Ideal Profile Method: A comparison between rating and ranking technique

50 SCIENCE & TECHNOLOGY DEVELOPMENT JOURNAL:<br /> ENGINEERING & TECHNOLOGY, VOL 1, ISSUE 2, 2018<br /> <br /> <br /> <br /> Ideal Profile Method: A comparison between rating and<br /> ranking technique<br /> Nguyen Quang Phong, Nguyen Hoang Dzung *<br /> <br /> <br /> 1 on its information, manufacturers can modify their<br /> Abstract—Ideal profile method (IPM) has been current product or create a new product to maximize<br /> proved to be an effective and useful method in product sales and marketing. That is the reason why most of<br /> development. This method is similar to QDA® except manufacturers always try to identify the ideal<br /> that the samples are not rated by trained panelists but<br /> naïve consumers. However, the rating technique is product. There are two types of methods for that<br /> often found to be difficult for consumers. This study purpose: conventional method and rapid method.<br /> proposed a new variant of IPM using ranking Conventional method is the so-called external<br /> technique to facilitate the data collecting by naïve preference mapping (PrefMap). Its data is a<br /> consumers. The samples were five commercial lemon combination of hedonic data and descriptive data.<br /> green teas available in Vietnam market. The Hedonic data are obtained by consumers, whereas<br /> participants were bottled tea consumers who were<br /> randomly assigned into two groups of 60. The first descriptive data are obtained by a trained or expert<br /> group performed the conventional IPM (aka “IPM- panel. From statistical point of view, PrefMap<br /> QDA”) using rating technique, in which the samples focuses on the sensory profiles of products, then<br /> were presented in randomized monadic order and the hedonic data will be regressed on the sensory<br /> participants rated both the perceived and ideal dimensions. Ideal product will belong to the area<br /> intensities of the attributes on the 10-cm line scales. where a maximum proportion of consumers would<br /> The second group, on the other hand, performed the<br /> new variant of IPM (aka “IPM-RDA”) using ranking like [2, 3].<br /> technique, in which the participants ranked the whole Due to training session about the vocabulary and<br /> set of the products (ties allowed) for each attribute at the scale using, trained panel provides good quality<br /> the same time. An empty cup representing the ideal data. However, it can take few weeks to several<br /> sample was then inserted into the ranked set of months to complete a study. Because vocabulary<br /> products at the most suitable position depending on and scale using must be adapted on the new product<br /> the ideal intensity. The result showed that two product<br /> spaces were highly similar. However, compared to space when it is changed. Therefore, the<br /> IPM-QDA, IPM-RDA better improved the shortcoming of the conventional method is time<br /> discriminability, increased the consensus among the consuming [4].<br /> assessors and reduced the variability of ideal profile. Rapid method is in fact a group of methods that<br /> These findings indicated that ranking was more collect descriptive data using consumers, such as:<br /> efficient than rating in gathering descriptive data JAR, CATA, Napping, etc. Among these methods,<br /> using naïve consumers.<br /> Ideal Profile Method (IPM) has been widely used<br /> Index Terms—Confidence ellipses technique, Ideal by researchers and practitioners. From the<br /> Profile Method, Multiple Factor Analysis, Ranking perspective of the task, for each product, consumers<br /> technique, Rating technique. are asked to rate both perceived and ideal intensities<br /> on each attribute using a 10 cm line scale, before<br /> 1 INTRODUCTION rating their overall liking using a 9 point scale [5].<br /> <br /> I DEAL product is assumed as a product that As a result, three blocks of data are collected:<br /> would maximize the consumer appeal [1]. Based sensory profiles, ideal profiles, and the hedonic<br /> scores. This method provides the profile of the ideal<br /> <br /> Received: on August 17th, 2018, Accepted: October 07th,<br /> 2018, Published: November 30th, 2018<br /> Nguyen Quang Phong, Nguyen Hoang Dzung, Ho Chi Minh<br /> City University of Technology, District 10, Ho Chi Minh City,<br /> Vietnam<br /> (E-mail: nqphong28@gmail.com, dzung@hcmut.edu.vn).<br /> TẠP CHÍ PHÁT TRIỂN KHOA HỌC & CÔNG NGHỆ: 51<br /> KỸ THUẬT & CÔNG NGHỆ, TẬP 1, SỐ 2, 2018<br /> <br /> product and the relative position of the real products sample consistency. At the beginning of the test, 20<br /> compared to the ideal [6]. milliliters of each sample were dispensed into<br /> By using consumers to profile products without lidded transparent plastic cups and stored in<br /> training session, IPM as well as other consumer- refrigerator for at least five minutes before serving<br /> based methods are less time consuming. In addition, to consumers. The maximum evaluation time was<br /> when hedonic and descriptive descriptions are 10 minutes and new samples were supplied if<br /> obtained from the same consumers, the link necessary to make sure that the serving temperature<br /> between the appreciation to the sensory perception was 5-10oC. The samples were presented to<br /> of the products for each consumer is more directly consumers coded with 3-digit random numbers,<br /> [7]. following Williams’ Latin square design.<br /> However, in the conventional IPM which is<br /> 2.2 Participants<br /> based on Quantitative Descriptive Analysis-QDA®,<br /> rating technique is applied to profile products. The Participants were recruited from the consumer<br /> limitation of this method (aka IPM-QDA) could be database of the research team. They were bottled tea<br /> that the products are evaluated independently and consumers who consumed bottled lemon green teas<br /> rating task is difficult to consumers, especially at least once a week. Most of them were students at<br /> when the number of attributes is high [6]. In HCMC University of Technology who were aged<br /> recently studies, several methods are developed to between 18 and 23 years old.<br /> identify the ideal product in which QDA® is 2.3 Procedure<br /> replaced by other consumer profiling 2.3.1 Study 1: Recruiting panels<br /> methodologies. Ares et al. applied Napping®, Preference of consumers is an important issue<br /> Check-All-That-Apply (aka CATA) in comparison that should be concerned when comparing their<br /> with intensity scale [8]. Brard et al. proposed IPaM ideal products. That is the reason why two<br /> as a variant of IPM which is based on Pairwise independent panels should be similar in preference<br /> Comparisons to apply to children panel [6]. Ruark before making a comparison between two methods<br /> et al. proposed CATA-I as a variant of IPM which (ie. IPM-QDA and IPM-RDA).<br /> is based on CATA to apply to adults panel [9]. In the study 1, 120 participants evaluated the<br /> In this study, we propose a new variant of IPM in overall liking of 5 products. Samples were<br /> which the ranking technique will be used instead of presented in sequential monadic order. The<br /> rating technique in the frame of IPM procedure. participants were asked to try samples and rating<br /> This method is so-called IPM-RDA which is based their overall liking scores on a 9-point hedonic<br /> on Ranking Descriptive Analysis [10]. The scale.<br /> objective of this study is making a comparison Hedonic data was collected in which liking<br /> between IPM-RDA and IPM-QDA in terms of scores were presented in a table crossing the<br /> gathering descriptive data for profiling both the real participants in rows and the products in columns. To<br /> and the ideal products using consumers. identify groups of consumers with different<br /> preference patterns, Principal Component Analysis<br /> 2 MATERIALS AND METHODS (PCA) and Hierarchical Clustering on Principle<br /> 2.1 Samples Components (HCPC) were performed. Then<br /> Five commercial teas were selected from local participants in each clusters were assigned into two<br /> supermarkets for testing. These samples were panels randomly and equally. Multiple Factor<br /> bottled lemon green teas corresponding to different Analysis (MFA) was performed to re-checking the<br /> brands in Vietnamese market, which were coded by similarity in preference of two panels.<br /> letters from A to E for confidentiality reasons. 2.3.2 Study 2: Comparing two methods<br /> Although the ingredients, sensory characteristics of To compare rating technique applied in IPM-<br /> these product were quite different, this was not a QDA and ranking technique applied in IPM-RDA,<br /> concern for the study. This highlights that the focus the same protocol was applied for each panels. In<br /> of this research was not on the particular results, but study 2, assessors were asked to profile both 5 real<br /> on the participants’ view on the methods. products and ideal product in their mind. The same<br /> All tea bottles were stored in refrigerator (0-4oC) list of descriptors was given to both of panels. Nine<br /> for at least 24 hours before testing session to ensure descriptors which attached their definitions were<br /> Color, Overall odor, Tea flavor, Lemon flavor,<br /> 52 SCIENCE & TECHNOLOGY DEVELOPMENT JOURNAL:<br /> ENGINEERING & TECHNOLOGY, VOL 1, ISSUE 2, 2018<br /> <br /> Sweetness, Sourness, Bitterness, Astringency and multivariate analysis (PCA, HCPC, and<br /> After-taste (cf. table 1). MFA) [11]. Similarity between the products<br /> In IPM-QDA method, samples were presented in spaces was evaluated using the RV<br /> sequential monadic order. For each product, coefficient between product configurations<br /> assessors rated both the perceived and ideal in the first two dimensions of the PCA [12].<br /> intensities of all attributes on the 10-cm line scales. - SensoMineR was used to perform the<br /> In QDA-RDA method, a whole set of five confidence ellipses technique [13].<br /> samples were presented with an empty cup Panellipse functions in SensoMineR was<br /> representing the ideal sample. Assessors were asked used to evaluate the sensory data quality of<br /> to try each of five samples and ranked them (ties each panels [6]. Panelmatch function in<br /> allowed) for each attribute. The ideal sample was SensoMineR was used to compare the the<br /> then inserted into the ranked set of products at the profiles provided by different panels [12].<br /> most suitable position depending on the ideal<br /> intensity. 3 RESULTS AND DISCUSSIONS<br /> The descriptive data provided by two panels were<br /> 3.1 Analyzing hedonic data<br /> collected into two blocks of data for each panel:<br /> - Sensory data including profiles of 5 real The results of cluster analysis using PCA and<br /> products was used to compare the quality of HCPC on overall liking scores were presented in<br /> descriptive data. The product maps were figure 1. The first plane of PCA factor map can<br /> compared by performing MFA. The sensory explain 50.77% of the total variance of the<br /> profiles quality was compared about the experimental data. Three identified consumer<br /> discriminability and the consensus among segments with different preference patterns were<br /> assessors by performing Confidence ellipses indicated: Cluster 1 was composed of 35 consumers<br /> technique for each panel. whose liking scores of 5 products were lower than<br /> - Ideal data includes not only the profiles of other clusters; Cluster 2 was composed of 47<br /> real products but also the profiles of ideal consumers who preferred A, B, and C; Cluster 3<br /> products given by each assessors. Ideal maps was composed of 38 consumers who preferred E<br /> were plotted together to compare the and D.<br /> variability of ideal profile by performing Variables<br /> Factorfactor<br /> map map (PCA)<br /> <br /> Confidence ellipses technique.<br /> 4<br /> 1.0<br /> <br /> <br /> <br /> <br /> cluster 1<br /> J.095<br /> Table 1. List of 9 descriptors using for cluster 2<br /> cluster 3<br /> 3<br /> <br /> <br /> <br /> <br /> J.094<br /> lemon green tea profiling J.022 J.097<br /> J.002<br /> A J.005<br /> 0.5<br /> <br /> <br /> <br /> <br /> J.112 A<br /> 2<br /> <br /> <br /> <br /> <br /> J.119<br /> Descriptor Definition J.080 J.069<br /> J.001<br /> J.032 B<br /> B<br /> Dim 2 (21.77%)<br /> <br /> <br /> <br /> <br /> J.064<br /> J.049<br /> J.038<br /> J.078 J.058<br /> J.061<br /> J.071<br /> CC<br /> Dim 2 (21.77%)<br /> <br /> <br /> <br /> <br /> J.100 J.060<br /> J.096 J.028<br /> Color How dark/light the color of tea is J.034 J.026<br /> J.031 J.011<br /> J.013<br /> J.101 J.106<br /> J.070J.075 J.107<br /> J.103<br /> J.009<br /> 1<br /> <br /> <br /> <br /> <br /> J.056<br /> J.044<br /> J.082 J.074 J.068J.030<br /> J.045 J.117 J.109<br /> J.019 J.104 J.052<br /> J.110<br /> J.083<br /> J.086J.046 J.053<br /> J.114 J.003<br /> J.113<br /> Overall Odor How strong/weak the overall odor in the J.067 J.015 J.024J.066<br /> J.042 J.057<br /> 0.0<br /> <br /> <br /> <br /> <br /> J.020J.035 J.118 J.055<br /> J.006<br /> J.010 J.099<br /> J.051<br /> J.089 J.039<br /> J.116<br /> 0<br /> <br /> <br /> <br /> <br /> J.098<br /> J.036<br /> J.041 J.077 J.065<br /> nose (orthonasal) is J.062<br /> J.063<br /> J.111<br /> J.047<br /> J.004J.108J.093<br /> J.105<br /> J.050 J.054<br /> J.072<br /> J.048<br /> J.059<br /> J.021<br /> J.018<br /> J.079<br /> J.007<br /> J.092<br /> J.088J.084<br /> J.085 J.027J.008<br /> J.073<br /> J.043<br /> Tea flavor How strong/weak the tea flavor in the J.040 J.102<br /> J.023 J.115<br /> -1<br /> <br /> <br /> <br /> <br /> J.017<br /> J.087<br /> J.090 J.029 J.037<br /> J.025 J.120<br /> J.081J.012<br /> J.076 J.014<br /> -0.5<br /> <br /> <br /> <br /> <br /> mouth and the nose (retronasal) is J.016<br /> J.091<br /> J.033 E E<br /> -2<br /> <br /> <br /> <br /> <br /> D<br /> D<br /> Lemon flavor How strong/weak the lemon flavor in the<br /> mouth and the nose (retronasal) is<br /> -1.0<br /> <br /> <br /> <br /> <br /> -4 -2 0 2<br /> Sweetness How strong/weak the sweetness on the<br /> tongue is (a) -1.5 -1.0 -0.5 1 (29.60%)<br /> Dim 0.0 0.5 1.0 1.5<br /> <br /> Sourness How strong/weak the sourness on the Variables factor map (PCA)<br /> Dim 1 (29.60%)<br /> tongue is<br /> 1.0<br /> <br /> <br /> <br /> <br /> Bitterness How strong/weak the bitterness on the<br /> tongue is<br /> 0.5<br /> <br /> <br /> <br /> <br /> A<br /> A<br /> B<br /> B<br /> Astringency How strong/weak the astringency in the C<br /> C<br /> Dim 2 (21.77%)<br /> <br /> <br /> <br /> <br /> mouth is<br /> 0.0<br /> <br /> <br /> <br /> <br /> After-taste How strong/weak the remained feeling in<br /> the mouth after tasting is<br /> -0.5<br /> <br /> <br /> <br /> <br /> E<br /> E<br /> D<br /> D<br /> -1.0<br /> <br /> <br /> <br /> <br /> -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5<br /> 2.4 Data analysis (b) Dim 1 (29.60%)<br /> <br /> All statistical analyses were performed using R Figure 1. The plots in the first and second dimensions of<br /> language. PCA and HCPC on hedonic data: (a) Representation of the<br /> - FactoMineR was used to perform the participants on the factor map, (b) Representation of the<br /> vectors of products on the correlation circle.<br /> TẠP CHÍ PHÁT TRIỂN KHOA HỌC & CÔNG NGHỆ: 53<br /> KỸ THUẬT & CÔNG NGHỆ, TẬP 1, SỐ 2, 2018<br /> <br /> The participants then were assigned randomly common (RV = 0.962). The representation of partial<br /> into two panels. The number of participants from individuals in figure 3a indicated that the structure<br /> each clusters was shown in table 2. of the product space established by the IPM-RDA<br /> Table 2. Number of consumers in each clusters is very close to the IPM-QDA s’ one. On the other<br /> and each panels<br /> hand, the representation of the vectors of<br /> Total<br /> Cluster 1 Cluster 2 Cluster 3<br /> by panel descriptors on correlation circle in figure 3b<br /> IPM-QDA panel 17 24 19 60<br /> indicated that two panels used attributes in the same<br /> ways. From these results, the sensory profiles<br /> IPM-RDA panel 18 23 19 60<br /> established by two panels were concluded similar.<br /> Total<br /> 35 47 38 120<br /> by cluster<br /> Individual factor map<br /> <br /> The results of comparing the preference of two IPM-QDA<br /> IPM-RDA<br /> <br /> <br /> <br /> <br /> 2<br /> panels using MFA was presented in figure 2. The D<br /> two first dimensions of the MFA can explain<br /> 60.87% of the total variance of the experimental<br /> <br /> <br /> <br /> Dim 2 (25.05%)<br /> <br /> 1<br /> data. Both groups share a large structure in common B<br /> (RV = 0.944). From these results, the preference<br /> <br /> <br /> 0<br /> patterns of two panel were concluded similar.<br /> E<br /> A C<br /> -1<br /> Individual factor map<br /> <br /> IPM-QDA -2 -1 0 1 2<br /> IPM-RDA<br /> E<br /> D<br /> Dim 1 (60.48%)<br /> (a)<br /> 1<br /> <br /> <br /> <br /> <br /> C<br /> Correlation circle<br /> Dim 2 (26.77%)<br /> <br /> 0<br /> <br /> <br /> <br /> <br /> IPM-QDA<br /> 1.0<br /> <br /> <br /> <br /> <br /> IPM-RDA After.taste<br /> After.taste<br /> Sourness<br /> Sourness Astringency<br /> Astringency After.taste<br /> After.taste<br /> -1<br /> <br /> <br /> <br /> <br /> B Sourness<br /> Sourness<br /> Lemon.flavor<br /> Lemon.flavor Bitterness<br /> Bitterness Tea.flavor<br /> 0.5<br /> <br /> <br /> <br /> <br /> Tea.flavor<br /> ColorLemon.flavor<br /> Color Lemon.flavor Astringency<br /> Astringency<br /> Dim 2 (25.05%)<br /> <br /> <br /> <br /> <br /> Color<br /> Color<br /> A Sweetness<br /> Sweetness Tea.flavor<br /> Tea.flavor<br /> -2<br /> <br /> <br /> <br /> <br /> Sweetness<br /> Sweetness Bitterness<br /> 0.0<br /> <br /> <br /> <br /> <br /> Overall.Odor<br /> Overall.Odor<br /> -3 -2 -1 0 1 Overall.Odor<br /> Overall.Odor<br /> -0.5<br /> <br /> <br /> <br /> <br /> Dim 1 (34.10%)<br /> Figure 2. The plots of products on the two first<br /> dimensions of MFA on hedonic data of two panels.<br /> -1.0<br /> <br /> <br /> <br /> <br /> Discussions: Although the consumers’ -1.0 -0.5 0.0 0.5 1.0<br /> preferences were not highly heterogeneous (cf.<br /> Dim 1 (60.48%)<br /> figure 1), the preference patterns of two panels were (b)<br /> highly similar (cf. figure 2). Because of the method Figure 3. The plots in the first and second dimensions of<br /> to recruiting panel, two independent panels in this MFA on sensory data: (a) Representation of the products<br /> study can be used to compare two methods. on the factor map, (b) Representation of the vectors of<br /> However, the number of consumers in each cluster descriptors on correlation circle.<br /> is too small that we cannot make comparisons in To assessing the quality of sensory data of each<br /> each clusters. In further studies, the sample size panels, 1000 virtual panels of 60 were generated<br /> could be enlarge to make the comparisons between using Bootstrap techniques. The p-value of 0.05<br /> homogenous groups of consumers. was set as the threshold above which a descriptor is<br /> not considered as discriminant according to AOV<br /> 3.2 Comparing sensory data model "descriptor=Product+Panelist". In figure 4,<br /> The results of MFA were presented in figure 3. each real product was circled by its confidence<br /> The two first dimensions of the MFA can explain ellipse generated by virtual panels. In figure 5, the<br /> 85.53% of the total variance of the experimental variability of each descriptor was drawn on the<br /> data. Both groups shared a large structure in correlation circle graph.<br /> 54 SCIENCE & TECHNOLOGY DEVELOPMENT JOURNAL:<br /> ENGINEERING & TECHNOLOGY, VOL 1, ISSUE 2, 2018<br /> <br /> As shown in figure 4, ellipses of products profiles Discussions: Ranking task in IPM-RDA method<br /> established by IPM-RDA panel did not overlap and helped to improve the discriminability, increase the<br /> we can consider that the products were well consensus among the assessors. In IPM-QDA<br /> discriminated by IPM-RDA panel, whereas the procedure, assessors evaluated one product at a time<br /> ellipses of products profiles established by IPM- on all attributes. In IPM-RDA procedure, a whole<br /> QDA panel (A, B, and E) overlapped and we cannot set of products were presented, assessors focused on<br /> affirm that the sensory evaluations are different. only one attribute at a time to rank them. It may lead<br /> These findings suggested a better discrimination by to the better using of descriptions by IPM-RDA<br /> the IPM-RDA panel. panel. We can notice that the vectors of descriptors<br /> As shown in figure 5, the variability between the used by IPM-QDA panel highly correlated together<br /> vectors of descriptors color, sweetness, lemon and correlated with dimension 1 (71.25%), whereas<br /> flavor, sourness, and overall odor established by the the vectors of descriptors used by IPM-RDA panel<br /> IPM-RDA panel was lower than which established dispersed and correlated with both dimension 1<br /> by IPM-RDA panel. The variability the vectors of (64.42%) and dimension 2 (23.19%). The IPM-<br /> descriptors tea flavor and astringency established QDA panel mainly discriminated products on the<br /> by two panels was high, as well as the variability first dimension which “tea related” attributes<br /> the vectors of descriptors bitterness established by towards the negative side and “non-tea related”<br /> the IPM-RDA panel was also high. With the p-value attributes towards the positive side. Moreover, the<br /> of 0.05 was set, the descriptor after-taste was variability between the vectors of descriptors used<br /> removed from the simulation of two both panels, the IPM-RDA was lower than which established by<br /> whereas the descriptor bitterness was removed from IPM-QDA panel. However, IPM-RDA is not<br /> the simulation of IPM-QDA panel. These findings suitable for a large number of products. It also<br /> suggested a higher consensus among assessors in requires careful temperature control or have<br /> IPM-RDA panel. persistent sensory characteristics [4].<br /> Confidence ellipses for the mean points Variables factor map (PCA)<br /> 1.0<br /> <br /> <br /> <br /> <br /> Overall.Odor<br /> Color<br /> Tea.flavor<br /> 4<br /> <br /> <br /> <br /> <br /> Lemon.flavor<br /> Sw eetness<br /> Sourness<br /> Astringency<br /> Sourness<br /> 0.5<br /> <br /> <br /> <br /> <br /> Astringency<br /> Tea.flavor Color<br /> D<br /> 2<br /> <br /> <br /> <br /> <br /> Overall.Odor Lemon.flavor<br /> Dim 2 (16.83%)<br /> <br /> <br /> <br /> <br /> Dim 2 (16.83%)<br /> <br /> <br /> <br /> <br /> Sweetness<br /> 0.0<br /> <br /> <br /> <br /> <br /> E<br /> 0<br /> <br /> <br /> <br /> <br /> C B<br /> A<br /> -0.5<br /> -2<br /> <br /> <br /> <br /> <br /> -1.0<br /> <br /> <br /> <br /> <br /> -4 -2 0 2 -1.0 -0.5 0.0 0.5 1.0<br /> <br /> Dim 1 (71.25%) Dim 1 (71.25%)<br /> (a) a)<br /> Confidence ellipses for the mean points Variables factor map (PCA)<br /> 1.0<br /> <br /> <br /> <br /> <br /> Color<br /> 4<br /> <br /> <br /> <br /> <br /> Tea.flavor<br /> Astringency<br /> Lemon.flavor<br /> Sw eetness<br /> Overall.Odor<br /> Sourness Sourness<br /> Tea.flavor<br /> Bitterness<br /> Astringency<br /> D<br /> 0.5<br /> <br /> <br /> <br /> <br /> Lemon.flavor<br /> 2<br /> <br /> <br /> <br /> <br /> Bitterness<br /> Dim 2 (23.49%)<br /> <br /> <br /> <br /> <br /> Dim 2 (23.49%)<br /> <br /> <br /> <br /> <br /> B Color<br /> 0.0<br /> <br /> <br /> <br /> <br /> Sweetness<br /> 0<br /> <br /> <br /> <br /> <br /> A<br /> <br /> Overall.Odor<br /> C<br /> E<br /> -0.5<br /> -2<br /> <br /> <br /> <br /> <br /> -1.0<br /> -4<br /> <br /> <br /> <br /> <br /> -4 -2 0 2 4 -1.0 -0.5 0.0 0.5 1.0<br /> <br /> Dim 1 (64.42%) Dim 1 (64.42%)<br /> (b) (b)<br /> Figure 4. Confidence regions around the real products: Figure 5. Confidence regions around the descriptors:<br /> (a) IPM-QDA panel, (b) IPM-RDA panel. (a) IPM-QDA panel, (b) IPM-RDA panel.<br /> TẠP CHÍ PHÁT TRIỂN KHOA HỌC & CÔNG NGHỆ: 55<br /> KỸ THUẬT & CÔNG NGHỆ, TẬP 1, SỐ 2, 2018<br /> <br /> the multiple ideal [7]. In comparison with CATA<br /> 3.3 Comparing ideal data<br /> with Ideal, nominal data collected in CATA-I was<br /> To compare the variability of ideal profile, ideal reported that have less power than ordinal data<br /> profiles of two panels were plotted together with collected in IPM-RDA. In comparison with<br /> profiles of real products (cf. figure 6). With respect Napping with Ideal, difficulty to interpret precisely<br /> to the MFA partial points representation, one ellipse the descriptions provided by the assessors in<br /> per product and per panel can be estimated. Napping [4]. In comparison with Pairwise<br /> The two first dimensions of the MFA can explain Comparison with Ideal, the experiment design in<br /> 82.69% of the total variance of the experimental IPM-RDA was not complex because all samples<br /> data. The structure of product spaces established by were ranked at a time. However, the limitation of<br /> two panels was similar in common. The ideal the IPM-RDA is also the ordinal data collected. In<br /> product was near the product D which is the most this study, the data collected from IPM-RDA was<br /> appreciated product of two panel (cf. table 3). analysis as numeric data instead of ordinal data as<br /> The ellipses related to the ideal products of IPM- its nature. In further studies, IPM-RDA data would<br /> RDA panel was smaller than which of IPM-QDA. be treated as an ordinal data and the data should be<br /> In other word, the variability of the description of checked the consistency before using for products<br /> the ideal product given by IPM-RDA panel is improvement and optimization.<br /> smaller than IPM-QDA panel.<br /> Confidence ellipses for the partial points 4 CONCLUSION<br /> By comparing IPM-RDA and IPM-QDA, the<br /> results showed that two product spaces obtained by<br /> 3<br /> <br /> <br /> <br /> <br /> Ideal<br /> the two methods were highly similar. Nevertheless,<br /> IPM-RDA was better in improving the<br /> 2<br /> <br /> <br /> <br /> <br /> discriminability among the products, in increasing<br /> the consensus among the assessors, and in reducing<br /> 1<br /> <br /> <br /> <br /> <br /> D<br /> Dim 2 (35.41%)<br /> <br /> <br /> <br /> <br /> the variability of the ideal profile. These findings<br /> implied that ranking technique might be more<br /> 0<br /> <br /> <br /> <br /> <br /> B<br /> E<br /> efficient than rating technique in gathering<br /> -1<br /> <br /> <br /> <br /> <br /> A<br /> C descriptive data using naïve consumers when<br /> applying IPM. IPM-RDA might be useful for<br /> -2<br /> <br /> <br /> <br /> <br /> collecting consumer data in the context of the final<br /> IPM-QDA<br /> IPM-RDA stage of product optimization process where a small<br /> -3 -2 -1 0 1 2 3 number prototypes were evaluated by a group of<br /> Dim 1 (47.28%) homogenous target consumers. For further studies,<br /> Figure 6. The plots in the first and second dimensions of this method can be applied not only in various<br /> MFA on hedonic data of two panels.