lable at ScienceDirect

Social Science & Medicine 70 (2010) 1317–1325

Contents lists avai

Social Science & Medicine

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

Preferences for long-term care services: Willingness to pay estimates derivedfrom a discrete choice experiment

Anna P. Nieboer a,*, Xander Koolman b, Elly A. Stolk a

a Erasmus University Rotterdam, Institute of Health Policy and Management, The Netherlandsb Faculty Technology, Policy and Management, Delft University of Technology, The Netherlands

a r t i c l e i n f o

Article history:Available online 12 February 2010

Keywords:PreferencesLong-term careSocial production function theoryDiscrete choice experimentThe Netherlands

* Corresponding author. Erasmus University Rotterdand Management, Burgemeester Oudlaan 50, Locatie WPA, The Netherlands. Tel.: þ31 10 4082804.

E-mail address: nieboer@bmg.eur.nl (A.P. Nieboer)

0277-9536/$ – see front matter � 2010 Elsevier Ltd.doi:10.1016/j.socscimed.2009.12.027

a b s t r a c t

Ageing populations increase pressure on long-term care. Optimal resource allocation requires an optimalmix of care services based on costs and benefits. Contrary to costs, benefits remain largely unknown. Thisstudy elicits preferences in the general elderly population for long-term care services for varying types ofpatients.

A discrete choice experiment was conducted in a general population subsample aged 50–65 years(N¼ 1082) drawn from the Dutch Survey Sampling International panel. To ascertain relative preferences forlong-term care and willingness to pay for these, participants were asked to choose the best of two carescenarios for four groups of hypothetical patients: frail and demented elderly, with and without partner.The scenarios described long-term care using ten attributes based on Social Production Function theory:hours of care, organized social activities, transportation, living situation, same person delivering care, roomfor individual preferences, coordination of services, punctuality, time on waiting list, and co-payments.

We found the greatest value was attached to same person delivering care and transportation services.Low value was attached to punctuality and room for individual preferences. Nursing homes weregenerally considered to be detrimental for well-being except for dementia patients without a partner.Overall, long-term care services were thought to produce greatest well-being for the patients ‘withouta partner’ and those ‘with dementia’. Individuals combining these two risk factors would benefit themost from all services except transportation which was considered more important for the frail elderly.The results support the notion that long-term care services represent different value for different types ofpatients and that the value of a service depends upon the social context. Examination of patient profilesconfirmed the notion that physical, mental and social vulnerability affect valuation of the services.Policy-making would profit from allocation models in which budgetary requirements of differentservices can be balanced against the well-being they produce for individuals.

� 2010 Elsevier Ltd. All rights reserved.

Introduction

Ageing populations, a broadening social definition of health andthe increasing prevalence of chronic diseases intensify pressure onhealth care budgets (WHO, 2005). While these changes may affectall types of health care, long-term care may be most affectedbecause long-term care expenses increase markedly with old age(de Meijer, Koopmanschap, Koolman, & van Doorslaer, 2009).Keeping long-term care affordable while raising quality to meetexpectations will be a big challenge (Miller, Booth, & Mor, 2008;OECD, 2005).

am, Institute of Health Policyoudestein, 3062 Rotterdam,

.

All rights reserved.

Effective response seems to be thwarted because the relativevalues of available services and supply methods for differentpersons have not been documented. Although evaluations oneffectiveness of long-term care products are plentiful (e.g. Bouman,van Rossum, Nelemans, Kempen, & Knipschild, 2008; Thompson,Lang, & Annells, 2008), information about the yield of individualservices in terms of improving consumers’ well-being is lacking.Consequently, it is not easy to determine how resources can best beallocated over beneficiaries, services, and/or modes of servicedelivery. Therefore, achieving efficient resource allocation deci-sions requires us to learn how individuals place value on particularaspects of long-term care.

Resource allocation decisions in curative care are often guidedby the outcomes of economic evaluations (Ham & Robert, 2003;Rutten & Busschbach, 2001). This is not common practice in long-term care, probably because suitable outcome measures are

A.P. Nieboer et al. / Social Science & Medicine 70 (2010) 1317–13251318

lacking. Support and care for people with physical, mental orsensory handicaps and/or illnesses is often needed for years or eventhe rest of their lives. As much of long-tem care is not aimed toimprove health outcomes, measures used in curative care – such asQuality Adjusted Life Years – are less suitable for outcome valua-tion. Consequently, evaluations of long-term care typically focus onthe extent to which services produce the intended effects. Reha-bilitation programs, for example, are evaluated by their effects onphysical functioning, and feeding policies by improvement innutritional status (Arinzon, Peisakh, & Berner, 2008; Forster et al.,2009). Impacts of such improvements on well-being remainunknown. The few available studies on the value of long-term careservices regrettably do not allow for a cross-service comparison(Howell-White, Gaboda, Scotto Rosato, & Lucas, 2006; Markle-Reidet al., 2006; Mitchell, Salmon, Polivka, & Soberon-Ferrer, 2006).

Research efforts need to be directed at developing tools that aresuitable to compare long-term care needs across populations and tomeasure benefits of various services. In this regard, Ryan andcolleagues developed the older person’s utility scale (OPUS) toevaluate the needs for (or the effects of) social care in the elderly,regardless of their condition (Ryan, Netten, Skatun, & Smith, 2006).Another example is the ICECAP measure, which produces a qualityof life index for elderly in a similar manner as the OPUS (Coast et al.,2008). Both measures may serve to evaluate improvements onwell-being as related to provision of long-term care services.Furthermore, their generic nature allows for assessing the relativevalues of those services. But as of yet, evidence of validity of the twomeasures is still rather limited, and relative values of differentservices for different people have not been established. For that, itseems we must wait until the ICECAP and OPUS techniques havegained wider acceptance. In the short run, however, valuation oflong-term care services is still needed.

In this article we present the results of a discrete choiceexperiment (DCE) performed to elicit preferences for long-termcare. The DCE exercise required respondents to choose betweenlong-term care scenarios for several hypothetical patients. Thescenarios were characterized by absence or presence of particularservices and by specific modes of services delivery. A cost attributeaddressed willingness to pay (WTP) for the scenarios and marginalWTP for their attributes. Thus, welfare gains related to long-termcare could be indirectly assessed. The results of our study mayguide resource allocation decisions across alternatives and/oracross groups of beneficiaries, and improve understanding of howservices affect quality of life of different patient subgroups.

Theory underlying attribute selection

In long-term care settings, a wide range of services may beoffered in various ways. In the current study, it is impossible toevaluate added value of each one. We focus on services and modesof service delivery with the largest potential to affect well-being –identified using the Social Production Function (SPF) theory (Lin-denberg, 1996; Lindenberg, 2001; Nieboer, Lindenberg, Boomsma,& van Bruggen, 2005). Below we first present this theory and nextdescribe how we used it to identify services and modes of servicedelivery with a possibly large impact on well-being.

Social production function (SPF) theory

In order to define a ‘core set’ of services and modes of servicedelivery, we had to conceptualize how different care servicescontribute to well-being. Linking goals with behavior, SPF theoryoffers a framework to this aim (Lindenberg, 2001; Lindenberg,1996; Nieboer et al., 2005; Ormel, Lindenberg, Steverink, & Ver-brugge, 1999; Ormel, Lindenberg, Steverink, & VonKorff, 1997).

SPF theory identifies two goals that all humans seek to optimize:physical well-being and social well-being. SPF theory assumes thatindividuals take an active role in pursuing their goals intelligently;i.e., they consider scarcities and differences in efficiency, theysearch for opportunities, and they substitute. Social circumstancesmay restrain and influence their beliefs and expectations, definitionof the situation, and ways to come to a decision. To optimizephysical well-being, people seek comfort and stimulation. Socialwell-being may benefit from affection, behavioral confirmation,and status. These five instrumental goals are assumed to havedecreasing marginal value for the realization of overall well-being.Substitution takes place on the basis of relative costs of alternativegoals. For example, when opportunities to realize status dwindle,one’s focus shifts towards affection and behavioral confirmation(Nieboer & Lindenberg, 2002). The production of overall well-beingoffers only limited leeway, however, for substitution of physical andsocial well-being (Steverink, 2001; Van Bruggen, 2004).

SPF theory further assumes that people deploy activities,services, and resources in such a way that these contribute to so-called metagoals, i.e., (immediate) efficiency and development ofwell-being over time. In the short run, well-being may be improvedmost by giving people control over their own well-being andproduction thereof, and by delivering multifunctional care thatcontributes to different dimensions of well-being. Avoiding (further)loss of well-being is efficient both in the short-term and the long-term (Lindenberg, 1996; Van Bruggen, 2004). Loss avoidance is animportant motive steering people’s behavior (Kahneman & Tversky,1979). Being caught in a so-called loss frame is particularly damagingfor well-being after life events (Nieboer, 1997) and strengthens theorientation towards respite care (Steverink, 2001). This situationhinders productive behaviors rather than stimulating them.

Linking services and modes of service delivery to SPF-goals

Below we indicate what specific services and characteristics ofcare may matter most in relation to the goals identified in SPF theory.

Physical well-being requires comfort and stimulation. One’s levelof comfort is optimal in the absence of pain, hunger or thirst, andbeing able to live in a pleasant and safe environment, suggestingthat comfort may be enhanced by personal, domestic, and nursingcare, but also by ensuring that clients have a say in where they liveand what care they receive. With regard to stimulation, careservices may fulfill people’s need for mental as well as physicalactivation. In particular, (social) activities that require physicaleffort contribute to physical well-being (Ormel et al., 1997).Transportation services and participation in organized socialactivities are important to this aim.

In relation to social well-being, Steverink and Lindenberg (2006)showed that affection, behavioral confirmation and status remainessential social needs at advanced age. Transportation services anddaycare are vital for continuation of social activities and maintainingcontacts with loved ones. Receiving care may also directly contributeto social well-being when affectionate interaction between client andcare provider develops. Having the same care provider over time isstimulating in this respect. Although elderly people’s need for statushas received little attention, there are good examples of how tofacilitate nursing homes residents’ need for status (Gerritsen,Steverink, Ooms, & Ribbe, 2004). The coordinated delivery ofcare services makes clients feel they are taken seriously (status). Timeon a waiting list is considered important in relation to all specifiedneeds.

One of the metagoals in SPF theory is having a sense of controlover one’s well-being. Being able to choose between services andcare provided will contribute to this sense of control (Groenewoud,van Exel, Berg, & Huijsman, 2008). Also the availability of

A.P. Nieboer et al. / Social Science & Medicine 70 (2010) 1317–1325 1319

transportation services is relevant for clients’ autonomy. Waitingtimes (lack of punctuality) need not be problematic; if short andknown, one can still manage one’s own time. A second metagoal isthe multifunctional production of well-being. Long-term care isexpected to enable people to continue current activities and developnew activities, thus contributing to social and physical well-being.Having a single care provider is a prerequisite for multifunctionality,in terms of social benefits and physical needs because the caregivergets to know the client’s preferences. The third metagoal, lossavoidance, steers people towards respite care, which may notnecessarily help to achieve the highest possible level of well-being.Elderly in general are reluctant to give up independence and move toan elderly home. They will only do so if the benefits clearly outweighthe disadvantages; for example, when one’s health is expected tofurther deteriorate, and one is not certain that a minimum level ofsocial and physical well-being can be sustained in the current livingsituation (Steverink, 2001). Acceptable waiting times are then alsodetermined by the perceived minimum level of well-being andsustainability of a given situation. Immediate care is called for ifwaiting would result in a serious loss of comfort, stimulation,affection, behavioral confirmation and status.

In sum, the following attributes are assumed to contribute tophysical well-being, social well-being and/or metagoals: personalcare, domestic care, adequate living situation, room for individualpreferences, same person providing care, punctuality, coordination ofcare, social activities, transportation, and short time on a waiting list.

Methods

Study design

The DCE involved choices between two hypothetical but real-istic long-term care scenarios, described by attributes derived fromSPF theory. Experimental design techniques served to decide whichattribute level combinations would be presented, in order to ensurethat information about attribute importance could be retrievedfrom the collected data.

Attributes and levels

The SPF framework identified ten types of care services andmodes of delivery that probably contribute to well-being of thebeneficiaries of long-term care. These were all included as attri-butes of long-term care in this DCE. An eleventh attribute wasadded to allow for estimation of WTP: co-payment. In phrasingthese attributes and deciding on their levels, consistency with long-term care options currently available was our prime concern. A

Table 1Attributes and levels.

Levels

Attributes 0 1

Number of hours of care per week 4 h 8 hOrganized social activities Not available 1 Half dTransportation service Available Not avaiLiving situation Living independently at home Apartme

proximiWho provides care Regular care provider VaryingIndividual preferences Standardized care The con

determiCoordinated care services deliverya Have to arrange little Have toPunctuality Max. 15 min waiting time Max. 1 hWaiting list in months Directly available 4 MonthCo-payment per week No co-payment 50 Euro

a Definition: ‘As an extra service your caregiver can coordinate all the care you receive.and you need additional care. Moreover, you do not have to arrange your own care with

literature search confirmed that all mentioned services and modesof services delivery are consistent with currently available long-term care options, either because they belong to public entitle-ments or because they are on offer in the private sector. Under theExceptional Medical Expenses Act (AWBZ, 1967) and the SocialSupport Act (WMO, 2005), patients are entitled to assistance in sixfunctional areas: personal care, nursing, supportive guidance,activating guidance, treatment, and living services. An independentauthority (Center for Indications in Care, CIZ) establishes what careis required and to what extent. Person may then opt for care in kind,a personal care budget or a combination of the two. One is free touse the personal care budget in a way that best suits one’s needsand situation. For example, specific preferences can be accountedfor by trading off elements of indicated care, or by seeking assis-tance in the voluntary sector.

Contrary to AWBZ regulations we did not distinguish betweennursing care and personal care, seeing that people are generallywilling to trade these off: the total number of hours counts ratherthan allocation to specific tasks. Similarly, we did not distinguishbetween the AWBZ entitlements supportive guidance (managea day schedule and activate patients) and activating guidance(modify behavior to cope with existing problems), because manyactivities contribute in both domains. Having a single care provider,coordination of care services, punctuality, room for individualpreferences and co-payment for extra comfort are importantreasons for clients to opt for a personal budget and choosetheir own care providers. Waiting lists may apply to all services onoffer.

The attributes were designed at 2 or 4 levels each, capturinga realistic range within the Dutch healthcare context (Table 1).Attributes related to modes of services delivery were presented asquality aspects that were present or absent. Punctuality and wait-ing lists varied over a realistic range. Living situation and trans-portation options reflected those available in the public or privatesector. With regard to hours of care and possibility to participate insocial activities, the indication will specify precise entitlements,limiting our freedom to vary levels over a wide range. However,because clients can spend a personal budget as they see fit we didnot feel restricted by indication, as long as levels were imaginable.The levels for personal and home care ranged from 4 to 16 h perweek; zero hours was not included because it is unrealistic to tradeoff all hours of care. Similarly, the amount of social activities wasvaried over an imaginable range; a ‘not available’ option wasincluded, however, because clients not interested in this kind ofsupport may trade it all off.

In the Netherlands, co-payments for long-term care are basedon the amount of care, the recipient’s income, and living situation.

2 3

12 h 16 hay per week 2 Half days per week 3 Half days per weeklablent building in the

ty of careSheltered accommodation Elderly or nursing home

care providerstent of care isned individuallyarrange a lotwaiting time Max. 2 h waiting time Max. 3 h waiting time

s 8 Months 12 Months100 Euro 150 Euro

This also means that they can respond quickly when your health deteriorates furtherindividual caregivers.’

A.P. Nieboer et al. / Social Science & Medicine 70 (2010) 1317–13251320

In our study, however, the co-payment level range simply repre-sented population minimum and maximum, from 0 (a realisticminimum level of WTP) to 150 euros per week (the maximum levelof co-payment that would have been required). Income-related co-payment levels were disregarded for lack of information andstatistical power. Co-payment levels were not fixed for living situ-ations or amounts of care, in order to facilitate measurement ofMWTP. We did not specify what other effects the presented livingarrangements might have on household budget (e.g. house rent).

The DCE was piloted in a convenience sample of experts,colleagues, friends and family. The responses were reason tocollapse two potential attributes (hours of personal and domesticcare were combined because of a compensatory relation betweenthe two) and to reword how another attribute needed to beinterpreted (individual preferences – see footnote Table 1). So,eventually the DCE included six 4-level and four 2-level attributes(Table 1). Irrespective of the nature of attributes (qualitative orquantitative) they all entered the regression analysis with codesdefined in the 0–3 range. The codes reflect dummy codes for thequalitative variables, and values on a continuous scale for thequantitative variables.

Experimental design

The DCE design contained 256 choice pairs – Fig. 1 provides anexample – representing a small share of the over 65,000 possibleattribute level combinations.

Example of a cho

Question 1

Please indicate which care package you find most suit

information, you will find a description of the content

click on the aspect in question. If you click on the ‘pa

patient again.

AegakcaP

sruoh8keewreperacfosruoH

Participation in organized social

activities

Not available

Transportation services Available

Living situation Living independent

Person delivering care Varying care provid

Content of care Standardized care

Coordination of services Have to arrange litt

Punctuality Maximum wait 15

Time on waiting list 12 months

Co-payments per week N eno

The most suitable care package for the patient is:

Care package A

Care package B

Fig. 1. Example of

The choice pairs were constructed according to the principlespresented by Street, Burgess, and Louviere (2005). From NeilSloane’s website (http://www.research.att.com/wnjas/oadir/) weobtained a strength three, 256-run orthogonal array that allowedfor estimation of 17 four-level factors. Seven columns were omitted,and four columns were collapsed into two level factors to match thearray to our set of attributes and levels (Sloane, 2003). The resultingorthogonal array represented the profiles in the first option of thechoice set. We applied a systematic set of level changes to pair eachprofile to a second profile (0 � 1, 1 � 2, 2 � 3, 3 � 0). The second setof choice options is also orthogonal and never has the same attri-bute levels as the first, maximizing information available from eachchoice. The main effects were all uncorrelated. The design was95.0% efficient compared with the optimal design for a choice set oftwo, and 65.8% compared with the optimal design for a choice set offour allowing for simultaneous consideration of all possible attri-bute levels.

Study sample

In December 2005, a stratified random sample of members ofthe Dutch Survey Sampling International (SSI) Internet panelbetween 50 and 65 years of age was invited by e-mail to participate.Strata were defined by age (50–54, 54–59, 60–64 years), gender,and education (low, middle, high education). Recruitment was inthree rounds, so that rounds 2 and 3 could compensate for strata-specific non-response in rounds 1 and 2, maximizing the chance to

ice set

able for the described patient. For your

of care and modes of service delivery when you

tient’ button you can read the description of the

BegakcaP

sruoh21

1 half day per week

Not available

ly at home Sheltered accommodation

ers Regular care providers

The content of care is determined

individually

le Have to arrange a lot

minutes Maximum wait 1 hour

8 months

orue05

a choice set.

Box 1. Four patient profiles.

Try to picture that you are acquainted with an elderly personsuffering from various physical problems. Outside of thehouse he or she cannot walk independently (not even withappliances), and already has experienced several falls.Medications are taken for a number of health complaints.Getting dressed independently is still possible but takes a lotof time and energy. This person lives alone and has contactwith only a few people. Although undertaking things alone isout of the question, the elderly person likes to keep activeand mean something to others. To help him or her to ach-ieve this goal, professional help is being sought.

Try to picture that you are acquainted with an elderly personsuffering from various physical problems. Outside of thehouse he or she cannot walk independently (not even withappliances), and has already experienced several falls.Medications are taken for a number of health complaints.Getting dressed independently is still possible but it takesa lot of time and energy. This person is married, but otherthan that has contact with only a few people. Althoughundertaking things alone is out of the question, the elderlyperson likes to keep active and mean something to others.To help him or her to achieve this goal, professional help isbeing sought.

Try to picture that you are acquainted with someone whosuffers from dementia. This person lives alone and tends toforget things that just happened. For example, a phone callor someone dropped by in the morning. Daily activities suchas making coffee and doing the dishes are still manageable.But gradually he or she becomes more dependent on othersfor the daily activities. As far as possible neighbors andfriends take care, but this is a heavy task also because itrequires constant supervision. The dementia patient isdepressed and often reacts snappy and angry. To help thisperson professional help is being sought.

Try to picture that you are acquainted with someone whosuffers from dementia. This person is married and tends toforget things that just happened. For example, a phone callor someone dropped by in the morning. Daily activities suchas making coffee and doing the dishes are still manageable.But gradually he or she becomes more dependent on othersfor the daily activities. As far as possible the partner takescare, but this is a heavy task also because it requiresconstant supervision. The dementia person is depressedand often reacts snappy and angry. To help the couple,professional help is being sought.

A.P. Nieboer et al. / Social Science & Medicine 70 (2010) 1317–1325 1321

obtain a sample that would be representative for the Dutch generalpopulation in these age groups.

The questionnaire

The questionnaire was made available online. Respondentswere asked to imagine that they have to decide on servicesoffered to a ‘relevant other’, like a spouse, parent or friend. Sincehealth state or social circumstances of the hypothetical benefi-ciary were expected to influence preferences, we distinguishedfour patient profiles: physically frail elderly and patients withdementia, both groups either living alone or with their partner.Descriptions are presented in Box 1. We expected the differentdimensions of care to be relevant for all hypothetical benefi-ciairies, but to a different extent. For example, patients with orwithout a partner may have different needs for social activation,and the urgency to meet long-term care needs may be strongestfor dementia patients living alone, for fear of their safety.

The 256 choices – arranged into 32 blocks of 8 – were admin-istered for each profile separately, so four times in total. All attri-bute levels appeared equally often in each block to preventrespondents from focusing too much on a particular outcome.Respondents were randomly assigned to one of the four profilesand one of the 32 blocks. We aimed to distribute each combinationof a profile and block to 8 respondents, so that we needed a sampleof 1024 (4 � 32 � 8) respondents. Bounds imposed to therandomization procedure guaranteed similar numbers of observa-tion across all blocks and patient profiles.

Respondents were first asked to read an introduction containinga motivation for the study, reasons why people may need long-termcare, a description of specific services that may be on offer (actuallya description of the attributes and levels), instructions onresponding to the questions, and a description of the hypotheticalpatient the respondent was to consider. Following the DCE ques-tions, we collected demographic information, information aboutdirect (as a user) or indirect (through family members) involve-ment with long-term care.

Statistical analysis

The DCE method is based on Lancaster’s characteristicstheory of value (Lancaster, 1966) and on random utility theory(Manski, 1977; McFadden, 1974). In the former, demand dependson the characteristics (attributes) of goods (alternatives), ratherthan on the goods themselves. Accordingly, choice data areanalysed on the basis of the idea that people will choose thealternative that provides most utility: that is, individual n willchoose alternative i over another alternative j if Uin > Ujn. Inrandom utility theory, the conventional utility function U (�) hastwo parts: one deterministic V (�) that contains observabledeterminants, and a random component 3 (�) that representsunobservable determinants. Therefore, the utility for individual iof alternative n is:

Uin ¼ Vin þ 3in (1)

The probability that individual n will choose alternative i overalternative j belonging to some choice set C is given by:

P ¼ Pr�Vin þ 3in > Vjn þ 3jn

�; cj˛C (2a)

P ¼ Pr�Vin � Vjn

���3jn � 3in

�; cj˛C (2b)

To estimate the choice probability of equation (2) we must makeassumptions about the nature of the error component. We assumed

that the error terms are independently and identically distributedwith a Type 1 extreme-value (Gumbel) distribution. Under thisassumption the probability of choice in equation (2) is satisfied bythe conditional logit model and is equal to:

P ¼ expðgVinÞPj exp

�gVjn

�; cj˛C (3)

where g is a scale parameter which is inversely proportional to thestandard deviation of the error terms and Vin is assumed to be linearin parameters:

Vin ¼X

k

Xikbik (4)

where X $ ik is the k attribute value of the alternative i and bik is thecoefficient associated with the k’th attribute. The marginal rate ofsubstitution between attributes is computed using

Table 2Characteristics of respondents.

Patient profile 1 2 3 4 Total

GenderMale (%) 48.9 50.0 49.6 45.9 48.6Female (%) 51.1 50.0 50.4 54.1 51.4

Mean age (SD) 56.51 (4.2) 56.69 (4.2) 56.20 (4.1) 56.74 (4.1) 56.53 (4.1)

EducationLow (%) 48.5 48.2 48.1 51.1 49.0Middle (%) 25.5 29.4 30.0 25.2 27.5High (%) 25.9 22.4 21.9 23.7 23.5

Income% Earning morethan average

43.8 44.1 42.6 44.0 43.6

A.P. Nieboer et al. / Social Science & Medicine 70 (2010) 1317–13251322

MRS ¼ � gba

gb¼ � ba

b(5)

�b

� �b

�

Because our design included a cost attribute it can be used toproduce an estimate of the ‘‘willingness to pay’’ (WTP) of attributea expressed in the units of the cost attribute by replacing thedenominator with the b estimate for this cost/price attribute:

WTPa ¼ ��

ba

bcost attribute

�(6)

These WTPs express the marginal WTP for a discrete change inan attribute level, and thus allow some understanding of the rela-tive importance that respondents give to attributes within thedesign. Confidence intervals were computed using a bootstrapapproach. For each bootstrap observation, we resampled a set ofrespondents with replication, equal in size to the original sampleand estimated the WTP. We repeated this procedure 10,000 timesto obtain a distribution of WTP estimates. Non-parametric confi-dence intervals were computed using the 2.5% cut-off value on bothsides of the distribution.

Results were broken down for low and high income groups(threshold: 40,000 euros per year as in Pavlova, Groot, & vanMerode, 2004) to account for the fact that WTP and preferences forspecific services may depend on income. Differences in WTP weretested using unpaired unequal t-tests based on the bootstrappedreplication results samples.

Table 3Willingness to pay for specified changes in the amount of care and modes of service del

Frail patient

Nopartner

4 h extra care per week 32*1 half day extra organized social

activities60*

Transportation service 120*

Living situationa

Apartment building 36Sheltered accommodation 35*Elderly/nursing home �3Regular care providers 36*Individual preferences 30*Coordinated care services delivery 35*Better punctuality (per change:

maximum wait reducedwith an hour)

25*

4 months shorter on waiting list 53*

*Indicates significance at 5% level.a In comparison with living independently at home (reference value).

Results

Respondents

In total 1859 (out of 3870 invited people) were administered theonline survey (48%), but 777 (20%) did not complete the ques-tionnaire. Accordingly, 1082 completed questionnaires wereobtained (28%). This is a normal response rate for this panel(personal communication, SSI). The obtained sample had the samedistribution of gender, age, and education groups as the generalDutch population in this age group: 51% was female, mean age was56.5 years (SD 4.2), 43% had a fulltime job, 72% was married orliving together, 56% earned less than 40,000 euros per year, 5% werecurrent users of long-term care, and 40% had indirect experienceswith long-term care through family and friends. The distribution ofrespondents’ characteristics was comparable over the 4 ‘‘patient’’profile groups (see Table 2).

Preferences for long-term care

Table 3 reports estimated WTP values. The regression results ofthe conditional logit model that were used to calculate these WTPsare reported in the Appendix. The WTP values show that respon-dents took all listed care dimensions into account: each dimensionwas significant for at least one patient profile. For three dimensionsthe WTP for one increment was comparatively low and differedonly marginally between patient profiles: amount of care (in

ivery (in euro’s).

Demented patient

Withpartner

Nopartner

Withpartner

21* 42* 26*41* 81* 26*

76* 88* 55*

28 177* 2412 64* 9

�18* 72* �649* 154* 88*22* 24 2116 154* 39*

17* 44* 30*

34* 104* 38*

Table 4Differences in willingness to pay between income groups.

Lowincome

Highincomeb

Ratio

Transportation service 113 183 0.62Regular care providers 101 158 0.64Coordinated care services delivery 71 92 0.784 months shorter on waiting list 42 62 0.671 half day extra organized social activities 38 61 0.63

Living situationa

Apartment building 61 17 3.56Sheltered accommodation 47 28 1.67Elderly/nursing home 42 55- �0.76*Individual preferences 30 61 0.494 h extra care per week 24 36 0.66Better punctuality (per change: maximum

wait reduced with an hour)22 34 0.64

*Indicates a significant difference.a In comparison with living independently at home (reference value).b A high income is defined as over 40,000 euros per year (gross).

A.P. Nieboer et al. / Social Science & Medicine 70 (2010) 1317–1325 1323

addition to a minimum level of 4 h), room for individual prefer-ences, and punctuality. Services that were valued high for mostbeneficiaries were transportation, social activities, and a regularcare provider. Other preferences differed across hypotheticalpatients.

All services were found to be of greater importance to peopleliving alone than to people with a partner. The value difference wasmost pronounced for the amount of social participation, protectedliving, coordinated care services, and time on a waiting list. It is thecontent of services rather than the modes of service delivery thatseemed to be particularly important for people without a partner.Less clear was the influence of the type of illness that makes peopledepend on long-term care. As expected, it was not dementia initself, but rather the combination of dementia with marital statusthat increased the value attached to a service. In particular, fordemented elderly without a partner, high value was attributed tohaving a single care provider, coordinated care services, shortertime on a waiting list, more participation in organized socialactivities and protected housing. Transportation was consideredless important for dementia patients than for fragile elderly. Thisservice was valued highest for physically frail elderly withouta partner. A possible explanation is that these persons’ opportu-nities to realize well-being need to be sought outside the home.

The results for dementia patients with a partner were mostlycomparable with frail elderly with a partner, except that fordementia patients more value was attributed to regular careproviders and coordinated care services delivery. For frail elderlywith a partner, living independently at home was preferred tomoving to an elderly/nursing home, even at extra cost.

Effect of income on WTP

Table 4 shows the results from the survey broken down for low/high income groups. In general, respondents with higher incomeswere willing to pay about 30% more than those with low income (seeratio). There were two exceptions. First, low income groups werewilling to pay more for protected housing than were high incomegroups. The latter were even willing to pay 55 euros extra per weekin order to be able to keep on living at home as compared to thealternative of moving to an elderly home or a nursing home. Second,the higher income groups valued care geared to individualpreferences, rather than standard care. The differences between thetwo income groups were consistent – in line with theory – butlargely statistically insignificant. The sole exception concerned

nursing homes, for which preferences were pointing in oppositedirections.

Discussion and conclusion

This paper examined the relative values of long-term careservices across beneficiaries. The results showed that well-beingmay be improved by services providing for physical and socialneeds. Given a minimum level of physical well-being, care servicesthat contribute to social well-being received high value. Extra hoursof personal and domestic care – in addition to the minimum of 4 h –were assigned less value. Organized social activities and trans-portation services for physically frail elderly and dementia patientswere given greater importance than additional hours of personaland domestic care.

The differences across patient groups (frail elderly and dementiapatients, both with and without partner) reflect differences in social,physical and mental vulnerabilities. This is why different value wasattributed to services when offered to different patient groups.Opinions on suitable living arrangements – for example – dependedon the characteristics of the beneficiary. Elderly or nursing homeswere mostly considered appropriate for dementia patients withouta partner, and WTP to reduce time on the waiting list was alsohighest for this group, probably because these persons are seen asmost at risk for loss of well-being. High income respondentsattached great value to avoid having to move to a home, reflectingthat they have more to lose in terms of living comfort.

The extent to which different aspects of care are judged tocontribute to people’s well-being is by and large in accordance withour expectations, offering support for the validity of our results. Forexample, high value was attached to having regular care providers,especially for dementia patients. This is in line with SPF theory(Gerritsen et al., 2004; Steverink, 2001): for these patients theability to build a relationship with the care provider may contributeto both physical and social well-being. Likewise, SPF theoryexplains why transportation received high value: it can be seen asan important resource for the multifunctional realization of socialand physical well-being that provides control over one’s life (to gowherever you like, whenever you like).

Some results seem contrary to SPF theory, such as the low valueattached to the extent to which a patient’s individual preferencesare taken into consideration. Further analysis of the data showedthat preferences for this attribute were related to income: respon-dents with higher incomes appeared somewhat more appreciativeof individual freedom of choice than low income groups. The lattermay be more concerned with ensuring that their needs are all met,whereas the former seek more control of individual life style. Thisresult therefore does not necessarily indicate limited validity of ourstudy. We were also surprised by the low WTP for care providers’punctuality. We hypothesize that respondents may have under-estimated the impact that waiting for care can have on dailyroutines and sense of control, perhaps because we did not provideselect options for how often care was delivered too late.

It may be seen as a limitation that preferences were not elicitedin patients, since patients are often considered the best judge oftheir own situation (Froberg & Kane, 1989; Sprangers & Aaronson,1992). We opted for proxy elicitation for several reasons. First, it isoften family members who decide on long-term care, not thepatient. In that respect, our choice context was realistic. Moreover,preferences of current users may be biased by ‘status quo’. Patientsmay have difficulty in imagining future care scenarios in the light ofexperiences with, and dependency on, current care practices. Thisbias will not affect preferences of future users (‘veil of ignorance’).Moreover, from an insurance perspective, the use of societal pref-erences is considered appropriate, because distribution of public

Ap

pen

dix

Frai

l,w

ith

out

par

tner

Frai

l,w

ith

par

tner

Dem

ente

d,w

ith

out

par

tner

Dem

ente

d,w

ith

par

tner

Coe

ffp-

valu

e(9

5%

CI)

Coe

ffp-

valu

e(9

5%

CI)

Coe

ffp-

valu

e(9

5%

CI)

Coe

ffp-

valu

e(9

5%

CI)

Nu

mb

erof

hou

rsof

care

per

wee

k0

.15

50

.00

0(0

.10

0,0

.21

0)

0.1

27

0.0

00

(0.0

73

,0.1

81

)0

.11

70

.00

0(0

.06

2,0

.17

2)

0.1

46

0.0

00

(0.0

91

,0.2

00

)O

rgan

ized

soci

alac

tivi

ties

0.2

85

0.0

00

(0.2

29

,0.3

41

)0

.24

70

.00

0(0

.19

2,0

.30

3)

0.2

27

0.0

00

(0.1

71

,0.2

83

)0

.14

40

.00

0(0

.09

0,0

.19

9)

Tran

spor

tati

onse

rvic

e�

0.5

72

0.0

00

(�0

.66

7,�

0.4

76

)�

0.4

59

0.0

00

(-0

.55

3,�

0.3

65

)�

0.2

47

0.0

00

(�0

.34

2,�

0.1

52

)�

0.3

06

0.0

00

(�0

.40

0,�

0.2

11

)

Livi

ng

situ

atio

na

Ap

artm

ent

bu

ild

ing

0.1

71

0.0

40

(0.0

08

,0.3

33

)0

.17

10

.03

8(0

.00

9,0

.33

2)

0.4

98

0.0

00

(0.3

32

,0.6

63

)0

.13

70

.09

9(�

0.0

26

,0.2

99

)Sh

elte

red

acco

mm

odat

ion

0.1

69

0.0

00

(0.0

75

,0.2

63

)0

.07

20

.13

3(-

0.0

22

,0.1

65

)0

.18

10

.00

0(0

.08

7,0

.27

6)

0.0

53

0.2

70

(�0

.04

1,0

.14

7)

Eld

erly

/nu

rsin

gh

ome

�0

.01

30

.62

8(�

0.0

68

,0.0

41

)�

0.1

11

0.0

00

(-0

.16

5,�

0.0

57

)0

.20

10

.00

0(0

.14

6,0

.25

7)

�0

.03

50

.19

9(�

0.0

90

,0.0

19

)W

ho

pro

vid

esca

re�

0.1

73

0.0

00

(�0

.26

7,�

0.0

79

)�

0.2

97

0.0

00

(-0

.39

1,�

0.2

03

)�

0.4

34

0.0

00

(�0

.52

9,�

0.3

39

)�

0.4

93

0.0

00

(�0

.58

8,�

0.3

98

)In

div

idu

alp

refe

ren

ces

0.1

45

0.0

03

(0.0

51

,0.2

39

)0

.13

30

.00

5(0

.04

0,0

.22

7)

0.0

68

0.1

57

(�0

.02

6,0

.16

3)

0.1

17

0.0

15

(0.0

23

,0.2

11

)C

oord

inat

edca

rese

rvic

esd

eliv

eryb

�0

.16

90

.00

0(�

0.2

63

,�0

.07

5)

�0

.09

40

.05

0(-

0.1

87

,0.0

00

)�

0.4

33

0.0

00

(�0

.52

9,�

0.3

38

)�

0.2

20

0.0

00

(�0

.31

4,�

0.1

26

)Pu

nct

ual

ity

�0

.11

90

.00

0(�

0.1

73

,�0

.06

4)

�0

.10

30

.00

0(-

0.1

57

,�0

.04

9)

�0

.12

20

.00

0(�

0.1

78

,�0

.06

7)

�0

.16

50

.00

0(�

0.2

20

,�0

.11

0)

Wai

tin

gli

stin

mon

ths

�0

.25

40

.00

0(�

0.3

10

,�0

.19

8)

�0

.20

40

.00

0(-

0.2

59

,�0

.14

9)

�0

.29

40

.00

0(�

0.3

51

,�0

.23

7)

�0

.21

20

.00

0(�

0.2

67

,�0

.15

7)

Co-

pay

men

tp

erw

eek

�0

.23

90

.00

0(�

0.2

94

,�0

.18

3)

�0

.30

20

.00

0(-

0.3

58

,�0

.24

5)

�0

.14

10

.00

0(�

0.1

96

,�0

.08

6)

�0

.28

00

.00

0(�

0.3

36

,�0

.22

3)

Logl

ikel

ihoo

d�

127

7�

12

84

�1

26

1�

127

0Ps

eud

oR

20

.16

00

.14

90

.15

80

.13

92

aIn

com

par

ison

wit

hli

vin

gin

dep

end

entl

yat

hom

e(r

efer

ence

valu

e)b

Defi

nit

ion

:‘A

san

extr

ase

rvic

eyo

ur

care

give

rca

nco

ord

inat

eal

lth

eca

reyo

ure

ceiv

e.Th

isal

som

ean

sth

atth

eyca

nre

spon

dq

uic

kly

wh

enyo

ur

hea

lth

det

erio

rate

sfu

rth

eran

dyo

un

eed

add

itio

nal

care

.Mor

eove

r,yo

ud

on

oth

ave

toar

ran

geyo

ur

own

care

wit

hin

div

idu

alca

regi

vers

.’

A.P. Nieboer et al. / Social Science & Medicine 70 (2010) 1317–13251324

funds affects everybody’s future lives, not just lives of current users,so everybody should have a say in their allocation. In spite of thearguments supporting use of the societal perspective, wewondered if values derived from a general population samplewould be informed and valid. Older persons might find it easierthan younger people to imagine living with impairments and toanticipate their own future care needs. Moreover, they are likely tobe aware of patient preferences and needs through stories, infor-mation and experiences of family members or friends. Accordingly,we restricted our sample to those aged between 50 and 65 years.We are reasonably confident that this approach enabled us toobtain preferences from a sufficiently informed sample, because45% of respondents indicated having had to think about this type ofcare for themselves (5%) or for others (40%).

Second, that the strength of people’s preferences was expressedin terms of WTP. The external validity of this approach is still beingdebated. WTP measurements have often been derived by contin-gent valuation methods (Diener, O’Brien, & Gafni, 1998), but thisapproach is now increasingly replaced by DCEs (Lancsar & Savage,2004). The latter have advantages related to their statistical prop-erties. They have greater assumed validity and reliability becausethey are grounded in theory on decision making and easily allowstudying both attributes and situational changes (Adamowicz,Boxall, Williams, & Louviere, 1998). Nevertheless, there is still littleempirical evidence that DCE based estimates are more valid thancontingent valuation (Ryan & Watson, 2008). So our findings shouldbe interpreted cautiously. It is unclear whether people would pay asmuch in real life as they stated that they would. External validity ofthe WTP estimates could further be limited for the lack of an ‘optout’ option. We left it out on purpose, because in real life people inurgent need of long-term care are also forced to select one of theavailable options. Moreover, we used WTPs merely as an estimateof the relative utility of services, and not as indication of true WTP.

The third limitation is that we referred to burden of caring andloss of well-being only in the description of the dementia patientwith a partner. From a societal perspective, patient- and caregiverwell-being should be considered as equally important in cost-effectiveness calculations of different care services and modes ofdelivery.

Our study provides insight into the benefits of a wide range oflong-term care services. These benefits are essential for optimalresource allocation as decision makers need to weigh these againstcosts for each service. Our results show that people’s preferencesfor long-term care depend on the patient’s illness and social envi-ronment as well as on the interaction between the two. Thesefindings support the Social Production Function theory, which hasit that care services may be assigned higher value through differ-entiation between (1) individuals in the extent to which theirphysical and social well-being are being threatened, (2) careservices to the extent they enable multifunctional activities,enhance a sense of control and avoid further loss.

In conclusion, the utility obtained from specific services differsbetween users. The benefit package of long-term care should bedefined by an optimal mix of services for specific patients. Therelative value people attribute to services differs because theirresources and restrictions differ. Disentangling how different careservices contribute to people’s social and physical well-beingfacilitates finding optimal ways to deliver care services in times ofincreasing scarcity.

Acknowledgements

The project was funded by the Federation of Patients andConsumer Organisations in the Netherlands (NPCF). The views inthe paper are those of the authors.

A.P. Nieboer et al. / Social Science & Medicine 70 (2010) 1317–1325 1325

References

Adamowicz, W., Boxall, P., Williams, M., & Louviere, J. (1998). Stated preferenceapproaches for measuring passive use values; choice experiments and contin-gent. American Journal of Agricultural Economics, 80, 64–75.

Algemene Wet Bijzondere Ziektekosten (AWBZ). (1967). Wet Van 14 December 1967, Stb.1967, 655, Houdende Regeling Van De Algemene Verzekering Bijzondere Ziektekosten.

Arinzon, Z., Peisakh, A., & Berner, Y. N. (2008). Evaluation of the benefits of enteralnutrition in long-term care elderly patients. Journal of the American MedicalDirectors Association, 9(9), 657–662.

Bouman, A., van Rossum, E., Nelemans, P., Kempen, G. I., & Knipschild, P. (2008).Effects of intensive home visiting programs for older people with poor healthstatus: a systematic review. BMC Health Services Research, 8, 74.

Coast, J., Flynn, T. N., Natarajan, L., Sproston, K., Lewis, J., Louviere, J. J., et al. (2008).Valuing the ICECAP capability index for older people. Social Science & Medicine,67(5), 874–882.

Diener, A., O’Brien, B., & Gafni, A. (1998). Health care contingent valuation studies:a review and classification of the literature. Health Economics, 7(4), 313–326.

Forster, A., Lambley, R., Hardy, J., Young, J., Smith, J., Green, J., et al. (2009). Reha-bilitation for older people in long-term care. Cochrane Database of SystematicReviews (Online), 1(1), CD004294.

Froberg, D. G., & Kane, R. L. (1989). Methodology for measuring health-state pref-erences–III: population and context effects. Journal of Clinical Epidemiology,42(6), 585–592.

Gerritsen, D. L., Steverink, N., Ooms, M. E., & Ribbe, M. W. (2004). Finding a usefulconceptual basis for enhancing the quality of life of nursing home residents.Quality of Life Research: An International Journal of Quality of Life Aspects ofTreatment, Care and Rehabilitation, 13(3), 611–624.

Groenewoud, A. S., van Exel, N. J., Berg, M., & Huijsman, R. (2008). Building qualityreport cards for geriatric care in the Netherlands: using concept mapping toidentify the appropriate ‘‘building blocks’’ from the consumer’s perspective. TheGerontologist, 48(1), 79–92.

Ham, C., & Robert, G. (Eds.). (2003). Reasonable rationing. International experience ofpriority setting in health care. Maidenhead, England: Open University Press.

Howell-White, S., Gaboda, D., Scotto Rosato, N., & Lucas, J. A. (2006). Creating needs-based tiered models for assisted living reimbursement. The Gerontologist, 46(3),334–343.

Kahneman, D., & Tversky, A. (1979). Prospect theory. An analysis of decision underrisk. Econometrica, 47, 263–291.

Lancaster, K. (1966). A new approach to consumer theory. Journal of PoliticalEconomy, 74, 132–157.

Lancsar, E., & Savage, E. (2004). Deriving welfare measures from discrete choiceexperiments: inconsistency between current methods and random utility andwelfare theory. Health Economics, 13(9), 901–907.

Lindenberg, S. (2001). Social rationality versus rational egoism. In J. H. Turner (Ed.),Handbook of sociological theory (pp. 635–668). New York: Kluwer Academic/Plenum.

Lindenberg, S. M. (1996). Continuities in the theory of social production functions.In S. M. Lindenberg, & H. B. G. Ganzeboom (Eds.), Verklarende sociologie:Opstellen voor Reinhard Wippler. [Rational choice sociology: Essays for ReinhardWippler] (pp. 169–184). Amsterdam: Thesis Publishers.

Manski, C. (1977). The structure of random utility models. Theory and Decision, 8,229–254.

Markle-Reid, M., Browne, G., Weir, R., Gafni, A., Roberts, J., & Henderson, S. R. (2006).The effectiveness and efficiency of home-based nursing health promotion forolder people: a review of the literature. Medical Care Research and Review:MCRR, 63(5), 531–569.

McFadden, D. (1974). Conditional logit analysis of qualitative choice behaviour. InP. Zarembka (Ed.), Frontiers in econometrics. New York: Academic Press.

de Meijer, C. A. M., Koopmanschap, M. A., Koolman, A. H. E., & van Doorslaer, E. K. A.(2009). The role of disability in explaining long-term care utilization. MedicalCare, 47(11), 1156–1163.

Miller, E. A., Booth, M., & Mor, V. (2008). Assessing experts’ views of the future oflong-term care. Research on Aging, 30(4), 450–473.

Mitchell, G., 2nd, Salmon, J. R., Polivka, L., & Soberon-Ferrer, H. (2006). The relativebenefits and cost of medicaid home- and community-based services in Florida.The Gerontologist, 46(4), 483–494.

Nieboer, A., Lindenberg, S., Boomsma, A., & van Bruggen, A. C. (2005). Dimensions ofwell-being and their measurement: the SPF-IL scale. Social Indicators Research,73, 313–353.

Nieboer, A. P. (1997). Life-events and well-being: A prospective study on changes inwell-being of elderly people due to a serious illness event or death of thespouse. Amsterdam: Thela Thesis.

Nieboer, A. P., & Lindenberg, S. (2002). Substitution, buffers and subjective well-being: a hierarchical approach. In. The universality of subjective wellbeing indi-cators (1st ed.). (pp. 175–189) Netherlands: Kluwer Academic Publishers.

OECD Health Project.. (2005). Long-term care for older people. Paris: OECDPublications.

Ormel, J., Lindenberg, S., Steverink, N., & Verbrugge, L. M. (1999). Subjectivewell-being and social production functions. Social Indicators Research, 46(1),61–90.

Ormel, J., Lindenberg, S., Steverink, N., & VonKorff, M. (1997). Quality of life andsocial production functions: a framework for understanding health effects.Social Science & Medicine, 45(7), 1051–1063.

Pavlova, M., Groot, W., & van Merode, G. (2004). Willingness and ability of Bulgarianconsumer to pay improved public health care services. Applied Economics,36(10), 1117.

Rutten, F. F. H., & Busschbach, J. J. V. (2001). How to define a basis package of healthservices for a tax funded or social based health care system? European Journal ofHealth Economics, 2, 45.

Ryan, M., Netten, A., Skatun, D., & Smith, P. (2006). Using discrete choice experi-ments to estimate a preference-based measure of outcome – an application tosocial care for older people. Journal of Health Economics, 25(5), 927–944.

Ryan, M., & Watson, V. (2008). Comparing welfare estimates from payment cardcontingent valuation and discrete choice experiments. Health Economics, 18(4),389–401.

Sloane, N. J. A. (2003). A library of othogonal arrays. Retrieved 11/17/2005, from.www.research.att.com/wnjas/oadir.

Sprangers, M. A., & Aaronson, N. K. (1992). The role of health care providers andsignificant others in evaluating the quality of life of patients with chronicdisease: a review. Journal of Clinical Epidemiology, 45(7), 743–760.

Steverink, N. (2001). When and why frail elderly people give up independent living:The Netherlands as an example. Ageing and Society, 21, 45–69.

Steverink, N., & Lindenberg, S. (2006). Which social needs are important forsubjective well-being and what happens to them with aging? Psychology andAging, 21(2), 281–290.

Street, D., Burgess, L., & Louviere, J. J. (2005). Quick and easy choice sets: con-structing optimal and nearly optimal stated choice experiments. InternationalJournal for Research in Marketing, 22, 459–470.

Thompson, P., Lang, L., & Annells, M. (2008). A systematic review of the effective-ness of in-home community nurse led interventions for the mental health ofolder persons. Journal of Clinical Nursing, 17(11), 1419–1427.

Van Bruggen, A. C. (2004). Individual production of social well-being: an explor-atory study. (Thela Thesis). Ph.D.-Thesis, University of Groningen, (Downloadableat: dissertations.ub.rug.nl/files/faculties/ppsw/2001/a.c.van.bruggen).

Wet Maatschappelijke Ondersteuning (WMO) (27 Mei 2005).WHO. (2005). World Health Organization 2005. Available from. www.who.int/topics/

chronic_disease/en.