefits they would get. More will have to be done to reach such people if the mandate of the Family Support Act is to be carried out that all people leaving welfare for work be informed of their ongoing coverage. And, if the goal is to ensure that everyone leaving welfare has some ongoing protection, extensions of the 4-month program, even with better outreach, will fall far short because most exits are not for earnings-related reasons.

There is significant evidence that the loss of Medicaid may be a deterrent for families with moderate to high expected medical costs to leave welfare. If the effect of a comprehensive insurance plan for the poor were similar to a reduction in the expected future medical expenses from the current average of $100 to $25, then we estimate exit rates would rise 10 to 20 percent. But our data and models are really not comprehensive enough to draw sweeping conclusions about the effects of expanding medical protection for those not on welfare.

Unfortunately, we cannot determine with our data how much effect the uncertainty surrounding medical expenses has on welfare exits. We cannot explore how much private and non-Medicaid public health insurance protection people have or could expect to have. Nor can we estimate what the effect will be of going from 4 months of transitional protection to 12 months, as called for in the Family Support Act. What we have seen here is clear evidence that many people are not getting transitional Medicaid they may be eligible for, and strong evidence that a lack of medical protection could be reducing exit rates.

Taken together our data suggest that loss of Medicaid is a deterrent to welfare exits and that transitional programs as implemented in the mid-eighties did little to improve the situation. It is possible that transitional benefits are inherently ineffectual. Lack of information may inevitably be a problem. Still, transitional benefits will always be time limited, by definition. If former recipients cannot get group-based medical coverage or cannot afford to buy individual protection, then they will always be at risk. They may be deterred from leaving welfare or discouraged from remaining independent once they have left, even if they have transitional benefits.

We strongly urge further research in this area. It may be that only a system of ongoing medical protection that is not linked to current or former welfare receipt would really offer the kind of protection that would encourage more recipients to leave welfare for work.