America’s Uninsured Children A Failure of State Policy Part 3

Table IV - Part A Face-to-Face Interview and Uninsurance Face to face interview I 0.0776 (1.47) II 0.0406 (0.74) III 0.0793 (1.37) IV 0.0255 (0.42) Hispanic 0.0854 (11.81)** -0.0197 (0.70) 0.0847 (11.70) -0.0170 (0.56) Black -0.0053 (0.69) -0.0052 (0.69) 0.0470 (1.34) 0.0033 (0.09) Age1to5 -0.0145 (0.99) -0.0459 (2.62)** -0.0001 (0.01) -0.0354 (1.65)* Age6to9 0.0161 (1.05) -0.0303 (1.69)* 0.0264 (1.53) -0.0331 (1.53) Age10~15 0.0357 (2.39)** -0.0116 (0.66) 0.0516 (3.06)** -0.0054 (0.25) Age16~18 0.1347 (7.59)** 0.0731 (3.57)** 0.1539 (7.65)** 0.0781 (3.14)** FFint* Age1to5 -0.0124 (0.30) 0.0208 (0.43) -0.0286 (0.62) 0.0027 (0.05) FFint* Age6to9 -0.0223 (0.55) 0.0316 (0.63) -0.0183 (0.39) 0.0582 (0.95) FFint* Age10to15 -0.0457 (1.22) 0.0096 (0.21) -0.0438 (1.01) 0.0316 (0.55) FFint* Age16to18 -0.0578 (1.52) -0.0138 (0.29) -0.0502 (1.11) 0.0113 (0.19) FFint* Hispanic 0.0923 (0.86) 0.1234 (1.07) Hispanic* Age1to5 0.0968 (2.82)** 0.0841 (2.28)** Hispanic* Age6to9 0.1437 (3.91)** 0.1472 (3.68)** Hispanic* Age10to15 0.1425 (4.03)** 0.1347 (3.53)** Hispanic* Age16to18 0.1636 (4.22)** 0.1580 (3.79)** FFint*Black 0.0401 (0.43) 0.0917 (0.90) Black*Age1to5 -0.0587 (1.80)* -0.0318 (0.85) Black*Age6to9 -0.0399 (1.16) 0.0094 (0.23) Black*Age10to15 -0.0601 (1.88)* -0.0181 (0.48) Black*Age16to18 Hispanic* FFint* Age1to5 -0.0582 (1.74)* -0.0694 (0.81) -0.0127 (0.32) -0.0615 (0.67) Hispanic* FFint* Age6to9 -0.1115 (1.51) -0.1214 (1.68)* Hispanic* FFint* Age10to15 -0.1273 (1.92)* -0.1341 (2.04)** Hispanic* FFint* Age16to18 -0.1070 (1.39) -0.1171 (1.56) Black*FFint*Age1to5 0.0685 (0.65) 0.0342 (0.33) Black*FFint*Age6to9 -0.0102 (0.11) -0.0631 (0.74) Black*FFint*Age10to15 0.0016 (0.02) -0.0554 (0.66) Black*FFint*Age16to18 -0.0178 (0.19) -0.0644 (0.75) For each regression, n = 28,726. (*) denotes significance at 90% level, (**) denotes significance at 95% level. Z-statistics in parentheses. Each column above denotes a different regression. In each regression, the dependent variable is a dummy tak ing on a value of 1 if the individual is uninsured, 0 otherwise. Dprobit regressions were used to generate estimates of coefficients on above explanatory variables. Column I includes double interaction terms involving "face to face interview" dummy and age dummies. Column II includes triple interaction terms involving "face to face interview" dummy, age dummies, and Hispanic dummy. Double interaction terms involving age dummies and "face to face interview" dummy, age dummies and Hispanic dummy, and Hispanic dummy and "face to face interview" dummy are included as well. Column III is structured in same way, but with Hispanic dummy replaced by Black dummy. Column IV includes both “Hispanic” and “Black” dummies, and all the interaction terms from Columns II and III. Controls were included for subject's family income relative to FPL, and state fixed effects in each regression. Controls for Asian or Native American background were included in each regression, but are not reported above. In Column I, the coefficients on the interaction terms involving the “Face-to-face interview” variable and the age dummies are all negative. These coefficients consistently grow 32 increasingly negative with older age groups. However, none are statistically significant at the 90% or 95% level. Additionally, the coefficient on the non-interacted “face-to-face interview” variable is positive, and larger in magnitude than any of the negative coefficients on the double interaction terms. My overall objective from this regression was to compare the probability of uninsurance for a child living in a state with a face-to-face interview requirement with that of a child of the same age bracket living in a state without this requirement. To do this, it is necessary to add the coefficient from the non-interacted “face-to-face interview” variable to the coefficient on the relevant double interaction term involving the age dummy and the “face-to-face interview” dummy. For example, to estimate the impact of a face-to-face interview requirement on children aged 16-18, I add the coefficient on the double interaction term involving the “Age 16-18” and “face-to-face interview” dummies to the coefficient on the non-interacted “face-to-face interview” variable. This computation indicates that the presence of a face-to-face interview requirement increases the probability of uninsurance for children aged 16 to 18 by 1.98 percentage points. Carrying out this exercise for all age brackets, I find that the face-to-face interview requirement increases the probability of uninsurance for children throughout the age spectrum. Interestingly, this effect appears to be larger for younger children. Indeed, for children aged 1-5, this requirement increases the probability of uninsurance by the largest amount: 6.31 percentage points. I use the regressions in Columns II, III, and IV to estimate the specific effect of the faceto-face interview requirement on Hispanic and Black individuals of each age bracket. To illustrate the procedure for calculating this effect, I outline the computation of the face-to-face interview’s impact on Hispanic children aged 1-5. I compare two samples: the sample of Hispanic children aged 1-5 residing in states with a face to face requirement, and the sample of Hispanic children aged 1-5 residing in states without face-to-face interviews. The difference in the probability of uninsurance among children of the two samples provides an estimate of the face-to-face 33 interview’s overall effect on Hispanic children in this age bracket. In order to compare the two previously mentioned samples, it is necessary to sum the coefficients on four terms in the Column II regression: the triple interaction term involving the “Age 1-5 dummy,” the “Hispanic” dummy, and the “face-to-face interview” dummy; the double interaction term involving the “Hispanic” dummy and the “face-to-face interview” dummy; the double interaction term involving the “faceto-face interview” dummy and the “Age 1-5” dummy, and the non-interacted “face-to-face interview” dummy. Adding these four coefficients from Column II, I estimate that the face-toface interview requirement increases the probability of uninsurance for Hispanic children aged 15 by 8.43 percentage points. In order to calculate the standard error on this estimate, I first square the standard errors on each of the four coefficients used in the calculation, and take the sum of the squared standard errors. The square root of this sum represents the standard error of the above estimate. Using this procedure, along with the coefficients in Column II, I estimate the effect of the face-to-face interview requirement on each of the age brackets within the Hispanic population. I then use the coefficients in column III to estimate the effect of this policy on each of the age brackets within the Black population. Finally, I use the coefficients in column IV to generate a second set of estimates of the effect of the face-to-face interview on Hispanic and Black individuals of different age brackets. Table IV (part B) summarizes these estimates. In this table, standard errors (rather than Z-statistics) are in parentheses. On the surface, it appears that there is nothing to be gleamed from the table on the following page. All of the standard errors are high relative to the estimated coefficients, none of which are close to being statistically significant. I believe that these high standard errors are partly the result of the small sample sizes that result from analyzing small subsections of the overall dataset. For example, the sample contains only 256 Hispanic children aged 1-5 living in states with a face-to-face interview requirement. However, if I momentarily leave aside the lack of statistically significant results and focus on the coefficients themselves, there are indeed interesting trends to examine. The positive coefficients in each column indicate that the use of a 34 face-to-face interview requirement appears to increase the probability of uninsurance for Hispanic and Black children individuals of all age groups. However, it is also clear that for both Black and Hispanic individuals, the coefficients grow less and less positive for older children. For example, the results of column II indicate that the use of a face-to-face interview increases the probability of uninsurance by 8.4 percentage points for Hispanic children aged 1-5, but increases this probability by only 1.2 percentage points for children aged 16-18. A similar trend of decreasing coefficients corresponding to older age groups among Hispanic children is observed in Column IV as well. Table IV - Part B Face to Face Interview: Effect on Specific Demographic Groups II III IV Hispanic Age 1-5 0.0843 (0.1592) 0.0901 (0.1744) Hispanic Age 6-9 0.053 (0.1510) 0.0857 (0.1656) Hispanic Age 10-15 0.0152 (0.1462) 0.0464 (0.1611) Hispanic Age 16-18 0.0121 (0.1500) 0.0431 (0.1641) Black Age 1-5 0.1593 (0.1695) 0.1541 (0.1787) Black Age 6-9 0.0909 (0.1551) 0.1123 (0.1754) Black Age 10-15 0.0772 (0.1540) 0.0944 (0.1609) Black Age 16-18 0.0514 (0.1528) 0.0641 (0.1602) Standard errors in parentheses. The above coefficients indicate the effect of face-to-face intervew requirement on individuals of the following demographic groups. The values in Column II were generated using the coefficients from Column II in Table V - Part A. The values in Columns III and IV were also generated using coefficients from the corresponding columns in Table V - Part A. Coefficients computed by summing the coefficients on the following four variables: the triple interaction term involving the relevant age dummy, the “Hispanic” dummy, and the “face-to-face interview” dummy; the double interaction term involving the “Hispanic” dummy and the “face-toface interview” dummy; the double interaction term involving the “face-toface interview” dummy and the relevant age dummy, and the non-interacted “face-to-face interview” dummy. I had previously noted that the face-to-face interview can have two competing effects on program take-up. On the one hand, it can reinforce take-up by serving as a venue for disseminating valuable information to applicants, while on the other, it can deter take-up by adding to the time and procedural burden of enrollment. I believe that the above results demonstrate both effects. The positive coefficients – indicating an increased probability of 35 uninsurance caused by the face-to-face interview requirement – on each age group within the Hispanic and Black population are likely due to the policy’s effect of adding to the time and effort needed to enroll in a given public insurance program. However, the decreasing coefficients indicate that this policy has a smaller effect on individuals of older age groups. This is likely because those families who initially enroll in public insurance (despite the presence of a face-toface interview) are less likely to drop out through attrition, due to the informational gain from the face-to-face interview. This mitigating effect on attrition serves to counterbalance the former effect that increases the likelihood of uninsurance. Given that a policy mitigating attrition will have the largest cumulative effect on children of older age groups, it is logical this latter effect reinforces public insurance take-up to the largest extent among older children. Accordingly, while the face-to-face interview requirement serves to increase the probability of uninsurance across the age spectrum, it has the smallest effect on individuals in the highest age bracket. In both columns III and IV, analysis of the face-to-face interview’s effect on the Black population again reveals positive coefficients across the age spectrum, with the trend of coefficients growing less and less positive with increasing age. It thus appears that for the Black population as well, the presence of a face-to-face interview requirement exerts the same two opposing effects observed for the Hispanic population. Interestingly, in Column IV, the coefficients on the Black age brackets are more positive than those on the corresponding Hispanic age brackets, indicating that on the whole, the presence of a face-to-face interview requirement increases the probability of uninsurance to a greater extent among Black individuals. I believe that this is because the face-to-face interview does not have nearly as much of a mitigating effect on attrition among Black individuals. I previously hypothesized that the language barrier is a major factor driving attrition among Hispanic individuals, hindering many in their ability to comprehend the necessary paperwork and enrollment procedures. I believe that the informational gain from the face-to-face interview is thus particularly large for Hispanics. For Black individuals, there is ostensibly no language barrier. Accordingly, while the face-to-face interview 36 may still mitigate attrition among Black individuals, this effect is not nearly as strong as it is in the Hispanic population. 5.5 Continuous Eligibility and Uninsurance I hypothesized that continuous eligibility may represent another policy with a mitigating effect on attrition among Hispanic children. The results from Table II indicate that as I had hypothesized, this policy lowers the probability of eligible uninsurance. States implementing this policy ostensibly reduce the administrative and procedural burden to remaining enrolled in public insurance by requiring individuals to re-enroll only once every 12 month period. Given my previous hypothesis that Hispanic individuals may be particularly averse to added administrative requirements to enrolling in public insurance, it is possible that children from Hispanic families will particularly benefit from the use of continuous eligibility. The set of regressions in Table V is structured the same way as the set presented in Table IV, with the “continuous eligibility” dummy replacing the “face-to-face interview” dummy. The explanatory variables in Column I are thus the “continuous eligibility” dummy, the four age dummies, and four interaction terms involving these two variables. Column II includes a series of four triple interaction terms generated by multiplying the “Hispanic” dummy with the double interaction terms in Column I, as well as a set of double interaction terms involving the age dummies and the “Hispanic” dummy. In Column III, the “Hispanic” dummy is replaced by the “Black” dummy. Finally, Column IV features both dummies, and all of the interaction terms from Columns II and III. In Regression I in the table below, none of the double interaction terms between the continuous eligibility dummy and the age dummies have coefficients are close to being statistically significant. These coefficients are all small in magnitude, and do not exhibit any discernible trend. The negative and statistically significant coefficient on the non-interacted 37 Table V - Part A Continuous Eligibility and Uninsurance I II III IV Continuous eligibility -0.0801 (1.97)** Hispanic 0.0849 (11.77)** -0.1010 (2.51)** -0.0743 (1.74)* -0.0333 (0.71) -0.0443 (1.34) 0.0852 (11.80)** Black -0.0185 (0.66) -0.0056 (0.73) -0.0059 (0.78) 0.0587 (1.41) -0.0154 (0.48) Age1to5 -0.0178 (1.04) -0.0526 (2.64)** -0.0010 (0.05) -0.0324 (1.81)* Age6to9 0.0166 (0.91) -0.0394 (1.93)* 0.0339 (1.64)* -0.0171 (0.93) Age10to15 0.0314 (1.78)* -0.0215 (1.06) 0.0519 (2.59)** 0.0007 (0.04) Age16to18 0.1261 (6.00)** 0.0559 (2.41)** 0.1468 (6.15)** 0.0687 (3.27)** CE* Age1to5 0.0056 (0.19) 0.0297 (0.81) -0.0052 (0.16) -0.0030 (0.09) CE* Age6to9 -0.0073 (0.25) 0.0417 (1.10) -0.0221 (0.70) -0.0105 (0.32) CE* Age10to15 -0.0039 (0.14) 0.0328 (0.91) -0.0120 (0.39) 0.0018 (0.06) CE* Age16to18 -0.0023 (0.08) 0.0377 (0.99) -0.0008 (0.03) 0.0137 (0.40) Hispanic* CE 0.07588 (1.25) 0.0307 (0.63) Hispanic* Age1to5 0.1261 (2.98)** 0.1062 (3.02)** Hispanic* Age6to9 0.2067 (4.45)** 0.1485 (3.96)** Hispanic* Age10to15 0.1897 (4.30)** 0.1543 (4.24)** Hispanic* Age16to18 0.2370 (4.77)** 0.1948 (4.70)** Black* CE -0.0094 (0.15) 0.0576 (0.94) Black* Age1to5 -0.0664 (1.77)* -0.0150 (0.43) Black* Age6to9 -0.0634 (1.64)* 0.0075 (0.20) Black* Age10to15 -0.0735 (2.01)** -0.0076 (0.22) Black* Age16to18 -0.0612 (1.55) 0.0072 (0.19) Hispanic* CE* Age1to5 -0.0630 (1.21) -0.0357 (0.77) Hispanic* CE* Age6to9 -0.1053 (2.30)** -0.0589 (1.33) Hispanic* CE* Age10to15 -0.0908 (1.92)* -0.0634 (1.49) Hispanic* CE* Age16to18 -0.1019 (2.17)** -0.0728 (1.68)* Black* CE * Age1to5 0.0551 (0.73) Black* CE * Age6to9 0.0728 (0.93) -0.0064 (0.11) -0.0212 (0.36) Black* CE * Age10to15 0.0412 (0.57) -0.0428 (0.79) Black* CE * Age16to18 -0.0025 (0.04) -0.0552 (1.00) n = 28,726. (*) denotes significance at 90% level, (**) denotes significance at 95% level. Z-statistics in parentheses. Sample restricted to children eligible for public insurance. Each column above denotes a different regression. In each regression, the dependent variable is a dummy taking on a value of 1 if the individual is uninsured, 0 otherwise. Dprobit regressions were used to generate estimates of coefficients on above explanatory variables. Column I includes double interaction terms involving "continuous eligibility" dummy and age dummies. Column II includes triple interaction terms involving "continuous eligibility" dummy, age dummies, and Hispanic dummy. Double interaction terms involving age dummies and "continuous eligibility" dummy, age dummies and “Hispanic” dummy, and “Hispanic” dummy and "continuous eligibility" dummy are included as well. Column III is structured in same way, but with “Hispanic” dummy replaced by “Black” dummy. Column IV includes both Hispanic” and “Black” dummies, and all the interaction terms from Columns II and III. For each regression, controls were included for subject's family income relative to FPL, and state fixed effe American background were included in each regression, but are not reported above. “continuous eligibility” variable does indicate that the policy lowers the probability of eligible uninsurance across the age spectrum. However, the statistically and economically insignificant 38 coefficients on the interaction terms indicate that this policy does not exert an increased effect on older individuals, as would be the case if it mitigated attrition in the general population. For example, the overall effect of continuous eligibility on eligible children aged 6-9 (calculated by adding the coefficient on the “continuous eligibility” dummy to that on the interaction term involving the “continuous eligibility” dummy and the “Age 6-9” dummy) is to reduce the probability of uninsurance by 8.74 percentage points. The policy’s overall effect on eligible children aged 16-18 is to lower the probability of uninsurance by 8.24 percentage points. As was the case in the previous section, I use the coefficients in Columns II, III, and IV in Table V to estimate the isolated effect of continuous eligibility on Hispanic and Black individuals of each age bracket. These estimates and their standard errors are calculated using the same methodology outlined in the previous section. The summed coefficients and their standard errors appear in the table on the following page. All of the standard errors from the following table are larger than the coefficients that they correspond to. Once again therefore, initial indications are that there is little information from the table above, as none of the coefficients are close to being statistically significant at either the 90% or 95% level. Once again though, I temporarily leave aside the high standard errors, and analyze the coefficients themselves to observe whether they reveal any notable results. Column II contains all negative coefficients, indicating that continuous eligibility, as hypothesized, serves to reduce the probability of uninsurance for eligible Hispanics across the age spectrum. While the three coefficients corresponding to the oldest age brackets are more negative than that corresponding to the “Age 1-5” bracket, they demonstrate no obvious trend of growing progressively more negative for older children. If a policy were to mitigate attrition across the age spectrum, such a trend would likely be present, given that if children of each age were kept from dropping out, a larger number of non-dropouts would accumulate in the age brackets corresponding to older children. Thus, while the use of continuous eligibility appears to lower the probability of uninsurance among eligible Hispanic children, it is unclear as to whether it serves 39 to mitigate attrition across the age spectrum. It is possible that this policy may lower the probability of uninsurance through some alternate mechanism, such as encouraging more eligible individuals to initially apply for enrollment. Table V - Part B Continuous Eligibility: Effect on Specific Demographic Groups II III IV Hispanic Age 1-5 -0.0584 (0.0977) -0.0583 (0.1149) Hispanic Age 6-9 -0.0887 (0.0935) -0.0720 (0.1179) Hispanic Age 10-15 -0.0831 (0.0939) -0.0642 (0.1114) Hispanic Age 16-18 -0.0893 (0.0940) -0.0617 (0.1081) Black Age 1-5 -0.0338 (0.1150) 0.0149 (0.1225) Black Age 6-9 -0.0330 (0.1179) -0.0074 (0.1209) Black Age 10-15 -0.0545 (0.1114) -0.0167 (0.1160) Black Age 16-18 -0.0870 (0.1081) -0.0172 (0.1106) Standard errors in parentheses. The above coefficients indicate the effect of continuous eligibility on individuals of the following demographic groups. The values in Column II of this table were generated using the coefficients from Column II in Table V - Part A. Coefficients computed by summing the coefficients on the following four variables: the triple interaction term involving the relevant age dummy, the “Hispanic” dummy, and the “continuous eligibility” dummy; the double interaction term involving the “Hispanic” dummy and the “continuous eligibility” dummy; the double interaction term involving the “continuous eligibility” dummy and the relevant age dummy, and the non-interacted “continuous eligibility” dummy. Column III suggests that continuous eligibility does appear to mitigate attrition among Black individuals. All four coefficients in this column are negative (though less negative than the coefficients on the corresponding Hispanic age brackets), and display a clear trend of growing more negative for children of older age brackets. However, the coefficients from Column IV do not lend much corroboration to the results in Column III. These coefficients are all relatively small in magnitude, and do not exhibit nearly as strong a trend of growing more negative for older children. The coefficients on the Hispanic age brackets in Column IV, however, remain relatively large in magnitude. It appears, for example, that the use of continuous eligibility reduces the probability of uninsurance among Hispanics aged 10-15 by 6.42 percentage points. The overall uninsurance rate for eligible Hispanic children aged 10-15 in the CPS sample is 26.41%. A 6.42 percentage point decrease in this uninsurance rate thus represents a 24.3% reduction in the overall rate of uninsurance for this age group. In column IV, the coefficients on 40 the Hispanic age brackets again do not exhibit any real trend of growing increasingly negative by age. Though it is impossible to definitively conclude anything from the results above, given the statistical insignificance of the coefficients, it does appear that the data from the above table is consistent with the hypothesis that the use of continuous eligibility lowers the probability of uninsurance for Hispanic children across all age brackets. Additionally, while there is some evidence that this policy mitigates attrition among Black individuals, it is again impossible to conclude this with confidence, given that this effect is not observed in the results from Column IV. 5.6 Continuous Eligibility, Self-Declaration of Income, and Different Income Levels In the previous two tables, I examined state enrollment procedures exerting different effects on individuals of different ethnic background. In the following table, I examine whether two procedures – use of continuous eligibility and self-declaration of income – may also have unique effects on individuals of different income levels. I hypothesize that continuous eligibility will have the strongest impact on those near the margin between eligibility and non-eligibility. This is because with this policy in place, an individual who has initially qualified for public insurance is able to remain enrolled in the program even if her family income exceeds the state threshold at some point during the 12-month period. Without this policy, the individual must reestablish eligibility each month, and will lose coverage if her family income rises above the state cutoff at any point during the year. Those with initial family incomes closest to the cutoff level are the ones who are most likely to see their incomes fluctuate over the cutoff level. Accordingly, they are likely to be benefited the most by the use of continuous eligibility. I hypothesize that the use of self-declaration of income will also benefit those near the margin between eligibility and non-eligibility. This is because those unsure of their eligibility status will ostensibly be less likely to apply for coverage if doing so entails an extensive financial audit that could very well reveal them to be ineligible. Allowing individuals to declare their own 41 incomes may also help those above the eligibility threshold, as it would seemingly make it easier for certain individuals to misrepresent their income in order to obtain coverage. To test the above hypotheses, it was necessary to know not only the family income of a particular child, but also where this income stood relative to the income eligibility cutoff in her state of residence. I created such a measure by dividing a given individual’s family income relative to the FPL by the FPL threshold for public insurance in her state of residence. This new measure expresses each individual’s family income as a proportion of the relevant income threshold. For example, in a state allowing only those with family incomes below 200% of the FPL to participate in public insurance, an individual whose family earns a salary equal to 100% of the FPL would have an income equaling 50% of the eligibility cutoff. Having created this continuous variable, I generated four dummy variables corresponding to specific ranges within this continuous variable’s distribution. Table VI uses these four dummy variables to test the hypotheses above. Column I employs as explanatory variables the four “% of eligibility cutoff” dummies, the continuous eligibility dummy, and the four interaction terms involving these two variables. These interaction terms are used to estimate the specific effect of continuous eligibility on individuals within the four different “% of eligibility cutoff” brackets. Column II is structured the same way, but includes instead the “self-declaration of income” dummy, and its interactions with the four “eligibility cutoff” dummies. Column I Uninsured = β1i*(Income Bracket)i + β2*(CE) + β3i*(CE)*(Income Bracket)i λ*Demographic controls + γ*State fixed effects + ε Column II Uninsured = β1i(Income Bracket)i + β2*(SD) + β3i*(SD)*(Income Bracket)i + λ*Demographic controls + γ*State fixed effects + ε 42 Table VI Continuous Eligibility and S elf Declaration of Income: Effects on Uninsurance Among Individuals of Different Income Levels I II 25-50% eligibility cutoff -0.0335 (4.17)** -0.0198 (2.94)** 50-90% eligibility cutoff -0.0525 (7.09)** -0.0527 (8.60)** 90-110% eligibility cutoff -0.0675 (8.01)** -0.0663 (9.27)** 110-150% eligibility cutoff -0.1041 (14.16)** -0.1093 (17.37)** Continuous eligibility 0.0521 (1.68)* Self-declaration of income CE* 25-50% cutoff 0.0091 (0.27) 0.0166 (1.21) CE*50-90% eligibility cutoff -0.0074 (0.60) CE*90-110% eligibility cutoff -0.0116 (0.75) CE*110-150% elirigiliby cutoff -0.0157 (1.12) SD* 25-50% cutoff -0.0653 (3.74)** SD*50-90% eligibility cutoff -0.0205 (1.17) SD*90-110% eligibility cutoff -0.0474 (2.26)** SD*110-150% elirigiliby cutoff 0.0128 (0.62) N = 41,027. (*) denotes significance at 90% level, (**) denotes significance at 95% level. Z-statistics in parentheses. Each column above denotes a different regression. In each regression, the dependent variable is a dummy taking on a value of 1 if the individual is uninsured, 0 otherwise. Dprobit regressions were used to generate estimates of coefficients on above explanatory variables. Sample restricted to children with family incomes below 150% of the eligibility cutoff in their respective states of residence. Column I includes double interaction terms involving "continuous eligibility" dummy and "% eligibility cutoff" dummies. Column II includes double interaction terms involving "self declaration of income" dummy, and the "% cutoff” dummies. Controls were included for subject's age, racial or ethnic background, and state fixed effects in each regression. The income dummies used in previous models are not included in any of the above regressions, as they are highly correlated with the "% eligibility cutoff" dummies used in the above regressions. In Column I, the coefficients on each of the double interaction terms involving the “continuous eligibility” and “income cutoff” dummies are statistically insignificant. While it does appear that the coefficients on these interaction terms appear to grow increasingly negative with increased income relative to the relevant public insurance cutoff, it is difficult to make any inferences from these coefficients given their low levels of significance. Additionally, in order to find the overall effect of continuous eligibility on individuals of different income levels, it is necessary to add the coefficients on the double interaction terms to that on the non-interacted “continuous eligibility” variable. This latter coefficient is not only positive and significant at the 43 90% level, but also larger in magnitude than any of the double interaction terms involving the “continuous eligibility” variable. Thus, it appears that if anything, the policy appears to increase the probability of uninsurance across the income brackets listed. In Column II, all four interaction terms involving the “self-declaration of income” dummy and the “% of eligibility cutoff” dummies have negative coefficients that are larger in magnitude than the positive coefficient on the non-interacted “self-declaration of income” variable (which is also statistically insignificant). As a result, adding this latter coefficient to the coefficients on the double interaction terms reveals that self declaration of income does indeed reduce the probability of uninsurance across the income brackets listed. Additionally, there are two double interaction terms with negative and statistically significant coefficients. One is the interaction term involving the “90-110% cutoff” dummy. This observation supports the hypothesis that use of income self declaration significantly reinforces public insurance take-up among those children with family incomes near the eligibility cutoffs in their states of residence. The coefficient on the interaction term involving the “110-150% cutoff” dummy, however, is positive and statistically insignificant. This finding refutes the hypothesis that use of income selfdeclaration would reduce uninsurance among those with incomes well above the relevant Medicaid or SCHIP cutoff by making it easier to deliberately misrepresent family income. Interestingly, the other interaction term in Regression II with a statistically significant coefficient is the one involving the “25-50% cutoff” dummy. This coefficient is also the most negative out of the four observed. Thus, it appears that this policy has a very strong reinforcing effect on public insurance take-up for those far below the eligibility cutoff. This surprising result may indicate that proper income documentation is a more difficult task for individuals at the lower end of the income spectrum. Thus, easing requirements for income documentation may preferably benefit these individuals, reinforcing public insurance take-up among them. 6. CONCLUSION This thesis began with two observations: first, that children are far likelier to be 44 uninsured as they grow older, and second, that this effect is particularly pronounced among children of Hispanic ethnicity. This paper hypothesizes that attrition is the main mechanism underlying the dramatic increase in uninsurance among older age groups in minority child populations, and that state policies tied to public insurance enrollment and re-enrollment can significantly impact the magnitude of this effect. A major recurring theme, very much in keeping with the spirit of Amartya Sen’s seminal work discussed earlier, is that state policies exert different impacts on different sectors of the population. The use of 12 month continuous eligibility (the effects of which are summarized in Table V, part B) appeared to have a large effect in mitigating uninsurance in the Hispanic population, but had a far less conclusive effect on the Black population. Meanwhile, the use of a face-to-face interview requirement appeared to have the dual effect of deterring initial enrollment, and of mitigating attrition among those who did enroll. I concluded that this policy’s mitigating effect on attrition is again far stronger in the Hispanic population than in the Black population. The above two results indicate that Hispanic individuals may exhibit the most sensitivity to the implementation of certain policies, and may benefit the most both from a simplification of the administrative procedures associated with public insurance enrollment, and from effective dissemination of program information. This is an important conclusion, given that Hispanic children comprise a disproportionate share of the uninsured child population. This paper thus suggests that by implementing a set of policies catered to Hispanic individuals, states can make a major inroad against uninsurance in this population. This paper also finds that state policy exerts differing effects not only on those of different racial groups, but also on those of different income levels. Indeed, Table VI shows that the policy of self-declaration of income appears to have the largest reinforcing effect on program take-up not among those near the income eligibility cutoffs in their particular states, but rather on those far below this cutoff. This paper thus suggests that the use of this policy will ostensibly be tied to relatively few instances of private insurance crowd-out, as these lower income individuals 45 among the eligible population are anyway fairly unlikely to have access to private health insurance. Additional work on this topic would undertake further investigation into the role of state policy on uninsurance. It would examine other state-level factors relevant to enrollment, such as outreach initiatives and program accessibility for parents, and examine the effect that these state factors have on the likelihood of uninsurance for children of different ages, ethnicities, and races. From it, a comprehensive picture can emerge of the specific policies that can be targeted towards the most needy in order to ensure their participation in public insurance. Additionally, the use of a larger data set for future investigation may allow for more significant results in Tables IV and V, which examine the impact of continuous eligibility and the face-to-face interview requirement on individuals of different demographic groups. Society pays a large cost for leaving individuals uninsured. In addition to the social welfare cost from the worse health outcomes of the uninsured, society bears a significant cost for funding charity care and emergency room visits for the uninsured, whose routine health needs often become emergencies due to neglect. Indeed, in Massachusetts alone, the estimated public cost of funding care for the uninsured was $500 million in 2004.14 If there is one simple conclusion that is present throughout this paper’s analysis, it is that society can make a significant inroad against child uninsurance – and all the costs associated with it – through simply addressing procedural issues tied to enrollment, without the need for any complex legislation expanding the existing public healthcare safety net. While several individual-specific characteristics do ultimately play a significant role in shaping public insurance take-up, public policy is of considerable importance as well. 14 Rowland, Christopher. “Mass. Health plan seems unlikely to be US model.” Boston Globe. April 14, 2006. 46 REFERENCES Bae, Jay P. and Katherine E. Gardner. (Summer 2004) “Low-Income Children’s Participation in a Public Health Insurance Program in Georgia.” Journal of Family and Economic Issues: 225-243. Blewett, Lynn; Davern, Michael, and Holly Rodin. (November/December 2004) “Covering Kids: Variation in Health Insurance coverage Trends by state 1996-2002. Health Affairs: 170-180. Currie, Janet. (December 1995) “Do Children of Immigrants Make Differential Use of Public Health Insurance.” NBER Working Paper 5388. Currie, Janet and Jonathan Gruber. (December 1996) “Saving Babies: The Efficacy and Cost of Recent Changes in the Medicaid Eligibility of Pregnant Women.” The Journal of Political Economy: 1263-1296. Currie, Janet, and Jonathan Gruber. (May 1996) "Health insurance eligibility, utilization of medical care, and child health." Quarterly Journal of Economics: 431-466. Cutler, David and Jonathan Gruber. (May 1996) “Does Public Insurance Crowd out Private Insurance?” The Quarterly Journal of Economics: 391-430. Davidoff, Amy J; Garnett, Bowen; Makuc, Diane M and Matthew Schirmer. (September 2000) “Children Eligible for Medicaid but Not Enrolled: How Great a Policy Concern?” The Urban Institute: 1-7. Eileen Ellis, Vernon K. Smith. (April 2001) “Eliminating the Medicaid Asset Test for Families: A Review of State Experiences.” Kaiser Commission on Medicaid and the Uninsured. pp. 1-64. Elbel, Brian and Karl Kronebusch. (June 2004) “Enrolling Children in Public Insurance: SCHIP, Medicaid, and State Implementation.” Journal of Health Politics, Policy and Law: 451-489. Flores G, Abreu M, Brown V, Tomany-Korman SC. (November-December, 2005) “How Medicaid and the State Children's Health Insurance Program Can Do a Better Job of Insuring Uninsured Children: The Perspectives of Parents of Uninsured Latino Children.” Ambulatory Pediatrics: 332340. Kaestner, Robert and Neeraj Kaushal. (November 2003) “Welfare Reform and Health Insurance Coverage of Low-income Families.” Journal of Health Economics: 959-981. Long, Sharon K; Coughlin, Teresa and Jennifer King. (February 2005) “How Well Does Medicaid Work in Improving Access to Care?” Health Services Research: 39-58. Lo Sasso, Anthony and Thomas Buchmueller. (2004) “The Effect of the State Children’s Health Insurance Program on Health Insurance Coverage.” Journal of Health Economics: 1059-1082. Santerre, Rexford and Stephen Neun. (2004) “Health Economics.” Shen, Yu-Chu, and Stephen Zuckerman. (June 2005) “The Effect of Medicaid Payment Generosity on Access and Use Among Beneficiaries.” Health Services Research: 723-745. Shore-Sheppard, Lara D. (October 2000)“The Effect of Expanding Medicaid Eligibility on the Distribution of Children’s Health Insurance Coverage.” Industrial Labor Relations Review: 59-77. 47 Sommers, Benjamin D. (February 2005) “From Medicaid to Uninsured: Drop-Out among Children in Public Insurance Programs.” Health Services Research: 59-78. Thorpe, Kenneth E. and Curtis Florence. (October 2000) “Medicaid Eligible, but Uninsured: The New York State Experience.” United Hospital Fund: 1-7. Wolfe, Barbara and Scott Scrivner. (Summer 2005) “The Devil May Be in the Details: How the Characteristics of SCHIP Programs Affect Take-Up.” Journal of Policy Analysis and Management: 499-522. Appendix: Enrollment procedures used by state Self declaration of income Asset test Initial face-toface interview Presumptive eligibility 12 month continuous eligibility Re-enrollment Interview ALABAMA ALASKA ARIZONA ARKANSAS CALIFORNIA COLORADO CONNECTICUT DELAWARE WASHINGTON DC FLORIDA N** N N** Y N N N N N N N N N N N N* N N N N N* N N N N N N N N N N N N N Y N N N N N Y N N** N Y N** N N** N N* N N N* N N N N N N N GEORGIA HAWAII IDAHO ILLINOIS INDIANA IOWA KANSAS KENTUCKY LOUISIANA MAINE Y Y Y N N N N N N N N N N N N N N N N N N N N N N N N Y N N N N N Y N N N N N N N N Y Y N N** Y N N* Y N N N N N N N N N N MARYLAND MASSACHUSETTS MICHIGAN MINNESOTA MISSISSIPPI MISSOURI MONTANA NEBRASKA NEVADA NEW HAMPSHIRE Y N Y N N N N** N N N N N N N N N N* N N* N N N N N Y N N N N N N Y Y N N Y N Y N N* N N Y N Y N N** N N** N N N N N Y N N N N N 48 Appendix (continued) NEW JERSEY NEW MEXICO NEW YORK NORTH CAROLINA NORTH DAKOTA OHIO OKLAHOMA OREGON PENNSYLVANIA RHODE ISLAND SOUTH CAROLINA SOUTH DAKOTA TENNESSEE TEXAS UTAH VERMONT VIRGINIA WASHINGTON WEST VIRGINIA WISCONSIN WYOMING N N N N N N Y N N N N N N N N N N N** N N N N N* N N N N N N N Y Y N** N N N N N N N Y N Y Y N** N N N N** N N N N N N N N N N N N N N N N Y N N N N Y N N N Y N* N N N N N N N N Y N Y N N N N N N N N N N N N N N N N N N* N N N N** N N** N** Y N Y N N Y N N N N N N N N Y refers to policy used for both state Medicaid and SCHIP (or Medicaid alone in states without separate SCHIP program). N** refers to policy used by state SCHIP but not state Medicaid. N* refers to policy used by state Medicaid but not SCHIP. N refers to policy used by neither program. 49