HYPOTHETICAL SURVEYS AND EXPERIMENTAL STUDIES OF INSURANCE DEMAND: A REVIEW

• Author: Johannes G. Jaspersen
• Date: 01 January 2016
• DOI: http://doi.org/10.1111/jori.12100
• Published date: 01 January 2016

©2015 The Journal of Risk and Insurance. Vol.83, No. 1, 217–255 (2016).

DOI: 10.1111/jori.12100

Hypothetical Surveys and Experimental Studies of

Insurance Demand: A Review

Johannes G. Jaspersen

Abstract

We offer a structured literature survey of experimental studies involving

insurance demand choices and their experimental methodology. With this,

we aim to fulfill two goals. First, we want to give an overview of the status

of the literature as is. Second, the overview of the methodology provides

researchers with an idea of how insurance demand experiments can be

designed and what the advantages and disadvantages of the different de-

sign aspects are. Wethus offer a resource for the design of future experiments.

Introduction

The primary motivation for studying any economic system in a laboratory setting is

the amount of control such a setting grants for the collection of empirical data (Smith,

1976). In contrast to other ways of observing economic behavior, the researcher is

able to control most aspects of the decision situation and can, to a certain extent,

measure the preferences of the decision makers. This advantage has led to a variety

of experimental studies on insurance demand with a great variety in both research

questions and employed methodology. This article offers a comprehensive overview

of prior studies in the field. It aims to give both an overview of the studied topics and

to discuss the experimental designs employed.

In his seminal article, Smith (1982) discusses the precepts for laboratory experiments

in economics. He particularly emphasizes the importance of salient payoffs. Crudely

speaking, experimental subjects should receive rewards (most commonly money) for

their participation in the experiment and these rewards should be increasing(decreas-

ing) in the good (bad) outcomes in the experiment. However, when considering the

literature, it is apparent that many studies of insurance demand have not included

Johannes G. Jaspersen is at the Munich School of Management, Ludwig-Maximilians-

Universit¨

at of Munich, Schackstr. 4, 80359 Munich, Germany. Jaspersen can be contacted via

Tel.: +49 (0)89 2180 2792; e-mail: jaspersen@bwl.lmu.de. I would like to thank the editor of

the Journal of Risk and Insurance; Keith Crocker, the guest editor of this special issue; Glenn

Harrison; and an anonymous reviewer for helpful comments. Thomas Kelly, Stefan Neuß,

Mark Ragin, and Harris Schlesinger provided immensely useful feedback and suggestions. All

remaining errors are mine.

217

218 The Journal of Risk and Insurance

salient payments and thus, are not experiments in the sense of Smith. Nevertheless,

these hypothetical studies can provide preliminary evidence, so they are still included

in this review. We will, however, highlight the potential weaknesses of such studies.

Irrespective of whether or not remuneration is salient, the studies reviewed here are

discussed with respect to other aspects of their methodology. Whether subjects receive

a neutral description of the decision (in terms of probabilities and possible gains

and losses) or the instructions feature a more practical description of an insurance

purchasing decision can influence decisions. Similarly, the exact nature of the decision

task with which the subjects declare their insurance demand has varied widely in the

reviewed articles and can make a difference in the results, as well. We discuss both of

these design aspects with references to known general and insurance specific results.

In most experiments, rewardsnot only need to be salient, they also need to be incentive

compatible. This implies that the rewards are structured in a way that subjects are

given incentives to make decisions in the experiment according to their own true

preferences. We discuss multiple procedures that implement incentive compatibility

and their behavioral implications.

This review is intended to serve two purposes. First, it makes it possible for peo-

ple unfamiliar with the literature to get an impression of which questions have been

addressed by experiments and which fields have not yet been covered in detail. Ad-

ditionally, with the discussion of important design aspects in the reviewed studies,

we provide a framework for the design of insurance demand experiments. Future re-

search can thus utilize our results for both the conceptualization and implementation

of new studies.

The rest of the article is structured as follows. The next section contains the method-

ology and results of the structured literature review. The “Areas of Research” section

discusses the different topics that have been addressed by experimental studies so

far. The “Design Issues in Insurance Demand Studies” section discusses the above-

mentioned design aspects; this section also provides results of general and insurance-

related literature on these design issues. The last section concludes.

Structured Literature Review

This review aims to identify what topics have been addressed and which design ele-

ments are common in insurance experiments. For this, we reviewed the high-ranking

insurance economics journals as well as articles published in certain general economics

journals since 1980. Wealso added recent working papers to capture the most current

research. We considered only studies meeting certain criteria (e.g., that the studies are

on insurance demand and that they study preferences on individual outcomes).

It needs to be emphasized that this literature review is not necessarily completely

comprehensive. The selection of the journals as well as the choice whether or not to

include a specific study is naturally subjective. However, the studies selected for the

review should be sufficient to establish an overview of the literature in terms of both

topics and experimental designs.

Hypothetical Surveys and Experimental Studies of Insurance Demand: A Review 219

The 95 articles surveyed here are listed in Table 1. Articles often report more than

one experimental study or hypothetical survey. We number these studies in order of

appearance in the article and report the relevant study number in the column Study.

It is also indicated how many studies the article includes in total. Wereport only those

studies within an article that concern insurance demand.1For every study within an

article, the table indicates whether it is an experiment (i.e., features salients payments)

or a hypothetical survey in the column Type.

Differentiating experiments from hypothetical surveys is not only a semantic issue.

Providing salient incentives can and often will change the behavior of the subjects.

Camerer and Hogarth (1999) conduct a literature review of 74 studies that vary in-

centives in their experimental design. The general finding of their study implies that

the existence and magnitude of salient incentives does affect the behavior of subjects

in experiments. The authors attribute this effect to an alignment of the subjects’ pref-

erences with the task at hand. A simple example of this is the choice between a risky

lottery and a riskless option. A subject can be assumed to have two goals: one is the

maximization of their preference functional over the monetary consequences of their

choice; the other is the minimization of boredom during the experiment. Since the

risky option of the lottery provides some excitement regarding the outcome, while

the sure option does not, one can see that choosing the risky option will bring some ful-

fillment of the second goal. Providing salient (most often monetary) rewards will lead

the subject to focus more on the first goal than on the second one. This effect increases

with the size of the incentives (Holt and Laury, 2002, 2005; Harrison et al., 2005).

Not only are subjects more risk averse in experiments, their behavior might also

change on a more fundamental level. Several of the violations of expected utility

theory reported in hypothetical surveys do not appear or are severely diminished

when the studies are repeated as experiments (Harrison, 1994; Lefebvre et al., 2011).

This observation also holds for studies on insurance demand. Laury et al. (2009) report

more demand for insurance against low-probability, high-consequence risks than for

insurance against high-probability, low-consequence risks when salient rewards are

present. As discussed below, such behavior is aligned with expected utility theory

(Browne et al., 2012). In hypothetical surveys, this behavior is not as clear cut or is

even reversed (Slovic et al., 1977).

Decisions of subjects in hypothetical surveys of insurance demand also seem to have

more noise associated with them (Camerer and Hogarth, 1999). Irwin et al. (1992)

show that bids in the Vickrey auction for insurance policies in their experiment are

more variable in studies without salient rewards. Similarly, the responses to a hypo-

thetical choice set of prescription drug coverages in Kesternich et al. (2013) render

coefficients in a multinomial logit estimation that have a much higher variance than

the corresponding coefficients from naturally occurring choice data.

1Thus, sometimes not all studies from a certain article arecovered. For example, only the second

of three studies in Arkes et al. (1988) investigates insurance demand. This is indicated by the

entry 2of3. In cases in which the design elements of multiple studies within one article are

consistent, these studies are reported in a single row.