Preference Refinement in Deliberative Choice Experiments for Ecosystem Service Valuation

I. INTRODUCTION

The importance of recognizing, demonstrating, and considering ecosystem services in environmental practice and policy assessment has gained considerable attention in the past 10 years (MEA 2005; TEEB 2010; Naturkapital Deutschland TEEB DE 2012; IPBES¹). In the policy arena this has led to a substantial interest in the assessment of economic value of ecosystem services, preferably in monetary terms (see the E.U. Biodiversity Strategy 2020, European Commission 2013). In recent years choice experiments (CEs) have gained considerable popularity among researchers for the valuation of nonmarket ecosystem services, due to their capability to obtain preferences for different policy characteristics (attributes) at the same time and the possibility to analyze trade-offs between them (Adamovicz, Louviere, and Swait 1998; Hensher, Rose, and Greene 2005). While this seemingly efficient way of preference elicitation for ecosystem services makes CEs attractive for researchers, this may be less the case for CE participants. Commonly implemented in a one-shot survey, stated preference methods place considerable strains upon respondents: On the one hand, respondents are asked to value changes to often unfamiliar and/or complex ecosystem services—about which they may have poor understanding and hold ill-defined preferences—in an unknown hypothetical market situation (Braga and Starmer 2005). On the other hand, standard survey formats give respondents little time and encouragement to think, learn, and reflect about their preferences, trade-off the benefits of the environmental change, decide on the amount of money they are willing to spend, and make a choice between different policy options (Brouwer et al. 2010; MacMillan, Hanley, and Lienhoop 2006). Given the complex task, instead of engaging in strenuous preference discovery, respondents may be influenced by different decision heuristics and framing effects that are theoretically inconsistent and lead to ill-rationalized responses and unfavorable biases (Braga and Starmer 2005; Bateman et al. 2008).²

The unfavorable combination of ill-defined preferences and one-shot surveys is also reflected in the literature on deliberative de mocracy. According to Fishkin (1991, 1), “An ordinary opinion poll models what the public thinks, given how little it knows.” Citizens’ beliefs and opinions are normally not based on high degrees of knowledge, sophistication, or consistency. Thus, in classical polls, a majority of respondents invent their opinion on the spot, and answers tend to be ill-rationalized and unstable (Fishkin 1991).

The scientific debate in political sciences early on recognized the role of deliberation in addressing this problem. Deliberation is regarded to be a prerequisite for the generation of accurate and rationalized opinions from citizens, as it provides them with the opportunity to discuss and with sufficient time to think, and thus enables them to discover and affirm their preferences on the issue at stake (Dahl 1989). Habermas (1984) describes a number of conditions for an “ideal speech situation” and refers to a “transcendental quality” of deliberation, in which participants first consider their individual interests, and through deliberation transcend these interests to adopt other-regarding perspectives and seek the common good. The underlying communicative rationality of deliberation envisages that individuals achieve a mutual understanding by means of exchange of argument and eventually reach consensus on of the issue under investigation.

Important aspects of deliberation include that citizens (1) are educated and informed about the issue, (2) have the opportunity to extensively reflect on their preferences, (3) are encouraged to ask questions, and (4) are spurred to express arguments for one outcome over another (Habermas 1984; Dahl 1989; Fishkin 1991). In applications of deliberative monetary valuation, Habermas’s “ideal speech situation” is an archetype, and the requirements for deliberation are adapted to the underpinnings of neoclassical economic theory, that is, rather than deliberation being a means of reaching mutual consent, it is here seen as a means to help participants refine their individual preferences. In this paper we therefore define deliberation in a narrow sense, namely, as (1) group discussion in which participants share information and raise advantages and concerns of the environmental change under investigation (see Lo and Spash 2012), and (2) the opportunity to reflect on preferences by giving them time (Habermas 1984).

CE applications can benefit from deliberation because discussion among participants and time for reflection can be crucial for participants’ comprehension of the environmental change under investigation and for preference discovery. This is particularly useful when the environmental change is complex and unfamiliar to the respondents (Dietz and Stern 1995; Dietz, Stern, and Dan 2009; Wilson and Howarth 2002; Svedsäter 2003). Recent research on respondents’ experience and knowledge levels in CEs provides important arguments for deliberation. Czajkowski, Hanley, and LaRiviere (2014) found that respondents become more certain about their preferences the more experience they have with the environmental good in question. Furthermore, LaRiviere et al. (2014) discovered that either more experience or more knowledge among respondents leads to more precise and higher willingness-to-pay (WTP) estimates. The authors argue that little experience with unfamiliar environmental goods or services might be compensated for by giving respondents solid information, in other words, knowledge (LaRiviere et al. 2014). This research implies that information provision and learning through deliberation are particularly helpful in valuation contexts that are unfamiliar to respondents. The critique of one-shot stated preference approaches has previously been addressed by the discovered preference hypothesis proposed by Plott (1996). The discovered preference hypothesis assumes that through practice and repetition, respondents gain experience with the task (hypothetical market, WTP question, or choice sets), called institutional learning, and discover their own preferences (value learning). Both institutional and value learning are assumed to lead to (1) preference discovery (impacts on preferences parameters), (2) choice precision (impacts on scale), and (3) a decrease in choice heuristics. Thus the last response in a sequence of repeated questions is assumed to be better rationalized and more stable (Plott 1996; Bateman et al. 2008). Furthermore, a number of repeated formats (test-retest studies) also provide evidence for changes in WTP, although not all studies explicitly asked respondents to reflect about their preferences in the meantime (Svedsä ter 2007; Loomis 1989; Teisl et al. 1995; Bateman et al. 2008; DeShazo and Fermo 2002; Holmes and Boyle 2005). According to Braga and Starmer (2005) repeated questions tend to stimulate institutional learning rather than value learning, because respondents are not provided with additional information between the rounds that might further encourage preference learning.

Another branch of stated preference research more closely adopts the different facets of deliberation. Contingent valuation applications using market stall or valuation workshops integrate solid information provision, discussion, and time for reflection into the valuation process (MacMillan, Hanley, and Lienhoop 2006; Lienhoop and MacMillan 2007; Lienhoop, Bartkowski, and Hansjürgens 2015) and find that all these aspects lead to an improved model fit in terms of the influence of independent variables and robustness (R²).

So far, deliberative CE applications are few. We summarize these studies and their findings as follows. Álvarez-Farizo and Hanley (2006), Christie et al. (2006), and Christie and Rayment (2012) compare CE results obtained in valuation workshops (elicitations before and after discussion), and a conventional survey method (only Álvarez-Farizo and Hanley 2006; Christie et al. 2006) in terms of preference parameter estimates and part-whole (implicit prices). The findings are mixed: Christie et al. (2006) observe an increase in the number of significant variables when moving from the first to the second elicitation in the valuation workshop, while the overall model fit improved, whereas Christie and Rayment (2012) find that all variables are significant before and after discussion, with a slight increase in model fit. Comparing the workshop models with the main survey models, no difference in the model results is detected between the two data collection modes by Christie et al. (2006), whereas Álvarez-Farizo and Hanley (2006) identify significant changes in preferences parameters and implicit prices (see also Torres et al. 2013). Reasons for these changes remain unexplained. Similarly Shapansky, Adamowicz, and Boxall (2008) look at the effects of information provision and respondent involvement on choice models and implicit prices with the difference that they use three independent treatment groups (a high-involvement group exercise, a low-involvement group exercise, and a mail survey). In line with Álvarez-Farizo and Hanley (2006) and Christie et al. (2006) they find less variance in choices in the high-involvement group compared to the low-involvement group. Interestingly though, when comparing the mail survey with the high-involvement group, no significant difference in preference and variance in preferences was detected. The authors conclude that further research is required to identify the contribution of deliberative CEs to improving CE results. Along those lines, Robinson et al. (2008) conducted a CE exercise before and after exposure to expert information in a jury setting. Here the model fit also improved between elicitation 1 and 2, and implicit prices for the attributes change. Two studies (Kenter et al. 2011; Álvarez-Farizo and Hanley 2006) find that the cost attribute became an insignificant determinant of choice after respondents were exposed to discussion and more information. Kenter et al. (2011) explains this by occurrence of lexicographic preferences, because respondents became more aware of their dependence on and cultural identity associated with the ecosystem services investigated in their study after discussion. Two studies were set up to compare individual and collective choice settings and found no difference in terms of implicit prices and model estimates (Álvarez-Farizo et al. 2007; Álvarez-Farizo and Hanley 2006). Two recent studies apply valuation workshops for unfamiliar public goods without further testing deliberation effects: Aanesen et al. (2015) elicit well-informed choices for the preservation of coldwater coral, whereas Colombo, Christie, and Hanley (2013) investigate attribute nonattendance in a CE on ecosystem services and biodiversity enhancements.

While the existing research makes important contributions to the role of deliberation in CEs, the findings are based on very small sample sizes that cannot offer strong evidence nor allow firm conclusions to be drawn.³ A number of questions remain unanswered: First, there is still very little understanding about why respondents choose differently after group discussion. The existing research has a solely quantitative nature, and apart from one qualitative study from the Solomon Islands (Kenter et al. 2011), none of the studies has explored the motives and considerations respondents take into account when making their choice and how they alter after respondents have discussed the environmental change under investigation. In order to gain detailed insights into the impact of discussion on choice changes and choice quality, qualitative data on respondents’ motives, arguments, and perceptions are essential. Second, previous research has looked at changes in model fit, parameter estimates, and implicit prices. In order to draw firm conclusions about the contribution of deliberation on choice quality, additional indicators, namely, how respondents perceive the discussion helps them learn about their preferences and whether choices become more certain, should be sought. Third, the existing research focuses on discussion effects. Time for reflection has not received any attention in the CE context. Given that respondents need to consider a number of attributes in CEs, it is worthwhile to investigate the effect of time for reflection on respondents’ choices.

In order to address these gaps, for this study we developed a deliberative CE design in an attempt to generate well-informed and rationalized ecosystem service values for policy advice. The overall aim is to test how deliberation affects preferences learning and facilitates respondents to make thoughtful and well-considered choices. For this purpose four hypotheses—two on discussion effects and two on effects of discussion and time for reflection—were formulated and tested using quantitative and qualitative data. Our value assumptions lean toward utilitarian rationality but draw from the discovered and constructed preference literature. Thus we assume that respondents’ choices are guided by their selfinterest and that these choices are not based on a predefined preference set, but are formed by a process of preference learning⁴ (Gregory and Slovic 1997; Brouwer et al. 2010; Braga and Starmer 2005). Respondents considering other-regarding and future generations’ needs in their choices are regarded to do so in a utilitarian sense, in that they obtain personal satisfaction. This differs from the approach of Habermas (1984), who assumes that respondents transcend from the individual to the other-regarding perspective in order to seek a common solution, and covers only one out of many types of social and shared values (see Kenter et al. 2015).

: Discussion leads to better-informed choices.

Group discussion exposes respondents to a range of viewpoints, perspectives, and knowledge of which they may not be aware when interviewed individually (Schkade and Payne 1994; Svedsäter 2003; Robinson et al. 2008). Thus, we assume that information sharing and preference learning lead to a broader consideration of aspects relevant to the environmental change under investigation, and that discussion facilitates choices that are better-informed, more thoughtful, and based on reason compared to more impulsive choices made in isolation (Robinson et al. 2008; Lienhoop, Bartkowski, and Hansjürgens 2015; Söderholm 2001; Wilson and Howarth 2002; Svedsäter 2003; Dietz, Stern, and Dan 2009). Qualitative data collected prior and after group discussion are used to test for these assumptions (see Section III).

: After discussion, self-interested and myopic motives decrease and considerations of other-regarding and long-run consequences increase.

A number of authors suggest that the exposure to different viewpoints is likely to encourage respondents to make choices that are not based merely on their direct personal needs, but consider also the benefits and costs accruing to others (Plott 1996; Sagoff 1998; Svedsä ter 2003; Dietz, Stern, and Dan 2009; Dietz and Stern 1995; Lienhoop, Bartkowski, and Hansjürgens 2015; Kenter et al. 2014; Kenter et al. 2015). Furthermore, discussion can lead to enhanced recognition of future generations in addition to short-term costs or benefits (Niemeyer and Spash 2001). As noted above, the consideration of other-regarding and future interests is here defined in a utilitarian sense—or as I-rationality according to Vatn (2009). These assumptions are tested using qualitative data on choice motives (see Section III).

: Discussion and time to reflect stimulate preference discovery.

One of the fundamental assumptions in economic theory and stated preference research is that “preferences are pre-existing, stable, and complete across all choice sets, and can therefore merely be called upon” (Spash 2007, 693). This assumption has been the target of considerable criticism, and the process of preference discovery is an ongoing matter of debate: The literature on preference discovery assumes that individuals’ preferences may be ill-defined, and at the same time that they hold a set of intrinsic values. Thus, preferences need to be discovered and choices need to be adjusted accordingly (Brouwer et al. 2010; Braga and Starmer 2005). In the environmental context, it is likely that ecosystem services exhibit high levels of unfamiliarity and complexity, thus requiring a process in which respondents can discover their preferences (Brouwer et al. 2010). Deliberation could be a mechanism to facilitate preference discovery, as it helps respondents to absorb and process complex information through discussion and time for reflection (Niemeyer and Spash 2001; MacMillan, Hanley, and Lienhoop 2006; Lienhoop and MacMillan 2007; Shapansky, Adamowicz, and Boxall 2008; Szabo 2011; Sagoff 1998; Wilson and Howarth 2002; Robinson et al. 2008; Lo and Spash 2012). As respondents adjust their choice behavior in accordance with preference learning, choice changes as well as changes to implicit prices elicited after discussion and time to reflect are used as indicators for preference discovery (see Section III).

: Discussion and time to reflect lead to more certain choices.

In line with Hypothesis 3, it is assumed that respondents gradually discover their exact preferences in response to discussion and time to reflect. Thus, we would expect that deliberation reduces initial choice uncertainty. Brouwer et al. (2010) discovered an increase in self-reported choice certainty through repetition of the choice task, but they did not find econometric evidence for this in terms of increasing variance of the error term in their random utility models. Czajkowski, Hanley, and LaRiviere (2014) found that experience reduces preference uncertainty as measured through the random component of utility. We assume that both discussion and time lead to higher levels of certainty. We explore changes to choice certainty from a respondent’s perspective in terms of self-reported certainty (similar to Brouwer et al. 2010), and from an econometrician’s perspective in terms of the variance of the error term in a random utility model (Brouwer et al. 2010; Czajkowski, Hanley, and LaRiviere 2014; see Section III).

II. DELIBERATIVE CE DESIGN

The underlying case study is concerned with public preferences for an extension of forest cover in West Saxony, Germany. West Saxony is one of the least densely afforested areas in Germany, with only 17% of the area being covered with forest and agriculture being the dominant land use (SMUL 2007). Currently, Saxony’s authorities aim to increase forest cover by planting mixed forests consisting of native species. The main motivation is to enhance forest-related ecosystem services, such as flood protection, erosion control, carbon sequestration, climate regulation, and landscape esthetics (SMUL 2007; Regional Plan West Saxony 2008; Sächs WaldG 2008).

In the CE, respondents were asked to choose between various afforestation options. Each afforestation option was described in terms of four attributes: landscape esthetics (including the share of land covered by forest), water purification, carbon sequestration, and change in annual household expenses representing costs (WTP) and savings (WTA).⁵ Table 1 provides an overview of the status quo and the levels used for alternative afforestation options.

Attribute levels were combined across the afforestation scenarios, so that the four attributes and their levels resulted in 7,056 possible combinations. A D-efficient fractional factorial main effects design (see Scarpa and Rose [2008] for a discussion of design efficiency in discrete choice modeling) run in NGene⁶ was applied to reduce this number, resulting in six choice cards per respondent. Each choice card consisted of two policy options and a status quo option (see example in Figure 1).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Choice Card Example

The CE was conducted in a deliberative setting using the market stall approach (MacMillan et al. 2002). The market stall exercise was adjusted in order to facilitate the investigation of discussion and time to reflect effects on respondents’ choices (Table 2). Important features included (1) detailed information provision in the form of verbal explanation by the moderator and information folders for respondents to read, (2) a discussion that gave respondents the opportunity to ask questions and encouraged them to voice the pros and cons of forest increase and associated ecosystem services, and (3) a weeklong time interval at home to reflect about the policy change and preferences. The market stall meetings lasted about two hours.

Respondents received the same set of six choice cards in each round. To avoid that respondents merely repeat their choices in the subsequent rounds, the order of choice cards was shuffled in Rounds 2 and 3. Each of the three rounds of CEs was followed up by a questionnaire asking respondents about their considerations and motives about forest increase, their choice certainty, perceptions of the choice task and the group discussion, and so on. The questionnaires contained both open-ended and closed-ended questions (see excerpt of the questionnaire in the Appendix). For Round 3, participants received a set of choice cards, a follow-up questionnaire, and a prepaid envelope at the end of the meeting. Respondents were asked to keep the documents at home and to fill in and return them via mail after a reminder phone call from one of the researchers after about one week.

A total of 150 persons were invited to participate in one of 12 market stall meetings. The meetings were held in three purposively selected small towns in the study area in February 2014, with an overall population amounting to about 33,000 households. Each meeting consisted of 8 to 15 participants. Participants were recruited by a market research firm that ensured an even age and gender distribution among participants. Of the 150 recruited participants, 125 showed up, resulting in a response rate of 83%. Ninety out of the 125 participants returned the choice cards and follow-up questionnaires via mail (Round 3).

III. METHODS AND MODELS FOR HYPOTHESIS TESTING

In order to test our hypotheses we draw upon data obtained with the follow-up questionnaires as well as the choice data. In this section, the procedure used to analyze the qualitative data from the open-ended followup questions and the econometric modeling of choice data are described in detail.

IV. RESULTS

The socioeconomic characteristics in our sample for the most part closely reflect the population of the state Saxony,⁹ with the exception that highly educated members of the general public are underrepresented¹⁰ (Table 3). Ninety-two percent of the respondents regard forest increase to be very or somewhat important. Forest is used for a number of activities including walks, cycling, mushroom picking, bird watching, running, berry picking, and making firewood (in order of decreasing importance). Twenty-four percent of the sample visit a forest at least once a week.

Better-Informed Choices

We hypothesized that discussion helps respondents to make better-informed choices. This was tested in two ways: (1) by the number and type of aspects participants consider when making their choices in Round 1 (before discussion) and Round 2 (after the discussion), and (2) by analyzing respondents’ perceived role of the discussion for preference learning.

Table 4 shows the results of a frequency analysis of the number and types of aspects considered before and after discussion. The results at the bottom of Table 4 show that the overall and average number of aspects considered is reduced after discussion (Round 2). This finding is contrary to our expectations but might be explained by the fact that respondents simply did not repeat the consideration they had already mentioned in the follow-up questionnaire in Round 1. Interestingly though, when looking at the type of aspects, we find that in Round 2 less obvious and more complex benefits of forest increase (e.g., impact on water purification, flood control, future generations) are raised, whereas obvious and clear aspects were stated less often (e.g., personal use, concern for environment, landscape esthetics, timber production).

In the follow-up questionnaire in Round 2, we asked respondents to raise all aspects that they liked or disliked about the discussion. A number of topics were raised, including the general atmosphere of the group meeting, the moderators, and the role of discussion for preference learning. Here, we focus on the latter aspect in order to further explore whether the discussion led to better informed choices. Eighty-three participants responded to the open-ended question. Of these, 69 respondents stated that the discussion was a good means for information provision and helped them learn about their preferences. Figure 2 summarizes these statements in three categories and provides exemplary quotations. The quotations show that the discussion made respondents aware of benefits and costs of forest increase that they had not thought of on their own, and helped them to obtain a broad insight into the topic and to form an opinion and preferences toward forest increase.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2

Perceptions of the Discussion with Respect to Preferences Learning

Although we cannot confirm that respondents base their choices on more aspects after discussion, there are strong indications that discussion helps respondents to consider other aspects that they would not have thought of on their own. We thus confirm Hypothesis 1 that discussion leads to better-informed choices.

Personal, Myopic, and Other-Regarding Considerations

According to Hypothesis 2 we assume that respondents focus on personal and myopic benefits of forest increase when they complete the choice task without group interaction, and that other-regarding and future generations’ needs receive attention in the choice task completed after discussion. Table 4 shows that personal aspects (personal value of forest increase and reference to payment) are less frequently stated, and considerations of benefits to others and future generations are more frequently stated in Round 2. It is likely that respondents did not repeat aspects that they had already written down in Round 1, hence we cannot conclude that personal and short-term motives are replaced by other-regarding and future considerations. Our finding simply shows that some respondents do consider aspects regarding benefits to others or future generations after having been exposed to other people’s views and perspectives. In a short discussion following Round 2 we asked respondents whether they considered new aspects in the CE following the discussion. We analyzed the data with respect to statements concerning others and future generations. In all groups, respondents voiced that it was interesting to hear why other participants supported forest increase, but only in two groups did respondents say that this led them to consider other people’s preferences in their choices. In 9 out of 12 groups a number of respondents stated that they had thought of the benefits of forest increase accruing to their children, as the discussion made them realize that forest benefits arise in the long-run and are particularly valuable to future generations. Overall, our findings suggest that future generations’ needs do play a role in some respondents’ choice decisions, but the evidence of considering other-regarding needs is relatively limited.

Preference Adjustment

In order to test whether respondents’ preferences change in response to discussion and time to reflect, we investigate the number of choice changes between the rounds and changes to WTP. On average, respondents make one change in the choice task between Round 1 and Round 2 in response to discussion (0.9 of 6 choice cards changed) and one change between Round 2 and Round 3 in response to time to think (1.1 of 6 choice cards changed). Figure 3 shows that about half of the respondents adjusted their choices on at least one of the six choice cards between the rounds. Only a small share of the respondents changed their choices on more than two choice cards. The output from the GMNL model in WTP space shows that WTP for the individual attributes (as represented by the coefficients) changes between the elicitation rounds (Table 5): Overall, all ecosystem services positively and significantly influence choice. Respondents’ preferences for landscape esthetics remain the same in Round 2 and are adjusted upward in Round 3. With respect to water purification, preferences also remain unchanged in Round 2 and slightly increase in Round 3. Preferences for carbon storage decrease in Round 2 and remain unchanged in Round 3. The tax attribute is a negative determinant of choice. The influence of tax becomes stronger in Round 2 and continues to be a highly significant determinant of choice in Round 3.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3

Percentage of Respondents Who Changed Their Choices between Rounds

Although none of the changes in marginal WTP between the three elicitation rounds is statistically significantly different from zero at the 5% level, the results offer preliminary evidence that the discussion affects both WTP for carbon storage as well as the consideration of the cost attribute, whereas time to reflect encourages respondents to change their preferences for all attributes apart from carbon storage. Hence, time for reflection during the weeklong interval at home seems to lead to preference refinement. The follow-up questionnaire in Round 3 sheds some light on respondents’ activities during the weeklong interval: 84% stated that they had talked with their family about the initiative to increase forest, another 84% said that they had spent further thought on the issue, and 66% had again looked at the information folder. Thus, respondents took into account preferences of other household members when making their choices in Round 3 and also spent more time thinking about environmental change. Based on the insignificance of our findings we cannot confirm our hypothesis that the group discussion and the weeklong interval at home are important aspects in the preferences discovery process, but yet the findings are suggestive of preference adjustment.

Choice Certainty and Choice Precision

In terms of self-reported choice certainty, respondents were asked to indicate their perceived choice certainty on a four-point Likert scale. Figure 4 reveals a slight increase in choice certainty, with 88% of respondents being very or quite certain in Round 1, 92% in Round 2, and 96% in Round 3. This increase is not statistically significant, but the result is suggestive of an increase in choice certainty as seen from a respondent’s perspective.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4

Self-reported Choice Certainty in Rounds 1 to 3

Choice certainty is further explored from an econometrician’s perspective in terms of the relative scale parameter (Table 6). The heteroskedastic logistic model reveals a similar scale, and thus a similar error variance, in Rounds 1 and 2 since the estimated scale parameter (λ = − 0.023) is very small and not statistically significant. This outcome suggests that choice certainty is not affected by the discussion. The comparison between Rounds 2 and 3 reveals a similar error variance (λ = 0.007) and thus indicates that respondent uncertainty is unchanged after time to reflect about the valuation task. Both changes in self-reported choice certainty and scale parameters are not strong enough to confirm our hypothesis of a reduction in choice uncertainty—both from a respondent’s and from an econometrician’s perspective—in response to deliberation, although there are indications for increased certainty from a respondent’s viewpoint.

V. DISCUSSION AND CONCLUSIONS

In this paper we describe a deliberative CE approach that facilitates the exploration of the role of group discussion and time to reflect for preference refinement. We found that (1) the discussion enables respondents to consider less obvious and more complex aspects of forest increase in their choices; (2) most respondents still focus on their personal needs after discussion, but there is increased attention toward future generations’ needs; 3) respondents adjust their preferences after discussion and time to reflect, although not in a statistically significant sense; and finally (4) self-reported choice certainty increases slightly after each elicitation round from a respondent’s perspective, but not in our econometrical analysis with the scientist’s perspective. The underlying assumption in this paper is that preferences for ecosystem services need to be discovered and that deliberation facilitates the elicitation of well-informed and rationalized preferences.

According to our findings it seems that the opportunity to discuss the environmental change under investigation with other group members encourages preference refinement. After the discussion, respondents pay attention to more complex and less obvious costs and benefits that they did not think of on their own. This finding suggests that the exposure to new information in the form of other participants’ thoughts and perspectives during the group discussion leads respondents to form their preferences in a broader, more comprehensive sense. This notion is supported by the number of respondents who stated in an open-ended question that the discussion facilitated deeper insight into forest increase. Against our expectations we found a relatively small increase in other-regarding and future generations’ considerations after discussion. Benefits to future generations were taken into account more frequently after discussion than other-regarding interests. Given the relatively short discussion time (lasting between 20 and 40 minutes) it could be that respondents were yet occupied with preference learning about their personal needs and were not yet prepared to look beyond their individual interests. Future generations were considered more often and were potentially easier to grasp. The group discussions indicated that respondents were aware of the longterm nature of forest ecosystem service provision and that these benefits would accrue to their children, whereas benefits to others were not raised at all in the discussions. In the context of shared social values, Kenter et al. (2014) state that the extent to which such values are expressed depends on the duration and intensity of the group interaction. It might thus be worthwhile to investigate in a future research project whether more time for discussion or repeated meetings would facilitate considerations of other people’s needs. Furthermore, in comparison to deliberative institutions aiming at mutual consent, deliberative CEs do not naturally encourage participants to think beyond their personal needs. Thus, the moderator plays an important role in stimulating the consideration of other-regarding and future preferences. In this study the facilitator took a neutral stance and did not steer the discussion topics.

Do preferences change in response to deliberation? The opportunity to discuss and time for reflection seem to be important elements for preference refinement in CEs. Our results show that half of the respondents make changes in one or more choice tasks between the three elicitation rounds and that marginal WTP estimates for the individual ecosystem services slightly change in Round 2 (after discussion) and again in Round 3 (after the weeklong interval at home). Our results are in line with similar research on contingent valuation: MacMillan, Hanley, and Lienhoop (2006) found that in response to discussion and time to reflect, respondents in a deliberative contingent valuation study changed their WTP significantly for an unfamiliar environmental change (reintroduction of the red kite to Scotland) and WTP for a more familiar change (increase in wind power) only led to insignificant changes in WTP. Before final conclusions can be drawn, it would be worthwhile to validate our finding by giving respondents new cards in each round, rather than simply shuffling the same cards, in order to exclude the potential of anchoring effects.

An interesting question that emerges is why respondents adjust their choices. The discussions and the various follow-up questionnaires after each round provide insight into the aspects raised during the group discussion and time for reflection at home that could have influenced respondent’s preference refinement process. An interesting observation, for instance, is the change in WTP for the attribute carbon storage, which was much stronger between Rounds 1 and 2 than for the other attributes. This pattern is directly mirrored by the group discussions: While participants’ arguments rose in the discussions asserted the positive side of forest increase for landscape esthetics and water purification, critical voices were raised about local efforts for carbon storage against the background of limited efforts to reduce carbon emissions and deforestation in other countries. Another topic in the group discussions was the question over what would be a reasonable amount for each household to spend for forest increase. Thus, the influence of the cost attribute became more significant in Round 2 and can be explained by the fact that participants paid more attention to the cost attribute in response to the discussion. Landscape esthetics and water purification increased only slightly in Round 2. The discussions revealed that respondents were able to easily envisage the landscape changes resulting from forest increase. They described their region as a homogeneous landscape dominated by large-scale agricultural production. Most respondents drive about 20 km to reach a “nicer mosaic landscape with more forest.” Respondents know both landscapes well and were therefore able to imagine how more forest would change their surroundings. Hence, the discussion seems to have confirmed their initial preferences for landscape esthetics and therefore led to only a marginal increase in WTP. With respect to water purification it turned out in the discussion that respondents were very interested in having better water quality in the river flowing through their region. The fact that forest increase could improve the water quality was new to most participants and required considerable explanation prior to Round 1. The discussion further confirmed the benefits of forest increase on water quality and therefore resulted in relatively stable preferences.

We know little about respondents’ preference refinement processes during their week-long interval at home, but the follow-up questions provide some insight into respondents’ activities: The majority of respondents discussed the topic with family, colleagues, or friends, took a second look at the information folder, or simply thought about the issue. This finding shows that not only further discussion with family members, but also time to reflect and information processing seem relevant factors for respondents’ preference refinement. Overall, the results presented here show that respondents change their preferences for all ecosystem services regardless of whether they are easy to comprehend (e.g., landscape esthetics) or more complex (e.g., water purification). We can therefore conclude that even though respondents are familiar with an ecosystem service, preferences elicited in a one-shot survey are not fully refined and undergo adjustments in response to arguments raised during the discussion and during the weeklong interval at home.

Our findings raise the question of when preferences are fully refined and stable. We do not have evidence for stabilized preferences in this study; all we know is that preferences still change during the weeklong interval at home. Interestingly, Bateman et al. (2008) and Brouwer et al. (2010) both find that preferences stabilize throughout a sequence of several repeated WTP questions and choice tasks, respectively. The difference in our study is that respondents repeat the choice tasks in isolation without additional information provision via discussion with others and without time to think about the environmental change in question. Thus, it seems that preferences stability is reached relatively quickly when choice elicitation is simply repeated without further information input between rounds, whereas preference discovery takes longer when respondents are encouraged to consider a wide range of aspects through discussion and have time to reflect about the topic (preference discovery is stimulated by new input and more thought). It would be worthwhile to devote additional research effort to separating out the effects of discussion, time to reflect, and pure repetition on preference stability to better understand the importance of these aspects for preference discovery. In this context it would also be important to explore to what extent and when (from the very beginning or at a later state in a repeated exercise) respondents use choice heuristics. It is not clear whether deliberation leads to an increase or decrease in choice heuristics: On the one hand, the constructed preference literature suggests that respondents holding ill-formed preferences apply decision heuristics or rules of thumb to tackle the choice task, whereas respondents who know their preferences and have gained experience with the hypothetical market and choices task do not rely on choice heuristics (Kahneman, Slovic, and Tversky 1982; Bateman et al. 2008; Braga and Starmer 2005). On the other hand, there are concerns that fatigue and boredom through repetition of the choice task discourage respondents from discovering their preferences and encourage the use of simplifying choice heuristics (Gregory and Slovic 1997).

Respondents’ self-reported choice certainty suggests that both discussion and time to reflect lead to a slight increase in respondents who feel certain that their choices reflect their preferences. However, this is not in line with the econometrician’s perspective, since the scale parameters are not statistically significant and no firm conclusions can be drawn about changes in error variance between the rounds. Brouwer et al. (2010) found no increase in scale parameter when respondents were asked to repeat the choice task, and Czajkowski, Hanley, and LaRiviere (2014) discovered that scale parameters increase when respondents have more experience with the good under investigation. An explanation for the ambiguity in the results are the two different perspectives—the respondent’s and the econometrician’s—that are taken in the two certainty tests. While our results indicate that respondents become more confident about their preferences in a deliberative setting, further testing is needed to draw final conclusions. For instance, further research could explore reductions in choice uncertainty resulting from deliberations about more complex, unfamiliar, and controversial environmental changes about which respondents might be highly uncertain and have no experience when first encountered with the choice task.

A major concern related to deliberative valuation approaches is their small sample size and lack of statistical representativeness. In the following we make three points to address this concern and to clarify when deliberative valuation makes sense: First, it is clear that statistical representativeness of small samples can be achieved only at a local scale. Thus, deliberative valuation approaches are restricted to small scales and are less suited for large scales if statistical representativeness is a requirement (Söderholm 2001). Second, it depends on the valuation purpose whether statistical representativeness is necessary. If the aim of the study is to aggregate individual preferences over the relevant population to provide input to a cost-benefit analysis, statistical representativeness is necessary in order to obtain accurate estimates. In such a situation, deliberative valuation should be implemented only in small-scale, local decision contexts. However, if the aim is awareness raising, for example, to demonstrate the importance of an environmental change to politicians, the requirements on the accuracy of aggregated estimates are somewhat lower (Brouwer et al. 2011). Third, it is worth distinguishing between statistical and political representativeness (Jacobs 1997; see also O’Neill 2001). Proponents of deliberative institutions argue that politically representative samples can be drawn, instead of aiming at statistical representativeness (Raymond et al. 2014; Goodin and Dryzek 2006). Thus, when statistical representativeness is not feasible because the population is large (too expensive), as on a national scale, political representativeness should be assured by recruiting participants that represent a diversity of social characteristics and a plurality of viewpoints toward the topic under investigation (Goodin and Drysek 2006).

With deliberative CEs being a relatively young research field, there are yet a number of underexplored aspects requiring attention in future research. We mention only two issues here. First, a concern associated with deliberative valuation designs is that they are prone to “various sources of contamination” (Holland 1997, 485); for example, participants might be influenced or biased by dominant participants who take a leading role in the discussion. So far no study has investigated group effects on respondents’ preferences (Völker and Lienhoop 2016). We would therefore like to reflect on the role of the moderator of deliberative valuation studies. The moderators in this study had a neutral stance toward forest increase. Their main role was to (1) counter contestable or overly exaggerated statements by participants in order to ensure that the overall information provision was substantiated (e.g., statements such as “forest increase threatens food security” were relativized), (2) ensure that no one was being dominated or feeling intimidated by other participants by encouraging quiet participants to raise their thoughts, and (3) answer and clarify questions regarding the topic. The question remains to what extent and how actively moderators of deliberative processes should steer the discussion in order to facilitate preference learning. The development of best practice guidelines to mitigate adverse group effects would be an important research agenda in this field. Second, the need for deliberative valuation designs may depend on the type of environmental change under investigation. We would assume that preference refinement in CEs varies depending on how familiar, complex, and controversial the goods or services to be valued are. It would therefore be worthwhile to explore the extent to which preferences for goods that have differing degrees of familiarity, complexity, and controversy are refined in response to deliberation and what the requirements for deliberation are in terms of frequency, length, and depth of interactions (see also Kenter et al. 2014).

It is still too early to provide crisp and conclusive methodological recommendations for stated preference elicitation. Our discussion has hopefully been fruitful in emphasizing the potential of deliberation and pinpointing open questions worthy of further research. With respect to the question whether deliberation delivers more appropriate estimates for policy advice than a one-shot approach, we believe that our study provides a response: Our overall research findings suggest that a deliberative approach to CEs that exposes participants to a variety of viewpoints during group discussion and gives them the opportunity to reflect about their preferences at home is suitable for the elicitation of public preferences for ecosystem services. First, it leads to a broader understanding of the topic and therefore generates better-informed and more rationalized choices. Second, both deliberation and the repetition of the choice tasks allow respondents to adjust their choices alongside preference learning and discovery. Third, respondents tend to become more certain about their choices. One-shot CEs do not give respondents the chance to learn about the environmental change and to refine their preferences in accordance. Since the deliberative design presented in this study leads to better-informed preferences throughout the elicitation rounds, we believe that the final choices are more knowledgeable and sophisticated compared to initial choices and should therefore be favored.