Estimation of the Thurstonian model for the 2-AC protocol

The 2-AC protocol is a 2-AFC protocol with a “no-difference” option and is technically identical to the paired preference test with a “no-preference” option. The Thurstonian model for the 2-AC protocol is parameterized by δ and a decision parameter τ, the estimates of which can be obtained by fairly simple well-known methods. In this paper we describe how standard errors of the parameters can be obtained and how exact power computations can be performed. We also show how the Thurstonian model for the 2-AC protocol is closely related to a statistical model known as a cumulative probit model. This relationship makes it possible to extract estimates and standard errors of δ and τ from general statistical software, and furthermore, it makes it possible to combine standard regression modelling with the Thurstonian model for the 2-AC protocol. A model for replicated 2-AC data is proposed using cumulative link mixed models. Examples will be given throughout the paper and the methodology is implemented in the authors’ free R-packages sensR and ordinal.     

Relating consumer and trained panels’ discriminative sensitivities using vanilla flavored ice cream as a medium

This research investigated the possibility of uncovering a relationship between the sensitivities of trained and consumer panels in three experiments. In Experiment I, five studies were conducted using vanilla flavored ice cream. In each study, two ice cream samples differing in formulation and/or their manufacturing process were used. They were compared by both panels using same–different tests with sureness judgments (degree of difference methodology). Using the appropriate probabilistic Thurstonian model, d′ values, a measure of the underlying sensory difference perceived between the products, were calculated and the underlying relationship between the two panels’ sensitivities uncovered. An additional study was then conducted (Experiment II). A new pair of ice creams differing in fat content was first evaluated by the trained panel. Based on the estimated d′ value (trained panel’s measured d′ = 2.69), the corresponding consumer d′ value was predicted using the relationship uncovered in Part I (consumer panel’s predicted d′ = 1.54). The same pair of ice creams was then evaluated by the consumer panel. The measured d′ value for the consumers was almost identical to that predicted by the uncovered relationship (consumer panel’s measured d′ = 1.56). Along with the discrimination component of these studies, paired preference tests were conducted (Experiment III) in order to study a link between perceived difference and expressed preference. The results give an indication of when a perceived difference might start translating into a change in acceptability of the original product. These results indicate the potential of such an approach to predict consumers’ perceptions from an in-house semi-trained or trained panel, providing a useful predictive tool and a means of reducing repetitive and costly consumer testing.     

The evolution of a model: A review of Thurstonian and conditional stimulus effects on difference testing

The application of Thurstonian modeling to difference tests is reviewed and explained non-mathematically. How experimental variables can affect the variance of perceptual distributions and reduce discriminability (d′ values) is described. The tendency of triangle and 3-AFC tests with one stronger stimulus and two weaker stimuli to be more sensitive than triads with one weaker and two stronger stimuli, is explained using the idea of conditional stimuli. From this, the development of sequential sensitivity analysis and the four distribution Thurstonian model is described. These ideas are then developed further to produce a potential eight distribution conditional stimulus model, to account for the effects of various interstimulus rinses and cognitive contrast. Also, the physical and physiological properties of the salt taste detection system, used in the development of these models, are outlined.     

Comparison of the triangle and a self-defined two alternative forced choice test

The triangle test and the 2-AFC test or paired comparison are commonly used discrimination tests. The 2-AFC is statistically more powerful than the triangle test, but can only be used when the quality of the difference is known (e.g. one sample has added sucrose and is expected to be sweeter). By allowing subjects to define their own criteria for use in the 2-AFC, the procedure might be used when the sensory difference between two samples is not known or easily defined, as is possible during a warm-up protocol. The research reported here investigated a modification of the sometimes extensive warm-up procedure but limited the number of samplings of the preview samples. This modified procedure we describe as a self-defined 2-AFC (SD-2-AFC). This method was compared to the triangle and conventional 2-AFC using four food systems, differing in the magnitude and quality of sensory differences. The AFC tests (including the SD-2-AFC) performed better than the triangle on a statistical basis. However, in some cases the SD-2-AFC produced lower d-prime values indicating lower absolute levels of sensory differentiation. Correlates of poorer discrimination with the SD-2-AFC included miss-identification of the critical attribute (as determined by majority response) or reversal of the direction of the difference during the inspection phase. Results were interpreted in terms of Thurstonian models. No advantage of adding a preview to the triangle or 2-AFC was observed. The overall statistical advantage of the SD-2-AFC suggests it should be considered as an alternative to the triangle test.     

Sequential use of the triangle, 2-AC, 2-AFC,and same-different methods applied to a cost-reduction effort: consumer learning acquired throughout testing and influence on preference judgements

Taste discrimination and preference of loyal, heavy users of a popular fruit beverage were measured to support a cost reduction effort. The research objective was to determine the highest reduction level permissible without effecting a noticeable perceptual difference among heavy users of the beverage. Judgements of an internal employee panel were considered for predictions of loyal, heavy-user consumer response. Heavy users were defined as individuals who consumed the beverage at least once a week or more. Internal panel 2-AFC and consumer triangle, 2-AC, 2-AFC, and same-different discrimination results collected using the sip methodology were compared using d′ — a measure of the degree of sensory difference to: (1) support the cost reduction effort; (2) study the relationship between scale methods; and (3) examine and establish the relationship between internal and external consumer sip test results. Results of these combined studies showed that while the ingredient reduction was perceived as different even at the lowest levels of reduction and delivered negative implications at 100% reduction, a mid-reduction level was successful in satisfying consumer preference. In light of these findings, the action-decision criteria for moving forward with the cost-reduction reformulation was modified from ‘no perceivable difference’ to ‘parity or superior preference’. Results of these studies led to a successful cost-reduction of the beverage.     

Observed discriminability is more variable than predicted by signal detection theory

Food scientists and technologists are interested in how sensitive judges and consumers are to changes in product formulations. Numerous approaches to measuring sensitivity have developed at a rapid rate in the last 25 years, however, the evaluation and assessment of sensory tasks is still ongoing. The current study compared the efficiency of four difference tests (A-Not A, A-Not AR, 2-AFC, 2-AFCR) across three concentrations (difficulty levels) of orange essence. To determine which of the four tasks had the lowest variance in d′, the decision strategy used by the judges first had to be ascertained. Using the rating-scale approach of signal detection theory, four estimates of sensitivity, d′, were obtained for each judge (n = 10) at each of the three difficulty levels, for all four tasks. Results indicate that the majority of judges adopted the β-strategy for the A-Not A and A-Not AR tasks, and the β/τ-strategy for the 2-AFC and 2-AFCR tasks. The A-Not A and A-Not AR tasks produced greater estimates of d′ than the other two tasks, and the A-Not A task exhibited a greater variance in d′ than the other three tasks.     

A Thurstonian model for the degree of difference protocol

In the degree of difference methodology (DOD), subjects are presented with pairs of samples, either identical or different, and must indicate how different the samples are using a t-point category rating scale. In this article, the Thurstonian model for the DOD assuming independent assessments is derived. The model permits the estimation of the size of the underlying difference between the products (δ), the variance of this estimate, as well as the sizes of the t − 1 τ criteria (scale boundaries on the difference perceptual distribution). This model expands further the collection of Thurstonian models already available for many discrimination, rating and ranking methodologies.     

Delta-tilde interpretation of standard linear mixed model results

We utilize the close link between Cohen’s d, the effect size in an ANOVA framework, and the Thurstonian (Signal detection) d-prime to suggest better visualizations and interpretations of standard sensory and consumer data mixed model ANOVA results. The basic and straightforward idea is to interpret effects relative to the residual error and to choose the proper effect size measure. For multi-attribute bar plots of F-statistics this amounts, in balanced settings, to a simple transformation of the bar heights to get them transformed into depicting what can be seen as approximately the average pairwise d-primes between products. For extensions of such multi-attribute bar plots into more complex models, similar transformations are suggested and become more important as the transformation depends on the number of observations within factor levels, and hence makes bar heights better comparable for factors with differences in number of levels. For mixed models, where in general the relevant error terms for the fixed effects are not the pure residual error, it is suggested to base the d-prime-like interpretation on the residual error. The methods are illustrated on a multifactorial sensory profile data set and compared to actual d-prime calculations based on Thurstonian regression modeling through the ordinal package. For more challenging cases we offer a generic “plug-in” implementation of a version of the method as part of the R-package SensMixed. We discuss and clarify the bias mechanisms inherently challenging effect size measure estimates in ANOVA settings.     

Investigation of test performance of the dual reminder A-Not A (DR A-Not A) in comparison to 3-AFC for discriminating samples of drinking water

Although the relative performance of sensory discrimination methods is well theorized in Thurstonian modeling and signal detection theory (SDT), empirical research is needed to investigate how such theorized models could be validated in complex sensory testing situations of food and beverages. This paper presents a practical procedure to utilize the existing SDT-based A-Not A sensory discrimination model based on a beta-strategy for detecting the off-sensory quality of samples of drinking water and validates the model for effective and efficient sensory quality management of food and beverages. Dual reminder A-Not A (DR A-Not A) using two tastings of the reference stimulus before evaluating every test stimulus is proposed as the optimal test procedure because it uses more sensitive test sequences and facilitates subjects’ familiarization with the standard quality of the reference. To test the performance of DR A-Not A, 3-AFC, which also uses three stimuli and is the recommended method for detecting odor or taste stimuli by ASTM, was used as a control method. In Experiment 1, 98 subjects each performed both DR A-Not A and 3-AFC. Based on these results, only sensitive subjects were selected for the next experiment, in which they were divided into two equally well performing groups. In Experiment 2, each group performed either DR A-Not A or 3-AFC over three repeated sessions. The results confirmed that the A-Not A beta-strategy was adopted for DR A-Not A and that its test performance was improved over replications. These results suggest that although DR A-Not A is an unspecified difference test method and does not use a forced-choice task, embedding the familiarization procedure for the reference renders it an effective sensory difference method for the sensory quality management of drinking water.     

The double discrimination methods

Theoretical aspects of the double discrimination methods are investigated in this paper. The main aspects involve statistical models, test powers, Thurstonian d′ s and variance of d′ s for the methods. The double discrimination methods can be treated under the same framework as the conventional discrimination methods, i.e. under the same binomial model but with different parameters. The double discrimination methods possess lower guessing probabilities and higher test powers than the conventional discrimination methods. The relationship between the double methods and the conventional methods in Thurstonian d′ s and variances of d′ s are derived in the paper. The tables of critical values for tests and the table of sample sizes required to reach 0.8 test power for the double two-alternative forced choice (the double 2-AFC), the double three-alternative forced choice (the double 3-AFC), the double triangular and the double duo-trio methods are given.     

Effects of Order of Tasting in Sensory Difference Tests Using Apple Juice Stimuli: Development of a New Model

ABSTRACT:  3-AFC discrimination tests between 2 confusable versions of apple juice were performed using a variety of interstimulus rinsing regimes, to investigate the effects of order of tasting on difference test sensitivity. The results were compared to predictions from the sequential sensitivity analysis (SSA) model and the conditional stimulus (CS) model, which have been developed to account for differences in model systems. These models did not account for the data for this beverage system. A new model was developed. This took a Thurstonian/signal detection approach but included new elements. It considered the gradual build-up of adaptation effects that would be encountered as a judge proceeded through a 3-AFC test with a relatively strong flavored food system. It further considered cognitive contrast effects, which had not been considered in the prior models. The model was named the sequential perception analysis (SPA) model and it accounted for the data found with the apple juice stimuli. The predictions of the model were precise enough to account for differences between individual triads, which the former models had not been able to do.     

Effect of Acylglycerol Composition and Fatty Acyl Chain Length on Lipid Digestion in pH‐Stat Digestion Model and Simulated In Vitro Digestion Model

In this study, a pH‐stat digestion model and a simulated in vitro digestion model were employed to evaluate the digestion degree of lipids depending on different acylglycerols and acyl chain length (that is, diacylglycerol [DAG] compared with soybean oil representing long‐chain triacylglycerol compared with medium‐chain triacylglycerol [MCT]). In the pH‐stat digestion model, differences were observed among the digestion degrees of 3 oils using digestion rate (k), digestion half‐time (t_1/2), and digestion extent (Φmax). The results showed the digestion rate order was MCT > soybean oil > DAG. Accordingly, the order of digestion half‐times was MCT < soybean oil < DAG. In simulated in vitro digestion model, digestion rates (k′) and digestion half‐times (t′_1/2) were also obtained and the results showed a digestion rate order of MCT (k′ = 0.068 min^?1) > soybean oil (k′ = 0.037 min^?1) > DAG (k′ = 0.024 min^?1). Consequently, the order of digestion half‐times was MCT (t′_1/2 = 10.20 min) < soybean oil (t′_1/2 = 18.74 min) < DAG (t′_1/2 = 29.08 min). The parameters obtained using the 2 models showed MCT was digested faster than soybean oil, and that soybean oil was digested faster than DAG.     

Exploration of a new consumer test method based on metacognitive certainty

Successful product development and marketing necessitate a study of the consumer concept of culture-specific or deep-positioned branded food. In this study, a new consumer test method was designed based on an authenticity test and used as a reference frame for the target concept without an upsetting story. The response format of this method included the metacognitive certainty response following the sensory authenticity response using the A-Not A test procedure. The method was applied to study the concept of goso flavor, as perceived by 91 female consumers with three commercial soymilk products, having each consumer evaluate each product 45 times over three days. The repeated responses of sensory authenticity were analyzed as mean scores and signal detection theory (SDT) d-prime (d′) values of the product difference. From the metacognitive certainty responses after the sensory authenticity response, a new quantitative group measure of d-prime metacognition (d′_MC) was calculated in the SDT context and compared with the other outputs. The measure ranged from negative to positive values, indicating a mismatch to a match for the concept of each product. Data analyses were conducted on both pooled data and segmented data, which was driven from the results of cluster analyses using the mean sensory authenticity scores and SDT C values (estimates of response bias about the concept tested). The results showed that d′_MC of each product corresponded to the mean scores and d′ with the advantage of easy interpretation. Overall, d′_MC can be a useful group measure for studying the consumer concept towards food and beverages.     

Unspecified duo–trio tests can be as powerful as the specified 2-AFC: Effects of instructions and familiarization procedures on cognitive decision strategies

Using signal detection theory (SDT) and Thurstonian analysis, it has been possible in the food industry to make various business decisions such as reformulations and cost reductions based on the degree of consumer sensory discriminability by applying equivalence/difference testing. To achieve more efficient decision-making, it is important to investigate a theoretically and operationally more powerful consumer discrimination test method. Cognitive decision strategies used in discrimination methods largely determine relative test power and thus are of theoretical and practical significance. The goal of this study was to investigate if test procedures using a constant-ref. duo–trio design enable untrained/naïve subjects to use a powerful 2-AFC strategy. Although previously the constant-ref. duo–trio design showed operational superiority to other unspecified test methods, the ‘comparison of distance’ (COD) strategy which is conventionally assumed for the duo–trio, lacks theorized power. The tested variables included two versions of duo–trio design, different familiarizations, instructions and experimental session designs. Three different iced tea samples were discriminated from a reference by 263 untrained/naïve subjects using 5 different duo–trio test procedures. Results showed that untrained/naïve subjects were able to perform the duo–trio with the reference presented twice, first and middle (DTFM) as a 2-AFC procedure, even with multiple pairs in a block design, after only two reference tastings with the information of the constant-reference. The duo–trio preceded by specified tetrad tests also showed potential as another efficient two-sequence test design. These results suggest that such constant-ref. duo–trio protocols using a 2-AFC strategy can be an alternative to the unspecified triangle and tetrad test.     

Variants of A Not-A and 2AFC tests: Signal Detection Theory models

Signal Detection Theory provides an approach to modelling difference test procedures that allows an estimate of the sensory difference between products to be obtained independently of the response bias and decision strategy used by the judge, and of the test procedure used to estimate that difference. The estimate of sensory difference obtained, d′, is therefore much closer to a ‘true’ estimate of sensory difference than many other measures employed by sensory scientists. The methods used to generate models based on Signal Detection Theory are standard, and these methods are applied to the development of models for two test procedures (A Not-A and 2AFC) and their reminder equivalents (A Not-AR and 2AFCR). 2AFCR is procedurally identical to the duo-trio procedure with a constant reference; however, it is shown that there is more than one decision strategy available to the judge in this task.     

Identification of optimal membrane morphological parameters during microfiltration of mosambi juice using low cost ceramic membranes

In this work, experimental investigations were carried out for the identification of optimal membrane morphological parameters (pore size distribution, average pore size and porosity) during microfiltration (MF) of mosambi juice using low cost ceramic membranes. Four different low cost ceramic membranes with different pore diameters (d_m) and porosities (?) were subjected to MF studies to evaluate the effect of d_m and ? on permeation characteristics (permeate flux and juice quality) of centrifuged mosambi juice (CJ) and enzyme treated centrifuged mosambi juice (ETCJ). Subsequently, a convenient factor namely effective permeable area factor (?_md_m²) was evaluated to provide greater insights in the fouling phenomena. Various physio-chemical properties such as colour, clarity, pH, citric acid content, density, total soluble solid (TSS) and alcohol insoluble solids (AIS) were measured for both feed and permeate juice samples to evaluate the effect of ?_md_m² on juice quality. Typical permeate fluxes were observed to vary from 5.78 × 10⁻⁶ to 13.45 × 10⁻⁶ m³/m² s for CJ and 14.07 × 10⁻⁶ to 60.64 × 10⁻⁶ m³/m² s for ETCJ at 82.7 kPa (ΔP) for different membranes whose ?_md_m² varied from 0.249 to 0.783 μm². Among different membrane pore blocking models, flux decline by cake filtration was found to be the best fitted model. The cake filtration model constant (k_c) was found to vary with ΔP and ?_md_m² and was empirically correlated. Phenomenological models were proposed to illustrate the dependency of total hydraulic resistance of membrane on ?_md_m², ΔP and time (t). Based on experimental as well as theoretical investigations, membranes with ?_md_m² up to 0.443 μm² for CJ and 0.294 μm² for ETCJ and a ΔP of 82.7 kPa were recommended for MF of mosambi juice.     

Antiproliferative and cytoprotective activities of a phenolic-rich juice in HepG2 cells

Antioxidant phytochemicals, especially the phenolics found in fruits and vegetables, have been proposed as the major bioactive compounds providing the health benefits associated with diets rich in plant-foods. In the present study, we aimed to evaluate the modulatory effect of a functional juice rich in phenolic compounds on cell proliferation and on the protection from oxidative stress caused by tert-butylhydroperoxide (t-BOOH) and hydrogen peroxide (H₂O₂) in HepG2 cells. Cell growth and cell viability were evaluated by the 3-(4,5-dimethyl thiazol-2-yl)-2,5-diphenyl tetrazolium bromide (MTT) assay. Lipid peroxidation was measured as the thiobarbituric acid reacting substances (TBARS) assay. Glutathione peroxidase (GPx) and glutathione-s-transferase (GST) activities were assayed by colorimetric assays. In order to elucidate the action mechanisms of juice phenolics, the protective effects of the juice were compared to those of the radical scavenger N,N′-diphenyl-1,4-phenylene-diamine (DPPD) and the iron chelator o-phenanthroline (o-phe). Cell proliferation was inhibited in a dose-dependent manner by juice extracts. Sub-toxic concentrations of phenolics (2–30 μg/mL of medium) increased cell survival and a decreased lipid peroxidation after exposure to both H₂O₂ and t-BOOH. Both juice and o-phe decreased the t-BOOH-induced GPx activation, but not that of GST. DPPD suppressed the t-BOOH-induced GST activation but not that of GPx. Our results support the efficacy of natural phenolics from grapes and berries in offering protection against oxidative stress. The possible role of an iron chelating mechanism, potentially of biological relevance, is proposed as being partially responsible for the health benefits provided by phenolic-rich processed foods.