The role of life experience in long-term care insurance decisions

Journal of Economic Psychology xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

Journal of Economic Psychology

journal homepage: www.elsevier .com/ locate/ joep

The role of life experience in long-term care insurance decisions

http://dx.doi.org/10.1016/j.joep.2014.04.0020167-4870/� 2014 Elsevier B.V. All rights reserved.

⇑ Corresponding author. Tel.: +82 2 2049 6275; fax: +82 2 450 4084.E-mail addresses: [email protected] (S. Tennyson), [email protected] (H.K. Yang).

1 Tel.: +1 607 255 2619; fax: +1 607 255 4071.

Please cite this article in press as: Tennyson, S., & Yang, H. K. The role of life experience in long-term care insurance decisions. JoEconomic Psychology (2014), http://dx.doi.org/10.1016/j.joep.2014.04.002

Sharon Tennyson a,1, Hae Kyung Yang b,⇑a Department of Policy Analysis and Management, Cornell University, 252 MVR Hall, Ithaca, NY 14853, United Statesb College of Commerce and Economics, Konkuk University, 1 Hwayang-dong, Gwangjin-gu, Seoul 143-701, Republic of Korea

a r t i c l e i n f o

Article history:Received 22 February 2013Received in revised form 21 March 2014Accepted 5 April 2014Available online xxxx

JEL classification:I13I11D03D12

PsycINFO Classification:295033773920

Keywords:Long-term carePrivate long-term care insuranceInformal care-givingInsurance purchase intentionEmotions and insurance purchase

a b s t r a c t

This study uses data from a unique survey of the retirement planning behaviors of latemiddle-aged individuals living in New York State, to test hypotheses regarding the roleof earlier life experiences on the demand for long-term care insurance. Our primary focusis on previous provision of informal long-term care, which some studies have found to becorrelated with demand for long-term care insurance. We add to the literature by provid-ing a test for causal relationships between previous care-giving and insurance demand, andby exploring the more generally the mechanisms through which previous life experiencesare linked to insurance demand. Results are robust to a variety of empirical specificationsand estimation methods, including consideration of current care-giving roles and endoge-nous selection into previous care-giving, and strongly support a causal relationshipbetween previous long-term care-giving and demand for insurance. Our estimates alsoprovide evidence that lifetime health trajectories and family relationships are associatedwith long-term care insurance demand, and suggest that both emotional and informationalforces influence demand.

� 2014 Elsevier B.V. All rights reserved.

1. Introduction

Long-term care is one of the major future liabilities of the aging population in many countries. In the U.S., annual expen-ditures on long-term care are $203.1 and are projected to rise to $399.7 billion per year by 2019 (Centers for Medicare andMedicaid Services (CMS), 2010). Expenditures on long-term care averaged 1.5% of GDP among the OECD countries in 2008,and are expected to double by 2030 (Colombo, Llena-Nozal, Mercier, & Tjadens, 2011, de la Maisonneve & Martins, 2013).Public expenditures on long-term care greatly outweigh private expenditures in all countries, and sustainability in the faceof projected cost growth is an important public policy concern. Some countries have social insurance programs for long-termcare but these often have strict eligibility thresholds and require user cost-sharing; moreover, most do not cover the costs ofboard and lodging associated with institutionally-provided care (Colombo et al., 2011). Continued cost-shifting to privateindividuals seems likely as public funding pressures increase (e.g. Mayhew, Karlsson, & Rickayzen, 2010).

urnal of

http://dx.doi.org/10.1016/j.joep.2014.04.002

mailto:[email protected]

mailto:[email protected]


http://www.sciencedirect.com/science/journal/01674870

http://www.elsevier.com/locate/joep


2 S. Tennyson, H.K. Yang / Journal of Economic Psychology xxx (2014) xxx–xxx

Private long-term care insurance is a potential funding mechanism for the costs of long-term care that must be borne byindividuals. However, despite the insurable features of the risk and the availability of private long-term care insurance inmany countries, relatively few of the elderly and near-elderly have purchased insurance.2 Private insurance penetration ishighest in the U.S., but even there only 7.6% of current long-term care expenditures are funded by private insurance (CMS,2010) and only 10% of the population at risk for future long-term care own private insurance (Kim, 2010). Low rates of privateinsurance purchase have also been noted as a public policy concern in other countries (Comas-Herrera, Butterfield, Fernandez,Wittenberg, & Wiener, 2012, chap. 4; Courbage & Roudaut, 2008; Zhou-Richter, Browne, & Grundl, 2010).

This study adds to the literature aimed at understanding the determinants of private insurance demand. Previous studiesof the low ownership rate in long-term care insurance suggest that the most important issues may be those related to con-sumer demand (Brown and Finkelstein, 2007). They find that while long-term care insurance policies are characterized byhigh prices and limited benefits, these do not appear to be the main drivers of the low rate of insurance purchase. Indeed,Cramer and Jensen (2006) find that the demand for long-term care insurance is relatively price-inelastic. Other authors havefound that individual differences in risk preference, knowledge, or attitudes influence long-term care insurance demand.Finkelstein and McGarry (2006) conclude that ‘cautious’ individuals are more likely to purchase long-term care insurance,but are no more likely than others to utilize long-term care services. Zhou-Richter et al. (2010) conclude that down-wardly-biased risk perceptions account for the low rate of long-term care insurance purchase in Germany. McCall,Mangle, Bauer, and Knickman (1998) and Coronel (2000) find notable differences between purchasers and non-purchasersin knowledge of and attitudes toward long-term care use and financing.

What is lacking in the literature is understanding of the factors that contribute to these differences across consumers. Toshed light on this issue, this study examines the role that earlier life experience with providing informal long-term care toothers may have in determining demand for long-term care insurance. The near-elderly are often care-providers to familymembers, or may have previously provided care to their own parents. Spillman and Black (2005) estimate that 93% ofU.S. elderly with disabilities who live in the community receive some informal care; and nearly two-thirds of them relysolely on informal care. Kaye, Harrington, and Laplante (2010) report that family members make up the vast majority ofinformal care providers. OECD reports document this same pattern in other countries as well (Fujisawa & Colombo, 2009).

Previous research studies have found an association between prior experiences with long-term care and decisions to pur-chase long-term care insurance. McCall et al. (1998) find a direct relationship between knowing someone (a close friend or arelative) who needed long-term care and purchase of long-term care insurance. Analyzing demand for long-term care insur-ance in France, Courbage and Roudaut (2008) find that having provided informal care to family positively affects the prob-ability of purchasing long-term care insurance. Our study adds to this literature by providing a stronger test for causalrelationships between previous long-term care-giving and the demand for long-term care insurance, and by exploring themechanisms through which previous care-giving is linked to insurance demand.

Psychological literature in insurance suggests that an individual’s previous experience may affect insurance decisions bychanging knowledge, attitudes and risk perceptions. For example, informal care-giving may increase awareness of long-termcare risk, affecting insurance demand by raising the perceived probability of needing care (Courbage and Roudaut, 2008).Alternatively, care-giving may affect insurance demand through emotional responses which increase the salience of long-term care risk. Ranyard and McHugh (2012) argue that emotional responses create a link between past risk and future insur-ance purchases, and provide experimental evidence confirming this link. Our study looks for evidence of these linkages ininsurance demand using household survey data.

The study uses data from a unique survey of the retirement planning behaviors of late middle-aged individuals living in thestate of New York. The survey includes information on respondents’ life histories of work, family, health and leisure, in addi-tion to measures of demand for long-term care insurance. These richly detailed data provide a window into individual’s expe-riences prior to (being at risk for) long-term care insurance purchase. This permits examination of links between previouslong-term care-giving and demand for long-term care insurance. Availability of other variables – including current care-givingroles, engagement in service activities, lifetime health trajectories and family relationships – provides an extensive set of con-trols that help to identify causal relationships and facilitates greater understanding of the related to insurance demand.

The remainder of the paper is organized as follows. Section 2 discusses theoretical perspectives on the determinants oflong-term care insurance demand with a focus on the institutional context in the U.S., and presents the main hypothesesto be tested. Section 3 describes the data, empirical models, and estimation strategies and Section 4 previews the respondentsample. Section 5 describes the results of estimation. The final section of the paper presents interpretations and conclusions.

2. Demand for long-term care insurance in the U.S

2.1. Traditional models of insurance demand

Long-term care risk resembles other types of risks that are made privately insurable through long-term contracts: it has alow probability of realization early in life, the lifetime risk varies greatly across individuals, and its realization involves high

2 Analysts note that an important reason for the lack of private long-term care insurance in some countries is crowd-out by public insurance schemes(Kessler, 2008, Mayhew et al., 2010).

Please cite this article in press as: Tennyson, S., & Yang, H. K. The role of life experience in long-term care insurance decisions. Journal ofEconomic Psychology (2014), http://dx.doi.org/10.1016/j.joep.2014.04.002


S. Tennyson, H.K. Yang / Journal of Economic Psychology xxx (2014) xxx–xxx 3

costs. Recent actuarial estimates for the U.S. suggest that while 20–30% of remaining life expectancy for 65-year-olds will bespent in a state of chronic disability, only 12.7% will experience severe disability in their remaining lifetime (Stallard, 2011).The Congressional Budget Office (CBO) forecasts that only 9% of future elderly will need long-term care for five or more years(Congressional Budget Office, 2004). However, the cost of such care is extremely high, with private rooms in a care facilitycosting an average of $219 per day ($79,935 per year) according to one recent survey (New York Life Insurance Company,2009). Home care services are a more affordable part-time care alternative and are typically covered by long-term care insur-ance policies, but nonetheless average $21 per hour (New York Life, 2009).

In the absence of private insurance, long-term care services are most likely to be received informally from family mem-bers or paid for out-of-pocket. In contrast to acute care services, health insurance and Medicare provide little insurance cov-erage for long-term care expenditures. Medicaid provides financing of long-term care for the elderly poor, but Medicaideligibility requires very limited assets and income. In many cases the elderly in long-term care facilities must spend downtheir own assets until becoming eligible for Medicaid. Thus, without private long-term care insurance the risk of needinglong-term care creates significant financial uncertainty in later life.

Because long-term care insurance must be purchased in advance of the need for care, the purchase decision is a multi-period planning problem under risk and uncertainty. Neoclassical models of rational, utility-maximizing decisions assumethat consumers deciding whether to purchase long-term care insurance face a known probability distribution of care needsin the future, at known expected costs of care (Pauly, 1990). Such models predict that the demand for long-term care insur-ance will be affected by the individual’s financial resources, health expectations, and risk aversion, in addition to the price ofinsurance. Desire for insurance should be higher among those with a higher probability of needing care, but if these expec-tations are correlated with variables observed by insurers they will lead to insurance denial or higher insurance premiums.Public funding through Medicaid (Sloan & Norton, 1997; Brown and Finkelstein, 2008) and the availability of informal carefrom family (Mellor, 2001; Pauly, 1990) will reduce demand for insurance.

Empirical demand estimates are consistent with many predictions of rational economic theory. Existing studies find theexpected relationships between long-term care insurance ownership and age, gender, income, wealth, health status and fam-ily structure.3 Studies of high-risk groups find that individuals who discover they are at high risk of needing long-term care aremore likely to purchase long-term care insurance (Oster, Shoulson, Quaid, & Dorsy, 2010; Zick et al., 2005); but no significantadverse selection effects are found in the elderly population as a whole (Finkelstein & McGarry, 2006). Medicaid does appear tocrowd-out some demand for long-term care insurance, but these effects are generally small (Brown, Coe, & Finkelstein, 2007).4

Mellor (2001) finds that the availability of adult children to provide care does not reduce long-term care insurance purchase.5

2.2. Accounting for non-rational behavior in insurance demand

These results notwithstanding, the low overall rate of purchase of long-term care insurance products has prompted con-sideration of non-rational determinants of (the lack of) demand. Theories that may explain consumers’ failure to purchaseinsurance include misperceptions of probabilities, myopia and information processing limitations.6 Each of these has somerelevance in the long-term care context.

For example, a wealth of literature has documented that individuals tend to be overly optimistic regarding future lifeevents or to assume that their personal risk is better (or no worse) than average.7 Studies have specifically documented opti-mism bias regarding mortality risk among relatively healthy populations (Bhattacharya, Goldman, & Sood, 2009; Hurd &McGarry, 2002); and over-optimism regarding future health (Weinstein, 1982). Researchers have also noted that individualsmay fail to purchase insurance due to high costs of obtaining information regarding insurance costs and benefits(Kunreuther & Pauly, 2005). This may be especially relevant in the long-term care context due to the complexity of the planningproblem. Qualitative research studies have documented that non-purchasers feel overwhelmed by information (Curry, Robison,Shugrue, Keenan, & Kapp, 2009), and that one response of families to long-term care risk is to ‘‘decide not to decide’’ (Stum,2001). These reactions are consistent with difficulties in weighing costs and benefits.

Experience with long-term care-giving could have a de-biasing effect on each of these demand inhibitors. Providing infor-mal care to others may increase understanding of long-term care service benefits, raise awareness of the need for planning,and provide more convenient access to information sources about long term care financing. Another possibility is that pro-viding long-term care creates emotional reactions to the prospect of receiving informal long-term care from others, and it isthese emotions (rather than more accurate information) that increase demand for insurance. Either or both of these effectsmay lead long-term care-giving experience to increase demand for insurance. Nonetheless, determining whether this is thecase requires more than simple observation of a positive relationship between previous care-giving and insurance demand,since an alternative is that both are correlated with another omitted or unobservable factor.

3 See for example Sloan & Norton, 1997; Mellor, 2000; Courbage and Roudat, 2008.4 Mellor (2000) finds evidence of an asset protection motive in long-term care insurance ownership, as non-housing wealth is more strongly correlated to

insurance ownership than housing wealth (which is protected from Medicaid spend-down requirements in most states).5 Nor does the impact of wealth on purchase decisions differ for those with direct heirs than for those without (Mellor, 2000; Sloan & Norton, 1997).6 The ideas stem from general theories of consumers’ bounded rationality (Simon, 1957) and empirical observation of decision biases (Tversky & Kahneman,

1974). See Cutler and Zeckhauser (2004) and Schwarcz (2010) for in-depth discussions of these ideas in the context of insurance decisions.7 A large research literature follows on from the findings of Weinstein (1980) and Svenson (1981). See also the literature reviewed in Harris and Hahn (2011).




3. Data and research methods

3.1. The data source

The data used in the study are taken from the Cornell Retirement and Well-Being Study (CRWB). The CRWB is a survey ofworkers and retirees aged 50–72, randomly selected from six major employers in upstate New York. The survey investigatesthe transition to and life in retirement, focusing on pathways in and out of paid work and unpaid community service, as wellas their implications for well-being (Moen, Erickson, Agarwal, Fields, & Todd, 2000). The survey was conducted via face-to-face, in-depth interviews. Our analysis uses the first wave of the CRWB study, which sampled 762 individuals.8 The sampleused in this paper includes the 693 respondents who gave valid answers to the questions regarding long-term care insuranceownership.

The CRWB survey has several features that make it particularly useful for studying consumers’ long-term care insuranceplanning and purchase decisions. First, the study respondents are sampled from a population of late-middle-aged individualsin middle income households. This is the population most ‘‘at risk’’ for purchase of long-term care insurance since they areboth unlikely to be eligible for Medicaid and unlikely to have sufficient resources to self-finance long term care. In addition,respondents were asked not only their current long-term care insurance status but also their estimated likelihood of futurepurchase if they do not currently own insurance.9 Data on intentions to purchase not only provide important insights into con-sumers’ planning behavior, but help to overcome a limitation seen in most previous studies – the small fraction of consumersobserved to purchase the insurance.

That the CRWB sample is drawn from a single state may raise concerns about generalizing the findings to other locations.However, while long-term care insurance ownership rates vary by state, research has found no significant impact of differ-ences in state policies, quality of long-term care facilities, or insurance marketing strategies on the differences in insuranceownership rates (Doerpinghaus & Gustavson, 2002). Moreover, in the LTCI context the single-state sample also presents anadvantage by ensuring that respondents face a similar market environment for long-term care services and long-term careinsurance. The single-state focus also helps to limit another problem common in studies of LTCI markets: the lack of priceinformation on long-term care insurance and long-term care services. We use the approach commonly used in priorresearch, which is to include proxies for prices in our empirical models (respondent age and health status for insurance price,and residential location for price of long-term care services). Due to the limited geographic variation of our sample, we canbe more confident that unobserved price variations are not due to differences in the characteristics of the long-term care orlong-term care insurance markets.

3.2. Empirical models

We develop and estimate empirical models of insurance demand to test for the impact of previous long-term care-givingexperiences after controlling for other characteristics that should influence the demand for insurance protection. The basicempirical specification is shown in Eq. (1). LTCIic indicates a respondent’s demand for long-term care insurance, where i indi-cates respondent and c indicates the sampled company. The term uc is an employer-specific fixed-effect to account for unob-served differences across companies (equivalently, locations in the state) that affect LTCI decisions; eic is a random error termassumed to be normally distributed.

8 Theabout lobeen sofor in-halone c

9 Toinsuranfrom 0%

PleaseEcono

LTCIic ¼ aþ bPastCareic þ hXic þ uc þ eic ð1Þ

The variable PastCareic is an indicator variable set equal to one for respondents who have previously provided long-termcare, and equal to zero for all others. CRWB respondents reported details of any past care-giving experiences, including thenature of care-giving. Reasons for care are categorized as short term or long-term illness or disability, child care, respite care,aging, help with repairs, driving, or shopping. In our analysis long-term care-giving includes care for a long-term illness ordisability, respite care and aging. Care recipients are categorized as parent, parent-in-law, sibling, other relative, friend,spouse, or other. We estimate two alternative specifications in which the measure of previous long-term care giving is var-ied. The first specification includes previous long-term care-giving provided for any person and the second includes only pre-vious long-term care-giving provided for an immediate family member.

The vector Xic includes demographic and personal characteristics that rational decision models predict will impact long-term care insurance demand, plus preference characteristics and family relationships. Basic controls include demographicvariables of age, gender, race, education, income, marital status, and number of children. Wealth is proxied by includingan indicator variable for home ownership. Life stage is measured by an indicator for retirement status (retired versus not).

first wave of the survey was undertaken in 1994–1995. Although these data are more than 15 years old, this is the only wave that included questionsng-term care insurance, and reports respondents’ detailed life histories which include episodes of care-giving and serious illness. Although there have

me changes in the U.S. long-term care insurance market since that time, including an increase in marketing through employers and expanded coverageome care, the structure of insurance coverage remains the same; America’s Health Insurance Plans (2012) notes that over 97% of policies offer stand-overage for long-term care services, and that employer-sponsored plans require employees to pay the entire cost of coverage.determine insurance planning for long-term care, CRWB survey respondents were asked, ‘‘What is the likelihood that you will obtain additional healthce that would cover only long-term health care (such as nursing home stay or in-home nursing care) for your retirement years?’’ The responses ranged

(absolutely no chance) to 100% (certain).

cite this article in press as: Tennyson, S., & Yang, H. K. The role of life experience in long-term care insurance decisions. Journal ofmic Psychology (2014), http://dx.doi.org/10.1016/j.joep.2014.04.002



Respondent’s current health status is measured using the Center for Epidemiologic Studies Depression Scale (CESD scale)score, the number of restrictions on activities of daily living (ADL score), and an indicator of whether the respondent eversmoked. The CESD score is based on weighted scores to 20 questions about mood or feelings in the past week, and rangesfrom 0 to 60 with higher scores indicating more depression symptoms. The ADL score is the sum of limitations in performingsix activities (walking six blocks, climbing stairs, performing household tasks, working for pay, moving around the house,and managing personal care).

We also include measures to control for the effect of lifetime health trajectories on long-term care insurance demand.Previous studies have been limited to studying the relationship between current health status and long term care insuranceownership. However, due to insurer underwriting current health status may affect insurance ownership through both sup-ply-side and demand-side effects. After controlling for current health status, previous negative health shocks or gradualdeclines in health during adult life may change an individual’s health expectations and subjective risk estimates of the needfor long-term care, thereby influencing long-term care insurance decisions.

Several variables are included in Xic to characterize the respondent’s lifetime health trajectory. The first is an indicatordefined to be equal to one for respondents who characterize their health trajectory as previous decline in health followedby recovery, and zero otherwise. The second is an indicator equal to one for respondents who characterize their health tra-jectory as constantly declining health, and zero otherwise. Because these self-reported health trajectories may be correlatedwith observable (to insurers) illnesses, the control variables also include the number of months of major illness the respon-dent experienced over the previous ten years.10 This measure is the sum of total months of reported illness beginning 10 yearsprior to the survey for each of six illnesses included in the survey questionnaire. If multiple illnesses are experienced concur-rently, each is recorded as a separate month of illness.

Finally, our data permit examination of the relationship between family ties and demand for long term care insurance. Inthe existing literature, family members are often viewed as preferred care-givers. Alternatively, if individuals are altruistictoward family the purchase of long-term care insurance may act as a commitment device to avoid burdening them with careresponsibilities. We use responses to questions about family relationships to examine which of these perspectives dominatesafter controlling for other characteristics. Measures of preferences and family ties included in Xic are an indicator variable setequal to one if the respondent sees relatives at least twice a month, an indicator set equal to one if the respondent reportsthat family gives them the most satisfaction out of life (compared with things such as work, friends, religion or hobbies), andan indicator equal to one if the respondent prefers to stay in his/her current home as (s)he ages. Inclusion of these variablesin the estimated models enriches the set of preference and family related variables beyond those used in previous research.

3.3. Measurement and identification

There are two major issues to be dealt with in identifying the effects of previous care-giving on long-term care insurancedemand. The first identification concern is measurement of demand for insurance. Reasons for lack of insurance ownershipinclude supply-side factors that are difficult to control for with available data, although observable health indicators andgeographic controls provide some proxies. A second important reason for lack of insurance ownership may simply be lifestage. If individuals are more likely to purchase long-term care insurance as they age, purchase data may understate (even-tual) demand for insurance in cross-sectional data.

To deal with the demand measurement issue, we estimate models in which the demand for long-term care insurance isthe respondent’s self-reported intention to purchase insurance. Purchase intentions may be a more accurate indicator thanownership of demand for LTCI among respondents who are still planning for future retirement. Estimating models of inten-tions may also better reflect the effects of life experiences if some individuals who own LTCI have not made an active deci-sion to purchase it, which could occur (for example) if a spouse or a child made the decision.

Using purchase intentions as the demand measure is especially relevant for the purposes of our study, because in our datawe are not provided the dates of purchase of long-term care insurance among current owners. This means that we cannotdetermine the health of insurance owners at the time of purchase, or whether they purchased insurance before or after pro-viding long-term care to others (if any). Indeed, intentions may be the most useful measure for the current study because anyeffects of previous long-term care-giving on demand for LTCI should come about by affecting purchase intentions.

In the CRWB survey, those who do not currently own long-term care insurance were asked to state the likelihood (from0% to 100%) that they would purchase it in the future.11 We use these responses as a measure of sentiment toward LTCI own-ership and estimate models of the form shown in (1) for this dependent variable. Respondents who already own LTCI are omit-ted from the sample for the reasons noted above.12

A second identification concern is that individuals who select into long-term care-giving may also be more predisposed topurchase long-term care insurance. This may occur, for example, if childless individuals are more likely to provide care toelders due to fewer competing time demands, and are also more likely to purchase long-term care insurance due to the lack

10 The CRWB survey collects detailed health histories by asking respondents to provide information on all previous spells of serious illness, including thestarting and ending dates (including ongoing), and the nature of the illness.

11 We are not concerned about response bias, because questions regarding long-term care insurance were asked as part of a lengthy series of questions aboutplanning for retirement. Thus, there is no suggestion in the survey question that purchasing long-term care insurance is desirable.

12 Results using the insurance ownership variable are available from the authors.



0

50

100

150

200

250

Num

ber

of r

espo

nden

ts

Likelihood of purchase (percent)

Fig. 1. Long term care insurance purchase intentions.


of potential care-givers in their old age. We address this potentially confounding effect in our basic model specifications byincluding a wide range of control variables for family and marital situation, family ties, and preferred living arrangements inold age.

It is also possible that unobservable personal characteristics create a correlation between long-term care-giving and long-term care insurance purchase. For example, care-givers may be more altruistic than others and may therefore also be morelikely to purchase insurance (due to concerns about burdening others). We address this issue in our basic models by adding acontrol variable into Xic indicating if the respondent is currently providing care to another. Including current care-giving isintended to control for time-invariant personality traits which portend greater care-giving by a respondent throughout theirlifetime, to separate this propensity from the specific experience of previous long-term care-giving. CRWB respondents wereasked detailed information regarding provision of care to anyone at the time of the interview. We exclude care for childrenand providing others help with repairs, driving or shopping, from the definition of current care-giving.

We also estimate models that formally control for endogenous selection into past care-giving. We model the probabilityof previously providing long-term care as shown in equation (2): a function of exogenous variables in Xic and other personalcharacteristics Zic not included in (1). Eqs. (1) and (2) are jointly estimated based on Heckman (1978) and using maximumlikelihood (ML) methods as described in Maddala (1983). This approach estimates a probit model for past long-term care-giving and a linear regression model for purchase intentions.13

13 Esti

PleaseEcono

PastCareic ¼ dþ cXic þ wZic þ mc þ eic ð2Þ

The instrumental variable Zic is an indicator of whether the respondent has ever previously provided care for another(ever a caregiver). The identifying assumption for the model is that having engaged in any previous care-giving activitiesincreases the likelihood of having provided informal long-term care, but is related to LTCI demand only through its effectof increasing that likelihood. We provide evidence in support of this assumption in the results section of the paper.

4. Sample characteristics

In the survey sample for our analysis, 6.8% of respondents report ownership of long-term care insurance. Among respon-dents who do not own LTCI, intentions of purchasing it ranged from 0% to 100%. The mean purchase intention was 30%, andthe median intention was 25%. Fig. 1 displays the distribution of purchase intentions among the 646 respondents who do notalready own insurance.

As is common with such questions, the figure shows evidence of focal points for the responses, especially at zero and at50%. About 32% of respondents had zero intention of purchasing long-term care insurance in the future, and 16.9% estimatedtheir probability exactly at 50%. Only 2.5% stated a purchase intention of 100%, and only 18% reported a greater than 50%likelihood of purchase. There is a strong tendency for reporting round numbers, with all respondents reporting their inten-tions at 5% intervals. The distribution of responses is nonetheless quite wide-ranging, with relatively equal proportions ofrespondents answering 0% purchase intention, 1–49% purchase intention, and 50–100% purchase intention, respectively.

Table 1 reports respondents’ experiences with current and past care-giving. Responses are reported for the sample as awhole, and by purchase intention, separating high-intention versus low-intention purchasers (among those who do notalready own insurance). In this table, respondents who report 0–40% intention of purchasing are categorized as low-inten-tion and those who report 60–100% intention of purchase are categorized as high-intention. To improve the clarity of cate-gorizations, those who report a 50% chance of purchasing insurance are omitted from this comparison (Bruine de Bruin &Carman, 2012).

mation using instrumental variables tobit methods yields similar results.

cite this article in press as: Tennyson, S., & Yang, H. K. The role of life experience in long-term care insurance decisions. Journal ofmic Psychology (2014), http://dx.doi.org/10.1016/j.joep.2014.04.002


Table 1Caregiving experiences by LTCI purchase intention.

Variables All non-owners High intention to purchase Low intention to purchase Test statistic v2

Mean (Percent) Mean (Percent) Mean (Percent)

Current caregiver 33.59 35.04 32.14 0.35Any past caregiving 70.90 71.06 72.07 0.04

Past long-term caregivingAny long-term caregiving 49.07 48.72 48.81 0.00Long-term care for family 45.82 45.30 45.24 0.00

N 646 117 420

⁄ p < 0.1.⁄⁄ p < 0.05.⁄⁄⁄ p < 0.01.


The table reveals that nearly 50% of respondents have provided long-term care at some point in their lives, and that long-term care recipients are usually family members. Nonetheless, respondents who express high intentions to purchase insur-ance in the future are about equally likely to have provided long-term care in the past than those who express low purchaseintentions (48.7 versus 48.8%).

Respondent characteristics more generally are summarized in Table 2. Statistics are reported for the sample as a whole,and for high-intention versus low-intention purchasers.14

Among the respondents who do not own long-term care insurance, those with high intentions to purchase are signifi-cantly more educated and have significantly higher incomes than those with low purchase intentions. They are also signif-icantly younger, less likely to be retired, and more likely to be female. While high-intention purchasers are significantly lesslikely to report ever having smoked, there are no statistically significant differences in current health or in self-reportedhealth trajectories between those with high purchase intentions and low purchase intentions. High-intention purchasersare less likely to prefer aging in their own homes than low-intention purchases, and this difference is statistically significant.The family relations of high-intention and low-intention purchasers are also statistically different, with high-intention pur-chasers more likely to report family as their main source of satisfaction but less likely to see their family often.15

5. Estimation results

5.1. Basic specifications

We estimate the models using two-limit tobit methods, because the outcome variable (purchase intention) is measuredas a proportion from 0 to 100 and we observe a substantial number of observations at either zero or one hundred(Loudermilk, 2007).16 Due to missing values for some demographic variables, the number of observations in the regressionmodels is reduced to 503. However, the results are not different if we impute the missing values, suggesting that the missingvalues are not creating any bias. The estimation results are reported in Table 3.

The table reports results of estimating four different model specifications. The two left hand columns report estimation ofthe basic specification (1), first measuring previous long-term care-giving experience as any long-term care-giving, and thenincluding only long-term care-giving to a family member. The two right hand columns report these same two models withan added control variable indicating any current care-giving by the respondent. Each model is estimated including all of thecontrols in Xic, as well as employer fixed-effects, but only the main variables of interest are reported in the table.

In each set of estimates, parameters for each of the measures of long-term care-giving experience are positive and sta-tistically significant. The coefficient estimates suggest that previous long-term care-giving increases stated LTCI purchaseintentions by 4.9–9.4% points. Current care-giving is also significantly associated with higher intentions to purchase LTCI.Inclusion of current care-giving in the models reduces the estimated impact of previous long-term care-giving, but coeffi-cient estimates on those variables remain positive and statistically significant. Nonetheless, these results suggest that someof the apparent effects of previous long-term care-giving may reflect unobservable individual characteristics that are relatedto the propensity to care for others.

Strength of family ties is significantly related to purchase intentions, but estimates suggest a nuanced interpretation.Respondents who report that family is their highest source of satisfaction have higher LTCI purchase intentions, and this rela-tionship is statistically significant. However, those who see relatives at least twice per month have significantly lower inten-tions to purchase LTCI. Thus, it seems that respondents in families that see each other frequently may prefer to rely on

14 Income categories are adjusted to current dollars. The three income categories reported in the raw data were characterized as less than $30,000, $30,000 to$60,000 and more than $60,000.

15 Insurance owners are demographically similar to non-owners, but insurance owners report more variation in their health over time than non-owners.16 OLS estimates are available from the authors upon request.



Table 2Characteristics of sample.

Variable All non-owners High intention to purchase Low intention to purchase Test statistic

Mean (Percent) S.D. Mean (Percent) S.D. Mean (Percent) S.D. t-stat (x2)

Age 60.68 5.92 58.83 5.57 61.46 5.80 4.37***

Female 50.46 58.12 49.05 3.01*

White 91.33 91.45 91.43 0.00

EducationHigh school or less 44.04 29.20 48.30 13.11***

Some college 21.14 20.35 21.12 0.03College degree or more 34.82 50.44 30.58 15.40***

Household incomeLess than 47.4 K 32.14 22.12 34.26 5.99**

47.4–94.8 K 38.80 35.40 41.31 1.28More than 94.8 K 29.06 42.48 24.43 14.08***

Retired 60.53 49.57 65.48 9.82***

Home owner 92.20 95.61 91.44 2.20Married 74.77 71.79 75.24 0.57Number of children 2.94 1.74 2.81 1.71 3.02 1.82 1.06Prefer to stay in current home 24.30 13.68 28.33 10.45***

Family relationshipsFamily highest source of satisfaction 89.53 95.69 87.02 6.92***

See family at least twice per month 72.24 61.74 74.88 7.69***

Current healthADL limitations 0.29 0.97 0.18 0.84 0.35 1.07 1.52CESD sum 3.89 5.11 4.14 4.98 3.75 5.10 �0.72Ever a smoker 57.12 48.72 58.33 3.44*

Health trajectoryMonths of serious illness 44.62 146.08 29.10 99.88 50.00 161.17 1.33Previous health decline then recovery 7.80 4.55 8.62 1.99Continuous decline in health 8.81 9.09 9.66 0.03

Number of observations 646 117 420

* p < 0.1.** p < 0.05.*** p < 0.01.


informal care from family, but that strong emotional ties with family may lead to desires not to burden them with care-giv-ing responsibilities.

Respondents who prefer to remain in their current home have lower LTCI purchase intentions and this effect is both largeand highly statistically significant. This may be due to concerns that owning LTCI may lead to an involuntary move out of thehome, an apparent belief which contrasts with the findings of Doty, Cohen, Miller, and Shi (2010) who conclude that LTCIfacilitates access to care services other than nursing homes. Among the LTCI claimants in that study, over 50% were receivingcare at home or in an assisted living facility, and insurance did not appear to restrict long-term care choices.

Finally, self-reported declining health trajectories are not associated with higher LTCI purchase intentions. In fact, declin-ing health status reduces intention to purchase LTCI and this relationship is marginally significant in most models. This couldindicate that some respondents have lower purchase intentions because they believe that they will be ineligible or becausethey have lower subjective estimates of life expectancy. Because this latter finding raises concerns that respondents whoreport zero purchase intentions may do so because they have been rejected for insurance, we report results of a robustnesstest in Table 4.

The table reports probit estimates of the likelihood of having zero intention to purchase LTCI; if zero intention to purchaseLTCI arises because insurers deem them ineligible, we would expect a positive and significant coefficient on the variablesindicating declining health trajectories. However, the health trajectory variables are not significant in these models. This pro-vides some additional confidence that insurer underwriting decisions are not the main driver of low intentions to purchaseinsurance. We note further that these estimates suggest previous long-term care-giving has a negative effect on zero inten-tion to purchase LTCI, which is consistent with our findings in Table 3.

5.2. Falsification test

We also provide a falsification test of the influence of previous long-term care-giving on LTCI demand, by estimating mod-els that include an indicator for previous spells of short-term care-giving (care provided for short-term illness or disability)rather than long-term care-giving. This provides a falsification test because a finding that short-term care-giving experiences



Table 3Tobit estimates of LTCI purchase intentions.

Variable All non-owners (N = 505)

Basic specification Control for current caregiving

1 2 1 2

CaregivingAny previous long-term caregiving 7.795*** 4.921*

(2.933) (2.756)Previous family long-term caregiving 9.384** 6.912**

(3.633) (3.327)Current (any) caregiving 8.526*** 7.912***

(2.713) (2.332)Health trajectoryMonths of serious illness*10�3 �13.624 �13.468 �12.219 �12.336

(13.075) (12.682) (13.120) (12.714)Decline in health, recovery �5.958* �6.250* �5.777* �5.969*

(3.190) (3.222) (3.462) (3.444)Decline in health, constant poor �9.782* �9.669* �9.222* �9.151

(5.614) (5.772) (5.574) (5.685)

Family and preferencesFamily highest source of satisfaction 15.114** 15.108** 14.838** 14.812**

(7.138) (7.177) (7.350) (7.367)See family at least twice per month �10.615*** �10.993*** �11.259*** �11.590***

(3.396) (3.649) (3.275) (3.522)Prefer to stay in current home �13.891*** �14.035*** �14.581*** �14.590***

(4.794) (4.824) (4.538) (4.594)

Other control variablesDemographics Yes Yes Yes YesCurrent health measures Yes Yes Yes YesFinancial measures Yes Yes Yes YesEmployer fixed effects Yes Yes Yes YesPseudo R2 0.195 0.199 0.201 0.204

Robust standard errors in parentheses; standard errors clustered by employer.Demographic controls include age, gender, race (white versus other), marital status, highest level of education (category), retirement status indicator;measures of current health include number of limitations on activities of daily living, score on CESD (depression) scale, smoker status (ever smoked);financial controls include household income (category), home ownership; employer controls consist of employer dummy variables.; McKelvey andZovoina’s Pseudo R2.* p < 0.1.** p < 0.05.*** p < 0.01.


are significantly related to LTCI demand would undermine our argument that previous long-term care-giving experienceaffects demand by increasing knowledge and salience of long-term care needs. If long-term care-giving experiences affectdecisions to purchase LTCI through means other than fixed personality traits or situational factors that lead to care-giving,then previous spells of other types of care-giving should have no significant relationship to LTCI demand.

Estimation results for the falsification test are reported in Table 5, using the same reporting format as the previous tablesof estimates. Estimates show a positive relationship between previous short-term care-giving and LTCI purchase intentions,but coefficients are small and are not statistically significant in any of the model specifications. These estimates thus showthat previous short-term care-giving is not significantly related to LTCI demand, providing additional confidence that ourmain estimates are measuring a meaningful relationship.

5.3. Selection into care-giving

A final robustness check is the joint estimation of (1) and (2) using methods that allow for endogenous selection into pastlong-term care-giving. Table 6 presents a comparison of summary statistics for previous care-givers versus others, to providesome insight into whether the two groups of respondents differ in important ways. The comparison reveals that previouslong-term care-givers are older, are more likely to be female, and are more likely to see family often. These differencesare strongly statistically significant. Other differences significant at the 10% confidence level are that previous care-giversare less likely to be highly educated, less likely to have high income, and are less likely to smoke. It is interesting to notethat several of the characteristics of previous care-givers are expected to be negatively correlated with demand for LTCI(i.e., see family often, lower education, lower income).



Table 4Probit estimates of zero intention to purchase LTCI (N = 505).

Variable Probit (marginal effects)

Basic specification Control for current caregiving

1 2 1 2

CaregivingAny previous long-term caregiving �0.105** �0.086*

(0.041) (0.048)Previous family long-term caregiving �0.129*** �0.114***

(0.037) (0.040)Current (any) caregiving �0.060 �0.053

(0.037) (0.033)

Health trajectoryMonths of serious illness 0.117 0.112 0.107 0.105

(0.154) (0.146) (0.155) (0.147)Decline in health, recovery 0.012 0.016 0.014 0.018

(0.022) (0.022) (0.021) (0.020)Decline in health, constant poor 0.089 0.089 0.086 0.085

(0.092) (0.097) (0.093) (0.097)Family and preferencesFamily highest source of satisfaction �0.124 �0.125 �0.122 �0.123

(0.111) (0.112) (0.112) (0.113)See family at least twice per month 0.065 0.069 0.069 0.072

(0.050) (0.052) (0.051) (0.053)Prefer to stay in current home 0.186*** 0.189*** 0.191*** 0.193***

(0.045) (0.047) (0.040) (0.043)

Other control variablesDemographics Yes Yes Yes YesCurrent health measures Yes yes Yes YesFinancial measures Yes Yes Yes YesEmployer fixed effects Yes Yes Yes YesPseudo R2 0.139 0.144 0.141 0.146

Robust standard errors in parentheses; standard errors clustered by employer.Demographic controls include age, gender, race (white versus other), marital status, highest level of education (category), retirement status indicator;measures of current health include number of limitations on activities of daily living, score on CESD (depression) scale, smoker status (ever smoked);financial controls include household income (category), home ownership; employer controls consist of employer dummy variables.* p < 0.1.** p < 0.05.*** p < 0.01.


Table 7 reports the estimation results after allowing for selection into previous care-giving.17 The first-stage F-statistics(based on linear IV specifications) are reported at the bottom of the table; these are all well above the critical values suggestedto assure that the instrumental variable is relevant. Direct tests of instrument exogeneity are not available, but the summarystatistics in Table 1 (showing that any previous care-giving and previous long-term care-giving are highly correlated) andthe estimation results in Table 5 (showing that non-long-term care-giving is not associated with demand for LTCI) provide evi-dence on this point. Additionally, supplementary estimates show that the instrument is not significant when added directly intoestimates of (2), and that the correlation of the residuals from (2) and the residuals from (1) is zero.18 Thus, we feel confidentthat our instrument is valid.

The estimates show that the selection effect is positive – respondents who are likely to express high LTCI purchase inten-tions are more likely to have selected into previous long-term care-giving – but is not statistically significant. Thus, theredoes not appear to be an important selection bias in our single equation estimates. Nonetheless, the estimated coefficientvalues for previous care-giving are somewhat larger than those obtained from the OLS estimates that do not account forselection effects.19 The selection-corrected estimates indicate that intentions to purchase LTCI are 6.7–6.9 points greater amongrespondents who have previously provided long-term care to others, and these are significantly different from zero at the 5%confidence level.20

17 Because current care-giving may also be endogenous, the variable is omitted from these models.18 The correlation coefficient for the instrument and residuals from the model using any previous long-term care giving is 0.0028; that for the instrument and

residual from the model using previous long-term care giving to family is �0.0073.19 Similarly, estimates using instrumental variables tobit methods yield coefficients on the previous care variables that are somewhat larger than the tobit

estimates reported in Table 4: 9.60 for any previous long-term care-giving and 10.48 for previous long-term care provision to family members. Both aresignificantly different from zero at the 5% confidence level.

20 The corresponding first-stage estimates of previous care-giving are available from the authors.



Table 5Falsification tests of LTCI purchase intentions (N = 505).

Variable Basic specification Control for current caregiving

1 2 1 2

CaregivingAny previous non-long-term caregiving 3.277 1.865

(4.230) (4.331)Previous family non-long-term caregiving 2.951 1.730

(4.195) (4.191)Current (any) caregiving 10.435*** 10.480***

(3.064) (2.951)Health trajectoryMonths of serious illness*10�3 �11.657 �11.776 �10.809 �10.872

(12.815) (12.974) (12.842) (12.982)Decline in health, recovery �6.620** �6.527* �6.128* �6.071

(3.365) (3.388) (3.648) (3.715)Decline in health, constant poor �10.140* �10.194* �9.292 �9.320*

(5.854) (5.819) (5.665) (5.653)

Family and preferencesFamily highest source of satisfaction 15.520** 15.481** 14.985** 14.959**

(7.178) (7.117) (7.403) (7.339)See family at least twice per month �9.257 �9.303*** �10.654*** �10.686***

(2.892) (2.849) (3.018) (2.956)Prefer to stay in current home �14.684 �14.566*** �15.178*** �15.115***

(5.173) (5.110) (4.882) (4.819)

Other control variablesDemographics Yes Yes Yes YesCurrent health measures Yes Yes Yes YesFinancial measures Yes Yes Yes YesEmployer fixed effects Yes Yes Yes Yes

Pseudo R2 0.188 0.188 0.199 0.199

Robust standard errors in parentheses; standard errors clustered by employer.Demographic controls include age, gender, race (white versus other), marital status, highest level of education (category), retirement status indicator;measures of current health include number of limitations on activities of daily living, score on CESD (depression) scale, smoker status (ever smoked);financial controls include household income (category), home ownership; employer controls consist of employer dummy variables.* p < 0.1.** p < 0.05.*** p < 0.01.


6. Discussion

The long-term care insurance purchase decision exhibits many of the hallmarks of decision problems for which observedbehavior is difficult to square with traditional models of rational utility-maximizing consumers. This may explain whyresearch has struggled to explain the low rate of ownership using rational decision models, and recent attention has turnedto behavioral and psycho-social determinants of long-term care insurance demand. The current study contributes to this lit-erature by examining the impact of previous life experiences on LTCI purchase intentions. We focus our analysis on previousexperiences with informal long-term care-giving, and further examine the role of lifetime health trajectories and familyrelationships.

We find that previous long-term care-giving leads to stronger insurance purchase intentions among late middle-agedindividuals. The results are statistically significant, are present even after controlling for current care-giving behaviors,and are robust to falsification tests and to estimates that account for selection bias. These findings add substantial weightto existing evidence by establishing that previous experience with providing informal long-term care for others is causallyrelated to an increased likelihood of purchasing long-term care insurance for oneself.

An important open question is the mechanism that underlies this relationship: does previous long-term care-giving raiseinsurance demand by increasing risk information and awareness, or by creating emotional responses to the risk of needingcare? Although our estimates cannot provide definitive evidence, some patterns in our findings are suggestive. For example,we find that previous provision of long-term care for family members has a larger impact on demand for LTCI than care-giv-ing for any person.21 If long-term care-giving experiences influence LTCI purchase intentions via information channels, we positthat providing care to any person should provide that information just as effectively as providing care to family members. In

21 The t statistics for the difference between the coefficients estimates of care-giving to any person and care-giving to family members are 7.648 and 10.356for the estimates in Table 3, and 8.056 in Table 7 which are all significant at less than 0.001 level.



Table 6Characteristics of previous long term care-givers and non-care-givers (N = 646).

Variable Care-givers Non-care-givers Test statistics

Mean (Percent) S.D. Mean (Percent) S.D.

Age 61.25 5.86 60.13 5.93 �2.39**

Female 55.84 45.29 7.18***

White 91.80 90.88 0.17

EducationHigh school or less 46.69 41.35 1.82Some college 22.40 19.87 0.60College degree or more 30.91 38.78 4.29**

Household incomeLess than 47.4 K 33.77 30.57 0.7247.4–94.8 K 40.40 37.26 0.64More than 94.8 K 25.83 32.17 3.00*

Retired 61.51 59.57 0.25Home owner 93.46 90.99 1.33Married 72.56 76.90 1.62Number of children 2.92 1.77 2.96 1.72 0.31Prefer to stay in current home 23.97 24.62 0.04

Family relationshipsFamily highest source of satisfaction 90.82 88.27 1.11See family at least twice per month 78.64 66.15 12.32***

Current healthADL limitations 0.34 1.00 0.25 0.95 �1.06CESD sum 4.12 5.16 3.68 5.06 �1.08Ever a smoker 53.63 60.49 3.10*

Health TrajectoryMonths of serious illness 53.22 176.82 36.33 108.19 �1.47Previous health decline then recovery 8.70 6.87 0.68Continuous decline in health 9.36 8.25 0.23Number of observations 317 329

* p < 0.1.** p < 0.05.*** p < 0.01.


contrast, emotional channels to LTCI purchase intentions should be more strongly triggered by care-giving to family members.Thus, the stronger relationship between family care-giving and LTCI demand suggests a potential role of increased emotionalsalience rather than (or in addition to) increased information.

Other aspects of the estimation results also suggest that emotions are at work in determining LTCI purchase intentions.Respondents for whom family is a greater source of life satisfaction express higher intentions to purchase LTCI, which is sug-gestive of a protective motive in purchasing insurance. Additionally, declining lifetime health trajectories and previoushealth shocks are found to be negatively (although generally not strongly significantly) related to LTCI purchase intentions.This is unexpected if adverse health experiences provide information regarding long term care risk, since experiences withdeclining health should raise the perceived probability of needing LTCI insurance.22

It would perhaps not be surprising to discover emotional influences on the demand for LTCI, since family relationships,declining health and vulnerability in old age are highly emotionally fraught. Others have begun to emphasize that the role ofemotions in insurance decisions is an important neglected area of study in general (e.g. Ranyard & McHugh, 2012), andunderstanding the role of emotions in LTCI may be particularly important. Financial inducements such as government taxcredits and state ‘‘partnership’’ programs which permit greater access to Medicaid benefits for those who purchase privateLTCI have not been effective in increasing insurance purchase rates (Wiener, Tilly, & Goldenson, 2000). To promote privateplanning, the federal government has launched a long-term care awareness program and state insurance regulators have

22 Although not reported in the paper, it is also interesting to note that estimates of LTCI ownership models (available from the authors) show that declininglifetime health trajectories are positively related to LTCI ownership, and statistically significant. Also in contrast to the results for purchase intentions, family asa major source of satisfaction is negatively related to insurance ownership; seeing family often, desire to age at home, and current care-giving are notsignificantly related to long-term care insurance ownership. Respondents who already own insurance are not asked about the likelihood that they will purchaseit in future, and therefore do not have to contemplate the implications of needing long-term care when answering the survey. Ownership of LTCI should be lesssensitive to current emotional triggers as a result, and this may explain the different effects of these covariates on ownership and purchase intentions.



Table 7Selection – corrected estimates of LTCI purchase intentions (N = 503).

Variable 1 2

CaregivingAny previous long-term caregiving 6.882**

(2.840)Previous family long-term caregiving 6.749**

(3.194)

Health trajectoryMonths of serious illness �0.010 �0.009

(0.008) (0.008)Decline in health, recovery �3.917 �4.300

(2.906) (3.004)Decline in health, constant poor �5.935** �5.989*

(2.917) (3.129)

Family and preferencesFamily highest source of satisfaction 10.723** 10.870**

(4.230) (4.221)See family at least twice per month �8.065*** �8.085***

(2.197) (2.236)Prefer to stay in current home �8.311* �8.575***

(3.179) (3.178)

Other control variablesDemographics Yes YesCurrent health Yes yesFinancial Yes YesEmployer Yes Yes

Robust standard errors in parentheses; standard errors clustered by employer.Demographic controls include age, gender, race (white versus other), marital status, highest level of education (category), retirement status indicator;measures of current health include number of limitations on activities of daily living, score on CESD (depression) scale, smoker status (ever smoked);financial controls include household income (category), home ownership; employer controls consist of employer dummy variables. From the linear IVmodels diagnostic tests indicates the instrument used is highly relevant. F statistics for the first-stage regressions are 446.25 and 322.83 (both p < 0.01). Therhos which show the level of correlation between the error terms of equations (1) and (2) are -0.059205 and 0.0015272 respectively. The likelihood ratiotest against H0: rho = 0 results are x2 = 0.30 (p = 0.5870) and x2 = 0.00 (p = 0.9893) respectively which imply the selection effects may not be important.* p < 0.1.** p < 0.05.*** p < 0.01.


begun to provide consumer information on LTCI (Rogal & Shiffrin, 2007). If demand for long-term care insurance has signif-icant emotional and experiential drivers, including a focus on family relationships and care-giving burdens could increasethe effectiveness of risk communication and education efforts. Future research should continue to address these questions.

References

America’s Health Insurance Plans (2012). Who buys long-term care insurance in 2010–2011? Washington, D.C.: America’s Health Insurance Plans.Bhattacharya, J., Goldman, D., & Sood, N. (2009). Market evidence of misperceived prices and mistaken mortality risks. Journal of Economic Behavior and

Organization, 72(1), 451–462.Brown, J. R., Coe, N., & Finkelstein, A. (2007). Medicaid crowd-out of private long-term care insurance demand: Evidence from the health and retirement

survey. In J. M. Poterba (Ed.). Tax policy and the economy (Vol. 21, pp. 1–34). Cambridge, MA: MIT Press.Brown, J. R., & Finkelstein, A. (2007). Why is the market for long-term care insurance so small? Journal of Public Economics, 91, 1967–1991.Brown, J. R., & Finkelstein, A. (2008). The Interaction of Public and Private Insurance: Medicaid and the Long-Term Care Insurance Market. American

Economic Review, 98(3), 1083–1102.Bruine de Bruin, W., & Carman, K. G. (2012). Measuring risk perceptions: What does the excessive use of 50% mean? Medical Decision Making, 32(2),

232–236.Centers for Medicare & Medicaid Services. (2010). National Health Expenditure Projections 2009–2019. <https://www.cms.gov/NationalHealthExpendData/

downloads/proj2009.pdf>Colombo, F., Llena-Nozal, A., Mercier, J., & Tjadens, F. (2011). Help wanted? Providing and paying for long-term care. Paris: OECD Publishing.Comas-Herrera, A., Butterfield, R., Fernandez, J.-L., Wittenberg, R., & Wiener, J. (2012). Barriers and opportunities for private long-term care insurance in

England: What can we learn from other countries? In A. McGuire & J. Costa-Font (Eds.), The LSE companion to health policy. Cheltenham: Edward ElgarPublishing.

Congressional Budget Office (2004). Financing Long-Term Care for the Elderly. <http://www.cbo.gov/ftpdocs/54xx/doc5400/04-26-LongTermCare.pdf>.Coronel, S. (2000). Long-term care insurance in 1997–1998. Washington, DC: Health Insurance Association of America.Courbage, C., & Roudaut, N. (2008). Empirical evidence on long-term care insurance purchase in France. The Geneva Papers on Risk and Insurance, 33,

645–658.Cramer, A. T., & Jensen, G. A. (2006). Why don’t people buy long term care insurance? Journal of Gerontology: Social Sciences, 61B(4), S185–S193.Curry, L. A., Robison, J., Shugrue, N., Keenan, P., & Kapp, M. B. (2009). Individual decision making in the non-purchase of long-term care insurance. The

Gerontologist, 49(4), 560–569.Cutler, D. M., & Zeckhauser, R. (2004). Extending the theory to meet the practice of insurance. In R. Litan & R. Herring (Eds.), Brookings-Wharton papers on

financial services (pp. 1–53). Washington DC: Brookings Institution.


http://refhub.elsevier.com/S0167-4870(14)00021-X/h0005










https://www.cms.gov/NationalHealthExpendData/downloads/proj2009.pdf

https://www.cms.gov/NationalHealthExpendData/downloads/proj2009.pdf





http://www.cbo.gov/ftpdocs/54xx/doc5400/04-26-LongTermCare.pdf











de la Maisonneve, C., Martins, J. O. (2013). Public spending on health and long-term care: A new set of projections. In OECD Economic Policy Papers N0. 06.Paris: Organisation for Economic Co-operation and Development.

Doerpinghaus, H. I., & Gustavson, S. G. (2002). Long-term care insurance purchase patterns. Risk Management and Insurance Review, 5(1), 31–43.Doty, P., Cohen, M. A., Miller, J., & Shi, X. (2010). Private long-term care insurance. Value to claimants and implications for long-term care financing. The

Gerontologist, 50(5), 613–622.Finkelstein, A., & McGarry, K. (2006). Multiple dimensions of private information: Evidence from the long-term care insurance market. American Economic

Review, 96(4), 938–958.Fujisawa, R., Colombo, F. (2009). The long-term care workforce: Overview and strategies to adapt supply to a growing demand. In OECD Health Working

Papers No. 44. Paris: Organisation for Economic Co-operation and Development.Harris, A. J. L., & Hahn, U. (2011). Unrealistic optimism about future life events: A cautionary note. Psychological Review, 118, 135–154.Heckman, J. J. (1978). Dummy endogenous variables in a simultaneous equation system. Econometrica, 46(4), 931–959.Hurd, M. D., & McGarry, K. (2002). The predictive validity of subjective probabilities of survival. The Economic Journal, 112, 966–985.Kaye, H. S., Harrington, C., & Laplante, M. P. (2010). Long-term care: Who gets it, who provides it, who pays, and how much? Health Affairs, 29(1), 11–21.Kessler, D. (2008). The long-term care insurance market. The Geneva Papers on Risk and Insurance - Issues and Practice, 33, 33–40.Kim, G. (2010). Medicaid crowd-out of long-term care insurance with endogenous medicaid enrollment. Ph.D. Dissertation, University of Pennsylvania.Kunreuther, H., & Pauly, M. (2005). Insurance decision-making and market behavior. Foundations and Trends in Microeconomics, 1(2), 63–127.Loudermilk, M. S. (2007). Estimation of fractional dependent variables in dynamic panel data models with an application to firm dividend policy. Journal of

Business and Economic Statistics, 25(4), 462–472.Maddala, G. S. (1983). Limited-dependent and qualitative variables in economics. New York: Cambridge University Press.Mayhew, L., Karlsson, M., & Rickayzen, B. (2010). The role of private finance in paying for long term care. The Economic Journal, 120, F478–F504.McCall, N., Mangle, S., Bauer, E., & Knickman, J. (1998). Factors important in the purchase of partnership long-term care insurance. Health Services Research,

33(2), 187–203. Part I.Mellor, J. M. (2000). Private long-term care insurance and the asset protection motive. The Gerontologist, 40(5), 596–604.Mellor, J. M. (2001). Long-term care and nursing home coverage: Are adult children substitutes for insurance policies? Journal of Health Economics, 20,

527–547.Moen, P., Erickson, W. A., Agarwal, M., Fields, V., & Todd, L. (2000). The cornell retirement and well-being study: Final report. Ithaca, New York: Bronfenbrenner

Life Course Center, Cornell University.New York Life Insurance Company (2009). Cost of Care Survey.Oster, E., Shoulson, I., Quaid, K., & Dorsy, E. R. (2010). Genetic adverse selection: Evidence from long-term care insurance and Huntington disease. Journal of

Public Economics, 94, 1041–1050.Pauly, M. V. (1990). The rational nonpurchase of long-term-care insurance. Journal of Political Economy, 98(1), 153–168.Ranyard, R., & McHugh, S. (2012). Defusing the risk of borrowing: The psychology of payment protection insurance decisions. Journal of Economic Psychology,

33, 738–748.Rogal, D. L., & Shiffrin, R. (2007). Consumer-preparedness for long-term care. Washington, D.C.: Academy Health.Schwarcz, D. (2010). Insurance demand anomalies and regulation. Journal of Consumer Affairs, 44, 557–577.Simon, H. A. (1957). Models of man: Social and rational. New York: John Wiley and Sons.Sloan, F. A., & Norton, E. C. (1997). Adverse selection, bequests, crowding out, and private demand for insurance. Evidence from the long-term care insurance

market. Journal of Risk and Uncertainty, 15(3), 201–219.Spillman, B. C., & Black, K. J. (2005). Staying the Course: Trends in Family Caregiving. Issue Brief #2005-17. Washington, D.C.: AARP Public Policy Institute.Stallard, E. (2011). Estimates of the incidence, prevalence, duration, intensity and cost of chronic disability among the U.S. elderly. North American Actuarial

Journal, 15(1), 32–58.Stum, M. (2001). Financing long-term care: Examining decision outcomes and systemic influences from the perspective of family members. Journal of Family

and Economic Issues, 22(1), 25–53.Svenson, O. (1981). Are we all less risky and more skillful than our fellow drivers? Acta Psychologica, 47, 143–148.Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131. New Series.Weinstein, N. D. (1980). Unrealistic optimism about future life events. Journal of Personality and Social Psychology, 39, 806–820.Weinstein, N. D. (1982). Unrealistic optimism about susceptibility to health problems. Journal of Behavioral Medicine, 5, 441–460.Wiener, J. M., Tilly, J., & Goldenson, S. M. (2000). Federal and state initiatives to jump start the market for private long-term care insurance. Elder Law Journal,

8, 57–101.Zhou-Richter, T., Browne, M. J., & Grundl, H. (2010). Don’t they care? Or, are they just unaware? Risk perception and the demand for long-term care

insurance. The Journal of Risk and Insurance, 77(4), 715–747.Zick, C. D., Mathews, C. J., Roberts, S., Cook-Deegan, R., Pokorski, R. J., & Green, R. C. (2005). Genetic testing for Alzheimer’s disease and its impact on

insurance purchasing behavior. Health Affairs, 24(2), 483–490.


















































Documents

The role of life experience in long-term care insurance decisions