Analyzng the determnants of wllngness-to-pay values for testng the valdty of the contngent valuaton method. Applcaton to home care compared to hosptal care Mare-Odle Carrère, Nathale Havet, Magal Morelle, Raphaël Remonnay To cte ths verson: Mare-Odle Carrère, Nathale Havet, Magal Morelle, Raphaël Remonnay. Analyzng the determnants of wllngness-to-pay values for testng the valdty of the contngent valuaton method. Applcaton to home care compared to hosptal care. Workng paper GATE 08-20. 2008. <halshs-00303725> HAL Id: halshs-00303725 https://halshs.archves-ouvertes.fr/halshs-00303725 Submtted on 22 Jul 2008 HAL s a mult-dscplnary open access archve for the depost and dssemnaton of scentfc research documents, whether they are publshed or not. The documents may come from teachng and research nsttutons n France or abroad, or from publc or prvate research centers. L archve ouverte plurdscplnare HAL, est destnée au dépôt et à la dffuson de documents scentfques de nveau recherche, publés ou non, émanant des établssements d ensegnement et de recherche franças ou étrangers, des laboratores publcs ou prvés.
GATE Groupe d Analyse et de Théore Économque UMR 5824 du CNRS DOCUMENTS DE TRAVAIL - WORKING PAPERS W.P. 08-20 Analyzng the determnants of wllngness-to-pay values for testng the valdty of the contngent valuaton method. Applcaton to home care compared to hosptal care Raphaël Remonnay, Nathale Havet, Magal Morelle, Mare-Odle Carrère Jullet 2008 GATE Groupe d Analyse et de Théore Économque UMR 5824 du CNRS 93 chemn des Moulles 69130 Écully France B.P. 167 69131 Écully Cedex Tél. +33 (0)4 72 86 60 60 Fax +33 (0)4 72 86 60 90 Messagere électronque gate@gate.cnrs.fr Serveur Web : www.gate.cnrs.fr
ANALYZING THE DETERMINANTS OF WILLINGNESS-TO- PAY VALUES FOR TESTING THE VALIDITY OF THE CONTINGENT VALUATION METHOD. APPLICATION TO HOME CARE COMPARED TO HOSPITAL CARE RAPHAEL REMONNAY, NATHALIE HAVET, MAGALI MORELLE, MARIE-ODILE CARRERE Unversty of Lyon, Lyon, F-69003, France; CNRS, UMR 5824, GATE, Ecully, F-69130, France; ENS LSH, Lyon, F-69007, France ; Centre Leon Berard, Lyon, F-69003, France Assstant Professor. Professor. Runnng head: Testng contngent valuaton valdty Key-words: contngent valuaton; valdty; protest responses; censored data; home care Name, address, telephone number and emal address of the correspondng author: Raphaël Remonnay Axe Econome de la Santé, GATE, UMR 5824 CNRS, Centre Léon Bérard, 28 rue Laennec, 69008 Lyon, France Tel: (33).4.78.78.27.80 Fax: (33).4.78.78.28.04 E-mal: remonnay@lyon.fnclcc.fr Word count (text ):5347 Fgures: 0 Tables: 3 Fnancal support: Our study receved fnancal support from the French Mnstry of Health. 1
SUMMARY The contngent valuaton (CV) method s an attractve approach for comparng home care to hosptal care n whch the only dfference s patents well-beng durng the treatment process and not health outcomes. We consdered the emprcal stuaton of blood transfuson (BT) n cancer patents and collected wllngness to pay (WTP) values among BT users. Our man objectve was to test the valdty of the CV method, namely ts ablty to elct true preferences. Frstly, possble determnants of WTP values and ther expected nfluences were dentfed, from both economc and non economc lterature and from the fndngs of a plot study. Secondly, they were compared to predcted nfluences resultng from approprate econometrc analyss of WTP values elcted by a bddng process. From the health economcs lterature t appeared that the double-hurdle model s the most approprate approach to account for zero values and protest responses. However, because the number of protest responses was too small, we used a truncated regresson model. None of the 7 hypotheszed nfluences was nvaldated by econometrc results. The anchorng bas hypothess was confrmed. The WTP for home BT compared to hosptal BT ncreased wth household ncome, wth prevous experence of home care, wth lvng far from the hosptal and wth low qualty of lfe. Conversely, t was lower for advanced-stage (pallatve or termnal) than for early-stage (curatve) patents. We conclude that the CV approach s acceptable to severely ll patents. Moreover, WTP values demonstrate good valdty gven that nfluences predcted by our model are consstent wth expected determnants. 2
1. INTRODUCTION In recent years, a growng number of studes conducted n the health care feld have used contngent valuaton (CV), as reported n several lterature revews (Dener et al., 1998; Olsen and Smth, 2001; Smth, 2003; Yeung et al., 2006). The CV method has been used for evaluatng all types of health care strateges, ether preventve, dagnostc or therapeutc. Meanwhle, ts ablty to provde the true preferences of the respondents has been extensvely questoned. Accordngly, numerous valdty propertes such as crteron valdty, content valdty and construct valdty have been defned and emprcally tested usng wllngness-to-pay (WTP) responses. Analyzng the determnants of WTP values can be a valuable way of assessng the valdty of the CV method. Indeed, possble determnants and ther expected nfluences can be derved from theoretcal predctons or emprcal lterature, then confrmed or not usng WTP data. For example, the respondent ncome should postvely nfluence WTP values accordng to construct valdty (Donaldson, 1999; Smth et al., 1999 b ; Drummond et al., 2005). However, analyzng the determnants of WTP has receved lmted attenton n health economcs. The emprcal lterature about WTP has focused more on the method used for data collecton and on possble consequences on the monetary values obtaned (Smth 2003; Smth 2006; Smth 2007 a, b ). The technque used for econometrc analyss of WTP data s nevertheless an mportant ssue because an napproprate choce can lead to erroneous nferences about the determnants of WTP and consequently about valdty. Econometrc modelng should ft the type of WTP queston asked (Donaldson et al., 1995; Donaldson et al., 1998; Dalmau-Matarrodona, 2001; 3
Kurth et al., 2004). A closed-ended approach and a dchotomous choce wth follow-up provde qualtatve bnary responses whle a payment scale approach, an open-ended queston and a bddng process generate contnuous monetary WTP values (Donaldson et al., 1998). In the 3 latter cases, the WTP dstrbuton s usually censored so that the data may contan a large proporton of zero values, for whch there s a range of possble explanatons such as protest responses or real zeros consstent wth economc decsons. The objectve of the present study was to analyze the determnants of WTP values n the framework of a CV survey amed at comparng hosptal care to home care. Home care s a promsng alternatve to hosptal care n many health care stuatons. However, to our knowledge, the two optons have never been compared usng the CV approach. We shall frst present the emprcal stuaton studed, the CV study desgn, the WTP questonng process, the selecton of possble determnants of WTP values and sample characterstcs. Secondly, the health economcs lterature on the econometrc methods currently used for analyzng determnants of WTP data wll be nvestgated, wth a specal focus on censored data and protest responses. Thrdly, the WTP values obtaned wll be analyzed usng approprate econometrc modelng, and results wll be compared to the expected nfluences of pre selected possble determnants. In concluson, we wll dscuss the mplcatons of our study regardng the valdty of the CV method. 4
2. EMPIRICAL SITUATION, CV STUDY DESIGN AND DATA COLLECTION 2.1. Emprcal stuaton CV s an attractve alternatve to the Qualty-adjusted lfe years (QALYs) approach when t comes to comparng two types of health care management whch dffer only n patents wellbeng durng the treatment process and not n health outcomes (Ryan and Shackley, 1995; Olsen and Smth, 2001). Ths s the case for some treatments whch can be delvered at home or n the hosptal day-care unt, such as blood transfuson (BT) n cancer care. BT s currently admnstered to advanced-stage cancer patents for curatve, pallatve or termnal ntent. BT s smple to admnster, wth dentcal effectveness and safety at home, n the framework of a homecare network, or n the hosptal (Idr et al., 1996; Madgwck and Yarduman, 1999). Admnstraton at home requres nether transportaton between home and hosptal nor watng tme n the hosptal and may decrease the patent's tredness; however, t may also ncrease the feelng of nsecurty and solaton. As a result, home admnstraton compared to hosptal admnstraton may be vewed dfferently from one patent to another. 2.2. Study desgn A CV survey conducted n the Comprehensve Cancer Centre of the Rhone-Alps Regon n Lyon, France, compared hosptal BT to home BT. As sad before, our analyss focused on WTP values. Wllngness-to-accept (WTA) values were also collected for further analyss. A prospectve plot study usng the same desgn as the man study was conducted n 40 elgble patents to assess whether a CV survey was feasble n ths stuaton,.e. among patents personally nvolved n the treatment under consderaton, most of them severely ll. 5
We also amed at dentfyng possble determnants of CV responses specfc to the emprcal stuaton under nvestgaton. The plot study demonstrated the feasblty of the CV survey: of the 44 patents planned, only 4 dd not consent to partcpate because they felt too tred for the 45-mnute ntervew. Therefore, a prospectve study was conducted on a 12 months perod n 2003 and 2004. All cancer patents needng a BT and aged more than 18 years were asked to partcpate. All were elgble to receve BT ether at home or n the hosptal. They were allocated between home BT and hosptal BT accordng to organzatonal possbltes. For example, home BT was only avalable n a few well defned geographcal areas. After obtanng nformed consent, and less than 48 hours after BT admnstraton, face-to-face ntervews were conducted by a traned ntervewer on the bass of a detaled gude. Although costly, ths type of ntervew was chosen for mnmzng hypothetcal bases and mprovng the qualty and rate of responses (Mtchell and Carson, 1989; NOAA, 1993, 1994; Smth et al., 1999 a ). Patents were frst asked whether they had already undergone home BT (ncludng the current procedure when t was admnstered at home) and whether they had already receved home care, other than BT. They were then gven a detaled presentaton of the two BT managements, ether at home or n the hosptal, and were told that effectveness and safety were dentcal n both cases. Fnally, they were asked where they would choose to receve ther next BT, would they need another one n the future and would the two BT managements be avalable. 6
2.3. WTP questonng process Once they have chosen the BT arrangements, all patents were gven a general presentaton of the CV method, whch was ntroduced as a way of measurng ther preference for the type of BT management they had chosen, as compared to the other one, wthout consderaton of the correspondng costs for the health care system. Then they were asked to magne a hypothetcal future stuaton where they would need another BT and where the only freely avalable management would be the one they had not chosen. Then the ntervewer asked them how much they would be wllng to pay to get ther preferred BT management rather than the free alternatve. We chose out-of-pocket expendture whch s the recommended payment vehcle when respondents are personally nvolved (Smth, 2003; Mtchell and Carson, 1989; O Bren and Gafn, 1996). Thus, n reference to prospect theory (Kahneman and Tversky, 1979), WTP was framed as a gan (Sayman and Öncüller, 2005). Indeed, we consdered that framng the WTP queston as a gan rather than a loss would be easer to understand. For askng WTP questons we preferred usng a bddng process rather than closed-ended questons because of our small sample sze (Donaldson et al., 1998). The process nvolved three steps. Frst, an ntal bd of money was proposed to the patents. If they agreed to pay, we proposed a hgher bd, whereas f they dd not we proposed a lower one. Ths step was repeated twce. Patents were randomly assgned to two ntal bds of money (38 and 76 ) to test for a possble anchorng bas affectng content valdty (Herrges and Shogren, 1994; Flachare and Hollard, 2007). The range from 38 to 76 was chosen because t was close to the range of fees normally covered by socal nsurance n France for a home vst by a specalst (ncludng call-out charges). 7
2.4. Selecton of possble determnants of WTP values As sad before, a possble anchorng bas was tested: accordng to health economcs lterature, choosng a hgher ntal bd value could ncrease WTP responses, but would n no way decrease them (Smth et al., 1999 b ). Apart from ths possble nfluence related to the framng of questons, we also nvestgated some personal characterstcs of the patents, selected on the bass of prevous knowledge of ther possble nfluence on patents preferences and WTP values. The health economcs lterature suggests that ncome postvely nfluences WTP values (Donaldson, 1999; Smth et al. 1999 b ; Drummond et al., 2005). Hence patents were asked ther household ncomes. The medcal lterature ndcates that prevous experence of home care promotes preference for home care compared to hosptal care. A Cochrane Collaboraton lterature revew by Shepperd and Ilffe compared home care to hosptal care n randomzed studes (Shepperd and Ilffe, 2001). They concluded that patents who had no a pror preferences (snce they agreed to partcpate n a randomzed tral) generally expressed hgher satsfacton after experencng home care compared to hosptal care. Accordngly, we asked our patents whether they had prevously experenced home care, ether for BT or any other procedure. In the plot study, patents were asked whch BT management they preferred, and why. More precsely, they were proposed a lst of possble reasons for ther choce and were asked to state the mportance of each reason usng a 5-pont Lkert scale rangng from not mportant at all to very mportant. Reasons were statstcally compared accordng to mportance usng a Fredman s test, and a follow-up multple comparson test. On the one hand, the most mportant reasons for preferrng home BT were that t avoded dsrupton n daly lfe and that 8
t spared transportaton and watng tme n the hosptal. On the other hand, hosptal BT was manly preferred because t allowed to clearly separate daly lfe from health care and because t was perceved as safer. We assumed that, all other thngs beng equal, patents would be all the more wllng to spare transportaton and watng tme as they lved far from the hosptal and as ther qualty of lfe (ncludng tredness) was low. We thus selected two possble determnants of patents' preferences: dstance between home and hosptal, and a health-related qualty of lfe ndex ncludng tredness as measured by the Functonal Assessment of Cancer Treatment General scale (FACT-G) (Cella, 1993). Consderng safety, we assumed that patents would be all the more senstve to ths concern as ther lfe was threatened, whch was taken nto account by collectng nformaton on ther dsease stage. Three stage levels were used: curatve, pallatve and termnal. Regardng daly lfe, patents preferences referred to contradctory reasons. Some wshed to avod dsruptons n daly actvtes, whereas others wanted to clearly separate daly lfe from health care. We collected data on patents characterstcs whch could nfluence ther feelngs about daly lfe,.e. martal status and havng chldren at home or not, wthout foreknowledge of ther nfluence on preferences. Other varables lke standard demographcs, gender, age and employment status were collected as controls, wth no a pror assumpton whatsoever on ther possble nfluences. 2.5. Sample characterstcs On the study perod, all 153 elgble patents consented to partcpate. However, 14 people could not be ntervewed wthn the 48 hours followng ther BT ether because of lack of 9
avalablty or because they felt too tred. As a result, 139 patents were enrolled, whch corresponds to a 90% response rate. Ther characterstcs are presented n Table I. All patents characterstcs were documented by the patents themselves, except stage of dsease whch was assessed as curatve, pallatve or termnal by the oncologst n charge of the Home Care Unt at the Cancer Centre. The qualty of lfe ndex measured usng the FACT-G scale vared from 0 (worst possble stuaton) to 108 (best possble stuaton). [Insert Table I over here] Medan monthly household ncome was between 1500 and 2300. Almost one n two respondents had prevous experence of home BT on the one hand, and of home care (except home BT) on the other (43.2% and 45.3% respectvely). Dstance from home to hosptal was close to 35 km n average, wth a hgh 100% varaton coeffcent. Qualty of lfe accordng to the FACT-G scale was rather poor, wth an average ndex of 61.7 (range 0-108). Regardng stage of dsease, patents were almost equally dstrbuted between curatve stage on the one hand and pallatve or termnal stage on the other. Standard demographcs were as follows: patents were 57.5 year old n average (SD=12.8) and 3.2% had a professonal occupaton. Fnally, patents were equally dstrbuted between males and females. 10
3. HEALTH ECONOMICS LITERATURE SURVEY: ECONOMETRIC MODELS FOR ANALYZING DETERMINANTS OF WTP VALUES RESULTING FROM A BIDDING PROCESS For Donaldson et al. (1998), the approprate technque for econometrc analyss of WTP data depends frstly on the type of queston asked. For example, closed-ended questons and dchotomous choce wth follow-up valuaton only generate qualtatve responses for WTP. Ths s why t s advsed to use dscrete choce models such as bnary logt and probt for regresson analyss (Johannesson et al., 1991; Johannesson et al., 1993). For data elcted by usng a payment scale, the most approprate econometrc methodology s grouped data regresson, also called nterval regresson or ordered logt/probt (Donaldson et al., 1998; Yasunaga et al., 2006; Bärnghausen et al., 2007). When usng an open-ended queston or a bddng process as was the case n our study - the WTP values obtaned are quanttatve and several modelng methods have been proposed n the lterature. For responses obtaned through the bddng process, the frst regresson analyss studes conducted n the health care feld began n the 1980s and mostly estmated standard lnear models by ordnary least squares (OLS) (Berwck and Wensten, 1985; Thompson, 1986; O Bren and Vramontes, 1994; Medzybrobzka et al., 1995). 1 However, the observed data for WTP responses are generally censored. When analyzng the dstrbuton of WTP, we generally observe that the WTP varable does not take values below zero and has postve densty at zero. The large proporton of zeros calls nto queston the contnuty of the dependent varable and consequently the use of the classcal multple regresson model. In the presence of data censorng, OLS estmaton yelds based and nconsstent estmates because 1 The dependent varable of WTP values could be expressed n level, n logarthm or n square roots. Besdes, Amn and Khondoker (2004) used a lnear specfcaton for WTP values, but supposed that they followed a log-normal dstrbuton. Consequently, ther estmatons were based on maxmum lkelhood estmators nstead of OLS. 11
t fals to account for the qualtatve dfference between lmt (zero) observatons and nonlmt (contnuous) observatons. Consequently, all conclusons on the determnants of WTP are potentally erroneous. The Tobt model s the correct alternatve frequently proposed for such censored data n contngent valuaton lterature n health economcs (Donaldson et al., 1995; Kurth et al., 2004). However, the underlyng assumpton n the Tobt model s that the same specfcaton s used both for the contnuous and the zero decson processes. Ths mples that the Tobt specfcaton s relevant only f all zero realzatons represent an economc decson,.e. a real zero preference for the health care program under evaluaton. Other qute reasonable determnants of zero observatons, such as the presence of protest zeros whch are common n contngent valuaton surveys, also exst. Protest responses do not reflect ndvdual patent preferences and have no economc sgnfcance (Lndsey, 1994; Jorgensen et al., 1999). They may reflect partcpants' objecton to beng asked to complete the questonnares, or ndcate an deologcal poston about the napproprateness of placng monetary values on health. Protest patents refuse to reveal ther real wllngness even though t s postve. On the contrary, respondents genunely place zero value on the WTP queston, through statng that they have no strong (nl or very small) preference for t or they cannot afford to pay. Ther responses have true economc sgnfcance. The ssue at stake n the lterature has thus been the development of approprate emprcal models to treat protest responses n addton to censored data. Several approaches have been proposed. The smplest and most frequently used soluton s to dscard protest zeros. Nevertheless, ths approach s generally ncorrect from a statstcal pont of vew: a sample selecton bas s ntroduced f the socoeconomcs and other personal characterstcs of the protest patents are 12
sgnfcantly dfferent from the rest of the sample. Under such condtons, dscardng protest responses produces based estmates, whch are not nterpretable and unusable to test valdty. The second soluton s to ndrectly address the ssue of protest responses. Because zero responses may have explanatons other than a genune zero WTP, Donaldson et al. (1998) consdered that postve WTP values and zero values could sgnfcantly dffer n ther determnants. In fact, postve WTP values stem from an economc decson-makng process whereas zero values are a mxture of sgnfcant economc responses and protest responses. They proposed a specfcaton of the censorng mechansm more flexble than Tobt, called the type II Tobt model (Amemya, 1984). 2 Ths model permts the coexstence of dfferent patterns for the queston of how much and whether to pay for the care under evaluaton. One set of parameters determnes the mpact of the characterstcs on the probablty to record a postve WTP value, and the second set characterzes the determnants of the postve WTP amount: y 1 * 1 f y 1 = X 1β1 + u 1 > 0 = 0 else ( postve WTP value) ( zero value for WTP), and y * y f y 1 = 1 = 0 else, * * * wth y = X β + u, y the WTP value and and latent varables. 2 2 2 y 1 y However, ths approach does not dfferentate zeros generated by economc decsons (genune zero values) from zeros generated by non-economc decsons (protest zeros), whch makes nterpretaton dffcult. 2 Ths econometrc specfcaton s also called the sample selecton model. The two-part model s a specal case of the type II Tobt model. 13
The thrd econometrc approach, the double-hurdle model used by Dalmau-Matarrodona (2001), represents an nterestng modfcaton of the type II Tobt model by explctly emphaszng protest responses. 3 It decomposes the behavour of ndvduals n the decsonmakng process n two parts: frst, the reasons for decdng to partcpate or not n the contngent market offered,.e. to gve or not a protest response (partcpaton equaton: frst hurdle), and second, the decson on the amount to consume, that s how much they are wllng to pay for the procedure (consumpton equaton: second hurdle). Partcpaton equaton: P * 1 f P = Z γ + u > 0 = 0 else; ( protest responses) Consumpton equaton: y * * y f y > 0 and = 0 f P = 0 or f P P = 1 = 1 and y * = 0, wth y = X β + u * Compared wth the type II Tobt model, the man advantage of the double-hurdle model s that the partcpaton equaton allows not only to determne the socoeconomc and personal characterstcs of the protest respondents, but also to better defne the WTP equaton. In fact, n the type II Tobt model, the expresson of the WTP equaton depends on a selecton equaton whch determnes whether a postve WTP value s recorded. In the double-hurdle model, the WTP equaton depends on the partcpaton equaton. But, because opposng protest responses to responses that reflect true preferences has more economc sense than opposng zero values to postve values, the selecton s better taken nto account n the 3 For a general presentaton of ths model, see Jones (1989, 1992), Blundell et al. (1987). 14
double-hurdle methodology. Consequently, the consumpton equaton n the double-hurdle model allows to obtan unbased and effcent estmatons for the determnants of WTP for the whole sample. 4 Ths s why the double-hurdle model seems to consttute the most robust approach for emprcally testng valdty n contngent valuaton n the presence of censored data and protest responses. 5 Even f the double-hurdle approach s attractve, t cannot provde relable estmates when the number of protest responses s too small. In that case, the most approprate econometrc methodology s a truncated regresson model (Mahmud, 2006). The estmaton s based on strct postve WTP only and takes nto account the elmnaton of all zero values to obtan * * vald results for all the patents: y = f y > 0 wth y = X β + u. y Parameters are estmated usng the maxmum lkelhood prncple rather than OLS. Indeed, wth only lmted numbers of protest responses, the type II Tobt model could be estmated, but the nterpretaton of the determnants of zero responses would reman problematc because the model does not dfferentate genune zero values from protest zeros. In ths case, t s preferable to estmate a truncated regresson model because all parameters have an economc nterpretaton. 4 5 Moreover, Flood and Gräsjo (1998) have shown, usng Monte-Carlo smulatons, that the double-hurdle specfcaton s more general than the type II Tobt model. If the data are generated by type II tobt, the double-hurdle model stll produces correct results, and f the data generaton process s double-hurdle, serous bases can be avoded usng double-hurdle nstead of type II tobt. The double-hurdle model can be vewed as a tobt method wth selectvty. 15
4. ANALYZING THE DETERMINANTS OF OUR WTP VALUES 4.1. Choce of modelng As reported n Secton 2, WTP data were collected n 139 patents. Forty-two patents responded that they would prefer to receve BT n the hosptal versus 97 at home,.e. 30% versus 70%, respectvely. Gven that a large majorty of patents preferred home BT to hosptal BT, we decded to favour WTP data assocated wth preference for home BT. Accordngly, the 42 patents preferrng hosptal BT were consdered to have real zero or negatve WTP values for home BT. Of the 97 patents preferrng home BT, 3 expressed comprehenson problems wth the WTP queston and ther responses were elmnated from emprcal analyss. Of the 94 remanng patents, 74 provded a strct postve WTP response and 20 chose a zero value ndcatng ether protest responses or real zeros. The dentfcaton of protest zeros was based on responses to a follow-up queston on the reasons for unwllngness to answer. Eght respondents reacted negatvely to the payment vehcle and gave reasons for ther zero WTP value such as I have pad health nsurance premums all my lfe and I should not have to pay anythng more, the goal of ths knd of study s that we pay more and more for care, I do not want to pay for blood when the donors are voluntary. These 8 respondents were consdered protest bdders. The other 12 patents who stated a zero response wthout any justfcaton or because of low ncome were consdered as real zeros. Apart from the 3 unusable and the 8 protest WTP responses, the mean WTP for home BT (as compared to hosptal BT) among the 86 remanng patents was approxmately 60 (60.9 ) wth a standard devaton close to 55 (54.8 ). 16
Censored data should receve a partcular attenton n our emprcal analyss because, wth the 42 patents who expressed a preference for hosptal BT and were treated as real zero or negatve WTP values and the 20 patents who preferred home BT but also stated a zero value (8 protest and 12 non protest responses), they represented 45% of the whole sample (62/139). Snce only 8 protest bdders were dentfed, t was not reasonable to estmate the partcpaton equaton of a double-hurdle model,.e. to dentfy the varables nfluencng the choce of a respondent to reveal or not ts real preferences. Nevertheless, dscardng protest responses, even n lmted number, could produce based results. In ths case, and as dscussed above, the most approprate econometrc methodology s a truncated regresson model. Ths model specfes the probablty to observe the amount of WTP, condtonal on the expresson of a postve wllngness to pay. The dependent varable y was defned as the logarthm of the expressed WTP value n the model equaton: y = y f y > 0 wth y = X β + u (1) * * where y s the value of the th observaton on the dependent varable log(wtp) and * y ts correspondng latent value, descrbed by the relaton y = X β + u * wth X the ndependent varables that nfluence the WTP amount, β the unknown parameters and u the error term such as u N(0, s 2 ). The lkelhood functon can be wrtten as follows: L = + P( y = y * y > 0) = + 1 s y X β f s X β F s (2) where refers to the product over postve observatons for y and F(.) and f(.) refers to the + standard normal cumulatve dstrbuton functon and densty functon, respectvely. The unknown parameters are the soluton to the maxmzaton of the lkelhood functon. 17
4.2. Results The results of the truncated model are reported n Table II. [Insert Table II over here] Accordng to the truncated regresson model, several patent characterstcs were sgnfcantly assocated wth the WTP value for home BT (as compared to hosptal BT): ncome, experence of home care, dstance from home to hosptal, stage of dsease, martal status (lvng wth a partner) and professonal occupaton (p < 0.05), as well as health-related qualty of lfe (FACT-G) and age (p < 0.1). Intal bd was also sgnfcantly assocated wth WTP (p < 0.01). However, nether havng chldren at home, prevous experence of home BT, nor gender seemed to nfluence WTP values. Compared wth econometrc results, none of the expected nfluences of pre selected possble determnants was nvaldated (Table III). [Insert Table III over here] Consderng patents characterstcs, the WTP value ncreased wth hgher ncome, as commonly reported n the health economcs lterature on contngent valuaton. Even f the WTP value dd not seem to depend on prevous experence of BT at home, t dd depend on experence of home care other than BT. As sad before, ths s consstent wth non-economc 18
lterature fndngs accordng to whch prevous experence of home care favours further preference for home care compared to hosptal care. Regardng the possble determnants selected on the bass of the results of our plot study, we observed, as expected, that patents lvng far from the hosptal stated a hgher WTP for home BT (compared to hosptal BT) than the others. Lkewse, those wth a hgher health-related qualty of lfe stated a lower WTP for home BT, whch means that home BT was all the more apprecated, as compared to hosptal BT, as patent qualty of lfe (ncludng tredness) was low. Ths s n agreement wth the results collected n the plot study, whch dentfed sparng panful transportaton and watng tme as strong reasons for preferrng home BT. WTP for home BT was hgher at early stages of the dsease (curatve) than at advanced stages,.e. pallatve or termnal. Referrng to the safety concern expressed n the plot study by patents preferrng hosptal BT, recevng hosptal BT nstead of home BT was all the more apprecated as patents' lves were threatened. We could not hypothesze on the possble nfluences of other patents characterstcs on WTP values. They were thus ntroduced n the model as control varables. Econometrc results dd not show any nfluence of ether gender or havng chldren at home. However, lower WTP values for home BT were obtaned for patents lvng wth a partner, whch means that these apprecated hosptal BT (compared to home BT) more than the others. Ths could result from a wsh of clearly separatng daly lfe from healthcare, as stated n the plot study by patents. All other thngs beng equal, WTP for home BT was hgher for older people, and t was lower n patents wth a professonal occupaton. These fndngs are not counterntutve but could not be antcpated from what was known at the begnnng of study. Fnally, the anchorng bas hypothess was confrmed: all other thngs beng equal, the WTP 19
value for home BT ncreased when the ntal bd was hgher, whch s consstent wth results publshed n the health economcs lterature. 5. DISCUSSION/CONCLUSION Home care may be an nterestng alternatve to hosptal care, especally n chronc dseases. Accordngly, t s more and more used for cancer care, partcularly for severely ll patents, ncludng those at a pallatve or termnal stage of ther dsease (Francks et al., 2000; Emanuel, 1996 ; Zmmerman et al., 2008). Home care has been extensvely studed as a possble substtute for hosptal care, regardng ether costs or patents qualty of lfe and satsfacton (Brumley et al., 2007; Shepperd and Illfe, 2001). To our knowledge, the CV method had never been used for comparng home care to hosptal care. However, t s an attractve approach for evaluatng dfferent types of health care management that dffer only n patents well-beng durng the treatment process and not n health outcomes (Ryan and Shackley, 1995; Olsen and Smth, 2001). We consdered the emprcal stuaton of BT n cancer patents and we collected WTP values among BT users. In so dong, we chose to elct use value, and not externalty or opton values. Indeed, we wanted to know whether a WTP questonng would be acceptable to severely ll patents,.e. whether they would be wllng not only to partcpate n the survey, but also to gve non protest responses. Protest responses could result from emotonal and ethcal concerns, as well as from socal responsblty consderatons (Jorgensen et al., 2000 ; Sayman and Onçüller, 2005 ; Meyerhoff et al. 2006), all dmensons whch were encountered n our stuaton. Emotonal and ethcal aspects could be partcularly mportant for patents at pallatve or termnal stages; some of them even clamed that health s nvaluable. Socal responsblty could also be an 20
mportant concern snce a survey usng a WTP approach could call nto queston the prncple of soldarty under whch the cost of cancer care s completely covered by natonal health nsurance n France. The response rate of our study was hgh (90%) and most patents (70%) expressed a preference for home BT. Of the 94 patents who preferred home BT and who dd not express understandng problems, only 8 gave protest responses, whch corresponds to a less than 10% protest responses rate. Overall acceptablty was thus hgh among BT users. Our man objectve was to test the valdty of the CV method, namely ts ablty to elct true preferences. For so dong, we chose a two-step approach. Frst, we dentfed possble determnants of our WTP values and ther expected nfluences, on the bass of both economc and non economc lterature. We also extrapolated from the fndngs of a prevous plot study. Second, we compared them to predcted nfluences as resultng from approprate econometrc analyss. The choce of approprate econometrcs for analyzng WTP values resultng from a bddng process s essental for obtanng relevant results. One must take nto account that censored data at zero value correspond ether to real zero values or to protest responses. As a consequence, sutable modelng should (1) nvolve the whole sample (ncludng non zero responses, protest zeros and real zeros), (2) take the dfferent behavors at work nto account (protest versus non protest responses). Based on a health economcs lterature survey, we concluded that the double-hurdle model would be the most approprate approach to fulfll these two condtons. However, the estmaton of the separaton equaton requres a suffcent number of protest responses. Because we had only few protest responses, we used a truncated regresson model rather than a type II Tobt model n whch estmated parameters for zero values would not be nterpretable. 21
None of the 7 hypotheszed nfluences was nvaldated by our econometrc results. Frst, the anchorng bas hypothess was confrmed. Regardng the possble nfluence of patents characterstcs, the WTP for home BT compared to hosptal BT ncreased wth the household ncome, wth prevous experence of home care, wth lvng far from the hosptal and wth low qualty of lfe. Conversely, the WTP for home BT was lower for advanced-stage (pallatve or termnal) patents than for early-stage (curatve) patents. These results frst underlne the nterest of usng all a pror relevant sources of nformaton for selectng possble determnants of WTP values. Second, because of the good consstency between expected and predcted nfluences of patents characterstcs on WTP values, our results may have some mplcatons for polcy-makng. They suggest that, all other thngs beng equal, home care should be consdered n prorty for people lvng far from hosptal and people wth low health-related qualty of lfe. Regardng the stage of the dsease, preference for home seemed to be lower among patents recevng pallatve or termnal care than among those recevng curatve care. Ths result may seem to contradct other medcal lterature fndngs accordng to whch home care could mprove satsfacton or qualty of lfe n pallatve patents (Zmmermann et al., 2008; Shepperd and Illfe, 2001). However, we beleve that there s no contradcton between the dfferent conclusons snce these authors dd not compare patents opnons across the dfferent stages of dsease. From our study we conclude that, based on WTP values, the CV approach s acceptable to severely ll patents. Moreover, WTP values even demonstrate good valdty, gven that predcted determnants are consstent wth expected ones. 22
Ths study focuses on the nterest of analyzng WTP values for documentng valdty ssues n the CV approach, but t mght be mportant to nvestgate also the potental role of WTA. For a number of reasons, the use of WTA for decson makng s not recommended. Studes nvestgatng gven health care programs have shown that WTA s generally hgher and more scattered than WTP (Smth et al., 1999 b ; Brown, 2005; Sayman and Onçüler, 2005; Whynes and Sach, 2007). On the bass of the WTA values also collected n our CV survey, we am to nvestgate, n a further analyss, the potental role of WTA for testng CV valdty. 23
ACKNOWLEDGEMENTS Specal thanks go to Dr Yves Devaux, wthout whom ths study would not have been possble. Yves Devaux s head of the oncology home care-coordnaton unt located at the comprehensve cancer center of Lyon. We are also grateful to Valere Kante, Josette Chalencon and the home care coordnaton team for ther help n patent recrutment, to Mare-Domnque Reynaud for edtoral assstance, and to Izabela Jelovac for valuable comments. 24
REFERENCES Amemya T. 1984. Tobt models: a survey. Journal of Econometrcs 24: 3-61. Amn M, Khondoker F. 2004. A contngent valuaton study to estmate the parental wllngness-to-pay for chldhood darrhoea and gender bas among rural households n Inda. Health Research Polcy and Systems 2: 3. Bärnghausen T, Lu Y, Zhang X, Sauerborn R. 2007. Wllngness to pay for socal health nsurance among nformal sector workers n Wuhan, Chna: a contngent valuaton study. BMC Health Servces Research 7: 114. Berwck D, Wensten M. 1985 What do patents value? Wllngness to pay for ultrasound n normal pregnancy. Medcal Care 23: 881-893. Blundell R, Ham J, Meghr C. 1987. Unemployment and female labour supply. Economc Journal 97: 44-64. Brumley R, Engudanos S, Jamson P, Setz R et al. 2007. Increased Satsfacton wth Care and Lower Costs: Results of a Randomzed Tral of In-Home Pallatve Care. Journal of the Amercan Geratrcs Socety 55(7): 993-1000. Cella DF, Tulsky DS, Gray G et al. 1993. The Functonal Assessment of Cancer Therapy Scale: development and valdaton of the general measure. Journal of Clncal Oncology 11(3): 570-579. Brown TC. 2005. Loss averson wthout the endowment effect, and other explanatons for the WTA WTP dsparty. J. of Economc Behavor & Org. 57: 367 379 Dalmau-Matarrodona E. 2001. Alternatve Approaches to obtan optmal bd values n contngent valuaton studes and to model protest zeros. Estmatng the determnants of ndvdual's wllngness to pay for home care servces n day case surgery. Health Economcs 16: 101-118. 25
Dener A, O'Bren B, Gafn A. 1998. Health care contngent valuaton studes: A revew and classfcaton of the lterature. Health Economcs 7: 313-326. Donaldson C. 1999. Valung the benefts of publcly-provded health care : Does 'ablty to pay' preclude the use of'wllngness to pay. Socal Scence and Medcne 49: 551-563. Donaldson C, Jones A, Mapp T, Olson JA. 1998. Lmted dependent varables n wllngness to pay studes: applcatons n health care. Appled Economcs 30: 667-677. Donaldson C, Shackley P, Abdalla M, Medzybrodzka Z. 1995. Wllngness to pay for antenatal carrer screenng for cystc fbross. Health Economcs 4: 439-452. Drummond MF, Sculpher MJ, Torrance GW, O Bren B, Stoddart GL. 2005. Methods for the economc evaluaton of health care programmes. Thrd Edton. Oxford, Oxford Unversty press. Emanuel EJ. 1996. Cost savng at the end of lfe. What do the data show? JAMA 275:1907-1914. Flachare E, Hollard G. 2007. Startng pont bas and respondent s uncertanty. Ressource and Energy Economcs 29: 183-194. Flood L, Gräsjo U. 1998. Regresson Analyss and Tme Use Data: A Comparson of Mcroeconometrc Approaches wth Data from the Swedsh Tme Use Survey (HUS). Workng Papers n Economcs no. 5. School of Economcs and Commercal Law, Göteborg Unversty, Sweden. Franks PJ, Salsbury C, Bosanquet N, Wlknson EK et al. 2000. The level of need for pallatve care: a systematc revew of the lterature. Pallat Med 14(2): 93-104. Idr S, Caton MP, S Al H, Patte R, Brere J, Baudelot J, Courtos F. 1996. La transfuson à domcle? Une alternatve à l'hôptal de jour. Transfuson Clnque et Bologque 3: 235-239. Johannesson M, Jonsson B, Borgqust L. 1991. Wllngness to pay for anthypertensve therapy- results of a Swedsh plot study. Journal of Health Economcs 10: 461-474. 26
Johannesson M., Johansson PO, Krstrom B., Gerdtham UG. 1993. Wllngness to pay for anthypertensve therapy -further results. Journal of Health Economcs 12: 95-108. Jones AM. 1989. A double-hurdle model of cgarette consumpton. Journal of Appled Econometrcs 4: 23-39. Jones AM. 1992. A Note on Computaton of the Double Hurdle Model wth Dependence wth An Applcaton to Tobacco Expendture. Bulletn of Economc Research 44: 67-73. Jorgensen BS, Syme GJ, Bshop, BJ, Nancarrow BE. 1999. Protest Responses n Contngent Valuaton. Envronmental and Resource Economcs 14: 131-150. Jorgensen, BS, Wlson, MA, Heberlen TA. 2000. Farness n the Contngent Valuaton of Envronmental Integrated knowledge for ecologcal economcs: a database to support ecosystem servces evaluaton. Ecologcal Economcs 36: 133-148. Kahneman D, Tversky A.1979. Prospect Theory: An Analyss of Decson Under rsk. Econometrca 47: 263 291. Kurth A, Waever M, Lodhart D, Belnsk L. 2004. The Beneft of Health Insurance Coverage of Contraceptves n a Populaton-Based Sample. Amercan Journal of Publc Health 94: 1330-1332. Lndsey G. 1994. Market models, protest bds, and outlers n contngent valuaton. Journal of Water Resources Plannng and Management 120: 121-129. Madgwck KY, Yarduman A. 1999. A home blood transfuson programme for betathalassaema patents. Transfuson Medcne 9: 135-138. Mahmud M. 2006. Contngent Valuaton of Mortalty Rsk Reducton n Developng Countres: A Msson Impossble? Keele Economcs Research Papers 1: 28. Meyerhoff J, Lebe U. 2006. Protest Belefs n Contngent Valuaton: Explanng Ther Motvaton. Ecologcal Economcs 57 (3): 583-594. 27
Medzybrodzka Z, Semper J, Shackley P, Abdalla M, Donaldson C. 1995. Stepwse or couple antenatal carrer screenng for cystc fbross? Women's preferences and wllngness to pay, Journal of Medcal Genetcs 32: 282-283. Mtchell RC, Carson RT. 1989. Usng Surveys to Value Publc Goods: The Contngent Valuaton Method. Baltmore: John Hopkns Unversty Press. Natonal Oceanc and Atmospherc Admnstraton (NOAA). 1994. Natural resource damage assessments: proposed rules. Federal Regster 59: 1062-1191. O'Bren B, Vramontes JL. 1994. Wllngness to pay: a vald and relable measure of health state preference? Medcal Decson Makng 14: 289-297. O Bren B, Gafn A. 1996. When do the "dollars" make sense?. Medcal Decson Makng 16: 288-299. Olsen J.A, Smth R. 2001. Theory versus practce: A revew of wllngness to pay n health and health care. Health Economcs 10: 39-52. Ryan M, Shackley P. 1995. Assessng the benefts of health care: how far should we go? Qualty n Health Care 4: 207-213. Sayman S, Onculer A. 2005. Effects of study desgn characterstcs on the WTA-WTP dsparty: A meta analytcal framework. Journal of Economc Psychology 26: 289-312. Shepperd S, Ilffe S. 2001. Hosptal at home versus n-patent hosptal care (Cochrane Revew). In: The Cochrane Lbrary Issue 1, Chchester, UK: John Wley & Sons. Smth R, Olsen JA, Harrs A. 1999 a. A revew of methodologcal ssues n the conduct of wllngness to pay studes n health care II: admnstraton of CV survey. Centre for Heath Program Evaluaton, Monash Unversty, workng paper 85. Smth R, Olsen JA, Harrs A. 1999 b. A revew of methodologcal ssues n the conduct of wllngness to pay studes n health care III: ssues n the analyss and nterpretaton of WTP data. Centre for Heath Program Evaluaton, Monash Unversty, workng paper 86. 28
Smth RD. 2003. Constructon of the contngent valuaton market n health care: a crtcal assessment. Health Economcs 12: 609 628. Smth RD. 2006. It's Not Just What You Do, It's the Way That You Do It: The Effect of Dfferent Payment Card Formats and Survey Admnstraton on Wllngness to Pay for Health Gan. Health-Economcs 15: 281-293. Smth RD. 2007 a. The role of 'reference goods' n contngent valuaton: should we help respondents to 'construct' ther wllngness to pay? Health Economcs 16: 1319-1332. Smth RD. 2007 b. Use, opton and externalty values: are contngent valuaton studes n health care ms-specfed? Health Economcs 16: 861-869. Thompson MS. 1986. Wllngness to pay and accept rsks to cure chronc dsease. Amercan Journal of Publc Health 76: 392-396. Yasunaga H, Ide H, Imamura T, Ohe K. 2006. Wllngness to pay for health care servces n common cold, retnal detachment and myocardac nformaton: an nternet survey n Japon. BMC Health Servces Research 6: 12. Yeung RY, Smth RD, Ho LM, Johnston JM, Leung GM. 2006. Emprcal mplcatons of response acquescence n dscrete-choce contngent valuaton. Health Economcs 15:1077-1089. Zmmermann C, Rechelmann R, Krzyzanowska M, Rodn G, Tannonck I. 2008. Effectveness of specalzed care : A systematc revew. JAMA 299:1698-1709. 29
TABLES Table I. Patent characterstcs Characterstcs Mean ± SD or number of patents (%) Income (1) < 800 16 (12.2%) 800-1499 37 (28.2%) 1500-2 299 34 (25.9%) 2300-2999 22 (16.8%) 3000 22 (16.8%) Experence of home BT (yes/no) 60 (43.2%) / 79 (56.8%) Experence of home care (except home BT)(yes/no) 63 (45.3%) / 76 (54.7%) Dstance from home to hosptal (km) (mean ± SD) 34.9 ± 40.6 FACT-G (2) 61.7±13.9 Stage of dsease Curatve 70 (50.4%) Pallatve 55 (39.6%) Termnal 14 (10.1%) Lvng wth a partner (yes/no) 102 (73.4%) / 37 (26.6%) Chldren at home (yes/no) 40 (28.8%) / 99 (71.2%) Male/female 69 (49.6%) / 70 (50.4%) Age (years) 57.5 ± 12.8 Professonal occupaton (yes/no) 60 (43.2%) / 79 (56.8%) (1) Net monthly household ncome before ncome tax (8 mssng data) (2) Functonal Assessment of Cancer Care General scale (6 mssng data) 30
Table II. Results of the truncated regresson model of WTP (log) values Explanatory varables Coeffcent SE p-value Intercept 3.17630 0.514 0.000 *** Income (1= 2 300 or more) 1.02655 0.142 0.000 *** Experence of home BT (1=yes) 0.16635 0.160 0.298 Experence of home care, except BT (1=yes) 0.41905 0.124 0.001 *** Dstance to hosptal (contnuous scale) 0.00365 0.002 0.045 ** FACT-G (contnuous scale) - 0.00944 0.005 0.060 * Stage of dsease (1=curatve; 0=pallatve or termnal) 0.30030 0.128 0.019 ** Lvng wth a partner (1=yes) - 0.45585 0.162 0.005 *** Chldren at home (1=yes) 0.21794 0.133 0.102 Gender (1=male) 0.17425 0.130 0.180 Age (contnuous scale) 0.01247 0.006 0.053 * Professonal occupaton (1=yes) - 0.49574 0.146 0.001 *** Intal Bd (1=76 ; 0=38 ) 0.37818 0.125 0.003 *** σ 0.46671 0.040 0.000 *** Number of observatons 68 Log lkelhood - 44.66932 * p<0.1; ** p<0.05;*** p<0.01 31
Table III. Expected and predcted nfluences of possble determnants of WTP (log) values Explanatory varables Expected nfluence Predcted nfluence Income (1= 2 300 or more) + a + *** Experence of home BT (1=yes) + b NS Experence of home care, except BT (1=yes) + b + *** Dstance to hosptal (contnuous scale) + c + ** FACT-G (contnuous scale) - c - * Stage of dsease (1=curatve; 0=pallatve or termnal) + c + ** Lvng wth a partner (1=yes)? c - *** Chldren at home (1=yes)? c NS Gender (1=male)? NS Age (contnuous scale)? + * Professonal occupaton (1=yes)? - *** Intal Bd (1=76 ; 0=38 ) + a + *** a Accordng to health economcs lterature. b Accordng to medcal lterature. c Accordng to plot study results * p<0.1; ** p<0.05; *** p<0.01 32