The Effects of Trigger Events on Changes in Children's Health Insurance Coverage

E. PRIOR EVENTS

Our principal findings with respect to the importance of individual trigger events in predicting changes in coverage are obtained from a logistic regression analysis that we present in the next section. The regression results are particularly useful in describing the relative strengths of individual events as predictors of changes in coverage, but they do not convey an intuitive sense of the overall importance of trigger events. To provide a context for viewing the regression results, we have prepared estimates of the frequency with which individual events and sets of events were found to have occurred in the months preceding recorded changes in health insurance coverage. After listing the events that we examined, we present these findings here. The results also give us additional information with which to assess the plausibility of some of the more surprising aspects of the transitions documented earlier.

1. Types of Events

We defined the following as potential trigger events for the transitions that we examined:

• Father's loss of employment
• Mother's loss of employment
• Fathers reduction in hours worked to less than 30 ⁽¹⁴⁾
• Mother's reduction in hours worked to less than 30
• Father's change in jobs (excluding re-employment or loss of employment)
• Mother's change in jobs
• Father's gaining employment
• Mother's gaining employment
• Father's increase in hours worked to 30 or more
• Mother's increase in hours worked to 30 or more
• Marked decline in family income (by one-third and at least $500)
• Marked rise in family income (by one-half and at least $500) ⁽¹⁵⁾
• Change in family headship (loss or addition of a parent)
• Reduction in family size
• Increase in family size
• Family's loss of AFDC

Family's enrollment in AFDC

All of these variables are measured as dichotomies, consistent with their use as events. The set covers a range of major family economic and demographic events that could be associated with changes in health insurance coverage.

2. Frequency of Prior Events

For each of the four sets of transitions--consisting of exits from (1) ESI, (2) uninsurance, (3) Medicaid, and (4) other insurance--we examined the frequency of alternative events occurring in conjunction with transitions into each of the three alternative destination statuses. We also examined the frequency of events occurring over the same time period among children who made no transition --that is, children who remained (1) covered by ESI, (2) uninsured, (3) covered by Medicaid, or (4) covered by other insurance. Comparing the frequency of individual events among children who made one of three transitions and children who made no transition provides a way of gauging the strength of the association between individual events and the transitions they preceded. An event that occurred with equal frequency among children who made a transition and those who did not is unlikely to have had a role in triggering the transition--no matter how often the event occurred among children who made the transition. On the other hand, an event that occurred ten times as often among children who made a transition as it did among children who did not make the transition has at least a strong association with the transition and may have had a role as a trigger as well.

For each of the four categories of coverage, the first column of Table 7 displays the average monthly number of children who were in that category prior to the transition month, the number who made each of three possible transitions, and the number who remained in that category through the transition month (that is, did not change coverage). ⁽¹⁶⁾ In the tables that follow, we show how often each of these outcomes was preceded or accompanied by a potential trigger event. The second column reports the one-month transition rates from each of the four initial categories of coverage to each of the four possible outcomes: transitions to any of three alternative categories versus no change in coverage. These transition rates were calculated by dividing the average number of transitions to each outcome by the average number of children who shared the same initial coverage in the prior month, and then multiplying the result by 100 percent. Children with ESI had the lowest exit rate, with only 1.4 percent moving to another coverage in the next month. By contrast, uninsured children and children with other insurance had exit rates in excess of 7 percent while children with Medicaid had an intermediate exit rate of 3.7 percent.

We looked at events over two time periods: one month and six months preceding the transition. Our findings are summarized in Table 8, which reports the percentage of children who experienced one or more events out of a set of possible events among children who experienced a particular transition. In each case the frequency of events among children who experienced a transition is contrasted with the percentage who encountered any of the same events among children who did not change coverage. For example, in the first row we find that 46.7 percent of the children who moved from ESI to uninsured experienced a possible trigger event in the prior month compared to only 8.8 percent of those who remained covered by ESI. If we look back six months instead of just one, the fraction of children who experienced a possible trigger event rises to 71.0 percent among children who moved from ESI to uninsured and 34.7 percent among children who remained covered by ESI. Generally, the set of events that we defined as relevant varied with the type of transition--even among children who shared the same coverage in the prior month. ⁽¹⁷⁾ Therefore, even though the comparison group of children without transitions is the same for all children who exited the same coverage, the frequency of prior events among children with no transition need not be the same in each case, and indeed it is not.

The comparative frequencies of events in the prior month show a clear association between the occurrence of possible trigger events and the occurrence of a change in coverage. In the prior month, possible trigger events occurred three to five times more often among children who experienced a transition than among children who did not. Depending on the type of transition, between 29 to 48 percent of transitions had at least one possible trigger event in the prior month, with most of the rates being above 40 percent. Possible trigger events occurred least often among children who moved between ESI and Medicaid (in either direction) and most often among children who moved between uninsured and other insurance (also in either direction).

Lengthening the reference period by five months increased the number of events, of course, but equally if not somewhat more so among children with no transitions than among children with transitions. ⁽¹⁸⁾ The net result is that the differentials between children with and without transitions are generally weaker when events as far back as six months as opposed to just one month are included. Why do the differentials become less pronounced over time? One possibility, of course, is that trigger events generate transitions relatively quickly rather than over a period of several months. A parent's losing employment may have an immediate impact on the ESI coverage of parent and child. Another factor contributing to this phenomenon is the seam bias in the reporting of transitions in the SIPP. With most transitions in coverage and many trigger events being "moved" to the nearest seam between interview reference periods, events that occurred as far apart as three or four months may be reported in the same month. In the tables discussed below, we report events occurring in the prior month. Six-month tables are included in the Appendix.

The incidence of prior events also provides additional information with which to assess the plausibility of different types of transitions. Earlier we presented evidence that changes in respondent--a potential source of reporting error--were no more common when transitions were reported to have been reversed within four months than when they were not reversed. Changes in respondent are only one source of response error. Here we ask whether transitions that were reversed within four months were less likely to have been preceded by possible trigger events than transitions that were not reversed. Table 9 summarizes our findings. Comparing the final two columns, we see that for eight of the ten transition types, potential trigger events were less common among transitions that were reversed than among transitions that were not reversed, but in every case these events were still much more common than they were among children with no transitions. Moreover, the transitions that seemed most likely to reflect respondent confusion--those between ESI and other insurance (in both directions)--are the most strongly supported by Table 9. We interpret these findings as suggesting that, at worst, transitions that were reversed were somewhat more likely to have been misreported than transitions that were not reversed. ⁽¹⁹⁾ Alternatively, it may be that trigger events of the kind we have examined here simply play a less important role in explaining transitions that are quickly reversed than they do in accounting for other transitions. ⁽²⁰⁾

3. Results by Type of Event

Having reviewed our findings with regard to the overall frequency of possible trigger events, we now examine the frequency of individual types of events for each type of transition and for children whose coverage did not change.

Children Losing ESI. Table 10 presents estimates of the frequency of alternative events occurring in the past month among children who lost ESI--and became uninsured, obtained Medicaid, or obtained other insurance--and children who retained ESI. The cell entries indicate the percentage of children who experienced each event, where the base of the percentage is the total children in that column. ⁽²¹⁾ For example, in the first row we see that the proportion of children whose fathers lost employment in the past month was 9.2 percent among children who lost ESI and became uninsured, 2.5 percent among children who lost ESI and obtained Medicaid, 3.2 percent among children who lost ESI and obtained other insurance, and only .4 percent among children who remained covered by ESI. At the bottom of the table we have repeated from the first column of Table 8 the percentage of children who experienced any one of a set of events that we defined as relevant to each transition. These events are denoted by asterisks beside the individual event frequencies in the column corresponding to each type of transition. By repeating the summary frequencies here we underscore how the results presented in Table 10 are related to those presented in Table 8. In Table 8 we also reported how often any event from each of the sets identified in the first three columns occurred among children who remained covered by ESI, but those three frequencies are not repeated here.

Nearly all of the event variables occurred with substantially greater frequency among children who made one of the three transitions than among children who remained covered by ESI. Generally, the employment-related events occurred with the greatest frequency among children who moved from ESI to uninsured. This was true of job changes and reductions in the number of hours worked for either parent and a loss of employment for the father. The mother's loss of employment was as common among children who moved from ESI to Medicaid as it was among children who moved from ESI to uninsured.

Table 10 includes not only events that involved a reduction in employment but events that represented either a gain in employment (from unemployed or out of the labor force) or an increase in the hours worked to 30 or more. While it runs counter to expectation that increased employment should be associated with the loss of a child's ESI, employment gains nevertheless did occur more often among children who lost ESI than among children who remained covered by ESI--but not as often as employment losses. In the regression analysis reported later, however, where we looked at the impact of trigger events on changes in children's insurance coverage, we found no significant impact of the employment gains on transitions out of ESI.

The occurrence of marked declines in income among children losing ESI is consistent with the parents' changes in employment, but compared to children who remained covered by ESI the relative frequency of these events is lower than that of losses in employment or reductions in hours worked. Only 4 percent of the children who remained covered by ESI had a marked decline in family income over the preceding month while another 4 percent had a marked rise in income. Between 14 and 23 percent of the children who lost ESI showed marked declines in income while 10 to 17 percent showed marked increases. Again, the occurrence of a marked rise in income appears inconsistent with a loss of ESI, but in the regression analysis we will show that declines in income did not have a significant effect on the likelihood of a child's losing ESI.

Changes in the headship of the family (among one parent, two parents, or no parents) occurred much more often among children who lost ESI and became uninsured or enrolled in Medicaid than they did among children who remained covered by ESI. ⁽²²⁾ Reductions in family size also occurred more often among children who became uninsured, enrolled in Medicaid, or obtained other insurance than among children who retained ESI.

Finally, near the bottom of Table 10 we report how often children's families obtained AFDC. To properly interpret these estimates, it is important to understand how the possible occurrence of a change in AFDC coverage is constrained by our limiting the observations to children with particular transitions in health insurance coverage and by the way we simplified the measurement of change over time, described earlier. Theoretically, all children covered by AFDC are covered by Medicaid as well. In the SIPP (and in the CPS), this relationship is forced by the Census Bureau's editing practices, which assign Medicaid coverage to all children who are reported to be receiving AFDC. A child who loses AFDC will not necessarily lose Medicaid, which does not require AFDC, but a child who obtains AFDC and was not already covered by Medicaid will always gain Medicaid coverage along with the AFDC. To be identified in Table 10 as obtaining AFDC, a child had to have had AFDC in month m. The only group for which this can be true is the group that left ESI for Medicaid. ⁽²³⁾ We found that 13 percent of the children who made the transition from ESI to Medicaid obtained AFDC at the same time, and for these children it is correct to infer that their enrollment in AFDC explains their transition in insurance coverage. The remaining 87 percent of children who enrolled in Medicaid acquired Medicaid without AFDC. The loss of AFDC, on the other hand, could not have occurred in any of the four groups, since coverage in the prior month was always ESI.

Children Becoming Insured. For children who become insured after a period without insurance, we would expect that the role of parents' employment changes in helping to trigger such transitions would depend on the type of insurance that children acquired. Gains in employment or hours worked may provide access to ESI that did not exist previously, so we would expect to see evidence of recent gains in employment among children who obtained ESI. For children who obtained Medicaid, however, it seems unlikely that we would see much incidence of parents becoming re-employed or increasing their hours. Rather, transitions from uninsured to enrolled in Medicaid would be more likely to be preceded by a loss of employment than a gain--or a reduction in hours rather than an increase. The scenario we imagine is that of a parent whose employer does not offer affordable coverage but pays the parent well enough to make the child ineligible for Medicaid. The parent's subsequent loss of employment or reduction in hours may qualify the child for Medicaid.

The findings presented in Table 11 are generally consistent with these expectations. Among children who made the transition from uninsured to ESI, parents' gains in employment clearly dominated losses. For example, 7 percent of fathers gained employment, and 11 percent increased their hours to 30 or more while only 1 percent lost employment and 1 percent reduced their hours below 30. For children who became covered by Medicaid rather than ESI, gains in employment or hours worked--by either parent--appear to have occurred with somewhat less frequency than among children who obtained ESI. But employment losses or reductions in hours--particularly for the mother--were actually more common than they were among children who obtained ESI, and this is consistent with our speculation that among children who are without insurance, parents' employment losses may help to qualify their children for Medicaid.

Transitions from uninsured to other insurance present something of a puzzle. In this group employment losses by both parents occurred about as often as they did among children who enrolled in Medicaid and much more often than among those who remained uninsured. In these respects, the transitions from uninsured to other insurance resemble what we found for the Medicaid transitions and not what we would expect to find for transitions into privately purchased insurance.

Changes in family income differentiate among the transition groups more clearly than changes in parents' employment. Marked increases and decreases in income occurred with about the same frequency among children who remained uninsured--each about 6 percent. But among children who became insured we find distinctly different patterns. Children who obtained ESI were much more likely to show a rise in income (22 percent) than a decline (6 percent) while reductions occurred with the same frequency as increases (about 14 percent) among children who obtained Medicaid. Children who obtained other insurance, however, were three times as likely (30 percent) to have had a marked decline in income as a significant rise (10 percent). Since other insurance presumably costs the purchaser a substantial amount, it is counter-intuitive that a marked reduction in income should trigger exits from uninsured to other insurance.

Changes in family composition occurred more often among uninsured children who gained coverage through ESI or Medicaid than among children who obtained other insurance or remained uninsured. Changes in family headship and increases in family size occurred disproportionately among children who obtained ESI or Medicaid. Reductions in family size occurred with the same frequency as these other events among children who enrolled in Medicaid but not those who enrolled in ESI. Finally, the frequency with which uninsured families obtained AFDC indicates that this path accounted for about 17 percent of the children who enrolled in Medicaid.

Children Leaving Medicaid. The wish to understand why children leave Medicaid--frequently without other coverage--was one of the objectives motivating this research, but the findings presented in Table 12 do little to advance our understanding. About 17 percent of those who became uninsured and 12 percent of those who obtained ESI had left AFDC in the past month, compared to only 1 percent of those who remained covered by Medicaid. We might have expected gains in parents' employment to emerge prominently among children who left Medicaid for ESI, but while such gains were certainly more common among children who made transitions out of Medicaid than among children who remained in Medicaid, employment gains, increased hours, and job changes were no more common among the parents of children who moved from Medicaid to ESI than among those who moved from Medicaid to uninsured. Reflecting the composition of Medicaid families, gains among mothers were more important than gains among fathers, with 6 to 9 percent of the children who left Medicaid for any destination having a mother who gained employment or increased her hours. Job changes by either parent occurred in 3 to 5 percent of the cases. Among children who left Medicaid for ESI, fathers were as likely to have lost employment or reduced hours as to have gained employment or increased hours.

Changes in family income reflected the mixed results of families losing AFDC and families gaining employment. Both losses and gains in income occurred more often among children who left Medicaid than among children who remained, with income gains outpacing losses by 18 to 10 percent among children who obtained ESI while gains and losses occurred with equal frequency--about 13 percent each--among those who became uninsured. Changes in family composition were little more likely among children leaving Medicaid than among those who remained.

Children Leaving Other Insurance. On the assumption that other insurance is generally privately purchased insurance, we would expect that children who leave such coverage tend to do so when their parents gain employment or change jobs, giving them access to ESI, or when their family income falls, making it difficult to sustain the costs of private insurance. The findings presented in Table 13 generally support these expectations, but we see less differentiation than we would have anticipated between children who left other insurance for ESI and those who simply became uninsured. The fathers of children who became uninsured were much more likely to lose employment or reduce their hours of work than the fathers of those who remained covered by other insurance or obtained ESI, but the fathers and especially the mothers of children who became uninsured were also more likely to gain employment than the parents of the other two groups of children. More consistent with our expectations, children who moved from other insurance to ESI were the most likely to have parents who changed jobs or increased their hours of work to 30 or more.

Both increases and reductions in family income were more common among children who left other insurance than those who remained. Children who became uninsured were somewhat more likely to have had marked reductions in income than those who obtained ESI (22 percent versus 15 percent) or retained other insurance (6 percent), but gains were about equally common among children who became uninsured (25 percent) or obtained ESI (22 percent) and much higher than for those who remained covered by other insurance (7 percent). It is difficult to interpret the complete loss of coverage among children who left other insurance despite rising family income. Table 3 showed that 10 percent of the children who made the transition from other insurance to uninsured were covered by ESI shortly thereafter; for these children the loss of coverage was merely transitional and may reflect waiting periods for ESI to become effective. For the others it is simply not clear what may have happened, but understanding such transitions is important to understanding and addressing the problem of uninsurance among children in the United States.

Increases in family size and changes in family headship were marginally more common among children who became uninsured than among those who obtained ESI or retained other insurance. Oddly, when we look back six months (Table A.5) we find that 20 percent of the children who became uninsured had experienced a reduction in family size compared to only 3 percent of those who remained covered by other insurance and less than 1 percent among those who obtained ESI. There is no hint of this in Table 13, which adds to the general ambiguity surrounding the impact of changes in family composition on transitions in children's health insurance coverage.

Footnotes:

14. We measured the change in hours worked in terms of movement across a threshold because employers generally base their offer of insurance coverage--or whether they contribute to premiums--on a specific number of hours worked. Whether the parent moved across this threshold is likely to be more important than the change in the number of hours worked in determining the implications for access to coverage. The number 30 is an industry standard used widely in the plans drafted by insurers. Individual employers may set their own thresholds, of course, but SIPP does not collect that information. Using the value 30 seemed appropriate under these circumstances, and the findings that we will present shortly seem to bear this out.

15. Despite the differing proportions the upward and downward changes in income can be viewed as equal but opposite in the following sense. If we start with X dollars, a 50 percent increase will raise the amount to 1.5X while a one-third reduction in 1.5X dollars will reduce the amount back down to X. In our sample these two types of change occurred with roughly equal frequency.

16. Recall that transitions were measured over the 12-month period from July 1993 through June 1994. The "prior" month, therefore, refers to the 12 months from June 1993 through May 1994.

17. We defined an event as relevant to a particular transition if there was a theoretically based reason for expecting the event to serve as a trigger or if there was empirical evidence of an association between the event and the transition that we could not dismiss as spurious. We erred on the side of being overly inclusive on the grounds that events with no net association would either add nothing to the count of children with one or more events or would add equally to the counts of children with and without transitions.

18. Using the 1984 SIPP panel, Short et al. (1988) estimated the frequency of selected events in the eight months preceding the interview in which Medicaid enrollment or disenrollment was first reported. They found that 83 percent of enrollments and 68 percent of disenrollments were associated with one or more events, but they did not compare the incidence of the same events among persons who made no transitions.

19. We must be careful not to ascribe a validating role to trigger events. That is, we must not take the view that a reported transition that was not preceded by a plausible trigger event did not occur. After all, past research has not assigned much importance at all to trigger events in explaining transitions (see the discussion below in Section F).

20. We also broke out the incidence of possible trigger events among children who moved from ESI to Medicaid by the parents' change in coverage and whether the child's transition was reversed in the next four months. Recall from the previous section that 71 percent of children's transitions from ESI to Medicaid occurred without either parent losing ESI. When the father lost ESI along with the child, a possible trigger event was observed in 79 percent of the cases, and it made no difference whether or not the transition was reversed in the next four months. When the father retained ESI (which happened with more than half the transitions), possible trigger events were observed less than 20 percent of the time, but this was still much larger than the 7 percent reported in Table 8 for children who remained covered by ESI. Moreover, possible trigger events were actually more common rather than less common when children returned to ESI within four months. When the father was either absent or had no ESI coverage before the child's transition, but the mother did have ESI, the incidence of possible trigger events depended on both the mother's loss or retention of ESI and whether or not the child's transition was reversed. When the mother lost ESI, possible trigger events occurred in 66 percent of the cases when the child's transition was not reversed but only 16 percent of the cases when the child's transition was reversed. When the mother retained ESI, possible trigger events were observed in only 16 percent of the cases when the child's transition was not reversed and in none of the cases when the child's transition was reversed. Perhaps these findings suggest that a significant number of the reported transitions from ESI to Medicaid may not have occurred at all, but that is not the only interpretation. Our selection of possible trigger events focused on employment-related changes and relatively rare demographic changes, so it should not surprise us that when children moved from ESI to Medicaid without their parents losing ESI, the incidence of possible trigger events was rather low. It may be that other types of trigger events dominate such cases or that trigger events play a less prominent role than other types of influences. Because of the policy interest in transitions between ESI and Medicaid, however, further research into the circumstances surrounding such transitions is clearly warranted.

21. The average monthly numbers of children to which the percentages reported in Table 10 apply were reported in Table 7.

22. Later, when we examine the effectiveness of these events in predicting transitions in coverage, we separate changes in family headship into gains and losses of parents, as we might expect these to have quite opposite effects. Here, where we condition on the transitions and look for prior events, we can ask simply whether any changes at all in family headship occurred.

23. Theoretically, a child could have had more than one source of insurance in a given month--the most likely scenario being one in which the child made the transition from one source to the other during the course of the month. Such transitions are reported only rarely in the SIPP, however, and to simplify our analysis of transitions in health insurance coverage we elected to assign only one source of coverage in any given month. Because Medicaid coverage is underreported in household surveys, we assigned a child to Medicaid if a child was reported to have had Medicaid coverage in that month--regardless of whether any other coverage was reported. We also favored ESI over other insurance coverage.

Where to?

Top of Page
Table of Contents

Home Pages:
Office of Health Policy (HP)
Assistant Secretary for Planning and Evaluation (ASPE)
U.S. Department of Health and Human Services (HHS)

Last updated September 5, 2000