From looking at differentials in health insurance coverage at a point in time we turn now to differentials in the lack of health insurance coverage over time. First we compare children who were uninsured at the end of a year with those who were ever uninsured during the year. Next we look at differentials in the duration of all new spells of uninsurance that began during FY93, followed by new spells of uninsurance coupled with Medicaid eligibility. We close by examining differentials in the duration of spells that were active at the end of FY93 and comparing these to differentials in the duration of new spells that began during the year.
1. Uninsured At a Point in Time Versus Ever in a Year
Table 12 compares demographic differentials in the number of children uninsured in September 1994 and the number ever uninsured during FY94, and Table 13 compares socioeconomic
differentials. In both tables we see that the percentages of children ever uninsured during the year are much higher than the percentages uninsured in September 1994, and this in itself is striking, but the differentials are not materially different except for race/ethnicity and region. Black non-Hispanic children are more likely than children in the "other" group of races to have been ever uninsured during the year but less likely to be uninsured in September 1994. With respect to region there is a very clear division between the Northeast and Midwest, on the one hand, and the South and West on the other in the percentage ever uninsured during FY94. In the Northeast and Midwest the proportions ever uninsured during the year vary from 12 to 17 percent while in the South and West these proportions range between 22 and 33 percent. We note as well that the very large differentials in the frequency of uninsurance by parents' education are even more impressive with respect to children ever uninsured during the year. The proportion of children ever uninsured in FY94 rises from 8 percent among children whose parents did some graduate work after college to 54 percent among children whose parents received less than 7 years of education
Consistent with the generally very limited change in differentials between children uninsured in September 1994 and children ever uninsured during FY94, shares of the two groups of uninsured show little difference. This holds true even where the differentials in the proportion ever uninsured during the year are clearly larger than the differentials in the proportion uninsured in September 1994. For example, while the difference between Hispanic children and non-Hispanic white children grows from 16 percentage points to 20 percentage points, the Hispanic share of children who were ever uninsured is actually 3 to 4 percentage points lower than the Hispanic share of children uninsured in September 1994. Because the white population is so much larger than the Hispanic population, a smaller percentage point change can nevertheless add more white children than Hispanic children to the number of uninsured. We see a similar phenomenon with respect to region. Despite larger percentage point increases in the proportion uninsured in the South and West, the share of uninsured children living in these regions is no higher for children ever uninsured in FY94 than for children uninsured in September 1994.
2. Duration of New Spells of Uninsurance
We have demonstrated that there are very pronounced demographic and socioeconomic differentials in the likelihood that a child is uninsured at a point in time or over the course of a year. Do these differences extend to the features of children's spells of uninsurance? In particular, are there important differences in the duration of spells of uninsurance by demographic or socioeconomic characteristics? We examined differentials in the distribution of completed spells of uninsurance among spells that started in FY93. Our findings with respect to demographic characteristics are presented in Tables 14 and 15. Socioeconomic differentials are reported in Tables 16 and 17.
As we explained in Appendix B, the durations of completed spells of uninsurance as measured in the SIPP incorporate a significant type of measurement error that greatly overstates the reported frequency of spells lasting exact multiples of four months. Consequently, the distributions reported in the tables that follow should not be interpreted literally. In particular, the 54 percent of all new spells that are reported as having been completed in 1 to 4 months is probably overstated, as some of the many spells identified as being completed in exactly 4 months undoubtedly lasted 5, 6, or even 7 months. Similarly, the percentage of spells lasting 13 months or more is probably understated, given the likely overreporting of spell durations of exactly 12 months. Intermediate spells are understated at the low end and overstated at the high end, with an uncertain effect on their overall frequency.9
Despite this measurement error in the reported durations of spells, differentials in the relative frequency of spells of different length may carry important information about differences in the experience of uninsurance among children across demographic and socioeconomic subgroups. We recognize, though, that measurement error in the reporting of durations may vary by these same characteristics and either weaken or distort the observed relationships.
The results reported in Table 14 suggest that, generally, demographic differentials in the reported duration of children's spells of uninsurance are weak at best. There are no meaningful differences by age. With respect to race and ethnicity we find that Hispanic children have the fewest spells of 1 to 8 months in length and the most spells of 13 months or longer, but the differences between Hispanic children and other children are modest. Turning to family composition, we find that children in mother-only and no parent families have the highest incidence of spells reported to have lasted 4 months or fewer while children in mother-only families also have the lowest incidence of spells exceeding 12 months. There are no differences by metropolitan residence but there are differences by region. Spells tend to be shortest in New England and longest in the East South Central and Pacific regions. The size of the Hispanic populations in these latter areas may contribute to the longer duration of spells, given that Hispanic children appear to have the highest frequency of long spells. Finally, the East North Central and mountain regions join New England in having relatively short durations, but beyond this the regional differences are inconsistent.
With relatively weak differentials in the duration of uninsurance spells by demographic characteristics, we would not expect the distribution of demographic subgroups of uninsured children to vary much by spell duration. Table 15 shows the distribution of demographic subgroups of uninsured children within each grouping of spells by duration. Age composition shows no meaningful variation by spell duration. Race and ethnic composition show very limited variation. Non-Hispanic white children represent a somewhat larger share of 1 to 4 month spells than they do of longer durations while Hispanic children represent a larger share of spells exceeding 12 months than they do of shorter spells. Non-Hispanic black children appear with relatively greater frequency among spells of 5 to 12 months than they do among shorter or longer spells. Children in mother-only families represent 35 percent of 1 to 4 month spells compared to only 21 percent of 13 month or longer spells. Children in two-parent families show the opposite tendency. They represent 69 percent of the longest spells but only 55 percent of the shortest spells. Metropolitan residence continues to show no consistent variation while there are small regional differences. Children from New England and the East North Central states appear with greater frequency among 1 to 4 month spells than among longer spells while children from the Pacific states account for only 18 percent of the 1 to 4 month spells but 28 percent of the 13 month and longer spells.
Like demographic characteristics, socioeconomic characteristics display a much weaker association with spell duration than with the incidence of uninsurance. In Table 16 we see that poverty has a weak, inverse relationship with spell duration. Children from the poorest families tend to experience a higher frequency of very short spells and a lower frequency of long spells than children from less poor or higher income families. We speculate that this may reflect the impact of Medicaid as a source of coverage that is more available to children in very poor families than it is to children in less poor families. The highest incidence of spells lasting 13 months or more occurs
among children between 100 and 300 percent of poverty. Long spells decline in frequency in the top income class. Spell length shows a very inconsistent relationship with parents' employment, with no category being associated unambiguously with long spells or short spells. For example, children with no parent present are most likely to have spells of 1 to 4 months and spells of 13 months or longer. To some extent this inconsistency may be due to the fact that parents' employment is itself a dynamic characteristic. Differentials in the duration of uninsurance may depend on when we measure parents' employment (before, during, or after a spell of uninsurance). In Table 16, as the title indicates, we measure characteristics at the start (that is, in the first month) of a spell, which gives us a uniform measure across spells and increases the likelihood that the characteristics we observe are associated with why the child is uninsured. If we measure socioeconomic characteristics at only one point in time, clearly this is he preferred point to do so. But this is not to say that there is nothing to be learned by examining differentials in duration by income and parents' education measured at another point in time.
The educational attainment of children's parents has a moderately strong, inverse relationship to the duration of spells of uninsurance. This is most evident in the proportion of children with durations of 13 months or longer, which declines monotonically from 33 percent to 12 percent as parents' education rises (excluding the children with no parent present). It shows up as well in the proportion of children with spell durations of 8 months or less, which rises--also monotonically--from 60 percent to 81 percent as parents' education increases.
Socioeconomic differentials are even less evident in the shares of children completing spells of different durations (Table 17). Children in families below 50 percent of poverty represent an increasingly smaller share of spells as duration rises, but there are no other consistent patterns by poverty level. This lack of consistency is true of differences in shares of spells represented by children classified by their parents' employment and parents' education. The fact that we see such weak differences in spell shares by parents' education despite the fairly strong differences in the distribution of spell length by parents' education appears to be due to the small numbers of uninsured children whose parents fall into the lowest and highest levels of education. Children whose parents completed less than 7 years of schooling account for twice as many of the longest spells as they do all shorter spells, but this represents only a five percentage point difference. Similarly, children whose parents completed some graduate school represent only half as many of the longest spells as they do spells of 8 months or less, but this is only a 4 percentage point difference.
3. Duration of New Spells of Medicaid-Eligible Uninsurance
As with all spells of uninsurance, the duration of spells of Medicaid-eligible uninsurance shows similarly weak relationships to the demographic and socioeconomic characteristics of uninsured children. Part of the explanation is the very high proportion of spells that are completed in 4 months or less: 75 percent (Table 18). This leaves little room for significant variation across population subgroups.
Only infants show a pattern that differs to any degree from that of other age groups. Infants are somewhat less likely to have spells of one 1 to 4 months than the other age groups--65 percent versus about 76 percent--but they are no more likely than other groups to have very long spells. Non-Hispanic white children have both the highest frequency of very short spells, at 81 percent, and the lowest frequency of very long spells, at just under 5 percent. Other differences by race and ethnicity are negligible. Children in two-parent or father-only families also have the highest frequency of very short spells, at 77 to 79 percent, and the lowest frequency of very long spells, at 5 percent. The corresponding percentages for children in mother-only families are 70 percent and 9 percent, respectively. Children from no parent families compare at 70 percent and 10 percent.
There is a 10 percentage point difference between metropolitan and non-metropolitan children in the proportion of spells completed in 1 to 4 months. The children to whom the metropolitan/non-metropolitan classification is not applicable differ from the other two groups in the proportion of spells that are completed in 1 to 4 versus 5 to 8 months and in the proportion running 13 months or longer, but the uncertainty about who this group represents makes it unclear what this is telling us.
Because the number of children experiencing spells of Medicaid-eligible uninsurance is much smaller than the number who experience any spells of uninsurance, sampling error makes a greater contribution to the differentials in these tables than the earlier tables, and this is nowhere more
evident than in the regional differences. The West North Central and Pacific regions have the highest percentages of very short spells, at 80 to 81 percent, and among the lowest frequencies of very short spells, at 0 to 3 percent. The South Atlantic and West South Central regions are similar. Unlike all of the earlier tables, the New England region shows nothing distinctive in its patterns. In general, the regional patterns in Table 18 deviate so much from what we have seen earlier that we should be cautious in attaching much credence to them.
Table 19 reports shares of spells within each duration group by demographic characteristics. In general, we see little variation across the columns. Non-Hispanic white children account for 56 percent of the shortest spells compared to 39 percent of the longest spells while non-Hispanic black children represent 26 percent of the longest spells compared to 17 percent of the shortest spells. Similarly, children in two-parent families account for 66 percent of the shortest durations compared to 51 percent of the longest durations while children from mother-only families account for 26 percent of the shortest durations but 40 percent of the longest. A comparison with Table 8 indicates that these findings are consistent with the relative frequency of Medicaid eligibility among uninsured children by race and ethnicity and family composition although they appear to run counter to the
differentials that we find in Medicaid participation (see Table 3). Clearly, a high participation rate in Medicaid does not imply that children leave spells of Medicaid-eligible uninsurance relatively quickly.
Metropolitan area children account for a bigger share of long spells than short spells at 76 percent versus 68 percent while non-metropolitan area children account for 31 percent of the shortest spells and 22 percent of the longest. Regional differences again show the effects of particularly large sampling error. Every region displays fairly wide variation in its shares of spells of different durations, but these differences do not yield a clear pattern.
Differences in the duration of spells of Medicaid-eligible uninsurance by socioeconomic characteristics are much weaker than the differences we reported in the duration of all spells of uninsurance (compare Table 20 with Table 16). There are negligible differences across the first three poverty classes, where sample sizes are largest, and the two highest poverty classes are not substantially different, given their very small numbers.10 Children with working parents, whether full time or part time, tend to have a greater frequency of short spells than children with no working parents or no parents present and a somewhat lower frequency of very long spells. Differences in spell duration by parents' education are very inconsistent, suggesting that what differences we do see are strongly affected by sampling error. Children whose parents attended graduate school clearly have the shortest spells, with 82 percent being 1 to 4 months in length and only 1 percent exceeding 8 months. Children whose parents completed less than 7 years of schooling have the longest spells, with 60 percent being 1 to 4 months and 23 percent 9 months or longer. But children whose parents completed college (and went no farther) look more like children whose parents completed only 7 to 11 years of schooling than they resemble children whose parents attended but did not complete college or completed college and went on to graduate school.
There are somewhat greater differences in the shares of spells of different durations by socioeconomic characteristics (Table 21) than by demographic characteristics (Table 19). Children in families below 50 percent of poverty account for smaller shares of spells 9 months or longer than they do of spells completed in 8 months or less, but patterns at higher family income levels are less clear. Children in families above 100 percent of poverty represent somewhat larger shares of the spells lasting 9 months or longer than they do of spells ending in 8 months or less. Children without working parents account for larger shares of spells 9 months or longer than they do of shorter spells while children with at least one parent working full time show the reverse pattern, but even here sampling error is evident in the volatility of the shares. Finally, children whose parents completed 7 to 11 years of schooling account for 33 percent of the longest durations compared to no more than 22 percent of any class of shorter durations whereas children whose parents attended but did not complete college represent 14 percent of the longest durations compared to 20 percent of the shortest.
4. Duration of Active Spells of Uninsurance
Earlier in this section we examined demographic and socioeconomic differentials in the lengths of completed spells of uninsurance. Spells that are active at a point in time have a different distribution--both currently and, especially, when completed--than do spells that started during a year. Furthermore, because active spells are generally not complete (some will end in that month, but most will continue), the relationships beween their duration and various characteristics of the uninsured differ from what we would observe with completed spells. More specifically, the shorter the active duration, the greater the proportion of these spells that will end up with longer completed durations. In terms of their characteristics, children who are in short active spells will look much more like children with completed spells of all durations than they do like children with completed spells of short duration. We would be much less interested in the characteristics of uninsured children by how long they have been uninsured were it not for the fact that state CHIP plans frequently limit coverage to children who have already accumulated some specified number of months of uninsurance. Table 22 compares the distributions of active spell lengths and completed spells by three characteristics: the child's age, family poverty level, and parents' employment.
Active spells are limited to those of children who were uninsured in September 1993 while the completed spells represent all spells that were started in FY93.
The active spells include a higher proportion of very long spells than do the completed spells; 46 percent of the children uninsured in September 1993 had already been uninsured for 13 months or more whereas only 20 percent of all the spells that started in FY93 (which may include multiple spells by the same individuals) extended beyond 12 months.
Infants have much shorter active spells than completed spells because spell lengths are bounded by their very limited life spans. We see no infants with spells exceeding 12 months, of course, because no infant, by definition, has been alive for more than 12 months. By contrast, there is no limit to the completed spell length of children who were infants when their spells began, and infants have nearly the same proportion of completed new spells exceeding 12 months as older children: 17 percent compared to 18 to 22 pecent for older children). The proportion of older children whose active spells exceed 12 months rises 20 percentage points--from 36 percent to 56 percent--between ages 1 to 5 and ages 11 to 15 but then drops slightly, to 49 percent, in the next age group. The proportion of new spells exceeding 12 months in length shows no increase with age after infancy, however. It is not clear why we should see an age difference in the current duration of active spells when there is no age difference in the completed durations of new spells. Active spells are in some sense sampled from new spells but with a probability proportional to (completed) length. It must be that the age differences that we see among active spells are due to age differences that do not appear in new spells until durations well beyond 13 months.
There are differences in the durations of both completed spells and active spells by poverty level, but the differences among active spells are weaker. The proportion of children with completed spells of 1 to 4 months declines from 63 percent to 51 percent over the first three income classes while the change among active spells is only 6 percentage points. Similarly, the proportion of new spells exceeding 12 months in length rises from 14 percent to 23 percent over the first three income classes whereas the increase among active spells is only 5 percentage points. Both distributions show an upturn in the relative frequency of short spells as income rises to more than 300 percent of poverty.
Parents' employment shows little relationship to the duration of completed spells. Children with no parent present have a somewhat higher frequency of short spells than other children, but there are no other consistent differences. Among children with active spells, those with no parents present have fewer spells exceeding 12 months than other children but comparable numbers of spells exceeding 8 months. Children with no working parents are somewhat more likely to have been uninsured for less than 5 months than children with working parents, but again the differences are modest.
Differences in the shares of active spells by demographic and socioeconomic groups are of interest because of what they tell us about the composition of children who may be eligible for state CHIP coverage. As we have pointed out, infants by definition cannot be uninsured for more than 12 months. States that would like to cover infants but choose to limit their coverage to children who have been uninsured for 12 months or more or even 6 months or more will have to consider defining eligibility among infants on some basis other than the length of their own spells of uninsurance, or infants will receive disproportionately little coverage. In Table 23 we see that while infants account for 7 percent of the spells of uninsurance that started 1 to 4 months earlier, they account for only 3 percent of the spells that started 5 to 8 months earlier and closer to 2 percent of the spells that started 9 to 12 months earlier. Shares of active spells that started more than 12 month earlier rise with age.
Children 1 to 5 account for 18 percent of such spells while children 11 to 15 account for 35 percent. There is little difference, however, in the age group shares of spells lasting 9 months or longer.
Both active spells and completed spells show the share of children in families between 100 and 200 percent of poverty rising with duration. This group accounts for 36 percent of the active spells that started 1 to 4 months earlier compared to 44 percent of the spells that started more than 12 months earlier. Completed spells show the share of children in this group rising from 34 percent to 41 percent. Other poverty classes show less variation in their shares of children with different durations of uninsurance.
Shares of the uninsured by parents' employment show minimal variation by parents' employment. For active spells, no employment group's share varies by more than 4 percentage points. There is somewhat greater variation among completed spells, but no group shows consistent growth or decline in its share of all spells as duration increases.
On the whole, then, spells active at a point in time show no greater evidence of demographic or socioeconomic differentials in duration than new spells that started over the course of a year.