RISK MISPERCEPTIONS AND SELECTION IN INSURANCE MARKETS: AN APPLICATION TO DEMAND FOR CANCER INSURANCE

• Published date: 01 September 2018
• DOI: http://doi.org/10.1111/jori.12180
• Date: 01 September 2018
• Author: David Hales,Mary Riddel

RISK MISPERCEPTIONS AND SELECTION IN INSURANCE

MARKETS:ANAPPLICATION TO DEMAND FOR CANCER

INSURANCE

Mary Riddel

David Hales

ABSTRACT

We test for the influence of optimism on selection in a hypothetical cancer-

insurance market using a survey of 474 subjects. We elicit perceptions of

baseline cancer risk and control efficacy and combine these with subject-

specific cancer risks predicted by the Harvard Cancer Risk Index to develop

measures of baseline and control optimism. Following Fang, Keane, and

Silverman (2008), we hypothesize that a variable may lead to advantageous

selection if it is positively correlated with both prevention effort and demand

for insurance. We find evidence of selection from both baseline and control

optimism, but little evidence of selection from cognitive ability and risk

aversion. We then estimate a model that allows us to classify subjects

according to their excess risk-reducing effort and excess insurance uptake

that occurs solely because of baseline and control optimism. We find subjects

who overinsure relative to a subject with accurate risk beliefs are likely to

have lower expected treatment costs, ceteris paribus. This indicates that on

net, baseline and control optimism leads to advantageous selection in our

sample.

Economists have long posited that asymmetric information with heterogeneous risk

types can lead to adverse selection in insurance markets (Rothschild and Stiglitz,

1976). Individuals with private information that they are high-risk types tend to buy

more coverage than low-risk types. High-risk types will also have higher claims,

leading to a positive correlation between insurance coverage and claims. Such

positive correlations have been found in some markets but rejected in others. For

example, Puelz and Snow (1994) find evidence of adverse selection in the market for

Mary Riddel is Professor of Economics at the Department of Economics, Lee Business School,

4505 S. Maryland PKWY, Box 6005, University of Nevada, Las Vegas, NV 89154-6005. Riddel

can be contacted via e-mail: mriddel@unlv.nevada.edu. David Hales is a graduate student at

the Department of Economics, University of California, Santa Barbara. The authors would like

to acknowledge helpful comments from Paul Thistle; the Harvard Risk, Perception, and

Response Conference held in March 2014; the CEAR/MRIC Behavioral Insurance Workshop in

Munich held in December 2014; and two anonymous referees.

© 2016 The Journal of Risk and Insurance. Vol. 85, No. 3, 749–785 (2018).

DOI: 10.1111/jori.12180

749

automobile insurance, but Chiappori and Salanie (2000) do not. Finkelstein and

McGarry (2006) reject the hypothesis of adverse selection in their study of long-term

care insurance.

These mixed findings concerning adverse selection have led researchers to look for

other sources of private information and selection in insurance markets. De Meza and

Webb (2001) develop a model where risk aversion leads to advantageous selection as

more risk-averse subjects buy more insurance and simultaneously engage in more

prevention behavior, leading to a negative correlation between coverage and claims.

Fang, Keane, and Silverman (2008) examine the market for Medigap insurance

(a supplement to Medicare). Controlling for cognitive ability, they find a negative

correlation between coverage and ex post claims that is indicative of advantageous

selection. They conclude that the correlation arises because cognitive ability is

correlated with both good health and the purchase of health insurance.

A recent article by Spinnewijn (2013) posits that heterogeneity in risk misperceptions

may also affect the relationships between coverage and claims. Spinnewijn recognizes

two dimensions of risk misperception. First, subjects may be relatively “baseline

optimistic,” meaning they believe their risk of damages is lower than it actually is.

Baseline optimistic subjects demand less insurance than their more pessimistic

counterparts. “Control optimists” overestimate the risk reductions from engaging in

preventative activities and avoiding risky activities. As a result, they overinvest in

risk-reducing activities relative to their true risk type, leading to lower expected

insurance claims. Assuming a simple model with two insurees with different

perceived risk types and incentive-compatible equilibrium contracts, Spinnewijn

shows that if one insuree is both more baseline optimistic and more control optimistic

than the other, a positive correlation between coverage and claims will occur. A

negative correlation results if instead the more control pessimistic type is also

relatively more baseline optimistic. Thus, depending on the net effect of control and

baseline optimism, either adverse or advantageous selection may result.

Ideally, we could test for selection arising from baseline and control optimism and

other possible selection variablesusing the Fang, Keane, and Silverman (2008) test for

multidimensional selection.They maintain that for a variable to lead to advantageous

selection it must have two properties: (1) the variable is positively correlated with

coverage and (2) the variable is negatively correlated with insurance claims.

Unfortunately, we are not aware of a database that includes subject-level coverage

and claims data as well as measures of controland baseline optimism related to cancer

risk and insurance. As such, we take a novel approach to investigating selection in

insurance markets. Rather than analyzing historical insurance coverage and claims

data, we examinethe relationship between prevention effort (a proxyfor ex post claims)

and willingness to pay for an insurance contract that covers all of the costs of cancer

treatment (a proxy for demand for insurance coverage). Prevention effort is a

reasonable proxyfor expected claims as increased prevention effort strongly correlates

with lower expected claims. Demand for insurance is increasing in willingness to pay

for a fixed contract, making willingness to pay a good proxy for coverage.

Using an online survey of 474 U.S. adults aged 18 and older, we test for

multidimensional selection in a hypothetical cancer insurance market arising from

750 THE JOURNAL OF RISK AND INSURANCE

baseline and control optimism, risk aversion, and cognitive ability while controlling

for observable demographic variables. We first elicit measures of baseline risk

perceptions and perceptions of the efficacy of prevention efforts for colon, prostate,

and bladder cancer in men, and colon, breast, and bladder cancers in women. We

query subjects about their behaviors that may either reduce or increase risks for

these cancers, allowing us to devise a measure of prevention effort. We elicit estimates

of the subject’s degree of risk aversion using Holt and Laury’s (2002) multiple

price-list elicitation method. We measure cognitive ability using a short intelligence

assessment.

1

Finally, we ask subjects if they would be willing to pay some stated

premium Pfor insurance that would cover all their cancer costs. This final question

allows us to develop a measure of willingness to pay for insurance as a function of the

selection and demographic variables.

To our knowledge, no other research has empirically evaluated the role of control and

baseline optimism in selection in insurance markets. This is likely because data on

actual and perceived risk as well as actual and perceived prevention efficacy that are

needed to create optimism variables are generally not available. Thus, the strength of

this study is that our measures of cancer risk misperception rest on combining

subjects’ survey responses with the Harvard Cancer Risk Index (HCRI) (Colditz et al.,

2000) to form measures of baseline and control optimism. The HCRI was developed at

the Harvard Center for Cancer Prevention by a working group of “epidemiologists,

clinical oncologists, and other Harvard faculty with quantitative expertise focused on

cancer and risk assessment” (Colditz et al., 2000). The HCRI provides quantitative

relative risk (RR) factors for each demographic or behavioral attribute that experts

believe bear on the risk of incidence of a given cancer. The HCRI can thus be used to

calculate the risk a subject will contract cancer, conditional on their behavioral and

demographic traits. We derive a measure of baseline optimism by asking subjects to

estimate their risk of incidence of each cancer and comparing this with corresponding

HCRI estimates. Similarly, we derive a measure of control optimism by asking

subjects how effective a series of preventative measures are in reducing cancer risk,

and how risky a series of unhealthy behaviors are, and then comparing their response

with corresponding HCRI RR factors.

Like Fang, Keane, and Silverman (2008), we hypothesize that variables that are

positively correlated with willingness to pay for insurance and prevention effort lead

to advantageous selection in the insurance market. Adverse selection results if a

variable is positively correlated with willingness to pay and negatively correlated

with effort. We first estimate models of the willingness to pay for insurance as a

function of baseline optimism, controlling for observable demographic variables. We

1

Recent articles find that a large proportion of experimental subjects have risk preference

functions that are best described by cumulative prospect theory where probability weighting

and loss aversion together with standard utility function-based risk aversion define

preferences (Conte, Hey, and Moffatt 2011; Barseghyan et al., 2013). In the current article,

we explicitly account for utility-based risk aversion. Some argue that optimism is a form of

probability weighting (e.g., Quiggin, 1993) and thus we also explicitly account for probability-

based risk aversion. Since the risks in the current article are defined solely over losses,

accounting for loss aversion is not necessary.

RISK MISPERCEPTION AND SELECTION 751