Understanding Consumer Preference for Functional Barley Tortillas Through Sensory, Demographic, and Behavioral Data
Ayako Toma,1 María Botero Omary,1,2 Kurt A. Rosentrater,3 Elizabeth A. Arndt,4 Naruemon Prasopsunwattana,5 Sutida Chongcham,6 Rolando A. Flores,7 and Sylvia P. Lee1
ABSTRACT Cereal Chem. 85(6):721–729
Consumption of whole grains as part of a health-promoting diet is en-couraged among Americans due to beneficial nutrients and phyto-chemicals. The main objectives of this study were to investigate the effect of whole barley flour particle size on consumer acceptance of whole grain and soluble fiber-enriched tortillas; to examine the effect of demographic and behavioral data on consumer acceptance; and to establish relation-ships between consumer acceptance and survey results to identify market opportunities among particular groups of consumers. Four experimental tortillas and two representative commercial brands were tested by 95 untrained panelists using a 9-point hedonic scale for appearance, color,
flavor, texture, and overall acceptability. Demographic and behavioral data on age, gender, ethnicity, frequency of tortilla consumption, daily effort to include fiber in the diet, consumption of tortillas containing ≥1 g of dietary fiber/serving, consumption of low-fat tortillas, and tortilla brands and type used were collected. Potential market segments for these tortillas include older consumers (41+), women, Hispanics, and Asians, and frequent tortilla consumers. A response surface relating flavor and texture to overall acceptability produced almost equivalent results to the multivariate PLS results in terms of predicting overall acceptability, which facilitates analysis and interpretation.
The U.S. population has not incorporated enough whole grains
and soluble fiber into their diet to meet the recommended con-sumption of at least 3-oz equivalents/day for whole grains (Die-tary Guidelines 2005) and 25 g of total dietary fiber, of which 6 g should be soluble fiber (FDA and DHHS 2005). USDA Dietary Guidelines for Americans (2005) recommend that at least half of the daily intake of grain should be from whole grains. Incorporat-ing whole grain barley into food products can serve as a solution to increasing whole grain intake, which may help reduce the risk of heart disease, type-2 diabetes, and certain cancers and help with weight maintenance (Dietary Guidelines 2005).
Whole grains differ from refined grains in the amount of vita-mins, minerals, fiber, and other protective substances that they provide. Also, different whole grains differ in nutrient content (Kantor et al 2001). The inclusion of a whole grain containing β-glucan-soluble fiber would also increase the amount of soluble fiber to help meet dietary recommendations. Whole grain barley has distinctive nutritional properties that include soluble and in-soluble fiber. The soluble fiber in barley ranges from 4% (Rieck-hoff et al 1999) to 12% (Arndt 2006) of the kernel weight and is primarily β-glucan, the viscous, cholesterol-lowering fiber also found in oats.
Incorporation of whole barley flour into tortillas is an approach to delivering whole grains and soluble fiber into the American diet, especially among Hispanics, who are the fastest growing minority in the U.S. According to the U.S. Census Bureau, the Hispanic population in 2005 was well over 42 million (www.census.gov/). In most parts of Mexico, flour tortillas are more popular than corn tortillas (Serna-Saldivar et al 1988). The Tortilla Industry Associa-
tion reported tortillas as the second most popular bread type in America, with Americans consuming ≈85 billion tortillas in 2000 (www.tortilla-info.com). Tortilla consumption is increasing in both Hispanic and non-Hispanic consumers.
Serna-Saldivar et al (1988) found that increasing the amount of fiber in flour tortillas decreases the product quality and consumer acceptance. Friend et al (1992) evaluated the sensory properties of tortillas using an untrained sensory panel. They found that the color of control tortillas (containing 0% whole white wheat flour) were similar to the tortillas made with 50% whole white wheat flour, but were significantly more preferred than the color of tor-tillas containing 50% whole red wheat flour. In terms of texture, the 100% whole wheat tortillas were less preferred than tortillas made with 0 or 50% whole wheat flour. Adding fiber by means of whole wheat flour affected dough processing and tortilla quality, but improved the nutritional properties of the tortilla (Friend et al 1992). The nutritional quality and sensory properties of tortillas can be optimized by determining the maximum fiber content through not only instrumental testing, but also by sensory testing to gauge consumer acceptability.
There is a growing interest in cross-cultural research on food perception and preference to better understand specificities of individual cultures (Mialon et al 2002). Evaluation of consumer preferences related to demographic and behavioral data can facili-tate the identification of potential market opportunities. Market share can be maximized by understanding what sensory proper-ties drive consumer liking. According to Thybo et al (2003), pref-erence data are to be interpreted through sensory, instrumental, demographic, and behavioral analyses. Attainment of reliable information needs to be based on statistical methods yielding observable patterns that point to differences among products, as well as possible segmentations among the consumers (Martens and Martens 2001; Thybo et al 2003). Therefore, in developing a new product, preference mapping can be helpful in two ways. First, it can be used to verify that a prototype product is accept-able, and second, it will help the product fall into the correct mar-ket segment (Helgesen et al 1997). Consumer data is used in internal preference mapping to determine consumer preference patterns. On the other hand, in external preference mapping, con-sumer preference data is used in conjunction with sensory infor-mation or instrumental data (Santa Cruz et al 2003). Meullenet et al (2002) determined overall preferences for tortilla chips as well as specific sensory characteristics in tortilla chips that were pres-ently on the market. Internal preference mapping showed that
1 Human Nutrition & Food Science, California State Polytechnic University, Pomona,CA 91768.
3 USDA-ARS, North Central Agricultural Research Laboratory, Brookings, SD57006. Mention of a trade name, propriety product or specific equipment doesnot constitute a guarantee or warranty by Cal Poly Pomona University or theUnited States Department of Agriculture and does not imply approval of a prod-uct to the exclusion of others that may be suitable.
4 ConAgraFoods, Inc., Omaha, NE 68102. 5 Current address: 3234 Bali Dr., West Covina, CA 91792. 6 Current address: 1465 W. Avenue 43, Los Angeles, CA 90065. 7 Dept. Food Sci and Technology, University of Nebraska, Lincoln, NE 68588.
flavor was the most important attribute to consumer overall ac-ceptance followed by texture and appearance.
Cho et al (2005) used descriptive analysis and consumer sen-sory testing on canned tea products to understand the sensory attributes that drive consumer liking. The study also determined the effects of consumer age and product information on the accep-tance of certain types of tea products. For the study, 500 tea drinkers were divided into five age groups and each age group was subdivided into two groups that tasted 10 canned tea samples, with or without accompanying information about each sample. Preference mapping was used to better understand the main sen-sory characteristics that drive consumer liking. Interestingly, when the information was presented along with the tea, accep-tance leaned toward teas that were marketed for health benefits such as bitter/astringent tasting tea products, whereas, when the information was not provided, a majority of consumers preferred the lemon-flavored black tea.
Kälviäinen et al (2002) used a trained sensory panel and three consumer groups (young adults, adults, and elderly) to rate six experimental and two commercial muesli oat flakes. Results showed that texture was comparatively more important to the elderly because they preferred muesli oat flakes that did not ad-here to their teeth and needed little mastication. Different types of muesli oat flakes appealed to different consumer segments.
In a study conducted by Kihlberg and Risvik (2007), market segmentation among organic consumers was conducted by using both a questionnaire and sensory data on breads scored for liking. The test subjects were split into two age groups: up to and includ-ing 30 years and more than 30 years of age. Results showed the majority of consumers in both age groups preferred the flavor of the organic bread compared with the conventional bread. How-ever, ≈50% in both age groups stated that they would not buy an organic food product if the prices were much higher than the con-ventional food product.
Principal component analysis (PCA), a type of multivariate preference mapping, is another method used by researchers to investigate relationships among sensory and demographic or pref-erence data. According to Rosentrater et al (1999), an advantage of principal component analysis is the ability to easily identify outliers through the examination of the scatterplots of the calcu-lated principal component scores. Correlational structure, clustering, skewness, and outliers in a data set can be easily assessed (Rosen-trater 2004).
The objectives of this study were to 1) investigate the effect of whole barley flour particle size on consumer acceptance of whole grain, soluble fiber-enriched tortillas; 2) examine the effect of demographic information, preference, and attitudes toward fiber-enriched and low-fat tortillas on consumer acceptance; 3) identify niche markets and market opportunities among particular groups of consumers based on the relationships between consumer ac-ceptance and survey results; and 4) explore mathematical models to help predict consumer acceptance of sensory attributes (inter-nal mapping) and demographic or preference data (external map-ping) to help streamline product development. To our knowledge, no similar study has been published to-date.
Agra Foods, Omaha, NE) containing 12% minimum β-glucan soluble fiber, and 30% minimum TDF was used in this study. All other ingredients were purchased from a local grocery store.
Tortillas were made with four different flour blends of 84% (flour basis) refined wheat (bread) flour and 16% (flour basis) whole barley flour (WBF). Each whole barley flour had a different parti-cle size distribution, while the bread flour particle size distribu-tion remained constant. Average particle sizes for control (C) bread flour, regular (R), intermediate (I), and microground (MG) whole barley flours were 72, 237, 131, and 68 μm, respectively. Average particle sizes for corresponding blends were 97, 81, and 72 μm, respectively.
Tortilla Preparation The formula and procedure for tortilla making was based on the
methods published by Mao and Flores (2001) and Bello et al (1991) with some modifications (unpublished data). See Table I for the formulation used. All dry ingredients, including bread flour, whole barley flour (Sustagrain), baking powder (Clabber Girl, Terre Haute, IN), and salt (Morton Iodized Salt, Chicago, IL), were mixed in a mixer (K45SS, Kitchen Aid, St. Joseph, MI) at speed 1 for 30 sec. Canola oil (Stater Brother’s canola oil, Colton, CA) and water (25 ± 2°C) were added and the doughs were mixed to full development for 2 min at speed 2 and another 1 min at speed 4. The absorption was directly proportional to the amount of β-glucan present in the flour blend. The dough was allowed to rest in a plastic covered mixing bowl at 25°C for 5 min. Afterwards, the dough was portioned into 73-g pieces and flattened for 10 sec using a dough press (DP2000, Doughpro Pro-process Corporation, Paramount, CA) with a top plate tempera-ture of 72.2°C and then turned over and pressed for another 10 sec. Tortillas were cooked on a cast iron pan at 232.2 ± 2.8°C for 10 sec, turned over for 30 sec, and lastly turned over for 20 sec, giving each side a total of 30 sec of cooking time.
Cooked tortillas were allowed to cool to room temperature. The cooled tortillas were stored in double resealable plastic bags and held for 24 hr before testing and sensory evaluation. During sen-sory evaluation, only a wedge of the tortilla was fed to panelists rather than the whole tortilla to avoid satiety before the conclu-sion of the experiment.
Barley tortillas prepared for this experiment weighed 66 g. However, nutrition composition data presented in Table II for the experimental tortillas and two commercial products are based on the Reference Amount Customarily Consumed (RACC) for tortil-las of 55 g. Based on 55 g, experimental tortillas contained 148 calories, 7.9 g of whole grains, 2.8 g of total dietary fiber (TDF), 1.5 g of β-glucan soluble fiber, 3.3 g of total fat, <1 g of saturated fat, and 0 g of cholesterol. One commercial tortillas was low-fat, low-carb, and high-fiber (LF/LC/HF) and one was a handmade flour tortilla (HM). Based on 55 g, the LF/LC/HF tortilla contained 157 calories, no whole grains, 16 g of TDF, no soluble fiber, 4 g of
TABLE IFormula for Functional Barley Tortillas
Ingredients
Control (%)
Regular, Intermediate, and Microground (%)
Control (% flour basis)
Regular, Intermediate, and Microground (% flour basis)
total fat, 0 g of saturated fat, and 0 g of cholesterol. Tortilla label information on the HM product indicated that the tortillas were handmade and contained 179 calories, no whole grain, no soluble fiber, 6 g of total fat, 1 g of saturated fat, and 0 g of cholesterol.
Demographic and Behavioral Data Demographic data and consumer attitude toward tortillas and
fiber were collected during sample evaluation (Table III). Ninety-five untrained panelists including students, faculty, and staff from Cal Poly Pomona participated in this survey. This survey was conducted to identify niche markets for these functional tortillas and to better understand the nutritional education level in the uni-versity population.
Panelists reported their age, gender, and ethnicity. They also answered other questions such as how often they consumed tortil-las, if they made a daily effort to include fiber in their diet, if they consumed tortillas that contain ≥1 g of dietary fiber/serving, and if they consumed low-fat tortillas.
Sensory Evaluation The same panel that participated in the survey evaluated six
randomly presented tortilla samples for overall acceptability (OA), appearance (A), color (C), flavor (F), and texture (T) using a 9-point hedonic scale. Products tested included four experimental tortillas: control (C), regular (R), intermediate (I), and microground (MG); and two commercial flour tortillas: a low-fat, low-carb, high-fiber tortilla (LF/LC/HF) and a handmade flour tortilla (HM).
Statistical Analyses All data was analyzed using a generalized linear model (v.9.1,
SAS Institute, Cary, NC). The least significant difference (LSD) was conducted at the 95% confidence interval.
Relationships between all collected data were then examined using Microsoft Excel 2003, SAS v.8.0 , Minitab v.14.11 (State College, PA), and TableCurve 3D v.4.0 (San Jose, CA) software. These analyses included linear regression, multiple polynomial regression, principal component analysis, and partial least squares regression analysis.
RESULTS AND DISCUSSION
Survey Results Table III gives details of the survey results. Among the 95 par-
ticipants, 86% were 18–40 years of age. Fifty-four percent (54%) of the participants were male. The population was diverse and balanced, including 33% whites, 29% Hispanics, 34% Asians, and 4% people of other ethnicities. Fifty-nine percent (59%) of panel-ists reported consuming tortillas at least once a week, and 23% at
least once a month. The remaining 18% included four panelists who consumed tortillas 3–4 times per week, four panelists who consumed tortillas twice a month, and seven panelists who rarely consumed tortillas. Fifty-seven percent (57%) of the respondents reported making a daily effort to include fiber in their diet, how-ever, when asked about consuming tortillas with >1 g of fiber per serving, 84% panelists either did not make an effort to consume tortillas with fiber or did not know. Seventy-six percent (76%) of participants indicated they do not seek or knowingly consume low-fat tortillas.
TABLE IINutrient Composition of Experimentala and Commercialb Tortillas
a C, control, no whole barley flour (WBF); R, regular, 84:16 bread flour and regular WBF; I, intermediate, 84:16 bread flour and intermediate WBF; MG, micro-ground, 84:16 bread flour and microground WBF.
b LF/LC/HF, low-fat, low-carb, high-fiber; HM, handmade. c Based on Reference Amount Customarily Consumed (RACC) of 55 g for tortilla. d Per serving size noted on the label.
TABLE III Background Data Including Age, Gender, Ethnicity, and Behavioral
Responses Toward Tortillas, Fiber, and Fat Contenta
n %
Age 18–40 81 86 ≥41 13 14
Gender Male 50 54 Female 42 46
Ethnicity White 30 33 Hispanic 27 29 Asian 31 34 Other 4 4
Frequency of eating tortillas Once a day 14 15 Once a week 40 44 Once a month 21 23 Other 17 18
Daily effort to include fiber in diet Yes 52 57 No 40 43
Consume tortillas with >1 g of dietary fiber/serving Yes 14 16 No 6 7 Don’t know 70 77
Type of tortilla consumed Flour 36 61 Flour and corn 14 24 Whole wheat and flour 9 15
a Some panelists left questions unanswered, thus some categories have fewer than 95 responses.
724 CEREAL CHEMISTRY
Fifty-three percent (53%) of the panelists reported consuming Mission brand tortillas, 31% consumed multiple brands, and the remainder were not sure. The last survey question asked for the particular tortillas consumed by the panelists; 61% indicated flour tortillas, 24% indicated corn and flour tortillas, and 15% con-sumed both whole wheat and flour tortillas, which corresponds to 16% of panelists consuming tortillas with ≥1 g of fiber/serving.
Treatment-by-Survey Interactions Overall results (Figs. 1–7) show that HM tortillas received the
highest score (7–8.5), followed by LF/LC/HF tortillas (6–8.5), the control (6–7), and WBF tortillas (4–7). In some cases, and for some interactions of attributes and demographic or preference questions, panelists reported similarities among one or more WBF tortillas and either one or both commercial products.
Effect of age. Results for age-by-treatment interaction (Fig. 1) showed that the younger group (86%) rated commercial tortillas higher than the WBF tortillas in most parameters evaluated. They also found no differences among the WBF tortillas (5.6). How-ever, the 41+ adults (14%) rated one or more WBF tortillas simi-lar to one or both of the commercial products. In other studies involving bread, tea products, and muesli oat flakes, age was also a factor affecting acceptance (Kälviäinen et al 2002; Cho et al 2005; Kihlberg and Risvik 2007).
Effect of gender. Males (57%) scored HM highest and both commercial products higher than the WBF tortillas in most pa-rameters evaluated (Fig. 2) and also found no differences among WBF tortillas. Compared with males, females (46%) gave higher overall scores in most attributes tested. Studies showed that females were more highly accepting of functional foods than males (Childs and Poryzees 1997; Gilbert 1997). The primarily household role
of females in food purchasing is important to consider. Lockie et al (2004) reported the selection of organics was strongly affected by gender and food purchasing role in the household.
Fig. 1. Effect of age on appearance, color, flavor, texture, and overallacceptability (A–E) of experimental tortillas made with whole barleyflour of different particle size and commercial tortillas. Control (C), regular (R), intermediate (I), and microground (MG). Commercial productsincluded a low-fat, low-carb, high-fiber tortilla (LF/LC/HF) and a hand-made tortilla (HM). Hedonic scores for the same tortilla type but differentage with the same letter (xyz) are not significantly different (P > 0.05). Hedonic scores for the same age but different tortilla type with the sameletter (abc) are not significantly different (P > 0.05). Lines above barsreflect individual standard deviation values.
Fig. 3. Effect of ethnicity on appearance, color, flavor, texture, and over-all acceptability (A–E) of experimental tortillas made with whole barley flour of different particle size and commercial tortillas. Explanation offeatures is given in Fig. 1.
Fig. 2. Effect of gender on appearance, color, flavor, texture, and overall acceptability (A–E) of experimental tortillas made with whole barley flour of different particle size and commercial tortillas. Explanation offeatures is given in Fig. 1.
Vol. 85, No. 6, 2008 725
Effect of ethnicity. Ethnicity-by-treatment interaction (Fig. 3) showed whites (33%) giving one or more WBF tortillas lower scores than Hispanics (29%) and Asians (34%). All ethnic groups scored WBF tortillas the same for most parameters tested, while giving either one or both commercial products higher scores.
Effect of frequency of eating tortillas. For the frequency-by-treatment interaction (Fig. 4), overall panelists in all groups rated HM and LF/LC/ HF tortillas highest. The most frequent consum-ers (once-a-day and once-a-week) rated some WBF tortillas the same as commercial products for most parameters.
Fig. 4. Effect of frequency of tortilla consumption on appearance, color,flavor, texture, and overall acceptability (A–E) of experimental tortillasmade with whole barley flour of different particle size and commercialtortillas. Explanation of features is given in Fig. 1.
Fig. 5. Effect of daily effort to consume fiber on appearance, color, flavor, texture, and overall acceptability (A–E) of experimental tortillas made with whole barley flour of different particle size and commercial tortillas.Explanation of features is given in Fig. 1.
Fig. 7. Effect of consumption of low-fat tortillas on appearance, color, flavor, texture, and overall acceptability (A–E) of experimental tortillas made with whole barley flour of different particle size and commercial tortillas. Explanation of features is given in Fig. 1.
Fig. 6. Effect of consumption of dietary fiber on appearance, color, flavor, texture, and overall acceptability (A–E) of experimental tortillas made with whole barley flour of different particle size and commercial tortillas.Explanation of features is given in Fig. 1.
726 CEREAL CHEMISTRY
Effect of daily effort to include fiber in the diet. Results for the frequency-by-treatment interaction (Fig. 5) showed both that panel-ists making a daily effort to consume fiber (57%) and panelists who did not (43%) scored either HM highest (7–8) or gave both commercial products the highest score (7–8); detected no differ-ences among experimental tortillas including C and WBF tortillas (5–7) or among C and LC/LF/HF tortillas for most parameters.
Effect of consumption of tortillas containing ≥1 g of dietary fiber/ serving. Results for this interaction (Fig. 6) showed that overall panelists in the group consuming tortillas with ≥1 g of dietary fiber/serving (16%) scored all experimental tortillas the same; found no differences among C and one of the commercial prod-ucts; and also detected no differences among the two commercial products for flavor, texture, and overall acceptability. At the same time, panelists in the group not consuming tortillas with ≥1 g of dietary fiber/serving (7%) scored most or all experimental tortillas the same as the commercial products for the same parameters noted above. The third group that was not aware of the content of ≥1 g dietary fiber/serving (77%) scored all WBF tortillas the same, and found no differences among the commercial products when rating flavor, texture, and overall acceptability.
Effect of consumption of low-fat tortillas. These results (Fig. 7) show that for all five parameters tested, the group that confirmed making an effort to eat low-fat tortillas (24%) found no differ-ences among all experimental tortillas (6.1); rated C and LF/LC/ HF tortillas similarly (6.5); and reported the highest score for the
HM tortilla (8.2). The groups who reported not making an effort to consume low-fat tortillas (48%) and who reported not knowing about the low-fat tortillas (28%) showed similar results as the group who made an effort to eat low-fat tortillas.
Multivariate Analysis Bivariate analysis. The data collected in this experiment was,
by its nature, multidimensional in structure, and to fully under-stand the data, further analyses were necessary. Toward this end, relationships between all survey and sensory data were investi-gated using linear regression analysis. The survey variables did not exhibit discernable relationships, either with other survey variables, or with any of the sensory variables based on linear regression analysis. The sensory variables alone, on the other hand, did indicate the possibility of linear relationships (Table IV, Fig. 8) between each other. There was scatter in the data though, and the highest coefficient of determination (0.51) was for color as a function of appearance (Table IV, #2). These results indicated the need to examine other techniques to more fully investigate possible variable relationships.
Multiple polynomial regression showed a relationship between overall acceptability as a function of flavor and texture (R2 = 0.61) (Table IV, #12) As indicated in Fig. 9, a linear response surface fit the data well, even though there was considerable scatter in the data. This relationship indicates the strong possibility to predict the overall acceptability of the tortilla products once the flavor and texture scores have been established.
Principal components analysis. To further investigate the rela-tionships and interactions between the survey and sensory data, a principal component analysis (PCA) was conducted using all the variables in the study. This type of analysis is typically used to reduce the dimensionality of multivariate data by summarizing the observed variance and projecting it onto a set of uncorrelated, orthogonal linear combinations (eigenvectors) based on the origi-nal variables that have the form yPC = a1X1 + a2X2 + . . . + azXz
where yPC. is a principal component value or score; a1 through aZ are principal component coefficients (eigenvectors), and X1 through XZ are the original property variables in vector form (Marascuilo and Levin 1983; Everitt and Dunn 1991).
A loading plot of the resulting principal component eigenvalues (Fig. 10A) visually indicates that the sensory data were approxi-mately twice as influential in terms of summarizing data compared
Fig. 8. Scatterplot matrix of sensory data indicates the possibility of linear relationships between the variables. Matrix provides a simultaneous view of all X-Y scatterplots possible between the variables, with each serving as vertical and horizontal axes, respectively. Numerical units are not provided for clarity.
TABLE IV Linear Regression Models Developed to Investigate Potential
with either brand or the survey data, as noted by the distance of each from the convergence point. A “scree” plot of the error ex-plained through the use of these principal components (Fig. 10B) indicates that three components may be sufficient in summarizing the data in multidimensional space. Examining the error ex-plained by each principal component (Table V), however, reveals that three components only accounted for 52.2% of all of the vari-ance in the data, which is not sufficient to encapsulate the data. To account for a cumulative error of 80%, seven principal compo-nents are necessary; but to achieve a 90% data summary, nine principal components are required, which is only a reduction of three dimensions (from the original 13). Thus it appears that PCA does not provide a convenient and comprehensive summary of the information contained in all the original variables in the study and cannot provide an adequate dimensionality reduction.
Even so, there was still some value to the PCA. Although the interpretation of principal components is very subjective in na-ture, it appears that the first principal component may be an indi-cation of sensory properties (as indicated by the eigenvector coefficients, the greatest of which had absolute values >0.41), while the second principal component might be an indication of nutritional characteristics (the largest eigenvector coefficients had absolute values >0.44), and the third principal component may be an indication of age and ethnicity (the eigenvectors had absolute values >0.50). Additionally, a benefit to using PCA to summarize multivariate data is the ability to identify curvature, outliers, and clustering through examination of low-dimensional scatterplots of the calculated principal component scores. Using this approach, very little was found. But it does appear that some delineation between the first two PC can be achieved according to fiber level (Fig. 10C), which lends support to the supposition that the second PC may be an indication of tortilla nutrient content or it may, in fact, be an indication of a consumer’s approach to daily food choices.
Partial least squares regression. To further explore the multi-dimensional nature of the data set, PLS regression was pursued to determine whether overall acceptability could be predicted as a function of all of the other sensory and survey data. PLS is a re-gression technique that relates a set of predictors to response vari-ables. PLS reduces the predictor variables to a set of uncorrelated linear components based on the covariance between predictors and the response variable, then performs least squares regression on these components (Esbensen 2004). As shown in the loading plot (Fig. 11A), all of the sensory variables had a high influence on the value of overall acceptability. Additionally, most of the survey variables appeared to be influential, especially gender, brand, and dietary fiber. Ethnicity and frequency, however, ap-peared to have very small influence on overall acceptability. Even so, flavor and texture were still the most influential of all of the
variables in the study. A model selection plot (Fig. 11B) indicates that a PLS model with three components should adequately pre-dict overall acceptability with a cross-validated model R2 = 0.655 (with a predicted R2 = 0.636). Parameter estimates for the linear combinations are given in Table VI. It appears that PLS does per-form much better at reducing the dimensionality of the multivari-ate data set in this study than PCA. Additionally, all 12 variables are included in this model, which thus more fully captures the information provided in the data.
Even so, considering all of the multivariate data together, it ap-pears that the polynomial regression results relating flavor and texture to overall acceptability may be almost equivalent to the PLS results; the coefficient of determination was only slightly
Fig. 10. Principal component analysis of all multivariate data. A, Loading plot indicates comparative influence of each variable; sensory data appearsmore influential than survey data. B, Error plot indicates that at least threecomponents are required. C, Component plot indicates possible segre-gation in multidimensional space according to daily effort to include fiberin diet.
Fig. 9. Multiple linear regression using flavor and texture as predictor vari-ables for overall acceptability appears to produce nearly similar predictioncapability as PLS regression (R2 = 0.61; F = 398.65; P < 0.0001).
728 CEREAL CHEMISTRY
lower (R2 = 0.610 vs. 0.655), and was much less computationally intensive to determine.
Implications of the Study The approaches used in this study are appropriate for all types
of product development endeavors. Combining sensory informa-tion with demographic and behavioral data provides a more com-
plete picture of the background and potential perceptions that may influence a given product’s acceptability. Examining the collected data with both a univariate as well as a multivariate approach allows the entire data set to be examined for trends but also allows specific relationships to be discerned.
Interestingly, in this study, when examining all of the data si-multaneously, flavor and texture were the overriding factors that affected the overall acceptability, which has also been reported elsewhere. For other food products, however, acceptability may be affected by other sensory as well as demographic and behav-ioral factors, and thus need to be quantified and analyzed. Other issues that may come into play also include why a given person makes certain food choices and the level of their knowledge con-cerning what constitutes a healthier choice.
CONCLUSIONS
Potential market segments for these functional tortillas are older consumers (41+), women, Hispanics and Asians, and fre-quent consumers. Additional formulation experiments should include descriptive analysis followed by consumer acceptance studies with a large group including the market niches identified from this investigation study. The effect of providing product information before conducting sensory trials should be evaluated as well.
TABLE VPrincipal Component Analysis (PCA) Used to Summarize All Multivariate Data
Fig. 11. Partial least squares regression using all multivariate data as pre-dictor variables for overall acceptability. A, Loading plot indicates sensory data is somewhat more influential than survey data. B, Model selectionplot indicates that a three-component model will optimally predict overallacceptability (R2 = 0.655; F = 325.3; P < 0.0001) for this data set.
TABLE VI Partial Least Squares (PLS) Regression Results Using
a Three-Component Modela
Parameter Estimate Value
Constant 0.6941 Age 0.1195 Gender –0.1033 Ethnicity 0.0030 Frequency –0.1016 Fiber –0.1980 Lowfat –0.0099 Dietary fiber 0.0606 Brand 0.0171 Apprearance 0.1025 Color 0.1710 Flavor 0.3439 Texture 0.3462
a Model incorporates all the multivariate data as predictor variables for overall acceptability (R2 = 0.655; F = 325.3; P < 0.0001).
Vol. 85, No. 6, 2008 729
The tortillas made with whole barley flour developed for this experiment contained 3.3 g of total fat/RACC. Reformulating the tortillas to ≤3 g fat/RACC would provide an opportunity to utilize the FDA-approved health claim relating to soluble fiber from oats, barley, or psyllium, and reduced risk of heart disease (21 CFR 101.81).
The consistently higher score given to the commercial hand-made flour tortilla (HM) also suggests a careful reformulation of the WBF tortillas to match them more closely to the HM tortilla without sacrificing the health benefits. This approach could im-prove acceptance among consumers who may not have a high awareness of the effect of fiber and low-fat content in their diet. The main differences among the two products are whole grain content and fat type.
Responses given to nutrition-related questions seem to indicate that the majority of the consumers involved in the study, although apparently aware of the importance of boosting daily consump-tion of fiber in the diet (57%), did not make an effort when it came to consuming tortillas containing ≥1 g of fiber/serving.
It appears that a response surface relating flavor and texture to overall acceptability produces similar results to the multivariate PLS results in terms of predicting overall acceptability. This methodology is much simpler in terms of analysis as well as in-terpretation, as flavor and texture are two of the primary parame-ters that consumers use when considering which foods to purchase and consume.
ACKNOWLEDGMENTS
We would like to extend appreciation to the California Agriculture Re-search Initiative for funds provided to support this project.
LITERATURE CITED
Arndt, E. A. 2006. Whole-grain barley for today’s health and wellness needs. Cereal Foods World 51:20-22.
Bello, A. B., Serna-Saldivar, S. O., Waniska, R. D., and Rooney, L. W. 1991. Methods to prepare and evaluate wheat tortillas. Cereal Foods World 36:315-321.
Childs, N. M., and Poryzees, G. H. 1997. Foods that help prevent disease: Consumer attitudes and public policy implications. J. Consumer Mar-keting 14:433-447.
Cho, H. Y., Chung, S. J., Kim, H. S., and Kim, K. O. 2005. Effect of sensory characteristics and non-sensory factors on consumer liking of various canned tea products. J. Food Sci. 70:S532-S538.
Dietary Guidelines for Americans. Available online at www.nal.usda.gov/ fnic/Dietary/9dietgui.htm. USDA: Washington, DC.
Esbensen, K. H. 2004. Multivariate Data Analysis—In Practice. CAMP Process AS: Oslo, Norway.
Everitt, B. S., and Dunn, G. 1991. Applied Multivariate Data Analysis. Halsted Press: New York.
FDA and DHHS. 2005. Food labeling. Health claims: Soluble dietary fiber from certain foods and coronary heart disease. 21 CFR 101.81. Code of Federal Regulations. Published online at www.gpoaccess.gov/ cfr/index.html. GPO: Washington, DC
Friend, C. P., Serna-Saldivar, S. O., Waniska, R. D., and Rooney, L. W. 1992. Increasing the fiber content of wheat tortilla. Cereal Foods World 37:325-328.
Helgesen, H., Solheim, R., and Naes, T. 1997. Consumer preference mapping of dry fermented lamb sausages. Food Qual. Pref. 8:97-109.
Kälviäinen, N., Salovaara, H., and Tuorila, H. 2002. Sensory attributes and preference mapping of muesli oat flakes. J. Food Sci. 67:455-460.
Kantor, L. S., Variyam, J. N., Allshouse, J. E., Putnam, J. J., and Lin, B. 2001. Choose a variety of grains daily, especially whole grains: A chal-lenge for consumers. J. Nutr. 131:473S-486S.
Kihlberg, I., and Risvik, E. 2007. Consumers of organic foods—Value segments and liking of bread. Food Qual. Pref. 18:471-481.
Lang, R., and Jebb, S. A. 2003. Who consumes whole grains and how much? Proc. Nutr. Soc. 62:123-127.
Lockei, S., Lyons, K., Lawrence, G., and Grice, J. 2004. Choosing organ-ics: A path analysis of factors underlying the selection of organic food among Australian consumers. Appetite 43:135-146.
Mao, Y., and Flores, R. A. 2001. Mechanical starch damage effects on wheat flour tortilla texture. Cereal Chem. 78:286-293.
Marascuilo, L. A., and Levin, J. R. 1983. Multivariate Statistics in the Social Sciences. Brooks/Cole Publishing: Monterey, CA.
Martens, H., and Martens, M. 2001. Multivariate Analysis of Quality. An Introduction. John Wiley & Sons: Chichester, UK.
Meullenet, J. E., Xiong, R., Monsoor, M. A., Bellman-Homer, T., Dias, P., Zvanovic, S., Fromm, H., and Liu, Z. 2002. Preference mapping of commercial toasted white corn tortilla chips. J. Food Sci. 67:1950-1957.
Mialon, V. S., Clark, M. R., Leppard, P. I., and Cox, D. N. 2002. The effect of dietary fibre information on consumer responses to breads and English muffins: A cross-cultural study. Food Qual. Pref. 13:1-12.
Rieckhoff, D., Trautwein, E. A., Mälkki, Y., and Erbersdobler, H. F. 1999. Effects of different cereal fibers on cholesterol and bile acid metabo-lism in the Syrian golden hamster. Cereal Chem. 76:788-795.
Rosentrater, K. A., Flores, R. A., Richard, T. L., and Bern, C. J. 1999. Physical and nutritional properties of corn masa by-product streams. Appl. Eng. Agric. 15:515-523.
Rosentrater, K. A. 2004. Analysis of factors affecting corn masa byprod-uct generation. ASAE/CSAE: St. Joseph, MI.
Santa Cruz, M. J., Garitta, L. V., and Hough, G. 2003. Note: Relation-ships of consumer acceptability and sensory attributes of Yerba Mate (llex Paraguariensis St. Hilarie) using preference mapping. Food Sci. Technol. Int. 9:347-352.
Serna-Saldivar, S. O., Rooney, L. W., and Waniska, R. D. 1988. Wheat flour tortilla production. Cereal Foods World 33:855.
Thybo, A. K., Kühn, B. F., and Martens, H. 2003. Explaining Danish children’s preferences for apples using instrumental, sensory, and demographic/behavioural data. Food Qual. Pref. 15:53-63.
[Received October 14, 2007. Accepted March 5, 2008.]
