modelling tenants' choices in the public rented sector: a stated

http://usj.sagepub.com

Urban Studies

DOI: 10.1080/00420980220119516 2002; 39; 665 Urban Stud

Bruce Walker, Alex Marsh, Mark Wardman and Pat Niner Preference Approach

Modelling Tenants' Choices in the Public Rented Sector: A Stated

http://usj.sagepub.com/cgi/content/abstract/39/4/665 The online version of this article can be found at:

Published by:

http://www.sagepublications.com

On behalf of: Urban Studies Journal Limited

can be found at:Urban Studies Additional services and information for

http://usj.sagepub.com/cgi/alerts Email Alerts:

http://usj.sagepub.com/subscriptions Subscriptions:

http://www.sagepub.com/journalsReprints.navReprints:

http://www.sagepub.co.uk/journalsPermissions.navPermissions:

http://usj.sagepub.com/cgi/content/refs/39/4/665 Citations

by Alireza Ehsanfar on October 29, 2008 http://usj.sagepub.comDownloaded from

http://www.gla.ac.uk/urbanstudiesjournal/

http://usj.sagepub.com/cgi/alerts

http://usj.sagepub.com/subscriptions

http://www.sagepub.com/journalsReprints.nav

http://www.sagepub.co.uk/journalsPermissions.nav

http://usj.sagepub.com/cgi/content/refs/39/4/665


Urban Studies, Vol. 39, No. 4, 665–688, 2002

Modelling Tenants’ Choices in the Public RentedSector: A Stated Preference Approach

Bruce Walker, Alex Marsh, Mark Wardman and Pat Niner

[Paper received in � nal form, February 2001]

Summary. This paper uses a stated preference (SP) approach to examine the potential housingchoices of tenants in the UK public housing sector. The paper begins by explaining the policysigni� cance of the choices that such tenants might make if alternative dwellings were offered tothem. It then discusses the SP approach in general before explaining the way in which it is usedin this study. The results of the SP modelling exercise are presented. These suggest that tenantsare unlikely to move to housing estates that they see as being worse than their current estatesolely in response to lower rents. This is because a number of factors other than rent are of moresigni� cance in their potential housing decisions.

1. The Policy Background

Approximately every 10 years since the early1970s, rent-setting principles and systems ofsubsidy to public rented housing have beenthrough major reforms in the UK. A furtherround of change and debate is currently tak-ing place. The question of how tenants re-spond to rent levels, changes anddifferentials is of central importance to thisdebate. This question is the main focus ofthis paper.

For much of their history, public landlordsin the UK operated rent-pooling policieswhich resulted in rent structures with limiteddifferentials between different types of prop-erty and between different locations. Thispolicy was implemented within a framework

of recurrent revenue subsidy that resulted inrents considerably below market levels. Thisrent-setting regime was thrown into seriousquestion in the 1970s and 1980s when issueswere raised about the distortionary effects ofprice subsidies. There was particular concernthat the level of average rents did not differsubstantially between areas of the countryand did not respond to local housing marketand labour market requirements (see, for ex-ample, Minford et al., 1987).

As is well known, legislative changes in1980 and 1989 changed the subsidy systembeing used by central government so as toeffect the shift from price subsidies to areliance on personal subsidy. In 1980, aver-

Bruce Walker is at the Centre for Urban and Regional Studies, School of Public Policy, J. G. Smith Building , University of Birmingham,Birmingham, B15 2TT, UK. Fax: 0121 414 4989. E-mail: [email protected] . Alex Marsh is in the School for Policy Studies,University of Bristol, 8 Priory Road, Bristol, BS8 1TZ, UK. Fax: 0117 954 5564. E-mail: [email protected] . Mark Wardmanis at the Institute for Transport Studies, University of Leeds, 36 University Road, Leeds, LS2 9JT, UK. Fax: 0113 233 5334. E-mail:[email protected] . Pat Niner is at the Centre for Urban and Regional Studies, School of Public Policy, J. G. Smith Building,University of Birmingham, Birmingham, B15 2TT, UK. Fax: 0121 414 4989. E-mail: [email protected] . This paper is basedon research undertaken for the Department of the Environment, Transport and the Regions (DETR). We would like to thank the DETRfor their � nancial and organisationa l support for this work and for their very helpful comments on initial drafts of this paper. However,the inferences and implication s drawn in this paper, and the opinions expressed, are those of the authors alone.

0042-0980 Print/1360-063X On-line/02/040665-24 Ó 2002 The Editors of Urban StudiesDOI: 10.1080/0042098022011951 6



BRUCE WALKER ET AL.666

age council housing rents in England were£7.70 which represented 6.9 per cent of aver-age earnings. In contrast, by 1989, they hadreached £20.70 (9.4 per cent of average earn-ings) and peaked at 13.2 per cent of averageearnings in 1996 (Wilcox, 1995, 2000). In-creases were not implemented uniformly: therevised subsidy system was geared to ensur-ing that rents rose fastest in areas wherecapital values and regional earnings werehighest, thereby leading to increasing differ-entiation in rents between regions.

Beyond concerns with regional rent differ-entials are more micro-level concerns withthe impact of pricing upon the mobility oftenants and, equally importantly, upon theef� cient utilisation of the public housingstock (for example, DoE, 1996). The argu-ment is straightforward: rent-pooling and � atrent structures give tenants little incentive tooptimise their housing consumption. Incen-tives are blunted even further by the 100 percent marginal subsidy provided to recipientsof assistance through the Housing Bene� t(HB) system. Hence, it was argued that therewas widespread underoccupation within thestock: older ‘empty-nester’ households wereseen as particularly prone to remain in largerproperties even when their current housingneeds could be met by a smaller property.The shortage of larger accommodation in thepublic sector, partly as a consequence ofsales under the Right to Buy policy, madefreeing-up larger accommodation a pressingpolicy issue. A belief in pervasive underoc-cupation has been a recurrent theme in cen-tral government thinking. Survey evidence(for example, Barelli, 1992) indicates thatthere may be a signi� cant proportion ofhouseholds with at least one ‘spare’ bed-room, although Barelli (1992) also suggeststhat most such households do not perceivethemselves as being ‘overaccommodated’.

While a number of incentive schemes havebeen put in place in an attempt to inducetenants to move to alternative accommo-dation, the � rst attempt by central govern-ment to address this issue directly throughpricing was section 162 of the Local Govern-ment and Housing Act 1989. The section

enjoined local authorities, when reviewingrents

to have regard in particular to the principlethat the rents of houses of any class ordescription should bear broadly the sameproportion to private sector rents as therents of housing of any other class ordescription (1989 Local Government andHousing Act, section 162, sub-section (3)).

The objective of this section is to break withthe principle of rent-pooling and to attemptto ‘mirror’ market differentials. By impli-cation, this would present tenants with theappropriate incentives to optimise their hous-ing consumption. Hence it represented a po-tentially important shift from the traditionalapproach of allocating public housing on thebasis of ‘objective’ need to a greater relianceon ability and willingness to pay. In practice,the policy was widely ignored (see Walkerand Marsh, 1995, 2000) and it confronted anumber of dif� culties at a conceptual level(Walker and Marsh, 1998).

While this concern with the ef� cient use ofthe housing stock persists, debate over rentpolicy has evolved. The impact of � at rentstructures has become a more urgent ques-tion, given that certain parts of the socialhousing stock are experiencing high turnoverand/or a high rate of voids. These and relatedphenomena are encompassed within the de-bate over ‘low’ or ‘changing’ demand forsocial housing (see, for example, Bramley etal., 2000; PAT7, 1999). The link to rentpolicy is the belief that if rents are undiffer-entiated spatially then tenants will have littleincentive to trade price off against quality,resulting in fewer tenants choosing to live inpoorer-quality, lower-priced neighbourhoods .The problems of poorer neighbourhoods arereinforced in some areas where the costs ofprivate housing are comparable—and, insome cases, favourable—to those in socialrented housing. Similar problems are beingexperienced across the two social housingsectors—local authority and RegisteredSocial Landlords—because of ‘incoherent’rent differentials within and between the twosectors (see, for example, Wilcox, 1997).



MODELLING TENANTS’ CHOICES 667

The recent housing Green Paper (DETR/DSS, 2000a) explores the options for im-posing a standard rent-setting method onsocial housing. This part of the currentgovernment agenda re� ects a desire to“provide a closer link between rents and thequalities which tenants value in properties”,“give tenants the opportunity to take moreresponsibility for their choice of housing”,and “encourage better management by sociallandlords of their stock” (DETR/DSS, 2000a,para. 10.1). The Green Paper sets out plansfor choice-based allocation systems and animportant goal of rent policy reform is thatrent structures should complement these newallocation systems. The statement of policythat emerged following consultation reassertsthe broad thrust of the Green Paper agenda(DETR/DSS, 2000b, ch. 9).

Marsh (2001) argues that underlying theGreen Paper discussion of rents are two in-terrelated, but separate, themes. On the onehand, there is a concern with developing arent-setting policy that conforms to a particu-lar de� nition of fairness. On the other, thereis a relatively simple conception of decision-making and a concern with tenants’ be-havioural responses to price signals. If rentsare more clearly differentiated by quality,then it would be expected that some tenantswould be willing to trade down to poorer-quality accommodation in order to realise asaving in weekly housing costs. This in turnwould shore up areas which may be at risk of‘low’ demand or even abandonment. Thegovernment’s concerns with the use of� nancial incentives to in� uence choices anddeal with underoccupation are similarly pre-sent in the underoccupation pilot schemerunning from April 2000 (DETR/DSS,2000a, para. 11.76). The scheme gives ten-ants receiving housing bene� t a lump sumequivalent to half of the HB saving thatresults from their moving to smaller accom-modation.

The issue of rent-setting and appropriaterent policies is central to the current policyagenda in England. Whether increasing rentdifferentials will have the impacts thatgovernment anticipates hinges on whether,

and how, tenants respond to rents and rentdifferentials. Addressing this question is fun-damental to assessing the scope for usingpricing mechanisms to achieve the govern-ment’s broader policy goals. In the remainderof this paper, we report on research thataddresses the question directly.

The research adopts a novel approach tothe issue by employing a stated preferenceapproach to modelling decision-making bylocal authority tenants. In section 2, we pro-vide an overview of stated preference mod-elling and the situations in which it has beenused. The methodology used to apply thestated preference approach to tenants’ hous-ing choices is summarised in section 3. Sec-tion 4 presents the principles of statedpreference modelling and details of themodel used in our research. Key results ofour modelling exercises are also presented.The results are interpreted in section 5 andsection 6 seeks to illustrate their application.Section 7 concludes by relating these resultsback to the policy agenda.

2. The Stated Preference Approach

2.1 An Overview of the Stated PreferenceApproach

Stated preference (SP) involves the simu-lation of choice situations. Those taking partin an SP exercise are presented with hypo-thetical alternatives and make choices be-tween them in much the same way as theymight do when making choices in market orother contexts. SP is a decompositional ap-proach which recognises, following Lan-caster (1971), that the overall attractivenessof a product is related to a set of relevantattributes (characteristics) and that changesto these attributes impact on the preferencestowards and willingness to pay for the prod-uct.

SP has a number of key features. Thedecision-maker taking part in an SP exer-cise—who can be an individual, household,group or organisation—is offered hypotheti-cal alternatives to evaluate. The evaluationmight take the form of a choice between the




alternatives or entail ranking or rating themon either an interval or categorical scale.More conventional revealed preference (RP)methods, based on the results of actualchoices, typically obtain one observation perrespondent. In contrast, SP participants arerequired to make a series of choices or eval-uations, typically between 8 and 16.

Each alternative presented to the decision-maker is characterised by a set of relevantattributes, with the number of attributes usu-ally limited to between three and � ve inrecognition that human cognitive abilities arelimited. The levels or values associated witheach attribute must be realistic and are usuallycombined according to the rules of an exper-imental design procedure, with orthogonalitybeing by far the most commonly used (Hen-sher and Barnard, 1990). Trade-offs betweenattribute levels must be offered. Statisticalanalysis allows the relative importance of theattributes to be estimated. These estimates canthen be used for purposes such as optimisingproduct design features or social cost–bene� tanalysis and to forecast behaviour.

The attractions of the SP approach stemfrom its ability to control the choice stimuli,thereby providing an approximation to labo-ratory conditions. The SP approach also hasa number of advantages over RP methods. Itcan avoid correlation problems, ensuresuf� cient variation in the data, offer bettertrade-offs between variables than often existin the real world, collect multiple choices perperson and avoid measurement error in theindependent variables, all of which will im-prove the precision of the parameter esti-mates. SP methods can also be used toanalyse markets, products or policy optionsthat do not currently exist in the real world orattribute levels beyond the range of currentexperience. SP scenarios can be created inorder to ‘design out’ those variables whichare problematic to represent or model and inwhich we are not primarily interested byspecifying them to be the same for eachalternative so that they cannot in� uencechoice. Such variables would have to beincluded in an RP model based on actualdecisions.

2.2 Some Issues in the Use of StatedPreference

The main drawback of the SP method, as withall approaches based on hypothetical ques-tions, is that the responses do not necessarilyprovide an accurate guide to actual prefer-ences. In other words, in contrast to RP datawhere respondents have actually made achoice, SP respondents are not committed tobehave in accordance with their stated prefer-ences. However, in circumstances where thealternatives offered to consumers are newand/or consumers have little experience mak-ing choices between such alternatives, RPdata may not be available. Similarly, suitableRP data may not be available where con-sumers are broadly familiar with the choiceson offer and have experience of choosing, butthe alternatives of interest to the researchdiffer signi� cantly from those consumers willpreviously have encountered. Where in-suf� cient RP data are available, or are verydif� cult to collect, there may be no practicalalternative to an SP approach in estimatingthe effects on consumers’ choices of (changesin) key attributes.

The errors that SP responses contain maybe random, such as might stem from respon-dents’ uncertainty, misunderstanding or fa-tigue. We would not expect random error tohave an adverse impact on the estimatedrelative importance of each attribute.1 Ofmore concern is error which is systematic andwhich may result in biased estimates. Thereare � ve principal forms of such responseerror. First af� rmation bias occurs where theSP responses are amended to af� rm the per-ceived study objectives or to provide answersthat the respondent expects the person con-ducting the SP exercise to want. Secondly,justi� cation bias arises where responses areamended to justify actual behaviour in orderto reduce ‘cognitive dissonance’. Thirdly, er-rors may arise from the effects of breaking ofhabits resulting from the fact that real-worlddecision-making involves far greater trans-action costs than decisions made under SPexercises. Fourthly, strategic or policy re-sponses bias can occur where respondents




attempt, through their answers, to in� uencethe likelihood or magnitude of changes in thereal world. Finally, social norm bias mayarise if SP responses fail to re� ect actualpreferences because actual preferences areperceived as socially unacceptable or ‘politi-cally incorrect’.

In addition, errors might arise from theinef� cient or inappropriate design of the SPexercise itself. If the variations in attributelevels offered are too small or unreasonablylarge, the levels of the attributes are pre-sented in unrealistic combinations or the ab-solute values are unrealistic, then they maybe ignored by respondents. In such cases, theimportance of the attributes in question willbe understated. Additionally, an SP exercisecould emphasise secondary variables whichwould have little impact on respondents’real-world choices, leading to an in� ated im-portance being ascribed to these attributes.

Some of these risks of error can be miti-gated by the appropriate design and conductof the SP exercise. First, the alternativesoffered should be of suf� cient signi� cance orimportance to participants for them to takethe exercise seriously. This might beachieved through basing sample selection onconsumers who actually or potentially facechoices in the relevant or associated markets.Secondly, the attributes of the alternativesoffered should both be understandable, andsuf� ciently realistic, to the participants.Thirdly, to reduce the likelihood of strategicor af� rmation bias, it is important that the SPexercise gives no special emphasis to the‘key’ attributes with which the research isconcerned. This can be facilitated by varyingthe levels of those attributes with which theresearcher is less concerned.

Clearly, how well SP performs will vary.For example, it has long been widely recog-nised (for example, Bates, 1988) that theperformance of SP in valuation, requiring ananalysis of coef� cient estimates relative toeach other, may be different from that inforecasting, where the absolute values of theestimated coef� cients are employed. Thequality of the responses obtained dependsupon, amongst other things, the questions

asked (Widlert, 1994) and the incentive tobias (Wardman, 1998; Wardman and Whe-lan, 2001). Good design can mitigate manyof these problems.

It is important, particularly for novel appli-cations of SP, to obtain RP data where poss-ible from a closely corresponding choicecontext. The performance of SP can then beassessed by examining whether RP data cor-roborate the evidence drawn from the SPexercise. Our attempts to do this are de-scribed later in this paper.

2.3 Using Stated Preference Analysis in theLocal Authority Sector

The research reported here was primarilyconcerned with assessing the in� uence ofrent levels and rent differentials on tenants’choices of dwellings. In particular, wewished to examine the impact of relative rentchanges on choices between the tenants’ cur-rent dwelling and those on better and worseestates. Previous housing policy research inthe social housing sector suggested threemain reasons why RP data alone, were theyavailable, would not provide an adequatebasis for estimating the signi� cance of thesefactors. Using an SP approach allows eachissue to be addressed.

First, previous moving behaviour withinlocal authority housing has been constrainedby allocation and transfer policies and bystock availability. As a result, tenants maynot have been able to reveal their preferencesthrough relocating. Secondly, rent differen-tials between dwellings within local authorityhousing are, typically, relatively narrow.Walker and Marsh (2000), for example, esti-mate that rent differentials widened some-what during the mid to late 1990s, but thatthe rent of a one-bedroomed dwelling wasstill 78 per cent of that of a three-bedroomdwelling in 1998/99. Such comparatively � atrent structures reduce the � nancial incentivefor tenants to adjust their housing consump-tion by relocation. It is therefore unsurprisingthat previous research (for example, Walkerand Marsh, 1995) suggests that rent levelsand differentials are not currently a major




factor in most tenants’ decisions to movebetween council properties. Indeed, qualita-tive work with tenants has indicated that rentlevels and differentials are not an issue thattenants tend to raise spontaneously, either asa particular concern or as something withwhich they were particularly satis� ed(Walker et al., 2000, Figure 1.1). If currentrent structures provide insuf� cient incentiveto move, then any existing RP data willinadequately capture the effects of higherrents and wider differentials were they to beintroduced in the future.

Thirdly, even if tenants had the oppor-tunity to choose their council housing in thepresence of more sharply differentiated rents,the majority would not face any increase intheir rent or would continue to pay no rent atall. Over two-thirds of tenants receive HB.Such tenants, in principle at least, do nothave to consider rent levels when makinghousing decisions, although there is someevidence that HB recipients are more sensi-tive to rents than might have been expected(see, for example, Walker and Marsh, 1995and 1998 for a discussion). To the extent thatreceipt of HB affects mobility at present,available RP data will give no guide to theeffects on location decisions that would arisefrom any reform of HB which made recipi-ents contribute (more) towards their rentsand to rent increases. Such reforms wereunder consideration at the time of this re-search, although the Green Paper subse-quently did not propose any material changesto the HB system (DETR/DSS, 2000a, ch.11).

2.4 The Use of Stated Preference Analysis inPrevious Work

SP has its origins in mathematical psy-chology and statistics, with Luce and Tukey(1964) often cited as the seminal work. Manymarket researchers and other practitionershave made strong claims for its validity intheir � eld (Green and Srinivasen, 1978; Lou-viere, 1988; Louviere et al., 2000), where itis often termed ‘conjoint analysis’. Accord-ing to Cattin and Wittink (1982), there had

been 1000 commercial applications in the USin the 1970s, whilst Wittink and Cattin(1989) estimate that there were about 400commercial applications per year in the early1980s. Carroll and Green (1995) provide amore recent review in this area.

Transport researchers were also among the� rst to employ the technique (Sheldon andSteer, 1982; Bates, 1983), particularly giventhe signi� cant developments in disaggregatemodelling which had occurred in this area.There have subsequently been a considerablenumber of transport applications (for exam-ple, Ortuzar, 2000; Louviere et al., 2000), aswell as a number of methodological develop-ments in this � eld. Similar developmentswere also apparent in geography (for exam-ple, Timmermans, 1984; Louviere and Tim-mermans, 1990). The technique has provedto be particularly useful where there is noactual market place in which preferences canbe expressed. Thus, there have been notableareas of practical SP application in healtheconomics (McClain and Rao, 1974; Windand Spitz, 1976; Ryan, 1999; Ryan andWordsworth, 2000), welfare economics(Donnelly et al., 1976; Hoinville, 1971) andenvironmental economics (Adamowicz et al.,1994; Boxall et al., 1996).

In relation to housing market choices, thesigni� cance of housing costs within thehousehold budget led to some early market-ing research in the � eld (Fiedler, 1972).Knight and Menchik (1976) provided anearly SP contribution to the residentialchoice literature, while more recent SP appli-cations have covered a wide range of issuesrelating to housing and locational choices(for example, Benjamin and Paaswell, 1981;Veldhuisen and Timmermans, 1984; Josephet al., 1989; van de Vyvere, 1994; Timmer-mans et al., 1996; Molin et al., 1997;Schellekens and Timmermans, 1997; Ward-man et al., 1998; Cooper et al., 2001).

However, the use of SP in a housingcontext to date appears to be exclusivelyconcerned with private-sector decision-making. There appear to be no publishedstudies using SP to model tenants’ choices inthe social rented sector. The absence of SP




work in this area means that the research wediscuss below represents an innovative ap-proach to the topic. It should therefore beseen as a pilot of the SP approach in theanalysis of such housing choices, particularlysince it examines both willingness to pay forspeci� c housing attributes and choice behav-iour in response to price (rent) changes. Mostresidential choice studies examine one or theother.

3. Applying Stated Preference Analysis toTenants’ Housing Choices

3.1 Selecting the Samples

The research, undertaken in 1999–2000, wasbased on two different samples, drawn usingdifferent sampling approaches. The � rst sam-ple was drawn to identify participants in theSP exercise. The second sample comprisedhouseholds who had actually moved withinthe past two years who would provide RPinformation. Where SP respondents had relo-cated in the past two years, they were askedalso to provide RP information.

The sample size for the SP exercise wasset at 400 in total, located in 3 local authorityareas in England. The areas were selected asbroadly illustrating the known geographicaldifferences in demand for social housing.Since these differences, and the associatedproblems, are particularly evident in largeurban areas, 3 heavily urbanised local author-ities were chosen, one in the North, one inthe West Midlands and one in London. Itfollows that generalising our � ndings to lessurbanised areas, and especially to rural areas,should be treated with caution.

Within these authorities, the SP samplewas further limited to a few selected estates.This allowed the SP scenarios to incorporatenamed nearby estates. Further, the estatesselected, in association with the local auth-ority, were of medium popularity on thegrounds that tenants could then choose totrade up or down from their current dwellingin response to cost changes. Selecting verypopular or very unpopular estates wouldhave reduced the plausible options that couldbe offered in the scenarios.

The 400 SP interviews were distributed asfollows: 135 in the West Midlands authority(hereafter, W. Midlands); 137 in the North-ern authority (North); and 128 in a LondonBorough (London). The SP sample � nallydrawn proved to be somewhat younger thanthe pro� le of English council tenants as awhole and more likely to be living in � ats orlarger properties with 3 or more bedrooms.However, 70 per cent of the sample werereceiving HB, broadly similar to the popu-lation of council tenants.

The RP sample was designed to include ashigh a proportion as possible of people whohad made or considered a housing choice ofthe type being offered in the SP experimentswithin the past two years and had been acouncil tenant at that time. To increase theprobability that RP respondents would besimilar in these respects to the SP partici-pants, the samples were drawn from the samethree local authority areas. Brie� y, the sam-pling frame for the RP attempted to identifythose who had moved within the local auth-ority sector and from that sector to othersectors, including through Right to Buy(RTB) and private purchase, using localauthority records.

The achieved RP sample comprised 468responses. It provided some useful infor-mation on relocation decisions (see Walkeret al., 2000, ch. 2). However, the RP respon-dents in general proved unable to providesuf� cient detailed information on the costs oftheir previous dwelling to enable us to modeltheir choices in the same way as the SP data.This prevented us from checking the accu-racy of our SP model against an RP model.As explained in the conclusion, this may notbe as major a drawback as it might havebeen, given the results of our SP analysis, butit is an admitted weakness in the interpret-ation of the results of the SP modelling exer-cise.

3.2 The Stated Preference Exercises

The SP exercises in this research presentedrespondents with a series of scenarios, eachscenario describing a number of alternative




dwellings with different combinations ofattribute levels. Respondents were asked tochoose the preferred alternative. The exer-cises were designed to allow any sensitivityto rent levels revealed by participants’choices to be isolated from the in� uence ofchanges in other variables, while minimisingbias from all sources. The key dimensions ofthe exercises were the attributes included inthe scenarios, the levels of each attribute andthe way in which choices were presented.Also fundamental to the research was theway in which rent changes were handledwhen respondents received HB. We considereach of these in turn.

The attributes included. The attributes in-cluded in the exercises were selected becausethey were hypothesised to be important inhousing decisions and because they could besimply described and varied in the scenarios.To keep the choices offered to respondents asclear and simple as possible, it was decidedthat no more than � ve attributes should beincluded in any one of the scenarios to bepresented. Inevitably, this process involved agross simpli� cation of the factors likely to beconsidered by tenants when making choicesin the real world, although we know of nocompelling evidence to indicate what the‘right’ number of such variables should be.

After piloting, the attributes offered wereselected from

—rent—the local authority estate on which the

property being considered was located;—the type of property;—the number of bedrooms;—the condition of the property;—the amenities of the estate;—the opportunity to change tenure.

The levels of the attributes. The variablesused in the exercise took the following val-ues or levels.

Rent. The changes in rent levels offered inthe scenarios were limited to the range 1 £10to 2 £10 relative to the weekly rent on the

current dwelling. This was thought to be areasonable and realistic range of rent changesin the context of current and recent rentpolicy. For example, a £10 per week rentincrease for the modal group of tenants—whose rents lay between £150 and £199 permonth—represents a 20–27 per cent increasein the rent charged.

The Local Authority Estate. Participantswere offered choices between a worse estate,the current estate and a better estate. Asexplained above, respondents were selectedfrom estates adjudged by the local authorityto be of medium quality or popularity. Theworse and better estates that appeared in theexercise were classi� ed according to thelocal authority’s assessment and the actualestates named in the scenarios were chosenby the participant as being the most familiarto them. Preceding the exercise, these re-spondents were asked to rate each estate interms of quality, environment, location,neighbours and security. Their assessmentclosely matched the local authority’sclassi� cation of better and worse estates.

The type of property. The dwelling cate-gories used were bedsit, maisonette, houseand � at.

The number of bedrooms. The values setwere one less bedroom, same number ofbedrooms, one more bedroom and two morebedrooms, compared with their current prop-erty.

The condition of the property. The condi-tions of the properties offered were describedas being in a poor, average, good or verygood state of repair.

The amenities of the estate. Participantswere asked to indicate whether security on anestate or the range of shops available wasmore important to them. If security was moreimportant, the amenity levels offered were noextra security measures, CCTV, or CCTVand approved neighbourhood patrols. Ifshops were chosen as more important, the




levels were set as no shops available within10 minutes walk, some shops and a goodrange of shops within that distance. In prac-tice, the great majority of participants chosesecurity as the more important amenity.

The opportunity to change tenure. In additionto local authority dwellings on the current oranother estate, participants were also offeredthe opportunity to change tenure. Participantscould choose to rent from a housing associ-ation, rent privately, buy their current homeunder RTB or buy in the private market. Thecosts of these options were set accordingto information collected in the local authori-ties about local rent and price levels.The cost of owner-occupation was based onassumed repayments on a 90 per cent mort-gage on an ‘average’ or indicative dwellingprice in the area. This understates the fullcosts of home-ownership by ignoring thecapital cost of the deposit and repair andmaintenance costs. The cost for RTB wasarbitrarily set at 66 per cent of the full own-ership level to allow for RTB discount.These costs did not vary for each respondent.To allow more variation in the costs of RTB,about half the participants, chosen at random,were presented with the full RTB costs andhalf with 80 per cent of that value.

Allowing for Housing Bene� t. HB was dealtwith by asking respondents in receipt of partor full bene� t to assume that they wouldhave to pay the increase in rent if they chosean alternative with a higher rent than the rentset for their current dwelling. Thus, if therent of the alternative was £10 per weekhigher than the rent set for their currentdwelling they were asked to assume that theywould have to pay £10 per week, but that HBwould continue to cover the remainder of therent. Respondents currently not receiving HBwere assumed not to be eligible for (partial)HB receipt after an increase in rent.

If the rent of the chosen alternative waslower than the rent set for their currentdwelling, participants receiving full HB wereasked to assume that they would receive thedifference in cash. Hence, if the rent of the

alternative chosen by them was £10 per weekless than that set for their current dwelling,they were asked to assume that they wouldreceive £10 per week. Participants on partHB were asked to assume that any decreasein rent would reduce the amount that theycurrently paid themselves with their HB enti-tlement remaining at the same level. Thus,part HB claimants were asked to assume thatthey would retain (rather than receive) anyrent savings.

These assumptions do not re� ect the wayin which the current HB system operatessince a rent reduction (increase) does notresult in any � nancial bene� t (cost) to HBrecipients. However, any future reform ofHB is likely to involve greater contributionsfrom recipients towards their rent. Our ap-proach here can be seen as a partial simu-lation of likely housing choices of recipientsunder such a reform. More importantly formodelling the in� uence of rents on choices,it was important that some cost be attachedto choices of better accommodation and that� nancial bene� ts should be received as aresult of choosing worse accommodation.Even on the basis of a very weak assumptionof consumer rationality, it would be expectedthat if better choices were free of charge toHB recipients and worse choices made nodifference to their � nancial position, theywould always choose the former. In thatsituation, there is no need for them to tradeoff an improvement or decline in their hous-ing consumption against higher or lowercosts. Consequently, the assumptions thatHB recipients were asked to make were in-tended to encourage them to consider‘trades’ in the same way as non-recipients.

Presenting choices in the stated preferenceexperiments. The SP experiment was con-ducted by showing participants scenarioscontaining a combination of those attributesand values discussed above drawn at randomfrom a set of scenarios consisting of a widerange of such combinations. The scenarioswere constructed in order to eliminate corre-lation between attributes. The participantchose their preferred alternative or indicated




that they could not make a choice. The exer-cise was then repeated with another ran-domly drawn scenario.

In order to constrain the complexity ofeach choice decision, the SP experiment wasdivided into two parts. In the � rst exercise(SP1), respondents were shown pairedchoices between their own dwelling andanother house on the same estate, describedas ‘close to where you live now’. Thiswas designed to reduce inertia due to anunwillingness to change estates while gettingrespondents to consider property size, con-dition and rent. Each participant was shown9 scenarios drawn at random from a set of 64possible scenarios.

The second SP exercise (SP2), also under-taken by those participating in the � rst exer-cise, offered a choice between the currentestate, better estate and worse estate withrents varying on each property offered. Thisexercise, again offering 9 scenarios from aset of 64, used fewer non-rent variables thanin the somewhat simpler � rst exercise.

3.3 The Results of the Stated PreferenceExercises

The results of the two exercises were com-bined to give 6297 observations of choices,each observation taking the form of ‘wouldchoose the alternative’ or ‘would not choosethe alternative’. This total falls short of maxi-mum number of choices that could in prin-ciple have been achieved (400 3 18scenarios 5 7200 choices) by about 14 percent. This was for three reasons.

First, some participants withdrew from theexercise before they had considered all 18scenarios, although this happened rarely.Secondly, in about 4 per cent of the exer-cises, the participants indicated that theywere unable to make a choice, even given thechoice of staying where they currently were,and these non-choices were excluded fromdata. Thirdly, it transpired that a notablefeature of the choices made by participantswas that only in around 10 per cent of casesdid they involve the choice of the non-local-authority alternative. This is perhaps not sur-

prising. Choosing to buy or to rent from adifferent landlord is arguably a much moresigni� cant change than choosing anotherdwelling in the same sector with the samelandlord since it involves changes in propertyrights and security of tenure and, if buying,incurring long-term debt. Further, the infor-mation constraints on the SP exercise meantthat the information presented about the non-local-authority alternatives was not as rich asthat desired for any actual choice. It is likelytherefore that the exercise underestimates theextent to which participants might makechoices of dwellings in the other sectors.However, given that one of our main re-search objectives was to explore intratenurechoice, we decided to exclude choices in-volving tenure change from the set to bemodelled.

4. Modelling Tenants’ Stated Preferences

4.1 The Principles of Stated PreferenceModelling

This study deals with a discrete dependentvariable indicating the choice made betweendifferent housing alternatives. Discrete de-pendent variables of this type are most com-monly analysed using discrete choice modelswhich analyse variations in choices acrossparticipants as a function of the characteris-tics of participants and of the alternativesbetween which choices are being made.

We can conceive of the alternatives be-tween which the household is makingchoices as yielding different levels of satis-faction or utility to the household. The totalutility of an alternative is made up of utilitiesassociated with each of the relevant attributesand the household is assumed to maximise itsutility by choosing the alternative with thehighest total utility. In making such a choice,the household is assumed to make trade-offsbetween characteristics.

The utility function for alternative i andhousehold n (Uin) can be presented in generalterms as

Uin 5 f(Xin, Sn, An) (1) by Alireza Ehsanfar on October 29, 2008 http://usj.sagepub.comDownloaded from



where X denotes, in our case, housing at-tributes; S re� ects socioeconomic factors;and A denotes any underlying attitudesin� uencing the utility of the alternative to thehousehold. In attempting to maximise utility,the household will choose alternative 1 if U1

is greater than Ui for all i.To examine some of the implications of

the approach, suppose for simplicity thateach housing alternative can be describedsolely in terms of a housing quality index (Q)and rent (R). The linear-additive form of theutility function, as typically used, in this casecan be represented as

Ui 5 a QQ 1 a RR (2)

To allow for the unobserved aspects of utilityarising from, for example, omitted variablesand, where SP data are concerned, errors thatarise because households are not committedto behave in accordance with their statedpreferences, we enter an error term for eachalternative (ei). Letting Vi 5 a QQ 1 a RR, thisgives

Ui 5 Vi 1 ei 5 a QQi 1 a RRi 1 ei (3)

Vi is known as ‘representative’ or ‘determin-istic utility’; whilst ei, capturing the net effectof omitted variables, is ‘stochastic utility’.The latter is entirely analogous to thespeci� cation of an error term in multipleregression to account for the net effect ofunobserved factors.

We can observe Qi and Ri and thereforeobtain some estimate of Vi. Since ei is unob-servable, the analysis of housing choice be-haviour must proceed on the basis of Vi

alone. However, we cannot be certain that,say, alternative 1 is preferred if V1 is highestsince the error may in� uence the outcome.We would, however, expect the probabilitythat a household chooses alternative 1 toincrease as V1 increases. The probability thatan individual chooses alternative 1 from theset of alternatives available can be repre-sented as

P1 5 Pr[(V1 1 e1) . (Vi 1 ei)] (4)

where, for all i, i Þ 1.By assuming some probability distribution

for the ei, the probability of a householdchoosing alternative 1 can be speci� ed solelyas a function of the component of utility thatcan be estimated (Vi). If we assume that theerrors associated with each alternative have atype I extreme value (Weibull) distribution,have the same variances and are uncorre-lated, then equation (4) yields the multi-nomial logit (MNL) model which we haveused in this study. The probability of choos-ing an alternative is then a function of thedeterministic utilities of each alternative, thatis

P1 5eV1

Si

eVi(5)

In turn, the deterministic utilities are a func-tion of relevant observable variables, whichin the case of our general utility function inequation (1), will be

Vi 5 f( X a Xi, X b Sn, X d An) (6)

where,

X 5p

Ï 6 r ei

The coef� cients of Vi are estimated by maxi-mum likelihood methods. Note that a scalingfactor ( X ) is common to all the parameterestimates. It allows for the effect of theunobserved in� uences on choice, where r ei isthe standard deviation of the errors. All thecoef� cients are scaled in units of residualdeviation. This is unique to choice modelsand has no parallel in regression-based ap-proaches. As the random error underlyingchoices increases, then the coef� cients falland the probabilities would tend to the equalshare. As the random error underlyingchoices falls, then the coef� cients increaseand the probabilities tend to the extremes.

4.2 The Model Used

The illustrative utility function below con-tains variables which are speci� c to thedwelling or to the household and outlines thegeneral principle involved in handling cate-gorical variables, such as housing type orratings of estate features, which are common




in this study. We consider two dwelling-speci� c attributes, rent level and the rating ofthe estate’s environment, along with a house-hold-speci� c effect relating to the extent towhich the household is satis� ed with its cur-rent property and a modi� er in terms of thearea in which the household lives. Let ustherefore specify a utility function for thecurrent dwelling (Uc) as

Uc 5 a R 1 b 2E2 1 b 3E3 1 b 4E4 1 c 2S2

1 c 3S3 1 a 2A2R 1 a 3A3R (8)

It is the purpose of model calibration todetermine the set of parameters ( a , b , c )which provides the best explanation of theSP responses given the levels of the explana-tory variables. In our example, R denotes therent level, the Es denote different ratings ofthe estate’s environment, the Ss denote dif-ferent levels of satisfaction of the currenthousehold and the As denote the differentareas. The latter three variables are categori-cal and hence we use dummy variables torepresent their effects.

Rent is a continuous variable and enterssimply as R. If the current estate is rated asvery good, good, adequate or poor, we canspecify three dummy variables, for say‘good’ (E2), ‘adequate’ (E3) and ‘poor’ (E4).Their coef� cients indicate the effect of eachof these ratings relative to the omitted en-vironmental category of ‘very good’. An en-tirely analogous procedure is adopted torepresent satisfaction levels, which are at-tributes speci� c to the household rather thanvariables relating to the alternative offered.In the utility function speci� ed above, threelevels of satisfaction require the two dummyvariables speci� ed (S2 and S3).

A similar procedure is adopted to handlethe modi� er effects, but now we allow thedummy variables which are household-speci� c to interact with the dwelling-speci� cattributes. In the example above, we are al-lowing households’ responses to rent levelsto depend on area. With three areas, twodummy variables are speci� ed (A2 and A3).These are each multiplied by the rent level toproduce interaction terms, the coef� cients ofwhich indicate the extent to which the sensi-

tivity to rent changes varies across areas. Forthe omitted category (area 1), the sensitivityto rent changes is simply represented by a .For area 2, the sensitivity to rent changes isa 1 a 2, and for area 3 it is a 1 a 3. Althoughseparate rent coef� cients by area can be ob-tained by estimating separate models foreach area, allowing the coef� cients to varywithin the model is statistically far moreef� cient. In contrast to estimating separatemodels, it allows only the coef� cients ofinterest to us (rent), rather than all themodel’s coef� cients, to vary across areas.

It is also necessary to include variablesthat are speci� c to the alternatives offeredand which will in� uence choice but whichwere not explicitly varied in the SP exer-cises. Given that the second SP exercise in-volved choices between properties ondifferent estates, factors speci� c to the es-tates must be included. In order to allow this,ratings of each estate were obtained fromparticipants in that experiment to indicatetheir perceptions of the overall condition andenvironment of the estate, the general con-dition of the properties, the local facilities,residents and neighbours and the conve-nience of the location for work and visiting.The ratings were located on a � ve-point scaleof very good, good, adequate, poor and verypoor. Although the property type was variedalso, with the best estate varying the propertybetween house and � at, a wider range ofproperty-type variables must be includedsince the property type of the current dwell-ing varies across households and will poten-tially in� uence choices. The differentproperty types are also represented bydummy variables.

4.3 Allowing for the Effects of Income

We would expect those with higher incomesto be less sensitive to rent variations of agiven absolute size on the grounds that themarginal utility of money is expected to dim-inish as income increases. One way of allow-ing for this is to segment the rent coef� cientsaccording to income-group, using a dummyvariable approach, and the proportion of the




full rent actually paid. This would involvedifferent rent variables for n 2 1 of n in-come-groups, the coef� cients of whichwould show the incremental effect on rentcoef� cients of being in a particular income-group. We would expect the implied rentcoef� cients to fall (be less negative) as in-come increases.

In this study, we have used a different andmore straightforward approach which in-volves dividing the rent variables by house-hold income. If for each household we enterinto the model rent divided by that house-hold’s income (Y), the sensitivity to rent willdepend on income. Ignoring terms other thanrent for the moment, the utility function forany alternative i used in our study took theform

Ui 5 a 1Ri

Y k (9)

The effect of rent on utility is now a 1/Y k

rather than simply a 1. The use of Y k ratherthan Y offers greater � exibility because itavoids assuming that a relationship existsbetween the sensitivity to rent changes andincome, and that any relationship is pro-portional. Neither assumption may be empir-

ically justi� ed. The value of the additionalparameter k is estimated by the model usingan iterative procedure.

A value for k between 0 and 1 willdampen the income effect. If, for example, kis 0.5 and a 1 is 2 50, the rent coef� cient fora household with a £10 000 annual incomewould be 2 0.5, falling to 2 0.35 for thosewith a £20 000 annual income. Thus, in thiscase, a doubling of income would lead to a30 per cent reduction in the sensitivity to rentchanges. In the limit, a value of zero for kwould mean that the sensitivity to rent varia-tions is independent of the level of income,whilst a value of 1 indicates that the relation-ship is proportional.

4.4 Results

Table 1 summarises the attributes (variables)from the SP exercises that were available forinclusion in the model.

Recall that the MNL model reported hereis based on 6297 observations relating tointralocal authority choices only. Table 2shows the attributes that were included in the� nal model yielding the best � t. The esti-mated model contains a large number of

Table 1. Explanatory variables (attributes) considered

Variables SP1 exercise SP2 exercise

Current and alternative dwelling attributesBedrooms Property typeCondition AmenitiesRent Rent

Overall environmentProperty conditionFacilitiesNeighboursLocation

Household attributesSatisfaction SatisfactionRetired Household Retired HouseholdChildren ChildrenLength of residency Length of residencyOn transfer list On transfer list

Other factorsIncome IncomeHousing Bene� t status Housing Bene� t statusLocal authority area Local authority area




Table 2. Attributes included in the reported model

Attribute name Attribute

RentC Rent on current estateRentW 1 Rent on worse estate (non-negative)RentB Rent on Better EstateRentO Rent of other dwelling, current estateA1 Area 1 (West Midlands)A2 Area 2 (North)A3 Area 3 (London)CCTV2 CCTVCCTV3 CCTV plus approved neighbourhood patrolsBeds 2 1 One less bedroomBeds 1 1 One more bedroomBeds 1 2 Two more bedroomsCondVG Property in very good conditionCondG Property in good conditionCondA Property in adequate conditionFlat Property is a � atMais Property is a maisonetteBedsit Property is a bedsitTlist Respondent on transfer listRet Respondent is retiredVsat – C Very satis� ed with current propertySat — C Satis� ed with current propertyWorse (A1, A2, A3) Worse estate in Area 1, 2 or 3Better A2 Better estate in A2Other (A1, A2, A3) Other dwelling on current estate in Area 1, 2 or 3

attributes with coef� cients statisticallysigni� cant at, as a minimum, the 5 per centlevel and of the correct sign (Table 3). The p2

goodness-of-� t statistic indicates how muchbetter the model predicts choices than chancealone. A value of 0.23 is, in our experience,somewhat higher than is typically achievedin more straightforward applications in trans-port research. Hensher and Johnson (1980)state that a � t between 0.2 and 0.4 is re-garded as ‘excellent’.

The properties of the estimated model areencouraging with regard to the design of theSP experiments, their presentation and con-duct, and the quality of the SP responsesobtained.

5. Interpretation

5.1 The Impact of Rents

To analyse the impact of rents on choices,separate rent coef� cients were estimated forthe current estate (RentC), the worse estate

(RentW 1 ), the better estate (RentB) and theother dwelling on the current estate (RentO).This allows the importance attached tochanges in rent to differ between the differ-ent estates.

Recall, however, that those in receipt ofHB who currently pay no or very little rentwould usually receive or retain money underthe assumptions of the SP experiment whenthe worse estate was chosen. Given that thereis a possibility that households respond dif-ferently to monetary outlay and to receivingmoney, we speci� ed separate rent variablesfor the worse estate according to whether therent to be paid by the participant was nega-tive (RentW-) or non-negative (RentW 1 ).The coef� cient for RentW- did not achievestatistical signi� cance (t 5 0.2) and hencewas not included in the � nal model. Thisresult implies that households have generallynot placed any importance on negative rents.Thus variations in rents that resulted in thepayment of negative rents appear to have no




Table 3. The results of stated preference model( q 2 5 0.23)

Attribute name Coef� cient (t-value)

RentC 2 0.0891 (7.0)RentW 1 2 0.0784 (5.2)RentB 2 0.0961 (7.1)RentO 2 0.0726 (5.6)CCTV2 0.6050 (3.0)CCTV3 0.9510 (5.5)Beds-1 2 2.1690 (11.6)Beds 1 1 0.6811 (5.8)Beds 1 2 0.3593 (3.0)Flat 2 1.0270 (10.8)CondVG 1.6070 (11.5)CondG 1.4950 (10.8)CondA 1.0590 (7.5)Tlist-C 2 0.2774 (3.7)Ret 2 1.4780 (16.0)VSat-C 1.7180 (16.6)Sat-C 0.6480 (9.4)Mais-C 2 0.2087 (1.9)Bedsit-C 2 0.3505 (2.1)Flat-C 2 0.4866 (5.4)Worse-A1 2 2.3210 (14.3)Worse-A2 2 5.3610 (10.0)Worse-A3 2 1.6830 (10.0)Better-A2 2 1.3120 (10.2)Other-A1 2 1.5800 (9.8)Other-A2 2 1.6740 (10.3)Other-A3 2 1.7753 (9.9)

that there is relatively little desire to movewithin the estate.

It is possible that the ordering of thecoef� cient estimates could have beenin� uenced by the operation of strategic bias.Respondents, in an attempt to in� uence pol-icy, could be signalling to decision-makersthat increases in rents on their currentdwellings or on better estates are much moreimportant, or less acceptable, to them, thanchanges in other rents. Even so, the lack ofsensitivity to reductions in the rents on worseestates is a notable � nding.

In order to capture any effects which liv-ing in the three different local authority areasmight have on the rent coef� cients, we ini-tially estimated separate models for eacharea, but found that there was little differencein rent coef� cients between these models.We would conclude from this that the rentcoef� cients in our model in Table 3 are‘transferable’ in that the estimated rentcoef� cient(s) accurately capture the impactof rents in different areas. Further, we did notobtain any evidence that the rent coef� cientsvaried across other categories of the sample,with the single exception of a slight incomeeffect which is discussed below.

5.2 The In� uence of Housing Bene� t

The approach used to establish whether HBin� uenced a household’s responses to rentchanges involved the speci� cation of dummyvariables. In brief, we created dummies indi-cating those who, after HB, paid 0 per cent,1–33 per cent, 34–66 per cent and 67–99 percent of the full rent, with the omitted cate-gory being those paying the full rent. Despiteextensive segmentations of the data accord-ing to the level of HB, no convincing, sys-tematic relationship with the rent coef� cientswas apparent. This was the case both withand without the speci� cation of the incomeeffect, as described below, and was indepen-dent of local authority area. The � ndingswere supported by the estimation of separatemodels for those with different HB entitle-ments.

The important implication of the in-

effect on utility and hence do not in� uencechoices in the SP experiment, although it ispossible that respondents found this part ofthe exercise implausible.

The rent coef� cients have the correct sign,since an increase in rent makes an alternativeless attractive. The coef� cient of –0.0784 fornon-negative rents (RentW 1 ) is lower (inabsolute terms) than the coef� cients forRentC and RentB, indicating that householdswere less sensitive to rents on the worseestates than to those set on better estates oron their current dwelling. This is not surpris-ing on the grounds that we might expectrespondents to be much more willing to‘trade-up’ than to ‘trade-down’ and hencereductions in rent on worse estates are notregarded as particularly relevant to them. Thecoef� cient for rent on other dwellings on thesame estate (RentO) is also low, indicating




signi� cance of the HB variables, allied to theinsigni� cance of the RentW- variable notedabove, is that, as modelled here, HB statusdoes not appear to affect, in a statisticallysigni� cant manner, households’ decisions asto whether to move in the face of changingrents. However, this does not mean thathousehold income, which is correlated withHB receipt, does not in� uence households’responsiveness to rent. We consider thisnext.

5.3 Allowing for the Effects of Income

To estimate the value of k in equation (9)above, income was de� ned as weekly house-hold income before deductions. The averageincome level was used for those who did notprovide income data. We had experimentedwith dividing the income variable by thenumber of people in the household, but thisdid not provide as good a � t as total house-hold income.

The best-� tting model was obtained for ak of 0.15. This means that, for example, adoubling of household income will reducethe sensitivity to rent changes by 10 per cent.This is only a weak effect, suggesting thathouseholds’ decision as to whether to choosean alternative dwelling in the presence ofvarying rent levels is not greatly affected bytheir income.

5.4 The Effect of Area

Our � ndings suggest that participants livingin the North (A2) were much more reluctantto move from their current dwelling to theworse or better estates. This may simply be are� ection of a regional difference in mo-bility, or it may be because the respondentsin the Northern authority viewed their cur-rent estate as signi� cantly better than thealternatives offered in the SP experiments.

The model also contains a set of ‘alterna-tive speci� c’ constants which re� ect the neteffect of variables not included in the model.If there are n alternatives, n 2 1 constants canbe estimated. We speci� ed constants forproperty on the worse estate (Worse), prop-

erty on the better estate (Better) and anotherproperty on the current estate (Other). Theseeffects are interpreted relative to the currentproperty. We also wanted to allow theseconstants to vary by area, which was foundto be warranted during our initial estimation,noted above, of separate models by area toestablish whether there was any variation inthe rent coef� cients.

The negative constants for other propertieson the same estate in all areas (Other-A1,Other-A2, Other-A3) denote that there is aninherent preference for remaining in the cur-rent property. We would expect the constantfor the worse estate to be more negativegiven that it discerns estate-speci� c effects aswell as an aversion to moving. This is thecase for two of the three areas (Worse-A1and Worse-A2). The constant for the betterestate in A1 and A3 is insigni� cantly differ-ent from that of the current estate, indicatingthat the higher quality of the estate is barelysuf� cient to overcome the inertia to moving.However, in the North the negative constanton the better estate (Better-A2) furtherre� ects the noticeably low desire to moveamong the households in this area.

5.5 The Effects of Other Attributes

The model identi� ed a range of statisticallysigni� cant ‘household-speci� c’ effects(Table 3), all of which are speci� c to thecurrent dwelling and denote variations in thepropensity to switch dwelling. Thecoef� cient for those on a transfer list (Tlist-C) is negative which denotes that, whengiven the possibility of moving house, thesehouseholds are more likely to move, as mighthave been expected. On the other hand, theretired (Ret) and those who are very satis� ed(VSat-C) or satis� ed (Sat-C) with their cur-rent property are less likely to move. Al-though correlated with degree of satisfaction,those whose current dwelling is a maisonette(Mais-C), a bedsit (Bedsit-C) or a � at (Flat-C) are more likely to move than those whocurrently live in a house or bungalow. These� ndings are intuitively reasonable.

Of the other variables explicitly controlled by Alireza Ehsanfar on October 29, 2008 http://usj.sagepub.comDownloaded from



in the SP experiments, it was found theCCTV and neighbourhood patrols hadsigni� cant positive values, but that house-holds were, in general, indifferent to theprovision of local shopping facilities. Thelatter attributes were not therefore retained inthe model.

One fewer bedroom (Beds-1) had a partic-ularly strong negative effect whilst two addi-tional bedrooms (Beds 1 2) had less of a(positive) impact on household choice thanone additional bedroom (Beds 1 1). This isexplicable in terms of the declining marginalutility of space and may re� ect the additionalcosts associated with a signi� cantly largerproperty. The results clearly suggest that par-ticipants did not feel that they had ‘toomuch’ room currently, but that some addi-tional room was preferable to substantiallymore.

The condition of the property is also animportant attribute. Given the absence of anynatural units in which to measure and rep-resent property condition, there is a degree ofsubjectivity surrounding respondents’ inter-pretations of the very good, good, adequateand poor categories speci� ed in the SP exer-cise. Dwellings in poor condition werede� ned as the base category. As expected,properties in very good condition (CondVG)were more likely to encourage an alternativeto be chosen than those in good condition(CondG) which in turn in� uence choice morestrongly than those in adequate condition(CondA). The largest increase in the proba-bility of choosing an alternative is when thatalternative is changed from poor condition toadequate condition, with the two furtherpossible improvements having a diminishingimpact.

6. Applying the Stated Preference Model

There are a number of uses to which this typeof SP model can be put. Here, we discusstwo applications: estimating the values whichhouseholds place on different dwelling andarea characteristics; and, forecasting house-holds’ probabilities of moving.

6.1 The Valuation of Housing Characteris-tics

The monetary values that households placeon different housing and environmental char-acteristics are implicit in the choices thatthey make when given the opportunity toconsume different bundles of characteristicsat differing rent levels. The valuations can bereadily estimated from the coef� cients of alinear additive SP model since the relativeimportance of any characteristic in monetaryterms in such a model is the ratio of thecoef� cient of that characteristic to that ofrent.

Recall from equation (9) above, however,that the rent coef� cient in our model is itselfdivided by household income raised to thepower of ( k 5 ) 0.15. As a result, monetaryvaluations will depend upon income and willincrease as income rises. Since monetary val-ues indicate the maximum willingness to payfor an attribute, it is plausible that monetaryvalues will be higher for those with higherincomes.

In our model, the value of a particularattribute with coef� cient a for a household iwould be

a

b /Y i0.15 5

a

bY i

0.15 (10)

where, Y is the household’s income and b isthe coef� cient estimated for rent divided byY 0.15. This implies that a doubling of house-hold income would increase values by 11 percent.

As we have seen, the rent coef� cients varyacross alternatives, implying different monet-ary valuations according to which rentcoef� cient is used. Here, we express themonetary valuations in terms of the rentcoef� cient estimated for the current dwellingso that the valuations relate to the character-istics of that dwelling. Table 4 presents sum-mary statistics for the valuations of a numberof attributes across our sample of house-holds.

The relative values all have the correctsign and, where the attribute has more thantwo levels, the relative magnitudes are plaus-ible. Thus, higher security is valued more




Table 4. Monetary valuations (£s per week), selected attributes

Characteristics Mean Standard error Minimum Maximum

Flat 2 24.3 0.09 2 19.2 2 28.9CCTV2 14.3 0.05 11.3 17.0CCTV3 22.5 0.08 17.8 26.7CondVG 38.0 0.14 30.1 45.1CondG 35.3 0.13 28.0 41.9CondA 25.0 0.09 19.8 29.8Beds-1 2 51.3 0.18 2 40.6 2 60.9Beds 1 1 16.1 0.06 12.8 19.1Beds 1 2 8.5 0.03 6.7 10.1

than lower security (CCTV3 . CCTV2) andprogressively higher valuations are placed onproperties in average, good and very goodcondition (CondVG . CondG . CondA). Thevalue placed on one extra bedroom is muchhigher than that placed on one less bedroombut, as noted earlier, is also higher than thatplaced on the signi� cant extra space impliedby two more bedrooms (Beds-1 ,Beds 1 1 . Beds 1 2). Flats are clearly un-popular.

One concern is that the absolute monetaryvalues in Table 4 generally appear to be high,particularly when the average weekly rent ofthe modal group of tenants is £42–50 perweek. In the speci� c case of property con-dition, where respondents appear to be indi-cating that, for example, they are willing topay £38 per week more for a dwelling invery good condition compared to poor con-dition, this may well be because of highcorrelations with the constants for the otherproperties on the current estate. Thecoef� cients for property condition will bein� ated if they are re� ecting some of theinherent features of the estate which ideallyshould be picked up by the constants. It isalso possible to argue that tenants in councilhousing currently pay sub-market rents, onaverage, but are in effect indicating throughthe experiment that this is the sum that theymight be prepared to pay in the market forthese characteristics. For example, paying anadditional £22.50 per week—or about 50 percent more in weekly rent—for high securitymay seem implausible, given current public-sector rents. However, it is quite possible that

tenants in the market may be prepared, andbe expected, to pay this sum for an otherwiseidentical dwelling in a safe environment inthe private sector.

More generally, it might be suggested thatthe high values are a result of participantsexhibiting strategic bias in respect of rent.However, were this present it would mostlikely be re� ected in an increase in house-holds’ reported sensitivity to rent in order tosend signals to policy-makers which reducethe likelihood or magnitude of a rent increaseand increase the likelihood or magnitude of arent reduction. Strategic bias operating onrent by increasing the size of the rentcoef� cient would thus tend to lead to rela-tively low monetary values. However, theremay be an element of respondents overem-phasising the importance of improved prop-erty conditions, the number of bedrooms,CCTV and dwelling type in order to increasethe chance that policy-makers will make im-provements in these areas. Comparisonagainst a well-controlled RP sample using alarge sample size would be able to test forthis.

A � nal issue which may be relevant here isthat a basic assumption of the SP approach isthat households will engage in ‘compensa-tory decision-making’, meaning that they areprepared to trade-off rent changes againstchanges in other housing attributes. How-ever, it is possible to argue that choices in thehousing market are made on a different basis.If households believe that there are minimumstandards or target levels that attributes suchas space and property condition must




Table 5. Choices between current property and properties on worse and better estates

West Midlands North London

Scenariosa C W B C W B C W B

Base 71.1 2.0 26.9 91.7 0.1 8.2 68.5 7.0 24.4Current 1 £1 70.4 2.1 27.5 91.4 0.1 8.5 67.8 7.2 25.0Current 1 £5 67.8 2.3 29.9 90.1 0.1 9.8 64.9 7.9 27.2Current 1 £10 64.5 2.5 33.0 88.3 0.2 11.5 61.0 8.8 30.2Worse 2 £1 71.0 2.1 26.9 91.7 0.1 8.2 68.4 7.3 24.3Worse 2 £5 70.8 2.4 26.8 91.7 0.1 8.2 67.7 8.3 24.0Worse 2 £10 70.6 2.8 26.6 91.6 0.2 8.2 66.9 9.7 23.4Better 1 £1 71.7 2.1 26.2 92.0 0.1 7.9 69.1 7.1 23.8Better 1 £5 74.2 2.2 23.6 93.1 0.1 6.8 71.4 7.6 21.0Better 1 £10 77.1 2.4 20.5 94.3 0.1 5.6 74.0 8.0 18.0Current 1 £0 Better 73.9 2.6 23.5 93.1 0.1 6.8 70.5 8.9 20.6

1 £5 Worse 2 £5Current 1 £5 Better 73.4 3.8 22.8 93.1 0.2 6.7 68.4 12.3 19.3

1 £10 Worse 2 £10Worse and Better and 86.9 0.9 12.2 97.2 0.0 2.8 85.9 3.1 11.0FlatWorse and Better 60.3 2.9 36.8 85.8 0.2 14.0 56.4 9.8 33.8

and Beds 1 1Worse 2 £5, Beds 1 1 65.8 10.9 23.3 91.1 0.8 8.1 53.3 30.0 16.7

and Very goodcondition

aThe factors shown in this column are the only ones in each scenario that change relative to the currentdwelling.

achieve, they may be far less willing to tradebetween these attributes than is assumed inthe SP exercise. This may lead to coef� cientestimates which imply a high willingness topay for certain attributes when in fact this isnot the case. We have already noted that therent coef� cient differs across estates eventhough these coef� cients relate to the samebasic commodity. This does not mean thatwe can state conclusively that households donot engage in compensatory decision-mak-ing, but the results can be read as consistentwith non-compensatory approaches. Wewould certainly conclude that this issue ofhigh relative valuations is one that meritsfurther attention.

6.2 The Probability of Moving

The SP model can be used to forecast theprobability of households choosing anotherdwelling (moving) as housing characteristicschange. Such forecasts are particularly im-portant for policy purposes, particularly

where the characteristics can be directlychanged by policy.

Table 5 indicates the probabilities (propor-tions) of tenants who will move in each ofour study areas in response to changes in,predominantly, the rents on the currentdwelling (C) and dwellings on the worseestate (W) and the better estate (B). The basescenario in each case indicates the proportionof tenants moving if offered a dwellingotherwise identical to their current dwellingexcept that it is on a better or worse estate. Inthis situation, around 70 per cent of tenantsin the West Midlands and London, and over90 per cent in the North wish to stay in theircurrent property. The low propensity to moveto a worse estate, particularly noticeable inthe West Midlands and the North, is to beexpected. It can be inferred that the tenantswho appear to be willing to move in thissituation do not actually regard the worseestate as being worse than the one on whichthey are currently located.

The table shows that an increase in the by Alireza Ehsanfar on October 29, 2008 http://usj.sagepub.comDownloaded from



rent of the current property by £10 per weekwould bring about a move by an additional7–8 per cent of households in the West Mid-lands and London, but only just over 3 percent in the North. The majority of thesemoves would be to a better estate. Loweringthe rent on the worse estate by £10 per weekhas only a marginal impact on the desire tomove there, with the additional 3 per cent ofhouseholds who would do so in London be-ing the largest quantitative effect. Raising therent on the better estate reduces the desire tomove there with about 6 per cent of house-holds in the West Midlands and Londonindicating a preference for staying in theircurrent dwelling in such circumstances. Onlya very small number of households wouldconsider moving to a worse estate in such asituation.

Offering a � at on the worse and betterestates (Worse Better and Flat) signi� cantlyreduces the desire to move there in all areas.Offering a property with one more bedroomhas little effect on the desire to move to theworse estate but increases by 6–9 percentagepoints the number of households willing tomove to the better estate (Worse and Betterand Beds 1 1). Perhaps the most striking re-sult in Table 5 is that for the London boroughwhen a property in very good condition, oneextra bedroom and a rent £5 lower than thatof the current dwelling is offered(Worse 2 £5, Beds 1 1 and Very good con-dition). While only an extra 9 per cent ofhouseholds would move to such a property inthe West Midlands, and very few additionalhouseholds would move in the North, a nota-ble 23 per cent of respondents would do so inLondon. One reason for this may be that theproximity of the alternative estates in theLondon borough compared with that of theWest Midlands and the North overcomesinertia. It may also re� ect the premium thathouseholds place on extra space in the capi-tal.

Perhaps the most important results for pol-icy concern the effects of rent changes aloneon the probability of public-sector tenantsmoving within the sector. Two scenariosmost closely re� ect current proposals to in-

crease rent differentials. One is where thecurrent rent remains the same but the rents ofthe better and worse estates increase anddecrease respectively by £5 per week (Cur-rent 1 £0 Better 1 £5 Worse 2 £5). The otheris where rent of the current dwelling in-creases by £5 per week, and those on thebetter and worse estates increase/decrease by£10 per week (Current 1 £5 Better 1 £10Worse 2 £10).

In the � rst case, where current rent isunchanged and there is a smaller increase indifferentials, the desire to stay put increasessomewhat in all areas. The desire to move tothe worse estate increases only marginally ornot at all while the property on the betterestate in all cases becomes less popular. Inthe second case, where current rents are in-creased and there are signi� cantly largerchanges in the rents elsewhere, the desire toremain in the current dwelling remainsbroadly the same or increases slightly com-pared to the base scenario. In particular, theimplied increase in the differential to £15 perweek between the current property and oneotherwise identical to it on a worse estateleads to only a 5 percentage point increase inthe number of households wishing to movethere in London. It leads to increases of lessthan 2 percentage points in the number ofhouseholds relocating to a worse estate in theWest Midlands and the North.

7. Conclusions

We conclude by re� ecting upon the methodused for this research and seek to highlightsome of the policy implications of our� ndings.

In respect of method, the SP approachmodels hypothetical choices and, as we haveacknowledged, because of data constraintswe have not been able to test these againstRP information in the way that we wouldhave wished. In our SP exercise, we haveestimated households’ preferences assumingthat the choices offered are available. In thissense, we have ‘arti� cially’ created a marketwhere one barely exists currently. However,we would argue that because social tenants—




both currently and historically—have hadvery little choice, SP is the only feasible wayof assessing how tenants might re� ect theirpreferences through choices. Nevertheless,even though the SP experiment itself wascarefully constructed, the issues aroundwhich it was based were relevant to partici-pants and the analysis based upon their re-sponses is statistically robust, the absence ofRP data means that we cannot be certain thatthe participants’ choices re� ect their prefer-ences. There are clear dif� culties in applyingchoice experiments in housing, given that wehave little theoretical guide as to how thosechoices are made and the relative weight ofvarious factors in decision-making. It is alsopossible that the assumption that householdswill engage in trade-offs is inappropriate tohousing, although if this is the case it hasserious implications for the great majority ofeconomic analyses of consumer behaviour inthe housing � eld.

It also needs to be recognised that house-holds would not repeatedly make the sort ofchoices offered in the exercises—i.e. house-holds generally make housing choices at in-frequent intervals, particularly in the publicrented sector. Another model would be re-quired to forecast when individuals wouldactually put themselves in the position ofchoosing. While this could be based onchanges in current or potential housing at-tributes and prices or on changes in householdcircumstances, a necessary requirement is,clearly, that housing authorities actually offeralternative housing where none existed previ-ously. Such a model would have to include atime-lag so that households did not evaluateand re-evaluate options more often than theywould in practice.

An extended speci� cation of the MNLmodel, such as that discussed by Timmer-mans et al. (1996) may, in future work, proveto be a potentially fruitful way of modellingthe sort of constrained choices faced by socialhousing tenants. However, it needs to berecognised that the major constraint facingmany such tenants is the reality of a lack ofchoice and a lack of alternatives within thesector at present. Further, our study was con-

cerned with piloting a complex decision-mak-ing exercise where it is dif� cult to identifyand model in detail the (other) constraintsfacing respondents. In some respects, the factthat the results of the SP experiment areintuitively reasonable in itself gives supportto the argument that the way in which wehave used the approach is valid. If partici-pants had not indicated that they had valued,for example, property condition and more, butnot too much more, space rather than less,there would be greater concerns about theusefulness of the method and/or the way inwhich it had been applied. Further, in indicat-ing that rent is not the most important factorin housing decisions and in being extremelyreluctant to move to worse areas despite muchlower rents, participants are con� rming whatmany in the housing � eld might have ex-pected.

Given these considerations, we are able todraw three main policy implications from ourwork. First, given the background that themajority of households display a preferencefor remaining in their current dwelling,households’ decisions to move from theircurrent dwelling in the light of changing rentsdo not appear to vary signi� cantly by area orhousehold income. Consequently, policies toencourage mobility through rent changes areunlikely to impact differentially on house-holds’ decisions as a result of their incomesor the (urbanised) area in which they live.

Secondly, whether or not a household re-ceives any HB does not appear to affect theirmoving decisions when they are faced withchanges in the relative rents of the alternativedwellings offered. This may be because HBrecipients (other than the retired) do not seetheir HB status as permanent and thus re-spond, or fail to respond, to rent changes ina similar manner to those who are paying allof their rent. This � nding also carries theimplication that, for example, reducing, orindeed increasing, recipients’ entitlements toHB while also changing the rents on differentproperties would not encourage either agreater or lesser number of such householdsto move compared to households not receiv-ing HB.




Finally, for tenants as a whole, relocationdecisions are comparatively insensitive torent changes and such decisions are particu-larly insensitive to reductions in rent on theworse estates. Our � ndings suggest thatwidening rent differentials signi� cantly be-tween better, average and worse estateswould have little impact on the moving deci-sions of most tenants. This suggests thatpolicies intended to encourage relocationthrough relative rent changes alone are un-likely to be very successful. Rather, movingdecisions are more likely to be affected byhigher levels of security on an estate, thecondition of the property, being offered ahouse there rather than a � at and being of-fered a somewhat larger property, all ofwhich positively in� uence the decision tomove. Further support to this argument isgiven by the � nding that households placehigh implicit monetary valuations on prop-erty condition, estate security and larger (butnot very much larger) dwellings.

The signi� cance of these factors can betaken as indication of the sort of consider-ations that social landlords should take intoaccount when devising investment strategiesintended to provide the types of new orrefurbished housing which will prove attract-ive to tenants. The emphasis tenants place onthe quality of the estate and security are alsovery much in line with current thinking in theUK about those factors which policies forurban regeneration and estate improvementought to address.

Note1. Although it will in� uence demand forecasts

through its effects on the absolute scale ofthe parameters even if their relative scale isunaffected (Wardman, 1991).

References

ADAMOWICZ, W., LOUVIERE, J. and WILLIAMS, M.(1994) Combining revealed and stated prefer-ence methods for valuing environmental ameni-ties, Journal of Environmental Economics andManagement, 26, pp. 271–292

BARELLI, J. (1992) Underoccupation in LocalAuthority and Housing Association Housing.London: HMSO.

BATES, J. J. (1983) Stated preference techniquesfor the analysis of travel behaviour. Paper pre-sented at the World Conference on TransportResearch, Hamburg, July.

BATES, J. J. (1988) Econometric issues in statedpreference analysis, Journal of TransportEconomics and Policy, 22, pp. 59–69.

BENJAMIN, J. and PAASWELL, R. E. (1981) Apsychometric analysis of residential location,Journal of Socio-Economic Planning Sciences,15, pp. 305–319.

BOXALL, P.C., ADAMOWICZ, W. L., SWAIT, J. ET AL.(1996) A comparison of stated preferencemethods for environmental valuation, Ecologi-cal Economics, 18, pp. 243–253.

BRAMLEY, G., PAWSON, H. and THIRD, H. (2000)Low Demand Housing and Unpopular Neigh-bourhoods . London: DETR.

CARROLL, J. D. and GREEN, P. E. (1995) Psycho-metric methods in marketing research: Part 1,conjoint analysis, Journal of MarketingResearch, 32, pp. 385–391.

CATTINK, P. and WITTINK, D. R. (1982) Commer-cial use of conjoint analysis: a survey, Journalof Marketing, 46, pp. 44–53.

COOPER, J., RYLEY, T. and SMYTH, A. (2001)Energy trade-offs and market responses intransport and residential land-use patterns:promoting sustainable development policy andpitfalls, Urban Studies, 38, pp. 1573–1588.

DETR/DSS (DEPARTMENT OF THE ENVIRONMENT,TRANSPORT AND THE REGIONS/DEPARTMENT OFSOCIAL SECURITY) (2000a) Quality and Choice:A Decent Home for All. London: DETR.

DETR/DSS (DEPARTMENT OF THE ENVIRONMENT,TRANSPORT AND THE REGIONS/DEPARTMENT OFSOCIAL SECURITY) (2000b) Quality and Choice:A Decent Home for All. The Way Forward forHousing. London: DETR.

DOE (DEPARTMENT OF THE ENVIRONMENT) (1996)Housing Revenue Manual (Amendment).London: HMSO.

DONNELLY, E. P., HOWE, S. M. and DESCHAMPS,J. A. (1976) Trade-off analysis: theory andapplications to transportation policy analysis.Preliminary Research Report 103, Planning Re-search Unit, New York State Department ofTransportation, Albany, New York.

FIEDLER, J. A. (1972) Condominium design andpricing: a case study in consumer trade-offanalysis, Paper presented at the Third AnnualConference of the Association for ConsumerResearch, Chicago.

GREEN, P. E. and SRINIVASEN, V. (1978) Conjointanalysis in consumer research: issues and out-look, Journal of Consumer Research, 5,pp. 103–123.

GREEN, P. E. and SRINIVASEN, V. (1990) Conjointanalysis in marketing: new developments withimplications for research and practice, Journalof Marketing, 54, pp. 3–19.




HENSHER, D. A. and BARNARD, P. O. (1990) Theorthogonality issue in stated choice designs, in:M. FISCHER, P. NIJKAMP and Y. PAPAGEORGIOU

(Eds) Spatial Choices and Processes, pp. 265–278. Amsterdam: North Holland.

HOINVILLE, G. (1971) Evaluating community pref-erences, Environment and Planning A, 3,pp. 33–50.

JOSEPH, A. E., SMIT, B. and MCILRAVEY, G. P.(1989) Consumer preferences for rural resi-dences: a conjoint analysis in Ontario, Canada,Environment and Planning A, 21, pp. 47–63.

KNIGHT, R. L. and MENCHIK, M. D. (1976) Con-joint preference estimation for residential landuse policy evaluation in: R. G. GOLLEDGE andG. RUSHTON (Eds) Spatial Choice and SpatialBehaviour, pp. 135–155. Columbus, OH: OhioState University Press.

LANCASTER, K. (1971) Consumer Demand: A NewApproach. New York: Columbia UniversityPress.

LOUVIERE, J. J. (1988) Conjoint analysis mod-elling of stated preferences, Journal of Trans-port Economics and Policy, 22, pp. 93–119.

LOUVIERE, J. J. and TIMMERMANS, H. J. P. (1990)Stated preference and choice models applied torecreation research: a review, Leisure Science,12, pp. 9–32.

LOUVIERE, J. J., HENSHER, D. A. and SWAIT, J. D.(2000) Stated Choice Methods: Analysis andApplication. Cambridge: Cambridge UniversityPress.

LUCE, R. D. and TUKEY, J. W. (1964) Simulta-neous conjoint measurement: a new type offundamental measurement, Journal of Math-ematical Psychology, 1, pp. 1–27.

MARSH, A. (2001) Restructuring social housingrents, in: D. COWAN and A. MARSH (Eds) TwoSteps Forward: Housing Policy in the NewMillennium, pp. 285–304. Bristol: The PolicyPress.

MCCLAIN, J. O. and RAO, V. R. (1974) Trade-offsand con� icts in evaluation of health systemsalternatives: a methodology for analysis,Health Services Research, 9, pp. 35–52.

MINFORD, P., PEEL, M. and ASHTON, P. (1987) Thehousing morass. Hobart Paper No 25, Instituteof Economic Affairs, London.

MOLIN, E. J. E., OPPEWAL, H. and TIMMERMANS, H.J. P. (1997) Modelling group preferences usinga decompositional preference approach, GroupDecision and Negotiation, 6, pp. 339–350.

ORTUZAR, J. DE D. (2000) Stated Preference Mod-elling Techniques. London: PTRC.

PAT7 (Policy Action Team 7) (1999) UnpopularHousing. London: DETR.

RYAN, M. (1999) Measuring bene� ts in healthcare: the role of discrete choice conjoint analy-sis. Paper presented at the International HealthEconomics Association Conference, Rotterdam.

RYAN, M. and WORDSWORTH, S. (2000) Sensi-tivity of willingness to pay estimates to thelevel of attributes in discrete choice experi-ments, Scottish Journal of Political Economy,47, pp. 504–524

SCHELLEKENS, M. P. G. and TIMMERMANS, H. J. P.(1997) A conjoint based simulation model ofhousing-market clearing processes: theory andillustration, Environment and Planning A, 29,pp. 1831–1846.

SHELDON, R. and STEER, J. K. (1982) The use ofconjoint analysis in transport research. Paperpresented to the PTRC Summer Annual Confer-ence, London, September.

TIMMERMANS, H. J. P. (1984) Decompositionalmultiattribute preference models in spatialanalysis, Progress in Human Geography, 8,pp. 189–221.

TIMMERMANS, H., NOORTWIJK, L. VAN, OPPEWAL,H. and WAERDEN, P. VAN DER (1996) Modellingconstrained choice behaviour in regulated hous-ing markets by means of discrete choice exper-iments and universal logit models: anapplication to the residential choice behaviourof divorcees, Environment and Planning A, 28,pp. 1095–1112.

VELDHUISEN, K. J. and TIMMERMANS, H. J. P.(1984) Speci� cation of individual residentialutility functions: a comparative analysis ofthree measurement procedures, Environmentand Planning A, 16, pp. 1573–1582.

VYVERE, Y. VAN DE (1994) Stated Preferencedecompositional modelling and residentialchoice, Geoforum, 25, pp. 189–202.

WALKER, B. and MARSH, A. (1995) Rent SettingPolicies in English Local Authorities. London:HMSO.

WALKER, B. and MARSH, A. (1998) Pricing publichousing services: mirroring the market?, Hous-ing Studies, 13, pp. 549–566.

WALKER, B. and MARSH, A. (2000) Rent SettingPolicies in English Local Authorities, 1998/99.London: DETR.

WALKER, B., NINER, P., MARSH, A. and WARD-MAN, M. (2000) Social Housing Tenants’ Atti-tudes to Rent Levels and Rent Differentials.London: DETR.

WARDMAN, M. (1991). Stated preference methodsand travel demand forecasting: an examinationof the scale factor problem, TransportationResearch, 25A, pp. 79–89.

WARDMAN, M. (1998) The value of travel time: areview of British evidence, Journal of Trans-port Economics and Policy, 32, pp. 285–315.

WARDMAN, M. and WHELAN, G. (2001) Valuationof improved railway rolling stock, TransportReviews, 2, pp. 415–447.

WARDMAN, M., BRISTOW, A. and HODGSON, F.(1998) Noise and air quality valuations: evi-dence from stated preference residential and




business choice models. Paper presented to the8th World Conference on Transport Research,Antwerp, July.

WIDLERT, S. (1994) Stated preference studies: thedesign affects the results. Paper presented at the7th International Conference on Travel Behav-iour, Santiago, January.

WILCOX, S. (1995) Housing Finance Review1995/96 . York: Joseph Rowntree Foundation.

WILCOX, S. (1997) Housing Finance Review1997/98 . York: Joseph Rowntree Foundation.

WILCOX, S. (2000) Housing Finance Review2000/2001 , York: Joseph Rowntree Foun-dation.

WIND, Y. and SPITZ, L. K. (1976) An analyticalapproach to marketing decisions in health careorganisations, Operations Research, 24,pp. 973–990.

WITTINK, D. R. and CATTIN, P. (1989) Commer-cial use of conjoint analysis: an update, Journalof Marketing, 53(July), pp. 91–96.



modelling tenants' choices in the public rented sector: a stated

Documents