The Journal of Choice Modelling - COnnecting REpositories · 2017-02-15 · Romy Greinera,n, Michiel Bliemerb, Julie Ballwega a Charles Darwin University, ... Darwin, NT, Australia

Page 1

Contents lists available at ScienceDirect

The Journal of Choice Modelling

The Journal of Choice Modelling 10 (2014) 34–45

1755-53http://d

n CorrE-m

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

Design considerations of a choice experiment to estimatelikely participation by north Australian pastoralists incontractual biodiversity conservation.

Romy Greiner a,n, Michiel Bliemer b, Julie Ballweg a

a Charles Darwin University, Institute for the Environment and Livelihoods, Darwin, NT, Australiab The University of Sydney Business School, Institute of Transport and Logistics Studies, Sydney, NSW, Australia

a r t i c l e i n f o

Article history:Received 6 March 2013Received in revised form14 November 2013Accepted 3 January 2014Available online 8 February 2014

Keywords:Choice experimental design‘Best-worst’ response mechanismBayesian D-efficient designWillingness to acceptFarmersBiodiversity conservation programme

45 & 2014 The Authors. Published by Elsevix.doi.org/10.1016/j.jocm.2014.01.002

esponding author. Tel.: þ61 418 242 156ail address: [email protected] (R. Gre

a b s t r a c t

This paper reports on the experimental design process and considerations of a choiceexperiment conducted in collaboration with farmers in northern Australia. The purpose ofthe research is to inform the design of effective and efficient payments-for-ecosystemservices schemes to safeguard north Australia’s biodiversity values. It promotes thecontractual provision of biodiversity conservation services by farmers, in particularpastoralists operating in Australia’s tropical savannas.

The paper focuses on the discrete choice experimental (DCE) aspects. The DCE isemployed to estimate farmers’ preference heterogeneity for supplying ecosystem services,specifically their willingness to accept remuneration for the on-farm conservation ofbiodiversity, based on potential programme attributes. The design of the choice experi-ment draws on best practice standards (Hoyos, 2010), a recognition of the benefits ofembedding design in a consultative process (Klojgaard et al., 2012) and recent advances inaccounting for response certainty (Brouwer et al., 2010; Hensher et al., 2012).

DCE design decisions relating to attribute selection, attribute levels, alternatives and choicetasks are explained based on literature, focus group discussions, expert input and an iterativeprocess of Bayesian D-efficient DCE design. Additional design aspects include measuring choicecertainty and stated attribute attendance, embedding the DCE within a discrete-continuousapproach, capturing relevant respondent-related attributes with socio-economic-psychologicalquestions and scales, and devising appropriate data collection logistics.

& 2014 The Authors. Published by Elsevier Ltd. Open access under the CC BY-NC-ND license.

1. Introduction

The tropical savannas of Australia cover around 1.9 million square kilometres (25% of the continent) across the north ofthe continent. Savanna landscapes support an abundance of endemic plants and animals, which are adapted to the harshclimatic conditions (Woinarski et al., 2007). Although savanna landscapes may appear relatively intact, their ecologicalcondition has widely declined since European settlement (Lewis, 2002). Land use practices, in particular over-grazing, andspread of exotic plant and animal species have caused widespread environmental degradation (Woinarski et al., 2007).

Tropical savannas endure a combination of relative under-representation in the formal conservation estate and lowparticipation of farmers in on-farm conservation. The states who share the tropical savannas, Queensland, the Northern

er Ltd.

iner).

Open access under the CC BY-NC-ND license.

Page 2

R. Greiner et al. / Journal of Choice Modelling 10 (2014) 34–45 35

Territory and Western Australia, have below-average proportions of land set aside for biodiversity conservation andprotection purposes (‘formal conservation estate’) with 1.6, 1.7 and 1.1 per cent, respectively (compared to the nationalaverage of 1.9%, ABS, 2011a). Conservation reserves in northern Australia are also not large enough, on their own, tomaintain viable populations of many endangered species and the ecological processes necessary to them in the long term(Bennett 1995). On-farm biodiversity conservation is therefore an important element of a strategy for safeguarding northAustralia’s natural heritage. A majority of farmers in the three states/territories report having native vegetation on theirholdings and report protecting at least some of it (ABS, 2011b). However, in the natural resource management regions whichcover the tropical savannas, no more than 41% of farmers protect native vegetation (ABS, 2011b).

“The contributions of all property holders and managers are needed to maintain the North’s natural values” (Woinarski et al.,2007, p. 88). The primary land use of Australia’s tropical savannas is extensive beef production. Individual beef grazingenterprises are up to 24 000 km2 in size and carry up to 65 000 head of cattle (Bortolussi et al., 2005). Nowhere isconservation action more critical than on farms that cover vast tracks of land with high ecosystem values, as one farmer’sland use decisions can have implications for soil, water and biodiversity conditions at the regional scale.

There have been a succession of biodiversity conservation programs in Australia over recent decades, but most have beenshown to be ineffective in targeting and inefficient in design (Hajkowicz, 2009). Designing incentive programs that areeffective and efficient requires that policy makers have a detailed understanding of (i) the financial resources required toincentivize a sufficient number of farmers to participate in on-farm conservation and (ii) the way in which programme andcontract design and administrative features influence participation. This research generates such understanding byexploring how programme attributes relate to farmers’ willingness to participate in contractual on-farm biodiversityconservation, and how much land and what type of land they would subscribe under what conditions.

This paper describes the design process of a choice experiment as the principal method for generating data which cananswer the research questions. Initial design considerations are presented, results of the DCE pre-test and pilot test(completed in late 2012) shown and the updated design discussed. The DCE is embedded in a larger socio-economic surveyof farmers so that choice decisions may be linked to social, psychological and economic models of decision making. Thepaper is targeted at applied choice modellers, particularly in the field of environmental management, and helps address thepaucity of literature which illustrates the multitude of choices the analyst has to make when designing a choice experiment.

2. Willingness to accept approach

Exploring agents’ behaviour in novel markets, in this case the question about participation in on-farm biodiversityconservation for money, poses a range of methodological challenges (Rolfe et al., 2004) and due to the absence of marketobservations a stated preference approach is required, such as a choice experiment (CE). CEs have become the method ofchoice to generate understanding which can support the design of new agricultural markets (Lusk and Hudson, 2004; Rolfeet al., 2008; Windle and Rolfe, 2005).

This application of CE explores the potential supply of an environmental service by farmers and has been used previouslyin the design of payments for ecosystem service programs (Beharry-Borg et al., 2013; Broch et al., 2013; Christensen et al.,2011; Espinosa-Goded et al., 2010; Kaczan et al., 2013; Ruto and Garrod, 2009). North Australian pastoralists have exclusiveproperty rights over their land, associated with land title, and are being asked to voluntarily give up elements of thatproperty right in return for remuneration, making willingness-to-accept (WTA) the correct conceptual construct to use(e.g. Broch et al., 2013; Carson et al., 2001; Kaczan et al., 2013). While WTA applications have been shown to be prone tostrategic bias when compared to willingness to pay applications (Grutters et al., 2008; Horowitz and McConnell, 2002;Mitchell and Carson, 1989), CE is arguably less prone to such bias than other stated choice methods (Burton, 2010).Respondents can be expected to have a high degree of task familiarity, which is important for reducing bias in statedpreference studies (Schläpfer and Fischhoff, 2012), as farmers are familiar with the concept of receiving payments for theprovision of environmental services through a series of government programs in recent decades, including grants, auctionsand cost-sharing programs.

CE elicits WTA indirectly, by asking respondents to choose between cleverly designed alternatives. CE assumes thatpeoples’ preferences are revealed through the choices they make. The method integrates concepts of conjoint analysis anddiscrete choice theory (Louviere and Hensher, 1982; Louviere and Woodworth, 1983). Respondents are presented withrepeated samples of hypothetical scenarios (choice tasks) drawn from all possible choice tasks according to statistical designprinciples (Ryan et al., 2008).

3. Design of the discrete choice experiment

The aim of a DCE is to estimate the weights that respondents place on each of the attributes which define thealternatives. A respondent acting rationally is expected to evaluate the alternatives in a choice task and choose thealternative which gives the greatest relative utility (Hensher et al., 2005). This premise of general utility theory, whenapplied to agricultural producers, offers an alternative to the profit maximisation paradigm, particularly in the presence ofrisk (Barry et al., 2009; Bond et al., 2011; Lin et al., 1974; Robison, 1982).

Thus, a pastoralists is expected to choose land management alternative A over B, if U (XA, Z)4U (XB, Z), whereU represents his/her indirect utility function from given land management alternatives, XA the attributes of land use A, XB the

Page 3

R. Greiner et al. / Journal of Choice Modelling 10 (2014) 34–4536

attributes of alternative B, and Z the personal (e.g. socio-demographic and attitudinal) and property characteristics (e.g. size,land productivity, farm profitability, ownership structure) that influence the pastoralist’s utility. Choices made in DCEs areanalysed using random utility theory, meaning a stochastic error term ε is included in the utility function to reflect theunobservable factors in the respondent’s utility function (Hensher et al., 2005). Thus, a pastoralist will choose alternative Aover B, if V (XA, Z)þεA4V (XB, Z)þεB, where V is the measurable component of utility estimated empirically, and εA and εBreflect the unobservable factors in the pastoralist’s utility function of alternatives A and B respectively.

Design dimensions fundamentally influence the results of choice experiments and resulting recommendations (Rolfe andBennett, 2009). In particular, design dimensions influence the fit of the econometric model applied to data analysis, asmeasured by the relative size of ε. A good design is able to explain more of the observed variance and minimises thestochastic element.

Decisions regarding experiment design that need to be made prior to construction of the experimental design matrixinclude (Bliemer and Rose, 2011; Hoyos, 2010):

▪

What alternatives, attributes and attribute levels should be included in the experiment? ▪ What response mechanism will be used? ▪ What will the utility function look like? ▪ What model will most likely be estimated after data collection? ▪ What statistical properties should the experimental design display? ▪ How many choice tasks should the design include? ▪ How will the survey be administered once the design has been generated?

Answers to these questions are provided below before the final design is exemplified.

3.1. Choice alternatives and response format

Responses in a DCE can take on different formats including ‘pick-one’, ‘best-worse’, and others. This research applies the‘best-worst’ format. While ‘pick-one’ better mimics real life decision making, it only captures the first preference. ‘Best-worst’ also reveals the first preference but elicits additional preference information per choice (Lancsar et al., 2013; Potoglouet al., 2011). In a situation where the sample size is expected to be low, this poses a distinct advantage. ‘Best-worst’ takesadvantage of an individual’s ability to identify extreme options and it is easy for respondents to understand (Flynn et al.,2007; Morrison et al., 2002).

‘Best-worst’ choices can be applied in a sequential fashion in order to obtain a full raking of all alternatives by askingrespondents to choose their preferred option, then to choose the worst option, then the best of the remaining options, etc. Whileimposing a higher cognitive burden on the respondent than ‘pick-one’, the tasks are easier to deal with than traditional rankingmethods (Marley and Louviere, 2005). Compared with ‘pick-one’, a sequential best-worst response format greatly reduces thenumber of choice tasks required to obtain the same number of observations (Lancsar et al., 2013). Best-worst scaling has beenfound to be superior when dealing with qualitative data such as the different conservation requirements and different monitoringarrangements explored in this choice experiment (Flynn et al., 2007; Goodman et al., 2005).

The number of alternatives in a DCE has a large influence on error variance. According to Caussade et al. (2005) it has thesecond largest influence on error variances out of all design dimensions with four alternatives being superior to three or fivein terms of scale effects. More alternatives increase the cognitive burden on respondents but Hensher (2006) illustrates thatrelevance of alternatives is more important than trying to limit cognitive burden.

A 3-alternative design is adopted. A ‘none’ option is also included to reflect unconditional demand and thus ensureconceptual validity of the design given the voluntary nature of farmer participation in a payments-for-ecosystem servicesprogramme. Rolfe and Bennett (2009) found that a 3-alternative design (with a ‘not sure’ option) generated moreparticipation compared to a 2-alternative design and was therefore preferable. However, Adamowicz et al. (2005) found thatrespondents in the 3-alternative version were more likely to choose a status quo option than in the 2-alternative version.

The alternatives in our DCE are of an unlabelled type (Louviere et al., 2000) and have generic titles (options ‘A’, ‘B’ and ‘C’)because this fits with the generic nature of the project’s investigation of the role that attributes of biodiversity conservationcontracts play in acceptance by farmers. Unlabelled designs have been shown to increase respondents’ attention toattributes and are therefore more suitable to investigating trade-offs between attributes (de Bekker-Grob, 2009).

3.2. Attributes and attribute levels: literature review and industry consultation

As this research requires the collaboration of members of an agricultural sector, it is critical to embed it within aconsultative process (Hoyos, 2010; Klojgaard et al., 2012). Choice experiments that include a policy question or a politicalchallenge must be included and explained when choosing attributes (Barkmann et al., 2008). A list of possible attributes canbe gleaned a priori from the literature but the list must be refined through focus groups and pilot studies (Ryan et al., 2008).In this case, the research was deemed of strategic importance to the north Australian pastoral industry and needed to havethe support of and be relevant to industry members to achieve sufficient participation and ultimately be successful.

Page 4

R. Greiner et al. / Journal of Choice Modelling 10 (2014) 34–45 37

Focus group meetings are essential for achieving relevancy in the eyes of the research agents by informing a number ofdesign choices (e.g. Rolfe et al., 2004). Focus group meetings were conducted in two central locations of the tropical savannas,in Croydon (Qld) on 12 September 2012 with six pastoralists attending and in Katherine (NT) on 1 October 2012, where ninepastoralists attended. The meetings pursued multiple objectives, (i) to narrow the framing of the choice experiments andimprove contextual presentation and relevancy of biodiversity conservation to pastoralists, and (ii) to establish whatprogramme attributes would critically influence pastoralists’ decision making and discuss potential attribute levels.

Key messages from the focus group meetings included: firstly, the alternatives would focus on broad-scale biodiversityconservation for the benefit of maintaining habitat integrity of flagship species and grazing-sensitive species, as opposed tothe conservation of localised natural assets such as waterholes and caves. Broad-scale conservation ensured relevancy ofconservation activities across a wide range of geographical and business situations and enabled remuneration to bemeaningfully expressed on a dollar-per-hectare basis. Secondly, to make the conservation alternatives relevant forpastoralists, they would need to be expressed in terms of their relationship with the principal enterprise, cattle grazing.Exclusion of cattle from areas designated for biodiversity conservation was to be expressed in terms of duration of exclusionand impact on grazing productivity. Management requirements associated with shifting focus of land management tobiodiversity conservation were to be specified, including weed, feral animal and fire management.

As per Ryan et al. (2008), a listing of possible attributes had been created from the literature (in particular: Broch andVedel, 2012; Horne, 2006; Peterson, 2011; Peterson et al., 2011; Rolfe et al., 2004; Rolfe and Windle, 2005; Ruto and Garrod,2009; Vedel et al., 2010; Windle et al., 2005; Yu and Belcher, 2011). Attribute suggestions were discussed with industrymembers in focus group discussions, the selection narrowed and attribute levels were also discussed. Focus grouprespondents found the following attributes relevant to their decision making: length of agreement (5-year—permanent);flexibility within the agreement to respond to ‘exceptional natural circumstances’ occurring (available—not available),where the funding for the programme would come from (government—philanthropy—commercial sector), and who wouldconduct the monitoring (self—funder—independent body).

To describe the monetary dimension of the alternatives, it was suggested that in all cases investment in fencing andwatering points necessary to implement any contract alternative should be provided up-front so that alternatives would becapital-cost neutral. Alternative-specific annual payments needed to be specified on a per-hectare basis—normalised tocommencement year and indexed for the contract duration. It was for respondents to decide whether the annual paymentswould cover the combination of property-specific opportunity costs (production and other), management costs, monitoringcost (if applicable), risk premium and profit margin.

Given the large influence of the monetary attribute on model outcomes (third largest influence on error variances out of alldesign dimensions; Caussade et al., 2005), the choice of compensation levels was further guided by (i) historical data about the landproductivity of the tropical savannas, in particular the value of cattle sales per hectare during 1992–2011 as derived from farmsurvey data (ABARES, 2012) and (ii) feedback from industry members.

The industry directives were further considered in the context of the relevant literature (see above) and the ecological merit andgrazing land management dimensions of the conservation options were discussed with rangelands ecologists and grazingproduction experts (see acknowledgements). Table 1 summarises the initial attributes and attribute levels derived in this manner.

When analysing pastoralists’ decision making in relation to participation in conservation programs, two attributesegments are relevant: (i) the conservation programme characteristics which are captured as attributes in the choices and(ii) respondent-specific factors, including farm and personal characteristics (Productivity Commission 2001). Respondent-specific attributes are not reflected in the choice experiment but captured directly or indirectly in the remainder of thesurvey. Farm and personal characteristics of respondents that are expected to be relevant to explaining pastoralists’propensity to engage in biodiversity conservation include, e.g., age of respondent, size of property and cattle herd, landproductivity, ownership structure, farm profitability and equity, risk perceptions, motivations and attitudes (Greiner andGregg, 2011; Greiner et al., 2009; Lankester, 2013; McAllister et al., 2006).

Respondent-specific parameters will be included in the CE model specification so that their influence on likelyparticipation in contractual biodiversity conservation can be quantified. For example, it is expected that land productivity(as approximated by average stocking rate) will be shown to be positively correlated to WTA, with pastoralists on better landrequiring more compensation per hectare for a given level of conservation service (due to higher opportunity cost) in orderto sign up to a contract (e.g. Yu and Belcher, 2011).

3.3. Bayesian D-efficient design

There are broadly two schools of thought about the statistical properties of experimental design display, efficient designversus orthogonal design. Orthogonality is defined and constructed in relation to the design codes in which the attributelevels between different attributes are uncorrelated (Louviere et al., 2000). A design is orthogonal when every pair of levelsoccurs equally often across all pairs of attributes, or when the frequency for level pairs are proportional instead of equal(Kuhfeld, 2006). While orthogonal designs are more prevalent in the literature, efficient design has recently emerged as analternative with new algorithms to facilitate the design. Efficient designs have been empirically shown to lead to smallerstandard errors in model estimation at smaller sample sizes compared to orthogonal designs (Bliemer and Rose, 2010;Bliemer and Rose, 2011; Bliemer et al., 2009; Rose and Bliemer, 2013). This is a distinct advantage for the proposed choiceexperiment given the small sample size envisaged for this research. Further, efficient designs are less restricted and easier to

Page 5

Table 1Attributes included in the DCE pre-test/pilot design.

Types ofattributes

Attribute definition Details; attribute levels

Conservationservice

Focus is on broad-acre species conservation (as opposed tolocalised hotspots) with flagship species group: birds such asbrolga (Grus rubicunda) and sarus crane (Grus antigone)

Defined relative to cattle grazing: expressed in termsof exclusion of cattle from the area subscribedto biodiversity conservation.3 levels

Opportunity cost and additional management costs defined SHORT exclusion period each year e.g., during nesting seasonwith zero reduction in cattle production from that landand no additional management requiredPROLONGED periods of cattle exclusion, e.g., wetlands duringdry season; riparian areas during wet season resulting in a 50%reduction in cattle production from that land; no ‘additional’management.TOTAL exclusion of cattle resulting in a100% reduction in cattleproduction from that land. Weed and feral animal control to beconducted and burning regime as defined necessary.

Remuneration Annual payment received, $/ha, indexed 5 levels: $2, $4, $8, $12, $16

Contractualconditions

Contract duration 4 levels: 5,10,20,40 years

Flexibility 2 levels:No flexibility, meaning stringent enforcement of contractconditions and/or potential penaltiesFlexibility: Option to ‘suspend’ participation in contracts of 45year duration in ‘exceptional’ circumstances – no paymentreceived, no penalty to be paid; frequency o¼1 in 5 years.

Monitoring 2 levelsExternal monitoring: the administrating agency undertakesregular monitoring or contracts an independent provider toundertake the monitoring.Self: the pastoralists provides the monitoring but random spot-checks are conducted by the administrating body to safeguardcontractual compliance and ensure validity of monitoringresults.

Sectorprovidingthe funding

Funding source 3 levelsGovernment: taxpayer funded programme;Corporate sector, e.g., as part of an off-set programmePhilanthropic sector.

R. Greiner et al. / Journal of Choice Modelling 10 (2014) 34–4538

find than orthogonal designs, often enabling much smaller designs in terms of the number of choice sets. In this paper wewill use the D-error criterion to optimise the efficiency of the experimental design.

The aim is to estimate a random parameter logit (RPL) model, in which the random parameters describe heterogeneity inpreferences. In stated choice surveys where respondents face multiple choice tasks, correlations between choices need to be takeninto account by considering the choices as panel data (Revelt and Train, 1998). As stated in Bliemer and Rose (2010), generatingefficient designs for the panel RPL model is extremely complex and mostly infeasible. However, they find that an efficient design forthe multinomial logit (MNL) model is often also efficient for estimating the panel RPL model. Therefore, we will generate anefficient design assuming an MNL model and use this design to create a survey for estimating an RPL model in the pilot study.

In order to generate an efficient design, priors (best guesses for the unknown parameters) are needed. The priors weredeveloped using a best-practice sequential process. The initial step, literature review and industry consultation, wasdescribed in the previous section of this paper. The second step involved the specification of an initial efficient DCE design,which would subsequently (third step) be pre-tested/piloted so as to validate the design in principle. In a fourth step, datafrom the pre-test would be analysed and resulting parameter estimates would be used as priors to inform an improved(more efficient) DCE design for the final survey.

Choice sets for the pre-test and pilot survey were developed on the basis of priors gleaned from the literature and focusgroup meetings with pastoralists (see above). Priors βk for parameters k were defined as Bayesian prior distributions,assuming a normal distribution of parameter value with a mean value μ̂k and standard deviation 6̂k so that βk �Nðμ̂k; 6̂

2k Þ.

The use of Bayesian priors, as introduced by Sándor and Wedel (2001), takes uncertainty about the prior parameter valuesinto account and therefore leads to a more robust efficient design.

The initial panel DCE design was subjected to a pre-test with industry experts (who answered the choice questions only)and a pilot survey with pastoralists (who completed the entire survey). The purpose of this approach was to generate asufficiently large preliminary choice data set from which to derive the efficient design for the pastoralist survey. To thateffect, a large number of choice tasks (36; 3 blocks of 12) was generated in Ngene, using the μ̂k and 6̂k values shown in the

Page 6

Table 2Attribute priors employed for Bayesian efficient DCE design, for pretest/pilot and full survey.

Attribute Pilot: initial prior estimates from literature and focus

group meetings as defined by μ̂ and 6̂Full survey: revised μ̂ and 6̂ obtained by RPL modellingof pre-test and pilot DCE responses

Conservationa,b requirement μ̂C-LONG¼�0.8, 6̂¼0.3 μ̂ C-LONG¼�0.7n, 6̂ ,¼0.4

μ̂C-TOTAL¼�1.6, 6̂¼0.8 μ̂C-TOTAL¼�1.6n, 6̂¼0.8Annual payment μ̂Pay¼0.2, 6̂¼0.1 μ̂Pay¼0.33nnn, 6̂¼0.1Contract length μ̂Years¼�0.05, 6̂¼0.03 μ̂Years¼�0.18nn ,6̂¼0.08Flexibilitya μ̂Flex¼0.6, 6̂¼0.4 μ̂Flex¼2.4nnn, 6̂¼0.9Monitoringa μ̂Mont¼0.4, 6̂¼0.3 μ̂Mont¼0.6, 6̂¼0.5

Attribute ‘Funding source’ not shown as eliminated from DCE after pre-test.a dummy variable,b effects variable in revised design.n ¼significant at po0.1,nn ¼significant at po0.05,nnn ¼significant at po0.01.

R. Greiner et al. / Journal of Choice Modelling 10 (2014) 34–45 39

second column of Table 2. The pre-test of the choice design was conducted in mid November 2012. Six north Australiangrazing industry experts from across Western Australia, the NT and Queensland, mostly with a background in extension,were asked to complete all 36 choice tasks. The pilot survey was conducted with seven pastoralists in Queensland and theNT in late November 2012. Pilot survey respondents completed the entire questionnaire, including a choice experimentconsisting of one block of 12 choice tasks. The purpose of the pre-test was threefold, (i) to review the survey instrument inits totality and identify any issues with comprehension or completeness, (ii) to contribute preliminary choice data foranalysis, and (iii) to review the choice experimental task complexity and cognitive burden for respondents and thepossibility of omitted relevant attributes. In QLD, the pilot survey was conducted in a one-on-one situation. In the NT, thepilot survey was conducted in a research meeting context, where respondents first completed the survey on their own andthen participated in a discussion about the methodology. A discussion of the survey instrument at the end of the meetingsfound that while respondents thought the survey was long (average time of completion 2.5 h), they also considered therealm of questions relevant but perceived the number of choice tasks to be excessive. There were no omitted variables.To the contrary, respondents were found to build their choices around four attributes in particular, namely conservationservice, remuneration, contract duration and flexibility. Source of funding and monitoring arrangements did not explicitlyfeature in the reasons given by respondents for choices made.

The pre-test and pilot DCE results were combined and analysed using both an MNL model and an RPL model. There are keyassumptions underlying the MNL model, namely (i) IID—ie. that unobserved effects are ‘extreme value 1’ distributed, independentand identically distributed, (ii) independence of observed choices and (iii) homogeneity of preferences (Hensher et al., 2005). RPL isa mixed multinomial logit model, which relaxes these assumptions, thus taking into account heterogeneity of the parameter valuesamong respondents (Hensher et al., 2005; Mariel et al., 2013; Marsh, 2012; Train, 1998). Marsh (2012) argues that RPL models havea number of benefits over MNL: they provide flexibility, are behaviourally more appropriate, provide the analyst with informationabout heterogeneity in the data while estimating unbiased parameter estimates. However, the estimation of a RPL model requiresthe specification of not only the parameters to include in the model, but also which parameters to treat as random parameters, thedistributional form(s), the type of draw and number of draws to use in the estimation, and, what correlations are to be consideredbetween parameters (Hensher et al., 2005).

Both models were run in NLOGIT 5 software (Econometric_Software_Inc, 2012). The RPL model delivered more compellingattribute parameter estimates. The RPL model used 100 Halton draws with model parameters assumed to be independent andrandom within a normal distribution for the non-qualitative attributes ‘remuneration’, ‘contract duration’ and ‘flexibility’. The RPLresults confirmed the direction of the attribute influence in all cases and the magnitude of prior in all cases except ‘flexibility’, and‘contract length’, both of which were found to have larger parameter values than anticipated. Based on the RPL parameterestimates, we have generated a Bayesian D-efficient design assuming the MNL model. The parameter means are used as priors forμ̂k and the standard errors of these means are used as 6̂k for the Bayesian priors βk �Nðμ̂k; 6̂

2k Þ. Although we generated an efficient

design for the MNLmodel, we inspected the efficiency under the panel RPL assumption to also verify efficiency for this model. Boththe pre-test/pilot and revised Bayesian design parameters are shown in Table 2.

Other changes to the choice experimental design specifications were the omission of ‘funding source’ as an attribute asneither the MNL nor RPL models detected preferences for any of the three alternatives and widening of the payment levelsto include $1, 2, 4, 8, 16 and 32. Widening the level range has been shown to have a significant positive influence on theefficiency of the design and therefore on the reliability of the parameter estimates (Rose and Bliemer, 2013). These decisionswere supported by respondent feedback volunteered during the survey pilot and the data analytical results. ‘Monitoring’was shown to be significant at po0.1 in the MNL model and was retained. Attributes included in the final CDE design areshown in Table 3.

All choice tasks were created in Ngene 1.1.1 (ChoiceMetrics, 2012) using a Bayesian D-efficient design approach (Sándorand Wedel, 2001). The design was generated without accounting for covariate effects. A constant representing the ‘none’

Page 7

Table 3Attributes included in the final DCE design.

Types of attributes Attribute definition Details; attribute levels

Conservation service As per Table 1 As per Table 1Remuneration As per Table 1 6 levels: $1, $2, $4, $8, $16, $32Contractual conditions Contract duration As per Table 1

Flexibility As per Table 1Monitoring As per Table 1

R. Greiner et al. / Journal of Choice Modelling 10 (2014) 34–4540

option was included in the design. Dominant and redundant choice tasks were prevented by defining design restrictions.The Modified Federov algorithm was used as an algorithm to generate an efficient design, which will not necessarily beattribute-level balanced.

The number of choice tasks answered by each respondent has been shown to impact on error variances (Hensher, 2006)but the effect is small compared to other design dimensions (Caussade et al., 2005). The question is what number of choicetasks is sufficient from a statistical design perspective. The general guideline is that there needs to be at least as much data(information) points as there are parameters to estimate. The number of parameters in our model is 13, taking all dummycoded variables into account. The amount of data from a single observation depends on the number of choice alternatives, J.If there are two alternatives, then a choice provides one data point. A choice among three alternatives provides two datapoints, knowing that the chosen alternative is preferred over both non-chosen alternatives. In general, having S choicetasks, the number of data points is Sn(J-1) (see Rose and Bliemer, 2013). In our case, we have four alternatives, we have that5n(4-1)¼15413, we need a minimum of S¼5 choice tasks. For our final survey we generate a design with 24 choice tasks,which is well above this statistical minimum. The experiment was blocked into four versions of six choice tasks, i.e. eachrespondent would answer six choice tasks. The global level of efficiency of the design is commonly expressed as theBayesian D-error, which minimises the determinant of variance-covariance matrix and hence the standard errors of theparameters in estimation. The smaller the Bayesian D-error, the more statistically efficient is the design. The Bayesian D-error for the final design is 0.0716.

3.4. Survey administration

The choice experiment forms the key part of a survey of graziers, which explores structural and financial aspects of thebusiness, land management system, cattle enterprise, risk attitudes and management, environmental attitudes andmanagement, and personal and family circumstances. To maximise opportunities for pastoralists to participate in theresearch and thereby maximise response rate and minimise participation bias of the sample (Wagner, 2012), the survey isadministered (i) in a face-to-face situation during visits by the lead researcher on pastoral properties and (ii) during researchmeetings with groups of pastoralists. Mode of participation depends on the preference of research respondents. In situationswhere several persons from the same pastoral property are present during the completion of the survey, the key decisionmaker is asked to apply the usual approach to decision making, which might mean the other attendees are consulted tovarious degrees or only get to watch and listen. Research meetings and on-farm visits take approximately 2–2.5 hours andrespondents receive an AUD 200 gratuity.

Station visits are arranged by telephone and due to the vast distances between stations (cattle stations can be tenthousands of square kilometres in size) are arranged along a travel path, which is often associated with the location andtiming of an industry event. All station managers who are prepared to participate in the survey and available at a timematching the travel itinerary are interviewed.

Research meetings are organised to value-add to industry events including branch meetings of industry associations orcommunity events organised by Landcare or regional natural resource management groups. To ensure integrity of thequantitative data in terms of independence of responses, meetings are moderated to be the antithesis to focus group discussions.Influence by the meeting moderator is limited to that of the interviewer in a face-to-face situation. Possible interaction betweenrespondents in research meetings is minimised by (i) seating arrangements—seating respondents physically apart—and (ii)moderation, whereby after the initial explanation of the research and the choice experiment in particular, and shared questionsof qualification, respondents complete the survey in their own time. The role of the moderator at this stage is limited tosupervising the setting and individually providing clarification of questions when required.

4. Additional choice experimental design dimensions

4.1. Attribute attendance

When completing choice experimental tasks, respondents often do not consider all attributes presented in the tasks butmake choices on only a sub-set of attributes. Not accounting for this ‘attribute non-attendance’ can cause biased parameterestimates and hence biased estimates of willingness to accept estimates (Hensher, 2007) as ignoring attributes indicates therespondent will not react and no improvement in attribute level will compensate the respondent even with worsening of

Page 8

R. Greiner et al. / Journal of Choice Modelling 10 (2014) 34–45 41

other attribute levels (Hensher and Rose, 2009), thus marginal rates of substitution between attributes cannot be reliablycalculated (Kragt, 2013). Accounting for attribute non-attendance could be “more important than unobserved heterogeneity,or at least explain much of it” (Scarpa et al., 2013, p. 177).

There are two principal approaches to estimating attribute non-attendance, by getting respondents to state theirattendance or by inferring it from the data (Hensher and Collins, 2011; Kragt, 2013; Scarpa et al., 2013). The statedmethod requires inclusion in the questionnaire of a supplementary question which asks respondents to indicatewhether they ignored different attributes in the choice tasks (e.g. Carlsson et al., 2010). Inferring attribute non-attendance is done by observing the response pattern. E.g. Scarpa et al. (2009) suggest a panel mixed logit model basedon a latent class structure which will detect respondents who belong to latent classes with zero utility weights forselected attributes.

This survey includes a stated approach. Respondents are asked how strongly they considered each attribute on a 5-pointresponse scale ranging from 1¼never to 5¼always. They are also asked to summarise any decision making heuristic theymay have applied in arriving at their preferred choices. This does not preclude the application of an inferred approach,which according to the recent literature may deliver superior estimates of attribute non-attendance due to respondent biasin the stated responses (Kragt, 2013; Scarpa et al., 2013).

4.2. Response certainty

Choice experiments give respondents hypothetical choice tasks and one critical question is as to the consistency betweenhypothetical and real choices, or ‘certainty’ (Hensher et al., 2012). Choice modelling typically assumes that hypothetical andreal choices are identical. Increasingly, choice experiments add a supplementary question after each choice task to obtain acertainty response is one way of accounting for the risk that one might attach to the choice of an alternative in a choiceexperiment. A supplementary question asks the extent to which a respondent is certain of his/her stated choice on a scalefrom 0 (very unsure) to 10 (very sure; Brouwer et al., 2010; Johannesson, 1999). The certainty score can be used to weightthe choice, and is one way to account for the perceived risk of the alternative (Blamey et al. 2000).

This survey measures choice certainty by including a supplementary question after each choice task, which asksrespondents to report the perceived surety of their discrete choice experiment answer. In particular, respondents are askedto give the perceived likelihood that they would choose this alternative if confronted with the choice in real life on acontinuous response scale from 0 to 100, where 100 means that the respondent is absolutely certain that he/she wouldmake this choice if given the opportunity and 0 is a zero per cent likelihood of the choice being adopted. This provides notonly a way of assessing the extent to which design attribute levels condition the degree of choice certainty, but accountingfor response certainty in the choice model can in some cases significantly improve its predictive power (Hensher et al.,2012) though not necessarily so (Brouwer et al., 2010).

4.3. Identifying protest responses

After completing the choice tasks, respondents who chose the ‘none’ option in all sets as their preferred option are askedto stipulate the reasons for doing so. The intention is to identify protest bids, i.e. situations where respondents do not agreewith the context of the valuation scenario, so their stated zero WTA is unlikely to be a reflection of their true WTA. Theserespondents will be eliminated from the choice data set for analysis (Windle and Rolfe, 2013).

4.4. Discrete continuous approach

This CE conceives the total supply of on-farm biodiversity conservation as a discrete-continuous problem with twoseparate but interdependent components (Hanemann, 1984). The first component is the biodiversity action choice made bypastoralists and the second component estimates the land subscribed to those choices. The first component is representedby the discrete choice experiment (DCE) which is the subject of this paper. It determines the weights that farmers attributeto various attributes of conservation contracts. The second component of the CE asks respondents to indicate how muchland they would supply if the chosen alternative was realized. Answers to both questions are needed to determine theaggregate supply function of on-farm biodiversity conservation services.

DCEs operate at the disaggregate level, i.e. the decision maker level, where choice behaviour can be defined in terms ofcommodity/service qualities or attributes (Truong and Hensher, 2012). They are not suited for describing aggregateconsumer demand and producer supply where the quantity decision is separate from the product choice. In such situations,however, the results of DCEs can serve as building blocks of aggregate demand estimation models.

Discrete-continuous approaches have been applied variously to study consumer behaviour e.g. to model the demand byhouseholds for energy (Buckley et al., 2012; Garrod et al., 2012; Mansur et al., 2005; Vaage, 2000), water (Olmstead et al.,2007) and alternative fuel vehicles (Ahn et al., 2008). The aggregate (second) component can use different types of methods,e.g. simulation models or computable general equilibrium models.

Examples in the agricultural-environmental literature which utilise discrete continuous choice include Lohr and Park(Lohr and Park, 1994; 1995) who explore farmers participation in a filter strip programme with the continuous choice beinghow many acres to plant. They found that willingness to accept payments and acres planted was not uniform across

Page 9

Table 4Illustration of a choice task.

R. Greiner et al. / Journal of Choice Modelling 10 (2014) 34–4542

locations. Lynch et al. (2002) studied landowners’willingness to plant or increase the size of streamside buffers and the totalarea they would plant. Significant variables investigated included land already planted with buffers, knowledge of buffersand previous participation in government programs. Lambert et al. (2007) examined landowner’s willingness to enrol in aconservation reserve programme. The discrete component compared programme components such as land retirement orworking land projects and participation in conservation reserve programs and the continuous component involved landenroled. Participation was dependent upon factors such as farm structure, personal attributes, farming experience andenvironmental factors.

The continuous element of the choice experiment is captured by the question, how much land the respondent wouldseek to subscribe to a scheme of the type captured by the chosen alternative. This information supports estimation of thetotal conservation area conditional on the discrete choices made.

Page 10

R. Greiner et al. / Journal of Choice Modelling 10 (2014) 34–45 43

There is also a supplementary question asking what type of land this is, so as to be able to ascertain the conservationvalue of the land on offer. Landowners make choices between competing land uses which depend on differences in landquality (Krosnick and Fabrigar 1997). Biodiversity and production values of different land types of each farm are explored inother survey questions.

Infrastructure costs, including fencing and additional stock watering points, can constitute a large share of biodiversityconservation expenses. They are excluded from the discrete choice experiment as they are dependent on the farm-individual situation such as current infrastructure, and the specific area of land the respondent intends to subscribe.To enable an estimation of the infrastructure cost component of on-farm biodiversity conservation to the funder of aprogramme, respondents were asked to provide estimates of new infrastructure required. One of the resulting choicesituations is shown in Table 4.

5. Concluding comments

This paper illustrates the various considerations involved in designing a discrete choice experiment, which seeks todetermine under what conditions north Australian pastoralists and graziers are willing to participate in voluntary on-farmbiodiversity conservation programs. This paper describes how a sequential process with different stages of industryconsultation and participation was used to derive at a Bayesian D-efficient design, and how matters of attribute attendanceand response certainty were addressed in the choice task. The resulting data set will be analysed using panel RPL and latentclass models and deliver important insights into how pastoralists perceive programme attribute trade-offs and howrespondent attributes influence preferences. The results from the DCE will also inform the continuous component of thediscrete-continuous approach, supporting the estimation of potential supply curves of agricultural land for biodiversityconservation purposes in different regions of northern Australia and under different conditions.

Acknowledgements

We had valuable discussions with John Rolfe, Jill Windle and Daniel Gregg about DCE design and with John Woinarskiand Mike Lawes relating to the ecological dimensions of conservation options. The research is funded by the NationalEnvironmental Research Programme – North Australia Hub. It is being undertaken in collaboration with the NorthernTerritory Cattlemen’s Association, the Pastoralists and Graziers Association of Western Australia, regional NRM groups acrossnorthern Australia and Landcare groups. We are grateful for the collaboration of staff and members of the Northern TerritoryCattlemen’s Association and the Northern Gulf Resource Management Group during the design, pre-test and piloting phases,and the feedback of participating pastoralists and graziers. A number of government grazing systems researchers, extensionofficers and private industry advisers from Western Australia, the NT and Queensland participated in the DCE pre-test. Anearlier version of this paper was presented at the 2013 conference of the Australian Agricultural and Resource EconomicsSociety. We are indebted to two anonymous referees for their insightful comments.

References

ABARES, 2012. AGSURF Data. Australian Government, Department of Agriculture, Fisheries and Forestry, ABARES, Canberra.ABS, 2011a. Data Cube: Land Management Practices – Environmental Conservation by NRM Region. In: Statistics, A.B.o. (Ed.), Canberra.ABS, 2011b. Land Management and Farming in Australia 2009–10. Australian Bureau of Statistics, Canberra.Adamowicz, V., Dupont, D., Krupnick, A., 2005. Willingness to pay to reduce community health risks from municipal drinking water: a stated preference

study, Working Paper. Report written for the United Stated Environmental Protection Agency. University of Alberta, Edmonton, CA.Ahn, J., Jeong, G., Kim, Y., 2008. A forecast of household ownership and use of alternative fuel vehicles: a multiple discrete-continuous choice approach.

Energy Econ. 30, 2091–2104.Barkmann, J., Glenk, K., Keil, A., Leemhuis, C., Dietrich, N., Gerold, G., Marggraf, R., 2008. Confronting unfamiliarity with ecosystem functions: the case for an

ecosystem service approach to environmental valuation with stated preference methods. Ecol. Econ. 65, 48–62.Barry, P.J., Sherrick, B.J., Zhao, J., 2009. Integration of VaR and expected utility under departures from normality. Agr. Econ. 40, 691–699.Beharry-Borg, N., Smart, J.C.R., Termansen, M., Hubacek, K., 2013. Evaluating farmers’ likely participation in a payment programme for water quality

protection in the UK uplands. Reg. Environ. Change 13, 633–647.Bliemer, M.C.J., Rose, J.M., 2010. Construction of experimental designs for mixed logit models allowing for correlation across choice observations. Transp.

Res. Part B: Methodol. 44, 720–734.Bliemer, M.C.J., Rose, J.M., 2011. Experimental design influences on stated choice outputs: an empirical study in air travel choice. Transp. Res. Part A: Policy

Pract. 45, 63–79.Bliemer, M.C.J., Rose, J.M., Hensher, D.A., 2009. Efficient stated choice experiments for estimating nested logit models. Transp. Res. Part B: Methodol. 43, 19–35.Bond, C.A., Hoag, D.L., Kipperberg, G., 2011. Agricultural producers and the environment: A stated preference analysis of Colorado corn producers. Can.

J. Agr. Econ./Rev. Can. Agroecon. 59, 127–144.Bortolussi, G., McIvor, J., Hodgkinson, J., Coffey, S., Holmes, C., 2005. The northern Australian beef industry, a snapshot. 1. Regional enterprise activity and

structure. Aust. J. Exp. Agr. 45, 1057–1073.Broch, S.W., Strange, N., Jacobsen, J.B., Wilson, K.A., 2013. Farmers’ willingness to provide ecosystem services and effects of their spatial distribution. Ecol.

Econ. 92, 78–86.Broch, S.W., Vedel, S.E., 2012. Using choice experiments to investigate the policy relevance of heterogeneity in farmer agri-environmental contract

preferences. Environ. Resource Econ. 51, 561–581.Brouwer, R., Dekker, T., Rolfe, J., Windle, J., 2010. Choice certainty and consistency in repeated choice experiments. Environ. Resource Econ. 46, 93–109.Buckley, C., Hynes, S., Mechan, S., 2012. Supply of an ecosystem service – farmers’ willingness to adopt riparian buffer zones in agricultural catchments.

Environ. Sci. Policy 24, 101–109.

Page 11

R. Greiner et al. / Journal of Choice Modelling 10 (2014) 34–4544

Burton, M., 2010. Inducing strategic bias and its implications for choice modelling design. Research Report No. 61. University of Western Australia, Perth.Carlsson, F., Kataria, M., Lampi, E., 2010. Dealing with ignored attributes in choice experiments on valuation of Sweden’s environmental quality objectives.

Environ. Resource Econ. 47, 65–89.Carson, R.T., Flores, N.E., Meade, N.F., 2001. Contingent valuation: controversies and evidence. Environ. Resource Econ. 19, 173–210.Caussade, S., Ortúzar, J.D., Rizzi, L.I., Hensher, D.A., 2005. Assessing the influence of design dimensions on stated choice experiment estimates. Transp. Res.

Part B: Methodol. 39, 621–640.ChoiceMetrics, 2012. Ngene 1.1.1, Australia.Christensen, T., Pedersen, A.B., Nielsen, H.O., Mørkbak, M.R., Hasler, B., Denver, S., 2011. Determinants of farmers’ willingness to participate in subsidy

schemes for pesticide-free buffer zones-A choice experiment study. Ecol. Econ. 70, 1558–1564.de Bekker-Grob, E.W., 2009. Discrete choice experiemnts in health care: theory and applications. Erasmus University, Rotterdam.Econometric_Software_Inc, 2012. NLOGIT 5, Australia.Espinosa-Goded, M., Barreiro-Hurlé, J., Ruto, E., 2010. What do farmers want from agri-environmental scheme design? A choice experiment approach.

J. Agr. Econ. 61, 259–273.Flynn, T.N., Louviere, J.J., Peters, T.J., Coast, J., 2007. Best-worst scaling: what it can do for health care research and how to do it. J. Health Econ. 26, 171–189.Garrod, G., Ruto, E., Willis, K., Powe, N., 2012. Heterogeneity of preferences for the benefits of environmental stewardship: a latent-class approach. Ecol.

Econ. 76, 104–111.Goodman, S.P., Lockshin, L., Cohen, E., 2005. Best-worst scaling: a simple method to determine drinks and wine style preferences.Greiner, R., Gregg, D., 2011. Farmers’ intrinsic motivations, barriers to the adoption of conservation practices and effectiveness of policy instruments:

empirical evidence from northern Australia. Land Use Policy 28, 257–265.Greiner, R., Patterson, L., Miller, O., 2009. Motivations, risk perceptions and adoption of conservation practices by farmers. Agr. Syst. 99, 86–104.Grutters, J.P.C., Kessels, A.G.H., Dirksen, C.D., van Helvoort-Postulart, D., Anteunis, L.J.C., Joore, M.A., 2008. Willingness to accept versus willingness to pay in

a discrete choice experiment. Value Health 11, 1110–1119.Hajkowicz, S., 2009. The evolution of Australia’s natural resource management programs: towards improved targeting and evaluation of investments. Land

Use Policy 26, 471–478.Hanemann, W.M., 1984. Discrete/continuous models of consumer demand. Econometrica 52, 541–561.Hensher, D.A., 2006. How do respondents process stated choice experiments? Attribute consideration under varying information load. J. Appl. Econom. 21,

861–878.Hensher, D.A., 2007. Attribute processing in choice experiments and implications on willingness to pay. In: Kanninen, B.J. (Ed.), Valuing Environmental

Amenities Using Stated Choice Studies, Springer, Dordrecht, pp. 135–157.Hensher, D.A., Collins, A.T., 2011. Interrogation of responses to stated choice experiments: is there sense in what respondents tell us? A closer look at what

respondents choose and process heuristics used in stated choice experiments. J. Choice Model., 4; 62–89.Hensher, D.A., Rose, J.M., 2009. Simplifying choice through attribute preservation or non-attendance: implications for willingness to pay. Transp. Res. Part

E: Logist. Transp. Rev. 45, 583–590.Hensher, D.A., Rose, J.M., Beck, M.J., 2012. Are there specific design elements of choice experiments and types of people that influence choice response

certainty? J. Choice Model. 5, 77–97.Hensher, D.A., Rose, J.M., Greene, W.H., 2005. Applied Choice Analysis: A Primer. Cambridge University Press, Cambridge, UK.Horne, P., 2006. Forest owners’ acceptance of incentive based policy instruments in forest biodiversity conservation – a choice experiment based approach.

Silva Fenn. 40, 169.Horowitz, J.K., McConnell, K.E., 2002. A review of WTA/WTP studies. J. Environ. Econ. Manag. 44, 426–447.Hoyos, D., 2010. The state of the art of environmental valuation with discrete choice experiments. Ecol. Econ. 69, 1595–1603.Johannesson, M., 1999. Calibrating hypothetical willingness to pay responses. J. Risk Uncertain. 18, 21–32.Kaczan, D., Swallow, B.M., Adamowicz, W.L.V., 2013. Designing payments for ecosystem services (PES) programme to reduce deforestation in Tanzania: an

assessment of payment approaches. Ecol. Econ. 95, 20–30.Klojgaard, M., Bech, M., Sogaard, R., 2012. Designing a CDE: the value of a qualitative process. J. Choice Model. 5, 1–18.Kragt, M.E., 2013. Stated and inferred attribute attendance models: a comparison with environmental choice experiments. J. Agr. Econ. 64, 719–736.Lambert, D.M., Sullivan, P., Claassen, R., Foreman, L., 2007. Profiles of US farm households adopting conservation-compatible practices. Land Use Policy 24,

72–88.Lancsar, E., Louviere, J., Donaldson, C., Currie, G., Burgess, L., 2013. Best worst discrete choice experiments in health: methods and an application. Soc. Sci.

Med. 76, 74–82.Lankester, A.J., 2013. Conceptual and operational understanding of learning for sustainability: a case study of the beef industry in north-eastern Australia.

J. Environ. Manag. 119, 182–193.Lewis, D., 2002. Slower Than The Eye Can See: Environmental Change In Northern Australia’s cattle lands. Tropical Savannas CRC, Darwin.Lin, W., Dean, G., Moore, C., 1974. An empirical test of utility vs. profit maximization in agricultural production. Am. J. Agr. Econ. 56, 497–508.Lohr, L., Park, T., 1994. Discrete/continuous choices in contingent valuation surveys: soil conservation decisions in Michigan. Rev. Agr. Econ., 16; 1–15.Lohr, L., Park, T.A., 1995. Utility-consistent discrete-continuous choices in soil conservation. Land Econ. 71, 474–490.Louviere, J.J., Hensher, D., Swait, J.D., 2000. Stated choice methods: analysis and application. Cambridge University Press, Cambridge.Louviere, J.J., Hensher, D.A., 1982. Design and analysis of simulated choice or allocation experiments in travel choice modeling. Transp. Res. Rec..Louviere, J.J., Woodworth, G., 1983. Design and analysis of simulated consumer choice or allocation experiments: an approach based on aggregate data.

J. Market. Res., 350–367.Lusk, J.L., Hudson, D., 2004. Willingness-to-pay estimates and their relevance to agribusiness decision making. Appl. Econ. Perspect. Policy 26, 152–169.Lynch, L., Hardie, I., Parker, D., 2002. Analyzing agricultural landowners’ willingness to install streamside buffers, Working Paper No. 02-01. Department of

Agricultural and Resource Economics. The University of Maryland, College Park. p. 35.Mansur, E.T., Mendelsohn, R., Morrison, W., 2005. A discrete-continuous choice model of climate change impacts on energy. Technical Paper, 40.Mariel, P., Ayala, A.d., Hoyos, D., Abdullah, S., 2013. Selecting random parameters in discrete choice experiment for environmental valuation: a simulation

experiment. J. Choice Model. 7, 44–57.Marley, A.A.J., Louviere, J.J., 2005. Some probabilistic models of best, worst, and best–worst choices. J. Math. Psychol. 49, 464–480.Marsh, D., 2012. Water resource management in New Zealand: Jobs or algal blooms? J. Environ. Manag. 109, 33–42.McAllister, R.R.J., Gordon, I.J., Janssen, M.A., Abel, N., 2006. Pastoralists’ responses to variation of rangeland resources in time and space. Ecol. Appl. 16,

572–583.Mitchell, R.C., Carson, R.T., 1989. Using surveys to value public goods: the contingent valuation method. Resources for the Future, Washington DC.Morrison, M., Bennett, J., Blamey, R., Louviere, J., 2002. Choice modeling and tests of benefit transfer. Am. J. Agr. Econ. 84, 161–170.Olmstead, S.M., Michael Hanemann, W., Stavins, R.N., 2007. Water demand under alternative price structures. J. Environ. Econ. Manag. 54, 181–198.Peterson, J.M., 2011. Buying environmental improvements from farmers who screen on contract attributes, AERE Inaugural Summer Conference, Seattle, p. 17.Peterson, J.M., Smith, C.M., Leatherman, J.C., Hendricks, N.P., Fox, J.A., 2011. The role of contract attributes in purchasing environmental services from

landowners. Kansas State University, Kansas.Potoglou, D., Burge, P., Flynn, T., Netten, A., Malley, J., Forder, J., Brazier, J.E., 2011. Best–worst scaling vs. discrete choice experiments: an empirical

comparison using social care data. Soc. Sci. Med. 72, 1717–1727.Revelt, D., Train, K., 1998. Mixed logit with repeated choices: Households’ choices of appliance efficiency level. Rev. Econ. Stat. 80, 647–657.

Page 12

R. Greiner et al. / Journal of Choice Modelling 10 (2014) 34–45 45

Robison, L.J., 1982. An appraisal of expected utility hypothesis tests constructed from responses to hypothetical questions and experimental choices.Am. J. Agr. Econ. 64, 367–375.

Rolfe, J., Alam, K., Windle, J., Whitten, S., 2004. Designing the choice modelling survey instrument for establishing riparian buffers in the Fitzroy Basin(Establishing the Potential for Offset Trading in the Lower Fitzroy River Research Report No. 3). Central Queensland University, Emerald, QLD.

Rolfe, J., Bennett, J., 2009. The impact of offering two versus three alternatives in choice modelling experiments. Ecol. Econ. 68, 1140–1148.Rolfe, J., Windle, J., 2005. Establishing the potential for offset trading in the Lower Fitzroy River. Research Report No. 6. Final report for MBI project 53.

Central Queensland University, Emerald, Qld.Rolfe, J., Windle, J., Bennett, J.W., 2008. Designing Choice Experiments to Incorporate Tests for Geographic Scale and Scope Differences. Crawford Institute,

Australian National University, Canberra.Rose, J.M., Bliemer, M.C.J., 2013. Sample size requirements for stated choice experiments. Transportation 40, 1021–1041.Ruto, E., Garrod, G., 2009. Investigating farmers’ preferences for the design of agri-environment schemes: a choice experiment approach. J. Environ. Plan.

Manag. 52, 631–647.Ryan, M., Gerard, K., Amaya-Amaya, M., 2008. Discrete choice experiments in a nutshell. In: Ryan, M., Gerard, K., Amaya-Amaya, M. (Eds.), Using Discrete

Choice Experiments to Value Health and Health Care, Springer, Amsterdam, pp. 13–46.Sándor, Z., Wedel, M., 2001. Designing conjoint choice experiments using managers’ prior beliefs. J. Market. Res. 38, 430–444.Scarpa, R., Gilbride, T.J., Campbell, D., Hensher, D.A., 2009. Modelling attribute non-attendance in choice experiments for rural landscape valuation. Eur. Rev.

Agr. Econ. 36, 151–174.Scarpa, R., Zanoli, R., Bruschi, V., Naspetti, S., 2013. Inferred and stated attribute non-attendance in food choice experiments. Am. J. Agr. Econ. 95, 165–180.Schläpfer, F., Fischhoff, B., 2012. Task familiarity and contextual cues predict hypothetical bias in a meta-analysis of stated preference studies. Ecol. Econ. 81, 44–47.Train, K.E., 1998. Recreation demand models with taste differences over people. Land Econ., 230–239.Truong, T., Hensher, D., 2012. Linking discrete choice to continuous demand within the framework of a computable general equilibrium model for the

analysis of wider economic impacts of transport investment projects. Transp. Res. Part B: Methodol. 46, 1177–1201.Vaage, K., 2000. Heating technology and energy use: a discrete/continuous choice approach to Norwegian household energy demand. Energy Econ. 22,

649–666.Vedel, S.E., Jacobsen, J.B., Thorsen, B.J., 2010. Contracts for environmental goods and the role of monitoring for landowners’ willingness to accept, Fourth

World Congress of Environmental and Resource Economists, Montreal.Wagner, J., 2012. A comparison of alternative indicators for the risk of nonresponse bias. Public Opin. Q. 76, 555–575.Windle, J., Rolfe, J., 2005. Diversification choices in agriculture: a Choice Modelling case study of sugarcane growers. Aust. J. Agr. Resource Econ. 49, 63–74.Windle, J., Rolfe, J., 2013. Assessing the tradeoffs of increased mining activity in the Surat Basin, Queensland. Preferences of Brisbane residents using non-

market valuation techniques. Aust. J. Agr. Res. Econ.. (In press).Windle, J., Rolfe, J., Whitten, S.M., Alam, K., 2005. Establishing the potential for offset trading in the Lower Fitzroy River. Research Report No. 5. Using Choice

Modelling to establish the supply of riparian services. Central Queensland University, Emerald, Qld.Woinarski, J.C.Z., Mackey, B., Nix, H., Traill, B., 2007. The Nature of Northern Australia: Natural Values, Ecological Processes And Future Prospects. ANU E

Press, Canberra.Yu, J., Belcher, K., 2011. An economic analysis of landowners’ willingness to adopt wetland and riparian conservation management. Can. J. Agr. Econ./Rev.

Can. Agroecon. 59, 207–222.