To assess the influence of relevant determinants of the WTP, we develop a framework as presented in Figure 1. The WTP can be affected by internal and external factors. In our framework, we have grouped these factors into (1) consumer characteristics, (2) product assessment, and (3) environmental factors.

Many studies found an impact of demographic characteristics on consumers' decisions. At the household level, the number and attributes of family members would motivate the buyer to purchase safe food products (Zheng et al., 2018; Chege et al., 2019; Kytö et al., 2019; Neill and Holcomb, 2019). At the individual level, the age, education, and income of the buyer have contradictory effects on their WTP. While a higher level of education or income led to higher WTP (Angulo and Gil, 2007; Mergenthaler et al., 2009; Li et al., 2016; Zheng et al., 2018; Chege et al., 2019; Riccioli et al., 2020; My et al., 2021), the older consumers had a lower WTP for premium food products (Yu et al., 2014, 2018). In addition, the food consumption habit of the consumers also significantly increased the probability to purchase the food (Kytö et al., 2019). Angulo and Gil (2007) found that the level of beef consumption is a key factor that influences consumers' WTP for beef products, while Yu et al. (2018) and Zheng et al. (2018) found similar results for salmon and vegetable products. Researchers also reported the positive impact of risk perception about food-borne diseases on the WTP (Angulo and Gil, 2007; Mergenthaler et al., 2009; Yu et al., 2018; Neill and Holcomb, 2019), while the enhancement in food safety knowledge might correspond to increasing WTP for safe vegetables (Mergenthaler et al., 2009).

Furthermore, consumers' exposure to information about the food product's characteristics can significantly affect their WTP. For example, many studies showed that the product description on the label strongly motivated consumers to pay more for food (Meenakshi et al., 2012; Zhang et al., 2012; Jin et al., 2017; Chege et al., 2019; Liu et al., 2019; Katt and Meixner, 2020; Riccioli et al., 2020). Important information that enhances consumers' WTP in many studies was the certification of the food products (Owusu-Sekyere et al., 2014; Ortega and Tschirley, 2017; Wang and Tsai, 2019), even if it is not a government certificate (Liu et al., 2019). Therefore, some alternative methods to make consumers distinguish between the different attributes among products should be considered so that they may reveal their true WTP. Nonetheless, providing consumers with messages about food safety risks right before they make their decision might be a critical point that affects WTP. Britwum et al. (2019) found that the message about reported cases of disease due to microbial contamination in food motivated consumers to pay more for safe food, while Bruner et al. (2014) reported a reduction in WTP for traditional food products due to the information about the estimated risk of the new food products. In contrast, the experiment by Hayes et al. showed that consumers were not affected by providing the figure about the probability of food-borne diseases (Hayes et al., 1995).

Finally, consumers' assessment of food products and food stores was identified as important determinants of their WTP for food. Owusu-Sekyere et al. (2014) indicated the hygienic condition surrounding the shop significantly affected consumers' WTP for the food product, while Zheng et al. (2018) found that a high assessment of the food by consumers motivated them to purchase the product.

In this study, conducted between October and November 2021, we used a Becker–DeGroot–Marschak (BDM) mechanism with a full bidding approach to measure the WTP of consumers for raw pork at traditional pork shops in Vietnam. This experiment design creates a market environment where participants can incorporate market feedback and reveal their value for the product via a bid (Lusk and Shogren, 2007). In other words, the participants will use real money to purchase the product from a set-up market, which can improve the reliability of the study (Koschate-Fischer and Schandelmeier, 2014), especially when compared with survey methods where participants tend to hide their true behavior (Kytö et al., 2019). In addition, Jaffee et al. (2019) indicated that the precision of consumers' WTP could be improved by presenting alternative products and letting them make decisions under their usual budget constraints.

This study was part of a larger project (SafePORK) aiming to improve food safety at traditional markets focusing on pork (ACIAR, 2016) through food safety interventions at different levels, including upgrading pork shops. The experiments were implemented at three traditional markets where some pork shops had been upgraded as part of the interventions. Two markets were located in Thai Nguyen province and one in Hung Yen province in the North of Vietnam. Traditional markets represent the most popular retail channel in Vietnam for distributing fresh food such as animal-source food, vegetables, or fish (Nga et al., 2014; Unger et al., 2019). In each market, two pork shops were set up for the experiments, as described in the following section.

On the day prior to the experiment, the research team came to the selected markets to recruit participants from potential consumers. One out of every three consumers who visited the market was asked to participate in the study. Upon their consent, the respondents who intended to buy pork on the following day would be interviewed using a structured questionnaire. Once completed, they received a coupon and were invited to attend the experiment on the following day. In total, 152 consumers (Nhai market—Hung Yen, n = 52; Dong Quang—Thai Nguyen, n = 50, and Dan market—Thai Nguyen, n = 50) were recruited, finished the interview, and participated in the experiment.

On the experiment day, two pork shops were set up at the selected markets. Each shop was supplied with 25 kg of pork shoulder sourced from the same slaughterhouse and delivered on the morning of the experiment day. The experiment was held on a day when the market was closed, to limit the interference of the market operation. According to previous studies (Nguyen-Viet et al., 2019; Thi Nguyen et al., 2019), pork shoulder is the most popular choice for Vietnamese consumers, so it was chosen as the product for the experiment. The experiment included two types of traditional pork shops: upgraded and typical pork shops. An upgraded pork shop had taken part in the food safety intervention from the SafePORK project. Both shops were equipped with fundamental tools for traditional pork shops, with a set of tools (cutting board, knives, scale, cloth to wipe hands, and other tools) and protection clothing (apron and mask) for the seller. The upgraded shop was provided with a disinfection package (sprayer, disinfection liquid for cleaning surfaces, and hand sanitation gel) and a poster to motivate the seller to frequently clean hands, surfaces, and tools, as well as to introduce recommended food safety practices to the consumers at a pork shop. In addition, the typical shop used paperboard to display pork on the granite table, while the upgraded shop displayed pork directly on the granite surface. The pork at both types of shops was supplied from the same slaughterhouse on the experiment day to make sure that the quality of pork was affected by the seller and the shop's facilities only.

The Becker–DeGroot–Marschak (BDM) mechanism (Becker et al., 1964; Lusk and Shogren, 2007), using a full bidding approach, was selected for this study. In this type of experiment, the participants compete against a random price by giving their full bid for 0.5 kg of shoulder pork. If the bid is higher than the random price, the consumer purchases the pork at the random price; if the bid is lower, the consumer does not purchase. To avoid demand reduction effects, only one type of pork shop was allowed for each individual buyer during the game (Lusk and Shogren, 2007). The BDM mechanism (Figure 2) had four main steps for each individual participants as follows:

The random price of pork shoulder was 40,000 VND per half a kilogram, while the random price was generated following uniform distribution and not shared with the participants. The range of the random price was 10,000–80,000 VND to cover the potential values and not limit the winning opportunity of the participants with reasonably high valuations (Lusk and Shogren, 2007). The mechanism is illustrated in Figure 2.

The research team developed a structured questionnaire that covered the potential variables in the conceptual framework and consisted of six parts: demographic information, pork preference, experience of food-borne diseases, perceptions about food safety, practices to prevent pork-borne diseases, and knowledge about pork safety. The demographic part included age, gender, occupation, education, and household characteristics (such as food expenditure and household size). The part assessing consumers' preference for pork covered their pork consumption habits (such as frequency, amount, or type of pork), while the experience of food-borne diseases focused on some common food poisoning symptoms (including stomachache, diarrhea, nausea, and vomiting).

The perception component comprised two parts: food safety practices at pork shops and the risk of pork-borne diseases. The first part included beliefs (measured by a five-level ordinal scale) in seven practices (that promote food safety) at the upgraded pork shop and one practice (that reduces food safety) at the typical one. The questions and results of this component are presented in Appendix A.

Practice to reduce the effect of pork-borne diseases was measured by asking the participants about the frequency and effectiveness of five practices on a five-level ordinal scale. The knowledge about pork safety was assessed by ordinal questions (yes/no/do not know) adapted from da Cunha et al. (2019), with 12 questions in total. The questions and answers on food safety knowledge are presented in Appendix B.

The internal consistency of the questions assessing perception about food safety at the pork shop (eight questions), practice to prevent pork-borne diseases (10 questions), and food safety knowledge (12 questions) was tested using Cronbach's alpha, with the results 0.903, 0.724, and 0.809, respectively, showing high internal consistency and adequate reliability of the questions (di lorio, 2005).

Interview data and experiment results were entered in Microsoft Excel. Descriptive analysis was applied to describe the characteristics of variables, while the mean WTP and pork shop assessment between different groups were compared using Wilcoxon signed rank test.

To assess the perception of consumers about the food safety practice of pork sellers, we calculated the overall score by adding up the score of each of the eight practices. The seven practices that promote food safety were graded from 1 to 5 points per question, while poor practice was graded from−5 up to−1 point. The overall perception score ranged from 2 to 34 points.

Regarding food safety knowledge, each response was marked 0 (for an incorrect answer) or one (for a correct answer) and then summed up to make the total score (ranging from 0 to 12 points). The Spearman rank correlation test was used to determine the relationship between the perception about food safety at pork shops, assessment of pork shops, food safety knowledge, and attitude about the risk of pork contamination.

Univariable analyses were implemented first to identify variables to include in the multivariable models. Variables were included if they had a p-value of ≤ 0.1 in univariable analyses. For the regression models, the dependent variables were the WTP (1,000 VND) for pork from the typical and upgraded shop and the difference in WTP for 0.5 kg between the two products. The linear quantile regression with the market variable as a random effect was implemented for all three models at 10, 50, and 90% quantile for the bids using the lqmm package in R (Geraci, 2014).

This study was reviewed and approved by the Institute Review Board at the Hanoi University of Public Health (No. 110/2018/YTCC-HD3). Verbal informed consent was obtained from each participant before conducting the interview.

Table 1 describes the characteristics of the participants. The average respondent was of middle age (51.9 years old on average), with 9.22 years of education, and most were female (90.1%). Many participants were small-scale vendors (37.5%) and rice farmers (28.9%), followed by those who took care of the household (16.4%). The mean participants' household size was 3.98 with many of them not having any children or being elderly (73 and 50.7%, respectively). In other words, most of the household members were working-age adults. Furthermore, on average, they had 6.44 pork dishes per week and purchase 1.08 kg of pork per shopping time. The most purchased pork type was bacon (65.1%), followed by the shoulder (21.7%).

Table 2 presents the experiences of participants with food-borne disease and their food safety practice and knowledge. The participants reported that they rarely suffered from common food poisoning symptoms. The most regular symptoms among the respondents and their families were stomachache (15.8%), followed by diarrhea (10.5%), nausea (7.9%), and vomiting (4.6%). In addition, the respondents reported that they regularly implemented food safety practices at home, especially eating well-cooked pork only (96.7%) and separating raw pork and cooked food (94.7%). Furthermore, the participants' average knowledge score was 10.96 (out of a maximum score of 12), with a standard deviation of 2.03. Pork-borne diseases seemed to be a concern for respondents since more than 90% worried about eating contaminated pork while more than half of them (57.2%) believed that slaughtering on a grid, instead of the floor, can improve the safety of pork.

Overall, participants gave positive feedback on the intervention packages at the upgraded pork shops. More than 80% of respondents believed that the suggested practices at upgraded shops would improve the safety of pork. In contrast, for the poor practice, which was carried out at the typical shop (placing the pork on the wooden table or carton board), half of the respondents believed this practice would improve the safety of pork, while only little more than one-third believed it would reduce the safety of pork. The details are presented in Appendix A. In addition, in the experiment, the participants evaluated the overall food safety condition of the upgraded shop (9.3/10) significantly higher than the typical one (7.6/10) (p < 0.05). The details are presented in Table 3.

The independence of variables was tested between pork shop assessment, food safety perception, and knowledge. The result showed a weak association between factors except for perception about food safety and concern about eating contaminated pork. The detailed correlation is presented in Table 4.

The experiment showed that the typical pork received the mean bid at 32,500 VND per 0.5 kg, while the figure for upgraded pork was 39,000 VND per 0.5 kg. Thus, the respondents were willing to pay a premium of ~6,500 VND per 0.5 kg (or 20%) for the upgraded pork compared with the typical one. The detailed result is presented in Table 5.

Table 6 describes the relationship between relevant factors to the bids and the difference in the bids of the two pork types. Overall, the pork consumption habits and household characteristics had a strong impact on the WTP for both pork products while the risk message had no effect. On the other hand, the difference in food safety assessment between the two shops was the only indicator that significantly affected the deviation of the WTP for each product. In addition, the market cluster effect did not cause a significant impact on the WTP. The tendency of each coefficient for each variable across quantile levels is presented in Figures 3–5.

For the upgraded pork, the number of pork dishes per week, the number of elderlies in the household, the difference in assessment between the two shops, and the perception about food safety practice significantly increased the consumers' WTP while the perception about food safety at pork shops show a negative impact. However, the number of pork dishes per week is not significant in the 25th percentile of WTP, whereas the perception score only affects in 25th percentile of WTP (Figure 3). Finally, the amount of pork in each shopping time, the participants' knowledge and perception about contaminated pork, and the risk message did not have any relationship with their WTP in any quantile.

The WTP for pork from typical shops was significantly affected by most variables except the knowledge score, the perception about food safety at pork shops, and the risk message. In detail, the consumption habit (the amount of pork in each shopping and the number of pork dishes per week) and the number of elderlies in the household caused a positive impact on WTP in all quantiles (Figure 4). On the contrary, concern about eating contaminated pork and the shop assessment difference had a negative effect, while the perception about pork shop practices only affects a low percentile of WTP (10th percentile).

The difference in shop assessment, the number of pork dishes per week, and the perception score are the only factors that affect the difference in WTP between the two types of pork (Figure 5). Although both variables significantly increased the difference, the first had an impact on all percentile of WTP while the other two only have an impact on the 60th percentile or higher.

This study did not cover participants in urban areas where supermarkets and convenience stores are strong competitors to traditional pork shops. Further studies should be implemented to find additional information about this group. Another limitation is the limited choice of pork type in the study, which may not completely reflect consumers' preferences. However, this design helped us to reduce bias in consumers' decisions due to the demand reduction effect. Finally, the convenience sampling procedure with a small sample size in this study may cause bias in the estimate of the regression model.