Sample display

Samples used in the study were beef strip loin steaks that graded USDA Choice with Modest and Moderate degrees of marbling (USDA, 1997). Samples were acquired from a commercial beef processor as case-ready packages in mother bags (Tri Gas, 69.6% N, 30% CO₂, 0.4% CO) and transported to the KSU Meat Laboratory under refrigerated conditions (2–4°C). Samples were delivered in mother bags of 6 packages. Samples were kept in their mother bags in the absence of light under refrigeration before simulated retail display.

During Phase 1, 5 of the 6 packages per mother-bag were assigned randomly to one of 5 retail case sections. Each of the 3 coffin-cases was equally divided into 3 sections, separated by traditional retail meat case dividers. A random package in each mother bag was assigned to pH and proximate analysis. Three mother bags were opened every other day, and packages were added to the cases to allow for variation in sample discoloration. Because the steak packages contained 2 steaks each, one steak per package was covered with a piece of matte-black tape so only one steak was visible. A single 3.8-cm-wide piece of tape was placed longitudinally across the center of the steak for the entire length of the steak to eliminate visibility. If one of the steaks had the gluteus medius present, that steak was covered; otherwise, the steak for display was randomly selected. Pilot work had indicated that when packaged 2 in the same package, steaks would discolor at different rates. Thus, covering one steak was done to ensure that sensory panelists and instrumental data were all consistently collected from the same steak. On their designated display day, samples were placed in one of 3 coffin-style cases (model DMF8; Tyler Refrigeration, Niles, MI, USA) at 2°C to 4°C under continuous fluorescent lights (32 W Del-Warm White 3,000 K; Philips Lighting Company, Somerset, NJ, USA) averaging 2140 Lux. Packages were rotated within each case section daily.

During Phase 2, all 6 packages per mother-bag were allocated to display: 2 packages per mother-bag in one of 3 retail case sections. One steak per package was covered with a piece of matte-black tape, leaving only one steak visible, similar to Phase 1. In both phases of the study, there was no pricing or weight information available to the consumers on a package label. Day 0 packages were placed in their respective case sections in the morning, approximately 6 h before sensory and objective evaluations. Samples for proximate and pH were randomly selected from paired mother bags.

Objective color measurement

Objective L*, a*, b*, and spectral data were collected using a HunterLab MiniScan Spectrophotometer (Illuminant A, 10° observer, 2.54 cm aperture; HunterLab, Reston, VA, USA). Three readings were collected from the surface of the displayed steak in each package, being cognizant of avoiding the external fat. The readings were averaged, then used to calculate chroma, hue angle, and percentage metmyoglobin and oxymyoglobin using the formulas provided in the American Meat Science Association (AMSA) Guidelines for Meat Color Measurement (King et al., 2023).

Trained sensory panel evaluation

For both phases, panelists were trained according to the AMSA Guidelines for Meat Color Measurement (King et al., 2023). Panelists were screened for color blindness with the Farnsworth Munsell 100 Hue Color Vision Test (Munsell, Grand Rapids, MI). Trained sensory panelists (N = 12–20 for Phase 1 and N = 14–23 for Phase 2) were required to attend at least 3 training sessions to familiarize themselves with the scales used. Panelists number varied on each evaluation day due to the time availability of panelists to participate in the study. Panelists were shown multiple packages and provided with reference pictures of varying discoloration, redness, and fat color during the training. Redness was assessed using the pictorial references published by Lybarger et al. (2023), and the fat color and percentage of discoloration were anchored by the references shown in Figure 1 and Figure 2. On panel dates, all samples were evaluated on 100-point line scales for percentage discoloration (0%–100%), redness – 0 being extremely dark red and 100 being bright cherry red, and fat color – 0 being brownish-white and 100 being bright white. Sensory panelists completed their evaluation using an electronic ballot (Version 2417833; Qualtrics Software, Provo, UT, USA) on electronic tablets (Lenovo TB-8505F, Morrisville, NC, USA). Trained panelists evaluated all samples rated by the consumer sensory panelists on each day of the panels in the same manner as consumer panelists. Trained sensory data were averaged across panelists for each sample on each day evaluated.

Fig. 1.

Training references used for training panelists for fat color score.

Fig. 2.

Training references used for training panelists for the visual discoloration percentage of beef strip steaks.

Consumer sensory panel evaluation

For both phases, consumers were recruited from the Manhattan, Kansas, community and surrounding areas and monetarily compensated for their participation. In Phase 1, consumer panelists (N = 200) evaluated 24 samples varying in discoloration, with 3 samples representing each day of display: 0, 2, 4, 6, 8, 10, 12, and 14 d. In Phase 2, consumer panelists (N = 176) evaluated 20 samples, all from the same day of display, with similar amounts of discoloration. In both phases, groups of 8 consumers evaluated different sets of steaks, with multiple 8-person panels held each night based on the study design. The same samples evaluated by trained sensory panelists were presented to consumers under the same lighting conditions as trained sensory panels, in a random order selected by the survey software on the electronic tablets. For each sample, consumers were asked to designate an overall appearance score on a 100-point continuous line scale with 0 being extremely undesirable and 100 being extremely desirable. They were also asked if they would purchase the sample at retail (yes/no). If they selected yes, they would move on to the next sample; however, if they selected no, they would be shown an additional question asking if they would purchase the sample if it was discounted (yes/no). No pricing, weight, or additional information was provided to the consumers about the samples at the time of evaluation. Consumers completed the evaluations using electronic tablets (Lenovo TB-8505F, Morrisville, NC, USA) using the same survey software as the trained sensory panelists (Version 2417833; Qualtrics Software, Provo, UT, USA).

Proximate and pH analysis

Thirteen steaks from Phase 1 and 12 steaks from Phase 2 were selected for proximate and pH analysis. Samples were trimmed of external fat and individually ground using a small table-top grinder (STX International Turbo Force II, Lincoln, NE, USA). Samples were then packed into plastic petri dishes, ensuring no air bubbles were present, and analyzed using the FOSS FoodScan 2 (FOSS Analytical, Hillerød, Denmark). Proximate analysis (moisture, protein, and fat) was recorded. Following proximate analysis, 5 g of each sample, in duplicate, was weighed into 100 ml beakers for pH analysis. 50 ml of Milli Q water (MilliporeSigma, Burlington, MA, USA) was added to each beaker. Each sample was then mechanically homogenized (Homogenizer 850; Fisher Scientific International, Hampton, NH, USA), and the pH of each sample was measured with an InLab Science Pro-ISM pH probe attached to a Seven Compact pH meter (Meter Toledo; Columbus, OH, USA).

Statistical analysis

The statistical analyses were performed using the procedures of SAS (SAS Institute, Cary, NC, USA), with α set at 0.05. Logistic regression models were calculated for the probability of a sample being identified as “would purchase” for both full-priced and discounted responses by consumer sensory panelists using PROC LOGISTIC for each independent variable. The PROC REG program was utilized to determine the simple linear regressions for consumer overall appearance ratings. PROC CORR was used to calculate Pearson correlation coefficients for sensory and objective measures. Data from each phase were analyzed separately.

Demographics

The demographics of the consumers who participated in the study are presented in Table 1. The consumers who participated in the current study were split between males (53.1% and 57.4% in each phase, respectively) and females (46.9% and 42.6% in each phase, respectively). Additionally, consumers were split between those greater than 30 years old (30.1%; 45.5%) and those less than 30 (69.9%; 54.5%). The consumers were primarily Caucasian (83.2%; 85.2%) and educated, with over 80% in each phase having at least some college education. The majority of consumers (64.3%; 58.0%) consumed beef at least 4 times a week, and purchased beef 1 to 3 times a week (41.8%; 40.3%). Overall, the demographic profile of the consumers used in the current study is similar to previous works conducted in Manhattan, KS (Davis et al., 2021; Harr et al., 2022; O’Quinn et al., 2024). When compared to the United States population, the consumers in the current work were overall slightly younger, less educated, and had a lower income in Phase 1, but were more aligned with national demographics in Phase 2 (U.S. Census Bureau, 2023).

Logistic regression models

The average pH of steaks in this study was 5.63 (SD = 0.06). Proximate analysis identified the average moisture of samples as 69.21% (SD = 2.44), protein as 20.25% (SD = 1.03), and fat as 8.66% (SD = 3.14).

Table 2 summarizes the descriptive statistics for the dependent variables used in the study for the development of the predictive models. Overall, in both phases, a wide range of all variables was accounted for. This was required for the calculation of robust regression models for the prediction of consumer purchase intent (yes/no). In both phases of the study, fresh, bright, cherry red colored steaks and steaks that were close to or were completely discolored were evaluated by consumers.

The developed regression equations for Phase 1 (Table 3) and Phase 2 (Table 4) indicated that most of the objective measures evaluated were predictors (P < 0.05) of consumer purchase intent in both full-priced and discounted scenarios. However, in Phase 1, the developed equations accounted for only a moderate amount of the variation in consumer purchase intent (R² of 0.30 to 0.78) in fully-priced samples and less variation (R² of 0.27 to 0.72) in the discounted samples. Still, the models were able to correctly classify more than 73% of fully-priced and more than 66% of discounted samples as would/would not purchase by consumers. In Phase 1, a* value was among the best objective measurements evaluated with R² values of 0.64 and 0.56 for full-priced and discounted models, respectively (Figure 3). Calculated chroma was determined from spectral data and resulted in high R² values of 0.56 in full-priced models and 0.58 in discounted models. Also in Phase 1, trained sensory panel discoloration score was a noteworthy predictor, with R² values of 0.56 in the full-priced model, and 0.47 in the discounted model (Figure 4).

Fig. 3.

Probability of a consumer purchasing a beef strip loin steak based on a* value and pricing: Phase 1.

Fig. 4.

Probability of a consumer purchasing a beef strip loin steak based on trained sensory panel discoloration score and pricing—Phase 1: Sensory discoloration scores: 0 = no visible discoloration, 100 = complete discoloration.

For Phase 2 of the study, the models were not as predictive as the models in Phase 1. For products sold at full price, the models accounted for only 11–57% of the variation in consumer purchasing intent, and the models were only able to classify 61–73% of the samples correctly as would/would not purchase. Phase 2 presented similar results among the objective color measurements as observed in Phase 1. Values for a* continued to be one of the strongest measurements, with R² values of 0.25 and 0.28 for full-priced and discounted models, respectively (Figure 5). Calculated chroma models also resulted in moderate R² values of 0.25 and 0.28 for full-priced and discounted models, respectively. The trained sensory panel discoloration score models (Figure 6) had R² values of 0.20 in the full-priced models and 0.21 in the discounted models. Overall, the Phase 2 models accounted for less variation among the variables than the Phase 1 models.

Fig. 5.

Probability of a consumer purchasing a beef strip loin steak based on a* value and pricing: Phase 2.

Fig. 6.

Probability of a consumer purchasing a beef strip loin steak based on trained sensory panel discoloration score and pricing—Phase 2; Sensory discoloration scores: 0 = no visible discoloration, 100 = complete discoloration.

Tables 5 and 6 present the various thresholds for consumer purchase intent from the logistic regression models. No value was reported for calculated thresholds that fell outside of the range of independent variables included in the study. Of particular note are the thresholds related to a* value and trained sensory panel discoloration score. The trained sensory panel discoloration score was a visual measure of the amount of metmyoglobin or browning on the surface of the steaks. Both variables were among the best evaluated for both phases of the current study. In Phase 1, a* values of 25.3, 29.9, and 34.4 were associated with a 50, 75, and 90% chance of a consumer purchasing the product. When discounted, these values shifted to 20.3, 25.8, and 31.3 for the same likelihoods, indicating consumers were more willing to purchase steaks that were less red when the product was discounted. For Phase 2 of the study, a* values of 23.6 and 34.2 were associated with a 50% and 75% chance of consumer purchase, and shifted to 15.9 and 25.5 when discounted. For the trained sensory panel discoloration score, in Phase 1, only 12% discoloration was associated with a 50% chance of a consumer purchasing the product. If the product was discounted, consumers were more tolerant of discoloration, with 35.8% and 8.3% discoloration associated with a 50% and 75% chance of purchase. For Phase 2, fully-priced samples that had 22.7% discoloration had a 50% chance of purchase, which increased to 56.8% discoloration for discounted samples.

Pearson correlations

Table 7 presents the Pearson correlation coefficients among the objective measures used in the study. All of the variables were closely associated (P < 0.05) in both phases of the study. a* value, as expected, was highly associated with trained sensory panel redness scores (r = 0.90) and consumer overall appearance scores (r = 0.87) in Phase 1 of the study. In Phase 2, a*’s relationship with trained sensory panel scores was shown to be just as strong (r = 0.95), but was much weaker (r = 0.69) with consumer overall appearance scores, further highlighting the diminished relationship between redness and consumer willingness to purchase when all steaks were of the same display period in the retail case. Other measures that were closely associated with consumer overall liking in Phase 1 included b* (r = 0.85), chroma (r = 0.88), trained sensory panel redness score (r = 0.89), and both lean discoloration scores (r = −0.76) and fat color scores (r = 0.83). For Phase 2, each of these variables was also related to the consumer’s overall appearance score, but showed much weaker (r = 0.67, 0.69, 0.64, −0.61, 0.67, respectively) relationships than in Phase 1.

Linear regression equations

Table 8 presents linear regression equations for predicting consumer overall appearance rating scores using the various objective measures, and Figure 7 and Figure 8 show the regressions for using a* value and trained sensory panel discoloration score for predicting consumer overall appearance rating, respectively. All of the objective measures were able to predict consumer overall appearance liking rating (P < 0.05) and explained 32–80% of the variation in the consumer scores for Phase 1 and 22–48% of the variation in Phase 2. Linear regression equations predicting consumer overall appearance liking ratings with objective measures also resulted in significant (P < 0.01) models, with R² values of 0.32 to 0.80 in Phase 1 and R² values of 0.22 to 0.28 in Phase 2 (Table 8). This indicates these models were able to account for a moderate to large amount of variation within the consumer’s overall appearance scores.

Fig. 7.

Linear regressions for predicting consumer overall appearance rating based on a* value; Consumer overall appearance scores: 0 = extremely undesirable, 100 = extremely desirable.

Fig. 8.

Linear regressions for predicting consumer overall appearance rating based on trained sensory panel discoloration score; Sensory discoloration scores: 0 = no visible discoloration, 100 = complete discoloration.

Steaks versus ground beef

In both the current study and the Lybarger et al. (2023) study, a* was one of the best predictors for consumer purchase intent. For steaks, consumers required higher a* values for the same level of purchase likelihood compared to ground beef. For example, fully-priced steaks required a 25.3 and 29.9 a* value for a 50 and 75% likelihood of purchase, whereas for ground beef, only 21.6 and 24.9 were needed for the same level of purchase likelihood (Lybarger et al., 2023). The same was observed in discounted samples, with an a* of 31.3 associated with a 90% likelihood of purchase in the steak study, but only an a* value of 25.0 was needed for a 90% chance of purchase in ground beef. As for discoloration, in the Lybarger et al. (2023) ground beef study, consumers had a 50% chance of purchasing samples with close to 38% discoloration. However, in the current study, only 12% discoloration in steaks reduced the consumer purchase intent to 50%. When comparing the ground beef results to the current study’s results for steaks, it is obvious that consumers were more tolerant of discoloration and poor color characteristics in the ground beef than they were in the steak samples.

One of the largest contrasts between the results for the steaks and the similar ground beef study by Lybarger et al. (2023) was in the predictive power of the logistic regression models. Lybarger’s models were very robust and highly effective at identifying points at which ground beef failed to meet consumer expectations for purchase. The results of the current study did not produce results that were as clear. Undoubtedly, this is a result of the differences in the products evaluated in each study. As discussed previously, ground beef and steaks discolor throughout retail display in very different manners, with ground beef uniformly discoloring and steaks forming discolored spots which eventually expand to the whole cut surface (King et al., 2023). It is clear that this difference in discoloration pattern has a significant impact on how consumers assess these products. But an even larger likely reason for the difference is related to the price point of the meat products. Ground beef is the most inexpensive beef product available for consumers, whereas steaks are often viewed as a premium product and are sold at a price point that is reflective of such. Thus, consumers’ willingness to purchase steaks that had even minor “problems” was less than with ground beef. Consumers viewed the higher price and premium nature of steaks as requiring the products to more fully and completely meet their expectations, as opposed to being willing to overlook some of the same deficiencies within the cheaper ground beef products. Consumer expectations and psychology related to product purchasing decisions and the impact they have on consumer satisfaction have long been studied, with consumer expectations before purchase directly impacting overall satisfaction (Oliver and Winer, 1987; Grewal et al., 2019). Additionally, consumers are more discriminating in products with a perceived higher quality or price point (Bertini et al., 2012). This highlights the need for ensuring that steaks sold at retail are maintained in an ideal condition, with practically no discoloration, for consumers to purchase, as only small amounts of discoloration and poor redness (darkened) will prevent consumers from purchasing the product, even if sold at discounted prices.

When compared to the ground beef study conducted by Lybarger et al. (2023), the models in the current work were not as predictive. In their study, when samples representing a range of discoloration were evaluated by consumers, models for fully-priced samples accounted for greater than 77% of the variation in consumer willingness to purchase, with the some of the best models, such as the one using a* value as a predictor, accounting for over 80% (Lybarger et al., 2023). Additionally, their models almost all classified greater than 90% of samples correctly related to consumers who would or would not purchase and had higher C-statistic values, indicative of models with a better fit for the binomial outcomes. Similar to the current work, Lybarger’s models for their discounted models were less robust and predictive than their full-priced models, but still accounted for more than 77% of the variation in consumer purchase intent compared to the 40–60% of the variation accounted for by the same variables in the current work.

Additionally, the Lybarger et al. (2023) study, similar to the current work, had models that were less predictive when all samples in the case were of the same day of retail display. Their models accounted for about half as much variation in their Phase 2 versus Phase 1, but still accounted for approximately 40% of the variation in consumer purchase intent. In the current study, these models accounted for a much lower percentage (< 25% for most variables). Moreover, these models had a very low predictive power for both full-priced and discounted samples, as evidenced by the low amount of variation accounted for by the models (R²), the relatively low percentage of correctly classified samples, and the low C-statistic values. Both studies highlighted the same theme—when consumers evaluate samples that are more similar than different, they inherently use different criteria and become more selective than when there is abundant variation. Previous studies have shown that consumers better discriminate between product options when evaluating products within a crowded quality space with minimum differences—such as Phase 2—rather than a more varied or sparse space for quality, such as in Phase 1 (Bertini et al., 2012). Thus, the predictors that are the most predictive in Phase 1—redness and discoloration—become less predictive when the variation is minimal, thus weakening the predictive strength of each. This is noteworthy as when consumers purchase beef from the meat case at retail, it is almost always in a scenario where there is only minimal variation, as much of the displayed meat would have been in the case for a similar display period. Taken together, this indicates that predicting consumer purchase of retail meat products in a representative retail setting is likely more variable and more difficult to predict than when consumers are presented with a range of variation in color and discoloration traits simultaneously. Thus, Phase 1 models may have only a limited utility when applied to a “real-world” setting of a retail meat case.

Consumer perceptions of steaks

The current results also demonstrate the challenges with attempting to quantify consumer willingness to purchase steaks at retail. All of the samples used in the current study were from the same production lot, of the same quality grade (upper 2/3 Choice), and packaged identically from the same commercial supplier. Despite these attempts to standardize all samples, slight differences in marbling amount, external trim level, loineye size, packaging variation, etc., that were not quantified in the study design may have had an impact on consumer willingness to purchase samples. This is clearly shown in samples with 0% discoloration, of which in Phase 1 of the study, only had a 70% chance of being purchased at full price, and in Phase 2 of the study, had only a 64% chance of being purchased. These probabilities were increased to only 81% for Phase 1 and 80% for Phase 2 if products were discounted. This means that there were samples that, when displayed fresh, at 0 d of retail display, consumers simply identified that they would not purchase. It is unclear as to what the reasons were for these responses, as it was not captured clearly in any of the objective measures evaluated, but it is likely related to the aforementioned package-to-package variance. This further indicates consumers’ unwillingness to purchase steaks at retail that have any perceived “defect,” regardless of what is related to it. Additionally, these preferences and aversions are likely highly specific to each consumer, making them further difficult to identify and quantify, and as a result, weakening the predictive power of the calculated regression equations in the current study. Previous work has identified the high level of subjectivity and varied expectations related to individual consumer perceptions of food quality, which is varied due to previous personal experiences, cultural backgrounds, individual situational factors, and numerous other influences that are closely tied to the individual (Cardello, 1995). Furthermore, these preferences are likely increased when viewing numerous samples simultaneously in a retail display case, allowing consumers to identify and show preference to only the samples that are “most ideal” to them, while discriminating against those that are less ideal, as discussed by Bertini et al. (2012). Thus, many of the packages consumers identified as lower on the liking scale or as “would not purchase” may have been viewed and rated differently had consumers had fewer samples in the case to compare to and show preference towards.

Color and discoloration thresholds

It is challenging to directly compare the results of the current work with other previously published reports that have attempted to assess consumer willingness to purchase beef steaks based on color and discoloration due to numerous methodological differences. However, a* value has repeatedly been shown to be one of the best predictors of consumer purchase intent (Carpenter et al., 2001; Holman et al., 2017; Najar-Villarreal et al., 2021). One such study identified a lower threshold a* value of 20.99 needed for acceptance of longissimus dorsi steaks (Najar-Villarreal et al., 2021). When compared to the current study, this identified threshold would have been well below the a* value needed for a 50% likelihood of purchase in both phases, and thus would likely be much too low for a reasonable indicator of consumer purchase intent. However, these authors used trained sensory panel scores for assessment of acceptance, which differs significantly from the consumer model used in our study.

Similar limitations within other published works provide a lack of a clear and direct comparison to the results from the current study for retail beef steaks. Holman et al. (2016) required consumers to evaluate 10 images of beef longissimus lumborum steaks using an online survey to determine consumer purchase thresholds and beef color acceptability. Photographed steaks were accompanied by instrumental color data, and participants were provided with explicit instructions to standardize screen settings. These authors found b* was more predictive of consumer acceptability of beef color than a*, which is contradictory to the current study. However, in a follow-up study, Holman et al. (2017) increased the number of steaks to be evaluated and included 80 images and found that an a* value was a more reliable predictor of consumer acceptance, which contradicts their previous work. This contradiction is likely due to an increased sample number and resulted in more accurate and robust results. However, in both studies (Holman et al., 2016; Holman et al., 2017) illuminant D⁶⁵ was used, which does not allow for direct comparison to the current study, which used illuminant A as advised by the AMSA Guidelines for Meat Color Measurement (King et. al., 2023). Additionally, the inability to verify whether participants adhered to the prescribed screen settings introduced the potential for added variance within their data sets, potentially influencing their results.

A study conducted by Hood and Riordan (1973) was one of the earliest in-store trials to evaluate consumer preferences for beef steaks with varying degrees of discoloration. Steaks from the semimembranosus, semitendinosus, biceps femoris, and vastus lateralus were displayed in PVC overwrap packaging and were subjected to room temperature until the desired discoloration level and then refrigerated through retail display (Hood and Riordan, 1973). The authors then collected instrumental color data and displayed steaks with discoloration levels ranging from 5% to 33%. Their findings indicated significant consumer discrimination against discolored steaks, particularly when discolored meat was displayed alongside bright-red meat, which aligns with the findings in the current study. While Hood and Riordan (1973) provided foundational insights into consumer preferences, the study did not encompass the full spectrum of discoloration, like the current work, which included discoloration from 0% to 100%. Furthermore, Hood and Riordan’s (1973) work was conducted over 50 years ago, drawing into question whether today’s consumer preferences and insights would be similar to this work or may have shifted over time.