Elsevier

Journal of Choice Modelling

Volume 29, December 2018, Pages 17-32
Journal of Choice Modelling

Testing the effect of changes in elicitation format, payment vehicle and bid range on the hypothetical bias for moral goods

https://doi.org/10.1016/j.jocm.2018.08.001Get rights and content

Highlights

  • We systematically investigate how DCE design influence preference structure and welfare measures, in the context of a moral public good. The design factors varied are elicitation format, payment vehicle and bid range.

  • Data spans seven samples collected in three main data collections. Two of these data collections are based on a standard, hypothetical DCE design, whereas a third is based on an incentivized DCE, using a real donation mechanism.

  • Depending on the characteristics of the good, we find that elicitation format, payment vehicle and bid range may influence preference structure and welfare estimates.

  • In many cases the effect of payment vehicle is negligible.

  • In the context of moral public goods, both stated and revealed welfare measures are likely to be sensitive to survey design.

Abstract

This paper explores how changes in survey design influence the conclusions reached from discrete choice models, a topic which is of particular interest in the context of stated and revealed preference comparisons investigating potential hypothetical bias. We systematically test the WTP of a good with no related market value, using two standard, hypothetical stated preference data collections and an incentivized stated preference data collection, using a real donation mechanism. The investigations into the nature of hypothetical bias typically involve changes in more than just the elicitation format. Therefore, we explicitly test the importance of changes in bid range, payment vehicle, and elicitation format upon the estimated hypothetical bias, while keeping the survey context constant. Our results show that depending on the characteristics of the good in question, the choice of payment vehicle, bid range and elicitation format may affect the results. In many cases, the importance of payment vehicle itself is negligible – especially when the good in question is distant to people. The implication of our study is that caution should be applied when conducting stated and revealed preference comparisons in the context of public good with strong moral components, as even very small design decisions may influence the observed WTP disparities.

Introduction

The valuation of non-market goods through the use of stated preference techniques is an established practice, allowing policymakers to infer the value placed on goods such as environmental services when no value can be inferred from market goods and related behaviour. Since the emergence and broad appliance of stated preference methods, many rich scientific discussions have taken place regarding the merits and drawbacks of the method (Kahneman and Knetsch, 1992; Carson and Mitchell, 1995; Carson et al., 1996; Johnston et al., 2017). The most prominent factor identified in this debate is the discussion regarding hypothetical bias, a term used to capture the idea that when faced with no real consequence of their choices, people will tend to overstate their willingness to pay for a certain good (List and Gallet, 2001; Murphy et al., 2005; Hensher, 2010). While for some types of goods it is possible to triangulate the results, for many other goods it is not. These other goods are typically described by a strong non-use component, also often involving morally challenging decisions. This can for example be conserving nature, or ensuring future generations’ living conditions – issues that are indeed discussed and taken into account in policy making. Consequently, if we want to inform policymakers on issues pertinent to policy design, stated preference methods must be applied, despite the existence of this possible bias.

In the literature, the origin of hypothetical bias has been ascribed to several factors. Examples of such factors are the elicitation format and the elicited welfare measure, e.g. willingness-to-pay (WTP) or willingness-to-accept (WTA) (List and Gallet, 2001), the warm glow of giving as termed by Andreoni (1990) which has been subsequently tested in relation to hypothetical bias in stated preference studies (Nunes and Schokkaert, 2003; Nunes et al., 2009; Johansson-Stenman and Svedsäter, 2012), social desirability bias whereby individuals’ valuation of a certain good is influenced by how socially attractive it is to show support (Crowne and Marlowe, 1960; Epley and Dunning, 2000; List et al., 2004; Lusk and Norwood, 2009), and lastly also the type of good being valued, e.g. public or private good (Alpizar et al., 2008; Johansson-Stenman and Svedsäter, 2012). In this study, we explore the role of elicitation format, payment vehicle and bid range, which are all factors that may influence the motivational reasons for hypothetical bias.

We use the term elicitation format to capture whether preferences are elicited through a hypothetical or a revealed preference mechanism. The typical approach used in investigating hypothetical bias has been to compare preference measures from stated and revealed preference (SP-RP) counterparts (List and Gallet, 2001). In this study, we are interested in investigating distributional preferences for climate policies, where no revealed preference measure exists. Our survey design scheme facilitates a partial SP-RP preference comparison, as we elicit preferences and welfare measures for climate policy in both two strictly hypothetical settings and in an incentivized stated preference setting, where individuals are making real donations to the same policies as presented to respondents in the hypothetical settings. This change in elicitation format is the closest approximation to a true stated and revealed preference comparison that is possible in the given context, as the most realistic payment vehicle for the climate policies investigated is a tax, which is a payment vehicle that is difficult to reproduce in a revealed preference context.1 However, in the survey design scheme examined in this paper, not only the elicitation format changes but also the type of payment vehicle and bid range of the cost attribute change, both of which could influence hypothetical bias. Therefore we need to look at the findings in the literature regarding the role of these two additional factors.

A branch of the valuation literature has investigated how a change in the type of payment vehicle in itself might influence inference from choice data, with some papers suggesting that a coercive instrument such as a tax results in higher estimates of willingness-to-pay compared to a voluntary instrument like a donation (Nunes and Schokkaert, 2003; Wiser, 2007). On the other hand, using an instrument like a donation mechanism has been shown to lead to warm glow of giving. While this is indeed a part of the people's utility function, it may not be solely linked to the good in question. As there is no “budget constraint” for this warm glow of giving, it may inflate the value for single studies, and consequently, since the NOAA recommendations (Arrow et al., 1993), it has been suggested that the donation payment vehicle should not be used in the estimation of welfare values. Related to this is also the possible diversion between preferences elicited when being asked to act as ‘homo economicus’ as opposed to ‘homo politicus’ (Nyborg, 2000). Nyborg argues that individuals may have a dual set of preferences, one relating to their personal well-being and one related to their subjective social welfare and that the latter may be the one that dominates when answering WTP question where their role as citizens are dominating. If this is the case, the underlying preferences will differ depending on the payment vehicle – as it emphasizes two different roles the individual may take. Another problematic component in the use of tax as a payment vehicle is that tax aversion may influence the choices made by respondents (Kallbekken and Sælen, 2011), including the possibility that a tax being coercive results in crowding-out of intrinsic motivation to contribute to public goods (Goeschl and Perino, 2012). We explicitly test for the effect of payment vehicle by comparing preferences elicited in subsamples where the only difference is the payment vehicle – a hypothetical tax and a hypothetical donation.

Another parameter that is typically varied in comparisons of SP-RP preference studies is the bid range – not least when the revealed comparison is conducted in a lab where individuals are faced with a real budget constraint, whereby participants for ethical reasons cannot be asked to leave the experiment worse off than when they entered. Consequently, any trade-off they are asked to make has to be made with the experimentally provided money (Kagel and Roth, 1995). Previous studies examining the effect of varying the level of the cost attribute in CE studies have typically found higher levels of WTP when the levels of the cost attribute increased (Carlsson and Martinsson, 2008; Ladenburg and Olsen, 2008; Mørkbak et al., 2010), but the literature also contains examples of no difference in WTP (Hanley et al., 2005; Kragt, 2013). Kragt (2013) proposes that the result of no difference in WTP when introducing an increased cost vector could be due to the presence of hypothetical bias, while also mentioning that a high level of the cost attribute might affect the unobserved error variance, leading to higher levels of variance. This effect on error variance has also been proposed by Mørkbak et al. (2010), but Kragt (2013) argues that the effect could also be reversed, in that high costs could lead to more certainty in choice, thus reducing the error variance. As in Kragt (2013), we vary the level of the cost attribute between samples, but the change in bid range in our setting is 10-fold, whereas it is only a 50% increase for the highest bid in her study. Thereby we push the bid range comparison into the extreme.

We keep the survey context constant when investigating the effects of varying elicitation format, payment vehicle and bid range. Respondents are asked to value climate policies, with distributional outcomes manifesting in the far future as well as in present time, across three specific regions of the world. The future distributional outcomes are described in terms of changes to the average income level as a result of lower impacts of climate change resulting from additional investments in climate policy. The present-time outcome is the provision of the co-benefit of cleaner air, brought about by increased climate mitigation activity in the fuel combustion sector. Compared to earlier SP-RP studies, we are able to explicitly distinguish the effect of payment vehicle, and vary the bid range and elicitation format. Furthermore, we are making the comparison for a good which has inherent moral components, which are relevant for many policy-informing SP studies, but also challenging.

Section snippets

Choice context

The data for this study consist of four samples using the same choice context, which is asking respondents to value distributional outcomes of climate policy.2 Climate policy is described as having a future and a present time impact on

Results

We start by showing the model estimates. Then we conduct the Swait & Louviere test of heterogeneity in preference structure and scale for all three specified tests and as a final test, we estimate and compare WTP of the different samples included in each of the three tests.

Table 2 lists the results of the four models. In all models, the estimated attribute parameters have the expected signs. The cost coefficients are all estimated as negative, indicating that respondents on average experienced

Concluding discussion

The presence of hypothetical bias has been an issue since the use of stated preference elicitation methods started. While in some fields, e.g. transportation, validation by the use of revealed preference methods is possible, this is a less viable option for many of the environmental goods where non-use values play a large role. Further, many environmental goods involve asking respondents to act as what Nyborg (2000) calls “Homo politicus” rather than “Homo economicus”, thereby potentially

Acknowledgements

The authors would like to thank Julian Sagebiel for providing his user-written Stata command, poetest. Furthermore, we thank three anonymous reviewers for constructive comments. Jette Bredahl Jacobsen acknowledge the funding from the Danish National Research Foundation for the Centre for Macro-ecology, Evolution and Climate (grant number DNRF96).

References (55)

  • S. Kallbekken et al.

    Public acceptance for environmental taxes: self-interest, environmental and distributional concerns

    Energy Pol.

    (2011)
  • J. Ladenburg et al.

    Gender-specific starting point bias in choice experiments: evidence from an empirical study

    J. Environ. Econ. Manag.

    (2008)
  • J.L. Lusk et al.

    Bridging the gap between laboratory experiments and naturally occurring markets: an inferred valuation method

    J. Environ. Econ. Manag.

    (2009)
  • A. Löschel

    The demand for climate protection—empirical evidence from Germany

    Econ. Lett.

    (2013)
  • P.A.L.D. Nunes et al.

    Identifying the warm glow effect in contingent valuation

    J. Environ. Econ. Manag.

    (2003)
  • K. Nyborg

    Homo Economicus and Homo Politicus: interpretation and aggregation of environmental values

    J. Econ. Behav. Organ.

    (2000)
  • R.H. Wiser

    Using contingent valuation to explore willingness to pay for renewable energy: a comparison of collective and voluntary payment vehicles

    Ecol. Econ.

    (2007)
  • F. Alpizar

    Does context matter more for hypothetical than for actual contributions? Evidence from a natural field experiment

    Exp. Econ.

    (2008)
  • J. Andreoni

    Impure altruism and donations to public goods: a theory of warm-glow giving

    Econ. J.

    (1990)
  • K. Arrow

    Report of the NOAA panel on contingent valuation

    Fed. Regist.

    (1993)
  • F. Carlsson et al.

    How much is too much?

    Environ. Resour. Econ.

    (2008)
  • R. Carson et al.

    Incentive and informational properties of preference questions

    Environ. Resour. Econ.

    (2007)
  • R.T. Carson

    Contingent valuation and revealed preference methodologies: comparing the estimates for quasi-public goods

    Land Econ.

    (1996)
  • J.B. Chang

    How closely do hypothetical surveys and laboratory experiments predict field behavior?

    Am. J. Agric. Econ.

    (2009)
  • ChoiceMetrics

    Ngene 1.1.1 User Manual & Reference Guide. Australia

    (2012)
  • D.P. Crowne et al.

    A new scale of social desirability independent of psychopathology

    J. Consult. Psychol.

    (1960)
  • M. Czajkowski

    Learning and fatigue effects revisited: investigating the effects of accounting for unobservable preference and scale heterogeneity

    Land Econ.

    (2014)
  • Cited by (19)

    • Hypothetical bias in stated choice experiments: Part I. Macro-scale analysis of literature and integrative synthesis of empirical evidence from applied economics, experimental psychology and neuroimaging

      2021, Journal of Choice Modelling
      Citation Excerpt :

      Therefore, the payment vehicle is often, by default, assumed to be a direct donation. The study of Svenningsen and Jacobsen (2018) discusses the potential role of payment vehicle on HB in such laboratory settings. Their testing in the context of climate policies suggested negligible difference between a hypothetical tax and a hypothetical donation, as reflected in stated WTP estimates.

    • Hypothetical bias in stated choice experiments: Part II. Conceptualisation of external validity, sources and explanations of bias and effectiveness of mitigation methods

      2021, Journal of Choice Modelling
      Citation Excerpt :

      Another factor that has been pointed out is the stake size, which is related to the range of the cost attribute (Isley et al., 2016; Ohler et al., 2000), suggesting that the larger the (hypothetical) monetary stake, the greater the HB. In public good valuation, the payment vehicle (e.g., tax versus donation) has also been suggested as a potential influencing factor (Penn and Hu, 2019; Svenningsen and Jacobsen, 2018). Methods of mitigating HB generally differ based on whether they are applied during the design and administration stages of the survey in order to counter inherent sources of bias, or through follow-up questions whose information can be used to correct HB in model estimation.

    • Attribute range effects: Preference anomaly or unexplained variance?

      2021, Journal of Choice Modelling
      Citation Excerpt :

      More recent applications, especially concerned with the impact of changing the range of the cost attribute, have also found mixed evidences. While in some cases, literature shows higher WTP estimates when the offered cost vector increases (Carlsson and Martinsson, 2008; Ladenburg and Olsen, 2008; Mørkbak et al., 2010; Contini et al., 2019), in others, the differences in marginal WTP are not always statistically significant (Hanley et al., 2005; Kragt, 2013; Glenk et al., 2019) or cannot be attributed solely to a change in the cost range (Su et al., 2017; Svenningsen and Jacobsen, 2018). In the light of this empirical evidence, researchers have suggested different reasons of why attribute range effects might occur.

    View all citing articles on Scopus
    View full text