Abstract
This paper provides field evidence on how price framing affects consumers’ decision to switch health plans. In 2009 German federal regulation required insurers to express premium differences between standardized health plans in absolute euro values relative to a federal reference price, rather than in percentage point payroll tax differences. Representative individual-level panel data and aggregated health plan level panel data on the universe of health plans show that the reform strongly increased enrollees’ willingness to switch plans, with demand elasticities increasing fourfold. The salience of premium differences and the default premium are obviously key driving forces in the decision to switch insurers.
I. Introduction
Price competition is a central component of managed competition. Under managed competition, consumers are provided with standardized benefit packages and transparent information about quality of care, along with various health plans from which to choose. In theory, insurers then face incentives to engage in price and quality competition. A necessary condition for this model to work in practice is cost consciousness among consumers and a sufficiently large health plan premium elasticity. The question of how to design insurance markets that foster price transparency and competition between insurers is critical to dampen health expenditure growth while improving quality of care.
Similar to the United States, Switzerland, or the Netherlands, Germany is a healthcare market with high per capita healthcare spending (OECD, 2015). However, the cost consciousness of consumers has been rather weak in these markets. For example, in Germany, despite considerable price differences between very similar health plans (called “sickness funds”), switching rates have been low (Schut et al. 2003; Nuscheler and Knaus 2005; Tamm et al. 2007; Bauhoff 2012; Eibich et al. 2012). Such weak consumer response has been identified as one driver of high health expenditures; as long as enrollees do not switch to cheaper plans, insurers do not have incentives to contain costs (Richter 2009).
This paper evaluates a German health insurance reform that intended to foster consumer price sensitivity. Effective 2009, the German federal legislature implemented a reform that changed the price framing rules for health plans. Before the reform, health plan premiums were expressed in percentage points of the gross wage and paid through payroll deductions (“contribution rates”). To foster competition among insurers, the reform introduced a unique contribution rate for all health plans and forced insurers to express price deviations from this standard price in absolute euro values.
To date, the large majority of the existing evidence on consumer price sensitivity of health plan choice comes from the United States where elasticity estimates vary tremendously. 1 We study the German public health insurance market. This market has some convenient features that allow us to shut down multiple potential confounders when estimating price effects. Germany has a universal but employer-based multipayer health insurance system. With heave federal regulation on essential benefits and cost-sharing, nearly 124 identical health plans are offered by sickness funds, which mainly compete on the basis of prices. Exploiting individual-level panel data as well as aggregated health plan level panel data, we study this unique natural price framing experiment from different perspectives.
Pre-reform, individual-level panel data show that neither high premium levels relative to the average market premium, nor premium increases relative to the average market increase induced consumers to switch to cheaper plans. Post-reform, however, monthly premium increases of €10 above the market increase roughly tripled the 5.3 percent baseline probability to switch. We also show that the increase in price responsiveness is equally large for enrollees who did not experience real premium increases post-reform, but only a pure change in price framing. The increase in consumer price responsiveness is robust to a plenitude of specifications. For example, aggregated health plan level panel data from across all German sickness funds show that the reform quadrupled demand price elasticities from −1 to −4.
The most likely explanation for the increase in consumer price sensitivity is the increased salience of premium differences after the price framing reform. Note that we broadly define price framing as how insurance premiums are generally labeled and presented to consumers. The paper studies the net effect of a change in the price framing regime that includes two main components: (i) the change from relative to absolute premium differences between plans; and (ii) the introduction of a fixed, and across health plans unique, reference price as defined by the regulator. Because both elements were introduced simultaneously, we cannot disentangle the separate effects. However, we see both elements as integral parts of the overall price framing regime; both increased the salience of premium differences between health plans.
Through laboratory experiments, various studies show that the framing of decisions matter (Tversky and Kahneman 1981; Pope and Sydnor 2015). Few studies exploit realworld settings (DellaVigna 2009; Lacetera et al. 2012; Choi et al. 2014). Chetty et al. (2009) demonstrate in a field experiment that the price framing of commodity taxes affects demand. In this respect, this paper also relates to the literature on the salience of taxes and its impact on welfare (Finkelstein 2009; Chetty 2009; Goldin and Homonoff 2013). Wallace and Huck (2010) provide a literature review of the impact of price framing on consumer decision making and conclude that there is limited evidence on price framing. Bertini and Wathieu (2006) also offer a concise introduction of how the framing of prices affects the perceived value of goods. In a recent paper, Goda and Manchester (2013) study the role of default options for retirement plan choices and find that the default increased enrollment by 60 percentage points. Another behavioral study that is close in spirit to ours is Hastings and Tejeda-Ashton (2008). In an experiment conducted in Mexico, they use a sample of financially illiterate individuals to show that presenting fees in pesos instead of annual percentage rates induces changes in choice behavior in a hypothetical investment setting.
By studying behavioral reactions in a health economics context (Just and Price 2013; Altmann and Traxler 2014; Royer et al. 2015), this paper also adds to the literature on Medicare Part D, which suggests that the health plan choices of the elderly are not consistent with optimization under full information (Abaluck and Gruber 2011; Ketcham et al. 2012; Heiss et al. 2006, 2010, 2009, 2013). On the other hand, Ketcham et al. (2015) find that neither switching rates nor price responsiveness decreased over time, and that 50 percent of all enrollees had switched plans between 2006 and 2010. Using U.S. data from a big employer, Bhargava et al. (2017) find that the majority of employees choose dominated options with excess spending amounting to almost half of the employee share of the premium. Providing real-world evidence from Switzerland, Frank and Lamiraud (2009) find that an increase in the number of choices (to more than 50 health plans) actually reduces enrollees’ willingness to switch plans. Schram and Sonnemans (2011) confirm the hypothesis of decision overload in laboratory experiments with undergraduate students from the Netherlands.
The next section outlines the institutional details of the German healthcare market and the health insurance reform. Section III provides information and results on the individual-level panel data set SOEP. Section IV provides complementary evidence from aggregated health plan level data. After a general discussion of the results, Section VI concludes.
II. Institutional Background and Reform
A. The German Health Insurance System
The German health insurance system comprises a private and a public system which coexist under universal coverage (Ziebarth 2010, 2014). This paper focuses on the public system that covers 90 percent of the German population. Employees whose gross wage is below a specific income threshold (in 2015: €54,900 per year) are compulsorily insured under the public scheme. Nonworking spouses and dependent children are covered at no extra cost by a public family insurance. Special regulations apply to specific groups, such as students and the unemployed. However, most of these individuals are publicly insured.
Under the public scheme, premiums are not risk-rated and depend only on income. They are collected in the form of “contribution rates. ” Sickness funds are nonprofit organizations meaning that, in the medium-run, their revenue must equal their expenses. There exists guaranteed issue and a free choice of providers, which is no provider networks. Reimbursement of providers is determined centrally and does not vary across sickness funds.
Everyone covered under the public scheme is entitled to a generous benefit package. These essential benefits are determined at the federal level and are codified in the Code of Social Law Book V (“Sozialgesetzbuch V”). Coinsurance rates and deductibles are prohibited in the public system and the relatively small copayment amounts are fixed at the federal level and do not vary across plans.
In 2015, a total of 124 public health insurers, called “sickness funds,” existed in Germany (German Federal Statistical Office 2015). Each sickness fund offers the standard health plan as determined by social law; thus, henceforth, we use the terms “health insurer,” “sickness fund,” and “health plan” interchangeably. Due to the federal regulations on cost-sharing and the generous essential benefit package, sickness funds have few options to differentiate their product. Basically, they only can vary their service quality2 or offer “not medically necessary” supplementary benefits in addition to the essential benefit package. (See Appendix D.) To summarize, the 124 German sickness funds offer very homogeneous products and mainly compete on the basis of prices.
At the time of writing, 43 of the 124 health plans operate nationwide on the federal level and every German resident can enroll. In addition, 53 health plans operate only in some of the 16 German federal states. Finally, there exist 28 firm-specific “closed” health plans (similar to the employer-sponsored health plans in the United States), where only the company employees can enroll.
Hence, the choice set—or the number of available sickness funds—varies across the 16 German states and over time. In 2002, depending on the state of residence, enrollees could choose between a minimum of 82 and a maximum of 125 health plans. By 2011 these numbers were steadily decreasing due to many mergers but enrollees still could choose from between 56 to 79 plans. We control for the size of the choice set in all regressions below.
B. Health Plan Premiums and Switching Regulations Pre- and Post-Reform
1. Framing of health plan premiums and price differences
German public health insurance is financed by mandatory payroll deductions. The reform evaluated in this paper became effective on January 1, 2009. Prior to the reform, premiums (“contributions”) were calculated as:
Hence, the total monthly premium of enrollee i equaled the contribution rate of her sickness fund k, multiplied by her gross wage, up to the contribution ceiling (for example, €40,500 per year in 2002).3 By law, this total premium had then to be split equally between employer and employees. The employee share of the premium, which basically represents a mandatory payroll tax, was automatically deducted from employees’ paychecks. Note that the contribution rates were independently set by the sickness funds and hence varied across funds. For example, assuming a specific contribution rate of 14 percent and a gross wage of € 2,500, 0.07 ·€2,500 = €175 per month were deducted from the employee’s paycheck. The employer contributed the same amount.
Effective January 1, 2009, the price framing reform altered the formula to:
As seen, contribution rates—which, pre-reform, were independently set by each fund—were equalized and frozen to 15.5 percent for all sickness funds. By law, if they could not cover their expenses by the revenues generated from the equalized general contribution rate, sickness funds were now required to charge an “add-on premium” in the form of an income-independent absolute monthly euro amount. Employees were billed for the entire add-on premium separately.
On the other hand, financially sound sickness funds directly could reimburse part of the enrollees’ premium in the form of a monthly euro amount (in other words, impose a negative add-on premium or refund).4
To sum up, the law forced sickness funds to no longer express premium differences in contribution rate percentage points differences relative to wages but instead in absolute euro values relative to a newly introduced standard premium. Now health plans had to bill (or reimburse) add-on premiums separately, whereas the 15.5 percent standard contribution still was deducted from the payroll. In practice, add-on premiums were typically paid by automatic deposit, with refunds paid by direct deposit on an annual basis. Note that health plans were legally prohibited from regional price discrimination; neither contribution rates (pre-reform) nor add-on premiums (post-reform) varied by geography.
2. The cheapest and most expensive plans pre- and post-reform
Pre-reform, in 2008, contribution rates varied considerably between sickness funds. An individual with an average gross wage of €2,552 (German Statutory Pension Insurance 2012), could have saved €50 per month by switching from the most expensive fund “City BKK” (contribution rate: 17.4 percent) to one of the least expensive funds, the BKK Gildemeister Seidensticker (contribution rate: 13.5 percent, see Table A1).5 Leaving the City BKK would have been easy due to guaranteed issue.
Post-reform, a total of 17 sickness funds charged add-on premiums, including two of the largest funds: the Deutsche Angestellten Krankenkasse (DAK) with 4.4 million enrollees and a nationwide health plan market share of 8.6 percent, and the Kaufmännische Krankenkasse (KKH) with 1.4 million enrollees and a market share of 2.4 percent. Add-on premiums varied between €6.50 and €15 per month. The majority of funds charged €8, which is, €96 per year (Table A1).
At the end of 2011, a total of 7.3 million Germans were enrolled in health plans that charged add-on premiums. This represents 15 percent of all those publicly insured. DAK and KKH enrollees alone account for 80 percent of this total figure. Hence, the main post-reform identifying variation in our individual level data set comes from employees enrolled in these two plans. Table A3 in the Appendix shows that enrollees in add-on premium plans are not a selective subsample. Their observable characteristics are very similar to the observables of enrollees in other plans.
In total, 15 sickness funds reimbursed their members between €2.50 (BKK A.T.U.) andV10 (BKK Würth) per month (Table A1). However, the average German is unaware of the existence of these relatively small health plans; the most prominent of which had a market share of just 0.5 percent. Most of these plans only operated in specific states with few branch offices, a very basic service infrastructure, and few employees. Only four out of the 15 reimbursing funds operated nationwide. Table A1 in the Appendix lists all sickness funds with add-on premiums and refunds along with their pre-reform contribution rate. As seen, pre- and post-reform prices are highly correlated.
3. Health plan switching regulations
When enrollees intend to switch health plans, they have to give written notice of their cancellation, which is feasible online. Enrollees can then switch two months from the end of the month in which notice was given. This means that there exists no fixed enrollment period like in the United States. In principle, the minimum contract period for newenrollees is 18 months. However, when health plans increase premiums, they are legally required to give written notice at least one month in advance. Regardless of the actual enrollment length, enrollees then have a right to cancel their contract and switch plans within two months. At the time of writing, individuals could choose between any of the 50 to 70 sickness funds in their choice set, which have to accept them without risk assessment.
None of the switching regulations changed in the course of the reform. Switching health plans via the Internet is uncomplicated and was possible pre- and post-reform. During this time period, specific websites existed to help consumers compare and switch health plans.
III. Evidence from Individual-Level Data
This section presents the main empirical analysis. It relies on rich individual- level panel data from the German Socio-Economic Panel Study (SOEP). The SOEP is a large and representative household panel that started in 1984 in West Germany and was extended to include East Germany in June 1990. More than 11,000 households comprising more than 22,000 individuals participate in the survey every year. The SOEP core questionnaire includes a wide array of questions on well-being, labor market activities, and health. Most of these questions are surveyed annually. More information on the SOEP can be found in Wagner et al. (2007). For our empirical analysis, we make use of the Waves 2002 to 2011.
Our empirical analysis focuses on sickness fund enrollees, not the total number of insured. The latter also would include family members insured at no extra cost under the public family insurance. More precisely, we focus on “paying” members in order to have one decision-making unit per observation. We define paying sickness fund enrollees as those who are gainfully employed, earn more than €800 gross per month, and pay the full health insurance premium as indicated in the SOEP.6 This restricts the sample to individuals aged between 18 and 64. In addition, we disregard those insured under the second tier of the German healthcare system—the private health insurance.
We end up with a sample of 41,942 person-year observations from 10,552 individuals. Exploiting the sickness fund information in the SOEP and the publicly available information on the interview date and the state of residence, we are able to assign all respondents to their health plans.
A. Potential and Realized Savings Pre- and Post-Reform, or: “How Much Money Has Been Left on the Table?”
In order to accurately calculate enrollees’ actual and potential monthly health plan premiums and measure price effects, we collect the publicly available health plan price information on a monthly basis for all sickness funds between 2002 and 2011. According to the interview date and indicated sickness fund, we then precisely assign enrollees their actual and potential contribution rates. To determine the monthly premium in euros, we only consider the employee premium share (in 2009 euros).7
In addition, we consider the actual health plan choice set of each enrollee at the time of the interview to calculate the exact monthly savings potential for each enrollee or “how much money are enrollees leaving on the table by not switching to the cheapest health plan in their choice set?”
Table 1 displays potential (Panel A) and, for switchers, realized (Panel B) monthly savings pre- and post-reform for our sample of representative SOEP respondents. It basically illustrates how the savings potential (determined at the health plan level) translates into realized savings (determined by the individual) using representative SOEP data.8
Potential and Realized Savings through Switching Pre- and Post-Reform
Panel A shows that, pre-reform, the maximal potential monthly savings—or the premium difference between the own and the cheapest plan in the choice set—were €23.32 on average (median = €20.64). After the reform, the premium distribution— especially the tails—had been compressed and the savings potential decreased toV8.60 (median = €8.50). This reflects the difference between the standard price and the €8 add-on premiums on the one hand, as well as the difference between the standard price and the up to €10 refund on the other hand. See also Table A1.
Next we investigate the savings potential with respect to the average market premium. When making switching decisions, enrollees might focus on the difference to the average market premium instead of the cheapest plan. As seen in Table 1, this average savings potential is significantly smaller at €2.37 pre- and €0.18 postreform. The compression of the premium distribution is nicely illustrated by the minimum and maximum values: pre-reform, enrollees in the cheapest plan paid €38.52 per month less and enrollees in themost expensive planV45.86more than the average consumer. The main hypothesis of this paper is that some consumers paid €550 per year more than the average consumer because actual premium differences were obscured by the price framing regulation and a lack of transparency. When the price framing reform made premium differences more salient, many consumers— particularly those enrolled in the expensive plans—switched, and the new market equilibrium resulted in less price dispersion. Post-reform, as compared to the average consumer, enrollees in the cheapest plans saved €8.50 per month and those in the most expensive plans left €14.48 on the table.
Panel B of Table 1 displays the realized savings per month and enrollee for those who switched health plans. Whereas the average realized savings were €9.47 pre-reform, post-reform realized savings remained remarkably stable at €7.80, despite the stark compression in the premium distribution.
One can conclude that, after the price framing reform, potential savings decreased significantly but realized savings remained constant. This is strong initial evidence of an increase in market price competition as a result of an increase in consumer price sensitivity. Indeed, as we show in the next section, enrollees who faced premium increases above the market increase were more likely to switch health plans after price differences became more salient.
B. Analyzing Individual-Level Switching Behavior
1. Dependent variable: Switching indicator
Our dependent variable in the following analysis indicates whether an individual switched health plans between two interviews. Interviews are typically carried out in the first five months of a year. In addition to eliciting the name of the current health plan, the SOEP explicitly asks whether the respondent had switched health plans between the end of the previous calendar year and the interview.
Our generated binary dependent variable is called Switch. As an example, a value of 1 in year 2006 (t) implies that the respondent switched plans between the interview in 2006 (t), and the interview in 2007 (t + 1). Appendix A2 shows that 5.3 percent of all person-year observations represent switches; in other words, we observe 2,224 health plan switches over all years and across all sickness funds.9
2. Switching model
The main switching regressions estimate the following linear probability model:10
(1)where Switchit indicates whether enrollee i switched health plans between t and t + 1. The vector Xit includes sets of standard socio-economic control variables of enrollee i in period t. These are demographics, educational, and labor market characteristics, health status, and healthcare utilization measures. (See Table A2.) ρst denotes the size of the choice set. State fixed effects σs net out permanent differences across the 16 German federal states and year fixed effects, φt, take account of common time shocks. Health plan fixed effects, δp, do not alter the findings but make the interpretation of the price regressors less straightforward, which is why we only include them in a robustness check in Table A4. As usual, εit represents the error term. All variable names, definitions, and descriptive statistics are shown in Appendix A2.
Prereformt is one in years before the reform and zero thereafter, whereas the opposite holds for postreformt. PriceIndicatorit measures the health plan premium of enrollee i in t as well as premium increases between t and t + 1. More specifically, PriceIndicatorit is representative of four different price regressors (all values are converted to 2009 euros):
1. Premium difference to lowest premium in choice set in t
2. Premium difference to average market premium in t
3. Premium increase between t and t + 1
4. Premium increase, relative to average market increase, between t and t + 1
3. Price regressors measuring premiums in levels and premium increases
Premium Levels. The first price regressor employed is Difference to lowest premium. We already displayed its descriptives in Table 1. Difference to lowest premium measures the difference between each enrollee’s actual monthly premium and the lowest premium in her state-year level choice set at the time of the interview.
The second price indicator, Difference to market premium, is also displayed in Table 1 and measures the premium differential to the average market premium using the average contribution rate. Note that these two price regressors measure premium differences in levels and indicate the savings potential exploiting individual choice sets.
Premium Increases. It is reasonable to assume that enrollees mostly consider switching health plans when they receive updated premium information. Premium increases are an important trigger, particularly in a highly price competitive market like Germany. German regulation stipulates that premium increases have to be announced by the insurer in a personal letter. This letter must explicitly explain that enrollees have an extraordinary right to switch plans, irrespective of the enrollment time. Another reason to focus on premium increases is the empirical irrelevance of premium decreases. For the whole of Germany, in December 2011, enrollees in health plans that awarded refunds accounted for only 1.4 percent of all enrollees (Table A1). Consequently, in the SOEP we only observe 39 respondents who received refunds—too few to obtain precise estimates.11
To identify premium increases, we calculate the premiums for each enrollee at the time of the SOEP interview in the two consecutiveWaves t and t + 1. Then we take the difference Premium Increase = premiumt + 1 - premiumt and use it as third price indicator. More precisely, because a premium increase between t and t + 1 might have induced a switch from health plan A (in t) to health plan B (in t + 1), we use the contribution rate of plan A in t + 1 to calculate Premium Increaset even if the individual has already switched to (the probably cheaper) plan B. Only this approach allows us to measure the effect of a price increase of sickness fund A. Moreover, we fix labor income in t. As such, we correctly assign premium increases to contribution rate increases and do not confuse them with wage increases.12
Table 2 shows that not every enrollee experienced premium increases in every year. It displays annual premium increases and the number of affected enrollees. Column 3 shows our third price indicator, Premium increase (absolute), conditional on an increase. For example, between 2003 and 2004, 39 percent of all health plans increased premiums—by an average of €11.11 per month (€133.32 per year) or 5.74 percent. Table 2 also illustrates that premium increases occur in annual waves due to healthcare cost shocks triggered by increases in consumption or provider reimbursement. It is easy to see that there is a lot of variation in the share of respondents who were affected by price increases—between 10 and 69 percent of all health plans increased prices in a given year.13 Hence the identified effects are based on a very broad set of enrollees and their switching behavior.
Size of Health Plan Price Increases Over Time
Premium increase (absolute) does not consider simultaneous premium increases of competing health plans. If all health plans simultaneously increased their premium by €10, enrollees obviously may not switch plans. Thus, our fourth (and preferred) price indicator is Premium increase (relative to market increase). It measures the difference between the premium increase of the own plan and the average market increase.14 If all plans simultaneously increased their premiums byV10, the relative increase would be zero.
The last column of Table 2 shows Premium increase (relative to market increase) conditional on increases. The difference to the third price indicator in Column 3 depends on how many sickness funds increased prices at the same time in a given year. In 2005, when only 22 percent of all sickness funds increased contribution rates, the absolute price increase was almost identical to the increase relative to the market increase (€5.71 vs. €5.67). In contrast, in 2007 69 percent of all sickness funds increased contribution rates and thus the difference between absolute and relative price increases was much larger (€9.51 vs. €4.44).
C. Increase in Consumer Price Responsiveness
Table 3 presents the baseline regression results and reads as follows: Each column represents one regression model as in Equation 1. All price indicators are measured in €10 units as of 2009.
Increase in the Salience of Premium Differences and in Consumer Price Sensitivity
Column 1 makes use of the first price indicator Difference to lowest premium. Prereform, a €10 premium difference between the own and the cheapest health plan in the choice set (the “savings potential”) increased the 5.3 percent baseline probability to switch by 1.2 percentage points. Hence, differences in price levels were quantitatively not very relevant switching determinants in pre- reform years. However, post-reform, a €10 premium difference increased the switching probability by ten percentage points, which represents a threefold increase of the baseline probability to switch.
Column 2 makes use of the second price indicator Difference to average premium. As compared to Column 1, the pre-reform behavioral reaction is slightly larger but is still small. This could indicate that enrollees benchmarked their premium to the average market premium rather than the cheapest plan in the individual choice set. In any case, the savings potential with respect to the average market premium matters a lot more post-reform. Here, a €10 monthly premium difference again triples the baseline switching probability.
Column 3 uses the third price regressor Premium increase (absolute) as the main explanatory variable. Before the price framing reform, a monthly premium increase of €10 increased the switching probability by 1.4 percentage points. Post-reform, the same price increase raised the probability of switching funds by 11.7 percentage points.
Finally, Column 4 accounts for the simultaneous price increases of other plans in the market, and uses the fourth price indicator Premium increase (relative to market increase). This model specification does not alter the coefficient estimates. Price increases after 2009 triggered a much stronger consumer reaction. Health plan switching probabilities increased by 12.7 percentage points when plans raised premiums by a monthly €10 above the market increase.
Using our preferred specification in Column 4, additional robustness checks are shown and discussed in the Appendix (Table A4). Among other checks, we vary the sample selection by excluding those who are optionally insured under the public system or by restricting the years under consideration.
Overall, we conclude that the findings in Table 3 are robust, regardless of whether we use the premium in levels or premium increases, or whether we exploit differences to the cheapest plan in the choice set or to the average health plan on the market. Even prereform, some enrollees were price-conscious and switched plans at significantly higher rates when premiums increased or were significantly higher than the average market premium. Still, the average switching probability was just above 5 percent. After the price framing reform that introduced a standard premium and made premium differences more salient to the consumer, the price sensitivity increased by several magnitudes.
D. Analyzing Pure Price Framing Changes
The previous section quantified the effects of different price levels and price increases pre- as compared to post-reform. In order to identify a pure price framing effect, this section employs a different specification with treatment and control groups. Thus, it adds to the previous evidence on price framing effects by carrying out a difference-indifferences (DiD)-type analysis. For this purpose, we focus on a subsample of individuals and the years 2005 to 2011. Table 4 summarizes the definition of treatment and control groups.
Definition of Treatment and Control Groups
As illustrated in Table 4, the Control Group consists of enrollees who (i) pre-reform, paid contribution rates of approximately 15.5 percent and had (ii) no add-on premiums or refunds post-reform. Hence, neither their premiums nor the framing of their premiums changed in the course of the reform.
Treatment Group 1 consists of enrollees whose contribution rate was likewise 15.5 percent pre- and post-reform but who were charged an add-on premium post-reform. This group underwent a premium increase that was framed in absolute euro values relative to the post-reform standardized contribution rate of 15.5 percent.
Finally, Treatment Group 2 consists of enrollees who experienced a pure change in price framing: Pre-reform, they were already members of an expensive health plan with contribution rates above 15.5 percent. Post-reform, their contribution rate decreased to 15.5 percent but instead they were charged an add-on premium.15 A good example of Treatment Group 2 is the City BKK, which, pre-reform, was the most expensive sickness fund in Germany (17.4 percent contribution rate in 2008, see Table A1). City BKK enrollees could easily have saved up to €50 per month by switching but the overwhelming majority did not switch health plans. Only after the reform, when the City BKK had to charge a monthly add-on premium of €15 and the equalized contribution rate was set to 15.5 percent, did many City BKK enrollees switch to other plans.
Table 5 reports the results of this DiD analysis, which makes use of linear probability models. In practice, due to the lack of health plans that precisely fulfill this criterion, it is not feasible to select plans with a contribution rate of exactly 15.5 percent in 2008 in order to define treatment and control groups. With this limitation in mind, we allow for slightly larger bandwidths. In the first column of Table 5, we focus on enrollees with contribution rates between 15.3 and 15.7 percent and obtain 6,343 person-year observations. Results for bandwidths between 15.2 and 15.8 percent (yielding more observations, not shown) and bandwidths between 15.4 and 15.6 percent (yielding less observations, Column 2) are qualitatively similar.
DiD Regression Results on Health Plan Switching Using Pure Price Framing
Table 5 demonstrates a large consumer response as a result of the price framing reform, even for consumers whose overall premiums did not increase at all. In Column 1, the coefficient of the interaction term Treatment Group1 × postreform indicates a ten percentage point increase in switching probabilities—in response to a premium increase in the form of a €8 monthly add-on premium. Recall that, pre-reform, the individual level switching probability only increased by 1.4 percentage points in response to aV10 monthly premium increase (Table 3, Column 3).
The highly significant coefficient of Treatment Group2 × postreform is even larger at 0.188 in Column 1. This means that enrollees in plans that always had been the most expensive and who solely experienced a differently framed (high) price were a substantial 18.8 percentage points more likely to switch post-reform, relative to the Control Group without any premium or price framing changes.
The differences in price sensitivities between Treatment Group 1 and 2 vanish in Column 2, where we use the smaller 15.4 to 15.6 percent bandwidth to define Treatment Group 1. In this specification, both highly significant coefficients increase to around 0.22. This implies that the pure price framing effect for the most expensive plans is quantitatively similar to the combined premium increase and framing effect of previously slightly cheaper plans that had now become the most expensive plans.16
E. Effect Heterogeneity by Income
Next, we test whether the reform effects differ by household income. The basic hypothesis is that, post reform, poorer households may have reacted more strongly to price increases (as compared to richer households) because aV10 add-on premium represents a steeper increase, relative both to their income and their baseline premium. On the other hand, one could argue that a flat €10 increase is now equally framed for both income groups (namely as aV10 increase), and was differently framed pre-reform. For example, depending on their exact income, pre-reform, aV10 increase may have been framed as a 1 percentage point contribution rate increase for poor households but for wealthier households it was only a 0.5 percentage points increase. If individuals do respond to percentage points contribution rate differences without converting them, then— pre-reform—the poor would have shown a larger price responsiveness than the rich. Hence we hypothesize that (i) pre- and post-reform, the price responsiveness of the poor with respect to identical euro amount increases was larger, and that (ii) post-reform, the price responsiveness increased for both groups.
Here, we return to the full sample specification using Premium Increase (relative to market increase) (as in Column 4 of Table 3) but split the sample at the median of the equivalized household income. The results are shown in Columns 7 and 8 of Table A4 in the Appendix. As can be seen, for both richer and poorer households, the prereform consumer response to premium increases was around 1 percentage point and insignificant. Post-reform, the price responsiveness increased strongly for both groups, to a highly significant 13 percentage points for both groups. Hence, maybe surprisingly, both groups reveal almost the same price responsiveness pre- and postreform.
In yet another specification exploiting the pure contribution rate as regressor (not shown)—and not the actual premium in euros—we find that, pre-reform, poor and rich enrollees reacted similarly to identical percentage point contribution rate differences. Because a one percentage point increase in the contribution rate translates into different euro amounts for both groups, these findings strongly suggest that (a) consumers predominately react to the scale in which prices are presented to them and do not mentally convert them into absolute euro amounts or relative shares of their income or premium. Moreover, this exercise shows once more that (b) consumers react strongly when premium level differences as well as increases are displayed in a more salient way in absolute euro values (relative to a standard price).
Additional effect heterogeneity tests which, however, do not reveal a clear pattern can be found in Schmitz and Ziebarth (2011).
IV. Evidence from Aggregate Health Plan Level Data
A. Data
This section provides complementary evidence of the reform effect. Here we rely on aggregated health plan level data. By law, sickness funds are obliged to collect information on their enrollees and report to the German Ministry of Health (“Bundesministerium für Gesundheit, BMG”) at regular time intervals. We collected annual health plan enrollment data for all German sickness funds from different sources over a ten year time period, which is, from 2002 to 2011.
The objective of this section is to evaluate the price framing effect from a different perspective. In Section III, we took an individual level perspective. Here, the unit of observation is the health plan and the net annual change in health plan enrollment.
In total, we observe 310 different health plans. As a result of mergers, the data set is an unbalanced panel at the health plan level. Table B1 in the Appendix provides descriptive statistics. As can be seen, enrollment varies tremendously between 242 enrollees and 6.5 million enrollees. Relative to the 50.1 million German public health plan policyholders, this translates into market shares between 0.0005 and 13.04 percent.
The identifying variation comes from the variation in health plan premiums across these 310 different health plans over ten years. The dependent variable measures the net annual enrollment changes at the sickness fund level. Health plan fixed effects net out time-invariant unobserved heterogeneity at the health plan level. In addition, to exclude that reverse causality drives the observed relationship, which is, that premiums vary in response to changes in enrollment, the model includes lagged price changes.
Table B1 shows that we observe price increases for 36 percent of all 1,979 health planyear observations. In contrast, only 6.9 percent of all health plan-year observations represent price decreases. This reflects the general trend of rising premiums due to increases in healthcare spending.
B. Empirical Specification
We estimate the following model to assess the impact of premium changes on changes in health plan enrollment by OLS:
(2)where ΔMst = Mst+1 – Mst/Mst measures (in percent) the change in health plan enrollment of sickness fund s during t, that is, between January 1 and December 31 of a given calendar year. Prereform and postreform are defined as above. PremiumIncrease and PremiumDecrease represent different price indicators and are explained in more detail below. Xst includes covariates that measure systematic enrollment variation due to mergers or other health plan characteristics. For example, in addition to various merger controls, a binary variable indicates whether the fund is “open” to everybody or “closed” so that only employees of specific companies can enroll (Table B1). σs stands for health plan fixed effects and nets out persistent growth rate differentials between plans. φt is a vector that includes year dummies and εst is the error term.
C. Results
Figure 1 provides a visual impression of what happened in the German health insurance market after premium differences became more salient. The X-axis displays calendar years 2002 to 2011. For every calendar year between 2002 and 2011 we categorize all German sickness funds into health plans (i) with price increases, (ii) with price decreases, and (iii) without price changes. Table 2 shows how price increases at the health plan level translate into euro values at the individual level: Pre- and post-reform, premium increases amounted to a monthly average of approximately €8. The increase relative to the mean market level increase was slightly less at below €7 per month.
Aggregated Health Plan Level Data—Changes in Premiums and Market Shares
Source: Müller and Lange (2009), Lange (2011), German Federal (Social) Insurance Office, National Association of Statutory Health Insurance Funds, Annual Reports of the Sickness Funds, information by sickness funds. Merger-related enrollee changes are corrected for.
The Y-axis of Figure 1 plots the product of enrollment changes and the health plans’ market share, and thus considers differences in the sizes of health plans. In other words, the Y-axis shows the percentage point changes in market shares for groups (i), (ii), and (iii). As can easily be inferred from Figure 1, in pre-reform years, the market share changes for groups (i) to (iii) were rather small. Pre-reform, plans with premium increases lost, on average, a market share of 0.07 percentage points. This figure translates into an average loss of about 50,000 enrollees. Post-reform, after the change in price framing, the market share loss for plans with premium increases shot up by factor 12 to 0.8 percentage points.
Table 6 provides estimation results from different regression models similar to Equation 2. Each column represents one model and all models use the same dependent variable: the net health plan enrollment change between January 1 and December 31 in percent. The scaling is necessary given the huge differences in the health plans’ sizes. Note that this approach assigns very small sickness funds the same weight as very large sickness funds. The regression results then can be interpreted as “a price change of X leads to an increase/decrease in net plan enrollment by Y percent. ”We believe that this setup and interpretation is superior to others because “health plan enrollment changes in percent of baseline members” seems to be the relevant parameter. Moreover, a priori, there is no reason to believe that findings for small health plans would not translate to large health plans. In fact, the results are robust to weighting the regressions with the market share of the health plans.17
Health Plan Premium Changes and Changes in Plan Enrollment
The first two columns of Table 6 differentiate between premium increases and decreases using binary indicators. The two indicators are interacted with pre- and postreform dummies. Column 1 only controls for year fixed effects. Column 2 adds health plan fixed effects and a set of additional health plan indicators. (See notes to Table 6 for more details.) First, it is reassuring that the coefficients only decrease slightly once we control for time-invariant unobservables at the health plan level. We find that, prereform, price increases triggered net membership losses of about four percent.18 Postreform, this number triples to more than 12 percent. Interestingly, whereas the premium decrease effect in pre-reform years is imprecisely estimated, in post-reform years we find significant increases in enrollment of about 7 percent as a reaction to premium decreases, meaning premium refunds.
Columns 3 and 4 now replace the two separate price increase and decrease dummies with one regressor that measures the monthly health plan premium in euros.19 The resulting estimates are all highly significant at the 1 percent level. Controlling for health plan fixed effects and further background information does not alter the results. A €10 increase in the monthly health plan premium decreases plan enrollment by about 5 percent. Related to the average monthly employee share of the premium (€189, Table B1), €10 represents a price increase of 5.3 percent. Because a 5 percent price increase triggers an enrollment decrease of around 5 percent, the pre-reform out-of-pocket price elasticity is approximately −1.
After the reform, the price sensitivity increased by factor four. Now, a €10 monthly premium increase leads to a net decrease in plan enrollment of approximately 20 percent. Likewise, the price elasticity quadrupled to −4.
Columns 5 and 6 split the price regressor into one premium increase and one premium decrease regressor for post-reform years. To estimate the pre-reform effect, we employ the sickness fund-specific contribution rate in percentage points of the gross wage. Again, controlling for additional covariates and health plan fixed effects barely changes the coefficient estimates.
Pre-reform, a one percentage point increase in the sickness fund-specific contribution rate decreased health plan enrollment by approximately 7 percent. This estimate is in line with Schut et al. (2003), who use similar data for Germany, but an earlier time period. They estimate the membership elasticity with respect to contribution rates to be −4.8.Aprice increase in the form of a 1 percentage point increase in the contribution rate represents an absolute price increase of approximately €25 per month. Relative to the average “full” premium (employee and employer share) over all years—€378 per month—this translates into a price increase of 6.6 percent. Hence, the implied full (“insurer-perspective”) demand elasticity with respect to health plan enrollment is around −1, which is in line with the findings above.
AddOnEuro × postreform and RefundEuro × postreform in Columns 5 and 6 yield the results for price increases and decreases in post-reform years. As can be seen, the price sensitivity increased substantially post-reform. A €10 higher premium relative to the standard premium decreased plan enrollment by 18.9 percent. Interestingly, a €10 lower premium increased plan enrollment significantly by almost the same amount: 16.3 percent. The standard errors indicate that these two estimates do not differ in a statistical sense.
V. Discussion
This paper finds a strong increase in consumer price sensitivity after a price framing reform. Here we discuss potential threats to a causal interpretation and examine the underlying mechanisms.
First of all, the reform debate itself may have increased consumer awareness, which is a general concern for any policy evaluation. However, healthcare issues are permanently under debate in the German media. Appendix C shows Google search rates for “health plan switching” and no evidence of an increase in consumer awareness. In addition, we did not find evidence of targeted government campaigns driving consumer reactions. As documented by the Ärzte Zeitung (2008 Physician’s Magazine), the only healthcare government campaign during that time period did not even mention “add-on premiums” or “health plan switching” but solely praised the advantages of the generous German healthcare system.
Second, the separate billing of add-on premiums might have contributed to the strong consumer response. While billing also could be interpreted as framing, the results of Section IV speak against billing as a main driver of the behavioral reactions: As shown, premium increases and decreases induced symmetric behavioral reactions. Premium refunds do not require any extra billing and come in the form of a direct deposit. Also note that refunds or add-on premiums did not alter the consumer information regarding prices. Pre- and post-reform, premium decreases and increases were announced in a personal letter to the enrollee.
Third, insurers could have altered their service quality or coverage of nonessential benefits as a supply-side response. Appendix D shows the absence of differential trends for both dimensions and health plans with and without add-on premiums.
Hence, fourth, the most likely trigger of the sharp increase in consumer price sensitivity was the increase in the price transparency and salience of premium differentials as a result of the change in price framing. Before 2009, to calculate their exact premium, employees needed to know (i) their exact monthly gross wage; (ii) the current contribution cap; (iii) the exact contribution rate that their insurer charged; and (iv) how their employee share of the total contribution rate was calculated. Although this was possible, in particular due to the wide availability of online calculators to compare price differences, the old price framing scheme was significantly less transparent than the new one.20
In addition, post-reform, enrollees were nudged to compare the premium differences in their plan to a clear reference price of zero (no add-on premium and no refund). Widely defining price framing as how premiums and premium differences are presented to consumers, the price framing reform consisted of two main elements: (i) expressing price differences in absolute euro values rather than percentage point payroll tax differences, and (ii) expressing these differences relative to a newly introduced standard price as set by the regulator. Because both elements were introduced jointly, we cannot disentangle the separate impact of either of these elements. However, both elements are relevant components of the overall price framing regime which, as we show, matters.
VI. Conclusion
This paper evaluates a price framing reform in the German public health insurance system. Germany has a unique employment-based, multipayer, universal coverage system. Due to historical developments, more than 100 non-profit health plans predominately engage in price competition. Heavy federal regulation determines a generous essential benefit package as well as identical cost-sharing amounts and provider reimbursement. Provider networks are unknown and the free choice of providers exists.
Prior to the reform evaluated here, health plan prices were labeled in the form of a mandatory payroll tax rate (the “contribution rate”). Each health plan set its contribution rate autonomously and independently. Hence, price differences between health plans were expressed in percentage point contribution rate differences. Pre-reform, monthly health plan premiums could easily differ by €50. Nevertheless, annual switching rates were low and approximately 5 percent of the enrollee pool switched in a given year. In 2009, in order to foster competition between insurers, a regulatory change standardized and fixed the contribution rate for all health plans. The law required health insurers to charge or refund differences to this standardized federal premium in absolute euro values. Thus, this natural experiment allows us to study consumer behavior under two different price framing regimes in a real-world setting. If enrollees were fully rational, the way in which premiums and premium differences are framed should not affect their switching behavior.
Using representative individual-level and health plan level panel data, we find that consumer price sensitivity significantly increased when premium differentials between plans became more salient. Post-reform, a €10 increase in the monthly premium— absolute or relative to the average market increase—tripled the switching probability from the baseline of 5 percent. We also show that enrollees who did not experience real price changes, but simply a change in the framing of their premiums, switched plans at significantly higher rates as compared to a control group. The general finding of a change in consumer behavior as a result of price framing is robust to a plenitude of specifications. For example, exploiting the universe of health plan enrollment data over ten years, we show that the most expensive plans (those which charged an €8 monthly add-on premium) lost 13 percent of their members post-reform. Pre-reform, comparable premium increases only resulted in enrollment decreases of 4 percent. We calculate a pre-reform demand elasticity of −1, which means, a 1 percent increase in health plan premiums decreased health plan enrollment by approximately 1 percent. Post-reform, this elasticity quadrupled to −4.
The most likely operating mechanism beyond the pricing reform goes back to the increased salience of price differences. However, not only did consumers become more price-responsive, we also find a decrease in the market price dispersion, and thus in potential savings. In contrast, after switching plans, the de facto realized savings remained very stable at approximately $120 per year. These findings strongly suggest that price competition intensified as a result of the price framing reform because price differences between plans became more salient.
Price transparency and consumer price sensitivity are essential to ensure health plan competition. Obviously, the manner in which prices are framed and presented to consumers is key for fostering such consumer price sensitivity and, thus, for designing competitive insurance markets. Moreover, price framing regulation is relatively costless, likely to gain bipartisan support, and easy to implement. The results of this study suggest that, to intensify price competition, differences in (health) insurance premiums would need to be expressed in absolute monetary values relative to a reference price set by the regulator.
Acknowledgments
The authors are very grateful for very helpful comments and suggestions by anonymous referees. They also thank Pio Baacke, Daniel Baumgarten, Dan Benjamin, Christine Benesch, John Cawley, Anna Choi, Damon Clark, Maria Fitzpatrick, George Georgopoulos, Markus Grabka, Dan Grossman, Tatiana Homonoff, Don Kenkel, Catherine Korachis, Martin Karlsson, Michael Kvasnicka, Sam Kleiner, Xianghong Li, Jean-Marie Lozachmeur, Sean Lyons, Thomas Mayrhofer, Nils aus dem Moore, Jordan Matsudaira, Andrea Podhorsky, Benjamin Protte, Ted O’Donoghue, Emily Owens, Imran Rasul, Steffen Reinhold, Wolfram Richter, Regina Riphahn, Ben Sand, Andreas Schmid, Piotr Spiewanowski, Barry Smith, Erdal Tekin, Nick Tilipman, Erik Schutt, Benjamin Schwab, Arthur van Soest, Stefano DellaVigna, Jürgen Wasem, Georg Weizsäcker, and conference participants at the 2012 Annual Meeting of the Association of German Health Economists (dggö) in Konstanz, the 2012 European Conference on Health Economics (ECHE) in Zurich, the 2012 Annual Congress of the European Economic Association (EEA) in Malaga, the 2012 Annual Conference of the European Society for Population Economics (ESPE) in Bern, the LAGV#11 Conference in Public Economics in Marseille, the 2012 Health. Development. Inequality Conference in Darmstadt, the 2012 Annual Conference of the Royal Economics Society (RES) in Cambridge, the 2012 International SOEP Use Conference in Berlin, the 2012 Annual Conference of the Society for the Advancement of Behavioral Economics (SABE) in Granada as well as seminar participants at the University of Augsburg, University of Dortmund, University of Duisburg-Essen, DIW Berlin, FU Berlin, University of Erlangen-Nuremberg, DICE Düsseldorf, the DIBOGS workshop in Hamburg, the Berlin Network of Labor Market Research (BeNA), York University, the Behavioral Economics Seminar at Cornell, and The Institute on Health Economics, Health Behaviors and Disparities at Cornell University. We take responsibility for all remaining errors and shortcomings of the article. The authors also thank Peter Eibich and Stefanie Thönnes for excellent research assistance and Jennifer Rontganger, Adam Lederer as well as Philip Susser for excellence in editing this paper. The research reported in this paper is not the result of a for-pay consulting relationship. Our employers do not have a financial interest in the topic of the paper that might constitute a conflict of interest.
The data used in this article can be obtained beginning six months after publication through three years hence from Hendrik Schmitz, University of Paderborn, CINCH, RWI, Warburger Str. 100, D-33098 Paderborn, Germany, phone: +49-5251 60 3213, email:hendrik.schmitz{at}uni-paderborn.de.
Appendix A
Overview of Sickness Funds with Add-On Premiums and Refunds (2012) and Their Pre-Reform Contribution Rates (2008)
Descriptive Statistics for Individual-Level SOEP Data
Enrollees of Add-On Premium Health Plans Do Not Differ from Enrollees in Other Plans
More Robustness Checks
Table A4 shows additional robustness checks for our main specification Premium Increase (relative to market increase). In Column 1 we run a probit model instead of a linear probability model and report marginal effects (averaged over all observations). The estimate is only slightly smaller than the baseline estimate in Column 4 of Table 3.
Column 2 restricts the sample to those enrollees who are compulsorily insured within the public health insurance system. In other words, we identify and exclude enrollees who could have selected themselves out of the treatment and could have opted out of the public system in reaction to the reform. Both point estimates, pre- and post-reform, are within the same confidence intervals as the baseline estimates although the pre-reform coefficient is no longer significant.
In Column 3, we only consider the years 2007 to 2011 to test whether long-term switching trends might affect the results. This exercise also provides evidence of whether serial correlation in case of long time horizons affects the estimated standard errors. The size and significance of the coefficients is comparable with our baseline estimates.
Column 4 excludes all socio-economic controls.
In Column 5, we additionally include lagged interaction terms. The results suggest that, for the likelihood to switch between t and t + 1, price increases between the interview in t−1 and the interview in t are as important as price increases between t and t + 1. Note that the specification does not alter the coefficients of our standard price increase indicator. However, it shows that—at least for some enrollees—price changes may require some time to diffuse before they lead to behavioral actions. Again, this lagged effect is stronger under the new framing regime than under the old one, reinforcing the interpretation of a higher salience of price increases after the reform, even for those who need more than one year to react.
Column 6 includes health plan fixed effects, and Columns 7 and 8 stratify the effects by the median equivalized household income. Section III.E discusses the latter results.
Robustness Checks
Appendix B
Aggregated Administrative Data
Descriptive Statistics for Aggregated Administrative Data
Appendix C
Google Search Rates
Figure C1 plots Google search rates for the terms “switch health plan,” “compare health plans,” and “cancel health insurance. ” The search was conducted with these keywords in German. Although we see several search peaks, there is clearly no break or trend in the search rates after the reform became effective. We rather observe a downward trend in search rates over a long time period from 2004 to 2011, which is reassuring and suggests that there was no increase in general awareness.
Google Search Query Rates—“Health Plan Switching” and Similar Terms
Source: Google Insights for Search. Linear trend is for the keyword “switch health plan. ”
Appendix D
Testing for Systematic Changes in Health Plan Service Quality
Since 2008 Focus Money, a German magazine, has provided an annual service quality ranking of sickness funds. We make use of these data in Table D1. The total score and ranking comprises 11 subcategories such as “supplementary health insurances offered,” “transparency,” “extra benefits offered,” “number of branches,” “call center service quality,” and “general service quality. ”We collected and digitized all Focus Money quality rankings and use the total scores and rankings of the health plans as outcome variables in the models in Table D1. All models can be interpreted as simple DiD estimates: Health plans that charged add-on premiums post-reform are in the treatment group and all other plans are in the control group. We include a full set of year dummies and interact a postreform indicator with the treatment group indicator.
In total, Table D1 carries out two main empirical tests: (i) whether more expensive health plans systematically offer higher (or lower) service quality or optional benefits, and most importantly, (ii) whether expensive health plans that (had to) charge more salient add-on premiums post-reform varied their quality or optional benefits systematically. Whereas the first test poses no real threat to our identification strategy, which is based on exogenous premium variation and the price framing reform, point (ii) is important. If expensive health plans whose premium differences became more salient had systematically extended their coverage or service quality in the post-reform period, we would not be able to disentangle the price framing effect from the quality effect.
Is Service Quality Correlated With Price Framing?
Table D1 shows that (i) plans that charged an add-on premium post-reform did not differ systematically in quality. In all four models, the time-invariant Treatment Group coefficients are not statistically significant and showno systematic sign. In addition, and as the only crucial condition for our price framing identification strategy, (ii) the change in framing was orthogonal to health plan quality. The four Treatment Group×postreform interaction terms are small in size across all four model specifications and the signs of the coefficients are ambiguous.
Footnotes
↵1. See, for example, Dowd and Feldman (1994), Buchmueller and Feldstein (1997), Cutler and Reber (1998), Royalty and Solomon (1999), Strombom et al. (2002), Atherly et al. (2004), Buchmueller (2006), Bundorf et al. (2012), Carlin and Town (2009), Chan and Gruber (2010), Lustig (2011), Gaynor and Town (2012), Handel (2013), Buchmueller et al. (2013), Starc (2014), Marzilli Ericson and Starc (2015), and Moriya and Simon (2015). Examples for non-U.S. literature on premium elasticities and consumer switching behavior are Schut and Hassink (2002), Andersen et al. (2007), Schut et al. (2003), Dijk et al. (2008), Frank and Lamiraud (2009) or Ortiz (2011). Guthmuller et al. (2014) study health insurance takeup in France. Miravete and Palacios-Huerta (2014) and Grubb and Osborne (2015) study consumer inertia in the market for cellular phone plans.
↵2. However, service quality is of limited relevance in the German system because enrollees typically do not interact much with their insurers. Insurers mainly process paperwork and pay providers.
↵3. Contribution caps existed pre- and post-reform. They are adjusted every year by the federal regulator. For example, in 2008, the cap was €42,300 ($54,000) and increased by approximately 2 percent per year to €49,500 ($61,875) in 2015.
↵4. The idea of the reform goes back to Richter (2005). His proposal aimed at reducing the implicit subsidy of low-income individuals by income-dependent contribution rates. Richter (2009) wanted to incentivize especially low-income individuals to switch to cheaper plans and to foster price competition between plans. The general political intention of the government that implemented the reform was to increase price sensitivity among all income groups in order to foster competition between sickness funds. As we will show below, switching rates and price elasticities increased significantly for all income groups.
↵5. The € 50 refers to the 50 percent employee share of the total premium of (17.4 percent–13.5 percent) × €2,552 = €49.76.
↵6. This excludes all those insured under family insurance, the unemployed—social security pays the health insurance premiums for some of these individuals, full-time students who pay an income-independent flat premium (2015:€66,33 per month) or they are insured under their parents’ family plan, pensioners, and special population groups, such as draft soldiers or low-income earners. Individuals with a wage of less than €800 per month do not have to pay the full contribution rate. The main reason to exclude retirees is that their income (subject to social security contributions) cannot be accurately measured in the SOEP. In addition, when relying solely on SOEP information, it is not always clear whether retirees have to pay the entire health insurance premium or only part of it. Evaluating the same reform using the SOEP, Wuppermann et al. (2014) focus solely on retirees.
↵7. We disregard the employers’ share because employees typically believe that they only pay the employee share. Hence, we assume that employees base their switching decision predominantly on their share of the premium. To the extent that employers pass through lower premiums in the form of higher wages and employees are aware of it (or believe that this was the case), our estimates below simply represent a lower bound.
↵8. Note that, while we can (almost) perfectly identify the initial health plan of all SOEP respondents, we have to rely on a subset for the “realized savings” exercise. The reason is that many enrollees switch from large sickness funds that are directly indicated in the SOEP questionnaire to smaller, cheaper funds that respondents indicate in handwritten form. Hence, the assumption in Panel B of Table 1 is that, for switchers, identified destination health plans are representative of all destination health plans.
↵9. When sickness funds merge and enrollees remain enrolled in the new fund, we treat them as not having switched. Only if enrollees switch to an entirely different health plan do we treat them as switchers. Premium changes after a merger are coded as premium changes.
↵10. Results in the Appendix show that probit models with marginal effects deliver similar results.
↵11. In the second part of the empirical analysis (Section IV), where we make use of health plan level data, we also assess the effect of price decreases.
↵12. The regression results are robust to fixing labor income in t + 1 instead of t. The robustness check is available upon request.
↵13. Note that we cannot unambiguously assign premium increases in 2009. This is because we only observe whether enrollees switched between the beginning of 2008 and the first months of 2009. However, we do not observe exactly when individuals switched and whether this happened before or after January 1, 2009. Consequently, we drop the year 2009 (meaning switches between 2008 and 2009). However, including 2009 and imposing assumptions does not alter the results.
↵14. The average increase includes zeros for all sickness funds that did not increase premiums.
↵15. Please note that enrollees in Treatment Group 2 may have experienced minor changes in their monthly euro premiums. Due to differences in wages, it is unlikely that the pre-reform premium was exactly replicated postreform. Applied to the average monthly gross wage ofV2,500, a monthly add-on premium ofV8 would equal a 0.6 percentage point higher contribution rate pre-reform. Thus, all employees with an average wage and a prereform rate of more than 16.1 percent experienced a slight decrease and the rest a slight increase in prices.
↵16. The smaller bandwidth specification in Column 2 uses only 1,527 observations. Note that, in Column 2, the size of Treatment Group 2 increases. This is because those with [15.6 percent <= contribution rates < 15.7 percent] now belong to Treatment Group 2, but belonged to Treatment Group 1 in Column 1.
↵17. Results are available upon request.
↵18. Note that multiplying this figure with the market share yields the change in market shares as displayed by the Y-axis in Figure 1.
↵19. PremiumEuro uses average annual sickness fund contribution rates, the annual contribution ceiling and the average annual German gross wage to calculate the average monthly health plan premiums, assuming that average wages are identical across all sickness funds.
↵20. Note that the monthly paycheck typically displays the premium in euros. However, in order to carry out a comparison of individual premiums between different plans, enrollees were forced to do the math which required the knowledge of (i) to (iv).
- Received August 2014.
- Accepted July 2015.
