ArticlePDF Available

Does the Threat of Insurer Liability for “Bad Faith” Affect Insurance Settlements?

Authors:

Abstract and Figures

Economic reasoning predicts that policyholders in states that treat for insurer bad faith in settling claims as a tort should receive higher payments from insurers because of the greater potential damages insurers face in claims disputed in court. We test this hypothesis using data on automobile insurance claims for accidents occurring during 1972–1997, exploiting differences in states “laws and variation in timing of states” adoption of bad faith rules to identify the effects of tort liability. We find that the presence of tort liability for insurer bad faith increases settlement amounts and reduces the likelihood that a claim is underpaid.
Content may be subject to copyright.
© The Journal of Risk and Insurance, 2014, Vol. 81, No. 1, 1–26
DOI: 10.1111/j.1539-6975.2012.01499.x
1
DOES THE THREAT OF INSURER LIABILITY FOR “BAD
FAITH”AFFECT INSURANCE SETTLEMENTS?
Danial P. Asmat
Sharon Tennyson
ABSTRACT
Economic reasoning predicts that policyholders in states that treat for insurer
bad faith in settling claims as a tort should receive higher payments from
insurers because of the greater potential damages insurers face in claims
disputed in court. We test this hypothesis using data on automobile insurance
claims for accidents occurring during 1972–1997, exploiting differences in
states “laws and variation in timing of states” adoption of bad faith rules to
identify the effects of tort liability. We find that the presence of tort liability
for insurer bad faith increases settlement amounts and reduces the likelihood
that a claim is underpaid.
INTRODUCTION
U.S. common law has long recognized the unequal bargaining power of insurance
companies and policyholders in the insurance relationship. The insurer not only
writes the contract terms, but settlements are negotiated at a time when the insured is
particularly vulnerable. For these reasons insurers are held to high standards of “good
faith” in dealings with policyholders (Jerry, 1994). Nevertheless, until relatively re-
cently policyholders who were treated unfairly in claims settlement had few available
legal remedies. Legal disputes over insurance claims were settled under standards
set in 19th-century English common law,1which limited policyholders to recovering
only the amounts specified in the insurance policy even if the insurer intentionally
breached the contract.
Over the past 30 years the compensation available to policyholders in cases where
insurers violate good faith standards has increased. A majority of states now recognize
Danial P. Asmat is a Ph.D. student in the Department of Business Economics and Public
Policy at the University of Michigan. Sharon Tennyson is Associate Professor of Policy Anal-
ysis and Management at Cornell University. The authors can be contacted via e-mail: da-
nial.asmat@gmail.com and sharon.tennyson@cornell.edu, respectively. The Insurance Research
Council (IRC) provided data for the empirical analysis. Useful comments were received from
Tom Baker and from seminar participants at Cornell Law School, Georgia State University,
Florida State University, Temple University, the World Risk and Insurance Economics Con-
ference, and the ARIA session at the Allied Social Science Association meetings. The authors
remain solely responsible for the content of the work.
1See the discussion in Tennyson and Warfel (2010).
2 THE JOURNAL OF RISK AND INSURANCE
the right of policyholders to file private lawsuits against insurers alleging unfair claim
settlement practices. However, the legal philosophies under which these cases may
be brought vary across the states. Some states view an insurer’s bad faith in settling an
insurance claim as a contract breach, but other states consider insurer bad faith to be a
tort (Stempel, 2008; GenRe, Palmer, and Dodge, 2008). Under tort law an injured party
may recover for all harm or injuries sustained, including legal expenses, economic
loss, and mental distress. Punitive damages may also be awarded if the conduct giving
rise to liability was particularly egregious. Thus, tort standards greatly increase the
potential costs to an insurer for bad faith dealings in settling a claim.
The theoretical law and economics literature posits that differences in the litigation
environment will alter the bargaining game for claims settlements between insurance
companies and their policyholders.2The threat of greater expected financial penalties
in the event of a bad faith ruling by the courts increases the relative bargaining power
of policyholders and reduces the incentives of an insurer to deny, delay, or underpay
claims (Abraham, 1994; Sykes, 1996; Crocker and Tennyson, 2002). If this threat is
salient, policyholders in states that permit tort liability for insurer bad faith should
receive more favorable insurance settlements.
We test this hypothesis using a large sample of automobile insurance claims from
accidents occurring in 42 states over the period 1972–1997 when tort liability for
insurer bad faith was expanding.3The use of repeated cross-sectional data allows us
to exploit the differences in states’ dates of adoption of tort standards to identify its
effects on insurance settlement amounts. Ours is the first study of bad faith regimes
to use this approach.4This research design addresses the difficulty in identifying
the effects of laws based on cross-sectional differences across states with different
laws. It also permits examination of the long-run effects of tort liability for insurer
bad faith. We test for statistically significant effects of tort-based bad faith regimes
on settlement amounts and examine how settlement generosity evolves over time in
relation to changes in states’ bad faith regimes.
We find that tort liability for insurer bad faith is associated with higher settlement
amounts and that this increase is statistically significant even after accounting for po-
tential changes in claimed loss amounts when bad faith liability is expanded. We find
further that tort liability reduces the likelihood of claims being underpaid. Examining
changes in the impact of bad faith liability over time, we find little evidence that its
effects on settlements have diminished over time. These results provide new evidence
that tort liability for first-party-insurer bad faith has real economic consequences for
the settlement of insurance claims.
The remainder of the article is organized as follows. After providing background on
liability regimes for insurer bad faith, we discuss the predicted implications of tort
2Cooter and Rubinfeld (1989) and Spier (2007) provide comprehensive reviews of the theoretical
law and economics literature on litigation and dispute resolution.
3The first court decision that allowed the application of tort liability to first-party insurance
bad faith was decided by the California Supreme Court in 1973. Courts in 27 states followed
this precedent between 1973 and 1997.
4However, Browne and Schmit (2008) use this research design to examine litigation patterns
over time.
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 3
liability for insurance settlements and the findings from previous empirical studies.
We next describe our research design and the relationship of our study to prior
research on the effects of state bad faith regimes, and then turn to a description of the
data and empirical models used in our study. The final sections of the article report
our empirical findings and their interpretation.
BACKGROUND
State Bad Faith Regimes
Tort liability for insurer bad faith was first adopted in the third-party insurance
(i.e., liability insurance) context in the late 1950s.5Third-party insurance provides
compensation for losses that the insured has caused to another party. In third-party
insurance, the courts have reasoned that insurers must be held to stringent standards
of fair dealing because of their disproportionate ability to influence the acceptance or
rejection of a settlement offer made by a claimant. Third-party insurers have a “duty
to settle,” and this standard requires an insurer to consider the insured’s interest in
addition to its own in deciding whether to accept or reject a settlement offer.6
First-party insurance provides compensation for losses experienced by the insurance
policyholder. In the 1973 Gruenberg v. Aetna Insurance Company decision, the California
Supreme Court extended the tort of bad faith to include first-party insurance coverage
disputes. The court reasoned that the insurer’s duty to settle in third-party claims and
in first-party claims “are merely two different aspects of the same duty,”7and so an
insurer’s breach of the good faith duty to settle in first-party claims may also give
rise to a tort action. This is an expansive application of tort because it applies to a
payment relationship governed by an explicit contract (the insurance policy) but does
not require a separate tort in order for the plaintiff to recover punitive damages.
Not every state recognizes tort actions for insurer bad faith. A number of states
have rejected the application of tort law in such cases but do allow private actions
against insurers under contract law. Many states that adjudicate first-party insurance
bad faith cases under contract law allow a broad definition of damages that include
damages such as prejudgment interest and legal expenses, whereas a few allow only
a narrow definition of damages. However, in all cases the contract standard removes
the possibility of punitive damages unless an independent tort such as fraud or
intentional infliction of emotional distress can be proven.
Another set of states has adopted legislation that permits a private cause of action
in cases of insurer bad faith, usually under an Unfair Trade Practices Statute. In
these states the statute may explicitly permit private lawsuits, or the courts may have
recognized an implied private cause of action under the statute. However, in most
states the damages allowed under statute are more limited than those in states that
recognize insurer bad faith as a separate tort (Browne, Pryor, and Puelz, 2004; GenRe,
Palmer, and Dodge, 2008).
5Browne, Pryor, and Puelz (2004) and Tennyson and Warfel (2010) provide more detailed
discussion of the history and variety of state bad faith regimes.
6See Comunale v. Traders and General Insurance Company, 50 Cal 2d. 654, 328 P. 2d 198 (1958).
7See Gruenberg v. Aetna Insurance Company, 510 P. 2d 1032 (Cal. 1973).
4 THE JOURNAL OF RISK AND INSURANCE
Law and Economics Framework
From a law and economics perspective, permitting bad faith lawsuits against insurers
may help insurers to commit to acting in good faith in claims settlement, by imposing a
direct financial penalty for doing otherwise (Abraham, 1986). This may be a necessary
supplement to the forces of market discipline. Competitive market forces provide
pressure for insurers to pay claims fairly and to settle promptly because an insurer
that systematically denies or underpays claims will face a reduction in demand for
their products (Sykes, 1996). Nonetheless, there is no guarantee that the negative
demand response will be large enough, quick enough, and/or long-enough lasting
to eliminate insurers’ incentives for bad faith practices in all cases.
Bad faith lawsuits may supplement market forces because insurers will also consider
the expected costs of dispute resolution when determining a claims settlement strat-
egy. Costs of dispute resolution depend on the likelihood that a policyholder actively
disputes the insurer’s settlement, the transactions costs of adjudicating the dispute,
and the potential settlement costs to the insurer if the dispute is resolved in the poli-
cyholder’s favor. Tort liability for insurer bad faith increases the expected value of a
disputed claim in litigation by decreasing the standards for a finding of bad faith and
by increasing the potential damages awarded to the policyholder. This increases the
insurer’s expected costs of dispute resolution and thus reduces expected benefits of
claims denial or underpayment (Sykes, 1996).
However, not every allegation of insurer bad faith arises from deliberate attempts by
an insurer to deny or underpay an obviously legitimate claim. Many claim disputes
between insurers and policyholders arise because of honest differences in beliefs
about the validity of a claim or the value of the losses experienced (Tennyson and
Warfel, 2010). Thus, a policyholder’s allegations of insurer bad faith may arise even
if the insurer believes it is acting in utmost good faith. In this situation the impact
of bad faith liability is to increase the relative bargaining power of the policyholder
in settlement negotiations. The possibility of a bad faith allegation increases the
expected value of the claim in litigation, increasing the threat point of the policy-
holder. This increased bargaining power will lead to more favorable settlements for
policyholders.
A potential unintended consequence of the threat of bad faith litigation over disputed
claims is that it will increase pressure on insurers to pay “reasonably disputable
claims” (Abraham, 1986). Insurers balance the benefits of reducing fraud costs by
contesting debatable claims against the expected costs of litigation associated with
such disputes (Sykes, 1996; Crocker and Tennyson, 2002). By increasing the expected
costs of litigation to insurers, tort liability for bad faith may reduce insurers’ incentives
to resist fraudulent claims and in turn raise the expected payoff to policyholders from
filing fraudulent claims. In this way insurer bad faith liability may also lead to higher
claimed losses (because of exaggeration of loss amounts), or to a higher prevalence
of claims with characteristics that are easier to falsify (Tennyson and Warfel, 2010).
In sum, by changing the balance of power between insurers and policyholders in
the settlement process, tort liability for bad faith may affect insurance settlements in
various ways. One specific prediction is that settlement amounts in states that permit
the tort of bad faith will be higher, holding all other factors constant. However, the
increased pressure on insurers to pay claims may also affect policyholders’ claiming
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 5
behavior, which means that claimed loss amounts and claim characteristics may
also be affected by the bad faith regime. When analyzing the effect of bad faith on
insurance settlement amounts we must be careful to control for differences in claim
characteristics as well as for the amount of claimed loss. Observed settlements may
be higher because insurers are more generous relative to a given claim amount or
because claimed amounts are higher, and this distinction is important.
A second set of predictions relates to the varying impact of bad faith liability on
insurer payment incentives relative to the claimed amount of loss. A policyholder
who is denied payment or who receives payment for far less than the full amount of
losses claimed will have a greater incentive to allege insurer bad faith. This means
that there is a larger threat of bad faith actions for settlement amounts that are farther
below the amount claimed by the insured. This suggests that another effect of bad
faith liability is that insurers will be less likely to severely underpay claims that they
choose to settle with some positive payment.
Empirical Literature
Only a few studies have attempted to empirically assess the impact of the threat of bad
faith on insurance settlements.8Because of the ready availability of data, most have
focused on private passenger automobile insurance. An exception is Hyman, Black
and Silver (2011), who examine medical malpractice and other commercial liability
claims from the state of Texas. These authors test the hypothesis that the potential for
bad faith liability in third-party claims provide insurers with an incentive to settle
claims at the insurance policy limit if the expected value of the claim at trial exceeds
that limit. Their results are consistent with this hypothesis, as many claims are settled
at policy limits and those claims settle more quickly than other claims.
Browne, Pryor, and Puelz (2004) provide the first analysis of the effect of first-party
insurance bad faith on automobile insurance payments. Using data on injury claims
settled under uninsured motorist (UM) policies from 38 states in 1992, the authors
test the hypothesis that settlement amounts are larger in states that permit private
actions for insurer bad faith. After controlling for characteristics of the claim, their
results support this hypothesis. They find additionally that the larger settlement
amounts occur in both the economic and noneconomic damages portions of the
settlements. However, use of cross-sectional data means that the study cannot rule
out the possibility that other state characteristics correlated with the bad faith regime
account for differences in settlement outcomes.
Hawken, Carroll, and Abrahamse (2001) study the effects of bad faith tort liability for
third-party automobile injury claims and find similar results. Their study focuses on
the particular effects of California court rulings, which allowed injured third parties
to file bad faith tort actions against liability insurers during years 1979–1988 only.9
The authors compare insurance settlements in California with those in other states
8There is, of course, a large empirical literature on the more general question of the impact of
legal settlement regimes on insurance or litigation outcomes. See, for example, Kessler (1995),
Lee, Browne, and Schmit (1994), White (1989), and Chang and Sigman (2000).
9In Royal Globe Insurance Company v. Superior Court (Koeppel), 592 P.2d 329 (Cal. 1979) the court
determined that the third-party claimant in a liability case could bring a bad faith action
against the liability insurer; that is, such actions were not reserved only for the insured party.
6 THE JOURNAL OF RISK AND INSURANCE
when the rule was in effect, and after the rule was overturned. Consistent with the
hypothesis that tort liability will lead to higher settlement amounts, estimates using
Insurance Research Council (IRC) claims data show that settlements in California
were 25 percent higher than similar claims in other states during 1979–1988, but not
after 1988 when the rule was overturned. Using aggregate state-wide data on claims
rates, the authors also find that the number of third-party automobile injury claims
in California was higher than in other states during 1979–1988 but not after 1988.
Unfortunately, the California court ruling that overturned the expanded tort liability
occurred in the same year as passage of Proposition 103, which made major changes
to insurance rate regulation in the state. The coincidence of these two major changes
in the legal environment for insurance claims makes it difficult to disentangle the
effects of bad faith liability changes and regulatory changes.
Tennyson and Warfel (2008, 2010) more directly examine the hypothesis that expanded
bad faith liability increases pressure on insurers to pay fraudulent or exaggerated
claims. Using data on first-party automobile insurance claims settled in 1997, these
authors compare selected characteristics of claims settled in states with tort-based bad
faith to claims settled in other states. Specifically, the authors examine the prevalence
of claim characteristics often associated with fraud (so-called fraud “red flags”), and
the frequency of insurer claim investigations, in these two sets of states. They find
that permitting private lawsuits for insurer bad faith is associated with more settled
claims that contain fraud “red flags” and with less intensive use of investigations
by insurers. The largest effects are observed in states that adjudicate bad faith suits
under tort liability. These findings are based on comparisons of mean values across
states but hold for different quartiles of the claim distribution as well. Nonetheless,
the use of cross-sectional data and univariate comparisons leaves the possibility that
other important sources of heterogeneity drive the results.
RESEARCH METHODS
This study examines the effects of first-party-insurer bad faith liability on insurance
settlement amounts using a large sample of first-party automobile insurance claims
for accidents occurring in different states and years. Combining data across states
and years permits a comparison of settlements before and after tort-based bad faith
regimes are in effect in a state, relative to settlements in other states in the same years.
Thus, the empirical estimates exploit differences in states’ bad faith regimes, and the
fact that states adopted bad faith liability rules at different points in time, to identify
the effect of bad faith liability on insurance settlements.
Based on insights from the law and economics literature on liability assignment,
we hypothesize that tort-based bad faith regimes will have the largest impact on
insurance settlement negotiations. Thus, we measure the impact of expanded liability
for first-party-insurer bad faith by comparing claims settled in states and years in
which insurer bad faith is recognized as a tort, to claims settled under other bad faith
regimes.10 We test two primary hypotheses:
This decision was overruled in Moradi-Shalal v. Fireman’s Fund Insurance Company, 758 P.2d 58
(Cal. 1988).
10 Tennyson and Warfel (2010) note that among states that recognize first-party-insurer bad faith
as a tort, some apply a negligence standard and others apply an intentional tort standard. In
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 7
H1: Settlement amounts are higher in states that permit tort actions for first-party-
insurer bad faith, after controlling for claim characteristics.
H2: The fraction of claims that are substantially underpaid is smaller in states that
permit tort actions for first-party-insurer bad faith, after controlling for claim
characteristics.
The form of the empirical models used to test H1 and H2 are as follows:
Settlement Generosityist=θTort
st +Xistβ+Wstδ+Ztφ+as+eist,(1)
where subscript irepresents a claim, srepresents the state in which the accident
occurred, and tdenotes the year of the accident. The key independent variable of
interest is an indicator of the state’s first-party-insurance bad faith regime (Tort st ),
coded as one if the state permits tort actions in the accident year and coded as zero
otherwise.
Xist is a row vector of accident and claimant characteristics, which are expected to
affect the settlement amount, and βis a column vector of unknown coefficients. Wst
is a row vector of time-varying characteristics of the accident state that may affect
average settlement amounts in the state, and δis the corresponding column vector of
coefficients to be estimated. Ztis a row vector of controls that vary only by accident
year, included to capture the effects of national phenomena that may affect average
settlement amounts over time; φis a column vector of coefficients to be estimated.
The variable asis a state-specific fixed effect included to ensure that the estimated
effect of bad faith does not inadvertently include unobserved effects of time-invariant
characteristics of a state in which the law is present; eist is a random error term assumed
to be normally distributed.
An important consideration is whether variables included in Xist may be endoge-
nously determined with the settlement amount. We address the endogeneity issue
through the use of instrumental variables estimation approaches. Although discus-
sion of the full set of control variables is deferred to later in the article, the endogeneity
concern clearly arises with regard to policyholders’ use of attorney representation.11
If a policyholder hires an attorney as a result of initial negotiations, or because the
nature of the claim or injury leads him to expect difficult negotiations, then settlement
amounts and the presence of an attorney will be jointly determined. In some estimates
we treat this variable as endogenously determined, because we do not observe the
reason for or the timing of the representation decision. Because attorney representa-
tion is a 0–1 indicator, we estimate the models using treatment effects (endogenous
theory, the intentional tort standard means that fewer cases are likely to support an allegation
of bad faith, and a smaller set of bad faith claims will warrant punitive damages. For this
reason we investigated whether negligence-based tort bad faith leads to significantly different
outcomes from tort-based regimes based on an intentional tort standard. Results, available
from the authors, suggest that the key distinction is tort versus nontort regimes, and thus our
analysis focuses on this distinction.
11 See also Browne and Puelz’s (1996) analysis of the impact of attorney use in automobile
insurance claims.
8 THE JOURNAL OF RISK AND INSURANCE
dummy variable) estimation based on Heckman (1978) and using maximum likeli-
hood methods as described in Maddala (1983).12
The dependent variable in each of our empirical models (Settlement Generosityist)isa
measure of the amount paid by the insurer.13 The primary dependent variable specific
to H1 is the natural log of the total settlement amount, which is the (log of the) sum
of all economic and noneconomic damages paid by the insurer. Models estimated
using the total settlement amount as the dependent variable represent a reduced-
form specification that relies on the assumption that settlements can be explained by
observed accident characteristics and claimant demographics after aggregate state-
level and time-level covariates are accounted for.
However, as noted previously a state’s bad faith regime may also affect the amount
claimed by an accident victim. This implies that—even after controlling for claim
characteristics Xist—claim amounts in states with tort-based bad faith regimes may
differ from those in other states. If settlement amounts are determined in relation to
the amount of the claim filed by the policyholder, the estimated impact of tort-based
bad faith in the model with total settlement as the dependent variable incorporates
both the effects of bad faith on insurers’ payment decisions (for a given claim amount)
and any effects of bad faith on the amount claimed. To better isolate the impact of the
bad faith regime we also test H1 using as a dependent variable the (natural log of the)
settlement amount relative to the claimed amount of loss: Ln(Total Settlement/Claimed
Loss), or equivalently, Ln(Total Settlement) – Ln(Claimed Loss).
To take further advantage of the time-varying nature of our data, we explore whether
the effects of tort actions for first-party-insurance bad faith change over the sample
period of our study. We are interested in exploring both the changes in average effects
(across all states) of bad faith over time, and changes within each state over the years
that a tort regime is in effect. Although there are no clear-cut predictions from theory,
there are two reasons to expect that the effects of tort regimes may change over time.
First, we may expect a lower impact of (new) tort liability regimes that are adopted
later in time, compared to those adopted earlier in time. As tort liability becomes a
more common standard across the states, insurers may adjust their settlement regimes
in all states to anticipate the possibility of a court adopting the tort standard in another
state (e.g., Malani and Reif, 2010).
Second, effects over time within a state may change as courts refine their application
of the tort standard. Efficiency arguments imply that courts should learn over time
how to properly apply the standard; reductions in court errors of the type discussed
by Sykes (1996) should reduce insurer uncertainty and thus reduce excessive claim
payments. Moreover, in the absence of court errors, insurers will learn the court’s
standard over time and will be better able to predict what actions may lead to a bad
faith judgment. This updating may lead to increased or decreased settlement amounts
12 These estimates are produced using the treatreg procedure in Stata.
13 Throughout the article, we use the term “settlement” or “total settlement” to refer to the final
amount that the insurer agrees to pay the policyholder (claimant). In contrast, the “claimed
amount” of loss or “claimed loss” refers to the economic loss that the policyholder asks
(“claims”) from the insurer.
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 9
over time, depending on whether insurers initially overestimate or underestimate the
liability standards.
Average (across all states) changes in the effect of bad faith over the sample period are
estimated using two alternative approaches. The first is to interact the Tor t st indicator
with a time trend indicator (Year t1972). The second approach is to simply allow
the estimated coefficient θfor the Tort st indicator to vary across decades. Within-state
changes in the effect of bad faith over time are estimated by interacting Tor t st with a
time trend indicator (Year t– Firstyearst ), where Firstyearst is the first year that state s
recognized insurer bad faith as a tort. We first estimate a model in which a common
time trend is assumed for all states. We also estimate a model in which the time trend is
allowed to vary across states based on the decade during which tort actions were first
permitted in a state. The dates of state initiation of tort liability for first-party-insurer
bad faith are listed in Appendix Table A1.
To provide evidence on H2, that tort liability for insurer bad faith affects the likelihood
of a settlement that is far less than full compensation, we estimate models in which
the dependent variable is an indicator for whether a claim is settled for less than the
amount of economic loss claimed. Specifically, the dependent variable is set equal to
1ifTotal Settlement – Claimed Loss <0 and set equal to 0 if Total Settlement – Claimed
Loss 0. Linear probability and probit models using this dependent variable test the
hypothesis that the probability a claim is settled for less than the economic loss amount
is affected by the bad faith regime. These estimates allow us to distinguish whether
the threat of insurer bad faith primarily affects the marginal settlement amount for
claims that are fully paid, or also settlements for claims that would otherwise be
underpaid.14
DATA AND SAMPLE FOR ANALYSIS
Our estimates use data on individual insurance settlements, combined with data on
state bad faith liability regimes and data on other relevant features of the claiming
environment, which may affect settlement amounts.
Insurance Claims Data
The claims data are obtained from periodic nation-wide surveys of automobile injury
claims undertaken by the IRC.15 We utilize IRC data on automobile UM claims settled
in 1977, 1987, and 1997. UM insurance provides compensation for bodily injury to
a policyholder in accidents in which another driver is at fault but does not carry
liability insurance to pay for the injury costs. UM insurance is considered to be first-
party insurance and courts have specifically upheld the applicability of first-party bad
faith remedies in the UM context (Browne, Pryor, and Puelz, 2004; GenRe, Palmer,
and Dodge, 2008).
14 This is an imperfect measure of claims underpayment, because claimed amounts may be
inflated relative to the “true” economic loss. However, it still has interpretive power for H2.
15 The IRC is an independent, not-for-profit organization supported by leading
property–casualty insurance organizations. Its mission is to provide “timely and reliable
empirical research to all parties involved in public policy issues affecting risk and insur-
ance.” (http://www.ircweb.org/).
10 THE JOURNAL OF RISK AND INSURANCE
Although UM claims are first-party claims in that they are filed by a policyholder
with his or her own insurer, because UM insurance compensates the policyholder for
losses that should have been compensated through the liability insurance of the at-
fault driver, UM insurance payments may include general damages. General damages
are intended to compensate injured persons for noneconomic losses such as “pain and
suffering.” The possibility of general damages means that policyholders often receive
insurance settlements that exceed their monetary losses,16 and this greatly increases
the scope for negotiations over settlement amounts. UM data are thus well suited to
hypothesis tests regarding the impact bad faith liability on settlement amounts.
The claims data are collected through surveys of automobile insurers’ closed claim
files, from accidents occurring throughout the entire United States. The data provided
include summary information on the injured claimant, the accident, the injury and
its treatment, the claimed amount of loss, and the settlement amount. The location of
each accident by state and the date of the accident are also reported, making it possible
to relate settlements to the bad faith regime under which each accident occurred. In
determining the final set of insurance claims for analysis, we omit all claims for
which the accident state, the date of the accident, the amount of loss, or the settlement
amount is missing in the survey data; these variables are essential for our analysis and
thus claims with missing data cannot be analyzed. We also omit claims for which the
claimed amount of loss exceeds the insurance policy limit, claims for which another
insurer contributes part of the settlement, and claims subject to minimum claiming
thresholds under a no-fault insurance regime.17 The reasons for deleting these claims
from the analysis stem from the desire to reduce heterogeneity in features of the
claiming environment that will partially determine settlement amounts.
Because of time lags in the reporting and negotiation of claims, claims settled in
each survey year arise from accidents occurring in that year and in prior years. For
example, the claims closed in 1977 represent payments for automobile accidents that
occurred in the years 1972–1977 and similar patterns are observed in the 1987 and
1997 surveys. Figure 1 displays the number of claims in our sample that arise from
accidents in each calendar year. The figure shows that accidents in our data span the
years 1972–1997, and at least one accident is observed in each year except for years
1978 and 1979. However, the vast majority of accidents in the database occur in the
survey years or a few years prior.
The variation in accident dates provides us with greater time variation in our sample
than might be suggested by the dates of the IRC surveys. However, it also raises
questions about possible joint determination of the accident date and the settlement
amount, because the IRC data are collected based on the year of claim closure and
not the year of the accident. For example, if claims that are relatively more disputed
16 Crocker and Tennyson (2002) and Loughran (2005) find this result for bodily injury liability
claims; Browne, Pryor, and Puelz (2004) and our data summarized below show this same
result for UM claims.
17 No-fault insurance limits the ability to file liability claims for automobile accident injuries to
those injuries exceeding a threshold level of severity as specified in each state’s law. Although
UM claims are first-party claims, a no-fault threshold will usually limit compensation for
pain and suffering or other general damages to those UM claims that exceed the specified
threshold.
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 11
FIGURE 1
Number of Claims by Accident Year
exhibit a longer delay between the accident and the date of claim closure, there may
be a negative association between accident year and settlement amount in the IRC
data. Conversely, if more severe injury claims display a longer time from accident date
to claim closing date, there may be a positive association between accident year and
settlement amount. Because of this joint determination problem we are careful to not
include accident year fixed effects in our empirical models; instead, we include other
time-varying variables in Zt, which are exogenous to settlement amounts (discussed
below).
Several additional features of the insurance data inform the research design and em-
pirical specifications. Importantly, the data exclusively contain automobile insurance
claims that have been settled with some payment from the insurer. This means that
we cannot examine the impact of bad faith liability on the frequency of claim de-
nials and that our tests are with regard to settlement amounts conditional on a claim
being paid. Moreover, although the IRC data report the full value of the settlement
amounts including economic and noneconomic damages, no data are provided on
the amount of noneconomic damages that were sought by the policyholder. Only the
claimed economic losses of the policyholder are reported, and the survey contains
no information on the bargaining demands of the policyholder for pain and suffer-
ing, mental distress, or other noneconomic damages. Thus we cannot examine how
closely the settlement amount meets the full demands of the policyholder. Instead, as
discussed previously, we examine insurer settlement amounts relative to economic
losses reported by the policyholder.
12 THE JOURNAL OF RISK AND INSURANCE
FIGURE 2
Percent of States in Sample With Tort Bad Faith
Data on State Bad Faith Regimes
Data on state bad faith laws are compiled based on Stempel (2008), GenRe, Palmer,
and Dodge (2008), and Ostrager and Newman (2008). Our study distinguishes claims
settled under tort-based bad faith regimes from those settled under other bad faith
regimes. States for which our sources disagree about the bad faith regime in place, or
for which the bad faith regime is unclear to us for other reasons, are omitted from the
analysis. States in which statutes permit significant penalties for insurer bad faith are
also omitted, to provide a clear distinction between states that allow the broad-based
damages permitted under tort and those that do not. Thus, our control sample of
states are those that adjudicate charges of insurer bad faith under contract law or
under a statute that allows only limited damages, and states in which the courts have
specifically rejected the rights of policyholders to file suits alleging insurer bad faith.
The resulting data set contains claims settled in 42 states, from accidents occurring
during years 1972–1997. Because state bad faith regimes have evolved over time, the
number of states that permit tort actions for first-party-insurer bad faith increases
over the sample period. Figure 2 displays the trend over time in the recognition of
tort-based actions for first-party-insurer bad faith by states in our sample. None of
the 42 states in our sample permitted tort actions for first-party insurer bad faith in
1972. By 1997, 66.7 percent (28) of these states adjudicated first-party-insurance bad
faith actions under tort.
Control Variables
The vast majority of our control variables consist of characteristics of the insurance
claim reported in the IRC surveys. In keeping with previous studies of automobile
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 13
insurance settlements, our empirical models control for the demographic character-
istics of the claimant, the nature and severity of the injury and its treatment, and
city-size characteristics of the accident location, as well as the accident state (e.g.,
Ross, 1980; Crocker and Tennyson, 2002; Doerpinghaus, Schmit, and Yeh, 2003, 2008;
Browne, Pryor, and Puelz, 2004; Loughran, 2005). A large set of variables for these
claim features are included in the empirical models, subject to the constraint that the
variables must be available in all three surveys. As noted previously, one of the control
variables is an indicator of attorney representation, which is treated endogenously in
some model specifications. Table A2 in the Appendix reports summary statistics for
the claims variables in the empirical models, for the sample as a whole and for each
survey separately.
We account for relevant time-varying characteristics of the state insurance claiming
environment, other than the bad faith regime, through Wst. Specifically, we control for
state legislation that places limits on damages awards in tort cases because Browne,
Pryor, and Puelz (2004) find that state punitive damages reforms are negatively related
to UM insurance settlement amounts.18 State limits on punitive damages awards will
reduce the expected penalties faced by defendants in all cases—including suits over
UM settlement amounts and suits over bad faith in UM claims settlement. We obtain
data on state punitive damages reforms from compilations provided by the Insurance
Information Institute.
As noted previously, we control for state-invariant factors that may affect settlements
over time through the use of time-varying variables rather than with accident-year
fixed effects. The specific variables included in Ztare the 1-year U.S. Treasury bill rate
in the accident year of a claim, and an indicator for the IRC survey (1977, 1987, or
1997) in which a claim appears in our data.
Sample Characteristics
Figure 3 displays the percent of accidents in states recognizing tort-based bad faith
for each year of the sample.19 As expected given the evolution of the law over time,
the fraction of claims settled under tort-based bad faith laws increases in each year of
the study period. The large variations in the data are because of the small numbers of
accidents that occur during years far removed from the survey year. The figure also
displays the best-fit trend line through the data, which shows a clear upward trend
in the prevalence of tort-based bad faith.
Figure 4 displays the same data for the sample of claims that includes accidents
occurring only 5 years before the survey year and forward (1972–1977, 1982–1987,
1992–1997). These data show the same trends in the applicability of tort-based first-
party-insurer bad faith, but much less irregularity in the data. The figure also displays
the best-fit log-linear trend line, which shows a much better fit than the trend line in
Figure 3.
18 The empirical models in Browne, Pryor, and Puelz (2004) include a variety of tort-reform
measures. Punitive damages reforms are the only tort-reform measure to have a statistically
significant impact on settlement amounts.
19 The bad faith law that applies to a claim is that which is in effect in the accident state in the
year of the accident.
14 THE JOURNAL OF RISK AND INSURANCE
FIGURE 3
Percent of Claims Settled Under Tort Bad Faith Regime
FIGURE 4
Percent of Claims Settled Under Tort Bad Faith Regime, Early Years Omitted
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 15
TABLE 1
Comparisons of Settlements Across Bad Faith Regimes
Obs With Tort Obs Without Tort
Bad Faith Bad Faith Difference
Variable Mean S.D. Mean S.D. in Means
Full sample of claims N=4,293 N=2,259
Settlement amount 4,733.720 297.750 5,604.090 234.830 870.37∗∗
Economic loss claimed 3,073.030 284.550 4,779.380 778.500 1,706.35∗∗∗
Settlement/claim amount 2.733 0.114 2.579 0.058 0.15
Percent of claims underpaid 0.138 0.005 0.198 0.008 0.06∗∗∗
Claimant hired attorney 0.485 0.007 0.467 0.010 0.02
Claims in 1977 survey N=749 N=1,138
Settlement amount 5,434.625 307.612 5,994.030 348.376 559.41
Economic loss claimed 2,790.446 186.327 3,593.209 232.424 802.76∗∗
Settlement/claim amount 2.605 0.995 2.791 0.094 0.19
Percent of claims underpaid 0.167 0.014 0.252 0.013 0.09∗∗∗
Claimant hired attorney 0.405 0.018 0.437 0.015 0.03
Claims in 1987 survey N=1,347 N=563
Settlement amount 5,292.356 250.225 5,650.796 469.402 358.44
Economic loss claimed 4,690.176 873.421 9,628.212 3,050.071 4,938.04∗∗
Settlement/claim amount 2.650 0.127 2.317 0.076 0.33
Percent of claims underpaid 0.117 0.009 0.147 0.015 0.03
Claimant hired attorney 0.526 0.014 0.508 0.021 0.02
Claims in 1997 survey N=2,197 N=558
Settlement amount 4,152.265 513.862 4,761.721 416.646 609.46
Economic loss claimed 2,177.880 131.937 2,306.199 160.784 128.32
Settlement/claim amount 2.827 0.206 2.409 0.110 0.42
Percent of claims underpaid 0.141 0.007 0.140 0.015 0.00
Claimant hired attorney 0.487 0.021 0.487 0.021 0.00
∗∗∗Indicates the difference in means is statistically significant at the 1% confidence level; ∗∗5%
confidence level; 10% confidence level, respectively. All t-tests are two-sided tests.
Table 1 compares summary statistics for our dependent variables of interest—
settlement amount, settlement/claim, percent of claims underpaid—in states that
utilize tort-bad-faith versus other states. The table also compares claimed amounts
and the percent of claimants who hire an attorney. Dollar amounts in each year are
normalized to 1987 values using inflation rates from the medical services component
of the U.S. Consumer Price Index.20 Means and standard deviations are reported for
the sample as a whole and for each survey separately. T-statistics are reported to
determine whether means are significantly different across the two sets of states.
Contrary to our hypothesis that bad faith liability will increase insurance settlements,
mean settlement amounts are significantly lower in states that permit tort-based bad
faith actions, as are mean claimed amounts. There are no statistically significant dif-
ferences in the ratio of settlement to claim amounts or in the percentage of claimants
20 Settlement amounts are adjusted based on the survey year, and claimed amounts are adjusted
based on the accident year.
16 THE JOURNAL OF RISK AND INSURANCE
who hire an attorney in any of the survey years or for the sample overall. The per-
centage of claims that are underpaid is significantly lower in states with tort-based
bad faith, as our hypothesis predicts. Most of the patterns, if not significance levels,
are observed for the full sample as well as for each individual survey year. Mean
settlement amount, mean claimed amount, and underpayment percentage seem to
be lower in the 1997 survey than in the other survey years; the mean claimed amount
in 1987 is much higher than in the other survey years and displays a much larger
variance as well.21
ESTIMATION RESULTS
Effects of Tort Bad Faith on Settlement Amounts
Our first set of estimates examines the impact of tort-based bad faith liability on
insurance settlement amounts. OLS estimates that assume that attorney use is exoge-
nous are presented along with the MLE (treatment effects) estimates that allow this
variable to be jointly determined with the settlement amount. The estimated models
of attorney use (not displayed here) include all of the explanatory variables in the
settlement models, plus state (real) per capita income in the accident year to account
for the demand for and supply of attorneys in a state and year.22
Results of estimation are reported in Table 2. The left-hand columns report estimates
based on the full sample of claims. To determine the influence on our estimates of
claims for accidents which occurred many years before the survey year, the right-
hand columns report the estimates after eliminating claims for which the accident
did not occur within 5 years of a survey year (i.e., including accidents in 1972–1977,
1982–1987, and 1992–1997). To preserve space, the estimated coefficients are reported
for only the key independent variables.
The estimated impact of bad faith liability is significantly positive in all
estimates—irrespective of the sample or whether attorney use is treated as endoge-
nous. When the sample includes observations from all years, treatment effects esti-
mates of tort bad faith are lower than the corresponding OLS estimates. This indicates
that of settlements from states with bad faith liability, some of those that took several
years to settle (because of a lawsuit) can be better explained by the claimant hiring
an attorney than by the bad faith regime. When early-year claims are excluded, as in
the right-hand side of Table 2, the coefficient on tort bad faith is nearly identical to
the OLS estimate. Nonetheless, if attorney use is taken as endogenous and the full
sample is considered, the estimates suggest that tort liability for insurer bad faith
is associated with 10–11 percent higher UM settlements (exp{0.101}-1), an increase
that is both statistically and economically significant. Other important independent
21 Investigation of the data revealed that 1987 data are affected by one large outlier in each
set of states. Removing the outliers reduces statistical significance of the mean and standard
deviations of claimed amounts in this survey but does not affect other results in the paper.
Because we have no rationale for removing these claims we do not omit them from the sample
for analysis.
22 Per capita income is positive and significantly related to the probability of attorney represen-
tation in these estimates. The first-stage partial F-statistic is F(1, 40) =52.87, which satisfies
suggested standards for rejecting weak instruments (Staiger and Stock, 1997).
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 17
TABLE 2
Estimates of Ln(Settlement Amount)
All Years Early Years Omitted
Attorney Attorney
OLS Endogenous OLS Endogenous
Tort bad faith indicator 0.242∗∗ 0.101∗∗ 0.228∗∗ 0.229∗∗
2.335 2.430 2.197 2.260
Punitive damages limits 0.232∗∗ 0.207∗∗ 0.230∗∗ 0.205
2.294 1.980 2.199 1.790
Claimant hired attorney 1.345∗∗∗ 1.601∗∗∗ 1.343∗∗∗ 1.588∗∗∗
48.052 6.150 47.894 5.630
1977 claim survey indicator 0.168 0.115 0.177 0.123
1.504 1.090 1.498 1.080
1987 claim survey indicator 0.102 0.068 0.105 0.068
0.631 0.460 0.615 0.430
Attorney treated as endogenous No Yes No Yes
Full set of control variables Yes Yes Yes Yes
State fixed effects Yes Yes Yes Yes
Number of observations 6,495 6,491 6,421 6,417
Note: T-statistics, based on standard errors clustered at the state level, are reported below the
coefficient estimates. ∗∗∗ indicates statistical significance at the 1% confidence level; ∗∗ indicates
significance at the 5% confidence level; and indicates significance at the 10% confidence level.
All models include the full set of control variables, the 1-year Treasury bill rate of return, and
state fixed effects. Empirical model for (endogenous) attorney use includes the same variables
plus the natural log of state per capita income in the accident year.
variables have the expected signs; for example, limits on punitive damages are as-
sociated with significantly lower settlements and the claimant’s use of an attorney
is associated with significantly higher settlements. Settlement amounts appear to be
rising over time, on average, as the indicators for the 1977 survey and 1987 survey
are each negative, though they do not achieve statistical significance.
As argued previously, if the threat of tort liability for bad faith affects the amounts of
losses reported by claimants (for a given set of claimant and accident characteristics),
the estimated impact of tort-based bad faith regimes will incorporate both the effect
on settlements offered and the effect on losses claimed. The results reported in Table 3
investigate the importance of these effects, by estimating the impact of bad faith
liability on the settlement amount relative to the claimed amount of loss, and on the
claimed amount of loss itself.
The left-hand columns of Table 3 report estimation results for models of
Ln(Settlement/Claim) and those in the right-hand columns report results for models
of Ln(Claim). Each model is estimated first using OLS and again using MLE methods
that allow the attorney-use indicator to be endogenously determined. To provide
the most conservative estimates, these regressions—and those in the remainder of
the article—are derived using the full sample of claims. The results for the sample
omitting early-year claims are virtually identical and can be provided by the authors
upon request.
18 THE JOURNAL OF RISK AND INSURANCE
TABLE 3
Estimates of Ln(Settlement/Claim)andLn(Claim)
Ln(Settlement)–Ln(Claim)Ln(Claim)
Attorney Attorney
OLS Endogenous OLS Endogenous
Tort bad faith indicator 0.1120.1160.145 0.143
1.860 1.840 1.415 1.430
Punitive damages limits 0.030 0.084 0.215∗∗∗ 0.146∗∗
0.360 0.870 3.080 2.390
Claimant hired attorney 0.211∗∗∗ 0.2481.124∗∗∗ 1.732∗∗∗
2.984 1.650 15.353 10.110
1977 claim survey indicator 0.0710.019 0.239∗∗ 0.117
1.711 0.330 2.139 1.040
1987 claim survey indicator 0.023 0.038 0.159 0.077
0.405 0.550 1.193 0.650
Attorney treated as endogenous No Yes No Yes
Full set of control variables Yes Yes Yes Yes
State fixed effects Yes Yes Yes Yes
Number of observations 6,495 6,491 6,495 6,491
Note: T-statistics, based on standard errors clustered at the state level, are reported below the
coefficient estimates. ∗∗∗ indicates statistical significance at the 1% confidence level; ∗∗ indicates
significance at the 5% confidence level; and indicates significance at the 10% confidence level.
Settlement/Claim models include the full set of control variables, the 1-year Treasury bill rate
of return, and state fixed effects. Claim models include the same variables except for degree of
claimant fault. Empirical model for (endogenous) attorney use includes the full set of variables
plus the natural log of state per capita income.
After controlling for the amount of losses claimed, the estimated impact of bad faith
liability on settlements remains positive and marginally statistically significant (10
percent confidence level, two-sided test). The estimated magnitude of the impact,
11–12 percent, is nearly identical to the lower bound in the previous specification.
The estimates also show that claimed amounts are on average 15–16 percent higher
in states and years in which insurer bad faith is treated as a tort. These point es-
timates are not statistically significant at conventionally accepted confidence lev-
els, however. Even if loss claims do increase in states that accept the tort of bad
faith, our results indicate that settlements are disproportionately higher in those
states.
Tort Bad Faith and the Probability of Underpayment
The estimates thus far suggest that tort liability for insurer bad faith increases policy-
holders’ bargaining power, as evidenced by the receipt of higher settlement amounts.
To provide additional insights into the effect of bad faith liability on insurer settlement
decisions, we examine whether tort liability for insurer bad faith affects the likelihood
that a (paid) claim is settled for less than the amount of economic loss claimed by the
insured. The dependent variable in the empirical models is an indicator set equal to
1 for claims paid less than the economic loss amount (Settlement <Claim), and 0 for
all other claims.
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 19
TABLE 4
Estimates of Probability(Settlement <Claim)
Attorney Exogenous Attorney Endogenous
OLS Probit 2SLS MLE
Tort bad faith indicator 0.071∗∗ 0.072∗∗ 0.072∗∗ 0.072∗∗
2.242 2.210 2.409 2.170
Punitive damages limits 0.004 0.000 0.026 0.056
0.114 0.010 0.648 1.330
Claimant hired attorney 0.125∗∗∗ 0.127∗∗∗ 0.1480.425∗∗∗
7.560 8.400 1.818 16.290
1977 claim survey indicator 0.034 0.028 0.088∗∗ 0.144∗∗∗
0.988 0.880 2.516 4.440
1987 claim survey indicator 0.028 0.031 0.008 0.039
1.052 1.150 0.263 1.170
Attorney treated as endogenous No No Yes Yes
Full set of control variables Yes Yes Yes Yes
State fixed effects Yes Yes Yes Yes
Number of observations 6,495 6,495 6,491 6,491
Note: T-statistics, based on standard errors clustered at the state level, are reported below the
coefficient estimates. ∗∗∗ indicates statistical significance at the 1% confidence level; ∗∗ indicates
significance at the 5% confidence level; and indicates significance at the 10% confidence level.
All models include the full set of control variables, the 1-year Treasury bill rate of return, and
state fixed effects. Empirical model for (endogenous) attorney use includes the same variables
plus the natural log of state per capita income.
Table 4 reports the estimates. The estimated coefficients are reported as marginal ef-
fects and may therefore be interpreted as the change in the predicted probability of
underpayment that is associated with changes in the explanatory variables (Greene,
2003). The left-hand columns of the table present linear probability (OLS) and probit
models in which attorney use is treated as exogenous to the probability of underpay-
ment. The right-hand columns present two-stage least squares and treatment-effects
(MLE) estimates in which attorney use is treated as endogenous.
The estimates show that bad faith liability reduces the probability that a claim is paid
less than the economic losses claimed, and the estimated impact is statistically signif-
icant in all models. The magnitude of the estimated impact is also consistent across
all models, which suggest a 7.1–7.2 percent lower likelihood of claim underpayment
when insurers face tort liability for bad faith. These effects are consistent with the es-
timates for settlement amounts, but provide stronger evidence that bad faith liability
has meaningful consequences for settlement outcomes. Beyond affecting settlement
amounts at the margin for fully paid claims, bad faith liability also increases the likeli-
hood of full payment of claimed losses. This finding is similar in spirit to the Browne,
Pryor, and Puelz (2004) result that bad faith liability affects both economic damages
paid and general damages paid.
Effects of Tort Bad Faith over Time
We now turn to an examination of changes in the impact of bad faith liability over
time. We first explore whether the average effect of bad faith decreases during our
20 THE JOURNAL OF RISK AND INSURANCE
TABLE 5
Estimates of Trends in Ln(Settlement/Claim)
Linear Time Trend Differences by Survey
Attorney Attorney
OLS Endogenous OLS Endogenous
Tort ×time trend (year—1972) 0.006 0.007
1.354 1.598
Tort ×1977 survey indicator 0.102 0.095
1.602 1.462
Tort ×1987 survey indicator 0.155∗∗∗ 0.131∗∗
2.598 2.141
Tort x 1997 survey indicator 0.086 0.140
0.829 1.232
Punitive damages limits 0.012 0.075 0.031 0.084
0.164 0.858 0.364 0.869
Claimant hired attorney 0.212∗∗∗ 0.2750.210∗∗∗ 0.257
3.029 1.868 2.960 1.639
1977 claim survey indicator 0.131 0.060 0.053 0.007
1.365 0.560 0.523 0.065
1987 claim survey indicator 0.066 0.015 0.031 0.031
0.840 0.160 0.399 0.300
Attorney treated as endogenous Yes Yes Yes Yes
Full set of control variables Yes Yes Yes Yes
State fixed effects Yes Yes Yes Yes
Number of observations 6,495 6,491 6,495 6,491
Note: T-statistics, based on standard errors clustered at the state level, are reported below the
coefficient estimates. ∗∗∗ indicates statistical significance at the 1% confidence level; ∗∗ indicates
significance at the 5% confidence level; and indicates significance at the 10% confidence level.
All models include the full set of control variables, the 1-year Treasury bill rate of return, and
state fixed effects. Empirical model for (endogenous) attorney use includes the same variables
plus the natural log of state per capita income.
sample period. Table 5 reports results of estimating two modifications to our basic
specification to test for these effects. The first model interacts the bad faith indi-
cator with a variable that measures the difference between the accident year and
the first year of our sample (Yea r t1972). This specification tests the hypothesis
that the average impact of bad faith increases or decreases linearly over the sample
period. The second model allows the estimated coefficient on the bad faith indi-
cator to differ by survey year (D77,D87,D97). This specification allows a test of
the hypothesis that the average impact of bad faith differs across the survey pe-
riods. Estimates are provided for Ln(Settlement/Claim), using both OLS and MLE
methods.
The estimation results show no linear trend in settlements over time: the coefficient
estimates on tort interacted with (Yea r t1972) are virtually zero in magnitude and not
statistically significant. In the MLE estimates the impact of tort bad faith by survey
decade is estimated to be smaller in 1977 than in 1987 (10 percent) and 1997 (14–15
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 21
TABLE 6
Estimates of Trends in Ln(Settlement/Claim) Within a State
Single Time Trend
Trends by Date of
Adoption
Attorney Attorney
OLS Endogenous OLS Endogenous
Tort bad faith indicator 0.1120.1180.071 0.131
1.810 1.820 0.940 1.460
Tort ×time (year – first tort year) 0.000 0.003
0.040 0.410
Tort ×time for states adopting 1973–1977 0.001 0.003
0.940 0.580
Tort ×time for states adopting 1978–1987 0.003 0.001
0.330 0.060
Tort ×time for states adopting 1988–1997 0.034 0.020
0.800 0.490
Punitive damages limits 0.030 0.083 0.025 0.079
0.350 0.850 0.310 0.780
Claimant hired attorney 0.211∗∗∗ 0.2570.211∗∗∗ 0.265
3.000 1.700 2.990 1.690
1977 claim survey indicator 0.040 0.011 0.080 0.015
0.720 0.090 0.790 0.115
1987 claim survey indicator 0.024 0.021 0.037 0.019
0.772 0.210 0.470 0.200
Attorney treated as endogenous No Yes No Yes
Full set of control variables Yes Yes Yes Yes
State fixed effects Yes Yes Yes Yes
Number of observations 6,495 6,491 6,495 6,491
Note: T-statistics, based on standard errors clustered at the state level, are reported below the
coefficient estimates. ∗∗∗ indicates statistical significance at the 1% confidence level; ∗∗ indicates
significance at the 5% confidence level; and indicates significance at the 10% confidence level.
All models include the full set of control variables, the 1-year Treasury bill rate of return, and
state fixed effects. Empirical model for (endogenous) attorney use includes the same variables
plus the natural log of state per capita income.
percent), but differences in the estimated coefficients are not statistically significant.23
Thus, we cannot conclude that there are any substantive changes in the impact of tort
liability for bad faith during our sample period.
Table 6 reports estimates to test whether the impact of bad faith liability within
a state changes as the number of years that bad faith actions have been decided
under tort law increases. We again construct two slightly different measures to test
for a time effect. The first model includes the bad faith indicator and an interaction
term that measures how many years the state has recognized bad faith as a tort
23 F-statistics for tests of differences in coefficient estimates in the linear model, and chi-square
statistics for such tests in the MLE model, never approach conventional levels of significance.
Results are available from the authors.
22 THE JOURNAL OF RISK AND INSURANCE
(Yea r t– Firstyearst). This specification tests the hypothesis that the average impact of
bad faith increases or decreases linearly with the number of years it is in place in a state.
The second model uses the same specification but allows the estimated coefficient on
the interaction term to vary across groups of states. We use this specification to test for
differences in linear changes in the impact of bad faith across states that first adopted
a tort regime during years 1973–1977, 1978–1987, and 1988–1997, respectively.
The estimation results again show no meaningful time effects: coefficient estimates
are virtually zero in magnitude and not statistically significant. Thus, there are no sub-
stantive changes over time in the impact of bad faith liability on settlement amounts
within a state. Settlement amounts are higher in states that recognize first-party-
insurer bad faith as a tort, and appear to remain so over time.
CONCLUSION
This article has analyzed the effect of permitting tort claims against first-party insurers
for bad faith in settling claims. We provide new evidence on the long-run impact of this
form of liability by examining its effects over three decades—the 1970s–1990s. The
use of individual claims data from several different time periods permits stronger
conclusions about the impact of bad faith liability than is possible from existing
empirical studies. Our empirical tests support the conclusion that claim payments are
higher in states that permit tort actions for insurer bad faith. Estimates also show that
the probability of a claim being underpaid is lower in these states. These findings are
consistent with the hypothesis that expanded liability for insurer bad faith increases
policyholder bargaining power.
These findings notwithstanding, one must be cautious in making normative infer-
ences regarding the consequences of tort-based bad faith regimes. Higher insurance
settlements under tort bad faith may be evidence of beneficial effects of bad faith lia-
bility if in its absence insurers would underpay valid claims. Alternatively, if higher
settlements indicate that insurers are reluctant to challenge suspicious claims, tort
liability for insurer bad faith leads to efficiency losses. Additional research on the
relationship between bad faith regimes and other important aspects of the settlement
process, such as the likelihood of fraud and the determinants of insurers’ denials of
claims, would help to shed light on the welfare consequences of permitting tort suits
for insurer bad faith.
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 23
TABLE A1
State Bad Faith Regimes
State Bad Faith Tort First Year State Bad Faith Tort First Year
Alaska Yes 1974 New Hampshire No n.a.
Alabama Yes 1981 New Mexico Yes 1974
Arkansas Yes 1984 Nevada Yes 1975
Arizona Yes 1982 New York No n.a.
California Yes 1973 North Carolina Yes 1976
Colorado Yes 1983 North Dakota Yes 1979
Connecticut Yes 1973 Ohio Yes 1983
Delaware Yes 1982–1995 Oklahoma Yes 1977
Hawaii Yes 1996 Oregon No n.a.
Iowa Yes 1988 Rhode Island Yes 1980
Idaho Yes 1986 South Carolina Yes 1983
Illinois No n.a. South Dakota Yes 1986
Indiana Yes 1985 Tennessee No n.a.
Kansas No n.a. Texas Yes 1987
Kentucky Yes 1977 Utah No n.a.
Maine No n.a. Virginia No n.a.
Michigan No n.a. Vermont Yes 1979
Minnesota No n.a. Washington Yes 1992
Missouri No n.a. Wisconsin Yes 1978
Mississippi Yes 1984 West Virginia No n.a.
Montana Yes 1982 Wyoming Yes 1990
Source: Authors’ calculations from GenRe, Palmer, and Dodge (2008), Stempel (2008)
and, Ostrager and Newman (2008).
24 THE JOURNAL OF RISK AND INSURANCE
TABLE A2
Summary Statistics for Claim Characteristics
Full Sample 1977 Survey 1987 Survey 1997 Survey
Variable Mean S.D. Mean S.D. Mean S.D. Mean S.D.
Claimant age in years 34.667 17.502 34.452 19.615 33.799 16.493 35.415 16.597
Claimant male 0.438 0.496 0.448 0.497 0.445 0.497 0.425 0.494
Claimant married 0.420 0.494 0.453 0.498 0.436 0.496 0.387 0.487
Claimant single 0.380 0.486 0.361 0.480 0.399 0.490 0.381 0.486
Claimant divorced 0.033 0.178 0.036 0.185 0.037 0.189 0.028 0.164
Claimant employed 0.425 0.494 0.334 0.472 0.515 0.500 0.424 0.494
Claimant was pedestrian 0.024 0.154 0.035 0.184 0.025 0.157 0.017 0.128
Accident occurred in large city 0.439 0.496 0.472 0.499 0.460 0.499 0.403 0.491
Accident occurred in suburb 0.194 0.395 0.136 0.343 0.214 0.410 0.219 0.414
Accident occurred in medium city 0.234 0.423 0.206 0.404 0.219 0.414 0.264 0.441
Accident occurred in small town 0.077 0.266 0.098 0.297 0.057 0.232 0.075 0.264
Number of vehicles in accident 2.199 0.699 2.252 0.620 2.197 0.801 2.164 0.673
Claimant seen in emergency room 0.462 0.499 0.473 0.499 0.458 0.498 0.457 0.498
Claimant hospitalized overnight 0.014 0.120 0.007 0.083 0.018 0.132 0.017 0.131
Claimant hospitalized up to 1 week 0.036 0.187 0.034 0.182 0.057 0.231 0.023 0.151
Claimant hospitalized over 1 week 0.021 0.143 0.043 0.203 0.021 0.145 0.005 0.074
Claimant had no disability 0.562 0.496 0.368 0.482 0.529 0.499 0.719 0.450
Claimant permanently disabled 0.046 0.210 0.074 0.262 0.039 0.194 0.032 0.175
Claimant deceased 0.007 0.083 0.012 0.110 0.008 0.091 0.002 0.047
Claimant had a fracture injury 0.069 0.254 0.120 0.325 0.067 0.250 0.036 0.186
Claimant had a strain injury 0.747 0.435 0.679 0.467 0.810 0.392 0.749 0.434
Claimant had a laceration injury 0.125 0.330 0.098 0.297 0.248 0.432 0.057 0.232
Claimant had other injury 0.181 0.385 0.316 0.465 0.158 0.364 0.106 0.308
Payment reduced because of fault 0.022 0.145 0.030 0.171 0.020 0.141 0.016 0.127
Number of observations 6,552 1,887 1,910 2,755
Source: Authors’ calculations from IRC data.
THREAT OF INSURER LIABILITY FOR “BAD FAITH” 25
REFERENCES
Abraham, K. S., 1986, Distributing Risk: Insurance Legal Theory, and Public Policy
(New Haven, CT: Yale University Press).
Abraham, K. S., 1994, The Natural History of the Insurer’s Liability for Bad Faith,
Texas Law Review, 72: 1295.
Browne, M. J., E. S. Pryor, and B. Puelz, 2004, The Effect of Bad-Faith Laws
on First-Party Insurance Claims Decisions, Journal of Legal Studies, 33: 355-
390.
Browne, M. J., and B. Puelz, 1996, Statutory Rules, Attorney Involvement, and Auto-
mobile Liability Claims, Journal of Risk and Insurance, 63: 77-94.
Browne, M. J., and J. T. Schmit, 2008, Litigation Patterns in Automobile Bodily Injury
Claims 1977–1997: Effects of Time and Tort Reforms, Journal of Risk and Insurance,
75: 83-100.
Chang, H. F., and H. Sigman, 2000, The Incentives to Settle Under Joint and Several
Liability: An Empirical Analysis of Superfund Litigation, Journal of Legal Studies, 29:
205-236.
Cooter, R. D., and D. L. Rubinfeld, 1989, Economic Analysis of Legal Disputes and
Their Resolution, Journal of Economic Literature, 27: 1067-1097.
Crocker, K. J., and S. Tennyson, 2002, Insurance Fraud and Optimal Claims Settlement
Strategies: An Empirical Investigation of Liability Insurance Settlements, Journal of
Law and Economics, 45: 469-508
Doerpinghaus, H., J. Schmit, and J. J.-H. Yeh, 2003, Personal Bias in Automobile Claims
Settlement, Journal of Risk and Insurance, 70: 185-205.
Doerpinghaus, H., J. Schmit, and J. J.-H. Yeh, 2008, Age and Gender Effects on Auto
Liability Insurance Payouts, Journal of Risk and Insurance, 75: 527-551.
GenRe, and E. A. Palmer, and L. L. P. Dodge, 2008, Bad Faith Laws for Property/Casualty
Claims (Stamford, CT: General Reinsurance Corp.).
Greene, William H., 2003, Econometric Analysis, 5th edition (Upper Saddle River, NJ:
Prentice Hall).
Hawken, A., S. J. Carroll, and A. F. Abrahamse, 2001, The Effects of Third Party Bad
Faith Doctrine on Automobile Insurance Costs and Compensation (Santa Monica, CA:
Rand Institute for Civil Justice).
Heckman, J. J., 1978, Dummy Endogenous Variables in a Simultaneous Equation
System. Econometrica, 46: 931-959.
Hyman, D. A., B. Black, and C. Silver, 2011, Settlement at the Policy Limits and the
Duty to Settle: Evidence from Texas, Journal of Empirical Legal Studies, 8: 48-84.
Jerry, R. H., 1994, The Wrong Side of the Mountain: A Comment on Bad Faith’s
Unnatural History, Texas Law Review, 72: 1317-1344.
Kessler, D., 1995, Fault, Settlement and Negligence Law, Rand Journal of Economics,
26: 296-313.
Lee, H.-D., M. J. Browne, and J. T. Schmit, 1994, How Does Joint and Several Tort
Reform Affect the Rate of Tort Filings? Evidence From the State Courts, Journal of
Risk and Insurance, 61: 295-316.
26 THE JOURNAL OF RISK AND INSURANCE
Loughran, D. S., 2005, Deterring Fraud: The Role of General Damage Awards in Auto
Insurance Settlements, Journal of Risk and Insurance, 72: 551-575.
Maddala, G. S., 1983, Limited-Dependent and Qualitative Variables in Econometrics (Cam-
bridge: Cambridge University Press).
Malani, A., and J. Reif, 2010, Accounting for Anticipation Effects: An Application to
Medical Malpractice Tort Reform, NBER Working paper.
Ostrager, B. R., and T. R. Newman, 2008, Handbook on Insurance Coverage Disputes
(New York: Aspen Publishers).
Ross, H. L., 1980, Settled Out of Court: The Social Process of Insurance Claims Adjustment,
2nd edition (New York: Alpine Publishers).
Shavell, S., 1987, Economic Analysis of Accident Law (Cambridge, MA: Harvard Uni-
versity Press).
Spier, K., 2007, Litigation, in: A. M. Polinsky and S. Shavell, eds., The Handbook of Law
& Economics (Oxford: North-Holland).
Staiger, D., and J. H. Stock, 1997, Instrumental Variables Regression With Weak In-
struments, Econometrica, 65: 557-586.
Stempel, J. W., 2008, Stempel on Insurance Contracts (New York: Aspen Publishers).
Sykes, A. O., 1996, “Bad Faith” Breach of Contract by First-Party Insurers, Journal of
Legal Studies, 25: 405-444.
Tennyson, S., and W. J. Warfel, 2008, The Emergence and Potential Consequences of
First Party Insurance Bad Faith Liability, Journal of Insurance Regulation, 28(2): 3-20.
Tennyson, S., and W. J. Warfel, 2010, The Law and Economics of First-Party Insurance
Bad Faith Liability, Connecticut Insurance Law Journal, 15: 203-242.
White, M. J., 1989, An Empirical Test of the Comparative and Contributory Negligence
Rules in Accident Law, Rand Journal of Economics, 20: 308-330.
... Insurer liability for bad faith increases pressure on insurers to pay fraudulent or excess claims. Tort bad faith liability gives rise to higher settlement amounts, and the likelihood that a claim is underpaid is reduced (Asmat and Tennyson 2014;Tennyson and Warfel 2010). The New Insurance Law of the People's Republic of China in effect as of October 1, 2009, stresses protecting the interests of consumers, thus courts may award higher amount, possibly weakening the severity of excess claim. ...
... The New Insurance Law emphasizes protecting the interests of consumers so some claims that, before the New Insurance Law, were supported by the court less or not at all are now supported or more supported. Accordingly, the court awards a higher amount for the same or similar injury after the passage of the New Insurance Law, weakening the severity of soft fraud by being more accurate than higher settlement amounts, and reducing the likelihood that a claim is underpaid (Asmat and Tennyson 2014;Tennyson and Warfel 2010). But, the excessive protection from the New Insurance Law is likely to induce moral risk in inflating claims made by consumers, so we will continue to track the excess claim effects of the New Insurance Law in the future. ...
Article
Full-text available
This paper assesses the effects of claimant demographics and other claim characteristics on the measurement of the severity of opportunistic fraud using 96 excess claim lawsuits in personal injury insurance in China in 2000–2012. The empirical result indicates that severe opportunistic fraud that results in death is more numerous than it is for fraud that leads to disability and nondisability, which may be due to the fact that more severe injury may create greater openings for opportunistic fraud. Second, the severity of opportunistic fraud in provincial cities is lower than that it is in small or midsize cities because the former does not imply greater severity of opportunistic fraud. Third, the severity of opportunistic fraud in injuries from daily activity is greater than that for injuries from work and traffic accidents, implying that a higher excess claim probability and greater severity of opportunistic fraud in injuries from daily activity are consistent.
... Our paper falls in the intersection of three research areas. The first is the study of insurance contracts when fraud is a possibility (Crocker and Morgan 1998;Crocker and Tennyson 2002;Dionne, Giuliano, and Picard 2009;Asmat and Tennyson 2014;Burgeon and Picard 2014;Cosconati 2020). ...
Article
We assemble homeowner insurance claims from 28 independently operated country subsidiaries of a multinational insurance firm. We propose a new insurance model, in which consumers can make invalid claims and firms can deny valid claims, as is common in the data. In the model, trust and honesty shape equilibrium insurance contracts, disputes, and claim payments, especially when disputes are too small for courts. We test the model by investigating claim incidence, dispute, rejection, and payment, as well as insurance costs and pricing across countries. The evidence is consistent with the centrality of trust for insurance markets, as our model predicts.
... In their paper, Woods and Weinkle collect data from 56 cyber insurance policies to examine how deliberate cyberattacks on corporations and organisations are viewed with respect to so-called war exclusions in insurance policies (see also Romanosky et al. 2019). We also learn that ʽwarʼ is in the eye of the beholder as war exclusions differ by policy type and different market practices, giving rise to ambiguity in payment and the risk of contract non-performance (see Doherty and Schlesinger (1990) and Peter and Ying (forthcoming) for more on this topic) or subperformance (Asmat and Tennyson 2014). Woods and Weinkle conclude that cyber insurance coverage has been converging to a new equilibrium whereby losses due to war, in its traditional sense, are still excluded, but losses due to terrorism are not. ...
... However, I investigate an additional source of contract non-performance in which the insurer may dishonestly assert that valid claims are not valid. This type of contract non-performance can clearly be characterized as an act of "bad faith" by the insurer, which has been investigated in the literature on law and insurance economics (see Crocker and Tennyson 2002;Tennyson and Warfel 2009a, b;Asmat and Tennyson 2014). My analysis provides theoretical evidence that bad-faith penalties alone are not sufficient to discipline insurers to treat policyholders fair. ...
Article
Full-text available
This paper investigates the dynamics of an insurance market on which insurance companies may dishonestly deny eligible claims. Behaving dishonestly can increase the current profit but also entails the risk of losing profit in the future due to a worse reputation. Depending on the reputation cost imposed by policyholders, the analysis either predicts the emergence of reputation cycles or convergence to a stable equilibrium in which all eligible claims are accepted and the insurers’ reputations remain at a high level. I also show that policyholders may discipline insurers using a buying strategy based on an image-scoring rule. My results lead to important insights. For instance, reputation campaigns may have a pro-cyclic effect which leads to more severe reputation crises in the future.
... Crocker and Tennyson (2002) show that some types of claims are less certain to be paid in full because they are easy to falsify. Tennyson and Warfel (2009) andAsmat and Tennyson (2014) provide evidence of claims underpayment and discuss the effect of the legal framework on the insurers' settlement and verification practices.Bourgeon and Picard (2014) develop an economic model of insurer "nitpicking", which reduces the efficiency of insurance contracts and can lead to substantial uncertainty about the performance of the contract. A related issue is that of nonverifiable losses as studied byDoherty et al. (2013) .3 ...
Article
We study optimal insurance demand for a risk- and ambiguity-averse consumer under ambiguity about contract nonperformance. Ambiguity aversion lowers optimal insurance demand and the consumer's degree of ambiguity aversion is negatively associated with the optimal level of coverage. A more pessimistic belief and greater ambiguity may increase or decrease the optimal demand for insurance, and we determine sufficient conditions for a negative effect. We also discuss wealth effects and evaluate the robustness of our results by considering several alternative models of ambiguity aversion. Our findings show how ambiguity about contract nonperformance can undermine the functioning of insurance markets, making it a concern for regulators. Caution is required though because demand reactions are only imperfectly informative about the welfare effects of ambiguity about contract nonperformance.
... Contract nonperformance risk is likely to be perceived as ambiguous, and it is the implications of this ambiguity that we study in this paper. 2 1 Crocker and Tennyson (2002) show that certain claims that are viewed as being easy to falsify, are less certain to be paid in full; also Tennyson and Warfel (2009) and Asmat and Tennyson (2014) provide evidence of claims underpayment and discuss the effect of the legal framework on the insurers' settlement and verification practices. Bourgeon and Picard (2014) develop an economic model of insurer "nitpicking", which reduces the efficiency of insurance contracts and can lead to substantial uncertainty about the performance of the contract. ...
Preprint
Full-text available
We study the optimal insurance demand of a risk-and ambiguity-averse consumer if contract nonperformance risk is perceived as ambiguous. We find that the consumer's optimal insurance demand is lower compared to a situation without ambiguity and that his degree of ambiguity aversion is negatively associated with the optimal level of coverage. We also determine sufficient conditions for biased beliefs and greater ambiguity to reduce the optimal demand for insurance and discuss wealth effects. Finally, we scrutinize several alternative model specifications to demonstrate the robustness of our main result and discuss the implications of our findings.
Article
We analyze how insurance law can mitigate moral hazard by allowing insurers to reduce or cancel coverage in some circumstances. We consider an incomplete contract setting in which the insurer may obtain information related to the policyholder's behavior through a costly audit of the circumstances of the loss. Court decisions are based on a standard of proof such as the balance of probabilities. We show that an optimal insurance law brings efficiency gains compared to the no‐audit case. We also highlight the conditions under which the burden of proof should be on the insured, provided that insurers are threatened with sanctions for bad faith.
Article
We examine whether laws designed to reduce distracted driving have consequences for the automobile liability insurance market. Our research is motivated by prior studies that suggest distracted driving laws lead to improvements in traffic safety. Following these studies, we propose that distracted driving laws should also lead to reductions in the frequency and cost of injury liability insurance claims. Consistent with this expectation, we provide evidence that cellphone bans lead to approximately 3,400 fewer injury liability insurance claims, on average, in any given state enacting a ban. Our analysis also suggests that cellphone bans lead to reductions in injury liability loss costs, and we estimate the total statewide savings to the insurance industry, on average, is approximately $32 million per year in any given state enacting a ban. Additional analysis that further considers other types of distracted driving laws confirms that laws designed to limit distracted driving lead to substantial reductions in the frequency and cost of injury liability insurance claims. Finally, we also present analysis that suggests the reduction in claims frequency and injury liability loss costs attributable to distracted driving laws leads to statewide automobile liability insurance premium savings of approximately 4.7%.
Article
In this article, we examine the effect of laws prohibiting the hand-held use of a cellphone while driving on the automobile insurance market. Our research is motivated by prior studies that present evidence that the enactment of such laws alters drivers’ behaviors in ways that reduce the risk of automobile accidents. We posit that that, by extension, these laws should also lead to reductions in the amount of losses paid by private passenger automobile physical damage insurers. Our analysis indicates that the enactment of a ban on the hand-held use of a cellphone while driving reduces the incurred losses and incurred loss ratios of automobile insurers by approximately 3 percent, suggesting that these bans have important economic consequences not previously documented in the literature. Additional analysis suggests that hand-held cellphone bans eventually lead to incremental reductions in premiums, but we do not observe these reductions in premiums until a couple of years following the enactment of a ban. Our analysis of automobile insurance losses also represents a departure from most prior studies of cellphone bans and therefore contributes to the ongoing debate in the public health literature regarding the extent to which hand-held cellphone bans have implications for traffic safety.
Article
The cost associated with an automobile liability incident in the United States has been hypothesized to be related to different tort reform statutes, the presence of no-fault rules, and the impact of a plaintiff's attorney. This article tests these relationships and reports the marginal impact of variables related to liability claims with individual loss data from a representative insurer. Among major tort reforms, our analysis reveals that punitive damage limits, caps on noneconomic damages, and minor reforms (sanctions on frivolous suits or defenses, prejudgment interest, and provisions for structured settlements) are associated with a reduced individual claim severity. Reform of the joint and several liability rule is associated with an increased individual claim severity in this insurance market. Low dollar thresholds and add-on no-fault rules increase liability claim severity, while no statistically significant difference in claim severity is found when the claim is subject to verbal threshold rules relative to tort law. Finally, attorney involvement is associated with a 64 percent increase in the average claim size.