Using WTP and WTA to value farmland preservation under ambiguous property rights and preference uncertainty Yicong Luo Researcher, Department of Resource Economics and Environmental Sociology, University of Alberta, Downloaded from by guest on December 8, 2023. Copyright 2022 Edmonton, Alberta, Canada Email: yicongluo5@163.com Mailing address: 9#704, Nanguojinhui, Fuzhou City, Fuijan Province, China Brent M. Swallow Professor, Department of Resource Economics and Environmental Sociology, University of Alberta, Edmonton, Alberta, Canada Email: brent.swallow@ualberta.ca Mailing address: 515 General Services Building, University of Alberta, Edmonton, Alberta, Canada, T6G 2H1 Wiktor L. Adamowicz Professor, Department of Resource Economics and Environmental Sociology, University of Alberta, Edmonton, Alberta, Canada Email: Vic.Adamowicz@ualberta.ca Mailing address: 515 General Services Building, University of Alberta, Edmonton, Alberta, Canada, T6G 2H1 1 doi:10.3368/le.98.4.060120-0074R2 Abstract: We use a paired-sample binary choice experiment to estimate willingness-to-pay (WTP) and willingness to accept (WTA) values when land is converted from agriculture to developed uses in Alberta, Canada. Validated principles for stated preference are followed in scenario design, elicitation format, experimental design and ancillary questions. Preference uncertainty is addressed through alternative calibration of uncertain responses. Reliability and incentive compatibility measures indicate that respondents found both WTP and WTA scenarios to be plausible and incentive compatible. WTA – WTP value gaps are smaller than most previous studies, with consequentiality increasing WTA, WTP Downloaded from by guest on December 8, 2023. Copyright 2022 and the gap between WTA and WTP. 2 1. Introduction The NOAA Blue Ribbon Panel (Arrrow et al., 1993) reviewed the state of stated choice valuation and recommended measurement of willingness to pay (WTP) over willingness to accept (WTA). A central reason for this recommendation was the large and consistent gap found between WTP and WTA estimates. For example, Tuncel and Hammitt (2014) found gaps of 3-10 times in their analysis of 76 previous valuation studies. WTA-WTP gaps may reflect real behavioral motivations such as loss aversion, reference dependence (Tversky and Kahneman, 1979), endowment effects, substitution effects (Hanemann, 1991) or commitment costs (Kim, Kling and Zhao, 2015). WTA-WTP gaps may also result from study design problems such as scenario rejection, protest voting, strategic behavior, social Downloaded from by guest on December 8, 2023. Copyright 2022 desirability bias, and hypothetical bias (Horowicz and McConnell, 2003; Tuncel and Hammitt, 2014). The Arrow et al (1993) recommendation against WTA has contributed to a dearth of WTA studies. Lloyd-Smith and Adamowicz (2018) report that their search of the Environmental Valuation Reference Inventory (https://www.evri.ca/) in February 2018 revealed 14 times more studies of WTP than WTA. Nonetheless, there has been consistent defense of WTA. Knetsch (2007) argues that the choice of WTP or WTA should depend upon context, particularly the context of implied property rights and respondents’ reference condition. The most recent comprehensive guidance for stated choice valuation (Johnston et al 2017) concurs that the choice of WTA or WTP should be guided by a combination of theory and empirical conditions. Johnston et al (2017, pp. 344-345) caution, however, that studies of WTA should pay special attention to the framing of incentive-compatible questions and the possibility for scenario rejection. Following the argument laid out by Knetsch (2007, 2020), we We use a paired-sample binary choice experiment to estimate willingness-to-pay (WTP) and willingness to accept (WTA) values when land is converted from agriculture to developed uses in Alberta, Canada. Validated principles for stated preference are followed in scenario design, elicitation format, experimental design and ancillary questions. Preference uncertainty is addressed through alternative calibration of uncertain responses. Reliability and incentive compatibility measures indicate that respondents found both WTP and WTA scenarios to be plausible and incentive compatible. WTA – WTP value gaps are smaller than most previous studies, with consequentiality postulate that farmland preservation in peri-urban areas can be framed in both WTP and WTA terms. The literature is replete with WTP framing: owners of farms at the peri-urban margin face the decision to continue farming or sell the land to a developer who will convert it into an urban use. Residents of nearby urban areas are asked their willingness to pay to 3 conserve some of that land in agriculture. The property right implied by that framing is the farm owner has the right to convert land to urban uses, and would need to be compensated to forego that right. The corresponding reference scenario is then assumed to be continual expansion at the city’s extensive margin with supportive changes in zoning. That reference scenario is consistent with purchase of development right programs that have been implemented in several American states over the last 45 years, including Connecticut, Maryland, Massachusetts, New Hampshire, New Jersey, Pennsylvania and Rhode Island (Kline and Wichelns, 1988). Data from referenda to earmark public funds for those programs, such as conducted in Pennsylvania and Rhode Island, provide insight into factors associated with willingness to pay for the purchase of development rights (Kline and Wichelns, 1988). Downloaded from by guest on December 8, 2023. Copyright 2022 In their review of the literature from the previous 20 years, Bergstrom and Ready (2008) identified eleven studies that used choice experiments to value amenity values of farmland in parts of the United States where referendum data are not available. All eleven studies used choice experiment methods to estimate net willingness to pay for farmland preservation of particular populations. Studies conducted in the last ten years have continued to rely on choice experiments to value willingness to pay for farmland preservation in the United States (e.g. Duke et al, 2012) and Canada, including Androkovich et al (2008) in British Columbia and Wang and Swallow (2016) in Alberta. Use of WTA approaches assumes a different framing of the policy and property rights context. Parcels of land in the city and surrounding region are zoned for specific uses to promote the public interest, optimizing investment in infrastructure, public services and overall quality of life. Some land is zoned for preservation of different types of open space (e.g. parks, nature reserves, recreation areas, urban farming, commercial-scale agriculture), producing different mixes of use and passive-use values. Rezoning allowable land use is a quasi-political process that produces winners and losers, with certain public benefits of open space diminished when agricultural or natural lands are rezoned into residential, commercial or industrial uses. With residents assuming the property right to open space and current land use as the reference scenario, residents should be asked their willingness to accept compensation for rezoning and the loss of open space values that will result. Despite the logic of the WTA scenario, we have identified very few studies of WTA for farmland preservation in a peri-urban context. Willis (1982) used a contingent valuation approach to estimate the willingness of local residents to accept compensation for the potential loss of the South Tyneside Green Belt in northeast England, UK. In Finland, Li et al (2004) blended WTA and WTP approaches by presenting survey respondents with choice scenarios in which they could choose one of 5 alternatives, 4 including the status quo level of farmland preservation, a 3% reduction in the area of farmland preserved that would be compensated by a reduction in income tax, and 3%, 6% and 9% increases in the area preserved, that would require increases in income tax. We find no recent WTA study of the public values of agricultural land that has followed the study design principles outlined by Johnston et al (2017) or validated for WTA study by Lloyd-Smith and Adamowicz (2018). Here we report a study that estimates and compares willingness to pay and willingness to accept compensation for farmland preservation in Alberta, Canada. Recommended practices for study design are followed and a split sample study design ensures comparability of methods and results. We compare indicators of study effectiveness for WTP and WTA, including measures of survey completion, scenario Downloaded from by guest on December 8, 2023. Copyright 2022 rejection, strategy rejection, and respondent uncertainty, as well as estimates of marginal willingness to pay generated through Random Parameters Logit models. While all other indicators were nearly equal, respondent uncertainty was somewhat higher for WTA than WTP scenarios. We are unable to ascertain the causes of the higher preference uncertainty, but postulate that it reflects respondents’ greater familiarity with the WTP framing or greater social desirability bias in the WTA framing. The literature provides little direction on uncertainty recalibration for WTA, so we report WTP and WTA estimates with alternative approaches. We conclude that both WTA and WTP approaches produce valid and useful information for planning and policy. The levels and gaps between WTA and WTP were highest for those who were most convinced of the validity of the scenario, implying that loss aversion is diminished by uncertain property rights and reference scenarios. 2. Study context Policy context Our study is set in the Canadian province of Alberta, whose economy is heavily dependent upon mining, oil and gas, forestry, tourism and agriculture, all land-extensive industries. The total area in agriculture (field crop, grassland, intensive vegetable) has remained relatively constant over the last 20 years (approximately 50 million acres in 2016, Statistics Canada, n.d.), with expansions at the northern frontier roughly offsetting conversions to developed uses (residential, commercial, industrial) in the more-densely populated south. Land use conversion has been particularly rapid around the major cities of Edmonton and Calgary and the highway 2 corridor that connects them. Between 2000 and 2012, the areas of agricultural land in that corridor area decreased by 7.0%, while the area of developed land increased by 39.4%. Across the province, 0.3 million acres of farmland was converted from agriculture 5 to developed uses between 2000 and 2012, 70.3% of which was classified as the highest or second highest farmland quality in the province (Haarsma and Qiu, 2017). We postulate that both WTP and WTA scenarios are logical approaches to estimate the public values of converted agricultural land in the peri-urban areas around Alberta’s cities. As in many other jurisdictions, the policy context encourages the preservation of land in agriculture. Specifically, the Alberta Land Use Stewardship Act (2009) and the Modernized Municipal Government Act (2016) mandate Regional Growth Boards and municipal governments to develop and implement land use plans that preserve agricultural land and enhance the viability of the agricultural sector. Two policy approaches are available to those governments. One approach is to compensate land owners for Downloaded from by guest on December 8, 2023. Copyright 2022 agreeing to conservation easements. In 2019-2020, the Alberta Government earmarked a limited amount of funds for investment in Land Trusts that negotiate conservation easements with land owners (https://www.alberta.ca/alberta-land-trust-grant-program.aspx). Currently, there are at least 13 land trusts in Alberta holding conservation easements (http://www.ce-alberta.ca/). WTP is the appropriate framing for assessing the willingness of Alberta residents to contribute to the purchase of conservation easements that preserve land in agriculture. The second approach is to mandate compliance with land use plans that preserve agricultural land, as has long been done by provincial governments in British Columbia and Ontario where over 13 million acres of farmland is protected through provincial farmland protection legislation. Kaplinsky and Percy (2016) postulate that municipal governments in Alberta are empowered to impose significant limitations on private use of land without compensation, provided that some reasonable private use of the property is left to the owners (Kaplinsky and Percy, 2016). WTA would be appropriate for assessing the amount that the public would need to be compensated for the loss of passive-use values when agricultural land is converted to developed land uses. Thus our premise was that it is equally valid for planners to consider WTP for farmland preservation and WTA for development. Our split-sample study design allows us to test whether survey respondents find preservation and development to be equally plausible reference scenarios for estimating WTP and WTA. Study Areas Most Alberta cities are located in the central and southern parts of the province. For this study, we selected the six most populous cities and their surrounding areas that could be developed in future years. For cities with larger populations, we use the Census Metropolitan Area (CMA) as our study area, and for the cities with smaller populations, we use the Census Agglomeration area (CA). Both CMAs and 6 CAs are formed by one or more adjacent municipalities centered on a core population center. A CMA must have a total population of at least 100,000 of which 50,000 or more people must live in the core. A CA must have a core population of at least 10,000. Calgary CMA is the most populous in Alberta and the 5th largest most populous in Canada. From 2011 to 2016, the Calgary CMA had the highest population increase (14.6%), followed by the Edmonton CMA (13.9%), which is the second most populous in Alberta. Red Deer CA and Lethbridge CMA rank third and fourth, with population increases of 10.9% and 10.8% respectively between 2011 and 2016. Medicine Hat CA has the fifth largest population, which increased by 5.1% between 2011 and 2016. The least populous urban area included in this study is Grande Prairie, which had a population increase of 13.5%. Together the Edmonton CMA Downloaded from by guest on December 8, 2023. Copyright 2022 and Calgary CMA currently comprise around 66.7% of the Alberta population. We used ArcGIS and interpreted remote sensing data to estimate the acres of agricultural land lost and the increase of developed land use between 2000 and 2016 for each of the six study areas. The Edmonton CMA map is shown as an example in Appendix Figure A.1. The map shows that the conversion in the Edmonton CMA has primarily taken the form of suburban development on the periphery of Edmonton and the smaller cities and towns in the area. Developed land increased by 128,710 acres, a 75.5% increase in 16 years. Approximately 92% of the newly developed land was converted from agriculture. This represented a 7.2% reduction in agricultural land in the Edmonton CMA. Appendix Table A.1 summarizes the development trend in the six study areas. 3. Methods This study uses binary choice experiment methods to estimate and compare WTP and WTA for preservation and conversion of agricultural land located around Alberta’s six largest urban areas. Our goal was to follow best practice in both WTP and WTA variants of the study (Johnston et al, 2017), then to compare results with respect to indicators of survey performance and estimation outcomes. Study design The population for this study is persons greater than 18 years of age who live in one of the 6 largest CMAs or CAs in Alberta, and are actually or potentially responsible for paying property taxes or rent. Our goal was to obtain data from samples of approximately 650 individuals for the WTP and WTA 7 surveys. The survey was implemented on-line through a commercial survey company (explained in the data collection section below). We began both variants of the questionnaire by establishing the age and region of residence of the respondent. If the potential respondent was over 18 and resident in one of the six regions, then the survey could continue. Information about the purpose of the study was provided and consent obtained. Consenting respondents were then provided with background information about the magnitude of recent development and conversion of agricultural land in their particular region, illustrated by a map of the relevant CMA / CA. The remainder of the questionnaire had four sets of questions. The first set probed respondents’ Downloaded from by guest on December 8, 2023. Copyright 2022 individual reference conditions, including Likert-scale questions regarding their perception of the primary functions of land in the city-region context, the ecosystem services that agricultural land provides, and whether it was desirable to live near agricultural land. We avoided using persuasive communications to minimize social desirability bias. The second question set presented the choice scenarios, including an introduction and 8 binary choice scenarios (WTP or WTA for the two variants of the survey), with follow-up uncertainty questions for each scenario. The third and fourth sections included ancillary questions about demographic characteristics and belief in the validity of the scenarios. Addressing hypothetical bias This study incorporates a number of survey design and ex post analysis features to strive to reduce the effects of hypothetical bias. Loomis (2014, p.35) defines hypothetical bias in the WTP context as “the difference between what a person indicates they will pay in a survey or interview and what a person would actually pay.” In the WTA context, hypothetical bias could similarly be expressed as the difference between what a person indicates they will be willing to accept in a survey or interview and the amount that they would actually be willing to accept. Lopez-Becerra and Alcon (2021) identify the following possible causes of hypothetical bias: 1) strategic bias and free riding; 2) social desirability and warm glow bias; 3) information and framing effects; 4) interview and interviewer biases; 5) starting point, status quo, and anchoring bias; 6) preference uncertainty due to unfamiliarity of the choice scenario or payment vehicle; and 7) complexity of choice and cognitive dissonance (Lopez-Becerra and Alcon, 2021). Incentive compatibility holds when the respondent has incentive to truthfully reveal private information requested. Vossler et al (2012) show that incentive compatibility holds for binary choice experiments of WTP when the following conditions are met: 1) participants care about the 8 outcome; 2) the authority can enforce payment by voters; 3) the elicitation involves independent yes or no votes on the choice sets; and 4) the probability that the proposed project is implemented is weakly monotonically increasing with the proportion of yes votes. Lloyd-Smith and Adamowicz (2018, p. 136) show that these conditions can be extended for incentive-compatible elicitation of WTA, with the following modifications to condition 2): “the authority can enforce voters to give up the good” and condition 3 “the elicitation involves a yes or no vote on a single project.” Condition 1) can be assessed through preliminary and ancillary questions, conditions 2) and 3) depend upon the elicitation design, and condition 4) – consequentiality -- depends upon the background and framing information provided to respondents. Downloaded from by guest on December 8, 2023. Copyright 2022 In the WTP framing of farmland preservation values, consequentiality holds when a respondent perceives a positive probability that their response will influence public policy or planning decisions related to farmland preservation and that their property taxes or rent might increase as a result of public decisions to preserve more farmland. In the WTA framing, the respondent must perceive a positive probability that the research output would influence the pace of development allowed by policy makers and planners and that additional development would reduce their taxes or rent. We framed the WTP and WTA choice experiment scenarios to enhance consequentiality as described in endnote i.i We used two additional approaches to reduce the effects of different sources of hypothetical bias. First, we used the dichotomous choice format to reduce the complexity of the choice problems: respondents could only choose the status quo or one alternative. This made the problem similar to a referendum vote (Carson and Groves, 2007; Kline and Wichelns, 1994). We tried to make the preservation and development scenarios as simple as possible, but recognize that some respondents may have experienced cognitive dissonance due to the relative unfamiliarity of the proposed scenarios and the relative complexity of the language that we used (see a further discussion of this in the results section below). Second, we incorporated certainty questions and conducted ex post uncertainty recoding / recalibration (Ready et al, 2010) to account for preference uncertainty or social desirability bias. After each dichotomous choice scenario, we asked respondents “how certain are you that this is the choice you would make in an actual vote?”, with five possible answers: “Very certain,” “somewhat certain,” “neither certain nor uncertain,” “somewhat uncertain,” and “very uncertain.” Given its importance to our empirical results, we next provide a more detailed discussion of uncertainty recalibration. Uncertainty re-calibration 9 Explicit concern about respondents’ preference uncertainty can be traced back many years, at least to Arrow (1993). Stated preference studies may require respondents to consider choice scenarios that they have not previously encountered, especially for amenity values of public goods. If a hypothetical situation were to become real, respondents may truly be undecided about how they would respond. Two survey design approaches have been taken to allow respondents to articulate such preference uncertainty in the dichotomous choice WTP framework. The approach recommended by Arrow (1993) was to allow “yes”, “no” and “don’t know” as possible responses. The alternative approach commonly associated with Champ and Bishop (2001) and Ready et al (2010) is to ask respondents two questions about each scenario: would they pay X amount for an item? and how certain are they of that choice? Certainty Downloaded from by guest on December 8, 2023. Copyright 2022 answers are either expressed on a numerical scale (eg 1-10) or on a Likert-type scale, for example “certain,” “somewhat certain,” “somewhat uncertain,”, and “uncertain”. The uncertainty formulation appears to be most favoured in recent studies. The “don’t know” or “somewhat uncertain” responses can then be used to recalibrate the yes / no responses. Several rules of thumb have emerged. In the WTP context, Champ and Bishop (2001) propose recalibration of some “yes, but uncertain” responses as “no”, based on case studies that have compared stated WTP with actual WTP. That is, there is a presumption that there will tend to be hypothetical bias in WTP studies, with stated WTP exceeding actual WTP, so that a certain amount of recalibration from yes to no will offset that bias and also produce a more “conservative” estimate of value, although Loomis (2014) warns that the Champ and Bishop (2001) approach may over-correct. Perhaps because so few studies of WTA have been conducted in recent years, there is very little guidance in the literature regarding uncertainty recalibration of stated choice WTA studies. One of the most fruitful for our purposes is Groothuis and Whitehead (2002), who evaluated “don’t know” responses to dichotomous choice WTP and WTA scenarios. The WTP case concerned a program designed to protect sea turtles in North Carolina, while the WTA case concerned siting of a hazardous waste landfill in rural Pennsylvania. The authors compared the don’t know responses to predicted values from multinomial logit and ordered logit models of WTP and WTA. They concluded that don’t know in the WTP scenario was similar to a no response (similar to the Champ and Bishop (2001) assumption), while don’t know in the WTA scenario was similar to an intermediate response between yes and no. We found no study of uncertainty recalibration in a split sample comparative study such as the present study. We reason that uncertainty in the WTP and WTA frameworks reflects some amount of cognitive dissonance, or difficulty understanding the choice scenario and the payment vehicle. We further speculate that there was a social desirability bias toward protection of farmland that could lead a 10 WTP respondent toward saying yes, he or she would be willing to pay, and a WTA respondent toward saying no, he or she would not be willing to accept the compensation offer. To produce conservative estimates of WTP and WTA, we thus adopted the following approach in our preferred model. Yes, but either very or somewhat uncertain WTP responses were recalibrated as no while no, but either very uncertain or somewhat uncertain WTA responses were recalibrated as yes. Given that we have no other evidence to support this approach, we also report the less conservative WTP and WTA results without recalibration of uncertain responses. Downloaded from by guest on December 8, 2023. Copyright 2022 Choice Scenarios Following Vossler et al (2012) and Lloyd-Smith and Adamowicz (2018), the study used a dichotomous choice format. The status quo was the same in both the WTP and WTA variants: continuation of the recent trend in conversion between agriculture and developed land uses and no change in property tax or rent. The alternative for the WTP study had three attributes: current agricultural land use, alternative developed land use, and a one-time payment that the respondent would make toward preservation of an incremental 1,000 acres of land somewhere within 10 kilometers of developed urban areas that would otherwise be converted into the developed land use. The WTA study had three similar attributes: current agricultural land use, alternative developed land use, and a one-time compensation that the respondent could receive as lower property tax or rent in order to allow the conversion of an incremental 1,000 acres of land from agriculture to the developed land use somewhere within 10 kilometers of the developed urban areas. Three agricultural uses, grain / oilseed production, grassland / hayland, and commercial vegetable production, and three developed land uses, light industrial, commercial and residential, were considered. These land uses correspond to the main agricultural and urban land uses around urban areas in Alberta (Haarsma & Qiu, 2015). Our payment vehicle was a one-time increase in property tax or rent for WTP and a one-time rebate of property tax or rent for WTA (all in Canadian dollars). Taxpayers in Alberta have long experience with paying property taxes and recent experience receiving carbon tax rebate payments of $300 to $630 per household. For both WTP and WTA versions of the study, we set a payment range of $50 to $1000 and three payment levels between. This range was set to simplify the number of design combinations, to ensure that respondents would find all alternatives credible, and to allow comparability with the previous study of the Edmonton area by Wang and Swallow (2016) who used a payment range of $25 to $600. 11 The payment range was discussed in some depth during both focus groups, discussed below. The wide range reflected differing opinions regarding the level of payment or compensation that would be considered meaningful and realistic for respondents with different income types and levels. Table 1 lists the attributes and attribute levels for the WTP and WTA surveys.ii Figure A.2 presents examples of the choice scenarios presented to respondents for WTP for additional farmland conservation (top panel) and WTA compensation for additional development of farmland into alternative urban uses. Lloyd-Smith and Adamowicz (2018) condition 3 for incentive compatibility is that there be a single discrete choice question. Resource and sample size constraints did not allow us to fully satisfy this condition, although we encouraged respondents to give independent consideration of choice scenarios by Downloaded from by guest on December 8, 2023. Copyright 2022 randomizing the order in which scenarios were presented to respondents, disallowing respondents from returning to previously answered questions, and including the following statement immediately before the choice scenarios. “In each set presented below, imagine that these are the only options available for you to choose from. For each set, please choose independently from other questions – please do not compare options from different sets.” (emphasis in original). Finally, to address the possibility of protest voting we asked the following open-ended question in the final section: “Do you have any other comments about the context of this survey or the reasons for your answers?” The study team read each of these responses and identified respondents who indicated protest against the premise of the WTP or WTA approach. Those who lodged protests were recorded and excluded from further analysis. Following Herriges et al (2010), in the last section of the questionnaire, we asked the following ancillary question to assess respondents’ perceptions of plausibility / consequentiality: “How likely do you feel the strategies presented in the survey could be really implemented? Possible answers were “very likely,” “somewhat likely,” “uncertain,” “somewhat unlikely,” and “very unlikely.” We labelled respondents who answered “very likely” or “somewhat likely” to be “strategy believers.”iii We hypothesized that strategy believers perceived the study to be more consequential, although we did not have a specific hypothesis regarding the WTA / WTP gap for strategy believers compared to the full sample. Below we present and compare results for the full samples and the samples of strategy believers. Experimental design The experimental design was symmetrical for the WTP and WTA versions of the study. As discussed above, choices were made over 3 attributes: current agricultural use, replacement development use, and 12 one-time payment or compensation level. There were three types of current agriculture use, 3 types of replacement development use, and 5 payment or compensation levels covering the range of $50 to $1000. Therefore, there are 45 combinations (3*3*5=45). A full factorial experimental design would include all 45 combinations to fully cover the attribute space so that all main and interaction effects are statistically independent. However, the full factorial design means fewer observations for each combination when sample sizes are constrained (Hensher et al., 2005). In this study, there are observations for about 650 respondents for the WTP and WTA variants of the survey: we judged that there would be insufficient observations for accurate parameter estimates if the full factorial design was applied. We instead chose the “d-optimal” efficient design that minimizes the inverse of the determinant of the variance-covariance matrix of the parameters (D-error) (Rose et al, 2008). Because we had no Downloaded from by guest on December 8, 2023. Copyright 2022 prior data, we assumed that all alternatives contained in the choice sets were equally attractive. We considered 16 different choice sets blocked into two groups of eight, with each respondent required to answer one block of 8 choice questions. The Ngene software was used for the d-optimal design and blocking. Questionnaire development and data collection We convened two focus groups, a pretest, and a soft launch of the internet-based survey. Focus groups of 10-12 persons were held in the Lethbridge and Edmonton areas, two of the six locations for this study, and convened by a local survey company. The survey company recruited participants from a panel of over 40,000 Albertans and random digit dial telephone contact.iv After giving their informed consent, focus group participants completed a trial version of the WTA or WTP survey (on paper), then participated in a facilitated conversation regarding ease of comprehension, respondent fatigue, language and format, hypothetical bias, consequentiality, and payment and compensation ranges. Respondents expressed good understanding of both frames and a range of opinions about the plausibility of either frame. Using the constructive feedback from the focus groups we adapted the survey to increase the credibility of the scenarios and consequentiality of the overall study, providing more complete information, highlighting the payment and compensation mechanisms, randomizing the order of some questions to minimize default effects, simplifying the language and formatting the questions and survey layout to alleviate respondents’ cognitive burden. We used a survey design software to format the questionnaire for online presentation. The online version was pre-tested with 10 volunteers known to the researchers. 13 Before recruiting the full sample of respondents, a soft launch of 10% of the intended sample was implemented. Analysis of the soft launch data showed levels of payment and compensation consistent with our expectations. We also ran simple multinomial models for both WTP and WTA that yielded a significant negative sign on cost for the WTP equation and a significant positive sign on compensation for the WTA equation, as expected. We then requested the survey company to open the surveys to large numbers of panelists (about 42,000 for the WTP survey and about 40,000 for the WTA survey), accepting entrants and completed responses until pre-determined quotas were filled for the WTP and WTA variants of the survey. Quotas of WTP and WTA respondents were set for each of the 6 areas. The survey company screened potential respondents by three criteria. Residents should be actual or Downloaded from by guest on December 8, 2023. Copyright 2022 potential taxpayers, over the age of 18 years, and residents of urban areas from which they were recruited. The survey was offered only in the English language; thus non-English speakers were excluded from participating in the online surveys. A “trap question” was inserted before the choice experiment questions to test whether respondents were reading the materials and paying attention to the questions.v Considering the sensitivity of questions about income, that question was asked last and respondents were allowed to avoid answering that question. All other questions were forced responses, requiring respondents to answer each question before moving to the next question. <Table 1 about here> 4. Data analysis Data analysis involved three steps. The first step was to compare demographic and socio-economic statistics for the respondents from the two online surveys. A two-sample t-test was used to determine whether there were statistically significant differences between the WTP and WTA samples. The second step was to use the choice experiment data to estimate marginal utility of the strategies and their attributes and calculate MWTP and MWTA using appropriate statistical models. The third step was to compare indicators of study reliability and welfare measures estimated by the WTP and WTA models. The framework for welfare analysis in choice experiments is the random utility model as initially proposed by McFadden (1974) and first used by Hanemann (1978) to place values on environmental and natural resources. In the random utility model, a decision maker chooses one option from a set of alternatives and the researcher observes the choice and attributes of the option. 14 When individuals make a choice among N mutually exclusive alternatives, it is assumed that they know their utility perfectly, so they make the choice that maximizes their utility. Zi and Zj reflect the utility from the land use attributes in alternatives i and j. Therefore, individual k will choose alternative i among a set of all alternatives C in the choice set if and only if： 𝑣𝑖𝑘 (𝒁𝒊 , 𝑦𝑘 − 𝑝𝑖 ) + 𝜀𝑖𝑘 > 𝑣𝑗𝑘 (𝒁𝒋 , 𝑦𝑘 − 𝑝𝑗 ) + 𝜀𝑗𝑘 ; ∀ 𝑗 ∈ C (1) The random utility model predicts the probability that an alternative was chosen given its systematic and error components. The general form for the probability that individual k choses alternative i from a Downloaded from by guest on December 8, 2023. Copyright 2022 choice set is: 𝑃𝑖𝑘 = 𝑃[𝑣𝑖𝑘 (𝒁𝒊 , 𝑦𝑘 − 𝑝𝑖 ) + 𝜀𝑖𝑘 > 𝑣𝑗𝑘 (𝒁𝒋 , 𝑦𝑘 − 𝑝𝑗 ) + 𝜀𝑗𝑘 ; ∀ 𝑗 ∈ C] (2) β0 (ASC) + β1 (vegetable) + β2 (livestock) + β3 (retail) + β4 (light industrial) + β5 (cost) (3) The observable utility function for the empirical analysis in our study is: Vik (Zi, yk – pi) = with variables are defined as in Table 2. Appendix Table A.2 defines the variables used in the econometric analysis. <Table 2 about here> ASC is an alternative specific constant which was defined as the utility of choosing the baseline strategy, which was to avoid the conversion of 1,000 acres of grain or oilseed farming to residential use in the WTP preservation case and converting an additional 1,000 acres of grain or oilseed farming to residential uses in the WTA development case. There are two other attributes for type of agricultural land: livestock grazing land and commercial vegetable production. The other types of residential land use are retail and light industrial. The parameters from the choice experiment are used to estimate marginal WTP and marginal WTA. Choice experiment methods are consistent with utility maximization and demand theory (Bateman et al., 2002), therefore, the marginal WTP and marginal WTA in a linear utility function is given by equation (4) (Hanemann, 1984). See Appendix B for a more detailed explanation of equation (4) in terms of Compensating Variation (CV) for the WTP case and Equivalent Variation for the WTA case. 15 𝜕𝑣 ⁄𝜕𝑍 ⁄𝜕𝑦 𝑀𝑊𝑇𝑃 = 𝑀𝑊𝑇𝐴 = 𝜕𝑣 (4) We estimate equation (3) as a Random Parameter Logit Model (RPL), which can capture heterogeneity through estimates of the mean and variance of the random parameter distributions. Like previous studies discussed above, we evaluated the WTA – WTP gap as one indicator of the performance of the two approaches. We also evaluated the credibility of the two methods by comparing rates of completion and protest. The completion rate is the ratio of the number of complete responses and the number of people who entered the survey. Protest was evaluated by responses to the concluding Downloaded from by guest on December 8, 2023. Copyright 2022 open-ended question in which respondents were asked to provide feedback on the overall survey. The protest rate is the number of protest votes divided by the total number of complete responses. The uncertainty rate is the number of uncertain responses divided by the total number of complete responses. The people who believed that the preservation strategies are likely or very likely to be implemented are identified as strategy believers. The proportion of strategy believers is the ratio of the number of strategy believers and the total number of complete responses. 5. Results In February 2019, the survey company opened the WTP variant of the survey to a panel of about 42,000 people and the WTA variant to a separate panel of about 40,000 people. Two panels were used due to small number of panelists in the smaller cities relative to the size of the quotas. Once the survey company received sufficient completed responses to meet the survey quotas, they closed the surveys to new participants. 1,900 respondents entered the WTP survey and 1,750 respondents entered the WTA survey. The survey company implemented a data scrubbing process to remove responses that were incomplete or clearly insincere. 34 WTP responses and 40 WTA responses were ruled out by the attention trap question. Finally, a total of 643 WTP and 660 WTA complete responses was regarded as the full valid sample. The survey completion rate was 33.84% for the WTP survey and 37.71% for the WTA survey. 28 responses from the WTP survey and 13 responses from the WTA survey were excluded from the analysis because their answers to the open-ended questions suggested that they rejected the premise of the study. We thus conclude that most respondents accepted the plausibility of the scenarios. On the 16 basis of the uncertainty questions, 8.56% of the WTP scenarios were recoded from willing to not willing to pay (842 / 9,840 scenario responses) and 15.34% of the WTA scenarios were recoded from not accept to accept (1,588 / 10,352 scenario responses). A possible limitation of the study design is the complexity of the language used to describe the WTP and WTA protection and payment scenarios. The Flesch-Kincaid grade level is 19.3 for the WTP scenario and 18.5 for the WTA scenario. 58% of our respondents had less than university undergraduate education (Table 2). The complexity of language, coupled with relative lack of familiarity with the WTA payment vehicle, may have led to greater preference uncertainty for the WTA questions. To evaluate this hypothesis, we compared the percentage of uncertain responses for respondents with Downloaded from by guest on December 8, 2023. Copyright 2022 bachelor’s education or more and for respondents with less than bachelor’s education. For the WTP survey, the level of certainty (somewhat certain or very certain) was very similar for the two groups of respondents: 82.2% for respondents with bachelor’s degree education and above and 83.1% for respondents with less than bachelor’s degree education. For the WTA survey, the level of certainty was 79.1% for respondents with bachelor’s education or above and 75% for respondents with less than bachelor’s education. Thus, the WTP – WTA uncertainty gap was 3.1% for the more educated group and 8.1% for the less educated group. This supports the hypothesis that cognitive difficulty could have been a more important issue for the less educated group. Respondent Characteristics Table 2 summarizes demographic and socio-economic statistics for all respondents to the on-line surveys. Eighty percent of the respondents are resident in either the Edmonton or Calgary regions. Here, we only report the results for the pooled samples, results are similar when estimated for the Edmonton and Calgary sub-samples. Two-sample t-tests are used to examine whether there are statistically significant differences between the WTP and WTA samples. Over 70% of all respondents live in cities. Age, residence, living in city, education, income, and ownership are very similar between the WTP and WTA surveys, while there are statistically significant differences in participation by gender, % retired, and incomes of $90,000-100,000. Our assumption is that the differences in demographic characteristics in the WTP and WTA samples reflect differences in the sampling frames from which the two samples were drawn. Unfortunately, it is not possible for us to easily compare the characteristics of our samples with the sampling frames or the broader Alberta population because of differences between the simple 17 demographic information we collected and information available through the Government of Canada census. <Table 2 about here> Pooling results across Alberta’s urban regions We first estimated equation (3) as a Random Parameter Logit Model for three sub-samples: respondents living in the Edmonton urban region; respondents living in the Calgary urban region; and respondents living in the other four smaller urban regions combined. We present the resulting mean willing to pay Downloaded from by guest on December 8, 2023. Copyright 2022 and mean willingness to accept results in Table A.2. The results are similar for Edmonton and Calgary, and somewhat lower for the other urban areas. For the present purposes, we focus on results for the pooled sample. For the welfare calculations, this pooling required an assumption of equal marginal willingness to pay and willingness to accept across the study areas. We consider the results most representative of the Edmonton and Calgary regions. WTP for Farmland Preservation For comparison purposes, we estimated and present three versions of the WTP model in Table 3. The first version is estimated for the full sample, with “yes but uncertain” responses recoded as no. This is our preferred model. The results are reported in the second and third columns of Table 3. The second version is estimated for the full sample, with no recoding, and the results are reported in the fourth and fifth columns of Table 3. That is, “yes, but uncertain” responses are treated in the same way as “yes, but certain.” The third model is estimated with data for the sub-sample of “strategy believers”, those who indicated in the ancillary question that they regarded the study to be consequential, and reported in columns six and seven. The estimates of the price coefficient are statistically significant and virtually identical in all three models, while the estimate of the ASC is lower for the preferred (recoded) model, and virtually identical for the other two models. The results for the different types of land use indicate few statistically significant differences between land uses, although high heterogeneity of land use preference in all models. The estimated value for the vegetable land use variable is statistically significant, but only for the model estimated without uncertainty recoding. 18 <Table 3 about here> We then used the estimated parameters from the RPL model to estimate MWTP for the farmland preservation scenarios according to equation 4 (Table 4). <Table 4 about here> The results of our preferred model indicate that the average respondent is willing to make a one-time payment of between $340.15 and $475.90 (depending on the land use change prevented) to preserve 1,000 acres of farmland that would otherwise be converted into non-agricultural uses. MWTP for the Downloaded from by guest on December 8, 2023. Copyright 2022 sub-sample of strategy believers is higher than for the whole sample, ranging from $472.16 to $832.02. Strategy believers are willing to pay 27% to 92% more than the whole sample. The finding of higher MWTP for respondents who feel that the survey is policy consequential is consistent with the finding of Vossler et al (2012). Table 4 also shows MWTP estimates for the full sample without recoding. MWTP are also somewhat higher than the base model, with a range of $602.35-$828.75, 48% to 119% higher than the base model. Clearly, uncertainty recoding had a material effect on the MWTP results. As suggested by Champ and Bishop (2001), uncertainty recoding produces a more conservative estimate of WTP. WTA Compensation for Development Next we focus on the results of the RPL models of willingness to accept compensation for the loss of passive-use value due to conversion of additional agricultural land into developed uses. The results presented in Table 5 show that price (a one-time tax reduction) always has a positive effect on respondents’ utility and that all respondents prefer the status quo level of conversion over additional conversion. Respondents from the full sample indicate no preference for conversion of commercial vegetable or grain land, but they are less likely to accept compensation for conversion of grazing land. Regarding alternative developed land use, respondents are more willing to allow conversion to light industrial compared to conversion to retail or residential. Significant coefficients of the standard deviation indicate that the RPL model captured large unobserved heterogeneity in WTA. MWTA estimates are presented in Table 6 for the preferred base model (with uncertainty recoding), for the strategy believers sub-sample, and for the full sample without uncertainty recoding. 19 Respondents who believed that these strategies would be implemented place greater value on the preservation scenarios compared with the current development trend. MWTA for those who believed in the consequentiality of the study (strategy believers) were willing to accept 21-93% more than the whole sample, as indicated by the Ratio 3/2 column. MWTA estimates without recoding were much higher than MWTA with recoding, ranging from 122% to 344% higher. As expected, the uncertainty recoding procedure produced a much more conservative estimate of MWTA. We do not have data on actual willingness to pay for farmland conservation in Alberta, so cannot independently verify the results with and without uncertainty recoding. <Table 5 about here> Downloaded from by guest on December 8, 2023. Copyright 2022 <Table 6 about here> Comparison of WTP and WTA Finally, we present our comparison of results for WTP and WTA, summarized in Table 7. The first indicator is the completion rate of each survey. The online survey involved a total of 1,303 complete respondents. The survey completion rate was 33.84% for the WTP survey and 37.71% for the WTA survey. The slightly higher completion rate for the WTA suggests that the WTA presented no greater cognitive dissonance than the WTP survey (p<0.05). As indicated above, however, we note that the Flesch-Kincaid reading level is 19.3 for the WTP survey and 18.5 for the WTA survey, and that less well-educated respondents expressed higher uncertainty in responding to the WTA questions. 28 WTP respondents (4.36%) and 13 WTA respondents were identified as project-rejecters (1.97%) on the basis of their responses to the open-ended question at the end of the survey. The number of WTP project rejecters is somewhat higher than for WTA, although still relatively low (4.63%). The difference is statistically significant). Overall, we conclude that most respondents accepted the plausibility of both the WTP and WTA scenarios. From the ancillary question regarding consequentiality, we found that 54.9% of respondents believed that the preservation strategies evaluated through the WTP are likely or very likely to be implemented, while 51.5% of respondents believed that the development strategies evaluated through the WTA are likely or very likely to be implemented. This 3.3% difference is statistically significant. This indicates 20 that belief in consequentiality was similar for the two groups, although marginally higher for WTP than WTA (p<0.01). A fourth indicator of the reliability of the survey design is the percentage of respondents who indicated “uncertain” or “very uncertain” in their reactions to the scenarios. The percentage of uncertain responses for the WTP survey was 8.6%, while the percentage of uncertain responses for WTA was 15.3%, a statistically significant difference. Thus, we conclude that respondents in the WTP case were somewhat more certain of their answers to the WTP survey than respondents in the WTA case. As discussed in the methods section, we assume that this preference uncertainty reflects a combination of cognitive dissonance and social desirability bias toward farmland preservation over development. Downloaded from by guest on December 8, 2023. Copyright 2022 Theses biases appears to have somewhat greater effect on WTA than WTP. To further illustrate the differences between estimates of WTA and WTP, the ratios of MWTA / MWTP for the nine strategies were calculated for the preferred base model and alternative models (Table A.6). For our preferred base model with the full sample and uncertainty recoding, we found that the ratio of MWTA to MWTP to range from -0.16 to 3.05, with an average of 1.21 across the nine scenarios. Overall, the ratios are higher for the strategy believers, ranging from 0.34 to 2.70, with an average of 2.11, and the full sample without recoding, ranging from 0.83 to 3.35, with an average of 1.42. All of these ratios compare favourably to the review conducted by Tuncel and Hammitt (2014). For the 76 studies that they considered, Tuncel and Hammitt (2014) found the average ratio to be about 3.00 across all goods, 6.23 for environmental goods, and 3.93 for other public or non-market good. It is noteworthy that we found a few MWTA / MWTP ratios to be less than 1. For example, for the ‘strategy believers’ group, grain-vegetable and livestock-light industrial are only 0.34 and 0.3, respectively, which means that MWTA is considerably smaller than MWTP. A potential reason is that although MWTA in these two categories are significant, the p values are just at the 10% level. Our WTP and WTA respondents included persons who were potential or actual tax payers. We estimated versions of both models for the sub-sample of renters. The RPL model results for the WTP and WTA models are reported in Table A.3 and the MWTP estimates are reported in Table A.4. While the WTP results are similar to the results for the base WTP model, the WTA results are much lower, not significantly different from 0. <Table 7 about here> 21 6. Discussion Few previous studies have applied the WTA framing to estimate the public values given up when agricultural land is converted into developed uses. We are aware of no previous study that has estimated WTA applying the consequentiality principles identified by Vossler et al (2012) for WTP and validated by Lloyd-Smith and Adamowicz (2018) for WTA. Yet, farmland preservation is a matter of public concern in many contexts. Policy makers and planners are expected to manage the tradeoffs between public and private interests associated with different land uses in urban-rural fringe areas. Downloaded from by guest on December 8, 2023. Copyright 2022 We conclude that the results of this study show that willingness to accept for the loss of the public good values of farmland can be reliably estimated using best practice choice experiment methods. This includes due attention to study framing to meet consequentiality requirements, cognitive burden, payment vehicle, use of follow-up questions, recoding for uncertainty and efficient estimation. Using four measures of survey reliability, we found two measures that indicated higher reliability of WTA -survey completion rate, rate of project rejection – and two measures that indicated higher reliability of WTP – strategy believer rate and scenario uncertainty.1 All of the differences in study performance are small, except for the rate of scenario uncertainty. Respondents expressed more uncertainty about their responses to the WTA framing. We postulate that this difference may reflect differences in respondents’ understanding of the consistency of scenarios with the de facto property rights situation to agricultural land in the peri-urban areas around Alberta’s cities. As indicated in the context section above, both the Alberta Land Use Framework and Municipal Government Act make it clear that private land rights are limited and that municipalities are expected to consider the public interest in allowing changes in zoning. This might lead to a social desirability bias for respondents to WTP scenarios to agree to high payments for preservation and WTA respondents to disagree to accept high levels of compensation for development. In practice, however, other supporting research shows that many farm owners believe that they have a right to convert their land to more developed uses and many municipal governments are reluctant to deny applications for land use redesignation made by those farm owners (personal communication with municipal authorities in Parkland and RockyView Counties). Municipal land use plans are not strictly followed, and municipal 1 We urge other researchers to pay attention to the ease of comprehension of the conservation and development scenarios. As reported in footnote 3, these surveys had high reading levels scores which may have contributed to respondents’ uncertainty. 22 governments are not held to account for complying with those plans. Another possible cause of the higher rate of uncertainty for WTA is that the idea of financial compensation for loss of public goods associated with agricultural land is less familiar than the idea of payment toward farmland preservation. We account for potential social desirability bias or cognitive dissonance by recoding “yes, but uncertain” WTP responses as no, and “no, but uncertain” WTA responses as yes. With this recoding, average WTA across the 9 land use change scenarios was only 21% higher than WTP; without this recoding, average WTA was 42% higher than WTP. Both of these gaps are considerably lower than gaps found in most previous WTA / WTP comparisons. This smaller gap may be partly due to the rigorous study design. However, we also need to accept that the smaller gap may reflect lower loss aversion. Downloaded from by guest on December 8, 2023. Copyright 2022 Respondents who had not previously thought of farmland conversion as the loss of public good benefits of open space might perceive less loss. One result that bolsters that proposition is that respondents who believed in the plausibility (and potentially consequentiality) of the preservation and development strategies had higher average willingness to pay and willingness to accept compensation than the full sample. The WTA-WTP gap was also larger, possibly indicating greater loss aversion. Following Hanemann (1991), lower loss aversion might also reflect the fact that respondents perceived there to be many substitutes for the public goods provided by agricultural land. Further exploration of this topic is warranted, possibly involving laboratory experiments that vary the certainty of property rights. We recommend that future studies of farmland preservation revive the logic of Knetsch (2007, 2020) and estimate the welfare measure that best matches the context. Ambiguous property rights and the de facto reference scenario of continual development may leave no clear choice between the two welfare indicators. In that context, accurate measurement of either welfare measure requires the analyst to present the scenario in terms that gain higher scenario acceptance by respondents, augmented by uncertainty questions and recoding. We argue that the WTA approach is more compelling than WTP, however, given the reversibility of land use changes. That is, today’s decision to preserve an area of farmland can be reversed tomorrow at relatively little cost, while today’s decision to develop that same area is unlikely to ever be reversed. Acknowledgements Financial support for this research was provided by the Alberta Real Estate Foundation, the Max Bell Foundation, the Calgary Regional Partnership, and the Alberta Land Institute. The authors thank Rachel 23 de Vos (Bocock) for first suggesting that we consider the duality between willingness to pay and willingness to accept compensation for preserved farmland. Haoluan Wang contributed methodological assistance for the survey and analysis. We acknowledge constructive comments from Edward Barbier and Ryan Abman, as well as the editor and two anonymous reviewers of Land Economics. References Androkovich, Robert, Ivan Desjardins, Gordon Tarzwell, and Peter Tsigari. 2008. “Land Preservation in British Columbia: an Empirical Analysis of the Factors Underlying Public Support and Willingness to Downloaded from by guest on December 8, 2023. Copyright 2022 Pay.” Journal of Agricultural and Applied Economics 40(3): 999-1013. Arrow, Kenneth, Robert Solow, Paul R. Portney, Edward E. Leamer, Roy Radner, Howard Schuman. 1993. Report of the NOAA Panel on Contingent Valuation. Federal Register, 58(10), 4601-4614. Bateman, Ian J., R.T. Carson, Brett Day, Michael Hanemann, Nick Hanley, Tannis Hett, Michael JonesLee, et al. 2002. Economic Valuation with Stated Preference Techniques: a Manual. Economic Valuation with Stated Preference Techniques: a Manual. Edward Elgar Publishing: Cheltenham, UK. Bergstrom, John C. and Richard C. Ready. 2009. What Have We Learned From Over 20 Years of Farmland Amenity Valuation Research in North America? Review of Agricultural Economics, 31(1), 2149. Boas, Taylor C., Dino P. Christenson, and David M. Glick. 2020. Recruiting Large Online Samples in the United States and India: Facebook, Mechanical Turk, and Qualtrics. Political Science Research and Methods 8(2): 232-250. Carson, Richard T. and Theodore Groves. 2007. Incentive and Information Properties of Preference Questions. Environmental and Resource Economics 37(1): 181-210. Champ, Patricia and Richard C. Bishop. 2001. Donation Payment Mechanisms and Contingent Valuation: an Empirical Study of Hypothetical Bias. Environmental and Resource Economics, 19 (4): 383-402. City of Calgary (2008). Land Use Bylaw. IP 2007. City of Calgary. City of Edmonton (2017). Zoning Bylaw 1280. City of Edmonton. 24 Duke, Joshua M., Allison M. Borchers, Robert J. Johnston, and Sarah Absetz. 2012. Sustainable Agricultural Management Contracts: Using Choice Experiments to Estimate the Benefits of Land Preservation and Preservation Practices. Ecological Economics, 74, 95-103. Groothuis. Peter A. and John C. Whitebread. 2002. Does Don’t Know Mean No? Analysis of `Don’t Know’ Responses in Dichotomous Choice Contingent Valuation Questions. Applied Economics 34 (15): 1935-1940. Haarsma, Darren, and Feng Qiu. 2017. Assessing Neighbor and Population Growth Influences on Agricultural Land Conversion. Applied Spatial Analysis and Policy, 10(1): 21-41. Downloaded from by guest on December 8, 2023. Copyright 2022 Hanemann, W. Michael. 1978. A Methodological and Empirical Study of the Recreation Benefits from Water Quality Improvement. Department of Agricultural and Resource Economics, University of California. Hanemann, W. Michael. 1984. Welfare Evaluations in Contingent Valuation Experiments with Discrete Responses. American Journal of Agricultural Economics, 66(3): 332-341. Hanemann, W. Michael. 1991. Willingness to Pay and Willingness to Accept: How Much Can They Differ? The American Economic Review 81(3): 635-647. Hensher, David, Nina Shore, and Kenneth Train. (2005). Households’ Willingness to Pay for Water Service Attributes. Environmental and Resource Economics 32(4): 509-531. Herriges, Joseph, Catherine Kling, Chih-Chen Liu, and Justin Tobias. 2010. What are the Consequences of Consequentiality? Journal of Environmental Economics and Management 59(1): 67-81. Johnston, Robert J., Kenneth J. Boyle, Wiktor Adamowicz, Jeff Bennett, Roy Brouwer, Trudy Ann Cameron, W. Michael Hanemann et al. 2017. Contemporary Guidance for Stated Preference Studies. Journal of the Association of Environmental and Resource Economists 4(2): 319-405. Kaplinsky, Eran and David Percy. 2016. A Guide to Property Rights in Alberta. Alberta Land Institute. (www.albertalandinstitute.ca/resources). Kline, Jeffrey and Dennis Wichelns. 1994. Using Referendum Data to Characterize Public Support for Purchasing Development Rights to Farmland. Land Economics 70(2): 223-233. Kline, Jeffrey and Dennis Wichelms, 1998. Measuring Heterogeneous Preferences for Preserving Farmland and Open Space. Ecological Economics 26(2): 211-224. 25 Knetsch, Jack L. 2007. Biased Valuations, Damage Assessments, and Policy Choices: The Choice of Measure Matters. Ecological Economics 63(4): 684-689. Knetsch, Jack L. 2020. Behavioural Economics, Benefit-Cost Analysis, and the WTP versus WTA Choice. International Review of Environmental and Resource Economics 14(2-3): 153–196. Li, Chuan-Zhong, Jari Kuuluvainen, Eija Pouta, Mika Rekola, and Olli Tahvonen. 2004. Using Choice Experiments to Value the Natura 2000 Nature Conservation Programs in Finland. Environmental and Resource Economics 29(3): 361-374. Lloyd-Smith, Patrick and Wiktor Adamowicz. 2018. Can Stated Measures of Willingness-to-Accept Be Downloaded from by guest on December 8, 2023. Copyright 2022 Valid? Evidence from Laboratory Experiments. Journal of Environmental Economics and Management 91: 133-149. Loomis, John B. 2014. 2013 WAEA Keynote Address: Strategies for Overcoming Hypothetical Bias in Stated Preference Surveys. Journal of Agricultural and Resource Economics 39(1): 34-46. Lopez-Becerra, Erasmo and Francisco Alcon. 2021. Social Desirability Bias in the Environmental Economic Valuation: An Inferred Valuation Approach. Ecological Economics 184: 106988. McFadden, Daniel. 1974. The Measurement of Urban Travel Demand. Journal of Public Economics 3(4): 303-328. Ready, Richard, Patricia A. Champ, and Jennifer L. Lawton. (2010). Using Respondent Uncertainty to Mitigate Hypothetical Bias in a Stated Choice Experiment. Land Economics 86(2): 363-381. Rose, John M., Michiel C.J. Bliemer, David A. Hensher, and Andrew T. Collins. 2008. Designing Efficient Stated Choice Experiments in the Presence of Reference Alternatives. Transportation Research Part B: Methodological 42(4): 395-406. Rose, John M. and Michiel M.C. Bliemer. 2013. Sample Size Requirements for Stated Choice Experiments. Transportation 40(5): 1021–1041. Del Saz-Salazar, Salvador. 2013. Respondent Uncertainty and Willingness-to-Accept Estimates: a Comparative Analysis of Different Approaches. (archiveo.alde.es/encuentros.alde.es/anteriores/xvieea/trabajos/s/pdf/24.pdf). Statistics Canada (no date). Table 32-10-0153-01. Total area of farms and use of farm land, historical data. DOI:https://10.25318/3210015301-eng. 26 Tunçel, Tuba and James K. Hammitt. 2014. A New Meta-Analysis on the WTP/WTA Disparity. Journal of Environmental Economics and Management 68(1): 175-187. Tversky, A., & Kahneman, D. (1979). Prospect Theory: An Analysis of Decision Under Risk. Econometrica 47(2): 263-291. Vossler, Christian A., Maurice Doyon, Daniel Rondeau. 2012. Truth in Consequentiality: Theory and Field Evidence on Discrete Choice Experiments. American Economic Journal: Microeconomics 4(4): 145-71. Wang, Haoluan and Brent M. Swallow. 2016. Optimizing Expenditures for Agricultural Land Downloaded from by guest on December 8, 2023. Copyright 2022 Conservation: Spatially-Explicit Estimation of Benefits, Budgets, Costs and Targets. Land Use Policy 59: 272-283. Willis, Ken. 1982. Green Belts: an Economic Appraisal of a Physical Planning Policy. Planning Outlook 25(2): 62-69. Yu, Jia and Ken Belcher. 2011. An Economic Analysis of Landowners’ Willingness to Adopt Wetland and Riparian Conservation Management. Canadian Journal of Agricultural Economics 59(2): 207-222. 27 Table 1: Binary choice experiment attributes and attribute levels for the WTP and WTA surveys Downloaded from by guest on December 8, 2023. Copyright 2022 Attribute Level Explanation Type of Current Grain or oilseed farming Agricultural Use Livestock grazing on native pasture (WTP & WTA) Commercial vegetable farm Type of development Residential Major types pf urban without preservation Light Industrial development without (WTP & WTA) Retail conservation in your areas 50 One-time additional 100 cost to each taxpayer 300 ($) (WTP) 600 Major types of agricultural in your area One-time additional increase in property tax or rent to each taxpayer in your areas 1000 50 One-time compensation 100 One-time reduction in paid to each taxpayer 300 property tax or rent to ($) (WTA) 600 each taxpayer in your area 1000 Source: Authors. 28 Table 2. Demographic and Socio-economic Statistics for the Sample （N=643 for WTP, N= 660 for WTA） Demographic variables Gender Downloaded from by guest on December 8, 2023. Copyright 2022 Residence (region) Household income (before tax) Education Employment City Ownership Age Description WTP 38.26 61.59 0.16 42.15 41.84 5.29 5.13 3.27 2.33 14.02 25.55 21.34 19.31 10.90 8.88 3.11 22.71 Male Female Other Edmonton Calgary Lethbridge Red Deer Medicine Hat Grande Prairie Less than $30,000 $30,000 - $59,999 $60,000 - $89,999 $90,000 - $119,999 $120,000 - $149,999 Greater than $150,000 Lower than high school Completed high school Completed post-secondary technical school Completed university undergraduate degree Completed post-graduate degree (e.g., Master or Ph.D.) Working part-time Working full-time Retired Student Unemployed Other, please specify In a city Outside of a city Others, please specify Own residence Rent residence Own agricultural land 18-64 65+ Sample percentage (%) WTA Significancea 46.21 *** 53.79 *** 0.00 39.85 41.82 6.82 4.85 4.39 2.27 15.05 25.53 23.25 15.35 ** 11.25 9.57 1.82 26.52 Total 42.29 57.64 0.08 40.98 41.83 6.06 4.99 3.84 2.30 14.50 25.48 22.26 17.27 11.05 9.21 2.46 24.64 32.66 30.45 31.54 32.66 31.06 31.85 8.86 10.15 9.52 17.11 48.06 17.42 3.89 7.78 5.75 74.03 9.64 16.33 69.83 29.86 5.75 86.16 13.84 14.39 43.79 24.24 4.39 8.03 5.15 70.30 11.52 18.18 71.82 29.70 5.91 81.52 18.48 15.73 45.89 20.87 4.14 7.90 5.45 72.14 10.59 17.27 70.84 29.78 5.83 83.81 16.19 ** a. The significance of the differences between WTP and WTA samples gained from two-sample ttest. *** p<0.01, ** p<0.05, *, no symbol represents no significant differences. (Source: authors) 29 Table 3. Random Parameter Logit Model (RPL) Coefficient Estimates (WTP) (preferred model, without recoding and strategy believers) Attributes Price ASC vegetable Downloaded from by guest on December 8, 2023. Copyright 2022 livestock retail light industrial RPL (Preferred model) Coefficient Coefficient (Std. Dev.) -0.00188*** (0.000121) 0.763*** (0.0776) 0.13 (0.124) -0.0735 (0.125) -0.00467 (0.124) 1.888*** (0.201) 1.737*** (0.206) 1.858*** (0.187) -0.0513 2.067*** (0.141) (0.199) 6014.723 6086.665 RPL (without recoding) Coefficient Coefficient (Std. Dev.) 0.00186*** (0.000117) 1.119*** (0.0810) 0.272** 1.732*** (0.120) (0.189) 0.208 1.663*** (0.130) (0.188) 0.145 1.819*** (0.123) (0.183) 0.0592 1.669*** (0.128) (0.180) 5949.394 6021.336 AIC BIC Log -2997.3614 -2964.697 likelihood Number of 615 615 Individuals Number of 9,840 9840 Choices Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 30 Strategy believers Coefficient Coefficient (Std. Dev.) -0.00188*** (0.000156) 1.110*** (0.108) 0.195 (0.166) -0.225 (0.156) 0.147 (0.153) 1.837*** (0.271) 1.348*** (0.256) 1.626*** (0.228) 0.255 2.202*** (0.196) (0.298) 3368.716 3434.877 -1674.3581 345 5,520 Table 4. Estimated MWTP for Farmland Preservation Scenarios in Alberta (per household, next year only) (RPL: preferred model, strategy believers, without recoding) Preservation Scenario Downloaded from by guest on December 8, 2023. Copyright 2022 Grain or oilseed farming; Residential Commercial vegetable farm; Residential Livestock grazing on native pasture; Residential Grain or oilseed Farming; Retail Commercial vegetable farm; Retail Livestock grazing on native pasture; Retail Grain or oilseed farming; Light industrial Commercial vegetable farm; Light industrial Livestock grazing on native pasture; Light industrial Mean of 9 land use changes Preferred Model ($) 406.67*** 475.90*** 367.50*** 404.19*** 473.41*** 365.01*** 379.33*** 448.55*** Strategy believers ($) 592.01*** 695.91*** 472.16*** 670.18*** 774.08*** 550.33*** 728.12*** 832.02*** Ratio (3/2) 1.46 1.46 1.28 1.66 1.64 1.51 1.92 1.85 Without recoding ($) 602.35*** 748.53*** 714.26*** 680.56*** 828.75*** 792.48*** 634.23*** 780.41*** 340.15*** 608.27*** 1.79 746.14*** 406.75 565.67 1.39 725.30 Ratio (5/2) 1.48 1.57 1.94 1.68 1.75 2.17 1.67 1.74 2.19 1.78 *** p<0.01. Ratio (3/2) is the ratio of the MWTP for strategy believers (column 3) and base model (column 2). Ratio (5/2) is the ratio of the MWTP without recoding (column 5) to base model (column 2). Source: Authors 31 Table 5. Random Parameter Logit (RPL) Model Coefficient Estimates (WTA) (preferred model, without recoding and strategy believers) Attributes Price ASC vegetable livestock Downloaded from by guest on December 8, 2023. Copyright 2022 Retail light industrial Preferred model Coefficient Coefficient (Std. Dev.) 0.000675*** (9.74e-05) -0.333*** (0.0694) 0.0494 0.176 (0.0782) (0.335) -0.270** 1.866*** (0.121) (0.196) -0.148 2.440*** (0.136) (0.220) 0.373*** -0.797*** (0.0885) (0.128) 6712.539 6784.988 -3346.2695 RPL (without recoding) Strategy believers Coefficient Coefficient (Std. Dev.) 0.000737*** 0.000849*** (0.000131) (9.76e-05) -0.499*** -1.087*** (0.0955) (0.0747) 0.0192 -0.00242 0.144 0.134* (0.117) (0.105) (0.311) (0.0781) 1.744*** -0.301* 1.686*** -0.601*** (0.185) (0.157) (0.227) (0.135) -0.567*** 2.009*** -0.297* 1.936*** (0.194) (0.163) (0.245) (0.139) -0.00790 0.315*** 0.316 0.404*** (0.146) (0.110) (0.241) (0.0804) 6298.831 3543.133 6341.28 3609.06 -3139.415 -1761.5665 AIC BIC Log likelihood Number of 647 647 Individuals Number of 10352 10352 Choices Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 32 337 5392 Table 6. Estimated MWTA for Conservation Strategy in Alberta (per household, next year only) (RPL: preferred model, without recoding and strategy believers) Conservation Strategy Downloaded from by guest on December 8, 2023. Copyright 2022 Grain or oilseed farming; Residential Commercial vegetable farm; Residential Livestock grazing on native pasture; Residential Grain or oilseed Farming; Retail Commercial vegetable farm; Retail Livestock grazing on native pasture; Retail Grain or oilseed farming; Light industrial Commercial vegetable farm; Light industrial Livestock grazing on native pasture; Light industrial Mean of 9 land use changes Preferred Model ($) Strategy believers ($) Ratio (3/2) Without Recoding ($) Ratio (5/2) 492.99*** 419.81*** 893.58*** 711.86*** 638.68*** 1112.45*** -59.69 -132.87 676.64*** 679.92*** 1084.67*** 1079.18*** 1082.45*** 1487.20*** 249.15* 252.42* 1.37 1.62 1.21 1.52 1.69 1.34 - 1280.23*** 1122.09*** 1987.31*** 1948.12*** 1789.97*** 2655.2*** 804.83*** 646.68*** 340.90* 657.17*** 1.93 2.60 2.67 2.22 2.74 2.80 2.39 4.44 490.86 805.42 1.64 1511.90*** 1527.3 3.11 *** p<0.01, * p<0.1. Ratio (3/2) is the ratio of the MWTP for strategy believers (column 3) and base model (column 2). Ratio (5/2) is the ratio of the MWTP without recoding (column 5) to base model (column 2). Source: Authors 33 Downloaded from by guest on December 8, 2023. Copyright 2022 Table 7. Comparison of indicators between WTP and WTA WTP WTA T test between the difference of two samples Survey completion rate 33.84% 37.71% +2.4969** Project rejectors 4.63% 1.97% -2.2878** Strategy believers 54.90% 51.50% -3.9906*** Scenario uncertainty rate 8.60% 15.30% 10.3194*** Mean WTP & WTA across 9 land use changes (Cdn $) (preferred model) $406.75 $490.86 NA Range of WTP & WTA across 9 land use changes (preferred model) $340.15 - $473.41 -$132.87 $1,112.45 NA Mean WTP & WTA across 9 land use changes (Cdn $) (without uncertainty recoding) 725.30 $1527.30 NA Range of WTP & WTA across 9 land use changes (without uncertainty recoding) $602.35 - $828.75 -646.68 - $1987.31 NA Mean WTP & WTA across 9 land use changes (Cdn $) (scenario believers) $565.67 $805.42 NA Range of WTP & WTA across 9 land use changes (scenario believers) $472.16 - $832.2 $249.15 $1487.20 NA *** p<0.01, ** p<0.05. Source: authors. 34 i To enhance consequentiality, we included the following statements in the introduction to the choice experiment scenarios. “Provincial and municipal governments make many decisions that affect land use. According to provincial policies, governments are expected to develop and follow land use plans that maintain the viability of agricultural operations. All municipalities and regions in Alberta are required to get input from local residents as they develop those plans. Some jurisdictions use survey like Downloaded from by guest on December 8, 2023. Copyright 2022 this to get such input. We will share our results with the provincial government and municipal governments in the areas covered by this study.” This statement was followed by the cheap talk text regarding the importance of honest responses: “It is very important that you choose as if this as a real vote. You need to imagine that you actually have to dig into your budget and pay the additional onetime increase in tax or rent costs associated with each of the proposed strategies. Remember, the results of this study will be shared with local municipalities. They may decide to change policy as a result. The cheap talk text for the WTA variant of the study was: “It is very important that you choose as if this was a real vote. You need to imagine that you would benefit from a one-time reduction in your property taxes or rent costs associated with each of the proposed strategies. You may have had experience with such incentives provided as tax credits. Remember, the results of this study will be shared with local municipalities. They may decide to change policy as a result.” In the ending ancillary questions, we asked “How likely do you feel the strategies presented in the survey could be really implemented?” with very likely, somewhat likely, uncertain, somewhat unlikely and very unlikely as possible responses. ii We considered the possibility that WTP and WTA responses might have been affected by the tax paying experience of our respondents. A one-time reduction or increase in property tax might be more credible than a one-time reduction or increase in rent. We considered the effects of this by estimating 35 the RPL models for sub-samples of renters and owners (not shown). There were no meaningful differences in results from the two samples, so we focus on the full sample results here. iii We now recognize that our consequentiality question was not ideal. The current preferred version of a consequentiality question is something like “do you feel that your vote (or the outcome of the survey) may actually have an impact on the policy outcome.” We acknowledge the input of the reviewer who Downloaded from by guest on December 8, 2023. Copyright 2022 raised this issue. iv Other studies have shown that samples collected by the survey company, Qualtrics, are good reflections of the population characteristics (Boaz et al., 2020). v The trap question is “Please select ‘agree’ for this line.” The five possible answers were “strongly agree”, “agree”, “neither agree nor disagree”, “disagree”, and “strongly disagree” were provided for participants to choose. Respondents who did not answer “agree” were removed from the sample. 36