1. Introduction
Food animal production is intrinsically linked to the health of animals, humans, and their shared environment. Consumers, particularly those of high-income countries, are increasingly concerned about the intangible aspects of food animal production such as its impact on climate change, farm animal welfare, food quality, and food safety (Messer et al., Reference Messer, Costanigro and Kaiser2017; Perry and Grace, Reference Perry and Grace2015). Consumers influence the way in which food animals are produced through market signals of premiums they are willing to pay for foods with specific qualities, and as constituents through voting and policy initiatives related to food animal production (McMullen and Halteman, Reference McMullen and Halteman2019; Mullally and Lusk, Reference Mullally and Lusk2018; Olynk et al., Reference Olynk, Tonsor and Wolf2010). In response, producers adopt these production practices and supply goods with the specific qualities demanded if it is profitable to do so or if they are mandated (Caswell and Mojduszka, Reference Caswell and Mojduszka1996). Understanding what drives consumer preferences for food animal products is important to ensure that the market signals incentivize production practices that are economically, socially, and environmentally sustainable.
In response to consumer concerns, producers can provide information regarding production processes using labeling to signal credence attributes (Olynk et al., Reference Olynk, Tonsor and Wolf2010). A credence attribute is a characteristic of a good for which quality cannot be determined even after the purchase and consumption of the product (Caswell and Mojduszka, Reference Caswell and Mojduszka1996). In other words, without the credence attribute label, a consumer cannot identify the difference between a product that possesses those attributes described by the attribute and a product without. The value that the consumers derive from purchasing a product labeled to signal credence attribute existence depends on the individuals’ beliefs about the credence attribute and what it means in terms of various quality dimensions (Costanigro et al., Reference Costanigro, Deselnicu and Kroll2015). Such quality dimensions include factors such as environmental friendliness, animal welfare, climate impacts, food safety, and origin. Consumers form beliefs about these attributes using the information they have available, and these beliefs are often persistent because the corresponding perceived quality is unverifiable – particularly those related to sustainability claims – because they are not directly tied to measurable or verifiable outcomes (Neuhofer et al., Reference Neuhofer, Lusk and Villas-Boas2023).
The organic label is in widespread use as a signal for credence attributes and is known for possessing a “halo-effect.” Though the organic label does not necessarily claim sustainability or health advantages, consumers associate organic food with being superior in numerous quality dimensions such as nutrition, environmental impact, food safety, and health (Ellison et al., Reference Ellison, Duff, Wang and White2016; Lee et al., Reference Lee, Shimizu, Kniffin and Wansink2013). However, these beliefs are not necessarily backed by scientific evidence. Multiple studies have found no difference between the health and safety of organically versus conventionally produced food (Smith-Spangler et al., Reference Smith-Spangler, Brandeau, Hunter, Bavinger, Pearson, Eschbach and Bravata2012), and others emphasize the need to describe the environmental impact of organic versus conventional food production in a more nuanced fashion because of the clear tradeoffs between the two systems (Klopatek et al., Reference Klopatek, Marvinney, Duarte, Kendall, (Crystal) Yang and Oltjen2022; van Wagenberg et al., Reference van Wagenberg, de Haas, Hogeveen, van Krimpen, Meuwissen, van Middelaar and Rodenburg2017). For example, (Klopatek et al., Reference Klopatek, Marvinney, Duarte, Kendall, (Crystal) Yang and Oltjen2022) found that grass-feeding beef cattle reduced smog formation potential relative to conventional production, but also had greater consumptive water use. For stakeholders, understanding consumer beliefs pertaining to organic and conventional food production is important when considering food marketing and policy options.
Consumer beliefs about credence attributes are becoming increasingly important factors in understanding what drives consumer preferences for food animal products (Costanigro and Onozaka, Reference Costanigro and Onozaka2020; Malone and Lusk, Reference Malone and Lusk2018; Neill and Williams, Reference Neill and Williams2016; Neuhofer and Lusk, Reference Neuhofer and Lusk2021). Many consumer demand studies have failed to distinguish beliefs from preferences, resulting in misinterpretation of the motives of food choice, incorrect inference, and misleading welfare analysis (Costanigro et al., Reference Costanigro, Deselnicu and Kroll2015; Lusk, Schroeder, and Tonsor, Reference Lusk, Schroeder and Tonsor2014). As far back as 1975, Fishbein and Ajzen described the role of beliefs, attitudes, and intentions in shaping behavior (Fishbein and Ajzen, Reference Fishbein and Ajzen1975). Other previous work has defined beliefs as subjective probabilities of attaining different outcomes (Lusk et al., Reference Lusk, Schroeder and Tonsor2014), or as the mediator of a consumers’ assessment of the quality of a good given the product cues (Costanigro et al., Reference Costanigro, Deselnicu and Kroll2015). As Steenkamp (Reference Steenkamp1990) puts it, “these cues [attributes] are valued because the consumer thinks [believes] they say something about the taste (and possibly other quality attributes) of the product” (p. 313). Utility is derived from the bundle of product attributes that can include both intrinsic and extrinsic attributes. Intrinsic attributes are those for which a direct benefit is derived from consumption. Extrinsic attributes are those for which benefits derived depend on an individual’s belief that the attribute implies a better product (Steenkamp, Reference Steenkamp1990). Preferences are the subjective ranking of attribute bundles, reflecting the relative importance an individual places on the intrinsic and extrinsic attributes. Costanigro and Onozaka (Reference Costanigro and Onozaka2020) explore this belief-preference dynamic via two discrete choice experiments. The first step involved a quality sorting task where respondents indicated which option they believe to be superior in terms of a quality dimension. The next step involved the product choice task where respondents indicated which option they would purchase. In doing so, they were able to distinguish beliefs from preferences to better understand how attributes affect food choice.
Traditionally, economists have viewed preferences as relatively stable constructs, whereas beliefs are assumed to be more malleable (Huffman et al., Reference Huffman, Rousu, Shogren and Tegene2007; Lusk et al., Reference Lusk, Schroeder and Tonsor2014). While there are countless studies that evaluate preference malleability to various interventions, relatively fewer studies exist that explore the malleability of beliefs in the context of food choices. Bayesian decision theory asserts that when individuals are presented with new information, their beliefs are updated by combining the prior belief with new information received, with the magnitude of updating being a function of the weights given to prior versus new information (Neuhofer et al., Reference Neuhofer, Lusk and Villas-Boas2023). In some cases, individuals place virtually no weight on new information (Lewandowsky, Cook, and Lloyd, Reference Lewandowsky, Cook and Lloyd2018). Consequently, individuals may hold unwavering prior beliefs, which could be attributed to identity-protective cognition (Kahan et al., Reference Kahan, Braman, Gastil, Slovic and Mertz2007) or lack of trust in the news source (Bayes and Druckman, Reference Bayes and Druckman2021). Other cognitive biases, such as confirmation bias, negativity bias, and illusionary correlations result in violations of Bayesian decision theory (Charness and Levin, Reference Charness and Levin2005; McFadden and Lusk, Reference McFadden and Lusk2015; Tversky and Kahneman, Reference Tversky and Kahneman1981), or surprising alterations of beliefs (Neuhofer et al., Reference Neuhofer, Lusk and Villas-Boas2023). Consequently, much remains to be discovered about belief formation, modification, and its relation to food purchasing decisions.
The objective of this research was to assess malleability in consumer beliefs about credence attributes and corresponding changes to consumer preference for the set of credence attributes relative to observable product attributes. We aim to demonstrate that the perceived quality of a product, as informed by its credence attributes, is not particularly stable. In this study we consider consumer preferences for ground beef, with a label for organic production representing the quality derived from a set of credence attributes. Fresh ground beef was selected as the product for this analysis because it has limited product variation – percent lean and packaging – and is widely consumed in the United States (Bina and Tonsor, Reference Bina and Tonsor2024). We employ two discrete choice experiments to address our research questions: 1) Does information alter consumer beliefs about organic label credence attributes for ground beef?, 2) Does the perceived quality attribute represented by the organic label remain stable relative to other ground beef attributes if, and when, beliefs change?, and 3) What are the effects on willingness-to-pay for organic ground beef if belief about the credence attributes change? By answering these questions, we contribute to the growing literature exploring the relationship between beliefs and preferences regarding food choices and provide insights for animal protein stakeholders interested in developing marketing strategies that resonate with consumers.
2. Materials and methods
2.1. Conceptual framework
The belief-preference conceptual framework we use was first developed by Costanigro and Onozaka (Reference Costanigro and Onozaka2020). Briefly, the framework follows the Lancaster (Reference Lancaster1966) approach and defines quality as a good’s effectiveness in its proposed purpose, and that the utility a consumer derives from a good is from the intrinsic quality of the good, not just the good itself. The quality of a food good is inherently multifaceted, as consumers enjoy a complex combination of qualities such as taste, smell, and nutrition. We can assume that the perceived quality of good j is the summation of a Q-dimensional vector of quality dimensions Qj= (Qj 1,Qj 2,…Qj Q). If individual i consumes good j, they will realize utility, Uij = (Qj,Pricej,γi), where Qj is the perceived quality of the good, γi is a vector of consumer specific preference weights, and Pricejis the price paid for good j (Costanigro and Onozaka, Reference Costanigro and Onozaka2020).
To evaluate the quality of a good, the framework assumes that consumers use intrinsic and extrinsic attributes of a good to generate perceived quality (Brunswik, Reference Brunswik1955; Dudycha and Naylor, Reference Dudycha and Naylor1966; Steenkamp, Reference Steenkamp1990). The extent to which a good’s attributes influence quality depends on a consumer’s subjective beliefs about the attribute itself. In other words, beliefs serve to map attributes into perceived quality.
To capture these belief and preference parameters from choice data, the food choice process is separated into two steps. The first step (deemed the quality sorting task) involves asking consumers which good is superior in a given quality dimension conditional on the good’s attributes. In the second step (deemed the product choice task), price information is added to the attributes and consumers are asked to indicate which, if any, good they would purchase. Consider a scenario where consumer i must choose which of good A or good B is superior in terms of quality dimension q. Consumer i’s perceived quality, Q, in dimension q, for alternative j in choice set t, becomes:
$${\rm Q}{{\rm _{ijt}}}^{{\rm q}}={\rm Q}({\bf X}_{{\bf ijt}};{\bf {\unicode{x03B2}}} {{\rm_ i}}^{{\rm q}})+\rm\varepsilon _{{\rm_ {ijt}}}^{{\rm q}}={\bf X}{{\bf _{ijt}}}{\bf \prime}{\bf {\unicode{x03B2}}} {{\rm_ i}}^{{\rm q}}+{{\rm _{ijt}}}^{{\rm q}}$$
Where Xijt
is a vector of good attributes,
${{\boldsymbol \beta} _{\rm i}^{\rm q}}$
is a vector of consumer-specific belief coefficients, and
${ \epsilon _{ ijt}^{q}}$
is the error term. The above equation is framed as quality because the question presented to respondents is to indicate which alternative they believe is superior in terms of the q quality dimension, not which alternative they would purchase. The probability that good A is perceived to be superior to good B in quality dimension q can be written as
${\rm Pr\; ({\rm Q_{iA}^{q}}\gt {\rm Q_{iB}^{q}})=Pr(({\bf X_{iA} - {\bf X _{iB}}})\prime {\bf \beta}_{\rm i}^{\rm q}\gt {\rm \epsilon}_{\rm iB}^{\rm q} - {\rm \epsilon}_{\rm iA}^{\rm q})}$
. Once subjective quality perceptions have been formed, the model for how consumers make choices about food purchases becomes:
$${\rm U}_{{\rm ijt}}={\rm U}({\bf Q}_{{\bf ijt}};{\rm Price};{\bf \gamma} )+{\rm v}_{{\rm ijt}}={\bf Q}_{{\bf ijt}}{\bf \prime}{\bf \gamma} +{\rm Price}_{{\rm j}}\gamma _{{\rm price}}+{\rm v}_{{\rm ijt}}$$
Where Uijt is the utility consumer i receives from alternative j in choice set t, γ is a Qx1 vector of coefficients that capture the marginal utility of each quality, γprice captures the marginal disutility of price, and vijt is the error term. The probability that good A is purchased rather than good B is given by: Pr (UiA>UiB) = Pr ((QiA−QiB
)′γ +(PriceA−PriceB)′γprice>viB−viA). As researchers, we do not directly observe Qijt. Hence, we can use estimates obtained from the belief model to predict perceived quality:
${\rm U}_{{\rm ijt}}={\rm U}(\widehat{{\bf Q}}_{{\bf ijt}};{\rm Price};{\bf\gamma} )+{\rm v}_{{\rm ijt}}=\widehat{{\bf Q}}_{{\bf ijt}}^{{\rm \prime}}{\bf\gamma} _{1}+{\rm Price}_{{\rm j}}\gamma _{{\rm price}}.$
Furthermore, we can allow for belief heterogeneity by estimating the above belief model using a random parameters (mixed) logit model.
As demonstrated by Costanigro and Onozaka (Reference Costanigro and Onozaka2020), the above specified belief-preference model allows for a decomposition of willingness-to-pay for an attribute via the Q-dimensional pathways of perceived quality. In the traditional utility model, an attribute’s marginal utility can be calculated as
${\partial U_{ij} \over \partial x_{k}}=\delta _{k}$
and the corresponding willingness-to-pay for the kth attribute is WTP
k
= δ
k
/ − δ
price
. In the belief-preference model, we can define the utility garnered from attribute xk as:
${\partial {\rm U}_{{\rm ijt}} \over \partial {\rm x}_{{\rm k}}}=\sum _{{\rm q}}{\partial {\rm U}_{{\rm ijt}} \over \partial {\rm Q}_{{\rm j}}^{{\rm q}}}\cdot {\partial {\rm Q}_{{\rm ijt}} \over \partial {\rm x}_{{\rm k}}}=\sum _{{\rm q}}{\rm {\rm \gamma}} _{{\rm q}}{\rm \beta} _{{\rm k}}^{{\rm q}}$
, and the part-worth decomposition of the WTP for attribute xk becomes:
$${\rm WTP}_{{\rm k}}={{\rm \gamma} _{1}{\rm \beta} _{{\rm k}}^{1} \over -{\rm \gamma} _{{\rm price}}}+{{\rm \gamma} _{2}{\rm \beta} _{{\rm k}}^{2} \over -{\rm \gamma} _{{\rm price}}}+\ldots +{{\rm \gamma} _{{\rm q}}{\rm \beta} _{{\rm k}}^{{\rm q}} \over -{\rm \gamma} _{{\rm price}}}$$
There is a summation across multiple belief and preference coefficients because one attribute can influence more than one perceived quality dimension.
2.2. Data and experimental design
An online Qualtrics survey was administered via Toluna Panels the first week of May 2023. Toluna Panels is a market research company that provides a platform for consumers to participate in surveys and other human subjects research. The survey was sent to a panel of U.S. consumers that was designed to be nationally representative according to age, ethnicity, income, and gender (Blakeslee et al., Reference Blakeslee, Caplan, Meyer, Rabe and Roberts2023). Prior to administering the survey, the research protocol was approved by the Institutional Review Board (IRB2023-348). To qualify for the survey, respondents had to be above the age of 18, red meat consumers, and the primary person in their household responsible for purchasing and preparing food. A total of 1,823 responses were collected. To ensure quality responses, participant data was dropped if: 1) attention check questions were answered incorrectly, or 2) the duration of the survey was less than 5 minutes. The descriptive statistics of the 733 participants retained for the analysis are provided in Table 1. The final column of the table lists demographic data collected from the U.S. Census Bureau. The composition of our sample pairs favorably with these estimates. Small deviations exist in terms of our sample containing a higher proportion of African American respondents and female respondents.
Table 1. Sample descriptive statistics
Notes: U.S. Age and Gender information from (US Census Bureau, n.d.) Income from (Guzman and Kollar, Reference Guzman and Kollar2023).
The questionnaire portion of the survey included demographic questions, questions about perceptions of the importance of environmental, animal welfare, and food safety concerns, and how conventional and organic beef cattle production impact these concerns. Interestingly, while most respondents indicated that the way beef cattle are raised significantly impacts the safety of beef products and the health of the environment, most were unsure if their food choices influence the way in which food animals are raised. The experimental portion of the survey included two discrete choice experiments intended to illicit respondents’ beliefs and preferences.
Given the importance of beef to the diet of U.S. consumers, ground beef was selected as the good of interest for the choice experiments. According to the Beef Research Center, over two-thirds of U.S. consumers eat beef on at least a weekly basis and 2021 per capita net beef consumption was 58.0 pounds (Beef Research, 2022). Beef production is also inherently tied to the health of the environment, humans, and cattle themselves through One Health challenges such as greenhouse gas emissions, antimicrobial resistance, and animal health and welfare. Correspondingly, we anticipated consumers to possess strong beliefs regarding the impact of beef cattle production on these holistic health challenges.
Following the approach of Costanigro and Onozaka (Reference Costanigro and Onozaka2020), the core of the survey included two discrete choice experiments. The first experiment, the quality sorting task, asked respondents to determine which ground beef option (if any) was superior (A, B, or “there is no difference”) in a specific quality dimension. The three quality dimensions (Q = 3) considered were environmental friendliness, animal welfare, and food safety. These quality dimensions were chosen based on previous literature exploring non-economic factors that are important to consumers when selecting to purchase organic or non-organic products (Danner and Menapace, Reference Danner and Menapace2020; Yiridoe et al., Reference Yiridoe, Bonti-Ankomah and Martin2005). The two attributes included were organic status and percent lean. The organic status included two levels (yes or no) and was the attribute of interest for this study. Percent lean included three levels (70 – 79%, 80 – 89%, >90%) and has been shown to be an important factor in consumer preferences for ground beef (Lusk and Parker, Reference Lusk and Parker2009). An example of the quality sorting task is presented in Figure 1.
Figure 1. Example choice set in discrete choice experiment 1 (Quality sorting task).
Notes: In the quality sorting task, respondents were asked to indicate which ground beef option they thought was superior in three quality dimensions: environmental friendliness, animal welfare, and food safety.
In the second experiment (product choice task), price information was added to the attributes described above and respondents were asked to indicate which option, if any, they would purchase. Price levels were based on the USDA Agricultural Marketing Service’s weekly advertised retail prices for ground beef from January – November 2022 for 70 – 79%, 80 – 89%, and 90% + lean ground beef (USDA-AMS 2023). From this data, a national weighted average retail price of $4.50. Variations in price were established by allowing for up to 33% deviation from the mean at $0.50 increments. For the experiment, therefore, the price per pound of ground beef varied from $3 to $6 in 50 cent increments. An example of the product choice task is presented in Figure 2.
Figure 2. Example choice set in discrete choice experiment 2 (Product choice task).
For both choice experiments, an orthogonal, efficient fractional factorial design was used to generate choice sets. For the quality sorting choice experiment, the design resulted in 16 alternatives randomly assigned into eight choice sets. The eight choice sets were then randomly allocated into two blocks, resulting in respondents receiving four choice scenarios for each of the three quality dimensions. For the second choice experiment, the design resulted in 52 alternatives, which would have required 26 choice sets. To ensure an even number of blocks and reasonable number of choices per participant, an optimal design was generated under the constraint of 50 alternatives in 25 randomly assigned choice sets. Those 25 choice sets were randomly allocated to five blocks, with each respondent receiving five choice scenarios for the product choice experiment.
In order to assess the malleability of beliefs, the following five information treatments were introduced: Organic Positive, Organic Negative, Neutral, Conventional Positive, and Conventional Negative. Respondents were randomly assigned to one of five information treatments, where they received written information that presented organic or conventional beef production in either a positive or negative light within each of the three quality dimensions (environmental friendliness, animal welfare, and food safety). Information was selected from news outlets that consumers would be likely to see and used language that would be understandable to the average consumer. The information treatments are provided in the Appendix (Rocha, Reference Rocha2023; TCFA n.d.; Mayo Clinic, 2023; Animal Welfare Institute n.d.). The Neutral information treatment was included to serve as a reference for potential movement of beliefs due to respondent attrition, and only presented general information about beef cattle with no reference to either organic or conventional production. Differences in the composition of respondents in each information treatment was assessed using the Kruskal–Wallis One-Way ANOVA. No statistically significant differences were found in terms of age (χ 2 = 2.06, P = 0.73), gender (χ 2 = 0.62, P = 0.96), income (χ 2 = 0.39, P = 0.98), or race (χ 2 = 0.78, P = 0.94). The composition and number of respondents in each information treatment group is provided in the Appendix. Briefly, the organic positive, organic negative, neutral, conventional positive, and conventional negative information groups contained 144, 150, 140, 150, and 149 respondents, respectively.
Sequentially, respondents first completed the belief choice experiment. Second, respondents were randomly allocated to one of the information treatments. Third, respondents were asked to complete the belief choice experiment again, but this time they received the alternative belief choice block. For example, respondent i was assigned to Block 1 Belief choice set, then received organic positive information, then was asked to complete Block 2 Belief choice set. Lastly, respondents completed the preference choice sets. Figure 3 presents a flow chart of the survey.
Figure 3. Flow chart of survey for one respondent.
2.3. Belief models
The respondent belief coefficients were obtained from the first choice experiment (quality sorting task). According to the conceptual model, this choice experiment provided respondent i’s perceived quality in quality dimension q of alternative j in choice set t (
${\rm Q _{ijt}^{\rm q}}$
). Respondents were asked to indicate which alternative they believe is superior in terms of a specific quality dimension. An “opt out” option that indicated “there is no difference in quality” was also included. For each quality dimension (environmental friendliness, animal welfare, food safety), respondent i’s perceived quality was a function of the good attributes, Xijt, and an alternative specific constant, ASCijt. The coefficients
${\boldsymbol \beta _{ki}^{\rm q}}$
, are respondent-specific beliefs that map the attributes of an alternative into perceived quality. To assess whether the belief coefficient for the organic status changed after receiving the information treatment, the pre and post belief choice experiments were grouped to estimate the model. We interacted the organic status with a dummy variable that indicates if the choice set was presented before or after the information treatment (Pt = 1if choice set was presented after information treatment, 0 if otherwise). The belief model becomes:
$$ {\rm Q}_{{\rm ijt}}^{{\rm q}}=({\bf X}_{{\bf ijt}}^{{\bf \prime}}{\bf \beta} _{{\bf i}}^{{\bf q}})+{\bf e}_{{\rm ij}}^{{\rm q}}={\rm \beta} _{1{\rm i}}^{{\rm q}}{\rm Org}_{{\rm ijt}}+{\rm \beta} _{2{\rm i}}^{{\rm q}}{\rm PL}80_{{\rm ijt}}+{\rm \beta} _{3{\rm i}}^{{\rm q}}{\rm PL}90_{{\rm ijt}}+{\rm \beta} _{4{\rm i}}^{{\rm q}}{\rm ASC}_{{\rm ijt}}+{\rm \delta} _{{\rm i}}^{{\rm q}}{\rm Org}_{{\rm ijt}}\times {\rm P}_{{\rm t}}+{\rm e}_{{\rm ij}}^{{\rm q}} $$
The equation has quality as the dependent variable, not utility, because respondents were asked to indicate which ground beef option is superior in quality dimension q, not which ground beef option they would buy. Hence, the alternative’s attributes and the respondent’s beliefs about how attributes inform quality will dictate which option they think is superior in terms of quality. In order to allow for heterogeneity of beliefs, the model was estimated using a mixed logit model simulated from 500 Halton draws and applied to each of the three quality dimensions. The results from the environmental friendliness belief model is presented in Table 2, and results from the animal welfare and food safety belief models are presented in Table 3 and 4, respectively. The coefficient,
${\rm \beta _{1i}^{\rm q}}$
, represents the mean of the belief distribution of how the Organic status influences perceived quality within quality dimension q prior to receiving information. The coefficient,
${\rm \delta _{i}^{\rm q}}$
, represents the average change in beliefs about the Organic status after receiving the information treatment. The sum of the two coefficients,
${\rm \beta _{\rm 1i}^{q}}+ {\rm \delta _{i}^{\rm q}}$
represent the mean of the updated beliefs after receiving the information treatment within each quality dimension.
Table 2. Belief mixed logit regression results: environmental friendliness
Notes: Mixed logit using 500 Halton draws. ASC = alternative specific constant. *** indicates significance at the 1% level, ** indicates significance at the 5% level, * indicates significance at the 10% level. Values in parentheses indicate the standard errors.
Table 3. Belief mixed logit regression results: animal welfare
Notes: Mixed logit using 500 Halton draws. ASC = alternative specific constant. *** indicates significance at the 1% level, ** indicates significance at the 5% level, * indicates significance at the 10% level. Values in parentheses indicate the standard errors.
Table 4. Belief mixed logit regression results: food safety
Notes: Mixed logit using 500 Halton draws. ASC = alternative specific constant. *** indicates significance at the 1% level, ** indicates significance at the 5% level, * indicates significance at the 10% level. Values in parentheses indicate the standard errors.
2.4. Preference models
As indicated in the conceptual model, Qijt is not directly observable. Hence, we used the respondent-specific organic belief coefficients obtained from the mixed logit belief models to predict an individual’s perceived quality of the alternatives in the preference choice experiment (quality sorting task)
$(\widehat{{\rm Q}}_{{\rm ijt}}^{{\rm q}}={\rm X}_{{\rm ijt}}{\rm ^\prime}{\rm \beta} _{{\rm ij}}^{{\rm q}}$
). Respondent-specific organic belief coefficients were obtained from the mixed logit regression outputs from the three belief models (environmental friendliness, animal welfare, and food safety). These coefficients were then multiplied by the presence or absence of the organic status of alternatives in the preference choice experiment. The predicted quality scores were then used as regressors in the preference model. To assess if the marginal utility of the perceived quality changed after receiving information, the perceived quality score was constructed using the organic coefficients both before and after receiving the information treatment. While constructing perceived quality scores for each quality dimension, it became clear that the organic status uniformly influenced perceived quality across all dimensions. Hence, including an explanatory variable for each perceived quality dimension would have resulted in multicollinearity. To overcome this problem, we used the approach of Costanigro and Onozaka (Reference Costanigro and Onozaka2020), and constructed a composite quality score, which was simply the average of the three quality dimension scores.
To assess whether the marginal utility of the composite quality score changed depending on if it was constructed using baseline belief coefficients or post information belief coefficients, we interacted the composite quality score with a dummy variable that indicated if the score was constructed using baseline or post information beliefs (P t = 1if constructed using post-information beliefs, 0 if otherwise). The preference model can be defined as:
$$ {\rm U}_{{\rm ijt}}={\rm U}({\bf Q}_{{\bf ijt}};{\rm Price};{\rm \gamma} )+{\rm v}_{{\rm ijt}}\\={\rm \gamma} _{{\rm Q}}\widehat{{\rm Qual}}_{{\rm ijt}}+{\rm \gamma} _{2}{\rm PL}80_{{\rm ijt}}+{\rm \gamma} _{3}{\rm PL}90_{{\rm ijt}}+{\rm \gamma} _{{\rm Price}}{\rm Price}_{{\rm ijt}}+{\rm \gamma} _{4}{\rm ASC}_{{\rm ijt}}+{\rm \delta} _{{\rm Q}}\widehat{{\rm Qual}}_{{\rm ijt}}\times {\rm P}_{{\rm t}}+{\rm v}_{{\rm ijt}} $$
With γk being interpreted as the marginal utility of attribute k. The above preference model was estimated using a multinomial logit model. The results are presented in Table 5.
Table 5. Multinomial logit preference regression results
Notes: ASC = alternative specific constant. *** indicates significance at the 1% level, ** indicates significance at the 5% level, * indicates significance at the 10% level. Values in parentheses indicate the standard errors. Composite Quality = (Environment + Animal Welfare + Food Safety)/3.
2.5. Decomposition of willingness-to-pay
Perhaps the most useful quality of the Costanigro and Onozaka (Reference Costanigro and Onozaka2020) belief-preference model is that it allows for a part-worth decomposition of willingness-to-pay for an attribute via the Q-dimensional pathways of perceived qualities. This approach allows for a tangible assessment of how each quality dimension (credence attribute) contributes to the overall willingness-to-pay for a credence attribute-indicating label. In our case, it also provides a mechanism to measure how willingness-to-pay changes given the altering of beliefs and preferences.
We perform a decomposition of baseline willingness-to-pay for the organic status using the means of the distributions of organic belief coefficients obtained from the belief models prior to implementing the information treatment. In this context, total willingness-to-pay for the organic status prior to receiving information can be defined as:
$${\rm WTP}_{{\rm organic}\_ {\rm baseline}}={\{[1/3({\rm \beta} _{{\rm org}}^{{\rm env}}+{\rm \beta} _{{\rm org}}^{{\rm anwf}}+{\rm \beta} _{{\rm org}}^{{\rm fdsf}})]\times {\rm \gamma} _{{\rm Q}}\} \over -({\rm \gamma} _{{\rm Price}})}$$
The premiums that the organic status generates due to a specific quality dimension, q, is
$\text{1}/{3} \Big[ \big( \beta_{\text{org}}^{q} \times {\gamma_{Qual}} \big)/- (\gamma_{\text{price}}) \Big]$
. The above equation is analogous to equation (3), where the belief coefficients for the organic status, βorg
q, are obtained from each belief model for a given information treatment. They are then multiplied by the preference parameter for quality, γQ, and divided by the parameter for price, − (γPrice, both of which come from the single preference model for the information treatment. Again, our hypothesis is that after receiving the information treatment, beliefs were modified, leading to a shift in willingness-to-pay. The updated willingness-to-pay for the organic status can be estimated using the same model above, but now including the interaction terms which captures the shift in beliefs,
${\rm \delta _{org}^{\rm q}}$
, and preference, δQ, values:
$$ {\rm WTP}_{{\rm organic}\_ {\rm updated}}={\left\{\left[1/3\left(\left\{{\rm \beta} _{{\rm org}}^{{\rm env}}+{\rm \delta} _{{\rm org}}^{{\rm env}}\right\}+\left\{{\rm \beta} _{{\rm org}}^{{\rm anwf}}+{\rm \delta} _{{\rm org}}^{{\rm anwf}}\right\}+ \left\{{\rm \beta} _{{\rm org}}^{{\rm fdsf}}+{\rm \delta} _{{\rm org}}^{{\rm fdsf}} \right\}\right)\right]\times \left({\rm \gamma} _{{\rm Q}}+{\rm \delta} _{{\rm Q}}\right) \right\} \over -\left({\rm \gamma} _{{\rm Price}}\right)} $$
Where − (γPrice) is obtained from the single preference model for the given information treatment.
3. Results
3.1. Belief model results
Prior to the information treatments, organic status was associated with increased perceived environmental friendliness, animal welfare, and food safety, as indicated by the positive and statistically significant coefficient for the organic status. However, the standard deviation of the organic coefficients suggests that this association significantly varied across respondents. The impact of the information treatments differed depending on the type of information presented. Across all quality dimensions (environmental friendliness, animal welfare, and food safety), the organic positive information treatment did not alter how respondents infer quality from the organic status by a statistically significant margin (P > 0.1). For example, in the environmental friendliness – organic positive treatment model, the Organic × Post coefficient is 0.109 (p > 0.05). This means that the organic positive information treatment did not statistically alter respondents’ beliefs about how the organic attribute relates to the environmental friendliness quality dimension. The same was true for the neutral information treatment, which suggests that beliefs are reasonably stable when presented with impartial information. The largest shift in beliefs regarding organic status occurred under the negative organic information treatment. Across all quality dimensions, the mean of the Organicijt × Pt coefficient is statistically significant and negative, indicating that after the information treatment, the association between the organic status and perceived quality decreased within each quality dimension (P < 0.01). For example, in the environmental friendliness – organic negative treatment model, the Organic × Post coefficient is negative (−1.028) and statistically significant. This represents a decrease in the belief that the organic status is associated with superior environmental friendliness. In fact, the quality inferred from the organic status in terms of environmental friendliness was inverted post information treatment. This suggests that respondents perceived non-organic ground beef to be superior in terms of environmental friendliness post information treatment. The conventional positive information treatment decreased the association between the organic status and quality within each of the quality dimensions, but the significance of the change differed depending on the quality dimension. For example, the most significant shift was seen in the environmental friendliness quality dimension (P < 0.05), with inferred animal welfare quality shifting minorly (P < 0.1), and food safety not shifting at all (P > 0.1). Lastly, the conventional negative information treatment increased the association between the organic status and perceived quality within the environmental friendliness (P < 0.1) and animal welfare (P < 0.01) quality dimensions, but not the food safety dimension (P > 0.1).
3.2. Preference models results
Across all preference models, the composite quality score was statistically significant and positive, indicating that composite quality generates positive marginal utility. Similarly, the coefficients for percent lean of 80 – 89% and >90% are positive and significant, indicating that respondents prefer ground beef with these lean ratios relative to 70 – 79% lean ground beef. The price coefficient is negative and statistically significant, demonstrating the marginal disutility of price, which conforms to a priori expectations. Within four of the five information treatments, the
$\widehat{{\rm Qual}}_{{\rm ijt}}\times {\rm P}_{{\rm t}}$
(recall that Pt is a dummy variable for post-treatment) is not statistically significant. This suggests the marginal utility of perceived quality did not change as a result of the information treatments. In other words, the way that respondents made tradeoffs between perceived quality and price did not change given the information. One exception was found under the organic negative information treatment, where
$\widehat{{\rm Qual}}_{{\rm ijt}}\times {\rm P}_{{\rm t}}$
is positive and statistically significant. We can interpret this as the marginal utility of perceived quality increasing after respondents receive the negative organic information treatment. This could be due to an introspective process, where respondents give greater consideration to their purchasing choices, thereby making quality a more valuable attribute.
3.3. Decomposition of willingness-to-pay results
Figure 4 depicts the decomposition of willingness-to-pay for the organic status estimated using baseline and updated beliefs and preferences. The total willingness to pay for the organic status calculated using baseline and post information beliefs and preferences is listed at the top of each bar. The decomposed willingness-to-pay by quality dimension is listed within the shaded regions within the bar. For example, the baseline total willingness-to-pay for the organic status for the organic positive information treatment was calculated as:
$\left[\left(1.410+1.187+1.365\right)\times 0.333\right]\times {0.439 \over -\left(-0.729\right)}=\$ 0.7945 \sim \$ 0.80.$
We can then decompose their willingness-to-pay for each quality dimension. For example, willingness-to-pay for the environmental friendliness quality dimension at baseline for the organic positive information treatment was calculated as:
${\frac{{1.410\times 0.439}{-\left(-0.729\right)}} {{3}}}=\$ 0.28.$
The willingness-to-pay after the information treatments was calculated using the same approach, except now using the updated belief (
${\rm \beta _{\rm org}^{q}}+ {\rm \delta _{org}^{\rm q}}$
) and preference (γQ + δQ parameters. First, the willingness-to-pay for the organic status under the neutral information treatment is around 89 cents. More importantly, the neutral information treatment shows that, even with no statistically significant differences in beliefs or preferences, there is a 10.48 cent, or 12%, variation in total willingness-to-pay due to random error. Similarly, though the organic positive information treatment did not alter beliefs or preferences by a statistically significant margin, it generated a 12.58 cent, or 16%, increase in total willingness-to-pay for the organic status.
Figure 4. Decomposition of willingness-to-pay for the organic status by quality dimension using baseline and updated beliefs and preferences.
Notes: Summation of WTP for individual quality dimension may not exactly equal total WTP for the organic label due to rounding.
The largest shifts occurred in both the organic negative and conventional negative information treatments. The organic negative information treatment reduced total willingness-to-pay for the organic status by 21.53 cents, or 70.50%. Within this information treatment, the environmental friendliness quality dimension experienced the largest shift. Prior to receiving any information, the organic status generated an 8.94 cent premium due to perceived environmental friendliness. After receiving the negative organic information, the organic status generated a 1.41 cent penalty due to perceived environmental friendliness. In this instance, beliefs about the environmental friendliness of organic versus conventional ground beef flipped, showing that consumers are now willing to pay a 1.41 cent premium for conventionally raised ground beef due to the perception that it is more environmentally friendly. Similar results are found in the conventional negative information treatment. The conventional negative information treatment increased total willingness-to-pay for organic ground beef by 51.56 cents, or 38.96%. Within this information treatment, the largest shift was seen in the animal welfare quality dimension. Prior to receiving information, the organic status generated a 39.21 cent premium due to perceived animal welfare. After receiving the conventional negative information, the organic status generated a 61.98 cent premium due to perceived animal welfare. The altering of beliefs and preferences in the conventional negative information group resulted in a 23 cent, or 58.07%, increase in the premium for organic due to increased perceived animal welfare. Lastly, conventional positive information treatment reduced total willingness-to-pay for the organic status by 35.66 cents, with the largest shift occurring within the environmental friendliness quality dimension, which experienced a 48.23% reduction in willingness-to-pay.
4. Discussion
Consumer beliefs are a critical component of the food valuation process as they serve to map credence attributes into subjective quality judgments of a good. The results of this study suggest that consumer beliefs about organic and conventional ground beef were malleable when presented with new information. The outcome of this belief updating process was an observed change in consumer willingness-to-pay for organic ground beef, attributed to revised associations between the organic status and perceived quality dimensions. This work specifically addresses a previously underexplored area of consumer choice by disentangling beliefs from preferences and exploring the malleability of beliefs and how they impact willingness-to-pay. Furthermore, our work was able to quantitatively assess belief malleability and changes in willingness-to-pay for specific quality dimensions associated with an attribute. We believe this advances our understanding of the belief-preference dynamic and malleability of this relationship in a hypothetical setting. Exploring this interaction for actual food purchase scenarios is recommended for future work.
The most important contribution this study makes to the literature is that it demonstrates that information modified the way in which respondents mapped credence attributes into perceived quality, one of the first studies to provide quantitative evidence that beliefs are malleable. This supports the general existing consensus of economists that beliefs are a flexible construct that can be modified through marketing and advertising activities (Lusk et al., Reference Lusk, Schroeder and Tonsor2014). Presenting information treatments in a narrative format could have limited the confirmation bias that has been identified in other studies that employed alternative methods of presenting information and their impacts on beliefs (Neuhofer et al., Reference Neuhofer, Lusk and Villas-Boas2023). The framing of the information also impacted the magnitude of belief adjustments, with negative information being associated with more significant changes. To demonstrate, the organic negative and conventional negative information treatments elicited a significant modification of beliefs in five out of six models, whereas the organic positive and conventional positive elicited a significant modification in only two out of six models. The greater influence of negative information could be attributed to prospect theory, where individuals weigh losses, or negative news, more than gains, or good news (Mizerski, Reference Mizerski1982; Tversky and Kahneman, Reference Tversky and Kahneman1981), or negativity bias (Kahneman et al., Reference Kahneman, Knetsch and Thaler1991; Mizerski, Reference Mizerski1982). The preexisting belief that organic was superior within all quality dimensions could have also explained why the organic positive information treatment did not elicit any significant change in beliefs.
Though beliefs were malleable, preferences were overall stable. This supports previous assumptions that assert preferences, or the way buyers tradeoff between quality dimensions and price, is relatively fixed (Lusk et al., Reference Lusk, Schroeder and Tonsor2014). In other words, the way respondents valued quality did not change given new information. The one exception was found in the organic negative information treatment, where the marginal utility of quality increased post information. This finding could be attributed to a modification in the quality valuation process, as it has been shown that preferences are, at least, partially related to perceived quality (Steenkamp 1986). In any case, this reinforces the fact that beliefs are worth studying in their own right, and that the disentanglement of beliefs and preferences is needed to truly understand how information affects food choice (Costanigro et al., Reference Costanigro, Deselnicu and Kroll2015; Lusk et al., Reference Lusk, Schroeder and Tonsor2014; Malone and Lusk, Reference Malone and Lusk2018).
We demonstrate that, due to the modification of beliefs, willingness-to-pay for organic ground beef is altered. To the authors’ knowledge, this is the first work to demonstrate an alteration in willingness-to-pay due to the modification of the quality valuation process under the Costanigro and Onozaka (Reference Costanigro and Onozaka2020) belief-preference model. We were able to decompose the change in willingness-to-pay of the organic status into changes in willingness-to-pay for specific quality dimensions. This provides evidence of how consumers economically value specific qualities associated with the organic status, as opposed to the organic status itself. The largest changes in willingness-to-pay occurred in the negative information treatments, where willingness-to-pay decreased by 70.50% and increased by 38.96% for organic negative and conventional negative, respectively. To demonstrate, after receiving the negative organic information treatment, respondents no longer perceived organic ground beef to be more environmentally friendly than conventional ground beef. Consequently, the 8.94 cent premium that the organic status garnered due to perceived environmental friendliness switched to a 1.41 cent penalty. This result has important implications for stakeholders interested in food marketing that combats label misinterpretation by consumers and overcomes cognitive biases such as the halo effect for credence attributes. Additionally, this work informs stakeholders as to which beliefs about given quality dimensions are malleable, and the relative economic importance of such quality dimensions to the consumer.
Inferential processing, the method by which consumers’ subjective beliefs are used to construct perceived quality from a credence attribute, lends itself to label misinterpretation (Apaolaza et al., Reference Apaolaza, Hartmann, Echebarria and Barrutia2017; Schuldt and Schwarz, Reference Schuldt and Schwarz2010), food stigmatization (Kanter et al., Reference Kanter, Messer and Kaiser2009), and potential reductions in agricultural productivity and safety (Caputo, Reference Caputo2020; Fox, Reference Fox2002; Messer et al., Reference Messer, Costanigro and Kaiser2017). We have shown, at least in a hypothetical setting, that information can modify beliefs about the organic status, thereby unraveling the halo effect and lessening the stigma of non-organic foods. Understanding what type of information and within which quality dimension beliefs are malleable can aid in crafting more effective marketing strategies and index-labels for consumers. Consequently, consumers can be more confident that their food choices align with their personal priorities for the food system.
Our analysis is not without limitations. Most obvious, belief modification was modeled in a hypothetical scenario. Therefore, an application that assesses belief malleability in a non-hypothetical scenario is needed to assess if it truly leads to changes in purchasing behavior. Second, it is possible that respondents misinterpreted or ignored the information treatments. In a real market context, a consumer could be exposed to both positive and negative information, and such a case was not considered in this study. Furthermore, it is possible that other important attributes of ground beef were not accounted for in the preference model. Lastly, having respondents complete the questionnaire before the choice sets could have introduced a pro organic bias. Future work should seek to assess if beliefs are malleable, and if belief modification leads to changes in purchasing behavior in a non-hypothetical setting. Additional work could also explore various avenues of providing information and its effect on belief modification, and belief malleability when exposed to both positive and negative information simultaneously.
5. Conclusion
We find information, particularly negative information, modified the way in which respondents mapped credence attributes into perceived quality, providing quantitative evidence that beliefs are malleable. The outcome of this belief updating processing was an observed change in consumer willingness-to-pay for organic ground beef, attributed to revised associations between the organic status and perceived quality dimensions. This work advances our understanding of the belief-preference dynamic and offers insights for developing more impactful food marketing strategies. Future work exploring if belief modification results in alterations of food purchasing behavior in a non-hypothetical setting is recommended.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/aae.2025.10030.
Data availability statement
Data is available upon request.
Acknowledgements
We appreciate the feedback of Dr Darren Hudson (Texas Tech University), Dr Jeff T. Larsen (University of Tennessee), and the participants of the 2023 American Agricultural Economics Association in Washington, D.C.
Author contribution
Conceptualization: B.A.Y., A.B.C., R.B.W; Data Curation: B.A.Y., A.B.C., ; Formal Analysis: B.A.Y., ; Funding Acquisition: R.B.W; Investigation: B.A.Y., R.B.W; Methodology: B.A.Y., A.B.C., ; Project Administration: ; Resources: ; Software: ; Supervision: ; Validation: ; Visualization: B.A.Y., ; Writing – original draft: B.A.Y., ; Writing – review & editing: A.B.C, R.B.W.
Financial support
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Competing interests
The authors declare no competing interests.
AI contribution to research
AI was not used in the creation of this manuscript.
Open access
