Selling Peace: How to Frame a Serbia/Kosovo Deal in a Referendum

Abstract Are referendum campaigns involving issues about sovereignty more likely to succeed if framed in a positive rather than a negative way? We ran a survey on a hypothetical referendum on a peace agreement between Serbia and Kosovo to answer this question, and we experimentally simulated both positive and negative frames. We found that the positive campaign frame, i.e. one that contains an invitation to support a lasting peace in the Balkans, economic prosperity, Serbia’s path to EU integration, and the protection of the Serbian population and cultural heritage in Kosovo, is more appealing than the negative one, which focuses on avoiding the risk of failure. Our finding contradicts previous works that relied on the prospect theory to argue that negatively framed messages can attract more attention because people try to avoid adverse outcomes. To explain our findings, we argue that positive referendum campaigns are more effective than negative ones when the reference point is low due to attributive framing.


Introduction
More than twenty years after the war, Serbia and Kosovo are still negotiating the final settlement of their relations.Kosovo declared independence in 2008, but Serbia refused to recognize it.Since 2012, the two sides have been normalizing relationships under the auspices of the EU (Bieber 2015), which is expecting the process to end with a legally binding agreement (Van Elsuwege 2017, 399).As the agreement would impinge upon the territorial integrity of Serbia, before coming into force, it would have to be ratified by a two-thirds majority in the Serbian Parliament and then accepted in a referendum in Serbia.However, the outcome of this referendum is somewhat uncertain as Kosovo is a highly emotional issue in Serbia.According to a recent poll, 68.5 percent of the respondents in Serbia still consider Kosovo to be the heart of Serbia (CSDRI 2019, 16).Accordingly, almost the equivalent (68.7 percent) percentage of respondents believes the recognition of Kosovo to be national treason (CSDRI 2019, 16).
Most Serbian opposition parties and the Serbian Orthodox Church are against such a deal and are likely to mobilize against it.Therefore, to secure the acceptance of a potential agreement, the Government of Serbia would have to carefully select the referendum question and its framing strategy during the referendum campaign.One of the critical dilemmas is whether the campaign should focus on the negative consequences of refusal to ratify the peace deal or positive results of its acceptance.
Contrary to the existing literature, our results indicate that voters in Serbia are more ready to accept the potential deal with Kosovo if it is framed in a positive rather than in a negative way.In other words, our research shows that the eventual Belgrade-Pristina peace deal would be more acceptable to voters in Serbia if the referendum campaign would focus on the positive aspects of its implementation such as peace, economic prosperity, EU accession than the negative aspects of its rejection such as conflict, economic hardship or international isolation.We argue that this is so because many citizens of Serbia already consider Kosovo to be a lost cause due to the redefinition (i.e.lowering) of a reference point over the past several years.
To explain this theoretically, we draw on the concept of attributive framing coined by Irvin Levin and Garry Gaeth (1988).Attributive framing is a kind of framing that involves an implicit evaluation of choices.It refers to peoples' tendency in evaluating positively framed objects ("silver medal on championship") more favorably than negatively framed ones ("lost the final game"), even though they are logically equivalent (Kreiner and Gamliel 2022).The reference point may affect such choices because if individuals decide about a thing that they do not have, they may support a trade-off for the gain which is positively formulated.In other words, as we assume in the article, if attributive framing is used, people will evaluate a choice better if it is framed as securing gains rather than avoiding losses.As for the majority of the citizens of Serbia, the issue of Kosovo is long-lasting but distant in terms of shaping everyday life, and it does it in a roundabout way.While on an abstract level, people in Serbia, as polling shows, do find Kosovo highly important, for many, the reference point in their choice between acceptance and rejection of a potential deal is either zero or next to zero.As a result, the positive attributes (aspects) of a potential deal (the positive frame) would have more effects than the negative ones. 1  The rest of the article unfolds as follows: We first describe the background against which discussions about the future Serbia/Kosovo peace deal are taking place.Then, we outline our methods and present the results of our experimental survey.Finally, we contextualize our results within the realm of political discourse, discuss the findings, and offer arguments for why we think the positive framing would work better in case of the potential referendum on the Serbia/Kosovo peace deal.
Background Analysis of the Serbia/Kosovo Peace Process Serbia and Kosovo have a long history of conflict, which surpasses the scope of this article (Judah 2002).For this article, just a brief contextual overview is necessary to set the background against which the ongoing peace negotiations are taking place.Since the 19th century, Kosovo has been construed as Serbia's holy land, the core territory of central importance for national history and identity (Ejdus 2020).Kosovo was annexed by the Kingdom of Serbia in 1912 and spent most of the 20th century as part of Yugoslavia.To quell separatist tendencies in its autonomous province of Kosovo, Serbia curbed its autonomy in 1989.In the early 1990s, Yugoslavia broke apart along its administrative borders, and former republics gained independence. 2 In 1998, a low-intensity conflict in Kosovo spiraled into a civil war triggering NATO intervention in 1999.In June, Yugoslav and Serbian forces withdrew from the province while UNSC Resolution 1244 allowed for the international military (KFOR) and civilian (UNMIK) presence in the province.
In the years ahead, the UN attempted to negotiate a final solution.Still, the two sides failed to reach a compromise as Belgrade was ready to grant Kosovo everything but independence which was the red line Kosovars did not want to negotiate.As a result, in 2008, Kosovo unilaterally declared independence which Serbia fiercely rejected as illegal.Soon after that, most western states and the majority of UN member states recognized the independence of Kosovo.However, permanent members of the UN Security Council (Russia and China), five EU members (Spain, Cyprus, Slovakia, Romania, and Greece), and many states in the global south have still not recognized it.Over the years, it has become apparent that without Serbia's approval, Kosovo will not be able to obtain full international recognition and a seat in the UN (Rossi 2014).Since 2011, the EU has initiated and facilitated a new round of talks which led to the conclusion of the First Agreement of Principles Governing the Normalization of Relations, also known as "The Brussels Agreement" in April 2013 (Beha 2018).The very name of the agreement suggested that this agreement was just the beginning of the normalization process, and the EU soon openly stated that it was expecting the process to conclude with a legally binding agreement (EU 2013).Although the EU does not officially insist on Serbia's recognition of Kosovo, the legally binding agreement is expected to serve at least a de facto recognition that would clear the ground for Pristina to get a seat in the UN.It would, therefore, have to be not only ratified in the Serbian Parliament but also approved by a simple majority in a referendum.

Previous Research on Referenda
We draw upon the works in political science, and our starting assumption is that framing plays a significant role in the formation of political attitudes (Klar, Robinson, and Druckman 2013).This applies to referenda, even if the electorate is well informed about the referendum issue (Dvořák 2013).To understand the effect of framing on political decisions, including referenda, political scientists have often drawn upon the prospect theory, which predicts how and under which conditions people's decision-making process deviates from norms of rationality (Kahneman and Tversky 1979).Specifically, this descriptive model of decision making predicts that people's decisions are systematically dependent not upon real values and probabilities but rather upon subjective estimations of both.It postulates that a decision-maker estimates the (given, objective) value of any offered option, comparing it to their own reference point, thus estimating whether he losses or wins by making that specific choice (Damnjanović and Janković 2014).
Furthermore, the fundamental notion of prospect theory is that losses matter more than gains, and people are inclined to make risky decisions to avoid losses (Kahneman andTversky 1979, Kahneman 2011).This theoretical assumption led many to argue that positive framing is counterproductive and that framers have more chances of succeeding when they put the risks of refusal front and center in their campaigns (Shuck and de Vreese 2009; Hancock and Weiss 2011;Hancock, Weiss, and Duerr 2010;Hancock 2011;Hager, Roy, Hancock, and Ensley 2019).Some authors, for example, show that the focus on positive aspects of the proposed EU constitution only mobilized sceptics to vote against it in the 2005 Dutch referendum (Shuck and de Vreese 2009).Similarly, Hancock argues that the strategy to focus on the risks of failure rather than on the benefits of success had led to the successful passage of the Good Friday Agreement in a referendum in Northern Ireland in 1998 (Hancock 2011).Following the same logic, Hancock and Weiss argue that the Oslo Agreement failed because its advocates focused on its expected benefits of a better future instead of emphasizing violent alternatives (Hancock and Weiss 2011).
We tested this line of argumentation in an experimental poll, which simulated the gist of a referendum campaign on the currently negotiated legally binding agreement on the future status of Kosovo.In the poll, we exposed respondents in Serbia to a specific frame and then asked them if they support a future arrangement with the Kosovo authorities.The poll was conducted online in May/June 2019 and consisted of 4,016 participants in total.In the poll, we tested whether the acceptance of the deal with Pristina is higher if it is framed in positive terms as a way to achieve gains or in negative terms as something needed to avoid losses.In addition to this, we tested which substantive arguments have the biggest saliency: peace/war; prosperity/poverty; yes/no EU membership; and protection/loss of the Serbian heritage in Kosovo.
Traditional view suggests that, when the issue has been known for some time, and the electorate has entrenched predispositions about it, voters are supposed to gather relevant information from political agents to make informed and predictable decisions about the referendum issue (Hobolt 2005).Our argument presupposes that framing matters greatly in formulating referendum questions and that most voters have no strong preferences (Zaller 1992).Even if the voters are quite familiar with the issue, it is possible to influence public opinion in specific directions (LeDuc 2002;2007).In other words, a preference reversal is possible even with entrenched beliefs and informed voters.
The most potent empirical examples to this effect are the two Danish referenda on the Maastricht treaty in 1992 and 1993 and the two Irish referenda on the Nice treaty in 2001 and 2002. 3Danes rejected the Treaty in June 1992 but then supported it in May 1993.A similar reversal took place in Ireland in June 2001 and October 2002.The explanation for both cases was not that at the second referendum, the Danish and Irish electorate learned something that they did not know at the first one.Instead, the proposal was reframed, which caused the preference reversal (Hobolt 2006;2009;Marsh 2007;Dvořák 2013).
To fully understand the current situation, one also needs to consider the changing relevance of Kosovo for the Serbian electorate.Traditionally, as discussed above, Kosovo has been regarded as a territory of high symbolic and identity importance for the Serbs, firmly entrenched in the Serbian history and culture (Bieber 2002;Ejdus 2020).Moreover, since the 1999 war, all Yugoslav and Serbian governments have continued to make a legal claim for their sovereignty over Kosovo based on the United Nations Security Council Resolution 1244.Moreover, for many years Serbia kept its effective control in Kosovo's north, mostly populated by Serbs (Clark 2014).However, in the past few years, we have witnessed the Serbian officials' attempts to gradually redefine this reference point.Since 2013, Serbia gave up its parallel institutions in the north of Kosovo.Moreover, Serbian President Aleksandar Vučić has invested efforts in lowering the reference point by repeating that "Serbia does not have anything in Kosovo [and that] even one meter [of the territory] would be a gain." 4 He has recently stated: "Kosovo to be fully recognized in the future." 5At times, he mitigates such radical statements by repeating that "we must understand that Kosovo is not ours, but it's not theirs either," 6 which indicates the willingness to accept the partition of the territory between the Governments in Belgrade and Pristina. 7 The campaign has already brought about a visible shift in public opinion.According to a recent survey, 47 percent of the respondents believe that Kosovo has been lost, although 74.5 percent would still never support the independence of Kosovo. 8According to another research, 33 percent of respondents consider Kosovo to be lost, while 52 percent do not think it is lost (CSDRI 2019, 17). 9  For our research, it is of relevance to note that all polls show that the younger and better-educated population, which is by and large the profile of our respondents, tend to think of Kosovo as the lost cause.For instance, while 42 percent of the 18-34 age group agrees that Kosovo is lost, only 22 percent of the over 55 age group agrees with this assertion (CSDRI 2019, 22).Also, the same poll showed that while 45 percent of the population with higher education concurs with the statement that Kosovo is lost, only 20 percent of the people with primary school agrees with it (CSDRI 2019, 23).Such a low reference point favors specific outcomes containing any gain and discourages gambles with unfavorable odds even if their expected value might be high.This suggests that the winning campaign strategy would be to prepare the ground by continuously insisting that Serbia has nothing to lose in Kosovo and then focus in the heat of the campaign on positive consequences of peace, stability, and prosperity of the country.In our experimental poll, to which we now turn, we tested these assumptions and studied the effect of framing on the potential referendum in Serbia on the peace deal with Kosovo.

Method
The above-presented findings of the empirical studies (see, e.g., Shuck and de Vreese 2009; Hancock 2011), along with the analysis of the political background of the treaty, posit several questions faced by campaigners of the possible referendum regarding the agreement.The first dilemma is whether the campaign should emphasize the positive consequences of the deal or its potential negative effects and dangers.Then, there is a second one: is it more efficient, and to what extent, when, in the campaign, to focus on pro-stance or the contra-stance.If we have the answers to the first two questions, if campaigners opt to employ framing as one of the mechanisms, they would have to map essential aspects which would serve as the means for advocacy.What precedes all these decisions is to know the current stance about the campaign issue in the general public.
When applied to the problem of our article, current stances in Serbia regarding the future status of Kosovo can be conceptualized and observed as the decision between refusal and the acceptance of the agreement.We ask whether it is more efficient to campaign by positively or negatively framing the future consequences after the supposed decision (question 1), to focus, or to frame, the acceptance or the refusal of the deal (question 2), and what is the most efficient "weapon" or "political token" in this particular matter.Finally, before that, it is important to pinpoint the public stance or the reference point regarding Kosovo.
One way of disentangling those dilemmas is to apply the empirical psychological paradigm of judgment and decision-making on a potential referendum of the hypothetical peace deal between Serbia and Kosovo.More precisely, in our research, we aim to shed light on the possible influence of the frames used during a referendum campaign.Frames, however, never exist in a vacuum, as they are always grounded in a specific context.Therefore, we merged, by using all theoretically possible combinations, the decisions about the deal and frames with political arguments or tokens.Our study consisted of twenty experimental situations, some just for experimental control and valid comparison.Each of them simulates questions that are expected to be politically salient in the future referendum, thus conveying the ecological validity of our study.This enabled us to establish the influence of the attribute framing mechanisms (emphasis framing) and the employment of the specific topics on the voters' decisions.Finally, it gave us an insight into how the formulation of the problem may shape that decision.

Sample
Our study used a convenience sample since we employed an experimental design since we aimed to measure the impact of our framing manipulation (Deming 1975;Hahn and Meeker 1993).The sample consisted of 4,016 participants who were on average 40 (SD = 12.7) years old, the oldest being 78 years old, and the youngest 18, 43 percent female; 65 percent of participants were employed, 21 percent unemployed, 14 percent retired.As for the participants' education levels, 0.5 percent had elementary school as the highest level of education, 18.6 percent high school education, 56.8 percent college education, while 21.7 percent had the graduate level (Figure 1).In terms of comparison of our sample with the population data, the average age in the population is 42.2 years old, 51.31 percent of the population is female, and the official employment rate is We employed an experimental rather than a questionnaire approach to avoid dependence on the demographic characteristic of the sample.To address a potential concern regarding the disproportional number of PhD-level of education among our participants, we repeated the analysis on the sample without those participants.As almost all of the obtained results were the same, including the effect sizes, we report only results from the whole sample.We conducted the study in May and June 2019.We reached participants via the online snowball recruitment technique by posting a web link for the questionnaire on social networks, Facebook, and Twitter and asking participants to share it further on.There was no monetary incentive for participation.The response rate was extremely high, above 95 percent, due to the simplicity of the procedure for the participants.
In the preparation of the study, we paid particular attention to the minimization of nonexperimental influence on the participants and maintaining neutrality regarding the topic.For that purpose, this questionnaire was programmed specially for this study and linked on the designated webpage, which did not contain any other materials except for the introduction, consent, instruction, and questionnaire.Once the participants completed their participation, their IPs were blocked to avoid multiple involvements.

Materials
The focus of this study was to create a hypothetical referendum situation to gain data about factors underlying binary choices, such as the acceptance of a possible peace deal at a referendum.To that end, each participant had a forced binary choice between supporting the acceptance and supporting the refusal of the peace agreement.The task had the form of a vignette in which the supposed referendum was announced.The second paragraph emphasized one of the possible outcomes (acceptance vs. rejection) and related consequences (positive vs. negative) related to important arguments or political tokens.As stated in the introduction, our goal was to simulate political dilemmas as closely as they are articulated in Serbia's public space.For that purpose, as previously stated, we have identified four principal arguments (or tokens) often used in public debates about a potential peace agreement between Serbia and Kosovo.Those were: peace, economic prosperity, EU membership, and the protection of the Serbian cultural heritage in Kosovo.In the following example, where a positive frame of economic prosperity was used, we will explain the basic structure of each vignette (Table 1).All versions of stimuli used in the research can be found in the Appendix.

Design
The study employed 5 (token)*2 (frame)*2 (acceptance/refusal) non-repeated design.The first factor, the token, had five levels, formed as previously stated arguments (peace, economic

The introductory part was the same for all participants
On the 14th of May, the referendum will be held for citizens of Serbia to decide whether to accept a legally binding agreement about the normalization of relations between Pristina and Belgrade.
Argument (ten variations, 5 tokens*2 frames) Acceptance of the Agreement enables long-term economic prosperity.

The question in two versions (A & B), support/refuse
A: Do you support the agreement?B: Do you refuse the agreement?prosperity, EU and population/cultural heritage).The additional fifth level was formed as the combination of all four.The second factor, the frame, had two levels: positive and negative consequences of the same option (e.g., peace or war).To avoid response bias caused by asking participants the same question and whether they accept the agreement, we also had to vary the question formulation.Therefore, we entered the acceptance/refusal as the third factor with two levels, assuming that participants were asked to accept the agreement in option 1 or to reject it in option 2 (Figure 2).By intersecting the factors of the frame with two formulations and five tokens, 20 tasks were formed [token (5)* frame (2)*response ( 2)].In the preparation of the study, special attention was paid to the minimization of type II errorfor detecting a statistically significant high attributive framing effect as reported by Galimel and Peer (2010, Cohens d = 0.8) by bivariate test at level p < .01 the sample of 78 subjects per experimental group is needed.Therefore, our sample includes more participants per group than the suggested minimum (Table 2).As the measure of the dependent variable, the percentage of acceptance of the agreement was used.

Procedure
Each participant was exposed to a single experimental manipulation.Tasks were randomly presented to the participants.Answering was preceded by written instructions and consent for participation.After that, on the next page, the single main question was presented to the participant, and their task was to choose one of two offered options by marking the preferred one.The main question was followed by questions about demographics (age, employment, gender, and education).Finally, on the last page, we provided our contact information.On average, the whole procedure lasted less than five minutes per participant.The number of participants was similar in all situations, and it was around 400 per situation.Only in the last situation in which we combined all four political tokens, the number of participants was slightly smaller, around 310.Most importantly, the number of participants was equalized for positively and negatively framed situations, which was achieved by randomly assigning questions to participants.Participants were debriefed once the experiment was completed, while the text explaining the main idea of the research was made publicly available.The study was ethically approved by the Institutional Review Board (anonymized for review) on November 23, 2019.

Results
To begin with, overall, 48.64 percent of our respondents accepted the agreement, while 51.36 percent refused it.The percentage of acceptance varied from 28.7 percent to 70.3 percent, depending on the frame and token.To test the frame (positive/negative) effects on the probability of referendum decision acceptance, we first used the Chi-square test to determine statistical significance and Phi coefficient as an effect size measure.Since the question formulation (acceptance/ refusal) did not show significant effects and was only used to control response bias, we will not show it in the analysis.Therefore, participants' answers for all situations were recoded in such a manner that they reflect agreement acceptance.Our research design was such that we tested framing for five different arguments (political tokens) which are expected to be used in a referendum debate: 1) peace/conflict in 2) economy development/decline, 3) protected/unprotected Serbian people and heritage in Kosovo, 4) EU membership/non-membership, and 5) a combination of all the previous four arguments.
Firstly, for each of them separately, we calculated significance for the framing effect (Table 2).In general, the percentage of agreement acceptance varied from 28.7 percent up to 70.3 percent of participants, depending on the task.This significant variation in the percentage of people accepting the agreement indicates that the referendum outcome could depend on the frame (positive/ negative), which we directly tested further on.
The results show that in all situations, the framing has a significant effect on the hypothetical referendum decision (Table 3).The effects are such that if agreement acceptance leads to positive outcomes (peace, economic development, protected Serbian people and heritage, EU membership), participants will mostly support the agreement on the supposed referendum (Figure 3).We can conclude that frame affects referendum decisionpeople will tend to vote for positive outcomes (positive framing) and to avoid/refuse negative ones (negative framing).If we look at the effect sizes for individual arguments (political tokens), they vary from 0.18 (small) up to 0.35 (medium to large).Of course, if we add all political tokens together, the effect increases to 0.38 (large).If we compare individual political tokens, it might seem that peace shows the highest effect, while the EU shows the lowest effects.These findings indicate that peace could be the argument that can be affected most by framing in the future referendum.EU membership is least affected, with the economic benefits and heritage showing effects between the previous two.We used binary logistic regression to directly oppose the effect of different political tokens (peace/conflict in Balkan, economy development/decline, protected/unprotected Serbian people and heritage, EU membership/non-membership) on people's vote.Since it is impossible to use all four political tokens simultaneously in regression, we had to exclude one because of the multicollinearity problem.The results remained the same when we repeated the analysis by omitting one by one and keeping the other three tokens.Accordingly, we will only show the results where the EU token was omitted since it did not appear significant in either of the regression analysis.So, in the final analysis, as predictors, we used three political tokenspeace, economy and population/ heritage, and a frame typepositive or negative.At the same time, the criterion was participants' decisionvoting for or against the agreement.
Binary logistic regression analysis showed a significant prediction of referendum decision with 63 percent of success.The only significant effect on the decision was shown by frame type.Since the exponential beta coefficient (odds ratio) is larger than 1 (2.89), it means that the frame increased approximately three times the chance for someone to accept the agreement, which includes positive outcomes, than when it includes avoiding negative ones (Table 4).
As for the political tokens, we can see that none of them shows significant effects, which means that they are similarly affected by framing in the context of agreement acceptance.Although Phi coefficients were somewhat higher for peace, and a bit smaller for the EU as a political token, directly confronting their effects did not reveal any differences among them.We can conclude that regardless of the token type, the positive frame would increase the chance for someone to accept the  agreement, which leads to positive outcomes.According to Phi coefficients, we can assume that combining all tokens and putting them in a positive frame might increase agreement acceptance compared to using just one of them.To sum up, while the argument that the Belgrade-Pristina deal would bring about peace seems to be slightly more convincing than the other three (economy, protection of Serbs and their heritage in Kosovo, EU membership), what really mattered is to frame the peace deal in terms of gaining benefits rather than avoiding losses.

Discussion and Explanation
The main finding of our research is that the positive frame is preferred to the negative one.This appears to run contrary to several recent works which draw on the prospect theory to claim that negative frames have a stronger impact on voters because losses matter more than gains (Shuck and de Vreese 2009; Hancock 2011).We argue that our findings are a product of a particular sort of framingnamely, attributive framing.It differs from two other types, risky choice frame and goal frame, in that it involves an implicit evaluation of choice.This section considers all three frames, thus explaining why attributive framing is most suitable for our findings.Tversky and Kahneman (1981) formulated the risky choice frame.People who were asked to choose between two options framed in positive and negative terms, where the positive frame involves gains and the negative frame involves losses, revealed they are unwilling to take risks when faced with a sure thing.However, they are more willing to do so when they face a loss.Tversky and Kahneman explained this by loss aversion, a cognitive bias under which losses are valued more than equivalent gains (Tversky and Kahneman 1979;1981).
We claim that risky choice, does not apply to the Kosovo referendum campaign, at least for the segment of the population in Serbia for whom Kosovo is lost.The essence of our argument is that the ruling elites in Serbia have already lowered the relevant reference point, thus assuming that Serbs have nothing (or very little) on Kosovo and can therefore only gain if the peace agreement is to be struck.Under such a reference point, it might be that many people experience that it is not possible to lose anything because all (or almost all) had already been lost.This practically makes risky choices inapplicable in this case.
There is a type of framing that resembles a risky choice in that the framer is interested in one unique goal.The only reason why the framers would use a positive or negative frame would be to establish which frame is more efficient to achieve this goal (Levin, Schneider, and Gaeth 1998).Goal framers are, therefore, indifferent between the frames.Instead, they are interested in their efficiency to bring about a unique outcome.A typical example of goal framing is the experiment about preventive breast examination, which showed that women who were faced with an argument to do breast examinations framed as a loss expressed more willingness to do the examination than the women who were faced with positive or neutral arguments (Meyerowitz and Chaiken 1987).
In contrast to risky choice and goal framing, attributive framing offers a possibility for an evaluation of choice.For example, in the experiment of branding meat as 75 percent lean or 25 percent fat (Levin and Gaeth 1988), selecting the former states that one dislikes fat provided that it is vilified within a community.Here, positive frames are found to generate more positive evaluations than negative ones because the majority of people prefer winning to losing, given the reference point has been set to be low or non-existent.
We argue that in our research on the hypothetical referendum on the Serbia/Kosovo peace deal, positive messages had more effect than negative ones because we framed the question as an instance of evaluative choice.The support for a positive referendum campaign, where people choose to accept the agreement to receive something, offers the voters a possibility to evaluate their choice, which is why positive frames appeal to voters more than the negative ones.We asked if the respondents would accept/reject the agreement with Pristina, given that this choice could result in two opposite worlds (e.g., peace/conflict, economic prosperity/poverty).If they choose lasting peace in the Balkans, they must also be against armed conflicts.Similarly, they must also be against economic poverty, blockade of EU accession, and the loss of the Serbian cultural heritage and population in Kosovo if they had previously chosen economic prosperity, EU accession, and the protection of the Serbian cultural heritage as well as Serbs in the province.
The importance of the reference point is further exemplified when we disaggregate the data through different arguments or tokens.Although differences between tokens were not significant, at least in our sample, respondents were slightly more inclined to accept the deal if it was designed to preserve peace than if it was designed to lead to membership in the EU.In this case, peace might be a more desirable carrot for our sample because it is something that is universally praised and within reach.By contrast, the EU membership is an uncertain and high-hanging fruit but also a contested foreign policy goal.According to an opinion poll from March 2019, 46.80 percent of Serbs are for, 25.38 percent against; while 27.82 percent remain undecided when it comes to EU membership (Pjevović and Subotić 2019, 2).Another research shows that 71.7 percent of respondents would reject recognition of Kosovo if it were a precondition for membership in the EU.In comparison, only 14.15 percent would accept such a bargain (CSDRI 2019, 17).But, as we mentioned before, differences in framing effect between tokens were not significant, which suggests that in a real referendum campaign, framing would not substantially depend on whether someone uses peace, economic prosperity, population/cultural heritage, or EU membership as an argument in the assumed referendum.

Conclusions
Since 2011, Belgrade and Pristina have been engaged in a normalization dialogue facilitated by the EU.This process is expected to result in a legally binding agreement which will probably have to be ratified in a referendum on both sides.While it is still uncertain if this deal will be struck and what its substance will be, in this article, we investigated what kind of framing would be beneficial for its success in a potential referendum in Serbia.Drawing on prospect theory, previous research on peace deals and referenda assumed that negative framing might be more successful in convincing the voters to accept difficult peace deals, as people are readier to take risks when faced with a prospect of loss than gains.
This article put these assumptions to the test in an experimental poll of over 4,000 respondents in Serbia.In contrast to our expectations, our respondents were more likely to support a possible peace deal between Belgrade and Pristina if it was framed positively, as something which will bring benefits, rather than in a negative way, whose failure might bring losses.This means that framing based on positive framing was more effective than framing based on loss aversion.We explained this by positing that a risky choice conceptualization of the referendum on the potential Belgrade-Pristina might lead to its refusal.For many Serbs, especially the better-educated ones, Kosovo is a lost cause, and even though most of them refuse the recognition of Kosovo, it seems like they do not consider that there is much to be lost in a deal.Given the low reference point, the referendum on the peace deal is better described as an attributive choice, where previous behavioral research established stronger effects of positive over negative frames.
Our research has several important theoretical and policy implications.To begin with, we have shown that prospect theory might not always be a good analytical framework to study the framing effects of referenda and peace deals.In other words, which framing effect will be the strongest is a context-dependent matter which varies from case to case.As we mentioned in Section 3, where we discussed the redefinition of the reference point, a significant number of the Serbian electorate believes that Kosovo has been essentially lost.This attitude has been reinforced by government officials' and the Serbian President's occasional public statements.We cannot claim that the research outcome would not be different from ours in other cases, where a large majority believed that the reference point was higher (i.e., the object of a referendum is owned and, therefore, can be lost).
Secondly, our research has important implications for future campaigners if and when the Belgrade-Pristina deal is struck.Our results indicate that a positive rather than negative campaign in favor of the deal could have stronger effects on the undecided voters, especially the young and better-educated ones.This holds since they are more prone to already having a relatively low reference point regarding Kosovo, making it easier for the electorate to accept the positive benefits of the yes-answer.Also, our findings can help the participants of the potential referendum by increasing their awareness of the framing effect and its implications on decisions.If participants are aware of the question framing during the referendum, they might pay attention to it, and hopefully, make decisions that are more by their own beliefs, not with the question framing itself.
Our conclusions point to several potential avenues for future research.First, the future research could replicate our experimental poll, adding to it a premeasured reference point of each respondent.This would provide more precise testing of the effect of the main theoretical insights developed in this article.Second, it would be useful to comparatively investigate a bigger sample of peace deals referenda to make a more generalizable claim on the effects of positive instead of negative claims.Third, our results suggest that the stance of the Serbian voters on Kosovo may be altered under certain framing conditions despite its high emotional salience.Further research should investigate if these mechanisms were successful in campaigns with less emotional and more practical stakes.Fourth, our respondents were younger and better educated than the general population in Serbia.Although our repeated control analysis shows that the effect exists irrespective of the level of education, it would be useful to replicate similar research with a more representative sample of respondents.Finally, in-depth focus groups with respondents could be conducted to make valid interpretations as to why the positive frames have stronger effects than the negative ones.

Figure 1 .
Figure 1.Education and employment status in our sample and the population of Serbia

Figure 2 .
Figure 2. Scheme of the design

Figure 3 .
Figure 3. Agreement acceptance depending on framing within various political token types

Table 1 .
Structure of the questionnaire

Table 2 .
Number of participants in each experimental situation

Table 3 .
Chi-square test for significance of the framing effects on referendum decision https://doi.org/10.1017/nps.2022.61Published online by Cambridge University Press

Table 4 .
Prediction of referendum decision based on frame and political token type