The Role of Prices Relative to Supplemental Beneﬁts and
Service Quality in Health Plan Choice ‡
Christian B ¨
unnings ∗Hendrik Schmitz †Harald Tauchmann ‡
Nicolas R. Ziebarth §
May 19, 2017
This paper links representative enrollee panel data to health plan data on (i) prices, (ii) service qual-
ity, and (iii) non-essential beneﬁts for the German statutory multi-payer market and the years 2007 to
2010. We ﬁrst show that, although heavy federal regulation ensures a simple choice architecture, the
majority of health plans are dominated—even when considering four non-price attributes. Enrollees
in dominated plans are older, less educated, and unhealthier. Second, we assess how switchers value
prices relative to non-price health plan attributes. Our mixed logit models incorporate a total of 1,700
health plan choices with each more than 50 choice sets. While prices are an important determinant for
nearly everyone, 40% of all switchers do not seem to value service quality and supplemental beneﬁts
when choosing health plans.
Keywords: dominated health plans, market frictions, choice determinants, service quality, non-essential
beneﬁts, health plan switching, German sickness funds, SOEP
JEL classiﬁcation: D12; H51; I11; I13; I18
3We would like to thank Dean Lillard, Normann Lorenz, Janina Nemitz, Carsten Schr ¨
oder and participants of
the iHEA 2014 in Dublin and the 2016 SOEP User Conference in Berlin for valuable comments and suggestions. We
also would like to thank Thomas Adolph for providing us with the data on health plan characteristics and Philip
Susser and Eric Maroney for editing this work. Financial support from the BMBF (F ¨
is gratefully acknowledged. We take responsibility for all remaining errors in and shortcomings of the article.
∗University of Paderborn, FOM Hochschule and RWI, e-mail: firstname.lastname@example.org
†University of Paderborn (Department of Economics), CINCH and RWI, Germany, phone: +49 5251 60 3213, e-mail:
at Erlangen-N ¨
urnberg, Department of Economics, phone: +49 (0)911 5302 635, e-mail: har-
§Cornell University, Department of Policy Analysis and Management (PAM), 106 Martha Van Rensselaer Hall, Ithaca, NY
14850, USA, phone: +1-(607)255-1180, fax: +1-(607)255-4071, e-mail: email@example.com
When it comes to regulating health insurance markets, policymakers around the world debate the fun-
damental question of how much choice would be optimal for consumers. On the one side of the spec-
trum are single-payer markets like in Canada or the UK which, except for supplemental private in-
surance, do not provide any choice. On the other side of the spectrum is the US where, despite an
increasingly stringent regulation, private insurers offering many different health plans still dominate
the scene. While the US has been increasingly moving into the direction of more insurance regulation,
other countries already run multi-payer systems that combine heavy state regulation with consumer
choice. Examples are Switzerland, the Netherlands and Germany.
Recent economic research also surrounds the question of how much regulation would be optimal
for consumers. A rich stream of papers analyzes the Medicare Part D market in the US (supplemen-
tal drug insurance for the elderly) and ﬁnd that many consumers do not choose the health plan that
would minimizes their out-of-pocket spending (e.g. Heiss et al.,2006,2013;Abaluck and Gruber,2011).
Basically, out-of-pocket spending is a function of (uncertain) demand for health care, deductibles, co-
insurance rates, copayments and stop-loss limits. Because many consumers seem to have difﬁculties to
understand complex insurance products (Loewenstein et al.,2013;Abaluck et al.,2015;Karaca-Mandic
et al.,2017), researchers have studied why consumers “leave money on the table”—explanations range
from general inertia or market frictions, to inattention, heuristic decision rules, and switching costs (Ce-
bul et al.,2011;Ericson and Starc,2012;Kling et al.,2012;Handel,2013;Ericson,2014;Ho et al.,2017;
Ericson and Starc,2015;Handel and Kolstad,2015).1Other papers have suggested—as well as rejected—
that consumers learn and that their choices improve over time (Ketcham et al.,2012,2015;Abaluck and
Gruber,2016). In one of the few non-Medicare papers, Bhargava et al. (2017) use data from a big US em-
ployer to show that most employees choose ﬁnancially dominated health plans with excess spending
amounting to almost half of the employee share of the premium.
This paper is based on uniquely compiled panel data that allows us to cast light on the choice sets
and health plan architecture of the German statutory multi-payer market. We link representative en-
rollee survey data to publicly available data on health plan prices and standardized quality informa-
tion. Then we exploit annual changes in these health plan characteristics across 120 plans and over
4 years. The standardized supply-side information stems from a private company that surveys and
ranks all German health plans. Thus, the empirical approach exploits the same health plan informa-
tion that German consumers can access in online portals and magazine rankings to select health plans.
In addition to prices, we make use of two non-essential beneﬁt indicators and two service quality in-
1There exists also a rich US literature on health plan choice that is not focused on irrational behavior (Dowd and Feldman,1994;
Cutler and Reber,1998;Royalty and Solomon,1999;Strombom et al.,2002;Atherly et al.,2004;Buchmueller,2006;Dusansky and
Koc¸,2010;Bundorf,2010;Dusansky and Koc¸,2010;Parente et al.,2011;Buchmueller et al.,2013). Excellent literature overviews
are provided by Kolstad and Chernew (2009) and Gaynor and Town (2012).
dicators. Non-essential beneﬁts are voluntarily provided on top of the generous German mandated
beneﬁt package. Examples of non-essential beneﬁts include ayurveda, homeopathy, osteopathy, urine
therapy, preventive check-ups, or speciﬁc immunizations. Service quality is measured by the density of
the brick-and-mortar branch network (where face-to-face customer service is provided but no medical
services) and the quality and accessibility of information via telephone hotlines and the internet.
The main theme of the paper is to study the role of prices as compared to these non-price health
plan attributes from a demand and supply-side perspective. The existing literature largely focuses on
prices, cost-sharing and ﬁnancial health plan attributes. Our descriptive supply-side analysis suggests
signiﬁcant market frictions between the 120 plans, but also demonstrates a clear trade-off between price
and non-price factors. We show that a large share of health plans are in fact not only ﬁnancially domi-
nated, but also dominated when considering the multi-dimensional attributes price, service quality, and
non-essential beneﬁts. Our demand-side analysis replicates over 100K choice options from 1,700 actual
plan choices with each more than 50 plans to choose from. Using mixed logit models, we ﬁnd that that
prices are the dominant choice determinant for the majority of consumers when switching plans. How-
ever, preference heterogeneity is large—a small share of consumers value non-price attributes such as
the quality of hotline services highly.
The paper makes several contributions to the literature. First, it is one of the few studies that com-
prehensively investigates the supply and demand-side of a multi-payer market outside the US.2The
German market has attractive features; due to its heavy federal regulation, many potential confound-
ing factors when studying health plan choice are shut down. Speciﬁcally, Germany has no provider
networks and reimbursement rates are centrally determined. Deductibles and co-insurance rates are
prohibited; only small and uniform co-payments apply to all plans. About 120 sickness funds (=health
plans3) compete for 90 percent of the population whose large majority is mandatorily insured. Many
health plans operate nationwide and, depending on the state of residency, the publicly insured can
choose between 40 and 70 health plans. As a comparison: In the US non-group market, which is or-
ganized at the state level around the so called ‘Exchanges’, on average 50.9 plans were offered by 3.9
insurers in 2014 (Dafny et al.,2015). The Medicare Part D market and the Swiss health care market also
offer comparable choice sets for consumers.
Second, in addition to prices, we exploit four consistently surveyed non-price attributes outlined
above (two for service quality and two for non-essential beneﬁts) and representative enrollee informa-
tion over four years. This allows us to cast a complete picture of the entire market representing a total
of 70 million insured.
2Schut and Hassink (2002) and Dijk et al. (2008) study the Netherlands, Frank and Lamiraud (2009) the Swiss market, and
Schmitz and Ziebarth (2017) as well as Wuppermann et al. (2014) the effects of a price framing reform in Germany. Christiansen
et al. (2016) studies switching in the private long-term care market in Germany.
3We use the terms “health insurance (company),” “sickness fund,” and “health plan” interchangeably.
Third, we show that substantial market frictions even exist in heavily regulated markets that sim-
plify the consumer optimization problem substantially. Between 70 and 90 percent of all plans in the
German statutory health insurance market are dominated, even when taking four non-price attributes
(plus prices) into account.4Just focusing on the one-dimensional price dimension (absent non-linear
cost-sharing), the share of dominated plans even increases to 95 percent. In line with Bhargava et al.
(2017), we ﬁnd that the stock of enrollees in dominated plans is older, less educated, and less likely to be
in very good self-reported health (as compared to enrollees in dominating plans). Switchers who chose
the absolute cheapest plan in the market are healthier and better educated.
Finally, as suggested by the title, we study the importance of prices as compared to the other four
non-price attributes from a supply and demand perspective.5We show that some plans, which would
clearly be dominated when just focusing on the price dimension, differentiate their product along the
non-price dimensions, and that there exists a clear trade-off between the different plan attributes. How-
ever, the selective subset of enrollees who actively switch plans seem to mostly value prices as the
dominating demand-side choice factor. While non-switchers in plans that diversify along the non-price
dimensions may not lack from inertia but simply value these quality characteristics, we can show that
consumers in “fully” dominated plans do leave money on the table due to inattention, inertia, or very
high switching costs. Overall, our analysis illustrates the importance of incorporating non-price at-
tributes into the analysis of consumer choice and inertia.
The remainder of the paper is organized as follows: The next section covers the institutional details
of the German public health insurance market. Section 3presents the data. The supply-side analysis is
in Section 4and Section 5contains the analysis of the demand-side. Section 6concludes.
2 Institutional Background
The German health insurance system is characterized by the coexistence of statutory health insurance
(SHI) and substitutive private health insurance (PHI). This paper focuses on the SHI, which covers
roughly 90% of the population most of whom are compulsorily insured. Insurance under the SHI is
mandatory for employees with gross wage earnings below a deﬁned threshold (in 2016: e56,250/$62K
per year). Nonworking spouses and dependent children under 25 years are covered free of charge by
SHI family insurance. Further regulations apply to speciﬁc groups of the population, such as students
and the unemployed, although most of them are covered by SHI. High-income employees, the self-
4By “dominated” we mean that at least one competitor dominates a plan in at least one of the ﬁve dimensions without scoring
worse in any other dimension.
5Not many existing papers explicitly study the role of quality information in the decision to choose health plans. Scanlon et al.
(2002) and Chernew et al. (2008) study the impact of quality report cards using US data from General Motors; Beaulieu (2002),
Wedig and Tai-Seale (2002), Dafny and Dranove (2008), Jin and Sorensen (2006), and Abraham et al. (2006) study health plan
ratings in “natural” US settings, whereas Harris (2002) conduct a discrete choice experiment in West Los Angeles to conclude that
large quality differences would be required for consumers to accept provider access restrictions. Kolstad (2013) studies the impact
of quality report cards for cardiac surgery on surgeon’s behavior.
employed and civil servants may opt out of SHI and buy substitutive PHI or remain publicly insured as
Currently, the SHI market consists of about 120 not-for-proﬁt health insurers, also called “sickness
funds”, roughly half of which are operating nationwide, while the remaining ones solely operate in
some federal states. Due to historical reasons, see below, each sickness fund basically only offers one
standard health plan, which is why we use the terms sickness fund and health plan interchangeably.
Switching sickness funds is uncomplicated: the minimum contract period is 18 months and there
is no enrollment period; guaranteed issue exists and several speciﬁc search engine websites help con-
sumers to compare and switch health plans. When sickness funds increase prices, enrollees have an
extraordinary right to cancel the contract and switch funds. In a given year, only about ﬁve percent of
all SHI insured switch plans (Schmitz and Ziebarth,2017) but from 1996 (when free health plan choice
became a legal right) to 2002, about a quarter of all policy holder had switched at least once (Pilny et al.,
2.1 Essential Beneﬁts and Cost-Sharing Regulation
About 95 percent of the SHI beneﬁt package is predetermined by social legislation at the federal level.
The federally mandated essential beneﬁt package is generous relative to international standards. It
basically includes all medically necessary treatments in addition to prescription drugs, birth control,
preventive and rehabilitation care as well as rest cures (c.f. Ziebarth,2010a).
German social legislation additionally heavily restricts cost-sharing. Deductibles and co-insurance
rates are prohibited in SHI and the law stipulates that only small copayments, which do not vary across
health plans, can be charged. Hence variation in cost-sharing parameters is not part of the objective
function when enrollees search for new health plans.
2.2 Price Regulation
Health plan premiums are charged in form of social insurance contributions. To calculate the employee
share of the premium, a sickness fund speciﬁc contribution rate is applied to the gross wage, including
all fringe beneﬁts, up to a deﬁned contribution ceiling (in 2016: e50,850 per year). One half of the
contribution rate is formally paid by the employee and the other half by the employer.
In January 2009 and as part of a health policy reform (GKV-Wettbewerbsst¨
framing in the SHI system was reorganized. Pre-reform, contribution rates were set individually by
each sickness fund, resulting in a variety of contribution rates, ranging from 12.2 to 16.9% of individ-
uals’ gross wages. The reform equalized the contribution rates across all health plans. Post-reform,
if allocated revenues generated by the 15.5% standardized contribution rate did not cover the funds’
expenses, sickness funds had to charge an additional monthly euro add-on premium. In contrast, if
allocated revenues exceeded expenses, sickness funds could reimburse their members a monthly bonus
payment. For our demand-side analysis, to make them comparable, we express health plan premiums
in monthly euro amounts—pre- as well as post-reform.
Because it has been shown that the price framing reform increased the likelihood to switch health
plans signiﬁcantly (Schmitz and Ziebarth,2017;Wuppermann et al.,2014), we exploit the reform in a
robustness check to correct for endogeneity in the switching decision. In the main analysis, we delib-
erately allow for selection into the switching decision in order to characterize the subset of switchers
empirically (c.f. Einav et al.,2013).
2.3 Non-Price Product Differentiation
To differentiate their product, in addition to engaging in price competition, sickness funds can add
supplemental non-essential beneﬁts to their beneﬁt package, or improve the quality of their customer
service. We extract two non-essential beneﬁt measures and two service quality measures to investigate
how these non-price attributes are traded-off by decision makers on the supply and demand-side.
Optional Non-Essential Beneﬁts
These are measured by one variable for (i) alternative medicine, and one variable for (ii) other non-
essential beneﬁts. Alternative medicine consist of complementary treatments such as ayurveda, home-
opathy, osteopathy, and urine therapy. Although the effectiveness of alternative medicine is discussed
controversially, there exists a steady demand for such treatments in Germany.
In addition to alternative medical treatments, sickness funds may also offer conventional “non-
essential” medical treatments. Examples are preventive check-ups (e.g. the ’J2’ check-up for adolescents)
or certain types of immunizations (e.g., malaria prophylaxis). Typically these beneﬁts have rather low
monetary values, e.g., a single combined vaccination shot against diphtheria and typhoid fever costs
about e15. However, non-essential beneﬁts may also comprise more expensive medical treatments,
such as coverage for in-vitro fertilizations.
Service quality is measured by accessibility of insurers and the quality of information provided. Most
sickness funds still operate a brick-and-mortar network of physical branches but also offer hotline ser-
vices. Operating a large number of physical branches may be preferable to (some) members—e.g., the
elderly—but also implies higher operational costs. To cut costs, some sickness funds have reduced their
branch network signiﬁcantly over time. Other funds do not run any brick-and-mortar stores but are
exclusively available by telephone or online (Direktversicherer).
Better customer service and accessibility by phone or the internet was part of the Health Care Re-
form Bill 2000 (GKV-Gesundheitsreformgesetz). The bill required all sickness funds to improve their
customer service and the quality of their consulting. As a result, sickness funds started to operate dif-
ferent types of hotlines. While some hotlines are fairly general, others provide detailed information
by trained health care professionals such as information about preventive care or drugs and their side
2.4 The German Risk Adjustment Scheme (RSA)
At its inception, in 1994, the German RSA was only based on the three factors age, gender, and dis-
ability status (c.f. Pilny et al.,2017). Today, the RSA also considers 80 different chronic (and expensive)
diagnoses, which is why it is sometimes called ’Morbi(dity)-RSA.’
In principle, the RSA works as follows: Employers automatically deduct the standardized contri-
bution rate (14.6% as of writing) from their employees’ gross wages. The total sum is collected in a
central Health Care Fund (Gesundheitsfond) which is administered by the Federal Insurance Agency
(FIA, Bundesversicherungsamt). The FIA then carries out the risk-adjustment based on the factors age,
gender, disability and 80 diagnosed diseases. Based on age and gender, enrollees are grouped into 40
age-gender groups. And based on 80 diagnoses, enrollees are categorized into 106 hierarchical morbid-
ity groups. Then, via regression models, standardized and risk-adjusted payments for providing health
care to SHI members in each of these cells are calculated. Finally, the FIA pays out the total sum of
risk-adjusted payments based on actual enrollee characteristics to each sickness fund.
If sickness funds cannot cover their total costs using the allocated capitated payments, they have
the option to (i) increase prices, or to (ii) reduce expenditures. The latter can be achieved by managing
the health care needs of their enrollees better (e.g., by providing preventive care, education, or special
bonus programs such as disease-management programs), by reducing customer service, by reducing
non-essential beneﬁts or by becoming more efﬁcient and cutting unnecessary administrative costs.
A constant criticism of the RSA has been the claim—supported by ofﬁcial reports (Buchner and
Wasem,2003;IGES,2004;Wissenschaftlicher Beirat zur Weiterentwicklung des Risikostrukturausgleichs
beim Bundesversicherungsamt,2011;Pilny et al.,2017)—that the capitated payments for sick enrollees
would not be sufﬁcient to cover their expenditures. In other words, the RSA would be incomplete and
there would still exist incentives for insurers to cream-skim. In an audit study, Bauhoff (2012) ﬁnd some
(limited) evidence that insurers try to actively cherry-pick good risks.
In the empirical analysis, in a robustness check, we will use changes in insurer risk pools as in-
struments for changes in prices. We construct our instruments by aggregating enrollees’ age and Self-
Assessed Health (SAH) up to the yearly plan level (excluding switchers’ contribution). As we will see,
changes in risk pools are highly predictive of price changes (but not of non-price attributes), yielding
support for the notion that the German RSA incompletely adjusts health risks across plans and that in-
surers primarily adjust price margins in the short-run. This is in line with Pilny et al. (2017) who show
that between 42 and 75 cents per additional euro revenue are passed through to lower prices (or charged
in form of higher prices when costs increase).
3.1 Individual Level Data
The German Socio-Economic Panel Study (SOEP) provides us with individual-level panel data. The
SOEP is a representative longitudinal survey that started in 1984 and collects annual information at the
household and at the individual level. Currently, the SOEP comprises more than 20,000 individuals
from more than 10,000 households (Wagner et al.,2007). We use the waves 2008 to 2011.
We use information on enrollees’ current health plan and insurance status. First, we exclude those
individuals who are covered by PHI.
Second, we restrict the sample to SHI policyholders, not the total number of insured. The latter
would also include family members insured at no cost under SHI family insurance. We focus on the
paying members to obtain exactly one observation per health plan choice decision. Paying members are
policyholders who gainfully employed and earn more than e400 gross per month.6
[Insert Table 1about here]
Third, to ensure that the empirical results are not driven by the high degree of state dependence—
only ﬁve percent switch health plans every year—the main part of the demand-side analysis deliberately
focuses on health plan switchers and those who leave the family insurance because they are not eligible
any more, not the stock of enrollees. There are several advantages to focusing on the subsample of
switchers. (a) Switchers likely inform themselves about the existing health plan options and trade-offs.
(b) Similar to Einav et al. (2013), we argue the we deliberately focus on individuals who switch plans
6This excludes all those insured under SHI family insurance, the unemployed for some of whom social security pays the
health insurance premium, full-time students who just pay an income-independent ﬂat premium (2016: e66,33 per month) or
who are insured under their parents’ family insurance, pensioners as well as special population groups, such as draft soldiers or
to be able to analyze their characteristics and study adverse selection. (c) In a robustness check, we
use the price framing reform of 2009, which exogenously increased the likelihood to switch, to study
the empirical relevance of endogenous switching. In addition to its external validity for the policy-
relevant subgroup of enrollees who actively switch plans, the ﬁndings likely also have validity for those
who express a willingness to switch plans. A representative survey among SHI enrollees (as of 2006)
revealed that almost half of all respondents ever switched plans and that almost one ﬁfth is currently
considering switching plans (Zok,2006).
Our ﬁnal estimation sample consists of 1,726 health plan choices from 1,594 different individuals.
We precisely replicate individuals’ choice sets. To do so, we consider enrollees’ state of residence because
some plans only operate in speciﬁc states. Panel A of Table 1shows that the average choice set includes
58 sickness funds. The minimum number of plans to chose from is 41 and the maximum is 73. Choice
sets are smallest for those living in Mecklenburg-Western Pomerania, ranging from 41 to 53 between
2008 and 2011. Individuals residing in North Rhine-Westphalia have the largest choice sets, ranging
from 55 (2011) to 73 (2008).7
Overall, the empirical identiﬁcation relies on 400 to 500 observed health plan choice per year, with
choice sets between 50 to 60 health plans. Thus, we observe about 100,000 total plan options over the
four years under consideration. We use the self-reported gross wages to calculate health plan premiums
for all potential choices in euro amounts for every year. Overall, the empirical identiﬁcation relies on
400 to 500 observed health plan choice per year, with choice sets between 50 to 60 health plans. Thus,
we observe about 100,000 total plan options over the four years under consideration. We use the self-
reported gross wages to calculate health plan premiums for all potential choices in euro amounts for
SOEP interviews are typically carried out in the ﬁrst quarter of the year, while sickness fund char-
acteristics were collected at the end of a calendar year in November and December. Thus we link the
respondents’ health plan choice sets at the time of the interview in the ﬁrst months of a year with the
health plan information as provided at the end of the previous calendar year. In the interviews at the
beginning of the year, enrollees indicate their current sickness fund and whether they had switched in
the course of the previous year. Consequently, we make use of switching and health plan data for the
years 2007 to 2010.
Panel B of Table 1shows the demographics for our sample of switchers: 55 percent are female and
the average age is 37 years. Thirteen percent self-rate their health as “very good” and 49 percent as
7Due to several mergers of sickness funds, the number of active health plans is decreasing over time.
3.2 Health Plan Level Data
Health plan characteristics (contribution rate, non-essential beneﬁts and service quality) are provided
by a private company (Kassensuche GmbH). The information are collected via questionnaires that are
sent out annually to all existing sickness funds. Sickness funds have a strong incentive to participate in
the survey, as Kassensuche GmbH operates a large German web portal where consumers can compare
a broad range of characteristics across all sickness funds. Moreover, at the end of each year, a popu-
lar weekly business magazine (Focus Money) publishes a detailed overview and ranking of the best 50
sickness funds that were surveyed by Kassensuche GmbH. This sickness fund ranking includes sub-
scores for several subcategories, measured on continuous scales. These subscores provide the basis for
the beneﬁt and service quality measures used in this study. Note that the information exploited in this
study is identical to the information provided to consumers. Being able to directly exploit its variation
is one main advantage of our approach.
The main drawback of using these data is the interpretation of the (sub)scores which is not straight-
forward as we discuss below. We use the subscore data as provided by Kassensuche GmbH because we
are unaware of the algorithm that had been used to calculate the subscores. Hence, we cannot directly
interpret the subscores and check, for example, how a higher density of the brick-and-mortar network
translates into a higher score.
To account for minor differences in the calculation of the sub-scores over time, the regression models
use the z-transformed subscores.8In total, we have information on 115 different sickness funds covering
the years 2007 to 2010. The health plans that are included every year have a total market share of around
80 percent and also represent 80 percent of all existing plans (M¨
uller and Lange,2010).
Annual prices for all German sickness fund are publicly available from M¨
uller and Lange (2010), the
National Association of Statutory Health Insurance Funds, and the annual reports of the sickness funds.
Post-2009, when contribution rates were uniﬁed, we also consider fund’s add-on premiums and refunds
(see Section 2). To calculate the exact monthly premiums (in euros) for each enrollee, we link plan
speciﬁc contribution rates with individuals’ gross wages and consider the federal contribution ceiling
of each given year.9
We ignore the employer share of the premium, which is legally ﬁxed at 50 percent of the total pre-
mium10, simple because employees likely make decisions solely based on their share of the premium.
8The z-transformation is carried out for each year separately. The number of subcategories has slightly changed over time,
therefore we use only those subscores which were part of the survey in each of the four years.
9The contribution ceiling is determined by the federal regulator and has been increasing every year by about two percent.
10Effective July 1, 2005, the strict equal sharing of contributions was altered. Between 2005 and 2015, the employees’ share was
[0.9 +0.5 ×(cr −0.9)] percent of their gross wage up to the contribution ceiling, where cr denotes the overall contribution rate. In
the example above, this amounts to an employee share of 7.45 percent and an employer share of 6.55 percent of the gross wage.
As seen in Panel C of Table 1, hypothetical average employee shares (assuming monthly gross wages
of e2,000) range between e133 and e178, with a mean value of e153. To the extend that employees
base their health plan choices on the full premium, we obtain a lower bound for the price effect in the
regression models below.
Two indicators measure health plan service quality: (i) hotline service and (ii) brick-and-mortar net-
work.Hotline service considers the different types of available hotlines (medical, non-medical) and
how many hours these hotlines are staffed. Differences in staff quality—e.g., the share of staff with
special qualiﬁcations such as social insurance clerks, physicians, nurses or pharmacist—are accounted
for by weighting the hotline’s operating hours accordingly, where higher staff quality receives a higher
The score of brick-and-mortar network measures the density of the brick-and-mortar network rel-
ative to the plan’s operating region. More precisely, the original score (before the z-transformation)is
derived from the log of the total number of brick-and-mortar stores divided by the number of federal
states in which the sickness fund operates.
Optional Non-Essential Beneﬁts
Two indicators also measure optional non-essential beneﬁts: (i) alternative medicine and (ii) other non-
essential beneﬁts. The score of alternative medicine is mainly based on the number of different alter-
native medical treatments offered by each sickness fund. Sickness funds are not entirely free to offer
any additional treatment, but have to choose from a list of approximately 20 approved treatments (e.g.,
ayurveda or homeopathy). The score also considers whether the treatments are restricted to certain
regions or physicians.
Other non-essential beneﬁts measures supplemental beneﬁts outside of the essential SHI beneﬁt
package (Kassenleistungen). Examples are certain immunizations (e.g., for tropical diseases) or preven-
tive screenings for breast or skin cancer in younger ages.
4 The Supply-Side of the German Health Plan Landscape
4.1 How Insurers Trade-Off Prices and Non-Price Attributes
The following analysis is based on our health plan data which contain prices and four non-price at-
tributes over four years. The purpose of this exercise is to investigate how insurers diversify health
plans and position themselves on the price vs. non-price dimensions. Since the z-transformation is
rather uninformative, Figures 1plot the original non-price scores, as observed and published by Focus
Money in their sickness fund ranking.
Prices vs. Service Quality
Figures 1a and b show the trade-off between prices and service quality as measured by hotline service
and brick-and-mortar network. One would expect a positive relationship between these plan char-
acteristics because fewer stores and hotlines imply lower (administrative) costs. Figure 1a visualizes
exactly this trade-off. The x-axis plots the sickness fund speciﬁc contribution rate, and the y-axis the
(non-transformed) hotline service quality score as calculated and published by Kassensuche GmbH.
We observe a clear positive relationship, where cheaper plans also exhibit signiﬁcantly less hotline
service quality (Figures 1a). The slope of a simple linear regression is 0.03 implying that for each point
increase of the hotline service score, on average, the contribution rate is 0.3ppt higher. This equals a
e3.82 higher monthly premium ($50 per year) when just considering the employee share of the pre-
mium. In the demand-side analysis below, we will investigate whether switchers are actually willing
to accept such a trade-off. Also note that, across the entire price distribution in the market, consumers
could easily choose plans that dominate others and get better hotline quality for less money. A detailed
analysis of dominated plans will also follow below.
[Insert Figure 1about here]
Figure 1b plots the trade-off between brick-and-mortar network and prices. Again, one observes a
clearly positive and signiﬁcant relationship between prices and the density of brick-and-mortar stores.
The slope here is even 0.073 implying that, on average, sickness funds with a one percentage point
higher physical store network score charge customers about $10 more per month. Here as well, at least
when just considering these two dimensions, we observe plans that clearly dominate others.
Prices vs. Optional Non-Essential Beneﬁts
Figure 1c shows a signiﬁcantly positive trade-off between other non-essential beneﬁts and the price.
The slope is 0.037 and the trade-off similar to the trade-off between hotline service and prices.
While all other non-essential beneﬁts and prices imply a signiﬁcant trade-off, this is not the case for
a very speciﬁc, supplemental beneﬁt: alternative medicine. As Figure 1d shows, the slope is ﬂat and not
statistically different from zero. This ﬁnding is plausible because the number of enrollees who actually
use alternative medicine is estimated to lie below ten percent (Robert Koch Institut,2002).
Changes in Prices and Quality over Time
Next, Figure 2shows the distribution of year-to-year changes in the health plan parameters price,hotline
service,other non-essential beneﬁts, and alternative medicine. Figure 2a shows that most health plans
increase prices over time. We observe a signiﬁcant share of plans that increased contribution rates by
up to 2ppt. As an example, increasing the contribution rate by 1ppt increases the employee share of
the monthly premium by e10, given the average gross wage of switchers in our sample. Figure 2a also
shows that very few plans decrease prices.
[Insert Figure 2about here]
Figure 2b illustrates changes in the hotline service quality score. The mass point is around zero and
one observes an almost symmetric distribution with some plans gaining up to 10 score points and some
plans loosing up to 10 score points. Loosing or gaining 10 score points actually represents a signiﬁcant
change, given that the overall scale ranges from 0 to 30 points.
Figures 2c and d likewise show substantial variation of other non-essential beneﬁts and alternative
medicine over time. Despite the mass points around zero, many health plans got upgraded or down-
graded by 10 (other non-essential beneﬁts) or 5 sore points (alternative medicine ) from one year to the
4.2 Analyzing Dominated Plans and Their Enrollees
Studying the joint distribution of all four non-price parameters, one ﬁnds positive correlations between
all four indicators, ranging from 0.10 (other non-essential beneﬁts and brick-and-mortar network) to
0.55 (alternative medicine and hotline service). As also suggested by Figure 1, these positive correlations
between non-price attributes show that sickness funds position themselves either at the high or low non-
price segment of the market. It also suggests that sickness funds are aware of the quality-price trade-off
when differentiating their product.
Considering all ﬁve health plan parameters jointly, the majority of plans in the market are dominated
by at least one competitor. That is, there is at least one competitor that dominates the plan in at least
one of the ﬁve dimensions without being worse in all other dimensions. This holds for every single
year. Figure 3illustrates the share of dominated plans in the German SHI multi-payer market. We
differentiate by plans that operate nationwide vs. all plans, including those that only operate in speciﬁc
states. In addition, the ﬁgure illustrates how the share of dominated plans varies, depending on the
[Insert Figure 3about here]
Because German social legislation mandates universal and minimal cost-sharing for all sickness
funds, price dominance is one-dimensional and straightforward. One or two health plans—the absolute
cheapest plans—always dominates all other plans. This holds for all years and whether we focus on the
black solid black line in Figure 3representing all health plans (whose number decreased from 94 in 2008
to 70 in 2011), or the dashed black line representing just nationwide operating plans (whose number
decreased from 45 in 2008 to 36 in 2011). Expressed in percent, about 97 percent of all German health
plans are dominated when just considering the one-dimensional price criterion.
When additionally considering the other four non-price factors, naturally, the share of dominated
plans decreases. As discussed above and illustrated by Figure 1, the reason is the trade-off between price
and non-price attributes; some plans clearly position themselves on the quality dimension. Deﬁning that
a plan is dominated when another plan is at least as good as the dominated plan in four dimensions but
better in at least one dimension, we obtain the red solid and dashed lines in Figure 3. Prior to the price
framing reform (see Section 2.2), the share of dominated plans was slightly below 70 percent; this share
increased to almost 90 percent in 2010 (due to the equalization of the contribution rate and the initial
hesitation of plans to charge add-on premiums) and then fell back to around 80 percent.
We can conclude that (a) considering non-price attributes decreases the number of dominated plans.
And (b) between 70 and 90 percent of all plans are dominated—even when using ﬁve different param-
eter dimensions—suggesting substantial frictions in the German SHI market.
[Insert Tables 2and 3about here]
Table 2shows a summary statistic of dominated vs. non-dominated plan characteristics. Not sur-
prisingly, on average, dominated plans exhibit poorer values for all plan characteristics, although none
of these differences is statistically signiﬁcant.
Table 3characterizes the stock of enrollees in dominated plans (columns (3) and (4)) as well as en-
rollees who actively choose these plans when switching (columns (1) and (2)). Each column represents
on linear probability model where the dependent binary variable indicates whether the SOEP respon-
dent is enrolled in a dominated plan. Columns (2) and (4) employ just the price as dominance criterion
and Columns (1) and (3) employ all ﬁve health plan attributes.
When applying solely the price criterion in columns (2) and (4), few health plans dominate all other
plans in a given year (Figure 3). Hence, given our sample size, it is challenging to identify signiﬁcant
enrollee characteristics that are correlated with the 97 percent of dominated plans. However, there is
some evidence that younger and healthier individuals are more likely to choose the absolute cheapest
plan in the market.
When applying all ﬁve dominance criteria, we ﬁnd: First, switchers into these plans are older, less
educated, and less likely to be in very good self-reported health (column (1)). Particularly the regression
coefﬁcient on “SAH bad”—the lowest of all ﬁve SAH categories—is very large (but also imprecisely
estimated). These ﬁndings are entirely in line with Bhargava et al. (2017).
Second, when looking at the entire pool of enrollees in dominated health plans—this includes indi-
viduals who never switch, e.g. due to inattention, inertia, or because of high switching costs (Kling et al.,
2012;Handel,2013;Ericson,2014;Ho et al.,2017;Handel and Kolstad,2015)—-the following mixed pic-
ture emerges (column (3)): Enrollees in these plans are younger, more likely to be male, less educated,
and more likely to be in “good” or “satisfactory” SAH (compared to “very good” SAH).
5 The Demand-Side of the German Health Plan Landscape
This section empirically studies consumer demand. The last section, which characterized enrollees in
dominated health plans, already investigated the demand-side a little bit. This section deepens the
analysis of the demand-side. Following the arguments in Einav et al. (2013), we deliberately focus on
the subset of enrollees who actively switch health plans. By explicitly allowing for selection into the
switching decision, we are able to proﬁle switchers’ socio-demographics and study adverse selection.
We exploit the fact that switchers make active decisions and reveal preferences about how they trade-
off price and non-price attributes of health plans. Note that we differentiate between all switchers and
those who switch because they (have to) leave the SHI family insurance.
According to a representative survey among SHI enrollees, 57 percent of actual switchers indicated
that the price played an important role in their switching decision. However, “better beneﬁts” were
still relevant for 34 percent and a “good service” for 26 percent of all respondents (insurer’s image: 10
percent). Among enrollees who were actively considering to switch, the shares were even higher (80
percent price, 85 percent beneﬁts, 59 percent service, and 18 percent image) (Zok,2006). While these
(hypothetical) numbers are informative and underscore the potential relevance of non-price factors in
the decision to switch, the analysis below will study actual choices and the trade-offs that switchers are
willing to make under real world conditions.
5.1 Prices and Non-Prices Differences of Switchers’ Old and New Plans
To obtain a ﬁrst impression about the relevance of prices, supplemental beneﬁts, and service quality in
individuals’ decisions to choose health plans, Table 4compares switchers’ new and old plan character-
[Insert Table 4about here]
11The results are based on a subset of 729 switches for which we know all ﬁve parameters for both the old and the new plan.
Note that, for the main analysis, we only need information on the new plan.
Table 4shows that the monthly premium of the new health plan is, on average, a signiﬁcant e2.37
lower than the premium of the old health plan. On average, there are no signiﬁcant differences for
brick-and-mortar network,hotline service, and other non-essential beneﬁts. However, new plans are
signiﬁcantly more likely to offer alternative medicine suggesting that (some) consumers value these
therapies. Overall, the non-price differences between old and new plans are very small, below 0.05 of a
standard deviation. On the other hand, if heterogeneity in consumer preferences is large, a mean of zero
may not be very informative.
5.2 Choice Determinants Identiﬁed by Mixed Logit Models
Next, we exploit discrete choice methods to model health plan choice determinants (c.f. Beaulieu,2002;
Wedig and Tai-Seale,2002;Jin and Sorensen,2006;Dafny and Dranove,2008). We use the random
parameters model (RPL), also called mixed logit model, which is a generalization of the conditional logit
model. The deﬁnition of the mixed logit model is based on the functional form of its choice probabilities;
it has the following strengths (McFadden,1973;Revelt and Train,1998;McFadden and Train,2000):
First, the mixed logit model explicitly allows for heterogeneity in consumer preferences by modeling
the preference parameters as random variables. Preferences for alternative medicine or service quality
are likely heterogeneously distributed across the population. Second, the RPL does not rely on the
restrictive independence of irrelevant alternatives (IIA) assumption. Third, the model is closely related
to the theoretical concept of a utility maximizing individual iwho reveals her preferences by choosing
health plan jand thereby maximizing utility uij according to the following representation:
iBene f itsj+ζ0
iServicej+αj+εi j (1)
where Priceij is the monthly insurance premium in euro which varies across enrollees and sickness
funds (because it is income-dependent in Germany). Bene f i tsjand Servicejinclude vectors of non-
essential beneﬁts and service quality.
To account for time-invariant unobservable health plan characteristics, we include 315 health plan
ﬁxed effects (αj). Essentially, the model assumes that the unobserved utility part consists of a health plan
speciﬁc ﬁxed effect and a random error term.12 εij is assumed to be iid and to follow a type I extreme
As discussed, we condition on individuals who make active health plan choices and, by revealed
preferences, thus provide information on how they value the trade-off between Priceij,B ene f its jand
12We do not directly include socio-economics in equation (1). Because socio-demographics are invariant over alternatives,
including them would require interacting them with each alternative in the individual choice sets. Because choice sets range from
41 to 73 health plans, for computational reasons, this approach is not feasible. This is similar to the approach adopted by Chernew
et al. (2004).
Servicej. The coefﬁcient vectors γi,δiand ζiare the preference parameters of interest.13 Because the
likelihood function has no closed-form solution, we use maximum simulated likelihood methods to
estimate the parameters.
[Insert Table 5about here]
Table 5reports the estimated means and standard deviations of the random coefﬁcients. Incorpo-
rating health plan ﬁxed effects nets out time-invariant plan characteristics, such as brand loyalty or the
reputation of the fund, and exploits the timely variation shown by Figure 2.
As indicated by the signiﬁcantly negative mean of price in the ﬁrst row and column of Table 5,
cheaper sickness funds are preferred by consumers who choose new health plans. The corresponding
standard deviation in the second column is close to zero and not statistically signiﬁcant, suggesting that
price is a homogeneous choice determinant among switchers.14
In contrast, the ﬁrst column also shows that the mean coefﬁcients for service quality and non-
essential beneﬁts are not statistically signiﬁcant, but of similar size than price. However, not only are the
standard errors much larger, the variables are also measured on different scales. Comparing the effect
of an increase by one standard deviation—which is about 8 for the premium and 1 for the non-price
measures (cf. Table 1, Panel C)—the response to a price increase equals eight times an increase in service
quality or supplemental beneﬁts.
The standard deviations of hotline service and other non-essential beneﬁts are relatively small and
insigniﬁcant, whereas those for brick-and-mortar network and alternative medicine suggest signiﬁcant
heterogeneity in consumer preferences. One ﬁnds that 40 to 60 percent of switchers either place a posi-
tive or negative value on brick-and-mortar network and alternative medicine.15
Although the effectiveness of alternative medicine not been proven in standard randomized con-
trolled trials, some consumers seem to value alternative treatment methods. For example, in one of the
few studies on actual usage of alternative medicine by SHI enrollees, Robert Koch Institut (2002) inves-
tigate claims of one nationwide operating German sickness fund (the BKK Securvita), which has 92K
enrollees and a speciﬁc reputation for offering generous alternative medicine packages. Robert Koch
Institut (2002) counted 21K alternative treatments in 2001—i.e. a share of 23 percent—but estimate that
this share is three times larger than for the average sickness fund.
13We assume that the individual preferences are normally distributed, a diagonal variance-covariance matrix of the coefﬁcients,
and hence uncorrelated coefﬁcients. We use the Add-On package mixlogit for Stata (Hole,2007). Estimation results are based
on 50 Halton draws. Any data or computational errors are our own.
14 Computing expected price elasticities for the mixed logit model is very cumbersome (cf. Train (2009), p. 141) because expected
price elasticities vary across both individuals and health plans. For this reason, it is not obvious which (average) elasticity to report
in order to provide the reader with an informative measure. We refer to Schmitz and Ziebarth (2017) who report an out-of-pocket
elasticity of about -1 for 2007 and 2008 using the same dataset as this study.
15The share of enrollees who value these characteristics (Train,2009) are for alternative medicine:P(X>0) = 1−FX(0) =
/0.371)≈0.57, where Φrepresents the cumulative distribution function of the standard normal distribution.
In summary, the results suggest that prices are the main determinant in health plan choice for those
who actually switch plans. Non-essential supplemental beneﬁts and service quality play, on average,
a negligible role in consumers’ decision to choose plans. An explanation could be that quantitative
premium differences are easy to understand. The monetary trade-off with service quality may only
become salient when customers actually need help (Schram and Sonnemans,2011). Another explanation
could be a low consumer awareness of non-price attributes across plans.
5.3 Testing for the Potential Endogeneity of Health Plan Characteristics
Endogenous Changes of Priceij ,Bene f i tsjand Servicej?
So far, we have assumed that annual changes in Priceij,B ene f its jand Servicejare not simultaneously
correlated with the error term and the health plan choice. It appears to be plausible that this assumption
is satisﬁed in the German SHI context, particularly when focusing on short-term (annual)changes in
plan characteristics. However, to be on the safe side, we solely interpret them as “determinants” and
abstain from a strict causal interpretation.
Unlike in the US, where employer-sponsored health plans may have very small numbers of enrollees
(and single enrollees may indeed affect prices and other plan characteristics), most German sickness
funds have a large number of enrollees. For example, in 2008, the average number of enrollees in all
155 sickness funds was 319,000 (Schmitz and Ziebarth,2017;Pilny et al.,2017). The largest health plan
had more than 5 million enrollees. Given these large enrollee pools, it is basically impossible that one
individual has an impact on prices (particularly given the German RSA, see Section2.4). Moreover, our
model includes health plan ﬁxed effects which net out unobserved time-invariant plan characteristics
such as a basic network infrastructure or a fund’s reputation.
Second, while annual changes in health plan attributes are likely to be exogenous from an indi-
vidual’s perspective, we have shown above that insurers position themselves on the price vs. qual-
ity dimensions (see Section 4). If this strategic decision was driven by unobserved and time-variant
plan characteristics that also matter for plan choice, the coefﬁcient estimates could be biased. However,
switchers are unlikely to have systematic private unobserved (insider) information about funds; their
decision is likely based on the same public quality and price information that we exploit in our empirical
model. For example, some insurers may decide to speciﬁcally engage in price competition, cutting back
customer service. This would only be threat to our main estimates if the reduction in service quality was
not captured by our service quality measure, which is unlikely.
Control Function Approach
To test for the relevance of potentially endogenous changes in health plan parameters, we apply a Con-
trol Function Approach (CFA) as suggested by Petrin and Train (2010). We consider this as a suggestive
robustness check and are aware that the crucial exclusion restrictions cannot be tested, but require good
reasoning. We use two variables that measure changes in the insurer risk pool (at the plan level) to in-
strument for annual changes in prices. We measure risk pool changes by aggregating the enrollee data
up to the yearly health plan level while excluding the information that switchers themselves contribute.
Speciﬁcally, we use one indicator that measures (changes in) the age of the risk pool and one indicator
that measures (changes in) in the health of the risk pool.16 By focusing on price changes and carrying
out a CFA, we provide suggestive empirical evidence on the empirical relevance of endogenous price
First, we carry out the ﬁrst stage regression at the health plan level. As discussed, we measure
annual changes in the risk pool of an insurer and exploit the fact that the German RSA is incomplete
(see Section 2.4). As shown below, changes in the risk pool are indeed highly correlated with changes
in prices (relevance of instrument). Moreover, the exclusion restriction requires that these risk pool
changes only operate through their impact on prices (validity of instrument). It would be violated if
short-term changes in risk pools also had a direct impact on enrollees’ decision to choose plans, which
is unlikely to be the case. We also test whether our instruments (and their interactions) are predictive
of changes in non-price attributes and found no evidence for that. This reassures us that prices are the
main margin that German insurers use to make adjustments in the short-run (recall that we use plan
ﬁxed effects) and that changes in risk pools are the main determinant of prices in a system without
provider networks, but ﬁxed reimbursement rates, and a generous essential beneﬁt package.
Second, we include the residuals from this ﬁrst stage as a control function in our structural model.
[Insert Table 6about here]
Table 6reports the CFA robustness checks using two different models. The ﬁrst just uses the price as
the only potentially endogenous regressor of interest, and the second uses our ﬁve health plan attributes
jointly, assuming that four of them are not signiﬁcantly correlated with the error term and the outcome
measure at the same time.
Panel A reports the ﬁrst stage results which include plan ﬁxed effects. As seen, changes in the age
structure of a risk pool as well as changes in the health structure of a risk pool are signiﬁcant predictors
of price changes. When the risk pool ages by one year, the contribution rate is 0.09 percentage points
16SAH is a comprehensive health measure that includes information beyond the 80 speciﬁc illnesses considered in the RSA.
Despite its simplicity, SAH has been shown to be a good predictor of true health (McGee et al.,1999). Issues related to report-
ing heterogeneity seem to be mostly limited to age and gender (Ziebarth,2010b). Moreover, SAH represents recent up-to-date
information on enrollees’ health.
higher. For both speciﬁcations, the F-statistic lies above the weak instrument threshold of “12” implying
that the instruments are relevant.
Panel B reports the second CFA stage, which includes the residuals from the ﬁrst stage. The residuals
are not statistically signiﬁcant and the results are quite robust in the CFA speciﬁcation which corrects
for potentially endogenous price changes. The benchmark estimate from the potentially biased regular
mixed logit model just using price as a main regressor is -0.057 (Std. Err. 0,007, not shown). The
benchmark estimate from the potentially biased regular mixed logit model with all ﬁve plan attributes
is in Table 5.
As seen, the price coefﬁcients in Table 6are almost identical and remain statistically signiﬁcant.
The standard deviations of brick-and-mortar network and alternative medicine also remain statistically
signiﬁcant and barely change in size. All other estimates are also very robust, reinforcing the notion that
annual changes in health plan characteristics (while netting out time invariant health plan ﬁxed effects)
can probably be considered exogenous from the perspective of a single consumer in a market where the
average health plan enrolls 319K individuals (and has few margins to adjust expenditures). However,
as mentioned, as the exclusion restriction cannot be formally tested, we remain cautious and interpret
the parameter estimates in the mixed logit models not as causal but call them “determinants.”
5.4 How Consumers Trade-off Price and Non-Price Factors
Let us now assess how switchers trade off the ﬁve health plan determinants. When using discrete choice
models—which are based on a linear index as in equation (1)—the ratios of the coefﬁcients represent
marginal rates of substitution (MRS). We are interested in −δk
i/γiindicating at which rate
enrollee iis willing to trade off additional beneﬁts or service quality against a lower premium, where k
indexes beneﬁt and quality characteristics.
Interpreting these ratios, however, requires scales that measure a one unit change. As no natural scale
is available for service quality and supplemental beneﬁts, we deﬁne a unit as one standard deviation
in the sample distribution. Similarly, we deﬁne one price unit; an increase in the monthly premium by
one standard deviation equals an e8 increase (cf. Table 1, Panel C). This value has intuitive appeal, as it
represents the average difference in health plans prices.
Mixed-logit estimation does not yield estimates for −δk
i/γiat the individual level and,
for this reason, does not allow for the calculation of individual marginal rates of substitution. Instead,
we obtain normally distributed population parameter estimates µkand σk. Because MRSkis a ratio
of normal random variables, its distribution involves a Cauchy-component rendering the mean (and
higher-order moments) undeﬁned. It cannot be consistently estimated (cf. Cohen Freue,2007), which is
why we discuss quantiles instead of means of MRSk. In particular, we focus on the median of MRSk.
Rather than directly interpreting −ˆ
µpremiu m as ML-estimate for the median of MRSk, we simulate the
percentiles of the MRS-distributions—-along with the corresponding 95-% conﬁdence bands.17
Prices vs. Non-Price Factors
Let us start with MRSstore. The simulated median of MRSstore is 0.128, indicating that the median enrollee
trades off lower prices against a denser brick-and-mortar network at a rather small rate: An increase in
the store network by one standard deviation (SD) is just valued as one-eighth SD of the premium, which
is e1 per month. However, the simulated 95%-conﬁdence interval of [−0.324, 0.564]indicates that this
value is imprecisely estimated.18 Nevertheless, even the upper conﬁdence bound is only 0.56 and lets us
exclude—with 95% statistical certainty—that consumers would trade an increase in the store network
by one SD for more than e8.50 per month.
This picture somewhat changes when we consider the estimated heterogeneity in the MRS. Although
the exact shape of the estimated MRS distribution depends heavily on distributional assumptions (and
should be interpreted with caution), assessing other quantiles may provide insights into the heterogene-
ity of consumer preferences. At the 95th percentile, the rate of substitution is more than ten times larger
than at the median (point estimate 1.307). This means that, according to our estimates, those ﬁve per-
cent of enrollees who have the strongest preferences for face-to-face services and a high store density
are willing to accept a 1.3 SD increase in the monthly premium (e10) for a one SD increase in the store
density. However, this number carries a lot of uncertainty (conﬁdence interval: [0.286, 3.064]).
On the other hand, according to the estimated distribution of MRSstore, 42 percent of all switchers
do not accept higher prices in return for more physical branches. However, we ﬁnd large consumer
heterogeneity and a small fraction of consumers who value physical branches highly.
The remaining MRSs resemble what has been just discussed for brick-and-mortar network. The es-
timated median MRSs range from 1
/10 (hotline service: 0.106; alternative medicine : 0.140; other
supplemental beneﬁts: 0.118), indicating a rather low median willingness to pay (WTP) for optional sup-
plemental beneﬁts like immunization shots, alternative medicine as well as hotline services. However,
the point estimates carry wide conﬁdence intervals. At the 95th percentile, the estimated MRSs (hotline
service: 0.808; alternative medicine: 1.379; other supplemental beneﬁts: 0.723) are 6 to 10 times larger
than the median. On the other hand, about 40 percent of active consumers are not willing to accept any
price increase in return for more beneﬁts or a better service.
17 The ratio of the means µk/µldoes not provide an accurate approximation of the median of the corresponding ratio
distribution—if the denominator distribution has a mean close to zero and a non-vanishing density at zero. To estimate the
percentiles, we draw 2 million random numbers from the relevant ratio distributions, with the point estimates ˆ
σpremium entering the normal distributions, and then average the simulated quantiles over 2,000 replications. Due to the large
size of the pseudo sample, the estimated percentiles exhibit very little sampling variability and averaging has almost no effect.
To simulate the conﬁdence bands, we also sample 2,000 times where, in each replication, the four relevant parameters are drawn
from the (estimated) jointly-normal distribution of the ML-estimator.
µpremium directly as estimate for med(MRSstore )and applying the delta-method for calculating conﬁdence intervals
yields results (point estimate: 0.128, conﬁdence interval: [−0.306, 0.561]) that just marginally deviate from the simulation-based
counterpart. This can be explained by the small value of ˆ
σpremium that lets the density of the denominator almost vanish at zero.
A Simulataneous Increase in all Non-Price Factors
In a ﬁnal step, we analyze how consumers would trade off a simultaneous increase in all four non-price
parameters by one SD. The calculation yields an estimated median value of 0.491 (conﬁdence inter-
val: [−0.181, 1.181]). Although—not surprisingly—the median WTP for joint improvements in quality
and beneﬁts exceeds the median WTP for single improvements, the point estimate is still smaller than
Again, the joint rate of substitution exhibits considerable heterogeneity. At the 95th percentile, the
rate is 2.438, i.e., ﬁve times larger than at the median. A third of all switchers are not willing to accept
higher premiums at all, even if all four non-price attributes would simultaneously increase by one SD.
5.5 The Selection into Health Plan Switching
Health Risks and the Potential for Cream Skimming
The German SHI system combines guaranteed issue with income-dependent contribution rates. In-
dividual risk rating is prohibited. This regulation, however, creates an incentive for sickness funds
to engage in active or passive risk selection. To minimize this incentive, the German RSA exists (see
Section 2.4). However, because the risk adjustment scheme is incomplete, the risk-adjusted capitated
payments for unhealthy enrollees are likely smaller than their actual costs. However, it is unclear to
what extent risk selection exists in the German market. Bauhoff (2012) ﬁnds evidence for risk selection
based on the state of residence of the insured—which is found to be very small, however, in quantitative
terms—whereas Nuscheler and Knaus (2005) and McGuire (2007) ﬁnd no evidence for risk selection in
the German SHI.
This paper cannot directly test whether plans actively cream skim. However, in the following we
assess the potential for risk selection by analyzing whether enrollees sort into plans with speciﬁc char-
acteristics, e.g., whether young and healthy switchers prefer speciﬁc plans. Section 4.2 has already
shown that dominated health plans have less educated and less healthy enrollees (and switchers into
[Insert Table 7about here]
To classify individuals into different risk types, we make use of the factors SAH, age, and gender.
However, the difference to Section 5.3 is that we do not aggregate these information at the plan-year
level to use them as instruments, but test whether switchers with these individual-level characteristics
are more likely to choose speciﬁc plans.
We construct mutually exclusive subsets of the two dichotomous indicators G1(group 1) and G2
(group 2) which represent different health risks. “Good health risks” report at least good health (SAH
category 1 or 2). Age is collapsed into two binary variables—younger than 50 (G1) and older than 50
(G2). We also use indicators for males (G1) and females (G2).
Then, we run regressions for each of these stratifying variables and interact them with all ﬁve health
plan characteristics. Table 7presents the results. Signiﬁcant differences between the distributions of the
preference parameters are indicated by +.
With respect to health plan prices, no signiﬁcant differences are found when we stratify by age,
gender, or health status. This implies that the sick and the healthy, the young and the old, as well as
males and females all seem to value (low) prices. The same holds for brick-and-mortar network.
With respect to hotline service, the null hypothesis of equal distributions is rejected when we stratify
by gender and health status (p-values: 0.0059 and 0.0179). The fraction of new enrollees who value a
good hotline service differs signiﬁcantly by the SAH speciﬁcation (G1(good health): 71 percent, G2(bad
health): 19 percent). Interestingly, good health risks seem to value hotline services more than bad health
The opposite holds for alternative medicine. The hypothesis of equal distributions for gender and
health is rejected (p-values: 0.0823 and 0.0309), but the fraction of those who value alternative treatments
is much larger among bad health risks and women (SAH: G1: 52 percent, G2: 85 percent; gender: G1
(men): 54 percent, G2(women): 98 percent).
With respect to other non-essential beneﬁts, the null hypothesis of equal distributions is again re-
jected for gender and health (p-values: 0.0823 and 0.0309), indicating that those in good health and
women value non-essential beneﬁts (like certain immunizations and preventive check-ups) more than
bad health risks (SAH: G1: 71 percent, G2: 56 percent; gender: G1(men): 75 percent, G2(women): 55
In sum, preference differences by age, gender and health status seem to be rather small. To the
extent that they exist, they do not point into one clear direction. Females value alternative treatments
and non-essential beneﬁts more than males. The unhealthy also value alternative medicine more than
the healthy, who have stronger preferences for other non-essential beneﬁts like immunizations. One
could argue that non-price attributes do not seem to be powerful tools for indirect risk selection in the
Using the Price Framing Reform of 2009 as Exogenous Shifter of the Switching Decision
As mentioned in Section 2, our sample period covers a price framing reform that became effective on
January 1, 2009. Schmitz and Ziebarth (2017) and Wuppermann et al. (2014) ﬁnd substantial effects
of this reform on the likelihood to switch plans. As seen in Figure 3, the share of dominated plans
decreased post-reform because market price dispersion shrunk (Schmitz and Ziebarth,2017).
This subsection tests whether consumers’ valuation of of prices, service quality and optional beneﬁts
structurally changed post- as compared to pre-reform. Using the price framing reform as an exogenous
shifter of the decision to switch health plans, this exercise also provides us with an intuition how our
results would change if we corrected the endogenous decision to switch plans.
To carry out our test, we split the sample into a pre- (2008/2009) and post-reform (2010/2011) pe-
riod.19 As can be seen in Table A1 (Online Appendix), the estimated mean coefﬁcient of the price is
about twice as large post-reform, which is in line with Schmitz and Ziebarth (2017) and Wuppermann
et al. (2014). The estimated coefﬁcients for alternative medicine and other non-essential beneﬁts, how-
ever, are very similar in pre and post-reform years. The estimates for brick-and-mortar network and
other non-essential beneﬁts remain non-signiﬁcant but increase in size which could indicate an increase
in their mean relevance.20 However, an LR-test fails (p-value 0.599) to reject the null hypothesis of equal
This paper studies a multi-payer market that combines consumer choice with heavy government regu-
lation. We exploit health plan level data on almost all German health plans. The health plan level data
contain price and non-price information of these plans. The non-price features are measured by con-
sistently collected information on service quality (hotline service,brick-and-mortar network) as well as
non-essential supplemental beneﬁts (alternative medicine,other non-essential beneﬁts).
Our supply-side analysis reveals, ﬁrst, that a clear and signiﬁcant trade-off between non-price and
prices attributes exists. Cheaper plans offer less service quality and less optional beneﬁts. Second,
we ﬁnd that the large majority of health plans are dominated—not only dominated on the ﬁnancial
dimension, but simultaneously dominated on the price and non-price dimensions. This suggests the
existence of signiﬁcant market frictions, despite the fact that consumer choice is signiﬁcantly simpliﬁed
through the regulatory framework (for example, the price dimension is one-dimensional). Enrollees
in dominated plans (as well as those who are observed actively choosing them) are signiﬁcantly less
healthy and less educated than enrollees in dominating plans. This ﬁnding is in line with recent evidence
from the US (Bhargava et al.,2017).
Our demand-side analysis is based on representative enrollee panel data, which we link to the health
plan level data to assess the relative roles of prices, non-essential beneﬁts, and service quality in the
decision to choose health plans. We model individuals’ health plan choices using a random parameters
model which accounts for health plan heterogeneity and time-invariant unobserved factors. In total, the
19We opt for assigning the wave 2009 to the pre-reform period, as SOEP interviews are typically carried out at the beginning of
a year and, hence, switching most likely refers to the previous year.
20 If this was true, the effects could either stem from the higher price sensitivity, the price compression in the market, or a
different subset of enrollees who actually switched plans.
empirical setting exploits 1,724 health plan switches and almost 100,000 potential choice sets between
2007 and 2010.
While substantial variation in all investigated health plan characteristics exists, we ﬁnd that prices
play the dominant role in consumers’ decision to choose health plans (when they decide to switch).
For the median enrollee, an increase in any of the non-price factors (density of brick-and-mortar stores,
hotline quality, alternative medicine, and other non-essential supplemental beneﬁts) by one standard
deviation is offset by a decrease in premiums by only one-eighth of a standard deviation, or e1 per
month. In contrast, from a supply-side perspective, the positioning of health plans between the vari-
ous price and non-price dimensions suggest a much higher trade-off that consumers are evidently not
willing to pay. For example, health plans trade-off e10 per month per standard deviation increase in
the density of the branch network. In other words, even when service quality and non-essential beneﬁts
are valued by consumers (which survey evidence suggests they do), they are only willing to trade them
for higher prices at a much smaller rate. This could explain why the number of sickness funds has been
decreasing substantially over the past decades.
Finally, we ﬁnd substantial consumer heterogeneity in the valuation of non-price attributes; a mi-
nority of consumers seem to value service quality very highly. On the other hand, we ﬁnd that up to
70 percent of enrollees who switch plans do not value non-essential beneﬁts and service quality at all, a
ﬁnding that is reinforced by survey evidence.
As in every empirical study, the strict interpretation of our results is limited to the speciﬁc setting,
in our case the German market. However, we believe that the ﬁndings are of broader relevance, in
particular because few papers are able to explicitly study the role of supplemental beneﬁts and service
quality relative to prices both from a demand and supply perspective. For example, we illustrate the
relevance of considering non-price attributes when analyzing dominated plans; non-price attributes
are important because we ﬁnd that a share of consumers value them highly and may stay in expensive
plans precisely for that reason. However, we also show that—among the selective sample of switchers—
prices are clearly the dominant factor in the decision to choose plans. Lastly, the paper illustrates that
substantial choice frictions also exist in markets with substantial government regulation and a much
simpler choice architecture than in the US.
Abaluck, J. and J. Gruber (2016). Evolving choice inconsistencies in choice of prescription drug insur-
ance: Do choices improve over time? American Economic Review 106(8), 2145–2184.
Abaluck, J., J. Gruber, and A. Swanson (2015). Prescription drug use under Medicare Part D: A linear
model of nonlinear budget sets. NBER Working Papers 20976.
Abaluck, J. T. and J. Gruber (2011). Choice inconsistencies among the elderly: Evidence from plan choice
in the Medicare Part D program. American Economic Review 101(4), 1180–1210.
Abraham, J. M., R. Feldman, C. Carlin, and J. Christianson (2006). The effect of quality information
on consumer health plan switching: Evidence from the Buyers Health Care Action Group. Journal of
Health Economics 25(4), 762–781.
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:I), 847–864.
Bauhoff, S. (2012). Do health plans risk-select? An audit study on Germany’s social health insurance.
Journal of Public Economics 96(9-10), 750 – 759.
Beaulieu, N. D. (2002). Quality information and consumer health plan choices. Journal of Health Eco-
nomics 21(1), 43–63.
Bhargava, S., G. Loewenstein, and J. Sydnor (2017). Do individuals make sensible health insurance deci-
sions? evidence from a menu with dominated options. The Quarterly Journal of Economics forthcoming.
Buchmueller, T. (2006). Price and the health plan choices of retirees. Journal of Health Economics 25(1),
Buchmueller, T. C., K. Grazier, R. A. Hirth, and E. N. Okeke (2013). The price sensitivity of Medicare
beneﬁciaries: A regression discontinuity approach. Health Economics 22(1), 35–51.
Buchner, F. and J. Wasem (2003). Needs for further improvement: Risk adjustment in the German health
insurance system. Health Policy 65(1), 21–35. Van De Ven Special Issue.
Bundorf, M. K. (2010). The effects of offering health plan choice within employment-based purchasing
groups. Journal of Risk and Insurance 77(1), 105–127.
Cebul, R. D., J. B. Rebitzer, L. J. Taylor, and M. E. Votruba (2011). Unhealthy insurance markets: Search
frictions and the cost and quality of health insurance. American Economic Review 101(5), 1842–71.
Chernew, M., G. Gowrisankaran, C. McLaughlin, and T. Gibson (2004). Quality and employers’ choice
of health plans. Journal of Health Economics 23(3), 471–492.
Chernew, M., G. Gowrisankaran, and D. P. Scanlon (2008). Learning and the value of information:
Evidence from health plan report cards. Journal of Econometrics 144(1), 156–174.
Christiansen, M. C., M. Eling, J.-P. Schmidt, and L. Zirkelbach (2016). Who is changing health insurance
coverage? Empirical evidence on policyholder dynamics. Journal of Risk and Insurance 83(2), 269300.
Cohen Freue, G. V. (2007). The Pitman estimator of the Cauchy location parameter. Journal of Statistical
Planning and Inference 147, 1900–1913.
Cutler, D. M. and S. J. Reber (1998). Paying for health insurance: The trade-off between competition and
adverse selection. The Quarterly Journal of Economics 113(2), 433–466.
Dafny, L. and D. Dranove (2008). Do report cards tell consumers anything they don’t already know?
The case of Medicare HMOs. The Rand Journal of Economics 39(3), 790–821.
Dafny, L., J. Gruber, and C. Ody (2015). More insurers lower premiums: Evidence from initial pricing in
the health insurance marketplaces. American Journal of Health Economics 1(1), 53–81.
Dijk, M., M. Pomp, R. Douven, T. Laske-Aldershof, E. Schut, W. Boer, and A. Boo (2008). Consumer price
sensitivity in Dutch health insurance. International Journal of Health Care Finance and Economics 8(4),
Dowd, B. E. and R. Feldman (1994). Premium elasticities of health plan choice. Inquiry 31(4), 438–444.
Dusansky, R. and C. Koc¸ (2010). Implications of the interaction between insurance choice and medical
care demand. Journal of Risk and Insurance 77(1), 129–144.
Einav, L., A. Finkelstein, S. P. Ryan, P. Schrimpf, and M. R. Cullen (2013). Selection on moral hazard in
health insurance. American Economic Review 103(1), 178–219.
Ericson, K. M. and A. Starc (2012). Heuristics and heterogeneity in health insurance exchanges: Evidence
from the Massachusetts Connector. American Economic Review 102(3), 493–97.
Ericson, K. M. M. (2014). Consumer inertia and ﬁrm pricing in the Medicare Part D prescription drug
insurance exchange. American Economic Journal: Economic Policy 6(1), 38–64.
Ericson, K. M. M. and A. Starc (2015). Pricing regulation and imperfect competition on the Massachusetts
Health Insurance Exchange. The Review of Economics and Statistics 97(3), 667–682.
Frank, R. G. and K. Lamiraud (2009). Choice, price competition and complexity in markets for health
insurance. Journal of Economic Behavior & Organization 71(2), 550–562.
Gaynor, M. and R. J. Town (2012). Competition in health care markets. In T. McGuire, M. Pauly, and P. P.
Barros (Eds.), Handbook of Health Economics, Vol. 2. Amsterdam: Elsevier North-Holland. Chapter 9.
Handel, B. R. (2013). Adverse selection and switching costs in health insurance markets: When nudging
hurts. American Economic Review 107(3), 2643–2682.
Handel, B. R. and J. T. Kolstad (2015). Health insurance for ”humans”: Information frictions, plan choice,
and consumer welfare. American Economic Review 105(8), 2449–2500.
Harris, K. M. (2002). Can high quality overcome consumer resistance to restricted provider access?
Evidence from a health plan choice experiment. Health Services Research 37(3), 551–571.
Heiss, F., A. Leive, D. McFadden, and J. Winter (2013). Plan selection in Medicare Part D: evidence from
administrative data. Journal of Health Economics 32(6), 1325–1344.
Heiss, F., D. McFadden, and J. Winter (2006). Who failed to enroll in Medicare Part D, and why? Early
results. Health Affairs 25(5), w344–w354.
Ho, K., J. Hogan, and F. S. Morton (2017). The impact of consumer inattention on insurer pricing in the
Medicare Part D program. RAND Journal of Economics forthcoming.
Hole, A. R. (2007). Estimating mixed logit models using maximum simulated likelihood. The Stata
Journal 7(3), 388–401.
IGES (2004). Klassiﬁkationsmodelle fr Versicherte im Risikostrukturausgleich.http://www.iges.de/
EndberichtRSA-Gutachten_ger.pdf, last accessed on September 4, 2016.
Jin, G. Z. and A. T. Sorensen (2006). Information and consumer choice: the value of publicized health
plan ratings. Journal of Health Economics 25(2), 248–275.
Karaca-Mandic, P., R. Feldman, and P. Graven (2017). The role of agentsand brokers in the market for
health insurance. Journal of Risk and Insurance, forthcoming.
Ketcham, J. D., C. Lucarelli, E. J. Miravete, and M. C. Roebuck (2012). Sinking, swimming, or learning
to swim in Medicare Part D. American Economic Review 102(6), 2639–2673.
Ketcham, J. D., C. Lucarelli, and C. A. Powers (2015). Paying attention or paying too much in Medicare
Part D. American Economic Review 105(1), 204–33.
Kling, J. R., S. Mullainathan, E. Shaﬁr, L. C. Vermeulen, and M. V. Wrobel (2012). Comparison friction:
Experimental evidence from Medicare drug plans. The Quarterly Journal of Economics 127(1), 199–235.
Kolstad, J. T. (2013). Information and quality when motivation is intrinsic: Evidence from surgeon report
cards. American Economic Review 103(7), 2875–2910.
Kolstad, J. T. and M. E. Chernew (2009). Quality and consumer decision making in the market for health
insurance and health care services. Medical Care Research and Review 66(1), 28S–52S.
Lange, W. G. (Ed.) (2011). dfg-Ranking – Liste der deutschen Krankenkassen, Ranking nach Mitgliederzahlen,
Stichtag 1. Juli 2011. Beitr¨
age zur Gesellschaftspolitik (BzG). MC.B, Verlag f ¨
Loewenstein, G., J. Y. Friedman, B. McGill, S. Ahmad, S. Linck, S. Sinkula, J. Beshears, J. J. Choi, J. Kol-
stad, D. Laibson, B. C. Madrian, J. A. List, and K. G. Volpp (2013). Consumers misunderstanding of
health insurance. Journal of Health Economics 32(5), 850–862.
McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior. In P. Zaremka (Ed.),
Frontiers in Econometrics, pp. 105–142. Academic Press: New York.
McFadden, D. and K. Train (2000). Mixed MNL models for discrete response. Journal of Applied Econo-
metrics 15(5), 447–470.
McGee, D. L., Y. Liao, G. Cao, and R. S. Cooper (1999). Self-reported health status and mortality in a
multiethnic US cohort. American Journal of Epidemiology 149(1), 41–46.
McGuire, T. G. (2007). Risk adjustment systems in competitive health insurance markets: Lessons from
the United States. In U. Schneider, V. Ulrich, and E. Wille (Eds.), Wettbewerb und Risikostrukturausgleich
im internationalen Vergleich: Erfahrungen aus den USA, der Schweiz, den Niederlanden und Deutschland, pp.
69–114. Nomos Verlagsgesellschaft mbH & Co. KG.
uller, A. and W. G. Lange (Eds.) (2010). Dfg-Krankenkassenatlas. MC.B, Verlag f¨
Loseblattausgabe in 3 Ordnern.
Murphy, K. M. and R. H. Topel (1985). Estimation and inference in two-step econometric models. Journal
of Business & Economic Statistics 3(4), 370–379.
Nuscheler, R. and T. Knaus (2005). Risk selection in the German public health insurance system. Health
Economics 14(12), 1253–1271.
Parente, S. T., R. Feldman, J. Abraham, and Y. Xu (2011). Consumer response to a national marketplace
for individual health insurance. Journal of Risk and Insurance 78(2), 389–411.
Petrin, A. and K. Train (2010). A Control Function Approach to endogeneity in consumer choice models.
Journal of Marketing Research 47(1), 3–13.
Pilny, A., A. W ¨
ubker, and N. R. Ziebarth (2017). Introducing risk adjustment and free health plan choice
in employer-based health insurance: Evidence from Germany. Journal of Health Economics forthcoming.
Revelt, D. and K. Train (1998). Mixed logit with repeated choices: Households’ choices of appliance
efﬁciency level. Review of Economics and Statistics 80(4), 647–657.
Robert Koch Institut (2002). Inanspruchnahme alternativer Methoden in der Medi-
zin. Gesundheitsberichterstattung des Bundes Heft 9. http://www.rki.de/DE/
alternative_methoden_inhalt.html, last accessed on August 3, 2016.
Royalty, A. B. and N. Solomon (1999). Health plan choice: price elasticities in a managed competition
setting. Journal of Human Resources 34(1), 1–41.
Scanlon, D. P., M. Chernew, C. McLaughlin, and G. Solon (2002). The impact of health plan report cards
on managed care enrollment. Journal of Health Economics 21(1), 19–41.
Schmitz, H. and N. R. Ziebarth (2017). Does price framing affect the consumer price sensitivity of health
plan choice? Journal of Human Resources 52(1), 89–128.
Schram, A. and J. Sonnemans (2011). How individuals choose health insurance: An experimental anal-
ysis. European Economic Review 55(6), 799–819.
Schut, F. T. and W. H. J. Hassink (2002). Managed competition and consumer price sensitivity in social
health insurance. Journal of Health Economics 21(6), 1009–1029.
Strombom, B. A., T. C. Buchmueller, and P. J. Feldstein (2002). Switching costs, price sensitivity and
health plan choice. Journal of Health Economics 21(1), 89–116.
Train, K. (2009). Discrete choice methods with simulation. Cambridge University Press.
Wagner, G. G., J. R. Frick, and J. Schupp (2007). The German Socio-Economic Panel Study (SOEP): scope,
evolution, and enhancements. Journal of Applied Social Science Studies (Schmollers Jahrbuch: Zeitschrift
f¨ur Wirtschafts- und Sozialwissenschaften) 127(1), 139–169.
Wedig, G. J. and M. Tai-Seale (2002). The effect of report cards on consumer choice in the health insurance
market. Journal of Health Economics 21(6), 1031–1048.
Wissenschaftlicher Beirat zur Weiterentwicklung des Risikostrukturausgleichs beim Bundesver-
sicherungsamt (2011). Evaluationsbericht zum Jahresausgleich 2009 im Risikostrukturausgle-
Meldungen/Gutachten_mit_Anlagen.pdf, last accessed on September 4, 2016.
Wuppermann, A., S. Bauhoff, and M. Grabka (2014). The price sensitivity of health plan choice among
retirees: Evidence from the German social health insurance. mimeo.
Ziebarth, N. R. (2010a). Estimating price elasticities of convalescent care programmes. The Economic
Journal 120(545), 816–844.
Ziebarth, N. R. (2010b). Measurement of health, health inequality, and reporting heterogeneity. Social
Science & Medicine 71(1), 116–124.
Zok, K. (2006). Beitragssatzkenntnis und Wechselbereitschaft in der GKV. Ergebnisse
asentativ-Umfragen bei GKV-Mitgliedern im Zeitverlauf. WIdO-monitor 2(3),
html, last accessed on August 2, 2016.
Figures and Tables
Figure 1: Trade-Off Between Health Plan Prices, Service Quality, and Optional Supplemental Beneﬁts
0 5 10 15 20 25
12 13 14 15 16
Price=13.8513 + 0.0292*Hotline [p-value=4.46]
0 5 10 15 20 25
12 13 14 15 16
Price=13.8963 + 0.0731*Density [p-value=4.93]
0 5 10 15 20 25
Other Non-Essential Benefits
12 13 14 15 16
Contribution Rate in ppt of Gross Wage ('Price')
Price=13.9183 + 0.0366*Benefits [p-value=4.19]
0 5 10 15 20 25
12 13 14 15 16
Contribution Rate in ppt of Gross Wage ('Price')
Price=14.2907 + 0.0172*AlternativeMedicine [p-value=1.44]
Figure 2: Changes in Health Plan Prices, Supplemental Beneﬁts, and Service Quality Over Time
0 .5 1 1.5 2
-1 0 1 2 3
Change in Contribution Rates btw. t=0 and t=1
0 .05 .1 .15
-20 -10 0 10 20
Change in Hotline Service Quality Score btw. t=0 and t=1
0 .05 .1
-10 0 10 20
Change in Non-Essential Benefit Score btw. t=0 and t=1
0 .05 .1 .15 .2
-10 -5 0 5 10 15
Change in Alternative Medicine Score btw. t=0 and t=1
Figure 3: Share of Dominated Plans Using different Dominance Criteria
2008 2009 2010 2011
All health plans, all 5 criteria All health plans, only price
Only national health plans, all 5 criteria Only national health plans, only price
Note: Own calculations, own illustration.
Table 1: Descriptive Statistics
Panel A: Sample Characteristics 2008 2009 2010 2011 Total
Observations (=switches×#plans) 25,920 25,476 27,352 21,140 99,888
Health Plan Switches 388 447 477 414 1,726
Standard Switchers 231 293 318 285 1,127
Exit Family Insurance 157 154 159 129 599
Mean 66.8 57.0 57.3 51.1 57.9
S.D. 5.4 4.4 4.7 3.9 7.1
Min. 53.0 48.0 48.0 41.0 41.0
Max. 73.0 63.0 63.0 55.0 73.0
Panel B: Individual Characteristics Mean S.D. Med. Min. Max. Obs.
Self-Assessed Health 2.34 0.84 2 1 5 1,724
Very Good 0.13 0.34 0 0 1 1,724
Good 0.49 0.50 0 0 1 1,724
Satisfactory 0.29 0.45 0 0 1 1,724
Poor 0.08 0.27 0 0 1 1,724
Bad 0.01 0.10 0 0 1 1,724
Age 37.03 12.63 37 18 80 1,726
Female 0.55 0.50 1 0 1 1,726
Monthly Gross Income [EUR] 1,959 1,413 1,700 400 12,885 1,726
Panel C: Health Plan Characteristics Mean S.D. Med. Min. Max. Obs.
Premium [EUR]+152.98 8.04 158 133 178 323
Brick-and-Mortar Network 0.00 1.00 -0.15 -2.42 2.48 323
Hotline Service 0.00 1.00 0.28 -2.95 1.39 323
Alternative Medicine 0.00 1.00 -0.17 -2.16 3.16 323
Other Non-Essential Beneﬁts 0.00 1.00 0.06 -4.73 1.93 323
uller and Lange (2010); Lange (2011), German Federal (Social) Insurance Ofﬁce, National Asso-
ciation of Statutory Health Insurance Funds, annual reports of the sickness funds, information by sickness
funds. SOEP.v28. Authors’ calculation based on the SOEP. +denotes ﬁctive absolute EUR premiums based
on a monthly gross income of e2,000 for illustrative purposes because premiums depend on three factors: the
contribution rate, the individual income, and the add-on premium (after 2009). In order to show average euro
amounts on the health plan level, we use a hypothetical monthly gross wage of e2,000. In the regression models,
exact premiums are calculated.
Table 2: Dominated vs. Non-Dominated Health Plans: All 5 Attributes as Dominance Criteria
All Dominated Non-dominated
health plans health plans health plans
Mean S.D. Mean S.D. Mean S.D.
Contribution rate 14.41 0.75 14.53 0.66 14.02 0.88
Brick-and-Mortar Network 5.71 2.47 5.26 2.09 7.16 2.99
Hotline Service 19.13 5.65 18.26 5.76 21.86 4.30
Alternative Medicine 4.97 4.07 3.79 3.12 8.70 4.48
Other Non-Essential Beneﬁts 12.80 4.93 12.33 5.06 14.29 4.21
Share of Female Enrollees 0.46 0.19 0.44 0.21 0.54 0.11
Average Age of Enrollees 43.04 3.63 42.83 3.90 43.71 2.57
Share of Enrollees SAH good 0.56 0.20 0.55 0.22 0.57 0.12
Observations 154 117 37
uller and Lange (2010); Lange (2011), German Federal (Social) Insurance Ofﬁce, National
Association of Statutory Health Insurance Funds, annual reports of the sickness funds, information by
sickness funds. SOEP.v28. Authors’ calculation based on the SOEP and on 154 plan-year observations.
Only plans that operate nationwide are included. The same set of plans is considered in Table A2 in the
Table 3: Characteristics of Enrollees and Switchers in Dominated Health Plans
Only Switchers All Enrollees
All attributes Only price All attributes Only price
(1) (2) (3) (4)
Age >50 0.035 (0.037)−0.001 (0.007)−0.053∗∗∗(0.008)0.004∗∗∗(0.001)
Female 0.030 (0.028)−0.002 (0.005)−0.088∗∗∗(0.007)−0.001 (0.001)
Education (=hs) −0.081∗(0.044)0.003 (0.009)−0.010 (0.015)0.003 (0.003)
Education (>hs) −0.095∗(0.051)0.015∗(0.009)−0.078∗∗∗ (0.016)0.005 (0.003)
Married 0.026 (0.029)−0.004 (0.005)−0.002 (0.008)−0.000 (0.001)
SAH good 0.078∗(0.041)0.004 (0.010)0.022∗(0.013)0.001 (0.002)
SAH satisfactory 0.071 (0.045)0.012 (0.010)0.027∗(0.014)0.002 (0.002)
SAH poor 0.058 (0.062)0.006 (0.014)0.017 (0.017)0.000 (0.003)
SAH bad 0.226 (0.139)0.016∗(0.009)0.004 (0.034)0.001 (0.005)
# Observations 1188 1188 15,998 15,998
Mean dep. var. 0.338 0.992 0.330 0.994
uller and Lange (2010); Lange (2011), German Federal (Social) Insurance Ofﬁce, National Association of Statutory
Health Insurance Funds, annual reports of the sickness funds, information by sickness funds. SOEP.v28. Authors’ calculation
based on the SOEP. Each column represents one linear probability model. The dependent variable in columns (1) and (3) is
a dummy that is 1 of the plan is dominated considering all ﬁve health plan attributes as dominance criteria; the dependent
variable in columns (2) and (4) is a dummy that is 1 of the plan is dominated only considering prices as dominance criterion.
Only plans that operate nationwide are included. The results are robust to including all health plans (available upon request).
Robust standard errors are in parentheses. ∗∗∗ p<0.01; ∗∗ p<0.05; ∗p<0.1.
Table 4: New and Old Health Plan Characteristics of Switchers
Price 133.12 135.49 −2.37∗∗∗ (0.20)
Brick-and-Mortar Network 1.01 0.96 0.05 (0.03)
Hotline Service 0.80 0.80 0.00 (0.02)
Alternative Medicine 0.96 0.89 0.07∗∗ (0.03)
Other Non-Essential Beneﬁts 0.45 0.42 0.03 (0.03)
Notes: Authors’ calculation. The table shows sickness fund characteristics of switchers’ new and
old health plans. The calculations are based on 729 observations. Standard errors of the differ-
ences in means are in parentheses. ∗∗∗ p<0.01; ∗∗ p<0.05; ∗p<0.1
Table 5: Health Plan Choice Determinants (Mixed Logit Models)
Price -0.061*** (0.008) 0.002 (0.014)
Brick-and-Mortar Network 0.063 (0.110) 0.354*** (0.128)
Hotline Service 0.052 (0.152) 0.210 (0.335)
Alternative Medicine 0.069 (0.071) 0.371*** (0.108)
Other Non-Essential Beneﬁts 0.058 (0.064) 0.182 (0.150)
Health Plan Fixed Effect yes
# Observations (=switches×#plans) 99,888
# Health Plan Switches 1,726
Notes: Authors’ calculation. The table shows estimated means and standard deviations of
the random parameters. Estimated standard errors are in parentheses. ∗∗∗ p<0.01; ∗∗
Table 6: Robustness—Health Plan Choice Determinants (Control Function Approach)
Panel A: First Stage—Contribution Rate Coefﬁcient SE Coefﬁcient SE
Instruments at Health Plan Level
Risk Pool Age 0.089∗∗∗ (0.021)0.086∗∗∗ (0.021)
Risk Pool Health −0.685∗(0.353)−0.691∗(0.361)
Other Non-Price Controls no yes
Health Plan Fixed Effect yes yes
F-statistic instruments 13.37 12.91
# Observations 316 316
# Health Plans 112 112
Panel B: Second Stage—Choice Model Mean SD Mean SD
First stage residual 0.056 (0.119)0.049 (0.127)
Price −0.060∗∗∗(0.010)0.000 (0.013)−0.064∗∗∗ (0.011)0.001 (0.016)
Brick-and-Mortar Network 0.054 (0.108)0.313∗∗ (0.139)
Hotline Service 0.110 (0.139)0.274 (0.198)
Alternative Medicine 0.062 (0.072)0.325∗∗∗(0.126)
Other Non-Essential Beneﬁts 0.036 (0.060)0.093 (0.178)
Health Plan Fixed Effect yes yes
# Observations (switches×#plans) 98,224 98,224
# Health Plan Switches 1,706 1,706
Notes: Authors’ calculation. Panel A shows OLS estimation results on the health plan level using annual changes in the risk pool (age and health
status) as instruments for annual changes in contribution rates. Panel B shows the standard mixed logit model on the enrollee level which controls
for the residual of the ﬁrst stage. Estimated standard errors in the second stage are in parentheses and corrected using the Murphy and Topel (1985)
correction for two-step models. ∗∗∗ p<0.01; ∗∗ p<0.05; ∗p<0.1
Table 7: Characterizing Switchers by Risk Types (Mixed Logit Models)
Age Gender SAH
(G1: age <50) (G1: males) (G1: SAH <3)
Mean SD Mean SD Mean SD
Price×G1−0.063∗∗∗(0.009)0.001 (0.015)−0.054∗∗∗(0.010)0.015 (0.028)−0.059∗∗∗(0.009)0.005 (0.020)
Price×G2−0.053∗∗∗(0.013)0.000 (0.028)−0.070∗∗∗(0.010)0.004 (0.015)−0.069∗∗∗(0.011)0.002 (0.022)
Stores×G10.056 (0.112)0.350∗∗ (0.145)0.083 (0.122)0.317∗(0.177)0.041 (0.119)0.302∗(0.183)
Stores×G20.004 (0.134)0.154 (0.398)−0.008 (0.112)0.264 (0.192)0.054 (0.124)0.422∗∗∗ (0.147)
Hotline×G10.187 (0.154)0.442∗∗∗(0.158)+0.128 +(0.167)+0.384∗∗+(0.178)+0.307∗+(0.166)+0.560∗∗∗+(0.155)
Hotline×G2−0.069 (0.172)0.153 (0.277)+0.236 +(0.174)+0.519∗∗∗+(0.172)+−0.098 +(0.149)+0.114 +(0.329)
Alternative Medicine×G10.064 (0.074)0.325∗∗∗(0.123)+0.058 +(0.087)+0.523∗∗∗+(0.105)+0.023 +(0.080)+0.499∗∗∗+(0.112)
Alternative Medicine×G20.021 (0.114)0.272 (0.262)+0.064 +(0.074)+0.030 +(0.219)+0.170∗+(0.090)+0.164 +(0.222)
Non-Essential Beneﬁts×G10.060 (0.065)0.143 (0.158)+0.121 +(0.077)+0.178 +(0.156)+0.083 +(0.069)+0.150 +(0.142)
Non-Essential Beneﬁts×G20.072 (0.114)0.357∗∗ (0.216)+0.042 +(0.077)+0.319∗∗+(0.141)+0.058 +(0.089)+0.375∗∗+(0.168)
Health Plan Fixed Effects yes yes yes
# Observations 99,888 99,888 99,790
# Health Plan Switches 1,726 1,726 1,724
Notes: Authors’ calculation. The table shows estimated means and standard deviations of the random parameters. G1and G2denote good health risks (age <50; males,
SAH <3) and bad health risks (age ≥50; females, SAH ≥3), respectively. +signiﬁcant differences (90% level) between distributions of estimated preference parameters.
Estimated standard errors are in parentheses. ∗∗∗ p<0.01; ∗∗ p<0.05; ∗p<0.1.
Table A1: Robustness—The Price Framing Reform of 2009 as Exogenous Shifter in the Decision to Switch
Premium ×pre −0.056∗∗∗(0.009)0.001 (0.014)
Premium ×post −0.092∗∗∗(0.031)0.069 (0.042)
Branch Network ×pre −0.010 (0.119)0.276 (0.200)
Branch Network ×post 0.203 (0.139)0.475∗∗∗ (0.167)
Hotline Service ×pre 0.152 (0.189)0.428∗(0.223)
Hotline Service ×post −0.008 (0.171)0.125 (0.341)
Alternative Medicine ×pre 0.073 (0.115)0.260 (0.168)
Alternative Medicine ×post 0.093 (0.107)0.651∗∗∗ (0.177)
Non-Essential Beneﬁts ×pre 0.010 (0.068)0.012 (0.180)
Non-Essential Beneﬁts ×post 0.076 (0.102)0.136 (0.167)
Health Plan Fixed Effects yes
# Observations 99,888
# Health Plan Switches 1,726
Notes: Authors’ calculation. The table shows estimated coefﬁcients of both, mean and standard deviation.
The binary indicator “pre” is one for the SOEP waves 2008 and 2009, while “post” is one for the waves
of 2010 and 2011. Estimated standard errors are in parentheses. ∗∗∗ p<0.01; ∗∗ p<0.05; ∗p<0.1
Table A2: Dominated vs. Non-Dominated Plan Characteristics–Only Price as Dominance Criterion
All Dominated Non-dominated
health plans health plans health plans
Mean S.D. Mean S.D. Mean S.D.
Contribution rate 14.41 0.75 14.43 0.73 13.60 1.16
Branch Network 5.71 2.47 5.74 2.49 4.53 0.83
Hotline Service 19.13 5.65 19.11 5.69 19.72 4.20
Alternative Medicine 4.97 4.07 5.03 4.10 2.88 1.30
Other Non-Essential Beneﬁts 12.80 4.93 12.78 4.92 13.50 6.02
Share of Female Enrollees 0.46 0.19 0.46 0.20 0.54 0.14
Average Age of Enrollees 43.04 3.63 43.12 3.64 40.09 1.76
Share of Enrollees SAH good 0.56 0.20 0.55 0.20 0.63 0.09
Observations 154 150 4
uller and Lange (2010); Lange (2011), German Federal (Social) Insurance Ofﬁce, National
Association of Statutory Health Insurance Funds, annual reports of the sickness funds, information by
sickness funds. SOEP.v28. Authors’ calculation based on the SOEP and 154 plan-year observations. Only
plans that operate nationwide have been considered.