Community characteristics and qualified health plan selection during the first open enrollment period.
Data Sources/Study Setting. Administrative data on qualified health plan selections at the ZIP code area merged with survey estimates from the American Community Survey.
Study Design. Descriptive and regression analyses.
Data Collection/Extraction Methods. Data were generated by healthcare.gov and from a household survey.
Principal Findings. Thirty-one percent of the variation in qualified health plan selection ratios resulted from between-state differences, and the rest was driven by local area differences. Education, language, age, gender, and the ethnic composition of communities contributed to disparate levels of plan selection. Medicaid expansion states had a qualified health plan selection ratio that was 4.4 points lower than non-Medicaid expansion states, controlling for covariates.
Conclusions. Our results suggest community-level differences in the intensity or receptiveness to outreach and enrollment activities during the first open enrollment period.
Key Words. Health insurance marketplace, qualified health plan
By the end of the third open enrollment period, 12.7 million consumers had selected a qualified health plan (QHP) or were re-enrolled in coverage through a Health Insurance Marketplace (Assistant Secretary for Planning and Evaluation [ASPE] 2016). Despite initial success, the Henry J. Kaiser Family Foundation [KFF] (2016) estimates that 54 percent of the potential market has yet to enroll, and there has been tremendous variation between the states. Existing estimates suggest that over 60 percent of the potential market in Florida has selected a plan versus less than 25 percent in states such as Minnesota and Iowa (KFF 2016). There is also substantial geographic variation within the states. For example, in the state of Texas alone, the share of the target market that had obtained a plan at the close of the first open enrollment period ranged from 6 to 51 percent across different communities (KFF 2014b).
Identifying the population and policy factors that contribute to geographic variation in take-up is critical to developing efficient outreach strategies and devising policy options that will decrease barriers to enrollment. High take-up and retention is needed to promote population health and financial well-being, while promoting stability in the marketplaces. The goal of this research brief was to examine the state and community factors that contributed to geographic variation in QHP selection in states that used health-care.gov during the first open enrollment period (OEP1).
Our analysis contributes to four developing strands of evidence on the Health Insurance Marketplaces. Analyses of individual-level administrative data have found that the majority of consumers are women and are older than 35 (ASPE 2014a). Most consumers (around 85 percent) receive financial assistance and appear sensitive to premiums (Burke, Mirsa, and Sheingold 2014). Survey research has focused on estimating changes in the number of uninsured and have found that (by 2015) between 14 and 17 million adults had gained coverage since the end of 2013, with larger gains for low-income, non-whites, and Medicaid expansion state residents (ASPE 2015; Carman, Eibner, and Paddock 2015; Long et al. 2015). Other studies have examined variation in premiums and suggest that the type of exchange (clearinghouse vs. active purchaser) and the composition of rating areas have important effects on premiums (Dickstein et al. 2015; Krinn, Karaca-Mandic, and Blewett 2015).
The research most similar to ours pairs aggregated counts of QHP enrollment with survey estimates to measure a selection rate defined as the number of people that have selected a QHP divided by the number of likely market participants (Enroll American 2014; KFF 2014a; Buettgens, Kenney, and Pan 2015; Drake, Abraham, and McCullough 2016). Previous estimates have been produced at the state, county, or Public Use Microdata Area (a statistical geography of about 100,000 people) and have been used to describe the extent of geographic variation in plan selection, but not the full range of potential community and policy factors associated with higher or lower levels (the exception being Drake, Abraham, and McCullough 2015). We contribute to this line of inquiry by examining much lower levels of geography than previously described: ZIP code areas. We did not aggregate our data to a higher level (e.g., rating area or county) because there is often substantial variation in the health and socioeconomic character of communities within the same county (cf. Yen and Kaplan 1999; Ponce et al. 2005; Yousey-Hindes and Hadler 2011). Similarly, we expected consumer demand, barriers to enrollment, and ultimately QHP selection to also vary at the subcounty level. Finally, our focus on ZCTAs aligns our study with evidence from qualitative studies of consumers that suggest that outreach tailored to the local community might benefit enrollment (Martin et al. 2014). In addition to examining lower levels of geography, we also provide new evidence on the association of QHP selection with community-level factors.
QHP data came from the Office of the Assistant Secretary of Planning and Evaluation (ASPE) (ASPE 2014b). For each ZIP code in the 36 states that used healthcare.gov in OEP1, the file indicated the count of individuals that selected a QHP. We focused on OEP1 because that allowed us to unambiguously interpret the plan count as the number of new enrollments. We chose not to use data from the subsequent open enrollment periods because ASPE pools new enrollment and renewals, which may be influenced by different factors. Nevertheless, we conducted a sensitivity analysis using OEP2 data to see if the results differed (Appendix).
To protect confidentiality, ASPE suppressed data for ZIPs that had less than 50 plan selections. This affected 5 percent of QHP selection, but 59.6 percent of ZIPs. As might be expected, many (82 percent) but not all of the missing ZIPs were rural. In our main analysis, we included the missing ZIPs by imputing plan selection counts. This was done by allocating the difference between the complete state count and the state count inferred by nonmissing cells within a state. The allocation was done proportional to population size, but no area was allocated more than 49 plans. Our regression models included an indicator that flags when a ZIP was imputed. Sensitivity analyses demonstrated that results did not differ when removing the imputed cells (Table A2). After imputing missing data, we converted ZIP codes to ZIP code tabulation areas (ZCTAs) to be compatible with Census geography.
QHP selection levels were determined by the size of the population eligible to select a plan and the proportion of eligible consumers that took-up coverage. Our primary interest was in variation in take-up rather than variation in the eligible population, so we had to account for differences in population size. We accomplished that by using a population denominator obtained from the 2009-2013 American Community Survey (ACS). The clear advantage of the ACS is that it provides a large set of demographic estimates at low levels of geography. The main disadvantage is that ZCTA estimates from the ACS are only available in prepopulated tables which precluded us from creating custom variables specific to this analysis.
Any legally residing, nonincarcerated individual can buy coverage through a Marketplace. This suggests that the true participation rate denominator would only exclude undocumented immigrants and incarcerated individuals. However, some consumers are very unlikely to ever select a plan, so it makes little sense to include such people in a QHP market definition. The Kaiser Family Foundation's (KFF) definition of the likely QHP market starts with people who are either uninsured or have nongroup or certain types of group coverage (KFF 2014a). They then exclude those with income below 100 percent of poverty or below Medicaid/CHIP eligibility levels (whichever is higher), those that they impute to have an employer offer, and those that they impute to be nonlegal residents. The Census Bureau's prepopulated ZCTA-level tables from the ACS do not include all the needed inputs to replicate the KFF definition. Instead, we used a broader denominator that was feasible with available data: nonelderly individuals that were either uninsured or had only an individual market plan. We divided QHP selection counts by this denominator to construct a measure we refer to as the QHP ratio. We multiply the result by 100 so that the ratio indicates the number of plan selections per 100 nonelderly people that were uninsured or covered in the individual market on an average day in the 2009-2013 period. We do not interpret the ratio as a QHP participation rate because our denominator does not reflect the exact universe of people that could potentially purchase a QHP.
Sociodemographic factors came from the 2009-2013 ACS and reflected the characteristics of the entire ZCTA (we were unable to control for characteristics of the denominator due to limitations of using prepopulated tables). The QHP ratio is likely influenced by income, immigration, and employer offers --factors that determine eligibility for premium subsidies, for example, advanced premium tax credits (APTCs). We controlled for the percent of the ZCTA that had family incomes between 100 and 400 percent FPL in states that did not expand Medicaid or between 139 and 400 percent in states that did. Unfortunately, the Census does not release detailed income by health insurance statistics, so we were unable to include income in the denominator of the QHP ratio in our ZCTA analysis. However, we explore this issue in more depth in a robustness analysis (see below). We also controlled for the fraction of the ZCTA that was noncitizen (the ACS lacks a more detailed measure of legal residency). Finally, we included an imputed measure of employer health insurance offer rates using industry-specific data for private employees, obtained from the Medical Expenditure Panel Survey--Insurance Component (MEPS-IC). The imputed variable was set equal to the product of the national, industry-specific offer rate and the fraction of workers employed in that industry for each given ZCTA. The other coefficients in the model are robust to excluding this imputed variable.
Other demographic covariates included gender, race/ethnicity (percent white, non-Hispanic), and age structure (percent age 35-84). Social covariates included the percent of 15+ year olds that were married, the percent of the nonelderly population that had a functional disability, the percent that spoke English less than very well, the percent of 25+ year olds that had less than a high school education, and the percent of the labor force that was unemployed. We also included an indicator for rural status which we obtained from the Missouri Census Data Center.
We hypothesized that premiums would be negatively correlated with QHP ratios. We obtained premiums for the second-lowest cost silver plan for 27 year olds from HIX Compare (Breakaway Policy Strategies 2014). Premiums varied by rating areas which we translated to ZCTAs. We excluded Nebraska, Alaska, and Idaho because they defined rating areas using three-digit ZIP codes that could not be cross-walked to ZCTAs.
We hypothesized that provider supply would be positively associated with QHP ratios. The availability of providers could make QHPs more attractive, or selection could be facilitated by medical professionals. This might be especially true of federally qualified health centers (FHQCs), which received substantial funding to support enrollment activities during OEP1 (Volk et al. 2014). However, the presence of safety-net clinics might discourage QHP selection if they act as substitutes for health insurance coverage. We obtained two measures of primary care supply (as of 2010), from the Health Resources and Services Administration's Data Warehouse, at the primary care service area (PCSA) (Goodman et al. 2003). We included the number of primary care physicians per 100,000 persons and the combined number of FQHCs, rural health centers, and "look-alikes" per 100,000 persons.
We considered three state policy measures (see Appendix for coding). The first indicated if the state had expanded Medicaid at the time of OEP1. The expansion decision could have had spillovers on the QHP market if outreach for Medicaid and QHPs were complements (leading to more QHP selection). However, the added complexity might have made enrollment and marketing more confusing or divisive (leading to less QHP selection). We also examined if state versus federal responsibility for Medicaid eligibility determination affected QHP ratios. Each state can opt to have the federal government ("federal determination states"), or the state ("assessment states") makes final decisions on Medicaid determinations in the Marketplaces. This policy parameter could affect who enrolled in Medicaid versus a QHP If states retained determination authority, they may have an incentive to be conservative in Medicaid determinations because of their financial liability in Medicaid enrollment (leading to more QHP selection). There could also be different experiences of eligibility determination by consumers (e.g., length until notification) that could affect QHP selection. Finally, we examined the type of Marketplace. We were not able to include State-Based Marketplaces, but we differentiated between states that were federally supported or partnership states (referred to as partnership states) and states that were federally facilitated. We had no specific expectations regarding how partnership status would affect QHP ratios, other than it shaped the institutional environment that consumers navigated.
Our units of analysis were ZCTAs, and our estimates should not be interpreted as reflecting individual behavior. Our final file consisted of 22,356 ZCTAs from 33 states. We used maps and box plots to inspect geographic distributions. We then estimated a regression model that used a beta distribution to account for a nonnormally distributed outcome (Ferrari and Cribari-Neto 2004). Our beta regression approach was designed for proportion data and is not appropriate when values are greater than or equal to 1. Thus, we did not include data from ZCTAs that had QHP ratios equal to or exceeding 1 (n = 50). The model and all descriptive statistics were weighted by the size of each ZCTA's nonelderly civilian noninstitutionalized population. We entered most covariates into the model as standardized variables. For example, the percentage of a ZCTA that was female was specified to have a mean of 0 and a standard deviation of 1. Standard errors were clustered on state.
Figure 1 presents ZCTA-level QHP ratios in the 33 states we observed. The average QHP ratio was 13.6 per 100 nonelderly individuals that were uninsured or covered in the individual market with a range of 0.4-98. Communities in Wisconsin, Michigan, New England, and Florida had relatively high QHP ratios. Communities in West Virginia, Iowa, and Texas had relatively low QHP ratios.
[FIGURE 1 OMITTED]
The box plots in Figure 2 further explore the distribution. The graph suggests substantial heterogeneity between and within states. To decompose the total variance into ZCTA and state variability, we fitted an empty random effects model allowing for random state intercepts (i.e., ZCTAs were nested within states). The model suggested that 31 percent of the variation was attributable to state variation (95 percent CI: 22 percent, 42 percent). In other words, the majority of the variation occurred between communities, not states. We also estimated the relative variation that occurred between and within counties by replacing the random state effects with random county effects. We found that 32 percent of variation occurred between counties. This held true if we nested ZCTAs within counties and counties within states.
[FIGURE 2 OMITTED]
Cook County, IL, provides a useful anecdote for exemplifying the utility of our ZCTA-level approach. If we had aggregated our data at the county level, we would have observed a QHP ratio of 10.0. However, the QHP ratios of the ZCTAs within Cook County ranged from 2.7 to 57.7 and had a standard deviation of 4.3. The within-county variation in QHP selection in Cook County may not be surprising given the broader geographic distribution of socioeconomics within the county (U.S. Census Bureau 2015).
Table 1 presents results from the beta regression. The first column lists the weighted mean and standard deviation (when appropriate) for a given characteristic. For example, 22.3 percent of ZCTAs are located in a Medicaid expansion state. The average premium for the second-lowest cost silver plan (for 27 year olds) was $215.20 with a standard deviation of $36.50.
The second set of columns describes the adjusted change in QHP ratios given a change in each community characteristic (marginal effect). Location in a Medicaid expansion state was associated with a 4.4 point lower QHP ratio (p < .001). Location in a federal determination state was also associated with a significantly lower QHP ratio. The association between partnership status and QHP ratios was nonsignificant.
A one standard deviation increase in the share that was 35-64 years old (a change of 5.3 percentage points) was associated with a 3.7-point increase in the QHP ratio (p < .001). The percent female was also positively associated with the QHP ratio. The percent that was Hispanic was negatively associated, but interestingly, a one standard deviation increase in the percent that was non-English speaking was associated with a 2.4-point increase in the outcome (p < .001). We were concerned that language may have been collinear with ethnicity and/or citizenship status. However, the coefficient on the language variable was 1.6 (p < .001) when excluding ethnicity and citizenship.
The percent that had incomes in the APTC range was positively associated with QHP selection as anticipated. However, the other variables related to APTC eligibility, citizenship status, and imputed employer offers did not have a statistically significant association with the outcome. Unemployment was associated with a statistically significant increase in the QHP ratio. A one standard deviation increase in the percent of adults with less than a high school degree was associated with a 4.6-point decrease in the QHP ratio (p < .001).
The association between premiums and the outcome was not statistically significant. Unemployment and primary care physician supply at the PCSA level were positively correlated with the QHP ratio at a significant level, but the number of per capita safety-net clinics was not.
The largest concern to internal validity we faced is that it was possible that the variables we included in our model were correlated with unmeasured determinants of the QHP ratio. For example, while we controlled for community-wide income, we did not control for the income distribution of the population in our denominator because such data are not available at the ZCTA level. To help understand how important this specific limitation was to our conclusions, we conducted a robustness test based on PUMA-level estimates derived from microdata (Appendix). In these data, we observed less geographic detail, but we had more control over the denominator which we defined as nonelderly citizens above 100 percent FPL and Medicaid/CHIP income eligibility limits, and either uninsured or enrolled in a nongroup plan. If the results we obtained from the PUMA-level regression were substantially different than our preferred ZCTA-level model, it would suggest that our results were biased by spurious correlations arising from variation in the size of the likely QHP mark that was not being measured by our controls. However, the results showed a similar pattern. In other sensitivity analyses, we found that our findings were not sensitive to alternative model specifications (Poisson, GLM, OLS), to excluding imputed cells, or to using OEP2 plan selection data as the dependent variable (Appendix). The last of these tests was particularly important because it demonstrated that the OEP1 results were not an artifact of the technological problems that affected healthcare.gov during OEP1.
Qualified health plans offered through Health Insurance Marketplaces are one of the main vehicles established by the Affordable Care Act for individuals, their families, and small business to purchase health insurance coverage. The Marketplace provides information to consumers, facilitates enrollment, and provides premium subsidies to income eligible enrollees. Identifying the population and policy factors that contribute to geographic variation in take-up is important to developing and refining policies and enrollment strategies. Robust Marketplace enrollment is important to public health, consumers' financial well-being, and to the stability of the exchanges. In this descriptive study, we present early evidence on the predictors of geographic variation in QHP ratios in 33 states that used healthcare.gov during the first open enrollment period.
The primary limitation of this study was that we could not perfectly measure APTC eligibility, the primary conceptual factor affecting the likelihood of QHP selection. It is possible that unmeasured variation in APTC eligibility was correlated with other variables in our model, leading to biased coefficients. However, community-level income, citizenship status, and imputed employer offer rates mitigated this risk. Furthermore, a robustness test showed that our ZCTA results were similar to PUMA-level estimates that came from microdata and used a narrower definition of market size. It is also important to note that while we used a different measure than the KFF, both approaches yielded similar geographic variation. KFF's OEP1 estimate for the states we observed was 26.4 percent (vs. our estimate of 13.1), but the Spearman rank correlation between our state-level estimates and those from Kaiser was 0.91 (Appendix).
Our results suggest that at the end of OEP1, a substantial share of the population likely to benefit from a QHP had not enrolled, a trend that continued through the end of OEP2 (Appendix). We found substantial geographic variation in QHP ratios that was concentrated at low levels of geography. Enrollment in communities with low levels of take-up might benefit from additional outreach efforts tailored to the local context. Our results complement recent qualitative research that stresses the importance of messaging tailored to local conditions and attitudes (Martin et al. 2014).
Our results suggest several factors that may help in targeting and tailoring outreach. We found that the percent of people aged 35-64 and the percent female were positively associated with QHP ratios. Interestingly, the share of non-English speakers was also positively associated with QHP ratios. Possible explanations could include more pent-up demand, or that the individual mandate was more salient in communities that are more likely to deal with immigration officials. The percent Hispanic, percent of adults with less than a high school education, and percent unemployed were associated with lower rates, suggesting that future outreach should target communities with these characteristics.
We observed a moderate but positive correlation between QHP ratios and primary care physician supply, suggesting that providers may play a role in increasing community-level plan selection either through increasing the value of insurance or by disseminating information. However, the evidence on the role of providers was mixed as we did not find evidence that the number of safety-net clinics mattered.
In our model, premium levels did not have a significant association with plan selection. Given that we observed actual premiums and not APTC-adjusted premiums and 85 percent of consumers received financial assistance, this may not be surprising. Previous research suggests that consumers opt for plans in their choice set that have lower premiums (Burke, Mirsa, and Sheingold 2014). Our results (from aggregate data) do not contradict such an effect.
While we found that within-state variation appeared to be paramount, there was still important variation between states. Most important, Medicaid expansion states tended to have lower QHP ratios even after controlling for income and measures of immigration and employer coverage offers. This association persisted in OEP2, in our PUMA-level robustness test (Appendix), and is consistent with other research (KFF 2014b; Buettgens, Kenney, and Pan 2015; Drake, Abraham, and McCullough 2015). Buettgens, Kenney, and Pan (2015) suggest that this pattern is driven by differential QHP selection among non-Medicaid eligible consumers with incomes less than 150 percent of poverty. It could be that states that expanded Medicaid added complexity to consumer options which may have made marketing more confusing and/or increased the perceived complexity of enrollment (leading to less QHP selection). Confusion would likely be the highest for expansion state residence near the margin of APTC and Medicaid eligibility (i.e., below 150 percent FPL). These results in combination with previous qualitative research suggest that clarifying coverage options in low-income communities located in expansion states may increase QHP enrollment (Martin et al. 2014). We also found that areas located in federal determination states had lower QHP ratios. The association we observed could result from a tendency by the federal government to be less conservative in Medicaid determination (resulting in more consumers ending up on Medicaid) and/or different consumer experiences that result in different QHP ratios.
In addition to describing the amount of geographic variation in QHP ratios during OEP1 and identifying important policy and sociodemographic factors associated with that variation, this brief demonstrated a relatively simple method of monitoring QHP selection levels. Our approach was based on publically available data and could easily be replicated by state analysts or other parties interested in tracking QHP enrollment.
Joint Acknowledgment/Disclosure Statement: This project was supported, in part, by a grant from the Robert Wood Johnson Foundation to the State Health Access Data Assistance Center. We thank Karen Turner for excellent CIS support and the editors and three anonymous reviewers for their comments.
Assistant Secretary for Planning and Evaluation [ASPE]. 2014a. "Health Insurance Marketplace: Summary Enrollment Report for the Initial Annual Open Enrollment Period." Department of Health and Human Services [accessed on December 15, 2014]. Available at http://www.aspe.hhs.gov/health/reports/2014/MarketPlaceEnrollment/Apr2014/ib_2014Apr_enrollment.pdf
Assistant Secretary for Planning and Evaluation [ASPE]. 2014b. "Plan Selections by ZIP Code in the Health Insurance Marketplace." Department of Health and Human Services [accessed on December 1, 2014]. Available at http://www.aspe.hhs.gov/health/reports/2014/MarketPlaceEnrollment/EnrollmentByZip/rpt_EnrollmentByZip.cfm
Assistant Secretary for Planning and Evaluation [ASPE]. 2015. "Health Insurance Coverage and the Affordable Care Act." Department of Health and Human Services [accessed on March 17, 2015]. Available at http://www.aspe.hhs.gov/health/reports/2015/uninsured_change/ib_uninsured_change.pdf
Assistant Secretary for Planning and Evaluation [ASPE]. 2016. "Health Insurance Marketplaces 2016 Open Enrollment Period: Final Enrollment Report." Department of Health and Human Services [accessed on April 16, 2016]. Available at https://aspe.hhs.gov/sites/default/files/pdf/187866/Finalenroll-ment2016.pdf
Breakaway Policy Strategies. 2014. "HIX Compare Dataset" [accessed on December 1, 2014]. Available at http://www.rwjf.org/en/library/research/2014/03/break-away-policy-dataset.html
Buettgens, M., G. M. Kenney, and C. Pan. 2015. Variation in Marketplace Enrollment Rates in 2015 by State and Income. Washington, DC: Urban Institute [accessed December 10, 2015]. Available at http://www.rwjf.org/content/dam/farm/re-ports/issue_briefs/2015/rwjf424382
Burke, A., A. Mirsa, and S. Sheingold. 2014. "Premium Affordability Competition and Choice in the Health Insurance Marketplace, 2014." ASPE Research Brief. Department of Health and Human Services [accessed on December 1, 2014]. Available at http://www.aspe.hhs.gov/health/reports/2014/Premiums/2014MktPlacePremBrf.pdf
Carman, K., C. Eibner, and S. Paddock. 2015. "Trends in Health Insurance Enrollment, 2013-2015." Health Affairs 34: 1044-8.
Dickstein, M., M. Duggan,J. Orsini, and P. Tebaldi. 2015. "The Impact of Market Size and Composition on Health Insurance Premiums: Evidence from the First Year of the Affordable Care Act." American Economic Review: Papers and Proceedings 105 (5): 120-5.
Drake, C.J. M. Abraham, and J. S. McCullough. 2016. "Rural Enrollment in the Federally Facilitated Marketplace." Journal of Rural Health 32 (3): 332-39.
Enroll American. 2014. "Changes in Uninsured Rates by County" [accessed on January 15, 2015]. Available at http://www.enrollamerica.org/state-maps-and-info/changes-in-uninsured-rates-by-county/
Ferrari, S., and F. Cribari-Neto. 2004. "Beta Regression for Rates and Proportions." Journal of Applied Statistics 27 (7): 799-815.
Goodman, D., S. Mick, D. Bott, T. Stukel, C. Chang, N. Marth, J. Poage, and H. Carretta. 2003. "Primary Care Service Areas: A New Tool for the Evolution of Primary Care Services." Health Services Research 38 (1 Pt 1): 287-309.
Henry J. Kaiser Family Foundation [KFF]. 2014a. "Methodology for Estimating the Number of People Eligible for Premium Tax Credits under the Affordable Care Act" [accessed December 15, 2014]. Available at https://kaiserfamilyfounda-tion.files.wordpress.com/2014/11/8509-methodology-for-estimating-subsidy-eligible-individuals.pdf
Henry J. Kaiser Family Foundation [KFF]. 2014b. "Marketplace Enrollment as a Share of the Potential Marketplace Population, April 2014" [accessed on December 15, 2014]. Available at http://kff.org/interactive/mapping-marketplace-enrollment/
Henry J. Kaiser Family Foundation [KFF]. 2016. "Marketplace Enrollment as a Share of the Potential Marketplace Population" [accessed on April 16, 2016]. Available at http://kff.org/health-reform/state-indicator/marketplace-plan-selections-as-a-share-of-the-potential-marketplace-population/
Krinn, K, P. Karaca-Mandic, and L. Blewett. 2015. "State-Based Marketplaces Using 'Clearinghouse' Plan Management Models Are Associated with Lower Premiums." Health Affairs 34 (1): 161-9.
Long, S., M. Karpman, G. Kenney, S. Zuckerman, D. Wissoker, A. Shartzer, N. Andersen, and K. Hempstead. 2015. Taking Slock: Gains in Health Insurance Coverage under the ACA as of March 2015. Washington, DC: Urban Institute [accessed on May 18, 2015]. Available at http://hrms.urban.org/briefs/Gains-in-Health-Insurance-Coverage-under-the-ACA-as-of-March-2015.html
Martin, L. T, N. Bharmal, J. C. Blanchard, M. Harvey, and M. V. Williams. 2014. Barriers to Enrollment in Health Coverage in Colorado. Santa Monica, CA: Rand Corporation [accessed on December 10, 2015]. Available at http://www.rand.org/content/dam/rand/pubs/research_reports/RR700/RR782/RAND_RR782.pdf
Ponce, N. A., K. J. Hoggart, M. Wilhelm, and B. Ritz. 2005. "Preterm Birth: The Interaction of Traffic-related Air Pollution with Economic Hardship in Los Angeles Neighborhoods." American Journal of Epidemiology 162 (2): 140-8.
U.S. Census Bureau. 2015. "2009-2013 5-Year American Community Survey." [accessed on November 15, 2015]. Available at http://factfinder.census.gov/bkmk/table/1.0/en/ACS/13_5YR/DP03/0500000US17031.14000
Volk, J., S. Corlette, S. Ahn, and T Brooks. 2014. Results from the First Year of Navigator Technical Assistance Project: Lessons Learned and Recommendations for the Next Year of Enrollment. Washington, DC: The Center for Health Insurance Reforms: Georgetown University Health Policy Institute [accessed on May 18, 2015]. Available at https://www.statereforum.org/system/files/navigatorreport_georgetownoct2014.pdf
Yen, I. H., and G. A. Kaplan. 1999. "Neighborhood Social Environment and Risk of Death: Multilevel Evidence from the Alameda County Study." American Journal of Epidemology 149 (10): 898-907.
Yousey-Hindes, K M., and J. Hadler. 2011. "Neighborhood Socioeconomic Status and Influenza Hospitalizations among Children: New Haven County, Connecticut, 2003-2010." American Journal of Public Health 101 (9): 1785-9.
Additional supporting information may be found in the online version of this article:
Appendix SA1: Author Matrix.
Table A1. State Policy Coding for OEP1 (All FFM States in OEP1).
Table A2. Comparison of OEP1 and OEP2 Beta Regression Models.
Table A3-1. Marginal Effects from PUMA-Level Data, OEP1.
Table A3-2. Marginal Effects from ZCTA-Level Data, OEP1.
Table A4. Marginal Effects Using Four Functional Form Assumptions.
Table A5. State-level QHP Ratio's versus KFF QHP Rates, OEP1.
Address correspondence to Michel Boudreaux, Ph.D., School of Public Health, University of Maryland, 4200 Valley Drive, #3310A, College Park, MD 20742; e-mail: email@example.com. Lynn A. Blewett, Ph.D., and Pinar Karaca-Mandic, Ph.D., are with the Division of Health Policy & Management, University of Minnesota, Minneapolis, MN. Brett Fried, M.S., is with the State Health Access Data Assistance Center, Minneapolis, MN. Katherin Hempstead, Ph.D., is with the Robert Wood Johnson Foundation, Princeton, NJ.
Table 1: State and Community Characteristics Associated with QHP Ratios, 33 Healthcare.gov States, First Open Enrollment Period Unstandardized Mean (SD) State policy characteristics Medicaid expansion state 22.3 Federal determination state 15.6 Partnership state 17.3 Demographic and social characteristics Rural ZCTA 17.3 Imputed plan selection 5.9 Percent age 35-64 39.4 (5.3) Percent female 51 (2.6) Percent Hispanic 14.3 (19.5) Percent white, non-Hispanic 65.5 (26.8) Percent married 49.3(11.1) Percent that does not speak English very well 6.9 (9) Percent with functional disability (nonelderly) 12.6 (4.7) Percent noncitizen 5.8 (6.8) Economic characteristics Percent income eligible for APTC 48.9 (10.7) Imputed ESI offer rate among private workers 86.1 (1.5) Percent of labor force unemployed (age 16+) 6.2 (2.6) Percent without HS diploma (age 25+) 14.1 (9.2) Health care market characteristics QHP premiums ($) 215.2 (36.5) PCSA-level primary care physicians per 100k 89.4 (57.2) PCSA-level health centers per 100k 4.0 (7.7) Marginal Effects from Beta Regression Using Standardized Means Est. SE P State policy characteristics Medicaid expansion state -4.40 (*) 0.72 0.000 Federal determination state -1.52 (*) 0.63 0.016 Partnership state -0.92 1.21 0.448 Demographic and social characteristics Rural ZCTA -0.12 0.42 0.777 Imputed plan selection -2.77 (*) 0.72 0.000 Percent age 35-64 3.70 (*) 0.33 0.000 Percent female 1.15 (*) 0.32 0.000 Percent Hispanic -0.72* 0.24 0.003 Percent white, non-Hispanic -0.55 0.29 0.060 Percent married -0.16 0.32 0.607 Percent that does not speak English very well 2.37 (*) 0.44 0.000 Percent with functional disability (nonelderly) -0.24 0.63 0.707 Percent noncitizen -0.36 0.31 0.247 Economic characteristics Percent income eligible for APTC 0.83 0.39 0.033 Imputed ESI offer rate among private workers 0.55 0.44 0.211 Percent of labor force unemployed (age 16+) 1.36 (*) 0.38 0.000 Percent without HS diploma (age 25+) -4.55 (*) 0.58 0.000 Health care market characteristics QHP premiums ($) 0.21 0.28 0.452 PCSA-level primary care physicians per 100k 0.36 (*) 0.08 0.000 PCSA-level health centers per 100k -0.01 0.23 0.981 Notes. QHP, qualified health plan. Characteristics apply to entire ZCTA, not the characteristics of the QHP ratio denominator. See text for a description of measures. Marginal effects are from a beta regression. SE is standard errors clustered on state (n = 22,322). (*) p < .05. Source. Office of the Assistant Secretary of Planning and Evaluation, 2009-2013 American Community Survey, and other sources as described in the text.