2.1. Data collection

For the sensory analysis and preference elicitation, we relied on a sampling of participants recruited from the Utah State university sensory lab during the summer of 2023. Participants included faculty and staff, undergraduate and graduate students, and members of the general public who responded to our open call for participation (emails were sent and flyers were posted around the campus). The study received ethical approval from the university’s Institutional Review Board (IRB number: 13656). Participants in the sensory tasting session provided their written informed consent prior to the study and were assured of the confidentiality and anonymity of their responses. The original collected sample size was 200. However, two participants were excluded from analyses due to response and recording errors, resulting in a usable sample size of 198. Summary statistics for the sample are included in Table 1. The sample was nearly equally represented by men and women and had good variability within education, income, and steak consumption frequency. Sample limitations included limited variability within ethnicity and low sampling of middle-aged individuals (45–64 years). The sample size and variability within the other demographics are similar to other beef consumer preference research (O’Sullivan et al., Reference O’Sullivan, O’Neill, Conroy, Judge, Crofton and Berry2021; Sitz et al., Reference Sitz, Calkins, Feuz, Umberger and Eskridge2006; Umberger et al., Reference Umberger, Feuz, Calkins and Killinger-Mann.2002; Umberger, Boxall, and Lacy, Reference Umberger, Boxall and Lacy2009). Despite the noted limitations in the sample, this analysis provides a strong starting point for future research with similar objectives to build upon and can give producers empirical evidence to consider when evaluating the adoption decision surrounding hydroponically produced barley fodder as a feedstuff.

Table 1. Sample demographics (n = 198)

^a How likely to consider the sustainability of production when making food purchasing decisions.

^b Other genders were available for participant selection but only those with responses have been included in this table.

^c Indicates the number of participants who were either informed or uninformed within the questionnaire about the feed/production practices of the samples.

2.2. Sample collection and preparation

Two types of beef samples were prepared: a control sample (CON) sourced from cattle fed a conventional mixed ration and a barley-fed (B-F) sample sourced from cattle fed hydroponically produced barley fodder as part of a total mixed ration. All beef used in this study was produced from cattle raised in the same herd at the Utah State University Agricultural Experiment Station. The cattle are typical of a commercial Angus-influenced herd in the region. The rations fed the B-F cattle were altered to include hydroponically produced barley fodder while still being nutritionally balanced similar to the conventional mixed ration. A summary of the rations fed to the CON and B-F cattle is contained in Appendix A (Table A1). All animals were fed to an industry standard average of 10 mm of backfat as measured by ultrasound (ExaGoFootnote ¹ ) and harvested at a commercial processing facility. Boneless strip loins were collected from one side of each carcass 48 hours post-harvest and aged for 12 days at 39.2°F. Each loin was subsequently fabricated into 1-in. thick steaks and two were randomly selected for sensory analysis. These steaks were vacuum packaged and stored in a refrigerator for four additional days before the sensory panel took place. Steaks from each treatment were cooked on separate electric clamshell-type grills (Blue Diamond Electric Sizzle Griddle, Model # CC002899-002, Hong Kong, China) to a consistent 159.8°F internal temperature measured by digital thermocouple thermometers (AMSA, 1995). Ten samples, approximately 1 inch square in size, were cut from each cooked steak and served to the participants. Separate individuals were randomly assigned to each treatment to cook and cut the steaks from within their assigned treatment to ensure no cross-contamination between treatments. The steaks were cooked directly prior to serving to participants and kept warm when fluctuations in the participant queue necessitated lag times between cooking and serving via heat lamps and heated bricks. No sample was under heat lamps for more than 10 minutes.

All participants received a tray containing both CON and B-F samples placed in small aluminum cups, covered with aluminum foil, in a randomized and balanced order identified only by three-digit random numbers as in Figure 1. Both tenderness and marbling levels were controlled for within the experimental design. This was accomplished by creating pairings (n = 11) of CON and B-F animals that were known to have marbling and tenderness levels within a similar range. The marbling score, provided by the harvest facility, was used in conjunction with Warner–Bratzler shear force (WBSF) measurementsFootnote ² taken three days prior to the sensory analysis to create pairings wherein marbling scores varied within a range of less than 24 units, while WBSF varied less than 0.08 kg. Table 2 contains a summary of the steak pairings and describes each pair according to WBSF and marbling score. Controlling for tenderness and marbling variability helped ensure participant ratings of sensory attributes were independent of the tenderness and marbling variability across samples. Within the study, 11 B-F animals were used and paired with 11 CON animals with similar tenderness and marbling levels. The target sample size for participants was n = 200. The experimental design called for 10 usable samples to be cut from two steaks from each treatment within a pairing. This resulted in 20 steak samples from each treatment within a pair. Thus, the 11th pairing was only prepared as a backup measure to be used if needed.

Figure 1. Tray with samples prepared for sensory analysis and subsequent questionnaire provided to participants.

Table 2. Sample parings tenderness and marbling scores

^a Pairings are such that the average difference in Tenderness between CON and B-F treatments is 0.078 with a standard deviation of 0.070 and the average difference in marbling is 23.18 with a standard deviation of 14.14.

^b Tenderness as measured using the Warner–Bratzler shear force method (kg).

^c Marbling score as reported by the commercial harvest facility providing an objective measure of the marbling for each animal.

2.3. Sensory analysis and questionnaire

After receiving the prepared tray (Figure 1), participants were instructed to remove the foil from the first sample and rate it according to “appearance” and “aroma” using a 7-point hedonic scale ranging from “1-dislike very much” to “7-like very much”. The participants were then instructed to eat the sample and continue rating it according to “overall acceptance,” “flavor,” “tenderness,” and “juiciness.” After completing the ratings for the first sample, the participants were asked to cleanse their palettes using the provided unsalted crackers and water. The rating exercise was then repeated for the remaining sample.

Following the sensory rating exercise, participants were randomly assignedFootnote ³ to one of two treatment groups – “informed” (n = 105) and “uninformed” (n = 93) before proceeding to complete the questionnaire. The “uninformed” group remained completely “blind” throughout the experiment as to sample characteristics and experimental design. The “informed” group, however, was informed directly after the sensory rating exercise about the different characteristics of the feed used to produce the steak samples. The educational script/nudge provided to the participants within the “informed” group not only revealed the difference in feeds used to produce the beef but also educated participants on some of the potential sustainability characteristics of the hydroponically produced barley fodder. The educational script seen by the “informed” group is contained in Figure 2.Footnote ⁴

Figure 2. Educational script provided to “Informed” participant group.

All participants were then asked to select which sample they would be most likely to purchase in a retail environment based on their sensory experience (and knowledge of the production system used to produce the samples if in the “informed” group).

To assess WTP for this study, we relied upon an open-ended contingent valuation question. Open-ended contingent valuation (CV) is a survey-based method used to estimate the economic value of goods and services. It involves asking respondents an open-ended question about the maximum amount they are willing to pay (WTP) for a particular good or service. Single-bounded and double-bounded contingent valuation are variations of this method that impose certain constraints (values) for the good and ask for a dichotomous response of WTP (Johnson, Bregenzer, and Shelby, Reference Johnson, Bregenzer and Shelby2019). Each approach has comparative advantages and disadvantages, suggesting that the choice between them ultimately depends on the objectives of the research and context of use.

Noted advantages of open-ended CV include simplicity, ease of administration, and potential reduced anchoring bias compared to bounded CV (Green et al., Reference Green, Jacowitz, Kahneman and McFadden1998). Open-ended CV surveys are often simpler and quicker to administer compared to their bounded counterparts. Respondents are asked a single, open-ended question, making the survey less complex. Open-ended CV may have reduced anchoring bias compared to bounded CV (Green et al., Reference Green, Jacowitz, Kahneman and McFadden1998). Anchoring bias occurs when respondent answers are influenced by the initial values presented to them. In bounded CV, researchers provide starting points, which can anchor respondent valuations. Open-ended CV avoids this issue by not providing such starting points. However, some argue that the double-bounded format is more familiar for respondents since typically consumers are faced with a given price in a retail setting and not afforded the opportunity to state their maximum WTP as in the open-ended format (Loomis, Reference Loomis1990). Additionally, open-ended questions may encourage free-riding and hence strategic overstatement (Arrow et al., Reference Arrow, Solow, Portney, Leamer, Radner and Schuman1993). In the context of contingent valuation, free-riding refers to a situation where respondents in a survey or study deliberately misstate their preferences or WTP for a particular environmental or public good. This typically occurs when individuals underestimate their true WTP for a good or service, often in an attempt to avoid financial responsibility or to reduce the perceived cost of the good or service being evaluated. This may in part explain why several studies have shown that the open-ended format yields lower estimates of WTP on average than double-bounded CV (Brown et al., Reference Brown, Champ, Bishop and McCollum1996). However, steak is a market good for which consumers are relatively familiar. Within our sample, responses indicate (Table 1) that 76.8% of participants consume steak at least once a month. This suggests that they should be generally familiar with the price of steak. Kealy and Turner (Reference Kealy and Turner1993) find no differences in WTP between open-ended and double-bounded methods in a familiar private market good (candy bar) suggesting the free-riding problem is mitigated when evaluating familiar market goods. Thus, as we are evaluating a good for which consumers are relatively familiar and seek to reduce anchoring bias, we chose to rely on an open-ended contingent valuation question within this study.

After selecting their preferred steak, participants were asked, “Assuming the average price of ribeye steak of this quality is $9.65 per poundFootnote ⁵ , how much MORE would you be willing to pay for your preferred steak relative to your least preferred steak?” The question relied upon a sliding scale response format that allowed participants to indicate any amount from $0.00/lb up to $3.50/lb in $0.10 increments.

Following the WTP question, we asked participants “how likely are you to consider the sustainability of production when you are making food purchasing decisions?” Participant responses were recorded using a 5-point hedonic scale ranging from “very likely” to “very unlikely.” This question was then followed up with standard demographic questions collecting information about participants’ age, gender, ethnicity, income, and education levels (demographics summary in Table 1).

2.4. Methodology

Attribute rating differences across treatments were evaluated using fixed effect ordinary least squares (OLS) regression modelsFootnote ⁶ for each attribute as in:

(1) $${\rm Rate}_{a,i,t}=\theta _{0,a}+\beta _{1,a}{\rm BF}_{a,i,t}+\beta _{2,a}{\rm Tenderness}_{i,t}+\beta _{3,a}{\rm Marbling}_{i,t}+e_{a,i,t}$$

where ${\rm Rate}_{a,i,t}$ is the 7-point hedonic rating for attribute a (overall acceptance, appearance, aroma, flavor, tenderness, and juiciness), individual i, and treatment t, ${\rm BF}$ is a fixed treatment dummy variable equal to 1 if the sample was a B-F treatment and equal to 0 otherwise, ${\rm Tenderness}$ is the WBSF value (kg), ${\rm Marbling}$ is the marbling score as provided by the harvest facility, and $e_{ait}$ is a random error term $e_{a,i,t}\sim N(0,\sigma _{e}^{2})$ . The primary focus for the sensory analysis rests on the evaluation of the sign and significance of the $\beta _{1}$ coefficients for each attribute. Though marbling and tenderness have been controlled for within the experimental design, we still include them within the equation to account for small variability remaining within sample pairings.

To model WTP premiums, we rely on a linear regression model as in:

(2) $$\begin{gathered}{\text{WT}}{{\text{P}}_i} = {\theta _0} + {\beta _1}{\text{B}}{{\text{F}}_i} + {\beta _2}tend\_dif{f_i} + {\beta _3}marb\_dif{f_i} + {\beta _4}highsu{s_i} + {\beta _5}informe{d_i} \\ \,\,\,\,\,\,\,\, + {\beta _6}highsus\_{\text{x}}\_{\text{informe}}{{\text{d}}_i} + {\beta _7}highsus\_{\text{x}}\_{\text{B}}{{\text{F}}_i} + {\beta _8}{\text{informed}}\_{\text{x}}\_{\text{B}}{{\text{F}}_i} \\ \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!+ {\beta _9}highsus\_{\text{x}}\_{\text{informed}}\_{\text{x}}\_{\text{B}}{{\text{F}}_i} + {e_i} \\ \end{gathered}$$

where ${\rm WTP}_{i}$ is the willingness-to-pay premium ($/lb) for $i$ th participant’s preferred steak, ${\rm BF}$ is a dummy variable equal to 1 if the participant’s preferred steak is the B-F sample and equal to 0 otherwise, ${\rm tend_diff}$ is the difference between the WBSF values (kg.) of the two samples (B-F tenderness relative to CON), $ lt; gt;\rm marb\_\ diff lt;/ gt;$ is the difference between the marbling score values of the two samples (B-F marbling relative to CON), ${\rm highsus}_{i}$ is a dummy variable equal to 1 if the participant self-identified as either “likely” or “very likely” to consider the sustainability of production when making food purchasing decisions and equal to 0 otherwise, ${\rm informed}$ is a dummy variable equal to 1 if the respondent was part of the “informed” group and equal to 0 otherwise and $e_{i,c}$ is a random error term $e_{i,c}\sim N(0,\sigma _{e}^{2})$ .

2.5. Evaluating preferences

We also analyzed the participant’s preferred steak choice selection to determine what attributes have the most impact on their selection. By combining participant sensory ratings with their choice selection, we can evaluate which attributes have the strongest impact on consumer intent to purchase. We model participant selections using a logistic regression model. An indicator variable ‘ $I$ ’ is created to represent the participant’s preferred steak selection and is set equal to 1 when the participant selects the B-F sample as their preferred steak and set equal to 0 otherwise. A logistic regression model is then estimated as

(3) $${I_i} = {\theta _0} + \sum\limits_{a = 1}^6 {{\beta _a}} {{\boldsymbol{\mathit{X}}}_{\bf{i}}} + {\beta _7}{{highsu}}{{\rm{s}}_i} + {\beta _8}{{educate}}{{\rm{d}}_i} + {\beta _7}{{highsus}}\_{{x}}\_{{educate}}{{\rm{d}}_i} + {e_i}$$

where ${{\boldsymbol{\mathit{X}}}_{\boldsymbol{\mathit{i}}}}$ is a matrix of sensory attribute ratings differences between samples (e.g., flavor rating B-F less flavor rating CON) for the $i$ th participant, $e_{i,c}$ is a random error term, and all other variables are as previously defined in equation (2). Within the X _i matrix we exclude rating differences for the “overall acceptance” attribute. It is assumed that when participants consider their overall acceptance of a sample, they would jointly consider the other attributes of appearance, aroma, flavor, tenderness, and juiciness when making their ratings. The overall acceptance attribute would thus be correlated with the other attributes and is excluded from the preference model (equation 3) to avoid multicollinearity.