Uninsured spells of the poor: prevalence and duration.

The number of persons without health insurance is increasing. Although research has focused on the uninsured poor and the duration of spells without health insurance, less attention has been paid to the dynamics of spells without health insurance among those in poverty. Here it is shown that the typical uninsured spell is longer for the uninsured poor (roughly 8.3 months) than for the uninsured non-poor (roughly 6 months) and that the duration of spells has increased over time. In addition, more than 40 percent of the uninsured at a point in time are chronically uninsured and poor or near-poor.

Introduction

The number of uninsured persons has continued to increase, from roughly 30.5 million in 1979 to 41.7 million in 1996 (U.S. Census Bureau, 1996a; U.S. Census Bureau, 1997). Some of this growth reflects increases in the U.S. population, and although comparisons of longitudinal data on the uninsured are clouded by recent changes in survey methodology, analysis indicates that roughly two-thirds of these recent increases are attributable to increases in the uninsured rate, after controlling for other factors (McBride, 1996). Coverage rates under public insurance programs have increased significantly in recent years, masking a substantial drop in those with private insurance (Holahan, Winterbottom, and Rajan, 1995; U.S. Census Bureau, 1996b). The number of uninsured persons, particularly persons lacking private insurance, is increasing. Given these recent trends, the problem of uninsurance has remained an important issue. Evidence of this is the passage of incremental measures, such as the Health Insurance Portability Act of 1996 (popularly known as the Kennedy-Kassebaum reforms) and the Child Health Initiative passed as part of the Balanced Budget Act of 1997.

Roughly two-thirds of the uninsured live in families whose income is below 200 percent of the poverty line; roughly 27 percent live in poverty. The uninsured poor are of particular interest because insurance is likely to be unaffordable for them, even if it is available; they also lack the resources to pay for health care without health insurance (Holahan and Zedlewski, 1992). In addition, their lack of health insurance may increase the likelihood of delays in the utilization of health care, thus affecting their health status, raising health spending, and contributing to the problem of uncompensated care.

Recent policy initiatives have been in part directed at helping the uninsured poor obtain insurance, especially expansions of Medicaid, State-level efforts to expand coverage to the poor, and the recent passage of the Child Health Initiative. Other programs, such as the recent passage of welfare reform and the creation of the Temporary Assistance to Needy Families (TANF) program, may actually exacerbate the problem of the uninsured poor in the long run. Initiatives to help the uninsured poor will be successful at helping the uninsured poor if they reach that population and if insurance or health care becomes more affordable for the uninsured poor population. Thus, to determine whether these initiatives will be successful, especially in helping the chronically uninsured poor, it is first important to determine the size and scope of the uninsured poor population and the persistence of uninsurance and poverty in this population. Considerable research has documented the insurance status of people in poverty at a point in time, the duration of spells without health insurance, and the duration of poverty spells. However, little attention has been paid to the dynamics of spells without health insurance for people in poverty. Longitudinal data from the Census Bureau's Survey of Income and Program Participation (SIPP) are used to explore this question.

Previous Literature

Numerous studies have investigated the duration of poverty spells, just as studies have investigated the duration of uninsured spells, but no study has focused on simultaneous poverty and uninsured spells or on the relationship between uninsured spell durations and health status and health utilization. Swartz and McBride (1990), in one of the first studies to look at the duration of uninsured spells, found that one-half of all uninsured spells end within 4 months, and only 15 percent last longer than 24 months. In addition, they found that people whose incomes were below the poverty line at the beginning of their spell were less likely to exit from the uninsured spell in the first few months. In similar analyses of more recent data, the Census Bureau (1990, 1995, 1996c) concluded that the average uninsured spell was becoming longer over time. Swartz, Marcotte, and McBride (1993a) confirmed the results of the previous studies and found that the distribution of durations of left-censored uninsured spells is not very different from the distributions of spells that began during the survey period. Swartz, Marcotte, and McBride (1993b) and McBride and Swartz (1991) found that monthly family income, education, and industry of employment had the biggest impact on the likelihood that a spell would end.

Although these studies focused on the duration of new uninsured spells, subsequent studies have focused on uninsured spell durations from different perspectives. McBride (1994), using a longitudinal measure of the duration of uninsured spells at a point in time, found that 75 percent of the uninsured were in spells that would last longer than 1 year, and only 3.5 percent were in spells that would last less than 5 months. Looking instead at the “uninsured flow” (that is, the number of persons experiencing at least 1 month without health insurance over a period of time), the Census Bureau found that roughly 20-30 percent of the population experienced at least 1 month without health insurance, depending on the period of time chosen to observe persons (U.S. Census Bureau, 1992, 1996c; McBride, 1996). These findings, of roughly 60-66 million persons experiencing a spell of uninsurance during a specified period, have been cited as estimates of the prevalence of the uninsured problem. In contrast to these measures of the uninsured flow, estimates of the uninsured stock (the number of persons remaining uninsured over an entire period of time, such as 1 year) yield lower estimates of the uninsured because only those who are uninsured throughout the entire period are counted as uninsured. For example, only 6-7 percent of the population (roughly 15-18 million persons) were uninsured for an entire year in recent years, and only about 4 percent of the population were uninsured for an entire 28-32 month period (U.S. Census Bureau, 1992, 1994). These findings of a smaller incidence of long-term uninsurance, as measured over a 28-32-month fixed window of time, have also been used as estimates of the “chronically uninsured” population.

Literature on Spells of Poverty

Concern about the contribution of Federal programs to the perpetuation of poverty over time has led to an explosion of research into estimating the duration of poverty spells since the early 1970s. Levy (1977), who defined the persistently poor as people with incomes below the poverty line for at least 5 years between 1967 and 1973, found that between 10 and 11 million persons, or 40-45 percent of the poor at a given point in time, were persistently poor. Using a longer time frame of 8 years (1967-75), Coe (1978) found that 3 million persons, or 12 percent of the poor at a point in time, were persistently poor. Subsequent studies confirmed the general scope of these results (Duncan, Coe, and Hill, 1984; Hill, 1981; Rainwater, 1981). However, Bane and Ellwood (1986), using techniques similar to those of the Swartz and McBride (1990) study focusing on uninsurance, found that only 12 percent of the non-elderly beginning spells of poverty would remain in poverty for 9 or more years, but that 51.5 percent of the non-elderly poor at a point in time are in the midst of a spell lasting 9 years or more. Ruggles (1990) found that 58 percent of poverty spells ended in the first 4 months, 81 percent were completed in the first 8 months, and only 5.3 percent of spells lasted 28 months or more. A comparison of all these studies suggests that a spell of poverty might typically be shorter than a spell without health insurance, although not appreciably so. As Ruggles (1990) explains, none of the previous studies accounted for the problems of left-censoring of spells.

Limitations of Previous Studies

Although these findings provide some evidence that people in poverty are more likely to experience long uninsured spells, the findings for the most part do not provide leads into the problem of uninsured spells for the poor. There are several reasons for this. First, none of the studies already cited looked specifically at coincident uninsured and poverty spells, and we know little about the characteristics of these individuals and how they obtain health insurance coverage, if they are able to.

Second, the Swartz and McBride (1990) study provides stratifications of the estimates of the duration of uninsured spells only by poverty status at the beginning of the uninsured spell. This means that many of the people who were in poverty at the beginning of their uninsured spell may have exited from poverty before—or at the same time as—they exited from an uninsured spell. Thus, the estimates of uninsured spell durations do not provide evidence about the durations of spells for the chronically poor, nor do they provide leads into the dynamics of poverty and uninsurance.

Finally, the relevance of earlier estimates of uninsured spell durations is limited because of the use of older data. Although these data are not seriously outdated (in comparison to other available data), legislation that went into effect in recent years could possibly affect the finding of many short spells without health insurance. For instance, the Consolidated Omnibus Budget Reconciliation Act (COBRA) of 1985 could eliminate many short uninsured spells because it requires that employers offer health insurance coverage to employees beyond the date of termination from their present job.

Data

This project uses data primarily from the 1990 panel of the SIPP, a multipanel, longitudinal survey conducted since 1984 by the Census Bureau. Participants in the 1990 panel were initially interviewed in February 1990 and were then interviewed every 4 months until April 1992. Data from the 1987 SIPP panel are also shown here in a few places for longitudinal comparisons. The SIPP is a nationally representative sample of adults (including the elderly) that provides detailed sociodemographic information as well as information on month-by-month fluctuations in household and individual income, health insurance status, labor force status, and participation in government-sponsored programs such as Women, Infants, and Children; Medicare; and Medicaid. The information is collected for the individual and the individual's household (including children under the age of 15 years) for the 4 months preceding each interview. The full 1990 panel consisted of eight interviews, covering a period of 32 months. The initial 1990 panel sample included approximately 23,600 households (or about 54,000 persons), but additional households and persons were added to the sample if they entered or exited from the original SIPP household.

This study is based on a large file that was created by merging data from three sources:

Extract from the 1990 Full Panel Research File (FPRF). The Census Bureau produces an FPRF by merging data from all eight interviews for every respondent. The FPRF includes data on basic demographic and economic variables, as well as longitudinal information on health insurance status. Because the 1990 FPRF contains only a subset of the data collected through the core questionnaire on each wave, relevant information on health insurance coverage, health status, and health utilization was not attached to the FPRF. Thus, two additional extracts were created and attached to the FPRF.

Extracts from 1990 Topical Module Files. Waves 3 and 6 of the 1990 panel included topical modules that included questions on health status and health utilization. These data were extracted from the 1990 wave files and merged with the FPRF extract.

Extracts from Core Wave Files. Because the FPRF does not include all of the information collected on health insurance coverage in each wave interview, additional variables were extracted from each wave file and merged with the other data. This extract included data on the exact source of employer-provided insurance coverage (e.g., present or former employer), the depth of insurance coverage (e.g., family or individual coverage), and the employer's contribution to insurance coverage.

The resulting large panel file includes extensive detail on the health insurance coverage, health status, and health utilization of a representative sample of persons in the United States.

Methods

Although the various studies just cited have produced estimates of uninsured spell durations for the entire population of uninsured persons, estimates of uninsured spell durations for the poor are of inherent interest, especially to test the hypotheses that these spells are likely to occur at the same time and that uninsured spells for the poor will be longer than uninsured spells for people who are not in poverty.

The variables of interest in this research are spells of poverty and spells without health insurance. Spells are defined as beginning when persons report that they have no insurance coverage (or, for poverty spells, when they enter poverty) and ending when they acquire insurance coverage (or their income rises above the poverty line). Insurance coverage is defined here to include any type of public insurance (e.g., Medicaid, Medicare, the Civilian Health and Medical Program of the Uniformed Services [CHAMPUS], or public employer insurance) or private insurance (employer-provided or other private coverage). Thus, only those who have no source of insurance are defined as “uninsured.” Initially, the official poverty line (Social Security Administration, 1996) was used for identifying persons in poverty, although alternative definitions of poverty were also applied to test the sensitivity of the results to this definition. Ruggles (1990) has pointed out that month-by-month estimates of poverty spells seem short because of the volatility of income flows and because the poverty line may be too low. To explore this problem, the sensitivity of the results to the definition of a “poverty spell” was explored by also looking at persons slightly above the poverty line, because this group is also likely to be at risk of financial difficulty from a health care expenditure. (Note that a $10,000 medical bill still represents 7 months of gross income for a family of three with an income equal to 150 percent of the poverty line.)

Conventional non-parametric methods (particularly life-table estimates produced by the SAS program) were used here to estimate the duration of new uninsured spells, controlling for the problem of right-censoring of spell lengths (Allison, 1995). The procedures developed by McBride (1994) were used to compute an estimate of the distribution of spell durations for the population of uninsured at a point in time. Fortunately, the process of estimating the total duration of uninsured spells at a point in time yields a more precise estimate of very long spells (that is, spells that have already lasted at least 24 months) here, because the SIPP sample includes spells that have lasted at least that long.

Uninsured Poor

A greater percentage of the uninsured are poor, compared with the insured population. Almost 50 percent of the uninsured population was poor or near-poor in 1991, in contrast to roughly 17 percent of the insured population. At a point in time in 1991, roughly 27 percent of the uninsured were poor—that is, their family income was below the official annual poverty line (Table l). In addition, another 21.9 percent of the uninsured had incomes placing them close to poverty—between 100 and 150 percent of the poverty line. In contrast, only 9.3 percent of the insured population was in poverty, and only 17.3 percent of the insured population had family incomes below 150 percent of poverty. Children (under age 18) are overrepresented in the poverty and uninsured populations, whereas the elderly are underrepresented in both groups. Nevertheless, the findings suggest that a lower proportion of uninsured adults age 18-64 are poor (24.9 percent) or near-poor (19.9 percent) than either children or the elderly. The comparisons in Table 1 indicate little change in the distribution of uninsured persons by poverty status between 1988 and 1991.

Table 1

Distribution of Uninsured by Poverty Status: 1988 and 1991

Poverty Ratio1	December 1988		December 1991
	All Uninsured	Uninsured Adults	All Uninsured	Uninsured Adults	All Insured
		Number in Thousands
Total Persons	29,238	20,409	31,308	23,219	192,621
		Percent Distribution
Total	100.0	100.0	100.0	100.0	100.0
Less than 1.0	26.0	22.7	27.0	24.9	9.3
Less than 0.5	8.3	6.9	8.1	7.7	3.3
0.5-1.0	17.7	15.7	18.9	17.1	6.0
1-1.5	21.8	19.0	21.9	19.9	8.0
1.5-2.0	15.8	15.6	16.7	16.5	9.5
2.0-3.0	17.5	19.5	18.9	20.3	21.8
3.0 or more	18.9	23.3	15.5	18.5	51.4

Ratio of annual family income to annual poverty line.

NOTE: Columns may not add to totals shown because of rounding.

SOURCE: McBride, T.; data from U.S. Census Bureau, 1987 and 1990 Survey of Income and Program Participation panels.

As previously noted, long uninsured spells are more of a concern for the poor because the threat of a health episode increases over time, and the likelihood of being able to afford health insurance decreases the longer the person is poor. To demonstrate this, Table 2 shows the distribution of uninsured spell lengths found in the 1990 SIPP panels. The results show that, for 1990, roughly 45 percent of new uninsured spells were completed within 4 months, and 60 percent were completed in the first 8 months. The median duration for a spell was 7.1 months. Close to 20 percent of spells lasted more than 2 years, and roughly one-third (31.6 percent) lasted more than 1 year. To see if the duration of spells is changing over time, the results from the 1990 panel are compared with results found using the 1987 panel. The evidence suggests that uninsured spells may be getting longer over time, at least until recently. The median spell length found in the 1990 panel was 7.1 months, compared with 6.3 months in the 1987 panel. On the other hand, the difference in median spell durations was not statistically significant, and new spells in 1987 were slightly more likely to last more than 2 years (although again, this difference is not statistically significant).

Table 2

Distribution of Uninsured Spell Lengths: 1987-89 and 1990-92

		Duration of New Uninsured Spells			Uninsured at a Point in Time (Total Spell Duration)
Spell Length	1987 Panel	1990 Panel			1987 Panel	1990 Panel
	All Persons	All Persons	Poor	Non-Poor	All Persons	All Persons	Poor	Non-Poor
		Percent Distribution
Total	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0
Less than 5 Months	47.0	44.9	40.6	47.2	3.5	2.6	1.5	3.0
5-8 Months	15.5	15.1	15.4	14.7	11.6	8.4	4.9	9.6
9-12 Months	8.1	8.4	7.8	8.6	9.8	7.2	5.2	7.9
13-16 Months	5.2	5.4	5.3	5.4	8.1	7.3	7.1	7.4
17-20 Months	3.4	4.4	5.2	4.0	6.9	6.6	6.5	6.6
21-24 Months	0.8	2.6	4.0	2.3	6.1	5.8	5.9	5.7
More than 24 Months	20.1	19.2	21.8	17.7	53.9	62.2	68.8	59.8
Median Spell Length (Months)	6.29	7.14	8.25	6.01	—	—	—	—
Total Number of Spells	5,786.00	11,971.00	2,941.00	8,365.00	—	—	—	—
Spells with Observed Ending	3,403.00	7,053.00	1,626.00	5,124.00	—	—	—	—
Number of Right-Censored Spells	2,383.00	4,918.00	1,315.00	3,241.00	—	—	—	—
Percent of Spells Right-Censored	41.20	41.10	44.70	38.70	—	—	—	—

NOTES: To compute an estimate of the distribution of spell durations for the population of uninsured at a point in time, samples of respondents were drawn from points in time (December 1988 and December 1991) from the 1987 and 1990 Survey of Income and Program Participation panels. Total spell durations to the end of a person's spell were then estimated for each person. The methods for obtaining these estimates are described in detail in McBride (1994). This process yields a good estimate of the chronically uninsured because the sample is drawn after observing people for more than 2 years. Thus, people whose spells last for more than 1 or 2 years (and people who are always uninsured during the panel) can be observed in this state. Columns may not add to totals shown because of rounding.

SOURCE: McBride, T.; data from U.S. Census Bureau,1987 and 1990 Survey of Income and Program Participation panels.

The finding that most uninsured spells are very short has led to considerable confusion about the uninsured (McBride, 1994; Swartz, 1994). In particular, the distribution of spell durations for new uninsured spells is considerably different from the distribution of durations of spells in progress. The long-term or chronically uninsured are more likely to be represented in the point-in-time population of uninsured (McBride, 1994). To see how much difference this effect makes, estimates of the duration of uninsured spells were obtained for the stock of uninsured at a point in time using the SIPP sample; the results are presented in Table 2. As in McBride (1994) the findings indicate that most of the uninsured at a point in time are experiencing long uninsured spells: 62.2 percent of the uninsured are in spells that last more than 2 years, and 81.9 percent in spells that last more than 1 year. In addition, a slightly larger proportion of the point-in-time uninsured population in 1987 were in spells that would last more than 1 year (compare 81.9 with 75.1 percent).

Duration of Uninsured Spells Among the Poor

People who are uninsured and poor for a long period of time may face a higher risk of financial problems associated with health care, compared with those who are not chronically uninsured and poor. The evidence presented in Table 2 implies that the poor are more likely to have longer uninsured spells than the non-poor, where poverty status is determined at the beginning of the uninsured spell. For instance, the median uninsured spell duration among the poor is roughly 8.3 months, compared with only 6.0 months for the non-poor uninsured, a statistically significant difference. In addition, among the population that was poor in the first month of their uninsured spell, 21.8 percent were starting spells that lasted more than 2 years, compared with only 17.7 percent of the non-poor uninsured. Most of the uninsured poor (88.4 percent) at a point in time were in uninsured spells that would last more than 1 year.

As reported in Table 1, almost one-half of the uninsured are chronically uninsured persons with low incomes. Roughly 49 percent of the uninsured at a point in time had incomes below 150 percent of poverty, and 27 percent were poor at a given point in time. Figure 1 shows that 23.7 percent of the uninsured at a point in time were both in poverty and in uninsured spells that would last more than 1 year, and 42 percent were in uninsured spells that would last more than 1 year and had incomes below 150 percent of poverty.

Figure 1

Distribution of Uninsured, by Spell Length and Poverty Status

Another way that researchers have studied the question of chronic uninsurance is to look at the uninsured stock, the number of people who remain uninsured for a specified period of time. Using this measure, the findings suggest that 62 percent of the uninsured poor at a point in time in 1991 were uninsured for all 12 months of 1991, in contrast to 56 percent of the non-poor uninsured who were uninsured for the entire year. Although this provides further evidence that the poor face more persistent uninsured spells than the non-poor, the difference in the proportions is not large.

Coincidence of Uninsurance and Poverty

As noted in the literature review, none of the previous studies have looked specifically at coincident uninsured and poverty spells. Table 3 presents evidence on people with those spells, showing the number of months people were without health insurance and poor during the year (i.e., in a coincident uninsured and poverty spell). The results indicate that 30.1 percent of the uninsured poor at a point in time in 1991 had no insurance and were poor for the entire year—their income never climbed above the poverty line (measured by the monthly poverty definition) in any month, and they never had insurance in any month.

Table 3

Months Without Insurance and in Poverty: 1991

Insurance and Poverty Status	Uninsured in December 1991			Insured in December 1991
	All	Poor	Non-Poor
	Percent
Months Without Insurance
0	0.0	0.0	0.0	92.1
1	2.9	1.5	3.4	1.0
2-4	11.6	9.7	12.4	3.8
5-8	18.6	16.6	19.3	2.1
9-11	9.3	10.3	8.9	1.0
12	57.6	62.0	56.0	0.0
Months Without Insurance and Poor
0	46.0	2.0	62.2	96.8
1	7.7	1.4	10.0	0.9
2-4	17.8	12.3	19.9	1.6
5-8	14.1	32.1	7.5	0.6
9-11	6.3	22.2	0.5	0.1
12	8.1	30.1	0.0	0.0
Months Without Insurance and Poverty Ratio Less Than 1.5
0	28.6	0.5	39.0	95.6
1	5.8	1.4	7.4	0.9
2-4	18.4	9.7	21.6	2.3
5-8	18.6	19.6	18.2	1.0
9-11	12.0	17.9	9.9	0.3
12	16.7	51.0	4.0	0.0

NOTES: To classify persons into categories, the annual definition of poverty is used. To determine if a person is poor in a given month, the monthly definition of poverty is used. Columns may not add to 100.0 because of rounding.

SOURCE: McBride, T.; data from U.S. Census Bureau, 1990 Survey of Income and Program Participation panel.

As previously described, Ruggles (1990) has pointed out that month-by-month estimates of poverty may overestimate the extent of persistent poverty because of the volatility of income flows. To counter this problem, estimates are presented that use the monthly definition of poverty and a threshold set at 150 percent of the poverty line. This shows that many persons (51 percent of the uninsured poor) will remain uninsured and poor or near-poor for the entire year. Roughly 61 percent of the uninsured with incomes above the poverty line spent some months with income above 150 percent of poverty and without health insurance. These findings suggest that many persons spend several months vulnerable to the risks of high health spending, if vulnerability is more broadly defined to include incomes below 150 percent of the poverty line.

Characteristics of the Uninsured Poor

To develop policies that are particularly suited to solving the problems of the uninsured poor, it is important to understand more about their characteristics. Are the uninsured poor similar to the uninsured non-poor or not? What do the differences between the groups suggest about appropriate strategies for meeting the needs of the uninsured poor?

Employment Characteristics

Health insurance in the United States is provided primarily through employers, so it is not surprising that the vast majority of the uninsured poor are people who are not employed. Only 29.3 percent of the uninsured poor were employed in December 1991, and only 15.3 percent were employed full time (Table 4). This is in sharp contrast to employment rates among the non-poor: 51.2 percent for the uninsured and 52.8 percent for the insured. Table 4 also indicates that almost one-third of the uninsured poor not in the labor force are children. Overall, 83.5 percent of the insured poor are not in the labor force, also not surprising because this group is dominated by people insured through public programs, which typically provide coverage only to those who cannot work.

Table 4

Characteristics of the Uninsured at a Point in Time, 1991

Characteristic	Uninsured in December 1991		Insured in December 1991
	Poor	Non-Poor	Poor	Non-Poor
	Percent
Employment Status
Employed	29.3	51.2	12.2	52.8
Full Time	15.3	35.3	7.2	42.3
Part Time	14.0	15.9	5.0	10.6
Unemployed	11.4	9.5	4.4	1.6
Out of Labor Force	59.4	39.3	83.5	45.6
Age 0-17 Years	31.9	23.3	44.9	21.8
Others	27.5	16.0	38.6	23.8
Occupation1
Professional or Technical	31.3	35.3	39.1	61.0
Service	31.2	23.2	28.9	11.8
Laborer	21.9	20.7	18.1	14.4
Other	15.5	20.7	14.0	12.8
Industry1
Agriculture/Mining	13.6	9.7	11.9	6.2
Manufacturing	9.3	10.4	12.9	19.5
Transportation/Construction	12.2	14.6	4.5	11.7
Retail Trade	32.2	29.2	24.8	18.4
Service	32.4	33.2	41.3	38.0
Public Sector	0.4	3.1	4.6	6.3
Age
0-17 Years	32.4	24.7	45.6	22.9
18-24 Years	16.8	18.2	8.4	8.6
25-44 Years	33.4	39.7	22.9	33.2
45-64 Years	17.1	17.2	10.0	20.8
65 Years or Over	0.4	0.2	13.1	14.5
Region
Northeast	9.5	14.2	18.7	21.5
South	55.5	41.6	37.4	31.3
Midwest	17.4	19.4	26.4	26.8
West		17.7	24.9	17.5
20.4
Race
White	68.2	79.9	63.4	87.3
Black	26.5	14.3	31.8	9.6
Other	5.3	5.8	4.9	3.2

Percent of workers.

NOTE: Columns may not add to 100.0 because of rounding.

But, as seen in Table 4, employment is not a guarantee of insurance coverage. For various reasons (e.g., low wages on the job, working for a small employer, pre-existing conditions), many workers do not have insurance through their jobs. Although studies have shown that one of the most important characteristics of a job is firm size (McBride, 1996), the FPRF does not contain this variable. Nevertheless, some circumstantial evidence can be obtained from looking at other employment characteristics. The number of hours worked (e.g., part-time versus full-time) is an important determinant of insurance status, because part-time workers are likely to earn lower wages and are less likely to be eligible for health insurance than full-time workers. In fact, 42.3 percent of the insured non-poor were working full-time, compared with only 35.3 percent of the uninsured non-poor and 15.3 percent of the uninsured poor. Table 4 also shows that the uninsured poor who are working are more likely than other groups to be employed in industrial sectors, such as wholesale or retail trade (sectors known for their low wages) and less likely to be in manufacturing (known for higher insurance coverage rates). Just over 30 percent of the employed uninsured poor are working in a service occupation, compared with only 11.8 percent of the working insured non-poor.

Sociodemographic Characteristics

Perhaps the most dramatic difference between the poor and the non-poor is in the number of children who are poor (Table 4). Among the poor, children account for 32.4 percent of the uninsured, in contrast to only 24.7 percent of the non-poor uninsured, and 22.9 percent of the insured non-poor. The higher proportion of uninsured poor who were children reflects a number of phenomena. First, children may not be covered by a parent's insurance policies, even if the parent is covered. Second, poor families are often larger than non-poor families. Third, parents may decide voluntarily to forgo insurance coverage for their children because of low health care costs for most children over the age of 3. Fourth, income increases with age, so older respondents are less likely to be in poverty. Efforts have been made in recent years, primarily through the Medicaid program, to provide insurance coverage for more children, and this is reflected in the higher proportion of insured children among the insured poor. However, these data do show that significant numbers of uninsured poor children remain.

The uninsured poor are more likely to live in the South than any other region of the United States. More than 50 percent of the uninsured poor live in the South, likely reflective of the other characteristics of the South, including higher overall poverty rates, lower wages, lower unionization rates, lower living costs, and less generous welfare programs than the rest of the country. The percentage of persons who are not white is higher among the uninsured poor (26.5 percent) than among the uninsured non-poor (14.3 percent) or insured non-poor (9.6 percent). However, an even higher proportion of the insured poor were not white persons (31.8 percent).

Uninsurance Entries and Exits

Most health insurance coverage in the United States is provided by employers or by the government, primarily through Medicaid and Medicare. Uninsured persons enter a spell without health insurance by losing one of these sources of coverage and exit an uninsured spell by gaining one of these sources. There are likely to be systematic differences among the types of people who are able to obtain coverage through these different sources. For example, Medicaid is a program designed to provide health insurance coverage for some persons with incomes below the poverty line. Coverage is typically available only to those with incomes significantly below the poverty line. Although point-in-time estimates suggest that only about 40 percent of the poor are covered by Medicaid (U.S. House of Representatives, 1996), it is nevertheless likely that exits from uninsurance to Medicaid coverage will be more prevalent for the poor than for others. To cite another example, persons who enter a spell simply because they lost a job that provided them with health insurance and income above the poverty line are probably likely to end this spell by obtaining private coverage.

Table 5 describes the type of insurance lost by people during the 1990 SIPP panel. A significant proportion of the uninsured (more than one-third) were without insurance throughout the SIPP panel or never gained insurance after losing it. This group is not included in this analysis because it is not known what coverage they lost or gained to start or end their spell. However, among the others, the type of insurance lost seems to depend on the person's economic status. As expected, a larger proportion (29.7 percent) of the poor than the non-poor (9.2 percent) lost Medicaid coverage, perhaps because of an increase in income or other change in eligibility standards. In contrast, a greater proportion of the non-poor lost private coverage (46.4 percent) than the uninsured poor (27.1 percent). Similar proportions lost coverage from others (usually coverage provided through a spouse's or parents' insurance policy).

Table 5

Reasons Why Uninsured Spells Began and Ended, 1991

Reason Spell Began	Total	Reason Spell Ended
		Gained Medicare Coverage	Gained Medicaid Coverage	Gained Private Coverage	Gained Coverage from Others
Persons in Poverty at Start of Spell	Percent of Uninsured
Total	100.0	1.0	34.6	26.1	38.3
Lost Medicare	0.4	0.3	0.0	0.0	0.0
Lost Medicaid	29.7	0.0	24.5	1.0	4.1
Lost Private Coverage	27.1	0.7	2.5	19.9	4.1
Lost Coverage from Others	42.9	0.0	7.6	5.1	30.1
	Percent by Reason for Spell Beginning
Lost Medicare	100.0	92.3	0.0	7.7	0.0
Lost Medicaid	100.0	0.0	82.6	3.5	13.9
Lost Private Coverage	100.0	2.4	9.2	73.4	15.0
Lost Coverage from Others	100.0	0.0	17.7	12.0	70.3
Persons Not in Poverty at Start of Spell	Percent of Uninsured
Total	100.0	1.8	10.9	48.3	39.0
Lost Medicare	0.4	0.2	0.1	0.2	0.0
Lost Medicaid	9.2	0.3	5.3	1.1	2.5
Lost Private Coverage	46.4	1.0	1.6	38.5	5.3
Lost Coverage from Others	44.0	0.5	3.9	8.5	31.2
	Percent by Reason for Spell Beginning
Lost Medicare	100.0	39.1	14.1	36.7	10.1
Lost Medicaid	100.0	2.8	57.7	12.3	27.2
Lost Private Coverage	100.0	2.1	3.5	83.0	11.4
Lost Coverage from Others	100.0	1.0	8.9	19.2	70.9

NOTES: Data are for spells with observed beginnings and endings only. Rows and columns may not add to totals shown because of rounding.

SOURCE: McBride, T.; data from U.S. Census Bureau, 1990 Survey of Income and Program Participation panel.

If there are systematic differences related to economic status in the types of insurance lost by the poor compared with the non-poor, are there differences in the types of insurance they obtain? Table 5 lists the types of insurance gained by people who lose and then gain insurance. In all cases, people are most likely to reacquire the same type of insurance that they lost. However, people in poverty are more likely than the non-poor to return to Medicaid coverage after losing it (compare 82.6 percent in the second panel of Table 5 with 57.7 percent in the bottom panel), perhaps because they become eligible for Medicaid through the “spend down” provisions under Medicaid. In fact, Medicaid is the path of exit from uninsurance for 34.6 percent of the poor, compared with only 10.9 percent of the non-poor. The non-poor are more likely to acquire private coverage (48.3 percent) or coverage from others (39 percent).

In general then, the poor are more likely than the non-poor to lose coverage because they lose Medicaid coverage and more likely to end their spell without insurance by gaining Medicaid coverage. This is of course expected, given that Medicaid is a program designed to serve the poor. However, it is worth noting that 70 percent of the uninsured spells for the poor start with a loss of private coverage or coverage from others, and roughly 65 percent of spells for the uninsured poor end with a return to one of these types of coverage. Thus, limitations on the availability of Medicaid coverage limit the assistance Medicaid provides to a majority of the poor who lose health insurance.

Discussion

Roughly two-thirds of the uninsured at a point in time live in families whose income is below 200 percent of the poverty line, and roughly one-quarter live in poverty at a point in time. Although policymakers have directed considerable public resources to closing the access gap for the uninsured poor, especially in recent years with Medicaid expansions and State-level efforts to expand insurance coverage, research has not focused extensively on the dynamics of the relationship between uninsured spell durations and poverty.

The results presented here show that the typical uninsured spell is longer for the uninsured poor (roughly 8.3 months) than for the uninsured non-poor (roughly 6 months). As a result, roughly 88 percent of the uninsured poor at a point in time will be uninsured for more than 1 year, and roughly 69 percent will be uninsured for more than 2 years. These findings are significant because many analysts have suggested that the finding of large numbers of persons uninsured and poor at a point in time is misleading, as many of the uninsured poor will not be poor or uninsured for a long time. In fact, the findings suggest that almost one-quarter of the uninsured at a point in time are poor persons who have been uninsured for more than 1 year, and 42 percent are persons with incomes below 150 percent of the poverty line and uninsured for more than a year.

The uninsured poor are of particular interest for a number of reasons. First, there are concerns about the affordability of health insurance and health care for the uninsured poor and near-poor. Insurance premiums are likely to be a large percentage of income for a poor or near-poor person even if provided by an employer, but especially if sought in the private market. This is likely to lead the uninsured to avoid purchasing health insurance coverage. Table 6 presents some evidence on this, in a self-reported question where SIPP respondents who report themselves as uninsured are asked why they are uninsured. The vast majority (73.2 percent) of the uninsured poor report that they cannot afford health insurance or it is too expensive, but only a simply majority of the uninsured non-poor report that affordability is a problem.

Table 6

Self-Reported Reason for Lacking Health Insurance, Point-in-Time Uninsured, 1991

Reason	All Uninsured	Uninsured Poor	Uninsured Non-Poor
Total	100.0	100.0	100.0
Too Expensive/Can't Afford	60.2	73.2	57.3
Employer Does Not Offer	14.5	10.4	15.4
Unemployed	5.7	5.9	5.7
Have Been Healthy	2.2	1.2	2.4
Don't Believe in Health Insurance	0.7	0.1	0.8
Can't Obtain (Pre-existing Condition)	1.3	1.2	1.4
Covered1	10.3	5.6	11.4
Other	5.0	2.5	5.6

Respondents reported no insurance coverage in response to questions on that status, yet reported they were still covered in response to this question.

NOTE: Columns may not add to totals shown because of rounding.

SOURCE: McBride, T.; data from U.S. Census Bureau, 1990 Survey of Income and Program Participation panel.

A second problem with the uninsured poor is that, without health insurance, medical care is likely to lead to significant out-of-pocket costs, which may lead them to use health care inefficiently (e.g., using emergency room care for primary care) and to account for significant percentages of uncompensated care (Holahan and Zedlewski, 1992). Estimates presented here show that close to one-half of the uninsured are poor or near-poor and might have difficulty financing large health care expenditures. Long-term poverty is likely associated with a continued inability to afford insurance; this is better captured in annual poverty rate calculations, the focus of this article. For example, the poverty line for a family of three was $11,140 in 1991. It is easy to see how a relatively small medical bill could cause serious financial hardship for a family whose income is less than this amount or less than 150 percent of this amount (roughly $18,500). High health care expenditures may not be a problem for a person who is only temporarily poor, however, and that is why this article focuses on the coincidence of poverty and uninsurance. If poverty or uninsurance is temporary, people could either delay a health care intervention until they have the money or insurance, or they could pay their health bills when they are no longer poor.

A number of policymakers have suggested solving the problem of uninsurance by offering people without health insurance tax incentives to purchase insurance. For example, people could be offered a tax credit or voucher to purchase health insurance. Other proposals would create a tax deduction for the purchase of health insurance. Because the tax incentive proposals often require some contribution from the uninsured and are often voluntary, these plans are likely to lead to less coverage of the uninsured poor than would be expected under other approaches, because these contributions would not be affordable. This would be especially true for the uninsured who are also in the midst of long poverty spells, a group described in detail here.

A third problem created by a large number of uninsured poor is that long spell durations can increase the likelihood of delays in the utilization of health care, exacerbating an already precarious financial position and leading to a perpetuation of the poverty spell and further declines in health status. Previous literature (U.S. Office of Technology Assessment, 1992) provides significant evidence on the relationship between health insurance, health status, and health utilization of the uninsured poor. The uninsured are found to have lower health utilization, tied to their lack of access to a mechanism to pay for health care, and this is likely to be exacerbated over time. There is significant support for this hypothesis, as uninsured spell durations are found to be significant predictors of fewer physician visits and fewer nights in the hospital, even after controlling for other factors. In addition, the long-term uninsured are found to be much less likely to use a physician as their usual source of care. Instead, they turn to other sources of care, such as clinics, hospital emergency rooms, or other hospital sources. However, the long-term uninsured poor are also more likely to have no usual source of care at all.

These findings could have considerable bearing on the choice of an appropriate public policy. For roughly one-third of the uninsured poor who exit from uninsurance spells, the Medicaid program provides such an exit. However, this also means that Medicaid is not closing the majority of insurance gaps; the vast majority of uninsured poor, if they do end their spell of uninsurance, do so through other mechanisms, particularly coverage from an employer or coverage from others.

Recent expansions of Medicaid, State-level efforts to expand coverage to the poor, the creation of the Health Insurance Portability Act, and the recent passage of the Child Health Initiative are likely to help alleviate the problem of the uninsured poor, although it is too early to tell how much of the gap will be closed by these initiatives. However, despite these efforts, gaps in coverage will remain difficult to close, especially for the uninsured poor, who are unlikely to be able to afford coverage under some of these alternatives or will not be eligible for many others. As shown here, the uninsured are disproportionately poor or near-poor, so it is unrealistic to assume that they can find a way to purchase insurance and become insured.

In addition, the recent passage of welfare reform and the creation of the TANF program may exacerbate the problem of the uninsured poor in the long run. The TANF program protects eligibility to Medicaid for a number of months and allows recipients to remain eligible while their incomes remain below recent Medicaid eligibility thresholds. However, TANF recipients who are actually successful in finding and keeping jobs may ironically lose Medicaid coverage if their employment pushes them above the eligibility thresholds after the phase-in period. This suggests that incremental reforms may not solve the problem of chronic uninsurance among the poor and near-poor.