OUT OF POCKET COSTS AND HEALTH INSURANCE TAKE UP RATES Euclid Avenue, RT South Broadway St.

Transcription

1 1 OUT OF POCKET COSTS AND HEALTH INSURANCE TAKE UP RATES Vasilios D. Kosteas Francesco Renna Associate Professor Associate Professor Department of Economics Department of Economics Cleveland State University The University of Akron Euclid Avenue, RT South Broadway St. Cleveland, OH Akron, OH Abstract: About half of all workers covered by employer sponsored health plans in the U.S. have the option to choose between multiple plans. In this paper we present a simple model examining the impact of workers out-of-pocket contributions on take-up rates. We then estimate the individual plan and total firm take up rates using data from the Employer Health Benefits Survey. Our fixed effects estimations suggest that workers respond to an increase in the out-of-pocket contributions for HMO plans by switching to PPO plans without impacting the overall take-up rate, while workers respond to increases in the out-of-pocket contribution for PPO plans by dropping out of the group coverage. In general we conclude that, even if the estimated elasticities are small, the increase in the cost of coverage for workers (represented by the changes in the out-of-pocket contribution as well as the change in deductible, coinsurance and copayment rates) are large enough to the entire drop in coverage during the period under analysis.

2 2 1. Introduction Over the first ten years of this century, the share of the US population covered by employer sponsored health insurance plans experienced a significant decline. In 2010, 58.6 percent of Americans under the age of 65 received health insurance coverage through an employer sponsored plan. This rate was 10.6 percentage points higher just a decade earlier. 1 This dramatic drop has been at the center of a vivid political debate which led to the passing of the Affordable Care Act in There are three possible explanations for this drop: (1) a decline in the rate of firms offering health insurance to their employees; (2) a decrease in eligibility rates for ESHI plans; or (3) a decrease in the take-up rate. The third explanation can account for about a quarter of the total decline in coverage (Clemans-Cope and Garrett, 2006). The increasing share of the premium that is paid by workers is often used to explain this decline in the take up rate (Cutler, 2003). However, the decision to seek health coverage is not only a function of the worker s contribution needed to participate in the ESHI, but also of the fees for using the insurance policy (Jacob 2009). In recent years the increase in copayments, deductible and coinsurance rate has far outpaced the increase in worker contribution ( In this study we analyze the impact of out-of-pocket (OOP) costs on the take-up rate for firms that offer multiple plan types. OOP costs consist of both worker s contribution and expected expenditure. While only 15 percent of all U.S. firms make more than one plan type available to their employees, these firms are much larger in size compared to those that offer only one plan, thus about half of all covered workers in the U.S. have the option to choose 1

3 3 between multiple plans. 2 These are also firms that are the least likely to drop health insurance from their compensation package. 3 As such, the decrease in coverage among this group of workers is almost completely explained by a drop in the take-up rate, making the estimation of the take up elasticity for firms who offer multiple plans particularly important. We found that the increase in the worker s contribution explains only half of the decline in the take-up during the time under analysis. The remaining part of the decline is due to the increase in coinsurance, copayments and deductibles. The present paper contributes to the literature in two ways. This is the first paper to jointly estimate the overall take up rate elasticity as well the individual plan take-up elasticites while controlling for expected expenditure. Second, we analyze plan switching behavior using a nationally representative dataset with more recent data. Existing studies either analyze data from a single or small number of employers (Cutler and Reber, 1996; Royalty and Solomon, 1999; Buchmueller, 2000; Strombon et al, 2002; Buchmueller and Feldstein, 1997) or data on retirees in the medicare system (Dowd et al 2003). One notable exception is Short and Taylor (1989) who analyze data from the National Medical Care Expenditure Survey. However, their data is from Indeed, existing studies generally use data from the 1980s or the 1990s. The significant increase in health insurance premiums as well as deductible, copayments and coinsurance experienced in the 2000s, raises the question of whether the relationships between OOP premiums and both take-up rates and plan choice have changed since that time period. In the following section we briefly discuss the existing literature. In section 3 we develop a simple model where individual plan take-up and overall plan take-up rates are linked to each health insurance plan s OOP costs. Next, we describe the data used for the estimation and in 2 See Exhibit 4.1 and 4.2 in 3 Kosteas and Renna (2014) show that large firms did not change their offer rate during the period under analysis. The drop in the offering rate was driven entirely by small firms, that typically offer only one plan to their employees

4 4 section 5 we explain the estimation strategies adopted in this paper. In section 6 we illustrate the results of our alternative estimations and then conclude with some final remarks. 2. Literature Review Typically employers require an OOP contribution from workers that want to participate in the group plan. Kosteas and Renna (2014) show that when premiums grow faster than wages, not only do employers increase the employee contribution required to participate in the firm plan, but premium increases in one type of plan (i.e. PPO) affect the OOP contribution for other plan types (i.e. HMO). In fact, the average annual employee contribution for individual (family) coverage grew from $318 ($2,196) in 1999 to $1,081 ($6,025) in How workers respond to increases in these contributions depends on the elasticity of the demand for coverage with respect to the OOP contribution. A number of studies have looked at the elasticity of demand with respect to the OOP contribution using a variety of data sets and estimation technique (Jacobs, 2009; Chernew et al, 1997; Gruber and Washington, 2005). Overwhelmingly all studies suggest the demand for coverage is quite inelastic, with estimates ranging from to Although these estimates are small, Cutler (2003) shows that the increase in the OOP contributions was large enough to explain the entire drop in take-up rates. Jacobs (2009), instead, finds that increases in worker contributions toward the premium explain about 60 percent of the decline in the take-up rate for ESHI and increases in insurance cost-sharing explain an additional 10 percent of that change. One empirical question raised by this literature is which OOP contribution to use when computing the elasticity of coverage for firms that offer multiple plans. Some studies have used a firm s weighted average premium (Jacobs, 2009; Polsky et al, 2005) while others have used the

5 5 lowest premium (Cutler, 2003; Jacobs, 2009; Polsky et al, 2005). Both approaches have some appeals and drawbacks as well. The least expensive plan is a fair measure of the opportunity cost of coverage only if the cheap plan is always a substitute for the better plan. However, this is not likely to be the case when employees face significant switching costs (Strombom et al, 2002). The average premium forces the price elasticity of demand to be the same among all plans. In both cases, the estimated elasticities may be biased and cannot predict the change in consumer behavior in response to the introduction of new plans available to them, or plan switching as the relative cost of the available plans changes. Another line of research has looked at individuals switching behavior between plans in response to changes in the relative OOP premiums. The previous literature has found that these switching elasticities are still in absolute value less than 1, but their magnitude seems larger than the estimates in studies looking at the take-up elasticity. For example Royalty and Solomon (1999) estimated switching elasticities from to depending on the type of plan. Dowd et al (2003), using an older sample of workers, find a smaller elasticity (-0.13). After controlling for differences in plan characteristics, Short and Taylor (1989) find that a $100 increase in the difference between the HMO and traditional plan premiums leads to a drop of 2.6% in the enrollment in the HMO plans. Welch (1986) finds that the demand for HMO coverage (compared to conventional plans) is more elastic in the long run: a 1 percent increase in the HMO premium leads to a 0.2 percent decrease in the HMO take-up rate in the short run but 0.62 percent in the long run. Similarly, a 1 percent increase in the premium of a conventional plan increases the HMO take-up rate by 0.16 percent in the short run and 0.49 percent in the long run. By combining the results of the elasticity to take-up rate with the results from the literature

6 6 on the switching between plans, we hypotheses that workers are more likely to switch between plan than to drop out of the coverage all together in response to a change in the OOP premiums. 3. Model One of the first issues to consider when modeling the demand for health insurance coverage is whether to use the full price of the health insurance plan or the out-of-pocket contribution required by firms in order for a worker to enroll in the group coverage. The former gives us a measure from the insurer perspective of the price elasticity of health insurance coverage, while the latter gives us a measure from the enrollees perspective of price elasticity of health insurance coverage. This distinction is not trivial as the estimated elasticities are very sensitive to which measure of health insurance cost is used. The employee-perspective elasticities are always less than 1, while the insurance-perspective elasticities are between and (Royalty and Solomon, 1999) with some estimates as high as (Atherly, Dowd, and Feldman, 2004). Since we set our model as the decision problem of an individual, we focus on the OOP contributions. Consider a firm that offers two health plans to its workers: a Preferred Provider Organization (PPO) and a Health Maintenance Organization (HMO) plan. Consumer i s utility function from consuming health plan j depends on the out-of-pocket costs (OOP j ), other plan characteristics (X j ) and an i.i.d. error term that is assumed to follow a type 1 extreme value distribution (ε ij ): UU iiii = αα jj XX jj + ββ jj OOOOOO jj + εε iiii where j = PPO, HMO (1) For now we do not include the option of foregoing coverage. Given a choice set of J plans, consumers choose a health plan that yields the highest utility. Starting from this random utility

7 7 model, one can derive the demand for plan j as the share of plan j (SS jj ) at the firm level relative to other alternatives: SS jj = eeeeee αα jj XX jj +ββ jj OOOOOO jj eeeeee αα jj =PPPPPP, HHHHHH jj XX jj +ββ jj OOOOOO jj The firm overall take-up rate is the sum of the two plan s take-up rates: tttttttttttt oooooooooooooo = SS HHHHHH tttttttttttt oooooooooooooo +SS PPPPPP tttttttttttt oooooooooooooo (3) = tttttttttttt HHHHHH + tttttttttttt PPPPPP Since each plan s share is a function of its own OOP cost as well as the cross OOP cost, the overall take-up rate is also a function of both OOP costs. As described above, the previous literature focused either on the individual plan shares or on the overall take-up rate, ignoring cross-premium effects in the former and focusing on the average or minimum premium in the latter. There is strong evidence that workers do respond to changes in the relative premiums by switching from one plan to another (see for example Buchmueller, 2000; Strombon et al, 2002; Buchmueller and Feldstein, 1997). If this change in the relative premium affected only the distribution of the workers among the plans available but not the overall take-up rate, then using only one summary variable for both premiums (such as a weighted average or the minimum premium) would be an acceptable approach. However, if the two plans are not perfect substitutes and if the increase in the premium of one plan would induce some workers to forego coverage altogether, then it is important to include both OOP premiums in the model. Our hypothesis is that, an increase in the OOP c will result in a reallocation of workers across plans as well as a decrease in the overall take-up rate. (2) 5. Methodology We start by specifying the take-up rate of firm j at time t for each plan as a function of the own and cross OOP costs (worker s contribution and the expected expenditures) for each plan which

8 8 is a function of key plan characteristics. Typically a firm will require a different contribution depending on whether the worker is seeking individual coverage or family coverage. Admittedly, the two rates are highly correlated over time, hence it does not matter much which one to use. The main estimations presented in the study use the family coverage premiums but we present a robustness check in which we use the single coverage worker s contribution rate. The OOP expenditures variables serve as measures of plan quality by incorporating information on deductibles, co-payment and coinsurance rates and are constructed in a fashion similar to Jacobs (2009). From the Medical Expenditures Panel Survey Household Component Summary Tables (MEPS 2015) we gather information on median expenditures for prescription drugs (MEPS 2015, Table 2), office visits (MEPS 2015, Table 8), inpatient hospital visits (MEPS 2015, Table 5) and total expenditures (MEPS 2015, Table 1) for each year in the sample. Using this information, we calculate expenditure shares for each of the three categories of health care expenditures. Next, we gather information on the average number of drug prescriptions from the Kaiser Family Foundation (12.5 prescriptions per person in 2014) and information on office visits (3.3 per person in 2010) and hospital visits (0.11 impatient visits per person in 2010) from the Centers for Disease Control National Ambulatory Medical Care Survey: 2010 Summary Tables. We then compute the average expected OOP expenditures for an individual for each HMO and PPO plan using the following formula: OOP EEEEEEEEEEEEEE = dddddddddddddddddddd + (eeeeeeeeeeeeee_hoooooo dddddddddddddddddddd eeeeeeeehaaaaaa_hoooooo hoooooooooooooooo 0.11) hoooooooooooooooo + hoooooooooooooooo (eeeeeeeeeeeeee_oooooooooooo dddddddddddddddddddd eeeeeeeehaaaaaa_oooooooooooo oooooooooooooooooooooo 3.3) oooooooooooooooooooooo + oooooooooooooooooooooo

9 (eeeeeeeeeeeeee_dddddddddd dddddddddddddddddddd eeeeeeeehaaaaaa_dddddddddd gggggggggggggggggggggggg 12.5) gggggggggggggggggggggggg + gggggggggggggggggggggggg 12 (4) where expends_hosp indicates median expenditures for hospital visits, expshare_hosp is the median annual expenditure on hospital visits, hospcopay is the copay for hospital visits, hospcoins is the coinsurance rate for hospital visits, and so on. We estimate the OOP expenditures for both plan types for family coverage. In order to do this, we assume a family of three where each covered person is subject to the individual deductible. 4 Between 2005 and 2012 expected expenditure associated with HMO plans increased by $1,355 while expected expenditure associated with PPO plans increased by $3,022. We recognize that the demand for coverage may be affected by the composition of the workers employed at the firm. Therefore we include a set of variables (Y) that capture firm specific characteristics such as firm size measured by the level of employment, the fraction of employees who are making less than $25,000 per year, and indicator variables for urban location, union membership, regional location (Northeast, Midwest, and West with South serving as the comparison group), whether the firm makes the health insurance plans available to part-time and temporary workers, a series of industry dummy variables, and indicator variables for whether the firm offers conventional or point-of-sale plans. tttttttttttt HHHHHH jj,tt = αα 0 + αα 1 OOPPPPPPPPP HHHHHH jj,tt + αα 2 OOPPPPPPPPP PPPPPP jj,tt + γγ 1 OOPEEEEEEEEEEEEEE HHHHHH jj,tt + γγ 2 OOPEEEEEEEEEEEEEE PPPPPP HHHHHH jj,tt + δδyy jj,tt + εε jj,tt (5) 4 The family plan OOP expenditures and single plan OOP expenditures are very highly correlated. Substituting the single plan expenditures yields highly similar results.

10 10 tttttttttttt PPPPPP jj,tt = ββ 0 + ββ 1 OOPPPPPPPPP HHHHHH jj,tt + ββ 2 OOPPPPPPPPP PPPPPP jj,tt + θθ 1 OOPEEEEEEEEEEEEEE HHHHHH jj,tt + θθ 2 OOPEEEEEEEEEEEEEE PPPPPP PPPPPP jj,tt + μμyy jj,tt + εε jj,tt (6) Then, following equation (3) we define the overall take-up rate as tttttttttttt oooooooooooooo jj,tt ππ 0 + ππ 1 OOPPPPPPPPP HHHHHH jj,tt + ππ 2 OOPPPPPPPPP PPPPPP jj,tt + φφ 1 OOPEEEEEEEEEEEEEE HHHHHH jj,tt + φφ 2 OOPEEEEEEEEEEEEEE PPPPPP oooooooooooooo jj,tt + ρρyy jj,tt + εε jj,tt (7) We estimate equation (7) with two different measures of the overall take-up rate. The first is the true overall take-up rate for health insurance in the firm. The second is the combined take-up rate for HMO and PPO plans. In the former case, the effect of an increase in the OOP premium for the HMO plan on the take-up rate is the combined effect on the take-up rate for each plan type, including conventional, point-of-service (POS) and high-deductible plans (HDHP). However, not all firms offering both HMO and PPO plans offer each of these types of plans. Thus, focusing on the latter measure (the combined HMO and PPO take-up rate) provides a better fit to the theoretical model. Alternatively, we could restrict the sample to include only firms not offering conventional, POS or HDHP plans. In those models, ππ 1 = αα 1 + ββ 1 and ππ 2 = ββ 2 + αα 2. However, restricting the model in this fashion leads to a loss of more than half of the sample, raising serious concerns about introducing selection bias and the applicability of the findings to the broader set of firms offering both HMO and PPO plans. Because workers can purchase only one of the insurance plans, we expect αα 1 < 0 and ββ 2 < 0, thus implying that a change in the OOP premium would lead some workers to switch health plans and others to drop out of coverage. Also, we expect to find the cross contribution effect (αα 2 and ββ 1 ) to be positive but smaller in absolute value than the own premium contribution

11 11 effect. Thus, both ππ 1 and ππ 2 should carry a negative sign. This simple model explains why we should expect that a change in the OOP premium would have a larger impact on the own take-up rate than for the overall take-up rate as found in the previous literature. Even with firm characteristics included in the model, there is still a significant potential for bias due to worker sorting across firms. Specifically, workers who place a high value on health insurance may sort into firms offering generous plans (making the expected expenditures variables endogenous) and the employers may try to satisfy these employees by requiring a lower OOP contribution (making the OOP premium endogenous). In this case, estimating the models via OLS will overstate the responsiveness of take-up rates to OOP premium contributions. To deal with this issue, we take advantage of the fact that we have multiple observations per firm, and run a fixed effects model excluding all time invariant variables. While this approach will not eliminate time varying unobservable factors that affect the demand for health insurance, these factors are likely to be stable over very short periods of time (Jacobs 2009). Most of the firms only appear in our sample for a few years, a time frame over which it is unlikely firms worker composition will change significantly. Thus, controlling for firm fixed effects should eliminate the bias introduced by unobserved demand factors. However, FE estimation may exacerbate the effect of measurement error resulting from our inability to ensure the HMO and PPO plans for which we have information are offered to the same employees. Thus, the FE estimates may be viewed as a lower bound for the responsiveness estimates. Given the information on health plan characteristics and take-up rates is provided by firm HR managers with access to firm records, we expect this to be a less significant issue than with individual level data where the individuals are providing this information. However, for firms with multiple establishments, it is possible that the largest plans in each category are offered to employees at different locations. To the

12 12 extent this occurs, it should push the coefficients on the cross-plan variables toward zero since there is no reason to believe premium prices at one location will have an impact on take-up rates at another location. As a robustness check, we estimate each model restricting the sample to include only single establishment firms, eliminating the measurement error that might arise from the PPO and HMO plans being offered to workers at different establishments. 4. Data The estimation employs the waves of the Employer Health Benefits Survey (EHBS) collected by the Kaiser Family Foundation and the Health Research and Educational Trust (Kaiser/HRET). The sample consists of about 1,800-2,000 firms per year with at least three employees. Both private and government firms are represented (except for federal government). Kaiser/HRET attempted to repeat interviews with the prior year s survey respondents. As a result, about 75 percent of the sample participated in at least two survey years. The information contained in the surveys was obtained through interviews with each firm s benefits manager or human resources manager. Employers that offered health insurance were asked to provide the total take-up rate as well as the take-up rate for each of the following plans: conventional plans, health maintenance organization (HMO) plans, preferred provider organization (PPO) plans, point of service (POS) plans and high-deductible health plans (HDHP) linked to either a health retirement or health savings account. If the firm offers any plans in a particular category, then information is gathered on the largest plan in that category. The estimation in this paper focuses only on firms that make both HMO and PPO plans available to their workers. We have a total of 4,233 firm-year observations for firms offering two or more plans. Among firms offering multiple plans, HMO and PPO plans are the most popular choices, covering about 72 percent of our subsample. When we restrict the sample only to firms

13 13 that offer both of these types of plans, we are left with 3,449 firm-year observations. Figure 1 shows the average worker contribution for family coverage toward either an HMO or a PPO plan for the firms offering both plan types. All monetary values are deflated using the CPI and presented in 2005 dollars. Both rates increase over the period, with a small slow-down in 2008 probably due to the economic recession. As expected, the worker contribution toward a PPO plan is much higher than a contribution toward an HMO plan. The average monthly contribution toward the HMO plans increased by $ (a 45.2 percent increase compared to the contribution required in 2005) and the average monthly contribution toward the PPO plans increased almost $ (a 39.3 percent increase compared to the contribution required in 2005). In figure 2 we show the average annual take up rate at the firm level, as well as the break down by plan type. Between 2005 and 2012, the overall take-up rate decreased by approximately 3 percentage points. By contrast, the offer rate remained steady at 91 percent. When looking at the plan specific take-up rates, we find the take-up rate for HMO plans declines by 6 percentage points, while the PPO plan take-up declined only by 2.5 percentage points. Note that the sum of the HMO and PPO plans do not necessarily add up to the overall take up rate because during this period firms also started offering high deductible plans. In fact, while HMO and PPO plans represent 100% of the take-up rate at the beginning of the period, they represent only 93% of the sample in Table 1 presents summary statistics for the key firms characteristics separately for the entire sample of firms that offer both HMO and PPO plans and the sample restricted to single establishment firms. For the full sample, the overall take-up rate is 84.26; HMO and PPO plans account for 93.5 percent of the overall coverage rate. The take-up rate for HMO plans is slightly lower than the take-up rate for PPO plans. The average monthly worker contribution for family

14 14 coverage plans when both HMO and PPO plans are available are $ and $ for HMO and PPO plans, respectively. The contribution schedule and the deductibles on both plans are on average higher when other plan types are available. Expected OOP costs (excluding premium contributions) are much higher for PPO plans than for HMO plans. The PPO plans in the sample have much higher deductibles on average as well as higher coinsurance rates for office visits and prescription drugs. By contrast, the sample of single establishment firms has a slightly higher overall take-up rate and a higher share of workers enrolled in HMO plans. There is also a larger average difference between worker contributions to (and premiums for) HMO and PPO plans. 6. Results Before turning to our main results, we present estimates for the overall take-up rate model using only one OOP premium even for firms offering multiple plans in the latter case we choose the lowest OOP contribution (table 2). This exercise allows us to directly compare our results with those obtained by Jacobs (2009). Results are presented for the full sample of firms as well as the sample used in all subsequent analyses (only firms offering both HMO and PPO plan types). Based on the OLS estimates, we find a $10 increase in the lowest OOP premium for a family plan reduces the take-up rate by roughly 0.13 percentage points. Using the mean take-up rate for this sample (81.95) and minimum OOP premium ($317.13) for this sample yields an implied elasticity of -0.05, which is slightly smaller than the OLS estimates obtained by Jacobs (2009) who reports an elasticity in the range of and -0.09, but generally comparable 5. Given the significant increase in OOP premiums over the twelve years from the start of Jacobs (2009) sample to the end of ours, it is not unreasonable to expect workers may have become more 5 Using the minimum OOP premium for a single plan, the implied elasticity is -0.13, which is slightly larger than the estimates in Jacobs (2009) but still comparable.

15 15 sensitive to changes in the OOP premium. The OLS results are similar when restricting the sample to firms offering both HMO and PPO plans. However, since the means are different for this subsample, the OLS estimates for the sample including only firms offering both HMO and PPO plans imply an elasticity of As expected, the FE estimates show a much smaller change in take-up rates for an increase in the minimum premium for the full sample. In Table 3 we report the results of the take-up rates for each plan using the sample of firms offering both HMO and PPO plans, including firms offering other plan types (i.e. POS and conventional plans). When the firm offers only HMO and PPO plans, the menu choices to the consumer is completely defined by equations (5-6). When a firm offers more than just HMO and PPO plans, we should include a take-up rate equation for additional plans offered. Moreover the additional plan information should be included in the HMO and PPO equation. However this information would be unavailable for firms that do not offer additional plans, thus including these additional variables would seriously affect the sample size of our estimation. Hence we do not control for the characteristics of plans besides HMO and PPO. Since HMO and PPO plans cover 93 percent of all insured workers in the sample, we do not expect this omission to seriously affect our estimation. As evidence to that effect, we note our results are highly similar whether or not we include the indicator variables for whether the firm offers a conventional or POS plan. In each table, we include a second set of OLS and FE estimates (FE 2) which only includes single establishment firms. As expected, in the HMO take-up rate estimation we find that the own price effect is negative while the cross price effect is positive. In the pooled OLS regression, an increase of the worker contribution toward an HMO plan by $10 per month decreases the HMO take-up rate by 0.35 percentage points, while a $10 increase in the PPO OOP premium increases the HMO take-

16 16 up rate by 0.32 percentage points. These estimates imply an own-price elasticity of the HMO take-up rate of and a cross-price elasticity of We observe some significant crossexpenditure effects as well: Higher expected expenditures for the PPO plan increase the HMO take-up rate, suggesting that as the quality of PPO decreases, people switch to HMO plans. Restricting the sample to single-plant establishments has a minimal impact on the key coefficient estimates, except for the expected OOP expenditures for the PPO plan, suggesting that the potential measurement error is not likely to be a big concern for the OLS estimates. Both fixed effects estimations show a much weaker relationship between OOP premiums and HMO take-up and neither coefficient is statistically significant. In this case, the implied own-price elasticity for the HMO take-up rate is while the cross-price elasticity is While the coefficients on the expected OOP expenditures variables have the expected signs, neither is statistically significant. Similarly, for the PPO take-up rate estimation we found a negative own-price effect and positive cross-price effect. The pooled OLS estimation indicates an increase in the worker contribution toward a PPO (HMO) plan by $10 per month decreases (increases) the PPO take-up rate by 0.32 (0.25) percentage points. These estimates imply an own-price elasticity of and a cross-price elasticity of 0.2 for the PPO take-up rate. Again we also find some evidence that the quality of the two plans affect the take-up rates. A $100 increase in the expected expenditures for the PPO plan reduces its take-up by 1.08 percentage points. Expected expenditures for the HMO plan are negatively correlated with the PPO take-up rate, however the coefficient is not statistically significant. The implied elasticity of PPO take-up with respect to expected OOP expenditures for the PPO plan is As with the estimates for HMO take-up, the FE estimates for the full sample (FE 1) of firms offering both HMO and PPO plans indicate the

17 17 HMO worker contribution does not seem to affect the take-up rate, while an increase in PPO worker contribution continues to show a negative and statistically significant impact on the PPO take-up. Restricting the sample to single establishment firms improves the coefficient estimates and their degree of significance; the implied own-price (cross-price) elasticity is (0.095). Table 4 reports the results of the estimation for the overall take up rate. In the case of firms offering only HMO and PPO this is simply the sum of the HMO take up rate plus the PPO take up rate. As such, the effect of the worker contribution schedule on the overall take-up rate should be given by the relationships ππ 1 = αα 1 + ββ 2 and ππ 2 = ββ 1 + αα 2. However, given that our sample includes firms which also offer other plan types, these equations do not hold for the overall take-up rate (although the estimates almost satisfy the restrictions above). The OLS estimates for the restricted sample show that increases in both OOP premiums reduce take-up, but take-up is more sensitive to the HMO worker s contribution. A $10 increase in the HMO (PPO) worker s contribution reduces the overall take-up rate by 0.06 (0.04) percentage points, with implied elasticities of for the HMO contribution and for the PPO contribution. Consistent with prior research, the magnitude of these estimates show that employee enrollment in ESHI plans is not very sensitive to OOP premiums. Even if the estimated effects are small, the change in average OOP premiums ($ for HMO plans and $ for PPO plans) were large enough to explain a bit more than half of the entire decline in the take up rate. In fact, the aggregate take-up rate in the sample declined by 2.26 percentage points between 2005 and 2012, while the coefficient estimates based on OLS estimation predict a decline in the average take-up rate of approximately 1.25 percentage points. Additionally, a $100 increase in expected out of pocket costs for a family of three covered under a PPO plan lowers the take-up rate by percentage points. Using these estimates, the average increase in OOP expenditures under PPO

18 18 plans (approximately $3,022) predicts a 0.79 percentage point decrease in the take-up rate. Combined, the increase in OOP premiums and expected PPO expenditure can explain 90 percent of the decline in the overall take-up rate. If the coefficient on expected HMO expenditure were to be significant, it would explain an additional 0.49 percentage point decline in the take-up rate. The fixed effects model estimated on the sample of single establishment firms shows a statistically significant link between the take-up rate and OOP premiums for the PPO plan only, with an implied elasticity of According to this model, the average increase in OOP premiums for family PPO plans alone can account for a 0.53 percentage point decline in the overall take-up rate, which is nearly one quarter of the decline in the average take-up rate. As expected, the results for the take-up rate defined as the sum of HMO and PPO take-up rates are similar to those for the overall take-up rate, with somewhat stronger relationships between key variables. The implied elasticity of the HMO+PPO take-up rate with respect to PPO workers contributions for family coverage is Here again we notice the HMO workers contributions for family coverage does not significantly affect overall take-up. This result together with the evidence presented in Table 3 seem to suggest that as HMO plans become more onerous for workers, they switch to PPO plans, while as burden of the cost of PPO for family coverage increases, workers either drop out of coverage all together or switch to the HMO plan. Given the generally small magnitudes of the estimated impact of rising OOP premiums on take-up rates, we turn out attention to estimating the effect of rising premiums on the share of enrolled workers who have chosen the HMO plan. This estimation will give us a measure of switching between plans in response to increases in the OOP premiums. The OLS estimates (Table 5) match predictions indicating a $10 increase (decrease) in the HMO (PPO) workers contribution decreases the fraction of enrollees choosing the HMO plan by 0.34 (0.38)

19 19 percentage points. Both the full and restricted sample estimates also show that higher expected OOP expenses for the PPO plan lead to a higher enrollment share for the HMO plan. The FE estimates based on the restricted sample continue to show a negative effect of rising HMO OOP premiums on the share of enrollees choosing the HMO plan; however the coefficient is no longer statistically significant. The model indicates a $10 increase in the OOP premium for the PPO plan increases the HMO share by 0.18 percentage points. Based on these estimates, the average increase in OOP premiums for the two plan types predicts a 1.1 percentage point increase in the HMO share of enrollment. Instead, we observe a 4 point increase in the HMO share of enrollees, indicating a possible change in preference for type of health insurance plan. The increases in the expected OOP expenditures for the two plan types essentially cancel each other out in terms of the enrollment share. Robustness Check In the main estimation, we employed the overall and OOP premiums for the family plans. As a robustness check, we estimate the models using single coverage premiums. The results (presented in Table 6) are obtained via fixed effects estimation on the sample restricted to single establishment plants. There is a high degree of consistency across these alternative measures of the OOP and overall premiums. For the overall take-up rate, the model using the single plan OOP premiums and expected expenditures yield the higher R-squared. For the remaining outcome variables, the models using the family premiums and expected cost variables perform better. In any case, the results are highly consistent, providing greater confidence in the accuracy of our main results.

20 20 7. Conclusions This study looks at how OOP costs for ESHI plans affect take up rates when firms offer a variety of health insurance plans. In particular, we conjecture that changes in the OOP contribution can lead both to switching between plans and a decrease in the take-up rate. While these two problems have been analyzed independently in previous studies, nobody has provided a comprehensive examination of the two phenomena. Consistent with previous research, we find that rising worker contributions towards HMO and PPO plan premiums has relatively little impact on overall health insurance take up rates for employer sponsored plans. However, even if the estimated effects are small, the changes in the OOP for HMO and PPO are large enough to explain most of the decline in the take-up rate during the period under analysis. Changes in OOP costs can only explain roughly one-quarter of the increase in the HMO share of workers enrolled in employer sponsored health insurance. Overall, we found the HMO take-up rate is less sensitive to changes in the worker contributions schedule than the PPO take-up rate. Moreover, FE estimates indicate the overall take-up rate is not responsive to changes in HMO worker contributions, but it decreases when worker s contribution for PPO plans increase. This result supports the hypothesis that workers respond to an increase in the OOP costs for HMO plans by switching to PPO plans, while increases in the OOP premium of a PPO plan induce some workers to switch to HMO plans and some workers to drop out of the group coverage altogether. In light of this finding we conclude that when the previous literature computed the elasticity of the demand for health insurance based on the lowest or the average worker s contribution, it used the incorrect measure. Our results indicate that the investigation of take up rates cannot neglect anymore the consideration of deductibles, copayments and coinsurance rates as they explain half of the drop

21 21 in the overall take-up. These fees are expected to play an even larger role with the rapid surge in enrollments high deductible health plans. Future research should investigate this recent trend and examine how the increased prevalence of these plans affects our understanding of employee responsiveness to shifting the cost of coverage from premiums to expected expenditure.

22 22 Bibliography Atherly A., B. E. Dowd, and R. Feldman, 2004, The Effect of Benefits, Premiums, and Health Risk on Health Plan Choice in the Medicare Program, Health Services Research, 39(4): Buchmueller T. C., 2000, The Health Plan Choices of Retirees under Managed Competition, Health Services Research, 35 (5): Buchmueller T. C., and P. J. Feldstein, The Effect of Price on Switching among Health Plans, Journal of Health Economics, 16: Chernew, M., K. Frick, and C. G. McLaughlin, 1997, The Demand for Health Insurance Coverage by Low-Income Workers: Can Reduced Premiums Achieve Full Coverage? Health Services Research, 32 (4): Clemans-Cope, Lisa, and Bowen Garrett. Changes in Employer-Sponsored Health Insurance Sponsorship, Eligibility, and Participation: 2001 to Kaiser Family Foundation, Kaiser Commission on Medicaid and the Uninsured (December 2006), Cutler, D. M., 2003, Employee Costs and the Decline in Health Insurance Coverage, in D. Cutler and A. M Carber (Eds) Frontiers in Health Policy Research, Vol 6, MIT press. Dowd, B.E., R. Feldman, and R.Coulam, 2003, The Effect of Health Plan Characteristics on Medicare+ Choice Enrollment, Health Services Research, 38(1): Gruber, J. and E. Washington, 2015, Subsidies to Employee Health Insurance Premiums and the Health Insurance Market, Journal of Health Economics, 24 (2005): Kosteas, V., and F. Renna, 2014, Plan Choice, Health Insurance Cost and Premium Sharing, Journal of Health Economics, 35:

23 23 Kaiser Family Foundation/Health Research & Educational Trust, 2014, 2014 Employer Health Benefit Survey, Jacobs, P. J., 2009, Forum in Health Economics and Policy, 12(2), Article 3. Medical Expenditures Panel Surveys Household Component Summary Data Tables mponent=0 Polsky, D., R. Stein, S. Nicholson, and M. K. Bundorf, 2005, Insurance, Prescription, Coverage and Pricing, Health Services Research, 40(5): Royalty, A. B., and N. Solomon, 1999, Health Plan Choice: Price Elasticities in a Managed Competition Setting, The Journal of Human Resources, 34(1): 1-41 Short, P., and A. Taylor, Premiums, Benefits and Employee Choice of Health Insurance Options. Journal of Health Economics 8, Strombom B. A, T. C. Buchmueller, and P. J. Feldstein, Switching Costs, Price Sensitivity, and Health Plan Choice. Journal of Health Economics, 21: Welch, W.P., The Elasticity of Demand for Health Maintenance Organizations, Journal of Human Resources 21,

24 24 Figure 1: Workers monthly contributions family coverage PPO HMO Figure 2: Firm and plan take-up rates percentage overall take-up HMO take-up PPO take-up

25 25 Table 1: descriptive statistics All Firms Offering Including Only Single HMO and PPO Plans Establishment Firms Mean Std Dev Mean Std Dev Overall take-up rate HMO+PPO take-up rate HMO take-up rate PPO take-up rate HMO Share of enrollment Worker's contribution to family HMO plan Worker's contribution to family PPO plan Monthly premium for family HMO plan Monthly premium for family PPO plan Expected OOP cost for HMO plan (100s $) Expected OOP cost for PPO plan (100s $) Firm offered POS plan Firm offered conventional plan Log total employment (capped) Percent employees low income Union indicator Wait period for ESHI Part-time workers eligible for ESHI Temporary workers eligible for ESHI Observations 3,193 1,243

26 26 Table 2: Take-up rates for full sample and only firms offering both HMO & PPO plans Full Sample Firms offering HMO & PPO Plans OLS FE OLS FE Lowest OOP contribution ** * ** to family plan (0.001) (0.0012) (0.0019) (0.002) Observations 12,605 12,605 3,434 3,434 Number of firms R-squared Standard errors corrected for clustering in parentheses. +,*,** indicate significance at the 10,5,1 percent levels, respectively.

27 27 Table 3: HMO & PPO take-up rates HMO Plan Take-up PPO Plan Take-up OLS 1 OLS 2 FE 1 FE 2 OLS 1 OLS 2 FE 1 FE 2 Worker's contribution to ** ** ** 0.025** family HMO plan (ζ 1 ) (0.003) (0.005) (0.003) (0.009) (0.003) (0.005) (0.003) (0.006) Worker's contribution to 0.032** 0.029** 0.005* ** ** ** ** family PPO plan (ζ 2 ) (0.002) (0.003) (0.002) (0.005) (0.002) (0.003) (0.003) (0.005) Expected HMO OOP expenditures (0.027) (0.044) (0.027) (0.076) (0.026) (0.039) (0.030) (0.049) Expected PPO OOP ** ** ** ** expenditures (0.022) (0.032) (0.018) (0.037) (0.022) (0.032) (0.019) (0.030) Observations Number of firms R-squared Standard errors corrected for clustering in parentheses. +,*,** indicate significance at the 10,5,1 percent levels, respectively. Models OLS 2 and FE 2 restrict the sample to single establishment firms.

28 28 Table 4: Overall take up rates Overall Take-up Rate HMO+PPO Take-up Rate OLS 1 OLS 2 FE 1 FE 2 OLS 1 OLS 2 FE 1 FE 2 Worker's contribution to ** * ** * family HMO plan (ζ 1 ) (0.002) (0.003) (0.002) (0.004) (0.002) (0.003) (0.003) (0.005) Worker's contribution to * family PPO plan (ζ 2 ) (0.001) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.004) Expected HMO OOP * expenditures (0.015) (0.029) (0.017) (0.054) (0.018) (0.030) (0.024) (0.059) Expected PPO OOP ** ** * expenditures (0.010) (0.015) (0.012) (0.018) (0.014) (0.017) (0.018) (0.029) Observations Number of firms R-squared Standard errors corrected for clustering in parentheses. +,*,** indicate significance at the 10,5,1 percent levels, respectively. Models OLS 2 and FE 2 restrict the sample to single establishment firms.

29 29 Table 5: Share enrolled in HMO plan OLS 1 OLS 2 FE 1 FE 2 Worker's contribution to ** ** family HMO plan (ζ 1 ) (0.004) (0.005) (0.003) (0.008) Worker's contribution to 0.042** 0.038** 0.009** 0.018** family PPO plan (ζ 2 ) (0.003) (0.004) (0.003) (0.005) Expected HMO OOP * expenditures (0.031) (0.048) (0.032) (0.079) Expected PPO OOP 0.061* 0.109** expenditures (0.025) (0.036) (0.021) (0.036) Observations Number of firms R-squared T-test for ζ 1 = ζ Standard errors corrected for clustering in parentheses. +,*,** indicate significance at the 10,5,1 percent levels, respectively. Models OLS 2 and FE 2 restrict the sample to single establishment firms.

30 30 Table 6: Single plan premiums and expected OOP costs Overall HMO + PPO HMO PPO HMO Share Worker's contribution to HMO plan (ζ 1 ) (0.012) (0.015) (0.022) (0.019) (0.021) Worker's contribution to 0.014* * 0.052* PPO plan (ζ 2 ) (0.007) (0.014) (0.020) (0.019) (0.023) Expected HMO OOP expenditures (0.153) (0.172) (0.211) (0.149) (0.219) Expected PPO OOP expenditures (0.052) (0.084) (0.108) (0.087) (0.104) R-squared Observations 1,243 1,243 1,243 1,243 1,243 Firms Standard errors corrected for clustering in parentheses. +,*,** indicate significance at the 10,5,1 percent levels, respectively. All results are for the fixed effects model estimated on the sample of single establishment plants.