Social context, financial stakes and hypothetical bias: an induced value referendum experiment

Abstract

This article investigates the effect of social context in an induced value, public good referendum experiment. In a split-sample treatment, social context is simulated by requiring participants to potentially have to disclose their vote to the group (voting disclosure) across both hypothetical and real settings. The experimental design also varies the cost (a coercive tax), and includes an uncertain level of benefit from the public good. The design allows investigation of the role of social context in both hypothetical and real referenda and its interaction with changes in the financial stakes involved (costs and potential benefits). Results show evidence of hypothetical bias, but also a social context effect that occurs in both real and hypothetical settings. This social context effect is larger than the effect of hypothetical bias, but is muted by the magnitude of costs and potential benefits. Hypothetical cases are also shown to be more prone to the social context effect.

Notes

¹ An increasing number of studies document the evidence of social preferences and other-regarding behaviour (e.g. Charness and Rabin, 2002; Ferraro et al., 2003; Messer et al., 2007). This is consistent with an emergent pattern of evidence indicating the importance of social factors in many experimental and behavioural economics studies. For instance, Prante et al. (2007) provide a meta-analysis of experimental economics studies that focus on Coasean bargaining and find that social contexts play a significant role in achieving efficient outcomes.

² According to Shogren (2005, p. 978): ‘Valuation experiments must confront the issue that people make decisions in both market and nonmarket settings, and that this interaction of exchange institutions could well affect the values that are elicited.’

³ Akerlof and Kranton (2005) introduce a utility function that incorporates self-image as an argument to explain a series of economic and behavioural implications of identity.

⁴ Mitchell and Carson (1989) referred to SD issue as a compliance bias; however, they may underestimate its significance with the following quote from DeMaio (1984, p. 279): ‘[T]o the extent that our actions as well as our responses to survey questions are influenced by what we see as socially desirable, perhaps the problem is not as overwhelming as it appears to be’.

⁵ Several studies have focused on whether phone or mail surveys, which involve different social contexts, should be standard protocol for CV applications. For example, the National Oceanic and Atmospheric Administration (NOAA) Panel on CV stated that it is ‘unlikely that reliable estimates of values could be elicited through mail surveys,’ and recommended in-person or phone-interviews (Arrow et al., 1993). However, Dillman (1996) argued that shortcomings of personal interviews are understated compared to mail surveys.

⁶ Rege and Telle (2004) do not use a referendum-type provision mechanism, but rather a public good experiment with a voluntary contribution mechanism. Similarly Alpizar et al. (2008a,b) investigate the role of social context for voluntary contributions in a field experiment. It may be inappropriate to use public good experiments with voluntary contribution mechanism to assess the validity of CV referendum surveys (Champ et al., 2002; Poe et al., 2002).

⁷ A number of CV studies (Ready et al., 2001; Berrens et al., 2002; Blumenschein et al., 2008) have used diverse approaches for incorporating respondents’ preference uncertainty in estimating WTP. While preference uncertainty has received a fair amount of attention, uncertainty regarding the good is relatively less addressed. Though environmental projects involve a wide variety of uncertainty, valuation studies are often presented as if there is no uncertainty about the good (Harrison, 2006a). Not only does this reduce credibility, it also makes respondents prone to speculations (Johansson, 1990; Crocker and Shogren, 1991). While Paradiso and Trisorio (2001) found that better knowledge about the good reduces hypothetical bias, Whitehead (1992, 1993) suggested that omitting uncertainty would distort WTP values. Fischhoff and Furby (1988), and recently Brouwer (2010), discuss how uncertainty in the level of benefit from the good, payment in question and social context can all affect measuring contingent values.

⁸ The design roughly mimics positive network externality effects as the benefits from the public good depend on the number of participants (Katz and Shapiro, 1985). Positive network externalities are common for many information commons (e.g. libraries), as well as many environmental services. For example, reduced risk of wildfire can be a function of the number of participants adopting wildfire risk mitigation activities in a community (e.g. Shafran, 2008).

⁹ To implement uncertainty, Vossler and McKee (2006) provide subjects with a range of equally-likely values, and then have one value selected through the roll of a die. In contrast, our design generates a more continuous spectrum of uncertain values contingent on subjects’ decisions. Also as Slovic et al. (1990) argue, people commonly obtain values for less familiar goods and services through an inferential process. So it is not unrealistic that subjects have to make inferences about others’ behaviour in stating their responses.

¹⁰ The College of Arts and Sciences in this university consists of around 20 departments ranging from humanities, social sciences to biophysical sciences, which ensures a large, heterogeneous pool of students to make our experimental subjects representative. The departments include: American Studies, Anthropology, Biology, Chemistry, Communication and Journalism, Earth and Planetary Sciences, Economics, English, Foreign Languages and Literature, Geography, History, Linguistics, Mathematics and Statistics, Philosophy, Physics and Astronomy, Political Science, Psychology, Sociology, Spanish and Portuguese, and Speech and Hearing Sciences.

¹¹ The signed ‘informed consent’ required by the Institutional Review Board (IRB) of the public university where the experiments were conducted allows subjects to not report their responses fully or partially. Earlier studies also document reluctance in responding to CV surveys (e.g. Berrens, 2000; Welch and Fischhoff, 2001).

¹² Average Net Benefit is the difference between Mean Average Benefit and Coercive Tax, where Mean Average Benefit is the average of possible maximum earnings (when everybody votes Yes) and minimum earnings (when a minimum six out of 10 subjects vote Yes since referendum only pass by majority votes).

¹³ This observed crowding out of the social context effect is consistent with results from a number of other studies. Frey and Oberholzer-Gee (1997) provide theoretical insights and empirical evidence that monetary incentives ‘crowd-out’ intrinsic motivation of civic duty in the context of residents’ willingness to host hazardous nuclear waste repositories in their neighbourhoods. Masclet et al. (2003) show that nonmonetary sanctions (e.g. social disapproval) work almost as good as a penalty with financial consequences.

¹⁴ In another CV study, Schkade and Payne (1994) used a verbal protocol that asked respondents to think aloud and document the factors that they considered when they were responding to a stated preference question. They found that more than 40% of the respondents considered how much other participants would be willing to contribute in their decision making process. From follow-up survey responses after their experimental sessions, Burton et al. (2007) provide evidence that hypothetical bias is prompted by other-regarding behaviour (acting for the good of group members) rather than free-riding or random voting mistakes. Welch and Fischhoff (2001) report two case studies investigating the relevance of social context in a CV task to elicit WTP for a river clean-up. They find that participants consider social contextual features just as relevant as the features of the non-market good and payment in question. In a related paper, Alpizar et al. (2008b) also explore other dimensions of social context (e.g. how information regarding a typical contribution and a token gift influence the voluntary contributions for improving the management of a national park).