Bio-based and Applied Economics 9(3): 241-262, 2020 ISSN 2280-6180 (print) © Firenze University Press ISSN 2280-6172 (online) www.fupress.com/bae Full Research Article DOI: 10.13128/bae-7928 Multi-country stated preferences choice analysis for fresh tomatoes Maria De Salvo1,*, riccarDo Scarpa2,3,4, roberta capitello2, Diego begalli2 1 University of Catania, Department of Agriculture, Food and Environment, Italy 2 University of Verona, Department of Business Administration, Italy 3 Durham University, Business School, United Kingdom 4 University of Waikato, Waikato Management School, New Zealand Abstract. In this study we investigate consumers’ preferences for fresh tomato attrib- utes in four European countries by assessing and comparing Marginal Willingness- To-Pay (MWTP) estimates from panel Mixed Logit (MXL) models with utility speci- fications in the WTP-space. We performed an in-depth post-estimation inference to identify what attributes are the main determinants of fresh tomato purchases in each domestic market. We also assess the choice probabilities for tomatoes of various ori- gins and types to illustrate how these post-estimation inference can be used to inform strategies designed to increase the market shares of Italian fresh tomato exports in new markets and to consolidate positions in markets where Italian fresh tomatoes are already appreciated by local consumers. Keywords. Mixed Logit Model, Marginal Willingness-To-Pay, WTP space, prefer- ence space, fresh tomato. JEL Codes. D12, Q13, Q18. 1. Introduction Fresh tomato is one of the most commonly consumed vegetable in Europe. Over the last decade its consumption has remained stable at about 15 kg/year per capita, although stark changes have been observed concerning the range of quality consumers demand (European Union, 2018). Italy is one of the major tomato producers in Europe (European Commission, 2020) with exports to German, Austrian, British, French and Romanian markets, where Italian fresh tomatoes are traditionally very appreciated. However, consumers’ preferences gradu- ally change, and year after year, the diversity of tomato types sold has increased every- where to meet a rapidly evolving and diversifying demand. Health, convenience, taste and type of packaging are nowadays some of the most important product values for consum- ers. In the case of tomatoes, as for other foods, the market for ‘specialties’ is growing at *Corresponding author. E-mail: mdesalvo@unict.it Editor: Meri Raggi. 242 Maria De Salvo, Riccardo Scarpa, Roberta Capitello, Diego Begalli a significant rate (Santeramo et al., 2018). New tomatoes varieties with attractive shapes, colours and tastes, innovative recyclable packaging, health claims and/or environmental certifications have been emerging as valuable product features that producers and retailers use to grow their market shares (Yue and Tong, 2009; Tonsor and Shupp, 2009; Alamanos et al., 2013; Oltman et al., 2014). Nevertheless, the demand for tomatoes shows substantive differences across coun- tries in terms of favorite shapes, packaging, origins and many other factors. Determin- ing consumers’ preferences and Willingness to Pay (WTP) for fresh tomato attributes is important to stakeholders in this industry (e.g., agricultural producers, intermediaries and retailers). It helps them determine which types of fresh tomato to grow and trade, how to manage the marketing mix, what communication content to emphasize in advertis- ing campaigns, and how to apply fair prices along the supply chain. This information is particularly crucial for small-scale farmers who experience a strong competitive pressure from bigger companies of producers and importers. For them it is essential to correctly identify and characterize the market segments to supply, so as to define the assortment of tomatoes to produce the following season. Against this background, the objective of this study is threefold. Firstly, this study aims at estimating consumers’ willingness to pay for fresh tomato attributes across four key importing countries in Europe. Secondly, it aims at identifying the main determinants of tomato purchases across such markets. Thirdly, it aims at exploring how structural esti- mates of heterogeneous preferences can be used to inform marketing strategies which could guide the growth of Italian fresh tomato exports. Alongside these research objectives, this paper also aims at achieving methodologi- cal and disseminative purposes. It will present and discuss the estimation strategies that could be implemented in a discrete choice cross-sectional analysis to face heterogeneity in preferences and take into account correlations between attributes. Frequently, in discrete choice applications, post estimation analyses are limited to the assessment of the Marginal Willingness-To-Pay (MWTP). But several additional results can be derived by the estima- tion of a discrete choice model with preference heterogeneity. In order to make concrete the methodological dissemination purposes of this study, the ‘Rmarkdown’ and ‘markstat’ codes we used in our analyses are made available to the reader. The data collection took place in Germany, Russia, the UK and Norway. These coun- tries were selected for different reasons. Two countries, Germany and the UK, are tradi- tional export markets for Italian fresh tomatoes. In particular, Germany has been for sev- eral years the main European country for Italian fresh tomatoes exports. In 2015, Germa- ny imported 28,188 tons of Italian fresh tomatoes, equivalent to 31% of the total fresh Ital- ian tomato export. In 2015 the UK ranked third in terms of imported quantity from Italy, with 8,250 tons of fresh tomatoes.1 The other two countries, Russia and Norway, instead, are marginal markets for Italian fresh tomato producers. Here Italian tomatoes compete with imports from other countries, such as the Netherlands, Spain, Egypt and Moroc- co. Nevertheless, the four markets under investigation in this study have all, to larger or smaller extent, the potential for future growth of Italian exports if producers will imple- ment strategies aimed at meeting consumers preferences. 1 Source: Trade Map (http://www.trademap.org). 243Multi-country stated preferences choice analysis for fresh tomatoes To achieve the study objectives, the same choice experiment was administered to four representative samples of consumers, one for each country. In the data analysis, to account for heterogeneity in preferences, we used Mixed Logit models (henceforth MXL, see Train, 1998; 1999, 2009) with utility specified in WTP-space, as suggested by Train and Weeks (2005) and Scarpa et al. (2008). Despite the well-argued methodological advantages of this approach, when compared with the more conventional preference-space specification, applications in food choice experiments are still infrequent (Balcombe et al. 2010, Balogh et al. 2016, Caputo et al. 2016, and Caputo et al. 2018). Researchers have generally opted for the more traditional preference-space approach (Loureiro and Umberger, 2007; Ortega et al., 2011; Zanoli et al., 2013; Liu et al., 2019), even to assess the MWTP for tomato attributes (Onozaka and McFad- den, 2011; Caputo et al., 2013; Carroll et al., 2013; Maples et al., 2014; Skreli et al., 2017). In the post-estimation stage of our analysis, estimates of the Marginal Willingness-To- Pay (MWTP) for fresh tomato attributes were derived and compared across the four sur- veyed countries. From an empirical point of view, MWTPs provide producers with evidence to adjust their price strategy in line with market preferences. Further, we estimated full cor- relation matrices for random taste coefficients of tomato attributes. We used these to esti- mate market shares for combinations of tomato shapes and certifications. Signs and magni- tudes of significant correlations between random attributes provide crucial information to producers and exporters. They are needed to define product profiles that meet consumers’ demand and identify those combinations of tomato traits that consumers dislike. Moreover, to illustrate, probability choice functions were derived for selected product profiles. These functions are useful to predict consumer behavior, since differences in choice probabilities are dependent on tomato attributes. Finally, marginal changes in choice probabilities within samples and for the whole population were simulated. This type of analysis serves as a tool to predict changes in consumer behaviour specific to the different export markets. The rest of the article is organized as follows. Section 2 discusses materials and meth- od, while Section 3 illustrates the econometric analysis. Section 4 reports and discusses results. We conclude with some final remarks in Section 5. 2. Materials and method Data used for this analysis were collected through a choice experiment designed to gather statements on hypothetical purchases of fresh tomato by consumers living in Ger- many, Russia, the UK and Norway. Preliminary focus groups and pilot surveys supported the final design of the questionnaire. Tomato attributes and levels were identified from previous studies (Yue and Tong, 2009; Onozaka and McFadden, 2011; Caputo et al., 2013; Carroll et al., 2013; Maple et al., 2014; Oltman et al., 2014; Meyerding, 2016; Skreli et al., 2017) and via discussion with experts in these export markets. Ten attributes were select- ed to profile fresh tomato characteristics. These were: tomato shape (which acts as a label for the product alternatives), colour, skin thickness, pulp type, packaging format, country of origin, production method, workers’ health and safety certification2, eco-sustainability 2 In the questionnaire this attribute was explained as follows: “tomatoes can be produced according to systems that ensure high health and safety standards to workers. The final product can have a label that certifies that these stan- dards were implemented in production”. 244 Maria De Salvo, Riccardo Scarpa, Roberta Capitello, Diego Begalli certification of production methods (including organic)3 and price4. Table 1 reports the attributes and their levels. To generate the alternatives, we used a fraction of the full factorial design, that was D-error minimizing within the sets that are orthogonal in the difference (refer to NGENE handbook for details). By using NGENE 12.0, 144 choice tasks were generated, blocked in twelve blocks of twelve each. Respondents were randomly assigned in a balanced rotation to one of the twelve blocks. Each was asked to complete the twelve randomized choice tasks in their assigned block. Given the complexity of the experimental design, the quali- tative attribute named ‘colour’ was not directly included in the experimental design but was paired with tomato shape. Consequently, the combination between shape and colour was constant in each block, but combinations changed between blocks and assigned to different people. In this way, each respondent always visualized the same pictures for the five tomato alternatives in all choice scenarios under his/her scrutiny. The other attributes were presented in a textual form. Figure 1 illustrates a choice card. The final questionnaire contained three sections. The first was designed to identify the respondent’s socio-demographic profile; the second relates to food consumption hab- its, with specific reference to fresh tomatoes; the final section was dedicated to the choice experiment. The survey was carried out in April 2016. The target population consisted of adult consumers that consumed fresh tomatoes in the last six months and were aware about the product characteristics. Country samples were selected to be representative of national populations in terms of age and gender. Interviews were conducted online and adminis- tered by Toluna (www.it.toluna.com), a market research company that deals with market 3 In the questionnaire this attribute was explained as follows: “tomatoes can be produced according to systems that ensure ecological sustainability and biodiversity protection. The final product can have a label that certifies that these standards were implemented in production”. 4 In the questionnaire, prices were expressed in the national currency. Table 1. Attributes and levels. Attribute Attribute levels Shape beef, salad (Salad), vine (Vine), cherry (Cherry), date (Date) Colour red (Red), not-red (i.e., yellow, orange or variegated) Skin thin, thick (Thick) Pulp juicy, rich (Rich) Packaging loose tomatoes, net (Net), tray (Tray) Origin Italy, Netherlands (NLD), Spain (ESP), Morocco (MAR), Egypt (EGY), others (OTH) Production method conventional, low environmental impact (Env), organic (Org) Workers’ health and safety certified not present, present (Safety) Eco-sustainable certified not present, present (Eco) Price (euro/kg) 1.18, 1.58, 2.37, 2.76 Note: qualitative attributes were coded using dummy variables. The price attribute was coded using a continuous variable. In bold font the reference level. In brackets the variable name. 245Multi-country stated preferences choice analysis for fresh tomatoes analysis and has a world opt-in panel with over 9 million consumers. The company sup- plied the availability of high-quality Internet panels (i.e., ISO certification and application of international quality standards for market research) and guaranteed an incidence rate equal to 0.70 for each country. The online questionnaire was completed by a total number of 2,600 respondents: 700 in Germany, Russia and the UK and 500 in Norway. The total choice observations generated were 31,200 (12 choice cards for 2,600 respondents). The number of products evaluated by respondents amounted to 156,000 (5 tomato shapes/scenarios for 2,600 respondents). Table 2 reports the summary statistics at country level and for the whole sample. Figure 1. Example of a choice card. Workers’ health and safety certification No No Yes No No Packaging Loose tomatoes Net Loose tomatoes Loose tomatoes Net Eco-sustainable certified Yes No No Yes Yes Pulp Juicy Juicy Juicy Rich Juicy Production method Low env. Impact Organic Low env. impact Conventional Organic Skin Thick Thin Thin Thick Thin Origin Morocco Italy Morocco Netherlands Netherlands Price (€/kg) 2.76 2.37 1.58 1.18 1.58 Table 2. Summary statistics. Germany (n = 700) Norway (n = 500) Russia (n = 700) UK (n = 700) All (n = 2,600) Mean SD Mean SD Mean SD Mean SD Mean SD White European ethnicity* 0.89 0.31 0.81 0.40 0.95 0.22 0.81 0.39 0.87 0.34 BMI 25.42 5.68 25.50 5.05 24.72 6.86 26.00 6.53 25.40 6.16 Vegetarian/vegan * 0.09 0.29 0.07 0.26 0.10 0.31 0.10 0.31 0.10 0.29 Female* 0.60 0.49 0.53 0.50 0.60 0.49 0.60 0.49 0.59 0.49 Age (in year) 39.01 12.09 40.34 16.74 38.70 11.09 39.23 12.40 39.24 12.96 Education** 0.29 0.45 0.39 0.49 0.75 0.43 0.43 0.50 0.47 0.50 Family size (n.) 2.54 1.23 2.47 1.39 3.23 1.17 2.84 1.34 2.79 1.31 Minor or dependent (n.) 0.68 0.92 0.61 1.02 0.93 0.94 0.79 1.04 0.77 0.99 *1 if yes; ** 1 if university graduate or post graduate. 246 Maria De Salvo, Riccardo Scarpa, Roberta Capitello, Diego Begalli 3. Econometric model and inference The choice data were analyzed by means of econometric models based on random utility maximization with heterogenous preference parameters (McFadden, 2001). We assumed a linear and additive indirect utility function: Unjt=-αnpnjt+βn’xnjt+εnjt (1) where pnjt is the price attribute, xnjt represents the vector of non-price tomato attributes, and αn and βn are random parameters which represent nth respondent’s taste intensities for each attribute describing the tomato profile of each jth alternative in the tth choice occa- sion in the sequence. For the random component, we hypothesized that εnjt~ i.i.d. Gum- bel. Assumptions imply that, conditional on βn, the probability of observing a particular sequence of 12 choices for each nth respondent (yn=yn1,yn2,…,yn12 ) is the product of stand- ard logit formulas: (2) Unconditional probability was calculated as the integral of equation (2) weighted by the density function g(αn,βn|μ,Ω): Pn(yn)=∫L(yn1,yn2,…,yn12|αn,βn)g(αn,βn|μ,Ω)dαndβn (3) This integral was approximated through simulation, by: i) taking draws from the g(.) function; ii) calculating the Likelihood function for each draw; and iii) averaging the results. The maximum simulated likelihood estimator is the value of the unknown param- eters that maximizes the likelihood of the sample simulated in this manner. Equation (3) represents the so-called panel mixed logit, which allowed us to use a mixed logit model specification in the context of repeated choices by respondents assuming specific taste distributions (Revelt and Train, 1998). To obtain a posterior distribution of αn,βn for each respondent, the procedure described by Revelt and Train (2000) can be used. Following Train and Weeks (2005) we specified the utility function in the WTP space5. With a Gumbel distributed unobservable component of utility, the error variance varies among respondents: (4) where kn represents a scale parameter for the nth respondent. To allow for random scale parameter, Train and Weeks (2005) suggested to divide equation (1) by the scale param- eter: 5 Sonnier et al. (2007) called this model the “consumer’s surplus model”. It is also known in literature as “expenditure function space” model, “valuation function”, or “money-space” (Thiene and Scarpa, 2009). 247Multi-country stated preferences choice analysis for fresh tomatoes (5) As a consequence, in equation (5), unjt~i.i.d. Gumbel, but with constant variance equal to π2/6. Assuming that λn=an/kn and cn=βn/kn, equation (5) becomes: Unjt=-λnpnjt+cn’xnjt+unjt (6) where λn=an/kn, wn = c’n/λn, cn=βn/kn and kn represents the scale parameter for the nth respondent. Equation (6) is the so-called utility function in the preference space. Given that, by definition, the MWTP for an attribute is the ratio between the attribute’s coefficient and the coefficient of the price attribute, equation (6) can be re-written as follows: Unjt=-λnpnjt+(λnwn)’xnjt+unjt (7) where wn = c’n/λn. Equation (7) is the so-called utility function in WTP space (Train and Weeks, 2005).6 Through the direct choice of specific random WTP distributions, the WTP space approach prevents situations where the implied MWTP distributions from the random preference coefficients show excessively long tails. This is often the case in preference- space utility specifications (Scarpa et al., 2008). The literature reports controversial results on what approach produces a better fit to the empirical data. However, there is a general consensus on the ability of WTP space specifications to generate more reasonable and less disperse estimates of WTP distributions (Scarpa et al., 2008; Balcombe et al., 2009; Hen- sher an Greene, 2009; Rose and Masiero, 2009; Daly et al 2012; Owusu Coffie et al., 2016). Our estimator was implemented in STATA 15.0 and employed the packages mixlog- itwtp (Hole, 2007). We did not find significant evidence of heterogeneity in preliminary estimations for tomato colour, skin, pulp and country of origin. So, we assumed these to be fixed, meaning that we hypothesized homogeneous preferences for these features. Conversely, we obtained significant variance estimates in preliminary results for toma- to shapes, packaging types and certifications. Hence, the associated random parameters were consequently assumed to be random and specifically distributed multi-variate nor- mal with a full correlation matrix. The coefficient for the negative of price was assumed to have a log-normal distribution, to constrain the price coefficient to be always negative. Estimates were obtained with 1,000 Halton draws, which despite the high number of ran- dom parameters, can assure sufficiently low simulation variance of the maximum simu- lated likelihood estimator according to Zeng (2016) and Palma et al. (2018). While all of the above is informative, it is also quite standard. In this study, however, we extended the range of inference in a more novel direction. We used the estimates of the vector of means μ and their variance-covariance matrix Ω=(LL’) for each country to infer the probabilistic choice behavior in the underlying population of consumers. Note 6 Sonnier et al. (2007) called this model the “consumer’s surplus model”. It is also known in literature as “expenditure function space” model, “valuation function”, or “money-space” (Thiene and Scarpa, 2009). 248 Maria De Salvo, Riccardo Scarpa, Roberta Capitello, Diego Begalli that estimates of the Cholesky decomposition L of the full variance-covariance matrix Ω enabled us to derive the correlation matrix for the random α and β. With this we iden- tified patterns of covariation across taste parameters β that we then used in behavioral inference. For example, we used them in the derivation of probabilistic demand functions based on the simulation of distributions of preference values β in the population from which to infer market choice probabilities. We replicated this for selected tomato attrib- utes (tomato profiles) and compared them across countries. Another type of inference was conducted at the sample level. Here information on the observed choice sequence of each respondent was brought to bear by deriving individual specific means for marginal WTPs. These are graphically represented for the samples by kernel smoothing plots for the sample of each country. Hypothetical choice prob- abilities were also simulated at the population estimates by modifying the choice sets to evaluate shares for what-if scenarios. Scenarios simulated the introduction in the choice tasks of specific tomatoes profiles at a given price. We provide an illustration of the latter obtained with the post estimation commands in Stata. This required the modification of one or more attribute levels in the choice set and the re-computation of the in-sample pre- dicted probabilities of choice to obtain changes in market shares following the introduc- tion of new tomato profiles. 4. Results and discussion Table 3 reports the coefficients estimates for each country. The model estimated for the pooled samples across countries are reported in the last columns to the right. The uninformed sequence of 12 choices between 5 alternatives has a log-likelihood of ln[(1/5)12]=-19.31, while the averages in our estimated model range between a maximum of -16.33 and a minimum of -16.67, respectively 0.85 and 0.86 percent of the uninformed probability. This implies a good explanatory power of the joint model. Findings suggest that red color (baseline yellow/orange/variegated) and country of origin (baseline Italy) are key determinants of choice, while a thick tomato peel (baseline thin) and a rich pulp (baseline juicy) do not seem to be relevant. Preferences vary across the investigated markets, especially for the country of origin. Italian tomatoes are always preferred to those coming from other origins for Germans. These, for example, are willing to pay an average premium of 0.90 €/kg for Italian tomatoes in comparison to those com- ing from Morocco. Russians are generally indifferent to country of origin when tomatoes come from Egypt, Italy or Spain. However, they significantly dislike those produced in the Netherlands, Morocco or “other countries”. For the latter, WTP is comparatively lower by about 0.19-0.21 €/kg. The UK consumers emerged as “origin-blind” as the country of ori- gin never emerges as significant. Country-level models’ results further suggest that Norwegians appreciate juicy toma- toes (+ 0.15 €/kg in comparison to rich-pulp tomatoes), while Russians prefer to buy thin- skin tomatoes (+ 0.23 €/kg in comparison to thick-skin ones). Interestingly, preferences for tomato shapes vary across countries and, at the same time, are also significantly heterogeneous within each country. For the coefficients of tomato shapes, standard deviations are significantly different from zero, with the excep- tion of “cherry” and “date” shaped tomatoes in the UK and “date” shaped tomato in Nor- 249Multi-country stated preferences choice analysis for fresh tomatoes Ta b le 3 . M ix ed lo g it m o d el s’ co effi ci en ts e st im at es . V ar ia bl es G er m an y N or w ay R us si a U K A ll M ea n St an da rd D ev ia tio n M ea n St an da rd D ev ia tio n M ea n St an da rd D ev ia tio n M ea n St an da rd D ev ia tio n M ea n St an da rd D ev ia tio n C oe f. C oe f. C oe f. C oe f. C oe f. C oe f. C oe f. C oe f. C oe f. C oe f. R ed 3. 77 ** *( 12 .8 7) 2. 26 ** *( 15 .7 0) 1. 39 ** *( 23 .2 6) 3. 95 ** *( 13 .6 9) 2. 52 ** *( 31 .0 2) Th ic k 0. 05 (0 .5 6) 0. 01 (0 .1 2) -0 .2 3 ** * (5 .7 9) -0 .0 3 (0 .3 5) -0 .0 6 (1 .8 6) R ic h -0 .0 7 (0 .8 5) -0 .1 5 * (2 .5 7) 0. 05 (1 .3 9) -0 .0 9 (1 .1 7) -0 .0 3 (0 .9 1) N LD -0 .3 9 ** (2 .6 4) -0 .0 9 (0 .9 3) -0 .1 9 ** (2 .9 2) -0 .0 1 (0 .1 0) -0 .1 8 ** (3 .4 1) ES P -0 .6 5 ** * (4 .2 8) -0 .1 5 (1 .4 3) -0 .1 2 (1 .8 9) 0. 09 (0 .6 3) -0 .2 1 ** * (3 .9 4) M A R -0 .9 0 ** * (5 .5 0) -0 .1 9 (1 .8 9) -0 .2 1 ** (3 .1 5) 0. 01 (0 .0 6) -0 .2 9 ** * (5 .4 4) EG Y -0 .8 6 ** * (5 .1 5) -0 .1 6 (1 .5 1) -0 .0 9 (1 .3 8) -0 .0 5 (0 .3 9) -0 .2 5 ** * (4 .6 4) O T H -0 .6 2 ** * (4 .0 0) -0 .2 3 * (2 .3 1) -0 .1 9 ** (2 .8 7) -0 .2 7 (1 .9 0) -0 .2 9 ** * (5 .3 4) Sa la d -2 .7 0 ** *( 10 .2 9) 2. 05 ** * (8 .8 0) -1 .3 5 ** *( 10 .0 5) 1. 17 ** * (8 .8 1) -0 .2 0 ** (2 .7 8) 0. 75 ** *( 10 .1 2) -2 .3 6 ** * (9 .2 0) 2. 48 ** * (9 .4 9) -1 .4 5 ** *( 18 .6 4) 1. 43 ** *( 18 .2 0) V in e 1. 11 ** * (7 .5 8) 0. 54 ** (3 .4 7) 0. 68 ** * (6 .4 1) 0. 50 ** * (5 .3 0) 0. 57 ** * (9 .2 1) 0. 48 ** * (9 .4 5) 1. 57 ** * (8 .5 6) 1. 16 ** * (7 .5 8) 0. 73 ** *( 14 .7 6) 0. 18 * (2 .1 2) C he rr y -2 .0 0 ** * (7 .8 9) 1. 81 ** * (7 .9 6) -0 .8 2 ** * (6 .4 6) 0. 80 ** * (6 .4 8) -0 .9 5 ** *( 11 .0 0) 0. 41 ** * (3 .9 6) -0 .6 8 ** * (4 .4 7) 0. 21 (1 .1 7) -1 .1 2 ** *( 14 .5 6) 0. 52 ** (2 .6 3) D at e -3 .6 2 ** * (9 .9 5) 0. 83 ** (2 .9 2) -1 .1 3 ** * (7 .5 7) 0. 10 (0 .7 7) -1 .2 1 ** *( 10 .5 3) 0. 71 ** * (5 .1 5) -1 .8 0 ** * (7 .8 1) 0. 02 (0 .1 4) -1 .7 5 ** *( 18 .4 1) 0. 07 (0 .5 2) N et -0 .2 5 * (1 .9 7) 0. 26 (0 .7 6) -0 .1 0 (1 .0 4) 0. 38 * (2 .3 7) -0 .1 4 * (2 .0 8) 0. 05 (0 .5 3) -0 .1 0 (0 .7 9) 0. 53 (2 .6 4) -0 .1 6 ** (3 .3 8) 0. 02 (0 .2 1) Tr ay -0 .7 2 ** * (5 .2 3) 0. 34 (1 .8 0) -0 .4 5 ** * (4 .6 6) 0. 32 * (2 .3 4) -0 .0 5 (0 .8 0) 0. 06 (0 .8 1) -0 .6 4 ** * (4 .6 3) 0. 05 (0 .3 1) -0 .3 9 ** * (7 .8 3) 0. 03 (0 .4 1) En v 0. 18 (1 .2 4) 0. 25 (0 .8 0) -0 .0 5 (0 .5 0) 0. 02 (0 .1 4) -0 .2 2 ** (2 .9 5) 0. 38 ** * (4 .4 3) 0. 23 (1 .7 5) 0. 15 (0 .6 9) 0. 01 (0 .2 2) 0. 08 (0 .8 1) O rg 1. 22 ** * (6 .2 3) 0. 99 ** * (4 .2 3) 0. 26 * (2 .3 2) 0. 80 ** * (6 .0 3) -0 .0 7 (0 .9 3) 0. 17 (1 .9 0) 0. 39 ** (2 .6 5) 0. 24 (1 .0 2) 0. 37 ** * (6 .5 8) 0. 56 ** * (3 .7 3) Sa fe ty 0. 56 ** * (4 .8 1) 0. 04 (0 .1 7) 0. 50 ** * (5 .3 6) 0. 23 (1 .6 6) 0. 23 ** * (3 .8 5) 0. 33 ** (3 .2 8) 0. 79 ** * (6 .2 6) 0. 19 (0 .9 7) 0. 43 ** *( 10 .1 2) 0. 13 (0 .8 1) Ec o 0. 60 ** * (4 .6 9) 0. 06 (0 .3 0) 0. 16 (1 .8 2) 0. 03 (0 .2 5) 0. 37 ** * (5 .5 6) 0. 36 ** * (3 .8 5) 0. 47 ** * (3 .9 8) 0. 31 (1 .4 9) 0. 44 ** * (9 .3 6) 0. 20 (1 .9 4) Ln -n eg .p -1 .2 1 ** *( 15 .6 9) 0. 13 * (1 .9 9) -0 .6 8 ** *( 11 .2 1) 0. 10 (0 .8 3) -0 .4 8 ** *( 10 .0 6) 0. 04 (0 .4 1) -1 .1 9 ** *( 16 .0 6) 0. 01 (0 .1 9) -0 .8 5 ** *( 27 .5 2) 0. 17 ** (2 .9 4) C ho ic es 42 ,0 00 30 ,0 00 42 ,0 00 42 ,0 00 15 6, 00 0 N 70 0 50 0 70 0 70 0 2, 60 0 ln -L /N -1 6. 33 5 -1 6. 42 9 -1 6. 48 8 -1 6. 48 8 -1 6. 67 4 A IC 23 ,0 39 .5 1 16 ,5 98 .8 7 23 ,2 52 .9 5 23 ,2 53 .0 1 86 ,8 75 .5 9 BI C 23 ,7 74 .3 7 17 ,3 05 .1 4 23 ,9 87 .8 1 23 ,9 87 .8 7 87 ,7 21 .9 9 *p < 0. 05 ; * *p < 0. 01 ; * ** p < 0. 00 1. z s ta ti st ic in p ar en th es is . V ar ia b le s co d in g is r ep o rt ed in T ab le 1 . 250 Maria De Salvo, Riccardo Scarpa, Roberta Capitello, Diego Begalli way. “Vine” tomatoes are always preferred to “beef ” (the baseline) across all countries, while the latter are always preferred to “salad”, “cherry” and “date” tomatoes. In general, consumers prefer to buy loose (the baseline) rather than packaged toma- toes. However, preferences for packaging types (in “nets” or “tray”) are not always signifi- cant and emerge as heterogenous at the country level. Tomatoes with certified credence attributes are preferred to those without certifi- cation. In particular, Germans are willing to pay, on average, 1.22 €/kg for organic-cer- tified tomatoes, even if the distribution is strongly dispersed in comparison to those in other countries. German consumers are also sensitive to certifications ensuring workers’ health and safety (+ 0.56 €/kg) and eco-sustainability (+ 0.60€/kg). However, the choices observed in the UK sample imply a higher willingness to pay for workers’ health and safe- ty certification (+ 0.79 €/kg). Preferences for certifications show a significant heterogeneity for organic products in Germany and Norway. Russians demonstrated significant heterogeneous preferences for low-environmental impact, workers’ health and safety, and eco-sustainable certified prod- ucts. Instead, organic certification is not appreciated by Russians, who in turn are the only consumers with a significant positive appreciation for certification for low-environmental impact production methods. Figure 2 displays the kernel smoothing of individual posterior means of MWTP sam- ple distributions for each country for tomato shape, whose coefficients showed a signifi- cant heterogeneity in the majority of the investigated countries. Sample distributions are displayed only for those tomato shapes which have both significant mean and standard deviation estimates. Some distributions differ significant in terms of range, number of modes and relative positions in the WTP space. As pointed out earlier, German and UK consumers do not appreciate salad tomatoes in comparison to beef ones. Their MWTP distributions are prevalently located in the negative range and are multimodal in both cas- es. This implies that, everything else equal, for most consumers, salad tomatoes need to be sold at a lower target price compared to beef tomatoes to induce a purchase; how much lower is different in the two countries. In contrast, MWTP for salad tomatoes in the other two markets are located to the right, especially for Russia, that present both positive and negative modal values. Nega- tive means of MWTP are shown also for cherry and date tomatoes in comparison to beef tomatoes. However, preferences for cherry tomatoes seem to be more similar among Nor- wegian and Russian consumers than Germans; while MWTP distribution for date toma- toes are less dispersed for Russians and more variable for Germans. Vine tomatoes, in contrast, are preferred to beef ones in all markets. Modal value estimates are always posi- tive, although the ranges of variation are extremely different between countries, with the widest one in the UK, where there is also the higher modal value. Table 4 reports the estimates of correlation coefficients (lower triangle), variances (diagonal) and covariances (upper triangle) between random MWTPs in each country. For some pairs of random attributes, correlations are significant across all models and have concordant signs, even if they have different magnitude. Correlations between salad and date tomatoes, for instance, are always positive and significant in all markets, mean- ing that these kinds of tomato could be jointly sold in these countries through focussed advertising strategies exploiting the “drag” effect of a tomato type on the other. Converse- 251Multi-country stated preferences choice analysis for fresh tomatoes ly, a negative correlation is estimated between salad and vine-shaped tomatoes. For Ger- many and Norway, in particular, the correlation coefficients are high and significant, -0.89 and -0.84 respectively. This finding is focal to support product marketing by the sellers: salad and vine tomatoes are antagonist in these markets and meet the preferences of dif- ferent consumers and consequently separate market targets. This could suggest locating these products on different shelves or even separate shops when the locations of these are correlated with one type of buyers. Country-level preferences for type of packaging vary and they are correlated with the tomato’s shape. In general, all consumers prefer to buy loose-packaged tomatos. However, Germans prefer to buy salad tomatoes that are tray- packaged (correlation coefficients: 0.77) and dislike trays for vine ones (-0.85); Norwe- gians like cherry tomatoes when packaged in a net (0.79). For the UK and Russia, some Figure 2. Kernel density plots for conditional WTPs for tomato shapes. 252 Maria De Salvo, Riccardo Scarpa, Roberta Capitello, Diego Begalli Table 4. Estimates of correlation and covariance matrixes in each country. Germany Salad Vine Cherry Date Net Tray Env Org Safety Eco Np Salad 4.20 -2.19 -0.38 4.58 0.47 1.25 0.14 0.46 -0.58 -0.14 0.06 Vine -0.89 1.44 1.05 -1.23 -0.34 -0.81 -0.15 0.30 0.46 0.18 -0.15 Cherry -0.08 0.37 5.77 4.82 0.10 -0.14 0.24 -1.09 0.62 -0.27 0.17 Date 0.66 -0.30 0.60 11.30 0.74 0.94 -0.30 -0.05 -0.08 -0.58 -0.15 Net 0.38 -0.46 0.07 0.36 0.38 0.19 -0.29 -1.17 -0.14 -0.36 0.10 Tray 0.77 -0.85 -0.07 0.35 0.40 0.63 0.29 -0.56 -0.12 -0.08 0.20 Env 0.07 -0.13 0.11 -0.10 -0.50 0.39 0.86 0.58 0.08 0.23 0.17 Org 0.09 0.09 -0.17 -0.01 -0.72 -0.27 0.24 6.98 -0.35 0.58 -0.37 Safety -0.37 0.50 0.34 -0.03 -0.29 -0.19 0.11 -0.10 0.59 0.61 -0.05 Eco -0.06 0.13 -0.09 -0.15 -0.51 -0.09 0.21 0.26 0.68 1.37 -0.21 -Price 0.05 -0.22 0.12 -0.07 0.28 0.42 0.32 -0.08 -0.12 -0.31 0.34 Norway Salad Vine Cherry Date Net Tray Env Org Safety Eco Np Salad 1.37 -0.93 -0.21 1.21 -0.14 0.14 -0.08 0.24 -0.13 0.07 0.07 Vine -0.84 0.88 0.21 -0.40 0.16 -0.05 -0.08 -0.33 0.04 -0.16 -0.23 Cherry -0.21 0.27 0.69 0.43 0.41 0.03 0.00 -0.06 -0.09 -0.06 0.04 Date 0.71 -0.29 0.35 2.14 0.26 0.23 -0.26 -0.01 -0.34 -0.16 -0.16 Net -0.19 0.26 0.79 0.29 0.39 0.08 -0.01 0.00 0.19 -0.01 -0.06 Tray 0.30 -0.13 0.08 0.39 0.33 0.16 0.01 0.09 -0.07 0.01 0.01 Env -0.20 -0.24 0.02 -0.48 -0.05 0.06 0.14 0.16 -0.11 0.01 0.11 Org 0.20 -0.35 -0.07 -0.01 0.01 0.21 0.43 1.02 0.03 0.36 0.10 Safety -0.12 0.05 -0.11 -0.24 0.31 -0.17 -0.31 0.03 0.93 0.27 -0.16 Eco 0.08 -0.24 -0.10 -0.15 -0.02 0.03 0.05 0.25 0.41 0.49 0.01 -Price 0.12 -0.46 0.09 -0.20 -0.17 0.07 0.54 0.05 -0.31 0.02 0.29 Russia Salad Vine Cherry Date Net Tray Env Org Safety Eco Np Salad 0.57 -0.08 0.12 0.71 -0.09 0.20 0.09 -0.06 -0.17 -0.12 -0.24 Vine -0.23 0.24 0.28 0.25 0.02 -0.02 -0.02 -0.06 -0.01 -0.01 -0.17 Cherry 0.21 0.75 0.60 0.51 -0.10 -0.08 -0.07 -0.26 -0.06 0.09 -0.44 Date 0.67 0.35 0.47 1.98 -0.22 0.21 0.14 -0.10 -0.26 -0.14 -0.39 Net -0.33 0.14 -0.36 -0.42 0.14 0.10 0.05 0.10 0.00 -0.14 0.06 Tray 0.53 -0.10 -0.22 0.29 0.55 0.25 0.14 0.08 -0.14 -0.28 0.05 Env 0.23 -0.09 -0.17 0.19 0.23 0.54 0.28 0.16 -0.02 -0.16 0.14 Org -0.13 -0.19 -0.54 -0.11 0.43 0.24 0.48 0.40 0.09 -0.05 0.12 Safety -0.36 -0.03 -0.13 -0.31 0.00 -0.45 -0.07 0.09 0.38 0.26 0.03 Eco -0.21 -0.03 0.15 -0.13 -0.49 -0.73 -0.40 -0.07 0.55 0.58 -0.14 -Price -0.39 -0.42 -0.71 -0.34 0.21 0.12 0.32 0.16 0.06 -0.23 0.64 253Multi-country stated preferences choice analysis for fresh tomatoes estimates of correlation coefficients between tomato shapes and package types are signifi- cant, but their values are lower than 0.60, showing a low-to-moderate correlation. Another interesting result concerns the relationship between certifications of workers’ health and safety protection and eco-sustainability. Their correlation is always significant and positive, suggesting that consumers who are willing to pay for an eco-sustainable tomato are also willing to pay for a health and safety-certified tomato. However, the correlation coeffi- cients are moderate ranging between 0.41 (for Norway) and 0.68 (for Germany). Table 5 reports estimates of the market shares for combinations of tomato shape and certifications. In all investigated markets, vine tomato shows the higher shares. In all countries, this tomato type could increase its market share when certified as produced with methods that promote workers’ health and safety, or eco-sustainability. Figure 3 displays the estimated choice probability functions for selected product pro- files along the price/kg dimension, when the baseline is a beef tomato without any addi- tional attribute. For all graphs we adopted a price ranging from the lower level assumed in the choice experiment (1.81 €/kg) to three times the higher level (3 x 2.76 €/kg). The top left plots of figure 3 show that as price increases the predicted purchase prob- ability by Germans of red rich-pulp Italian tomatoes drops rapidly, so that at a price of 6€/ kg it is basically zero except for vine tomato sold loose and certified. The function for Norway investigates the simulated effects of eco-sustainable and organic certification of red Italian tomatoes, while the function for Russia is about the effect of tomato shape. Finally, the function for the UK investigates the role of certifica- tion for the same tomato profile and demonstrates that in this country consumers have the same reactions to price changes regardless the type of certification. Figure 4 focuses on co-variation of preferences for tomato shapes, specifically salad and date, both of which tends to be disliked compared to the beef tomato baseline. It reports the iso-quantile plots for all the countries of bivariate kernel densities of MWTP UK Salad Vine Cherry Date Net Tray Env Org Safety Eco Np Salad 6.14 -2.66 -0.69 5.70 0.43 0.92 -0.30 -0.02 -0.58 -0.50 0.12 Vine -0.68 2.50 -0.16 -2.63 -0.25 -0.63 0.47 0.51 0.29 0.04 -0.62 Cherry -0.53 -0.19 0.27 -0.50 -0.04 -0.09 0.00 -0.15 0.11 0.14 0.14 Date 0.98 -0.71 -0.41 5.48 0.38 0.73 -0.16 -0.03 -0.42 -0.39 0.09 Net 0.25 -0.23 -0.11 0.24 0.47 0.35 0.10 -0.27 -0.29 -0.41 0.14 Tray 0.49 -0.53 -0.23 0.41 0.68 0.58 -0.04 -0.32 -0.29 -0.40 0.28 Env -0.18 0.44 0.01 -0.10 0.22 -0.07 0.46 0.14 0.06 -0.22 -0.15 Org -0.01 0.28 -0.25 -0.01 -0.35 -0.38 0.19 1.28 -0.07 0.26 -0.20 Safety -0.24 0.19 0.22 -0.18 -0.44 -0.39 0.09 -0.06 0.94 0.53 -0.15 Eco -0.22 0.03 0.29 -0.18 -0.65 -0.57 -0.35 0.21 0.59 0.86 -0.15 -Price 0.08 -0.66 0.44 0.07 0.33 0.61 -0.37 -0.11 -0.26 -0.26 0.36 Note: estimates of variances are reported in the diagonal. Covariance and correlation estimates are reported above and below the diagonal, respectively. Correlations are in italic. In bold the estimates which are significant with p<0.05. 254 Maria De Salvo, Riccardo Scarpa, Roberta Capitello, Diego Begalli estimates in a range of change between -5 to +5 Euro/kg compared to the baseline prod- uct profile. The price change combinations along each iso-quantile curve represent the proportion of the population with the same probability of selecting tomatoes with one of the two shapes rather than the baseline beef tomato. The isoquantiles highlight a positive correlation between salad and date-shaped toma- toes in the four countries, but the price set combinations with which they relate to the baseline are quite different. For German consumers, with a correlation estimate of 0.66, the curves cover a much larger set of MWTP values than in the Russian and Norwegian sam- ples. For the UK consumers, the MWTP ranges are similar to those shown in Germany. However, because of the much stronger correlation of 0.98 between the shape attributes, the room for a differentiated pricing policy is much reduced. Norwegian and Russian con- sumers show quite similarly sets in terms of preferences and willingness to pay. Finally, the estimated models can be used to simulate marginal changes in probability of choice within the samples rather than in the population. For example, what would the distribution of choice probability be, within the German sample, if all choice tasks includ- ing the baseline Italian tomato were offered with certification for workers’ health and safe- Table 5. Market shares for combinations of tomato shapes and certifications. Shape Certifications Germany Norway Russia UK Salad org-safety 4% 6% 11% 7% org-safety-eco 4% 4% 8% 5% org-eco 6% 6% 10% 7% safety-eco 5% 5% 17% 8% env-safety 4% 2% 10% 7% env-safety-eco 3% 1% 7% 4% Vine org-safety 45% 29% 28% 45% org-safety-eco 37% 22% 21% 36% org-eco 41% 28% 25% 40% safety-eco 54% 32% 46% 54% env-safety 39% 17% 20% 48% env-safety-eco 32% 11% 14% 34% Cherry org-safety 11% 5% 0% 4% org-safety-eco 8% 3% 0% 3% org-eco 8% 6% 1% 4% safety-eco 12% 5% 6% 8% env-safety 12% 3% 1% 6% env-safety-eco 8% 2% 1% 5% Date org-safety 7% 8% 5% 10% org-safety-eco 6% 6% 3% 7% org-eco 6% 8% 5% 9% safety-eco 8% 7% 8% 11% env-safety 6% 2% 5% 10% env-safety-eco 4% 2% 3% 6% 255Multi-country stated preferences choice analysis for fresh tomatoes ty at a price increased by ten percent? This comes down to computing the choice prob- abilities for all five alternatives in each choice task, i.e. the probability vectors with the Figure 3. Country-level demand functions for some types of tomato. 256 Maria De Salvo, Riccardo Scarpa, Roberta Capitello, Diego Begalli price increase for certitication (p1) and without such change (p0) for the baseline tomato. Then the difference between the two sets of predicted selection probabilities (p1 - p0) for the alternatives with the profile of interest is computed and the distribution of these val- ues examined. In our case we have 13,071 choice sets containing the baseline profile in the German sample. An increase of ten percent would always result in a decreased selec- tion probability, as shown in Figure 5. This suggests that either the price change should be lowered, or some additional positive features should be added, for example organic certi- fication, that the German consumers seem to strongly appreciate. One can also envisage iterating this exercise at gradually lower price increases until a sufficient fraction of the within sample predicted choices show a positive value. Figure 4. Iso-Quantile Plots of Bivariate Kernel Density distributions for MWTP estimates for salad and date-shaped tomatoes in the four countries. 257Multi-country stated preferences choice analysis for fresh tomatoes 5. Conclusions We conduct identical surveys across four countries to estimate the marginal WTPs for a set of attributes of fresh tomatoes. Estimates were obtained in WTP-space, which several authors encourage practitioners to adopt to obtain more reliable, interpretable and plausible MWTP distributions. Specific differences in preferences across countries have been highlighted in terms of sign and magnitude of the coefficient estimates, condi- tional MWTPs, correlation coefficients and market shares. Further, simulations of choice purchase behaviour were inferred within-sample and at the population level. These were discussed with regards to their effects of price changes on tomato profiles in the four mar- kets, to explore marketing implications of population distributions of marginal MWTPs and to exemplify the range of uses analysts can make of these model post-estimations. The method can produce evidence that could be used to support the design of strate- gies aimed at consolidating the position of Italian tomatoes on traditional European mar- kets, such as Germany and the UK; and at the same time, it could help Italian producers to identify what types of tomato produce to improve their share in Norwegian and Rus- sian markets. The tomato profile, which shows the highest probability to be purchased in all mar- kets is vine, red and sold loose (unpackaged). However, some specific tomato profiles Figure 5. In-sample simulation of selection probabilities for workers’ health and safety certification at 10% price increase. 258 Maria De Salvo, Riccardo Scarpa, Roberta Capitello, Diego Begalli have been identified for each market. In Germany, where Italian tomatoes are preferred to those coming from other countries, consumers ask tomatoes whose quality is certified for workers’ health and safety and eco-sustainability, but only within a restricted price range, as shown by the in-sample inference, where a ten percent increase was found too high. Salad-shaped tomatoes is more likely to be purchased when packaged in trays, while the use of this package should be avoided for vine-shaped tomatoes. In the UK, the same types of tomato certifications are also appreciated. However, the UK consumers seem to be not interested in the country of origin, unlike German consumers. Norwegian and Russian consumers adopt an intermediate behaviour. Consequently, tomatoes from Italy may not enjoy the same level of competitive advantage abroad, as it is generally assumed, and hence export penetration strategies should vary across countries. To sell more toma- toes in Norway, Italian producers should offer juicy-pulp tomatoes and certify their qual- ity with organic and worker’s health and safety labels. Cherry tomatoes are more appreci- ated in the UK market if are packaged in a net. Finally, Russians prefer thin-skin tomatoes and appreciate certifications for workers’ health and safety and eco-sustainability, rather than for organic production. Further research should address some of the limitations of our study in order to con- firm or disconfirm our findings, which were only illustrative in their nature. In fact, we are aware about a number of limitations of our study. They arise from the choices we have been forced to make regarding the experimental design and the data analysis. Firstly, to assure that the survey respect international quality standards for market research in a cross-country context, we decided to collect data engaging a market research company. The use of such online survey has grown rapidly in social science and policy research in the last ten years (Lehdonvirta et al., 2020). However, it is well known that data generated in this manner could be affected by self-selection issues and non-random and non-representativeness of the samples, and these limitations should be taken into account in evaluating the external valid- ity of our results. Further, to reduce the choice task complexity, we simulated a forced choice decision context, asking consumers to imagine they had to decide to buy one of the pro- posed options, without including an opt-out alternative. This decision has been supported by Dhar and Simonson (2003) who suggested that forced choice may generate more accu- rate and complete results in categories of familiar commodities in which the deferral option is available but rarely exercised. We assumed that this is the case of our research given that participants in our survey are consumers of fresh tomatoes, fresh tomatoes are character- ized by high versatility in cooking and individual diets, and the expenditure of this product has a low impact on the individual/household budget. However, we are aware that this can be seen as a limitation of our study. Therefore, market shares estimates could be affected by the adopted choice design. This possibility must be taken into account by the reader. Moreo- ver, each choice card includes several attributes and levels and, despite this well simulates the real-life scenario faced by consumers when purchasing fresh tomatoes, at the same time, respondents may not have attend to a certain number of attributes. An attribute-not-attend- ance phenomenon (Hensher, 2010) could consequently affect this survey as a limitation. We plan to analyse this eventuality through a further paper, given that it is not the focus in this one. Another limitation is related to the econometric approach. We chose to use Halton draws for simulations, despite the use of Scrambled Sobol draws could be more appropri- ate, as demonstrated by Czajkowski and Budziński (2019). Our choice stemmed from the 259Multi-country stated preferences choice analysis for fresh tomatoes fact that one of the aims of this paper is to provide the reader with estimation and post- estimation codes used in data analysis to facilitate dissemination. Further, it is worth observ- ing that we took the exporter viewpoint, and consequently we did not adjust prices accord- ing to the national purchasing power given that results are mainly presented at a country level. Therefore, it is important to underscore that, in the case of a country comparison, the same tomato profile could be perceived as relatively cheap or expensive in countries with different purchase powers. These cases could affect choice probability estimates. Finally, we used maximum likelihood estimators, which suffer from the limitation of local optima, and assumed normal and log normal distributions of qualitative attributes and price, respectively, for the random parameters. Assumptions of unimodal symmetric distributions surely affect our estimates and the analysis might also have been conducted with more flexible semi-par- ametric mixtures (Train 2016, Caputo et al. 2018, Scarpa et al. 2020). Despite these limitations, this study presented useful insights into consumer choices and their impact on market competitiveness for food producers. It demonstrated how the use of stated food choice experiments in a multi-country context is focal to support deci- sion makers in determining which types of product to grow and promote, how to manage the marketing mix, what communication content to emphasize in advertising campaigns and how adopt price differentiation strategies in different markets to face consumers’ demand. References Alamanos, E., Bourlakis, M., Tzimitra‐Kalogianni, I. (2013). Segmenting Greek tomato consumers: policy and marketing insights towards a healthy diet. British Food Jour- nal 115(4): 488-507. Balcombe, K., Chalak, A., Fraser, I. (2009). Model selection for the mixed logit with Bayesian estimation. Journal of Environmental Economics and Management 57(2): 226-237. Balcombe, K., Fraser, I., Di Falco, S. (2010). Traffic lights and food choice: A choice exper- iment examining the relationship between nutritional food labels and price. Food policy 35(3): 211-220. Balogh, P., Békési, D., Gorton, M., Popp, J., Lengyel, P. (2016). Consumer willingness to pay for traditional food products. Food Policy 61: 176-184. Caputo, V., Nayga Jr, R. M., Scarpa, R. (2013). Food miles or carbon emissions? Exploring labelling preference for food transport footprint with a stated choice study. Austral- ian Journal of Agricultural and Resource Economics 57(4): 465-482. Caputo, V., Scarpa, R., Nayga Jr, R. M. (2016). Cue versus independent food attributes: the effect of adding attributes in choice experiments. European Review of Agricultural Economics 44(2): 211-230. Caputo, V., Scarpa, R., Nayga Jr, R. M., Ortega, D. L. (2018). Are preferences for food quality attributes really normally distributed? An analysis using flexible mixing dis- tributions. Journal of choice modelling 28: 10-27. Carroll, K. A., Bernard, J. C., Pesek Jr, J. D. (2013). Consumer preferences for tomatoes: the influence of local, organic, and state program promotions by purchasing venue. Journal of Agricultural and Resource Economics 38(3): 379-396. 260 Maria De Salvo, Riccardo Scarpa, Roberta Capitello, Diego Begalli Crastes Dit Sourd, R., Beaumais, O., Mahieu, P. A., Martinez-Camblor, P., Scarpa, R. (2020). A contingent valuation test for measuring the construct validity of willing- ness-to-pay estimates derived from choice-experiments. Land economics, accepted and forthcoming. Daly, A., Hess, S., Train, K. (2012). Assuring finite moments for willingness to pay in ran- dom coefficient models. Transportation 39(1):19-31. Dhar, R., & Simonson, I. (2003). The effect of forced choice on choice. Journal of market- ing research 40(2): 146-160. European Union (2018): EU agricultural outlook for markets and income, 2018-2030. European Commission, DG Agriculture and Rural Development, Brussels. European commission (2020): DG agri-dashboard: tomatoes (https://ec.europa.eu/info/sites/ info/files/food-farming-fisheries/farming/documents/tomato-dashboard_en.pdf). Goett, A., K. Train, Hudson, K. (2000). Customer Choice Among Retail Energy Sup- pli- ers: The Willingness-to-Pay for Service Attributes. Energy Journal 21:1–28. Hensher, D.A. (2010). Attribute processing, heuristics and preference construction in choice analysis. In: Hess, S., Daly, A.J. (Eds.), State-of Art and State-of Practice in Choice Modelling: Proceedings from the Inaugural International Choice Modelling Conference, Chap. 3, pp. 35–70. Emerald, Bingley, UK. Hensher, D. A., Greene, W. H. (2011). Valuation of travel time savings in WTP and prefer- ence space in the presence of taste and scale heterogeneity. Journal of Transport Eco- nomics and Policy (JTEP) 45(3): 505-525. Hole AR. (2007). Fitting mixed logit models by using maximum simulated likelihood. Sta- ta Journal 7: 388-401. Hole, A.R. Kolstad, J., (2012). Mixed logit estimation of willingness to pay distributions: a comparison of models in preference and WTP space using data from a health- related choice experiment. Empirical Economics 42(2): 445-469. Layton, D.F., Brown. G. (2000). Heterogeneous Preferences Regarding Global Climate Change. The Review of Economics and Statistics 82(4): 616–24. Lehdonvirta, V., Oksanen, A., Räsänen, P., & Blank, G. (2020). Social Media, Web, and Panel Surveys: Using Non‐Probability Samples in Social and Policy Research. Policy & Internet. Liu, R., Gao, Z., Nayga Jr, R. M., Snell, H. A., Ma, H. (2019). Consumers’ valuation for food traceability in China: Does trust matter? Food Policy 88: 101768. Loureiro, M. L., Umberger, W. J. (2007). A choice experiment model for beef: What US consumer responses tell us about relative preferences for food safety, country-of-ori- gin labeling and traceability. Food Policy 32(4): 496-514. Mabit, S., Caussade, S., Hess, S. (2006). Representation of taste heterogeneity in willing- ness to pay indicators using parameterization in willingness to pay space. In Euro- pean Transport Conference 2006. European Transport Conference 2006. Maples, M., Morgan, K. L., Harri, A., Hood, K., Interis, M. (2014). Consumer willingness to pay for environmental production attributes in tomatoes: a southeastern con- sumer survey, Selected paper prepared for presentation at the Southern Agricultural Economics Association 2014 Annual Meeting, Dallas, TX. Meyerding, S. G. (2016). Consumer preferences for food labels on tomatoes in Germany– A comparison of a quasi-experiment and two stated preference approaches. Appetite 103: 105-112. 261Multi-country stated preferences choice analysis for fresh tomatoes Meyerding, S. G., Trajer, N., Lehberger, M. (2019). What is local food? The case of con- sumer preferences for local food labeling of tomatoes in Germany. Journal of Clean- er Production 207: 30-43. Morey, E., Rossmann, K. (2003). Using Stated- Preference Questions to Investigate Varia- tions in Willingness to Pay for Preserving Marble Monuments: Classic Heterogenei- ty, Random Parameters, and Mixture Models. Journal of Cultural Economics 27(3/4): 215–29. Oltman, A. E., Jervis, S. M., Drake, M. A. (2014). Consumer attitudes and preferences for fresh market tomatoes. Journal of food science 79(10): S2091-S2097. Onozaka, Y., McFadden, D. T. (2011). Does local labeling complement or compete with other sustainable labels? A conjoint analysis of direct and joint values for fresh pro- duce claim. American Journal of Agricultural Economics 93(3): 693-706. Ortega, D. L., Wang, H. H., Wu, L., Olynk, N. J. (2011). Modeling heterogeneity in con- sumer preferences for select food safety attributes in China. Food Policy 36(2): 318- 324. Owusu Coffie, R., Burton, M. P., Gibson, F. L., Hailu, A. (2016). Choice of rice production practices in Ghana: a comparison of willingness to pay and preference space esti- mates. Journal of Agricultural Economics 67(3): 799-819. Palma, M. A., Vedenov, D. V., and Bessler, D. (2018). The order of variables, simulation noise, and accuracy of mixed logit estimates. Empirical Economics 58: 2049–2083. Revelt, D., Train, K. (1998). Mixed logit with repeated choices: households’ choices of appliance efficiency level. Review of economics and statistics 80(4): 647-657. Revelt, D., Train, K. (2000). Customer-specific taste parameters and mixed logit: House- holds’ choice of electricity supplier, working paper No. E00-274, University of Cali- fornia at Berkley, Department of Economics, Berkley, California. Rose, J. M., Masiero, L. (2010). A comparison of the Impacts of Aspects of Prospect Theo- ry on WTP/WTA Estimated in Preference and WTP/WTA Space. European Journal of Transport and Infrastructure Research 10(4): 330-346. Santeramo, F. G., Carlucci, D., De Devitiis, B., Seccia, A., Stasi, A., Viscecchia, R., Nar- done, G. (2018). Emerging trends in European food, diets and food industry. Food Research International 104: 39-47. Scarpa, R., Thiene, M., Train, K. (2008). Utility in willingness to pay space: A tool to address confounding random scale effects in destination choice to the Alps. Ameri- can Journal of Agricultural Economics 90: 94-1010. Scarpa, R., Thiene, M., Marangon, F. (2007). The value of collective reputation for envi- ronmentally-friendly production methods: the case of Val di Gresta. Journal of Agri- cultural & Food Industrial Organization 5(1). Scarpa, R., Franceschinis, C., Thiene, M. (2020). Logit mixed logit under asymmetry and multimodality of WTP: a Monte Carlo evaluation. American Journal of Agricultural Economics, accepted and forthcoming. Skreli, E., Imami, D., Chan, C., Canavari, M., Zhllima, E., Pire, E. (2017). Assessing con- sumer preferences and willingness to pay for organic tomatoes in Albania: a con- joint choice experiment study. Spanish journal of agricultural research 15(3): 1-13. Sonnier, G., Ainslie, A., Otter, T. (2007). Heterogeneity distributions of willingness-to-pay in choice models. Quantitative Marketing Economics 5(3): 313-331. 262 Maria De Salvo, Riccardo Scarpa, Roberta Capitello, Diego Begalli Street D.J., Burgess L., Louviere J.J. (2005). Quick and easy choice sets: constructing opti- mal and nearly optimal stated choice experiments. International Journal of Research in Marketing 22(4): 459-470. Thiene, M., Scarpa, R., (2009). Deriving and testing efficient estimates of WTP distribu- tions in destination choice models. Environmental and Resource Economics 44: 379– 395. Train, K.E. (1998). Recreation demand models with taste variation. Land Economics 74: 230–239. Train, K. (1999). Mixed logit models for recreation demand, in J. Herriges & C. Kling, eds, ‘Valuing Recreation and the Environment’, Edward Elgar, Northampton, MA. Train, K., Weeks, M. (2005). Discrete choice models in preference space and willingness- to-pay space. In Applications of simulation methods in environmental and resource economics (pp. 1-16). Springer, Dordrecht. Train K., (2016). Mixed Logit with a Flexible Mixing Distribution. Journal of Choice Mod- elling 19: 40–53. Tonsor, G. T., Shupp, R. (2009). Valuations of ‘sustainably produced’ labels on beef, toma- to, and apple products. Agricultural and Resource Economics Review 38(3): 371-383. Yue, C., Tong, C. (2009). Organic or local? Investigating consumer preference for fresh produce using a choice experiment with real economic incentives. HortScience 44(2): 366-371. Zanoli, R., Scarpa, R., Napolitano, F., Piasentier, E., Naspetti, S., Bruschi, V. (2013). Organic label as an identifier of environmentally related quality: A consumer choice experiment on beef in Italy. Renewable Agriculture and Food Systems 28(1): 70-79. Zeng, T. (2016). Using Halton sequences in random parameters logit models. Journal of Statistical and Econometric Methods 5(1): 59-86. Investigating determinants of choice and predicting market shares of renewable-based heating systems under alternative policy scenarios Cristiano Franceschinis, Mara Thiene Multi-country stated preferences choice analysis for fresh tomatoes Maria De Salvo1,*, Riccardo Scarpa2,3,4, Roberta Capitello2, Diego Begalli2 “Not my cup of coffee”. Farmers’ preferences for coffee variety traits. Lessons for crop breeding in the age of climate change Abrha Megos Meressa, Ståle Navrud* Does the place of residence affect land use preferences? Evidence from a choice experiment in Germany Julian Sagebiel1,*, Klaus Glenk2, Jürgen Meyerhoff3 The use of latent variable models in policy: A road fraught with peril? Danny Campbell*, Erlend Dancke Sandorf