Abstract
Existing agri-environment schemes have suffered from poor uptake and high burden. This article leverages behavioral economics to test choice framing in three hypothetical policy scenarios. Using a randomized survey experiment on U.K. farmers, we find that changing policy framing based on mental accounting and loss aversion can significantly influence agricultural policy–related decision making. Our findings highlight the following considerations for the design of future policy: (1) whether application costs are integrated into or segregated from a subsidy is important, (2) the labeling of agricultural schemes may affect expenditure allocation, and (3) reference points can affect the evaluation of new scheme alternatives.
1. Introduction
Under the European Union’s Common Agricultural Policy (CAP), member countries receive funding to subsidize agricultural activity. Farmers can apply for basic payments that depend on land entitlements and/or funding under rural development schemes. Unlike basic payments, rural development funds are often competitive and usually incentivize specific practices that are largely aimed at improving or protecting the rural environment. A major component of rural development funding takes the form of agri-environment schemes (AESs). AESs were originally introduced in the European Union in the mid-1980s and have been mandatory for member states since 1992 (Hodge and Reader 2010). They were designed to improve environmental outcomes such as biodiversity and shift farmer dispositions toward environmental protection (Ovenden, Swash, and Smallshire 1998; Riley 2016). However, participation in AESs has been lower than desired (Hejnowicz, Rudd, and White 2016; Pellegrin et al. 2018), often below what would be predicted given the level of financial return (Vanslembrouck, Huylenbroeck, and Verbeke 2002). This leads to a natural question: how could we improve scheme uptake without necessarily increasing the level of financial compensation?
Investigations into what drives farmers to participate in these schemes have found that a range of economic factors such as compensation, transaction costs, application burden, and farm structural factors significantly influence the decision to participate (see Lastra-Bravo et al. 2015 for a review; Siebert, Toogood, and Knierim 2006; Defrancesco et al. 2008; Hynes and Garvey 2009; Christensen et al. 2011; Pedersen et al. 2012; Murphy et al. 2014). Opportunity costs coupled with restrictive prescriptions on practice associated with schemes have also been put forward as an explanation for lower-than-expected participation (Defrancesco et al. 2008; Espinosa-Goded, Barreiro-Hurlé, and Ruto 2010). This is particularly true for smaller farms (Ruto and Garrod 2009). In addition, there is a rich literature highlighting the importance of sociological and psychological constructs in understanding the behavior of farmers in this area. Some of the factors highlighted as important are farmer identity (e.g. Cullen et al. 2020), value orientations (Gasson 1973), and social embeddedness/peer influence (de Krom 2017; Cullen et al. 2020).
Given the lower than anticipated uptake, literature has emerged that leverages insights from social psychology to test the effectiveness of behavior-oriented interventions for encouraging participation in AESs (Behaghel, Macours, and Subervie 2019; Thoyer and Préget 2019). These studies have typically used experimental designs ranging from randomized surveys with informational treatments (Kuhfuss et al. 2016b; Howley and Ocean forthcoming), laboratory experiments (Holst, Musshoff, and Doerschner 2014; Dörschner and Musshoff 2015; Buchholz, Holst, and Musshoff 2016; Ferré, Engel, and Gsottbauer 2018; Peth et al. 2018; Thomas et al. 2019), and to a lesser extent randomized controlled trials (Wallander, Ferraro, and Higgins 2017) as a way to test the effectiveness of behavioral interventions (or “nudges”). Notwithstanding recent developments, using nudges in the agricultural sphere is still very limited in contrast to other sectors (e.g., health, development, and education). As an illustration, Dessart, Barreiro-Hurlé, and van Bavel (2019) noted that the four annual reports by the pioneering U.K. Behavioural Insights Team include only one mention of agriculture, despite showcasing hundreds of behavior-oriented interventions for policy.1
Moving forward, it is likely that a focus on the environment will remain one of the cornerstones of further CAP reform. Given its withdrawal from the European Union, the United Kingdom’s situation is more complicated once the current CAP agreement ends. It is still unclear which direction the U.K. government will take when it comes to policy support (see Franks 2016 for a discussion). What seems relatively clear is that the environment is likely to remain a point of focus for any support measures. Although there could be a reduction in overall funding, there may be an opportunity to reform and simplify legislation or take advantage of behavioral interventions designed to encourage pro-environmental practices. In light of this, our goal here is to explore policy framing as a tool for improving the design of future agricultural policy by making schemes more attractive, and therefore potentially increase the net impact of AESs.
Many other studies in the agricultural literature have made suggestions for policy design changes through choice experiments. Choice experiments allow for the inference of willingness to pay for different policy features or attributes (Ruto and Garrod 2009; Espinosa-Goded, Barreiro-Hurlé, and Ruto 2010; Christensen et al. 2011; Garrod et al. 2012; Schulz, Breustedt, and Latacz-Lohmann 2014; Villanueva et al. 2015; Kuhfuss et al. 2016a; Rocchi, Paolotti, and Fagioli 2017; Villamayor-Tomas, Sagebiel, and Olschewski 2019). Although these studies have been beneficial in providing a better understanding of policy preferences from a farmer’s perspective, they generally cannot provide a direct comparison of the effects of different policy modification “treatments” in the same way a traditional randomized experiment can. As an alternative to choice experiments, a few recent studies in this field have used surveys that randomly assign different question treatments to respondents to test for between-subject differences in responses (e.g., Kuhfuss et al. 2016b; Pellegrin et al. 2018). Although their use in agricultural settings has been limited, the randomized survey experiment has been extensively used in nonagricultural contexts to study issues such as immigration attitudes (e.g., Hainmueller and Hiscox 2010; Naumann, Stoetzer, and Pietrantuono 2018; Kaufmann 2019) and preferences for income redistribution (e.g., Cruces, Perez-Truglia, and Tetaz 2013; Kuziemko et al. 2015). The advantage of randomized surveys relative to choice experiments is that by establishing a counterfactual (i.e., using a control condition), one can estimate the causal effect of a proposed policy change.
We leveraged ideas stemming from mental accounting and loss aversion (Thaler 1985; Tversky and Kahneman 1991) to experimentally test three hypothetical policy modifications that we argue can be used to improve the design of AESs. We used these concepts as a basis for our experimental interventions chiefly because of the large body of evidence that supports them in the literature, coupled with the fact that the theories appear to be closely related to behavioral factors that we suggest can influence the uptake of AESs and subsequent expenditure decisions. Outside of the agricultural sphere, prior research has pointed toward the importance of mental accounting and loss aversion in explaining a variety of behaviors. The tendency to separate money into different mental accounts can explain a variety of seemingly puzzling consumption and savings decisions, such as under- and overconsumption (e.g., Heath and Soll 1996; Prelec and Loewenstein 1998; Read, Loewenstein, and Rabin 1999; Zhang and Sussman 2017). Recent experimental studies have shown that mental accounting can explain high levels of tax compliance (Muehlbacher, Hartl, and Kirchler 2017), and the lack of fungibility of money (Abeler and Marklein 2017). In a farming context, we posit that mental accounting may affect (1) whether farmers believe a scheme is financially worthwhile to adopt based on how the costs and benefits are combined (i.e., whether they are integrated or segregated), and (2) how the label attached to an incentive scheme would influence how payments would be allocated across different types of expenditure.
“Loss aversion” refers to the tendency for losses to be given more weight than equivalently sized gains. Empirical studies have shown that loss aversion drives investors to hold on to losing investments too long to avoid realizing a loss via the “disposition effect” (e.g., Odean 1998; Weber and Camerer 1998; Barber et al. 2007), it leads to low-profitability sectors being given more protection in international trade policy (Freund and Ozden 2008; Tovar 2009), and it may be partly responsible for the endowment effect (Morewedge and Giblin 2015). In agriculture, for example, loss aversion has been shown to partly explain a lower-than-expected demand for rainfall index insurance (Lampe and Würtenberger 2019). In this article, we posit that loss aversion may be a possible barrier to the uptake of new agricultural schemes (i.e., due to perceived losses relative to the status quo).
Our randomized survey experiment generated three main results regarding the framing of agricultural policy that may improve the success of any new schemes that may be devised in the future. First, farmers are willing to exchange a larger amount of money for reduced application time when it is framed as a reduction in the initial subsidy offered rather than when it is framed as a standard subsidy combined with a separate cost that a farmer must pay. This suggests that farmer welfare may be improved if application costs are integrated into the initial subsidy payment. Second, how AESs are named can change how a subsidy payment is allocated across expenditure categories. We found that using an environmentally oriented name for a financial incentive scheme may encourage farmers to spend additional income on environmental initiatives, even if no restrictions are placed on expenditure. Finally, changing the framing of an existing scheme can affect the preference relationship between new scheme alternatives. In practice, this means that new schemes should be designed carefully to avoid features that compare unfavorably with existing schemes, because this may adversely affect their attractiveness even if the new scheme provides other benefits. Although the United Kingdom is our case study, our findings illustrate the importance of leveraging insights from behavioral economic theories for agricultural policy redesign more generally. Our findings also add to the empirical evidence surrounding the cognitive factors of farmer decision making (see Dessart, Barreiro-Hurlé, and van Bavel 2019). The three experiments are discussed in more detail in the next section.
2. Experimental Design
We discuss background literature and theory to help us form three separate hypotheses. From this, we describe the design of three different experimental questions that test these hypotheses.
Trading Money for Time: Segregation versus Integration
The complexity of the application process appears to be one important reason for the poor uptake of recent AESs, particularly for higher-level stewardship schemes (Hejnowicz, Rudd, and White 2016). Many farmers appear to rely on paid advisers to manage the bureaucracy involved in applying for and managing these schemes (Hejnowicz, Rudd, and White 2016). This adds operating costs that could otherwise be used for more productive purposes. It may also be possible that smaller farmers who cannot afford an adviser are discouraged from applying entirely. Previous work has suggested that farmers may be willing to accept a lower payment if application assistance was free of charge (Christensen et al. 2011). Given that the overall pool of agricultural subsidy may be reduced in the United Kingdom when the CAP is replaced, if farmers were willing to trade subsidy for reduced application burden, the government may be able to compensate those who may have to accept less money by reducing (or eliminating) application burden. Although reducing application burden will no doubt be an effective strategy for increasing uptake, drawing on insights from mental accounting (Thaler 1985), we suggest that trading the same amount of money for a unit reduction in application burden will generate higher utility when it is framed as an integrated trade-off (the loss is reflected in a lower lump-sum subsidy) rather than a segregated one (the loss is a lump-sum subsidy coupled with a separate payment). Therefore, we are interested in two questions: (1) how much a farmer is willing to pay to reduce application time, and (2) the difference between paying to reduce application time and accepting less subsidy in exchange for a reduced application time.
To answer these questions, we designed a hypothetical task with multiple conditions, based on mental accounting (Thaler 1985, 1999). Suppose we have two sums of money x ≥ 0, and y ≥ 0, such that x > y. Using Thaler’s (1985) notation, consider a mixed gain of (x,−y), and a value function v(.) of the form described in Kahneman and Tversky (1979), which reflects the psychological valence associated with a monetary outcome. This value function captures the idea that losses are felt more strongly than equal-sized gains. Based on this, the following inequality must hold true: v(x−y) > v(x) + v(−y). This is shown graphically in Appendix A. In other words, because losses loom larger than gains, receiving a combined gain and loss (integration) is preferred to receiving a gain and loss separately (segregation), that is, receiving one sum equal to x−y would be preferred to separately receiving x and having to give up y. To use an illustrative example, the net utility from receiving a subsidy of £10,000 would be greater than if farmers were presented with a subsidy of £11,000 along with having to make a payment of £1,000. In theoretical terms, this means that if we were to allow farmers to indicate either how much of their subsidy (x) they would be willing to pay to reduce their application time (y1) or how much payment they would accept overall in exchange for the same application time reduction (x−y2), then for both options to be valued equivalently, it must be that y2 > y1 because v(x−y1) > v(x) + v(−y1), and so v(x−y2) = v(x) + v(−y1) only when y2 > y1. That is, the reduction in the single payment one is willing to accept should be greater than the outgoing payment one would be willing to pay.
To test whether our prediction holds empirically in an agricultural policy context, we devised a simple trade-off task with two conditions (Figure 1). We posed a hypothetical situation where an initial subsidy scheme offers £10,000 in exchange for 10 hours of application time (framed as time taken to complete an application form for a financial incentive scheme). A new time-saving scheme is made available that cuts the application time to one hour but requires a separate fee to be paid. Application costs are therefore segregated from the subsidy. The respondent provides the maximum fee they would be willing to pay to prefer the time-saving scheme. In the treatment condition, instead of a fee, the respondent must state the minimum subsidy they are willing to accept to prefer the time-saving scheme. Standard utility theory predicts that the willingness to pay (WTP) in the control condition and the willingness to accept (WTA) in the treatment condition should be identical, since the tradeoff is identical (a nine-hour reduction in application time). However, based on the theoretical argument at the end of the previous paragraph, we hypothesize that the mean fee one is willing to pay will be less than the mean reduction in subsidy one is willing to accept. If this hypothesis is correct, in practical terms it would imply that farmers may be more likely to take up a scheme if application costs were integrated into the initial subsidy provided, rather than if they were provided with a higher headline level of subsidy that required separate administration costs to be paid.
The Control (top) and Treatment (bottom) Conditions for a Question Testing the Effect of Mental Accounting on the Amount of Money Traded for a Reduction in Application Time
Although it was not a primary hypothesis, we included a third condition for the money versus application time question (see Appendix B). This treatment was identical to the control condition in that it tested the WTP to reduce application time in a segregated fashion. However, in this treatment, the application time in the time-saving scheme was zero hours rather than one hour. This treatment allowed us to estimate the marginal value of a one-hour reduction in application time when the reduction is near an extreme (i.e., zero). We anticipated this result may prove interesting, given the usual theoretical assumption that utility or value functions have a very steep gradient in the neighborhood of zero.
Scheme Labeling
The majority of subsidy payments in the United Kingdom under the CAP are currently provided within the basic payment scheme (BPS) stream. These payments are based on the size of any eligible land holding of at least 5 ha that is used for “agricultural activity.”2 This is defined as producing, rearing, or growing agricultural produce, as well as land that is kept in a state suitable for grazing or cultivation (Rural Payments Agency 2019). About a third of the payment is contingent on a set of greening rules being met. Any subsidy received can be used relatively freely by the recipient. Whereas classical economic theory suggests that a lump-sum transfer like this will be allocated by the recipient in a way that maximizes utility, regardless of how it is presented, previous work has shown that labeling payments can affect how individuals choose to spend the money. Labeling grants affects the marginal propensity to consume goods related to the grant label. For example, a disproportionate amount of money was spent on eligible foods from Supplemental Nutrition Assistance Program payments (the successor to the food stamp program) in the United States (Beatty and Tuttle 2015; Hastings and Shapiro 2018), winter fuel payments were used to a disproportionate extent to pay for fuel in the United Kingdom (Beatty et al. 2014), and a disproportionately high proportion of child benefits were spent on children’s clothing in the Netherlands (Kooreman 2000). In the latter two cases, spending behavior changed, even though there were no restrictions on how the payment must be used. Labeling effects have also been found in experimental work in the lab and in the field (Abeler and Marklein 2017), though the contexts for these experiments did not focus on realistic policy scenarios. The results from this literature provide empirical support for the theoretical predictions from mental accounting and narrow bracketing (i.e., ignoring other available income sources when making a decision) and violate the fungibility of money assumption that is often taken as a given in economics (Read, Loewenstein, and Rabin 1999; Thaler 1999).
Drawing on this literature, our objective was to test whether changing the name of a generic farm subsidy could influence spending intentions. We devised a question that posed a hypothetical new subsidy payment akin to the BPS and asked farmers to allocate a percentage of the subsidy to one of three broad expenditure categories: (1) farm/operating costs investment, (2) environmental management, and (3) household expenses and savings. Under these headings, a short list of examples of goods and services was provided to assist the respondent. To test the effect of labeling, three question versions were formulated. The only difference between the questions was the name of the new payment scheme. In all cases, it was made clear that the money could be used freely across expenditure categories. In the control condition, the scheme was called the simple payment scheme (SPS). This name was chosen to broadly correspond to the existing BPS, in that the name is neutral in relation to the kind of expenditure the subsidy might be expected to be used on. In the first treatment, the scheme was renamed the agricultural payment scheme (APS). This subtle change was designed to exploit the lack of fungibility observed in previous studies. In this condition, we expected that a lower portion of the income would be allocated to household expenditure and savings relative to the SPS condition, because one would expect that the word “agriculture” induces individuals to classify the subsidy into a mental account associated with expenditure related only to agriculture. Finally, in the second treatment, the scheme was renamed to the environmental protection scheme (EPS), and artwork of a green flower was placed next to the scheme title. This treatment was designed to induce respondents to associate the subsidy with a mental account for environmental practices. Therefore, we expected that a higher proportion of subsidy would be allocated to environmental management in the EPS condition, relative to the SPS condition.
Reference Points and Choosing between New Schemes
Given the United Kingdom’s exit from the European Union, policy makers may be interested in how alternative schemes are perceived by farmers who are already participating in a scheme, because this relative perception is likely to have an effect on the uptake of any new scheme. We hypothesize that framing alternative schemes differently depending on one’s current scheme may have an effect on choice, as a result of loss aversion. Loss aversion is a well-documented characteristic of prospect theory that suggests individuals experience losses with greater valence than an equivalent gain. This is represented by the “kink” at the origin of the value function (see Appendix A). An implication of loss aversion in situations of riskless choice is that the preference between two indifferent alternatives, X and Y, can be reversed depending on the reference point against which the alternatives are being compared (Tversky and Kahneman 1991). The second graph in Appendix A presents this in a diagram. X is more likely to be preferred than Y when the reference point is R1 rather than R2. This is because the difference between two disadvantages (i.e., reductions in an attribute) is viewed more strongly than the difference between two advantages (i.e., gains in an attribute). For example, from R1, X represents a loss of one in attribute 1 and a gain of one in attribute 2, whereas Y represents a loss of two in attribute 1 and a gain of two in attribute 2. The additional loss of one unit in attribute 1 will override the additional gain of one unit in attribute 2. The reverse argument holds for Y, which is more likely to be preferred from R2 than R1. Classical utility maximization would suggest that the initial endowment should not affect the choice between X and Y when they are both feasible options.
To test whether this finding is reproduced for farmers in the context of choosing between subsidy schemes, we presented a randomized choice question. We designed a classification system for financial incentive schemes that provide a rating for each scheme on two attributes: environmental benefits and profitability benefits. Each attribute has five possible values, ranging from very poor to very good. These values are made more quantifiable and easier to visually compare by attaching a number of stars to each one (very good is five stars, good is four stars, moderate is three stars, poor is two stars, and very poor is one star). Individuals were first told that the environmental and profitability benefits of a current scheme had been rated according to this system and were shown the results of this rating. This was designed to generate a reference point to which new schemes would be compared. We presented participants with two hypothetical financial incentive schemes: scheme X and scheme Y. Scheme X was defined as having good (four-star) environmental benefits and moderate (three-star) profitability benefits, and scheme Y was defined as having moderate (three-star) environmental benefits and good (four-star) profitability benefits. Individuals were asked whether they preferred scheme X or scheme Y.
The attributes of schemes X and Y do not change across treatments. However, the reference scheme against which individuals are comparing against does. The reference scheme is defined as having very good (five-star) environmental benefits and poor (two-star) profitability benefits in the control condition, and these attribute values are reversed in the treatment condition. Figure 2 shows the matrix of options presented under the control condition (top) and the treatment (bottom). If choices are made independent of the reference point (i.e., if initial endowments do not matter, as would be suggested by utility theory), there should be no difference in the proportion that chooses scheme X relative to scheme Y.3 However, we hypothesize that due to loss aversion, scheme X will be preferred by a greater proportion of farmers in the control condition than in the treatment condition. This is because the value function is steeper in losses than in gains, and the combination of a small loss and an equivalently small gain does not generate as much disutility as the combination of a large loss and equivalently large gain. If our hypothesis holds true, this suggests that any new formulation of agricultural policies after Brexit should take into consideration the advantages and disadvantages of new policies relative to existing ones. We posit that farmers will partly judge the desirability of any new scheme not only based on its absolute costs and benefits but also on how the features are framed relative to the features of current schemes. In practice, this means that the perceived attractiveness (and hence uptake) of new AESs may be adversely affected by unfavorable comparisons to previous schemes.
The Control (top) and Treatment (bottom) Conditions for a Question that Tests Whether a Change in the Framing of the Current Scheme Affects Preference between New Schemes X and Y
3. Method
Randomized Survey
An online survey was created with Qualtrics software, using the principles of tailored design (Dillman et al. 2014). The reason for using this method is twofold. First, online surveys allow for more sophisticated randomization and display logic. Second, although we could have included mail and online surveys, aside from the additional development cost, there is evidence that offering a choice of response methods can reduce response rates (Medway and Fulton 2012). Appendices B–D show all treatments for the three main questions surrounding mental accounting and loss aversion. Each respondent was shown one randomly selected treatment for each question, though the overall order in which the questions were presented was predetermined and the same across farmers. Randomization was performed at the question level, rather than the respondent level, and each treatment had an equal probability of being shown. As well as these main questions, the survey included questions that elicited attitudes toward four different aspects of the agricultural subsidy process: application difficulty, application time, application pleasantness, and the level of perceived control a subsidy scheme provides. Finally, questions were included to capture demographic variables such as age and differences in farm characteristics, such as farm size, type, and income.
Data
A sample frame of farm addresses was formed using publicly available data on recipients of CAP subsidy payments in the United Kingdom between 1999 and 2013 from farm-subsidy.org. After removing records with incomplete postcodes, 138,379 addresses remained. We dropped addresses that received aggregate payments totaling > €10 million and < €15,000, to remove large corporations and very small holdings, for whom our study was unlikely to be relevant. This resulted in an overall sample frame of 84,159 addresses. Given that our survey randomizes across three conditions, we required a sample size that was large enough to detect significant differences between means using a t-test. Using G*Power (Faul et al. 2007, 2009), we estimated that a minimum sample size of 64 per group was necessary to detect differences at the 5% level with a medium effect size of d = 0.5 and power = 0.8. In practice, we aimed for and collected a sample larger than this. Because response rates have been typically close to or below 10% in similar farmer surveys (Kuhfuss et al. 2016a; Pellegrin et al. 2018), and because our data source meant that some addresses may be incorrect or out of date, we randomly selected 12,000 addresses from the sample frame to ensure that we obtained a sufficient sample size. Addresses corresponding to councils, nonfarm organizations, and trust settlements were excluded. Invitation letters were mailed to the 12,000 sampled addresses, and included a shortened URL for the online survey and a colored slip with a handwritten thank-you message and smiley face to encourage participation. We avoided including name information in the address because this can increase nonresponse (Dykema et al. 2019) and because farm ownership may have changed since the CAP data was obtained. Follow-up reminder letters were sent three to four weeks after the initial invitation letter.
The data were collected in summer 2019. Out of the 12,000 letters initially sent, 624 were returned as undeliverable or we received correspondence from the recipient explaining that they were unable or unwilling to complete the survey. The main reasons given for not completing the survey were either that the farm was inactive (e.g., due to retirement or closure) or that the respondent had no access to a computer. Of the remaining letters, we received 860 total survey responses (7.6%). From these 860, 799 answered at least one question of the survey overall; 727 responded to experiment 1; 647 responded to experiment 2; 630 responded to experiment 3; and 583 surveys were fully complete, including all demographic questions at the end.
Survey designs like this suffer from a degree of response bias because of systematic differences between easy- and hard-to-reach respondents (Heffetz and Rabin 2013; Heffetz and Reeves 2019). This is not a primary concern for the present study because our main objective was to test the differences between randomly assigned treatments in a group of farmers. Even so, we briefly compared our data to national data to put our sample into perspective. Respondent characteristics appear to correspond broadly to U.K. data on farmers and farms. For example, the proportion of cropping farms in the sample (26%) was similar to that in the United Kingdom (29%) according to the 2016 Farm Structure Survey, though our sample contained a lower proportion of livestock and dairy farms (44% in sample versus 64% nationally) and a higher proportion of mixed farms (24% in sample versus 6% nationally).4 Sixty-one percent of U.K. farms had an income below €50,000 and 14% were €250,000 or higher. This corresponds reasonably well to our sample, where 49% are below £45,000 and 16% are above £190,000. Our sample does skew toward larger farms: 55% of our sample were 100 ha and over, but only 19% were this size in U.K. data from 2017.5 However, it contained the same proportion of women (15%) and had the same median age group (50–59) as in U.K. data.
4. Results
Summary Statistics and Opinions about the Application Process
Before turning to our main results for the experimental questions, we discuss some summary statistics on the characteristics of the farmers in our sample. As described already, respondent characteristics were broadly reflective of the U.K. farming population: 0.5% of responses were from Northern Ireland, 13.9% were from Scotland, 6.2% were from Wales, and the remaining 79.4% were from England. Two hundred thirty-five reported that they were currently enrolled in an AES from the 607 who responded to this question (38.7%). A set of summary farm and demographic statistics from the sample can be found in Appendix E, as well as the distribution of characteristics across the treatment groups for the three experiments. We asked four opinion-based questions designed to understand views toward the application process for agricultural subsidy schemes. Figure 3 summarizes the breakdown of responses to these questions. The majority of farmers seem to find applying for agricultural funding difficult, time consuming, and unpleasant. They also feel that they have too little control over their land and farming activities. Due to response bias, one cannot claim these results are representative for all farmers. However, the findings are unsurprising given previous research around the reasons behind poor adoption of AESs, and our sample does contain variation in terms of farm and farmer characteristics.
Attitudes towards Agricultural Schemes. (a) Perceptions of Application Difficulty; (b) Perceptions of Application Time; (c) Perceptions of Application Unpleasantness; (d) Feeling of Control within a Scheme
Main Results
Segregation versus Integration
The first of our three main results concerns how money is traded for reduced application time when losses are segregated, compared with when they are integrated (see Appendix B for the conditions). For easier comparison, we converted the minimum WTA in the treatment condition into an effective WTP by subtracting the reported WTA from £10,000. The means and confidence intervals are shown in Figure 4. In the control condition, the mean WTP from a £10,000 subsidy to reduce the application time from 10 hours to 1 hour was £835.61 (n = 217, 95% confidence interval [£650.58, £1,020.65]). The mean WTA in the treatment condition was £6,868.71, so the effective mean WTP condition was £3,131.29 (n = 220, 95% confidence interval [£2,679.59, £3,582.99]). A two-sample t-test comparing these means, assuming unequal variances, yields p < 0.00005. This result fits the prediction from mental accounting outlined in Section 2, that is, the value of an integrated loss is greater than the value of an identical segregated loss under a mixed-gain outcome scenario. Therefore, for the values under integration and segregation to be identical, the loss under integration (i.e., the effective WTP in the treatment condition) should be greater than the loss under segregation (i.e., the WTP in the control condition). Indeed, this is what we observe in the data. For robustness, we repeated the analysis including only those farmers who fully completed the survey, since there may be a concern that survey attrition is correlated with responses. This has virtually no effect on the results. The number of observations in the control and main treatment groups falls to 200 in both groups. The mean WTP in the control (segregated payment) group increases from £835.61 to £844.89 and in the treatment (integrated payment) group falls from £3,131.29 to £3,018.56.
Mean WTP for Reduced Application Time with 95% Confidence Intervals. Losses Are Segregated in the Control (n = 217), T2 (n = 210), and Integrated in T1 (n = 220)
The results suggest that first, farmers have a high WTP to reduce the application time associated with a new subsidy scheme, which means that the disutility associated with application burden is substantive. Second, our findings suggest that reducing application burden and providing farmers with a lower subsidy (e.g., £9,000, consisting of an integrated mixed gain of £10,000 − £1,000) is preferred over the provision of a higher subsidy (e.g., £10,000) that requires the same cost (£1,000) to be paid to someone who can reduce the burden by the same amount. In essence, the negative repercussions from asking farmers to sacrifice some of the subsidy they receive in exchange for application assistance weighs much more heavily than simply subtracting the costs from the initial subsidy to begin with.
To see whether reducing the time from one hour to zero hours under the control (segregated) condition had any effect on WTP, we added a second treatment as described in Section 3. As can be seen in Figure 4, there is no significant difference in the mean WTP between this second treatment and the control condition (p = 0.882). Our motivation for including this treatment condition was to see if completely eliminating application burden (i.e., moving from one hour to zero hours) would have a disproportionately large marginal utility associated with it. We did not find evidence for any such effect here.
Scheme Labeling
We look at whether the name attached to a subsidy is likely to affect expenditure allocation as a result of mental accounting. Figure 5 shows the mean percentage of subsidy that farmers allocated to each of three different expenditure categories across three different labeling conditions. The full text for each treatment can be found in Appendix C. Our first treatment condition was designed to elicit a mental account for agriculture, and we hypothesized that it would result in greater expenditure on farm operations. However, the results show this not to be the case. The mean proportion of subsidy allocated to operating costs was 63.09% (n = 238, 95% confidence interval = [59.71%, 66.47%]) in the SPS condition and 60.46% (n = 245, 95% confidence interval = [57.09%, 63.84%]) in the APS condition. There is no significant difference between these means (p = 0.279). Therefore, we cannot conclude that the APS treatment is likely to significantly increase spending on farm costs.
The Effect of Scheme Name on the Mean Percentage of Subsidy Allocated to Three Spending Categories. Treatments Are SPS, n = 238; APS; n = 245; EPS, n = 244
The second treatment condition, EPS, replaced the words “simple payment” with “environmental protection” in the scheme name. It also added some artwork of a green flower next to the scheme title (see Appendix C). This treatment was designed to encourage the formation of a mental account for the environment. Hence, the alternative hypothesis is that the mean percentage of subsidy allocated to environmental management activities should be greater in the EPS condition, relative to the SPS and APS conditions. The mean proportion of subsidy allocated to environmental management was 26.73% (n = 238, 95% confidence interval = [23.84%, 29.61%]) in the SPS condition, 26.96% (n = 245, 95% confidence interval = [23.90%, 30.02%]) in the APS condition, and 32.70% (n = 244, 95% confidence interval = [29.62%, 35.79%]) in the EPS condition. There is no difference between the means in the SPS and APS conditions (p = 0.9135). A one-sided t-test of the alternative hypothesis between the SPS and EPS conditions yields a p-value of .0028 (p = 0.0056 for a two-sided test), and a one-sided t-test of the alternative hypothesis between the APS and EPS conditions yields p = 0.0047. This suggests that an environmental label, even in the absence of any scheme conditions, may encourage farmers to allocate a greater portion of their payment to the environment. Repeating the analysis for only those with full survey completion as a robustness check, we observe a slightly larger effect of the EPS condition (reduced n = 205) on environmental management spending allocation relative to the SPS condition (reduced n = 191). Environmental management spending in the EPS condition attracted 6.73 more percentage points relative to the SPS condition using this restricted sample. Therefore, a significantly greater proportion of subsidy is allocated to environmental spending when the scheme is framed as being related to the environment. As with previous studies on labeling, this violates the fungibility of money assumption because there is no explicit restriction on the type of expenditure that is permitted in either treatment.
Reference Points
Our final experimental question tests whether different reference points can change preferences for a new subsidy scheme depending on the relative differences between attributes (see Appendix D). Loss aversion implies that a smaller proportion of people would prefer scheme X to scheme Y in the treatment condition than in the control condition. The null hypothesis reflects the prediction of classical utility theory, which suggests that there should be no difference in the proportions choosing X relative to Y when the reference point is changed. The result rejects the null hypothesis of no difference in choice proportions between control and treatment, that is, the reference point does affect the proportion of farmers choosing X relative to Y (Figure 6). Scheme X was chosen 44.4% of the time in the control condition (n = 313), and only 27.8% of the time in the treatment condition (n = 317). One-sided and two-sided t-tests of these differences yield p-values < 0.00005. By repeating this analysis with only fully completed surveys as a robustness check (new n = 297 in the control condition and n = 286 in the treatment condition), we see that the proportion choosing scheme X remains virtually unchanged at 44.1% in the control condition, and it falls slightly from 27.8% to 26.9% in the treatment condition. Therefore, the same result emerges: the reference scheme has a significant effect on the choice between the new schemes X and Y. This result suggests that existing schemes can influence how new schemes are evaluated against each other, even when the existing scheme is not in the choice set.
Proportion Choosing Scheme X (Environmental Option) in Each Treatment. The Reference Scheme Has Higher Environmental Benefits in the Control Condition than in the Treatment
5. Discussion
This article tested whether we could devise design improvements for agricultural policy based on the principles of mental accounting and loss aversion. We found that all three of our main predictions were supported by the data. First, when exchanging money for reduced application time, farmers are willing to sacrifice substantially more of their subsidy for a time reduction than they are willing to pay in a separate transaction. This is consistent with Thaler’s (1985) theory of mental accounting and consumer choice—the integration of a mixed gain is preferred to segregation. This suggests that one way of at least partly compensating farmers at least in terms of overall utility (as opposed to money) for reductions in future agricultural support is by integrating any application costs into the subsidy itself. Aside from mental accounting, another possible contributor to this result is that farmers may implicitly associate paying a fee with paying an adviser, even though the question did not explicitly specify to whom the fee would be paid. This may indicate that there is a disutility from transferring subsidy money to a private entity rather than to other farmers or the government. Regardless of the underlying mechanism, the overall result suggests that if financial incentives to farmers were to reduce in the United Kingdom as a result of leaving the European Union, then policy makers should seek to integrate the cost of application burden into the final subsidy offered, rather than expecting farmers to pay for assistance. In the treatment condition (where the costs of application time were integrated rather than segregated), farmers were willing to give up an average maximum of 31% of their subsidy in exchange for a reduction in application time from 10 hours to 1 hour. This suggests that there may be significant gains in farmer welfare if the burden of the application process was shifted away from farmers, a suggestion that is supported by largely negative attitudes toward the application process (Figure 3).
Second, the label attached to a subsidy appears to affect how farmers would allocate expenditure from additional subsidy income. This result adds to the existing evidence against the fungibility of money. The likely explanation for our result is that individuals are creating a new mental account for the additional income, meaning that it is treated differently than an unlabeled cash transfer. Therefore, a simple and cost-effective nudge to promote more sustainable or environmentally beneficial behaviors would be to rename any basic holding subsidy to include a label that promotes the kinds of expenditure it would be welfare-increasing to encourage. It should be noted that our results also suggest that the label would need to be salient and specific enough to induce the creation of a new mental account in one’s mind. The bottom line here is this: given that mental accounts are created based on scheme labels, governments should determine which aspect of farmer expenditure would generate the greatest societal benefit when deciding how to label any future agricultural subsidy scheme.
Third, leaving the European Union may lead to a change in the features/attributes of new agricultural subsidy schemes available to farmers. Our results emphasize that the relative desirability of any new scheme will depend not only on the characteristics of those new schemes but also on the characteristics of any existing scheme a farmer is currently participating in. Loss aversion may lead farmers to undervalue a new scheme if it contains some features considered to be directly inferior to features of the existing scheme, even if the new scheme offers additional benefits. This finding is especially important for incentive schemes that suffer from low uptake or renewal rates. Irrespective of relative treatment differences, the responses to our loss aversion question also highlight an absolute focus on the profitability benefits from any subsidy scheme. In both conditions, less than half of farmers chose scheme X over scheme Y (Figure 6), that is, the scheme with more environmental benefits. This suggests that the marginal utility of perceived profitability benefits may be greater on average than the marginal utility of perceived environmental benefits. Therefore, there may be a more general need to ensure farm profitability before policy makers can expect farmers to take up policies offering strong environmental benefits.
Written comments at the end of the survey highlight other agricultural policy issues not directly addressed in this survey, which could be useful for future studies. Many farmers expressed concern not just about application time but also inefficiencies in the bureaucratic process, citing poor experiences in dealing with administering bodies and late payments. As one respondent commented, “I would like to see the application of environmental schemes made much easier. The present protocols are Kafkaesque, byzantine, tortuous and bloody difficult.” Another pointed out that “If farmers are to rely on environmental payments the payments must arrive on time, appeals must be finalised promptly and the people administering and checking them must allow some leeway and understand that farms are natural places and are affected by weather conditions and that not all fields are the same size.” Others emphasized the need for different schemes for different regions and farm types. This is clearly a difficult balance to achieve, as greater individualization comes with greater administration costs. Some responses also drew attention to potential positive externalities that are not adequately captured by existing subsidies. Farmers argued that they create public goods for the environment during regular farming practice for which they are not explicitly compensated.
Limitations
A limitation of the questions posed in this article is their hypothetical nature. While care was taken to frame them in a context that would potentially be relevant to farmers, it is quite likely that they would not be used to making decisions from information presented in this way. Indeed, based on correspondence from respondents, at least some explicitly stated that they had difficulty in answering some of the questions because of their abstract nature. One would assume that any errors arising from this would not be biased in any particular direction, and therefore t-tests of mean differences should suffice. In addition, hypothetical questions may overestimate the proportion of choices that are socially desirable, since any cost associated with such behavior is likely to be less salient (El Harbi et al. 2015). In this study, this may have led to an overestimate of the mean proportion of a subsidy a farmer would actually spend on environmental management in the scheme labeling question. Given that one would expect this to be equally true across all treatments, it is unlikely to interfere with the objectives of this study.
6. Conclusions
We conducted an online randomized survey experiment on a sample of U.K. farmers to show how two insights from behavioral economics—mental accounting and loss aversion—can inform the design of new agricultural policy schemes to improve uptake (and perhaps effectiveness). The United Kingdom will have to develop a new set of agricultural subsidy schemes after it formally detaches itself from the EU agricultural policy framework. Our results show that relatively small changes in the way a policy is framed can potentially have a large effect on behavior. These changes have negligible costs compared with the potential benefits they could have. Improving policy design by taking behavioral aspects of choice into account may serve to increase adoption and adherence, encouraging behaviors that boost social welfare. Importantly, better-designed policies could improve the welfare of farmers themselves.
Acknowledgments
This work was supported by the Global Food Security’s Resilience of the U.K. Food System Programme with support from Biotechnology and Biological Sciences Research Council, Economic and Social Research Council, Natural Environment Research Council, and the Scottish Government. We thank Laura Harpham and Sam Durham from Championing the Farmed Environment for their insights. Ethical approval for the survey was granted by the University of Leeds Research Ethics Committee.
Footnotes
Appendix materials are freely available at http://le.uwpress.org and via the links in the electronic version of this article.
↵1 The single reference to agriculture we found in the Behavioural Insights Team reports was the following very brief mention in the 2016–2017 report: “We’ve expanded the range of our policy focus, covering issues as diverse as sustainable food consumption and farmers’ biosecurity behaviours” (Behavioural Insights Team 2017, 29).
↵2 The 2019 regulations state that payment can be claimed on land “at your disposal” on May 15, 2019, which includes owner-occupiers and tenants with a farm business tenancy under an appropriate agricultural tenancy act. It does not include landlords of tenant farmers, contractors, or farming of land under license (Rural Payments Agency 2019).
↵3 Note that although it may be interesting to see whether scheme X or scheme Y was preferred overall, which scheme individuals preferred is irrelevant to our hypothesis in this experiment. We are primarily concerned with whether the aggregate choice proportions are significantly different between the two treatments, which would be sufficient to show that reference schemes matter.
↵4 See https://www.gov.uk/government/statistics/farm-labour-profiles-from-the-england-and-uk-farm-structure-survey.
↵5 See https://www.gov.uk/government/statistical-data-sets/agriculture-in-the-united-kingdom.
