Draime 1 Did the Massachusetts Health Care Reform Lead to Smaller Firms and More Part-Time Work? By Alex Draime Professor Bill Evans ECON 43565 April 19, 2013 Abstract:: The Massachusetts health care reform of 2006 dramatically increased health insurance coverage rates statewide. The legislation required employers with 11 or more full-timeequivalent employees to supply health insurance to their workers or face a tax. While research has shown improvements in coverage, there has been little examination of the impact on firm behavior as a result of these new obligations. Using data from the March Current Population Survey from 2003 to 2012, I examine how the reform impacted firm size and par-time work. My results suggest that while the reform did not impact these variables in aggregate, certain demographics particularly low-skilled workers were less likely to work for small firms and more likely to work part-time as a result of the reform.
Draime 2 Leading the charge in health insurance reform, the state of Massachusetts enacted comprehensive legislation in April 2006 to provide virtually universal healthcare coverage to state residents. This legislation served as a template for the March 2010 Patient Protection and Affordable Care Act (the ACA ) and featured many of the same provisions as its Federal counterpart. The core of the Massachusetts legislation was an individual mandate designed to maximize coverage coupled with subsidies for low income residents to ensure affordability. As shown in Long et al. (2009) and other sources, the legislation has succeeded in raising coverage rates statewide. As part of its mechanism to increase coverage, the reform also created new obligations for employers. By July 1, 2007, employers with 11 or more full-time-equivalent employees had to choose to either supply health insurance for their employees or pay a fair share tax of up to $295 per year per employee not insured (Kaiser 2012). Additionally, employers were required to offer cafeteria plans that allow workers to purchase health insurance with pretax dollars or face a free rider tax if employees utilize the statewide Health Safety Net, an uncompensated care pool (Kaiser 2012). These regulations did not apply to firms with fewer than 11 full-timeequivalent employees. Firms could avoid providing health insurance and paying the fair share tax by downsizing or switching to part-time labor. With the forthcoming full implementation of the ACA and its similar employer requirements but much larger tax for noncompliance, it is relevant to consider this question as potential foreshadowing of firm behavior under the ACA. Using data from the 2003-2012 March Current Population Survey (CPS) for Massachusetts and the other New England states (New Hampshire, Vermont, Maine, Connecticut, and Rhode Island), I construct difference-in-difference (DD) estimates in an effort
Draime 3 to answer this important question. DD models, as used here, isolate the treatment effect of a shock like a policy change by comparing differences in averages over time between a control and treatment group over time and then taking the difference of those results (Woolridge 2013). Using this method, I confirm the efficacy of the reform in raising insurance coverage rates and find that while there are not statistically significant aggregate effects for firm size and part-time work, certain demographic groups experience especially low skilled workers qualitatively large and statistically significant increases in both part time work and employment in smaller firms. This paper will proceed as follows. Section 1 details the study design and data collection, while Section 2 provides confirmation on the general efficacy of the reform in increasing insurance rates in Massachusetts with reference to some existing literature along with an overview of the relevant dependent variables. Section 3 details experimental methodology and presents results. Section 4 concludes. I. Study Design and Data Collection A. Study Design The study design is a straightforward difference-in-difference (DD) model, exploiting the shock to Massachusetts caused by the 2006 health care reform. In this case, I compare the changes in outcomes in Massachusetts over time to the same changes for the rest of New England. This latter groups serves as a control group that provides an estimate of the secular changes in outcomes that would have occurred in Massachusetts over time had there been no reform. I utilize three distinct outcomes to examine the possible effect of reform on firm behavior. The first is whether individuals have health insurance coverage. The estimates from
Draime 4 this model should match previous estimates and provides a reality check that the data set and model are providing reasonable results. The second outcome measures the number of workers in small firms. Specifically, the outcome is a dummy variable that equals 1 if a working CPS respondent was employed by a firm with fewer than ten employees and it equals zero otherwise. The third outcome is a dummy variable that identifies whether the working CPS respondent was a part time worker defined as working less than 35 hours per week. In an attempt to further parse these results, the DD models are estimated for all workers as well as for specific demographic groups. B. Data I use data from the March CPS from 2003 to 2012 for nonelderly adults (ages 18-64). The CPS is a monthly US survey of roughly 60,000 civilian, non-institutionalized households conducted by the US Census Bureau and the Bureau of Labor Statistics which collects a wide array of demographic and labor force data. The March Annual Social and Economic Supplement features additional elements related to factors like work experience, benefits, insurance, and more. I source all of the March CPS data from the Integrated Public Use Microdata Series (IPUMS), a project sponsored by the University of Minnesota. The data set includes weightings that enable one to construct aggregate level statistics at the state or national level. To measure health insurance coverage, I use the IPUMS HCOVANY variable, which is the broadest available measure of insurance coverage. It indicates whether the respondents had any health insurance coverage public, private, subsidized, etc. during the previous year. Note that this means that it is possible that a respondent could be uninsured at the time of the survey
Draime 5 while having been insured at some point over the prior year. This variable is constructed by the State Health Assistance Database Center (SHADC) at the University of Minnesota and includes a number of modifications to the health insurance variables available in the raw CPS. For example, the SHADC edited pre-2005 data to reflect changes in the CPS related to the assignment of private health insurance coverage to non-policy holders. To measure whether a respondent was employed full-time or part-time, I use the IPUMS FULLPART variable. Full-time work is defined as 35 or more hours per week and the variable indicates whether the respondent worked full-time or part-time in the previous year. To measure firm size, I use the IPUMS FIRMSIZE variable, which indicates the total number of persons who worked for the respondent s primary employer during the previous year, including all locations where the employer operated. If the respondent is self-employed, the reported number is the number of the respondent s employees. Responses are grouped into ranges. For the purpose of this paper, I generate a dummy variable to indicate when firm size is less than 10 employees, non-inclusive. The IPUMS data is grouped as follows: under 10 employees, 10-24 employees, etc. Under the reform, the employer obligations go into effect once a firm has 11 employees, so the cutoff in the data set does not perfectly match the cutoff in the legislation. It is possible that this mismatch has a confounding effect on my DD results. II. Replication of Previous Results and a Discussion of Dependent Variables A. Replication of Previous Results Figure 1 plots the fraction of non-elderly adults with in insurance in Massachusetts and the rest of New England from my analysis sample. These estimates shows that from 2003 to
Draime 6 2006, the insurance rate in Massachusetts for non-elderly adults hovered around 88%, which matches the rate found in Long et al. (2009). From 2007 to 2008, the insurance rate jumped to 94% in Massachusetts, while the rate for the rest of New England remained at 88%. Table 2 reports the the results of a basic DD model specification that attempts to replicate previous estimates on the impact of the Massachusetts reform on insurance rates. The basic estimating equation is of the form (1) InsCov ist = β 0 + β 1 Treatment ist + X ist β + Ɵ s + λ t + μ ist InsCov is a dummy variable that indicates whether or not an individual was insured at any time in the year prior to the administration of the survey. The data varies across people (i), states (s) and years (t). Treatment is the interaction term for being in Massachusetts post 2006 and hence β 1 is the coefficient of interest. The variable Ɵ s represents state effects and λ t represents year effects. The variable X ist represents a vector of observed characteristics, which in this case consist of age, sex, education, marital status, and race/ethnicity. The variable μ ist is a random error term. In the first row of the table, I report results for the entire sample of non-elderly adults. In subsequent rows, I report results for important subgroups. Overall, the results of this DD model are unsurprising. At the aggregate level, the coefficient is a statistically significant 5.5 percentage points (interpreted as an increase in the insurance rate attributable to the reform). For comparison, Long et al. s (2009) DD model found that the uninsurance rate among non-elderly adults in Massachusetts drops by 6.6 percentage points in the first year under the plan. Long s
Draime 7 (2008) pre-post model showed a drop of 5.6 percentage points for the same sample. My results are in-line with these findings. Looking into demographic breakdowns, the effect is especially pronounced for groups that are generally more likely to be uninsured, including minorities, people under age 30, and people with lower levels of educational attainment. Note that all results are statistically significant at demanding levels. B. Discussion of Dependent Variables Figure 2 shows the percentage of working respondents that are employed by establishments with 10 or fewer employees for the treatment and comparison samples while Figure 3 shows the same for employees at establishments with 20 or fewer employees for reference. The percentage of people at firms with 10 or fewer employees falls from about 21% to 18% from 2007 to 2008 in Massachusetts. Given that the new obligations for firms became effective in 2007, one might expect to see a move in the opposite direction. This jump may be attributable to the most recent recession, which may have had the effect of shaking out some of the smaller firms. A similar, though less dramatic pattern is observed for the rest of New England. Figure 4 shows the percentage of the working population that works on a full-time basis and Figure 5 shows the percentage of the working population that works on a part-time basis. In both cases, the levels fluctuate in the aftermath of the implementation of the reform in 2006 and 2007. The percentage of the population working part-time spikes to roughly 27% in 2010 before returning to pre-reform levels. It is difficult to gauge the effect of the reform on either of these
Draime 8 metrics from the visuals of the graph. Note that the widest swings also coincide with the time period encompassing the nadir of the recent recession. Overall, the firm size data and the full-time and part-time data hint at something going on beneath the surface, but it is impossible to draw conclusions from a visual appraisal, as one could with the insurance coverage data. The DD estimates serve to demystify these trends to an extent. III. Methodology and Results A. Methodology To isolate the effect of the reform on firm size and part time work, I employ multivariate DD models. The two equations are as follows: (2) FirmSize ist = β 0 + β 1 Treatment ist + X ist β + Ɵ s + λ t + μ ist (3) PartTime ist = β 0 + β 1 Treatment ist + X ist β + Ɵ s + λ t + μ ist These DD equations follow the same form as the DD model used to generate Table 2. FirmSize is a dummy variable constructed from the IPUMS FIRMSIZE variable indicating whether or not an individual worked for a firm with fewer than 10 employees in the year prior to the administration of the survey. PartTime is a dummy variable constructed from the IPUMS FULLPART variable indicating whether or not an individual worked part-time or full-time the year prior to the administration of the survey. In both cases, the data varies across people (i), states (s) and years (t). In both cases, Treatment is the interaction term for being in Massachusetts post 2006 and hence β 1 is the coefficient of interest. The variable Ɵ s represents
Draime 9 state effects and λ t represents year effects. The variable X ist represents a vector of observed characteristics, which in this case consist of age, sex, education, marital status, and race/ethnicity. The variable μ ist is a random error term. B. Results Firm Size Table 2 shows the DD estimates for the effects of the reform on firm size. At an aggregate level for all non-elderly adults, the treatment effect is minuscule and statistically insignificant with a p-value of 0.173. In short, there is no broad-based effect on firm size due to the reform. This comports with the inconclusive nature of the visual data shown in Figure 5. However, the DD estimates for specific demographic groups yield some interesting results. Non- Hispanic Blacks exhibit a statistically significant treatment effect of 5.1 percentage points with a p-value of 0.018, indicating that this group saw employment in firms with 10 or fewer employees increase by about 5 percentage points as a result of the reform. The Hispanic population experienced the opposite result with a statistically significant treatment effect of -12.7 percentage points at a p-value of 0.000. The firm size results by educational attainment are also notable. While there are no statistically significant effects for those who report having attended or finished college, there are significant effects for those with lower levels of education. Individuals who completed high school show a statistically significant treatment effect of -2.9 percentage points at a p-value of 0.005. Those who failed to complete high school show a statistically significant treatment effect of -6.3 percentage points at a p-value of 0.003.
Draime 10 C. Results Part-Time Employment Levels Table 3 shows the DD estimates for the effects of the reform on part-time employment levels. As with the firm size estimates, the treatment effect at an aggregate level for all nonelderly adults for part-time employment levels is small and statistically insignificant with a p- value of 0.548. The DD estimates by demographic yield some interesting results. The Other, Non-Hispanic group shows a statistically significant treatment effect of 9.3 percentage points with a p-value of 0.000. This means that this group saw its part-time employment level increase by 9.3 percentage points as a result of the reform (said another way, the group s full-time employment level dropped by 9.3 percentage points). The results by educational attainment are similarly interesting. There are no statistically significant effects for individuals those who report having attended or finished college. But, those who achieved a lower level of education experienced an increase in part-time employment. Those who completed high school experienced a part-time employment rate increase of 1.6 percentage points at a p-value of 0.096. Those who failed to complete high school experienced a part-time employment rate increase of 4.0 percentage points at a p-value of 0.044. IV. Conclusion The Massachusetts health care reform certainly had the desired effect of increasing insurance rates for the state population. And, in aggregate, the new employer obligations appear to have had virtually no significant impact on firm size or part-time employment. However, as
Draime 11 my DD estimates show, statistically significant effects begin to emerge once the data is broken down into demographic groups. Impacts on individuals with lower levels of educational attainment are consistent across both the firm size and part-time employment DD models. In both cases, individuals who had at least some college experience were unaffected, while those with less education experienced significant effects. Individuals with less education found themselves less frequently working for firms with fewer than 10 employees. They also experienced an increase in part-time employment at the expense of full-time employment. One possible explanation is that instituting more onerous requirements on firms with more than 10 full-time equivalent employees forced employers to make decisions on the margin about the level of skill required from their workers to operate. The employer obligations impose a penalty on inefficiency the option of taking on additional employees as a means of improving performance becomes a more expensive proposition under the reform. The effect on the different racial/ethnic groups is a bit more difficult to parse. The fact that the results are inconsistent across the two DD models is somewhat puzzling. An explanation related to marginal decisions based on skill level would be reasonable, but coming to a definitive conclusion would require more targeted research on those groups. Based on the results from Massachusetts data, one can reasonably conclude that insurance coverage rates will improve nationally with the full implementation of the ACA. It is more difficult to draw conclusions on how the ACA will impact firm behavior, given that the ACA s employer obligation provisions feature a higher cutoff at 50 employees and a different penalty structure. Still, the Massachusetts data yields some interesting results which are worthy of additional research.
Draime 12
Draime 13 References Kaiser Family Foundation (2012). Massachusetts Health Care Reform: Six Years Later. Retrieved from http://www.kff.org/healthreform/upload/8311.pdf. King, M., Ruggles, S., Alexander, J., Flood, S., Genadek, K., Schroeder, M., Trampe, B., & Vick, R. (2010). Integrated Public Use Microdata Series, Current Population Survey: Version 3.0. Minneapolis, MN: Minnesota Population Center. Retrieved from http://cps.ipums.org Long, S. (2008). On the Road to Universal Coverage: Impacts of Reform in Massachusetts at One Year. Health Affairs, 27(4), 270-284. Retrieved from http://content.healthaffairs.org/content/27/4/w270.full.pdf+html Long, S., Stockley, K., & Yemane, A. (2009). Another Look at the Impacts of Health Reform in Massachusetts: Evidence Using New Data and a Stronger Model. American Economic Review: Papers & Proceedings, 99(2), 508-511. Retrieved from https://www3.nd.edu/~wevans1/class_papers/long_stockley_aea.pdf Wooldridge, J. M. (2012). Introductory Econometrics, a Modern Approach. South-Western Pub.
Percentage of Firms Percentage of Firms Percent Insured Draime 14 Figures and Tables 98 96 94 92 90 88 86 84 82 80 FIGURE 1 - PERCENT NON-ELDERLY ADULT POPULATION INSURED 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Massachusetts Rest of NE 23 22 21 20 19 18 17 16 15 FIGURE 2: PERCENTAGE OF FIRMS WITH 10 OR FEWER EMPLOYEES 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Massachusetts Rest of NE 18 16 14 12 10 8 6 4 FIGURE 3: PERCENTAGE OF FIRMS WITH 20 OR FEWER EMPLOYEES 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Massachusetts Rest of NE
Percentage Part Time Percent Full Time Draime 15 FIGURE 4: PERCENTAGE WORKING POPULATION - FULL-TIME 79 78 77 76 75 74 73 72 71 70 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Massachusetts Rest of NE 28 27 26 25 24 23 22 21 20 FIGURE 5: PERCENTAGE WORKING POPULATION - PART-TIME 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Massachusetts Rest of NE
Draime 16 TABLE 1: DIFFERENCE-IN-DIFFERENCE ESTIMATES, TREATMENT EFFECT OF MA REFORM ON INSURANCE COVERAGE, BY DEMOGRAPHIC CHARACTERISTICS n Treatment Effect p-values All Adults 18-64 94,521 0.0547 0.000 (0.0061) By Race/Ethnicity White, Non-Hispanic 81,079 0.0405 0.000 (0.0064) Black, Non-Hispanic 3,267 0.1087 0.002 (0.0356) Other, Non-Hispanic 4,094 0.0754 0.007 (0.0278) Hispanic 6,081 0.1786 0.000 (0.0310) By Sex Male 47,883 0.0644 0.000 (0.0094) Female 46,638 0.0437 0.000 (0.0077) By Marital Status Married 57,348 0.0240 0.000 (0.0051) Unmarried 37,173 0.0910 0.000 (0.0120) By Age Group Under 30 19,320 0.1100 0.000 (0.0166) Over 30 75,201 0.0376 0.000 (0.0062) By Educational Attainment High School 27,145 0.0710 0.000 (0.0142) Some College 16,669 0.0541 0.000 (0.0152) College 44,078 0.0315 0.000 (0.0068) Less than High School 6,629 0.1751 0.000 (0.0320)
Draime 17 TABLE 2: DIFFERENCE-IN-DIFFERENCE ESTIMATES, TREATMENT EFFECT OF MA REFORM ON PERCENTAGE OF FIRMS WITH 10 OR FEWER EMPLOYEES, BY DEMOGRAPHIC CHARACTERISTICS OF SURVEY RESPONDENTS n Treatment Effect p-values All Adults 18-64 105,334 (0.0073) 0.173 (0.0054) By Race/Ethnicity White, Non-Hispanic 89,816 (0.0021) 0.722 (0.0059) Black, Non-Hispanic 3,903 0.0509 0.018 (0.0216) Other, Non-Hispanic 4,672 0.0116 0.641 (0.0248) Hispanic 6,943 (0.1272) 0.000 (0.0213) By Sex Male 53,427 (0.0185) 0.018 (0.0079) Female 51,907 0.0047 0.517 (0.0073) By Marital Status Married 63,359 (0.0020) 0.781 (0.0071) Unmarried 41,975 (0.0131) 0.113 (0.0083) By Age Group Under 30 21,860 (0.0202) 0.066 (0.0110) Over 30 83,474 (0.0032) 0.600 (0.0062) By Educational Attainment High School 30,622 (0.0290) 0.005 (0.0104) Some College 18,645 0.0153 0.233 (0.0128) College 48,495 0.0061 0.423 (0.0076) Less than High School 7,572 (0.0627) 0.003 (0.0213)
Draime 18 TABLE 3: DIFFERENCE-IN-DIFFERENCE ESTIMATES, TREATMENT EFFECT OF MA REFORM ON PART-TIME EMPLOYMENT, BY DEMOGRAPHIC CHARACTERISTICS OF SURVEY RESPONDENTS n Treatment Effect p-values All Adults 18-64 105,334 0.0031 0.548 (0.0052) By Race/Ethnicity White, Non-Hispanic 89,816 (0.0011) 0.840 (0.0057) Black, Non-Hispanic 3,903 (0.0394) 0.140 (0.0267) Other, Non-Hispanic 4,672 0.0927 0.000 (0.0258) Hispanic 6,943 0.0133 0.523 (0.0208) By Sex Male 53,427 0.0049 0.406 (0.0059) Female 51,907 0.0024 0.785 (0.0087) By Marital Status Married 63,359 (0.0040) 0.518 (0.0062) Unmarried 41,975 0.0039 0.661 (0.0088) By Age Group Under 30 21,860 0.0126 0.346 (0.0133) Over 30 83,474 (0.0037) 0.491 (0.0083) By Educational Attainment High School 30,622 0.0158 0.096 (0.0095) Some College 18,645 (0.0129) 0.352 (0.0138) College 48,495 (0.0039) 0.602 (0.0073) Less than High School 7,572 0.0402 0.044 (0.0204)