Sunteți pe pagina 1din 7

Food Quality and Preference 11 (2000) 121127

www.elsevier.com/locate/foodqual

Skin sensory performance of individual personal care ingredients and marketed personal care products
Vincent A.L. Wortel*, Johann W. Wiechers
Uniqema, PO Box 2, 2800 AA Gouda, The Netherlands Received 20 October 1998; received in revised form 21 May 1999; accepted 22 July 1999

Abstract Cosmetic products and personal care ingredients were evaluated by a Quantitative Descriptive Analysis1. Three skin sensory panels performed the evaluation of 10 marketed products and 55 personal care ingredients. A detailed exploration of the data was performed in order to understand the variances in the data. Subsequently, a suitable data pre-treatment was applied to reduce the variance in the data. Principal component analysis (PCA) was performed on both data sets (personal care ingredients and marketed products) individually. In addition, the ingredients were projected in the marketed product PCA-space. This projection showed clear overlap between personal care ingredients and some marketed products. # 1999 Elsevier Science Ltd. All rights reserved.

1. Introduction The skin sensory performance of personal care products is an important factor for the sales potential of any cosmetic product. Ecacy claims can be made for these products based on clinical trials. However, if the clinical activity in one way or another can be supported by sensorial perception, this personal care product is more likely to be successful in the market place. Skin sensory characteristics of a product are the result of the ingredients in the marketed products as well as the physical form in which they have been formulated. The objective of this study was to identify the sensory characteristics of a large series of personal care ingredients and to compare these with a selection of marketed skin care products. In this way, it should be possible to identify those characteristics that are common to both ingredients and nalised products and those relevant to only one of the two groups. The sensory technique used was Quantitative Descriptive Analysis1 (QDA), where people in a panel are carefully selected based on their capability to dierentiate between products and to verbalise these dierences. During intensive training sessions, this panel develops its own vocabulary and test methodology that

is most appropriate to describe and dierentiate the products. Dierent panels therefore evaluated personal care ingredients and marketed products. 2. Materials and methods Quantitative Descriptive Analysis (QDA1) (Meilgaard, Civille & Carr), developed by the Tragon Corporation (Redwood City, CA), is based on the capability of a panellist to verbalise perceptions of a product in a reliable manner. The method embodies a formal screening and training procedure, development and use of a sensory language and the scoring of products on repeated trials to obtain a complete, quantitative description (Stone, 1992). Two QDA1 panels evaluated 55 personal care ingredients. The subdivision of the ingredients over two panels was necessary to reduce the time required to evaluate all the products. When a study takes too long, subjects tend to lose their interest, resulting in reduced performance and/or dropouts. Nineteen personal care ingredients were evaluated by both panels in order to check consistency to allow the datasets of the two panels to be merged. Evaluation took place on the subject's left and right forearms, on both the inside and the outside. The study was conducted in two stages. The purpose of the rst stage of the study was to generate phrases and words that adequately described the sensory

* Corresponding author. Fax: +31-182-542-747. E-mail address: vincent.wortel@uniqema.com (V.A.L. Wortel).

0950-3293/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved. PII: S0950-3293(99)00057-9

122

V.A.L. Wortel, J.W. Wiechers / Food Quality and Preference 11 (2000) 121127

characteristics of the neat personal care ingredients. This process was facilitated by a panel leader, who was not part of the judging. The panels developed their own list of attributes and were subsequently trained to apply the scorecard. Each panel consisted of 13 women who had been selected for their high degree of sensory acuity. Both panels were given a subset of the test products reecting the dierences present between these products. Independently, they developed in their own words a language to describe the sensory characteristics of these products. Panel discussions focused on reaching agreement on the number and type of attributes necessary to describe the products, their denition and the sequence in which they should be tested. The panel leader left both panels free to use the terminology they wanted to. In cases where no clear decision could be made, the panel leader suggested a certain wording to ensure consistency between the two panels. Following this phase, both panels were individually trained to use scorecards properly. the intensity of each attribute was scored by marking a point on a 6-inch horizontal line, anchored one-half inch from each end with directional terms (e.g. weak/strong, slight/very) that best represented the intensity of that attribute. The score was calculated by the distance times 10, which resulted in a scale from 0 to 60. The panellists were checked on a regular basis by the panel leader to ensure consistency and retrained on specic attributes if necessary. This extensive training on sensory perceptions and scaling allowed the use of these panels as ne-tuned assessment ``instruments''. Interestingly, both ingredient panels identied independently the same 31 attributes to fully describe these products in the second stage of the study. These attributes could be grouped in three classes: appearance, feel and sensation. Relevant attributes were evaluated before rubbing (BR), during rubbing (DR) and after rubbing (AR). The non-formulated personal care ingredients were evaluated in the same way as in the rst phase as the study. Another QDA1 panel was trained for the evaluation of 10 marketed skin care products. This panel identied 41 attributes to describe the products. These attributes could be grouped in four classes (appearance, aroma, feel and sensation). Relevant attributes were evaluated BR, DR and AR and 1 h after application. The products were scored using scorecards as described above. Both the ingredients and marketed products were evaluated three times on three dierent days according to an incomplete block design to balance the serving order of the products. 2.1. Data visualisation To explore the data, two visualisation tools were applied: a matrix plot and a multiple box plot. A matrix

plot is a collection of scatter plots of all attribute combinations. The attribute names are written in a diagonal line from the upper left corner to the lower right corner. The data points were presented as dots in the cross-related boxes. Fig. 1. shows an ingredient matrix plot of four attributes: thickness (BR), shininess (DR), absorbency (DR) and thickness (DR). For instance, the plot in the upper right corner showed the strong positive correlation between the thickness before and during rubbing. A box plot (Massart, Vandeginste, Buydens, de Jong, Lewi, & Smeyers-Verbeke, 1997) is a univariate presentation of the data distribution. The length of the box represents the range of the middle 50th percent of the data. A horizontal line in the centre is drawn at the median (50th percentile). A vertical line is drawn from the box to the farthest data point. Potential outliers are displayed as separate dots if the distance between these data points and the box is larger than a selected value (normally 1.5 quartile). Fig. 2 shows a box plot of the thickness (DR) attribute of the 10 marketed products. For product 1, the distance between the two separate data points (value approximately 50) and the box is too large. Consequently, these points should be considered as outliers. The complete data set could be quickly visualised by combining all attributes in a multiple box plot. This was very useful during the exploration phase of the data analysis. 2.2. Principal component analysis (PCA) General univariate statistics were applied to describe the data. Furthermore, PCA was performed; this

Fig. 1. Matrix plot of four attributes for the ingredients.

V.A.L. Wortel, J.W. Wiechers / Food Quality and Preference 11 (2000) 121127

123

Fig. 2. Box plot of thickness (DR) of the marketed products. Fig. 3. The variation of the subjects for the thickness (DR).

multivariate technique is frequently used in chemometric applications. Especially when the number of attributes is high, PCA has proven to be a very ecient tool to reduce the number of dimensions. In sensory data, several attributes could be present that contain essentially the same information. These attributes could be combined to a new latent variable (principal component or PC). Moreover, the correlation between attributes is utilised, which results in a stronger comparative method in relation with other univariate methods. 3. Results 3.1. Data exploration and data pre-treatment The matrix plot generally is a very powerful tool to detect a trend or correlation between attributes. Unfortunately, due to the combination of a large number of observations and a large variance in this particular study, these trends were dicult to observe. Therefore, the data was rst studied in detail by multiple box plots. The 41 box plots (formulations set) were all plotted on equal y-axis scales to facilitate the comparison of the dierent plots. This showed distinct dierences between products for some descriptors. In contrast, other descriptors included too much variance. In sensory data, a certain amount of variance is expected; however, the variance was much larger than expected. Therefore, the data was analysed in more detail, up to the individual data point to check the repeatability of the panellists. To visualise the repeatability, the range (maximum minimum score) was calculated for each panellist (per product) and displayed in scatter plots. The range of the thickness before rubbing is displayed in Fig. 3. Each dot represents the range of one product. The precision of some panellists for some products was superb; a range of approximately 5 points on a scale of 60 was much better than expected. Unfortunately, the performance of other panellists was less good. These panellists had differences in replicate observations of 3045 points. Such

dierences caused more variance in the data set than was acceptable for proper product dierentiation. About 10% of the data had a range exceeding 30 points. In a substantial fraction of these cases, panellists evaluated two out of three replicates reasonably similarly and one observation was totally dierent. The similar observations showed that panellists were able to do a reproducible sensory measurement. Consequently, this suggested a systematic error (e.g. writing errors, score card exchange and sample swapping) but there was no evidence for this in the study documentation and therefore the fundamental cause of this variance was not found. Using the mathematical mean of the replicates would have reduced this eect. However, average values will level out the dierences between products. Therefore, instead of the mathematical mean, we used the median value that is less sensitive than the mean to the presence of outliers. The data was compressed further by taking the mean of the medians of all panellists, which results in one observation per product. 3.2. Ingredients PCA A PCA was performed on the ingredient data set (55 products 31 attributes). The attributes were not scaled because all attributes used the same 60-point scale. In addition, when the attribute scores are very low (with small standard deviations), a scaling procedure would create inuential attributes. Cross-validation (11 segments of ve products) was performed because the number of available samples was too low for a test set validation. The rst principal component explained 68% of the variance. The loading plot (Fig. 4) of this PCA clearly shows that the rst PC is mainly related to thickness on one side of the scale and absorbency on the other. This relationship between these attributes was also evident in the matrix plot (Fig. 1). The second PC explained 12% of the variance; the cloudiness, spreadability and the odour of the ingredients contributed greatly to this factor. The total amount of variance that was explained at three PCs was 85%. The score plot

124

V.A.L. Wortel, J.W. Wiechers / Food Quality and Preference 11 (2000) 121127

Fig. 4. 2D loading plot of ingredients.

(Fig. 5) showed an area with most of the ingredients; however, three ingredients with a higher spreadability performance are in dierent areas of the score map. 3.3. Marketed products PCA The marketed products set (10 products 41 attributes) was used for a second PCA. The number of samples was small and therefore we performed a full cross-validation. The colour had an unreasonable strong inuence on the second PC. This was mainly caused by the lack of a good distribution of the data. In addition, from an ingredient supplier's point-of-view, colour was not interesting to us and we therefore decided to remove the colour attribute and recalculate the PCA. The rst principal component clearly showed product absorbency versus thickness (Fig. 6). The second principal component was mainly related with the aroma of the product. The PCA explained 72% by two principal components. Thickness and absorbency were predominantly present in the rst principal component and aroma in the second principal component. This model was less powerful in comparison with the ingredient model as it was based on only 10 products. The marketed products in the score plot (Fig. 7) are shaped like a triangle, the corners represented by the absorbency, thickness and aroma. 4. Discussion 4.1. Projection of ingredients in marketed products PCA The individual PCAs of the ingredients and the marketed products resulted in an accurate overview of the

Fig. 5. 2D score plot of ingredients.

sensory attributes. Unfortunately, the ingredients scores could not be projected in the marketed products PCA. However, this projection would be very interesting from a marketing point of view as it would identify which ingredients are suitable for which kind of market product. In order to project the ingredients in the marketed products PC-space, both sets should have identical attributes. However, the marketed products panel utilised more and dierent descriptors in comparison with the ingredient panel. A set of 20 overlapping descriptors was identied, and these were applied by both the ingredient and the marketed products panels. Fortunately, the most important PCA attributes from both sets (ingredient and marketed products) were present in the overlapping descriptor set. The PCA of the marketed products was recalculated based on the overlapping set of attributes. As expected, the rst principal component was predominantly based on the absorbency and thickness of the product. The second principal

V.A.L. Wortel, J.W. Wiechers / Food Quality and Preference 11 (2000) 121127

125

Fig. 6. 2D loading plot of marketed products.

Fig. 8. 2D loading plot overlapping attributes (ingredients + marketed products). Fig. 7. 2D score plot of marketed products.

component (Fig. 8) was based on shininess (BR and DR), sheen, absorbency (AR and DR) and thickness (BR). In contrast to the other models, the moisturising and the silkiness played a less important role. The marketed products were nearly ideally spread in the score plot (Fig. 9). Four products were located in the direction of both ends of the rst and second principal component. The other products were somewhere within this `diamond'-shaped area. The shape of the cluster of the ingredients cluster was recognisably comparable with the shape in Fig. 5. The strong overlap between the ingredients and marketed products at the left-hand side of the plot indicated signicant similarities between these products. In contrast, no ingredients were present that were overlapping the products in the fourth quadrant.

Fig. 9. Projection of ingredients in a 2D score plot of marketed products.

126

V.A.L. Wortel, J.W. Wiechers / Food Quality and Preference 11 (2000) 121127

The marketed products were selected based on several criteria to include a wide range of products, such as the region where the product was sold. Interestingly, the easily absorbed personal care ingredients were projected near the Asian products, which are renowned for their light feel characteristics. 5. Conclusion The evaluation of sensory data of cosmetic ingredients and marketed products was complex because of the large number of descriptors and observations. However, when using the PCA technique, a good evaluation of both ingredients and marketed products was possible. The example with this data set clearly showed the importance of data exploration to understand the data. This is an essential step prior to the data pretreatment. In the skin sensory data, the variance was decreased by using medians, based on interpretation of the underlying causes of the variance. The marketed products were described by more and other descriptors than the cosmetic ingredients. A comparison of these sets was only possible by using a set of descriptors used by both panels. The overlapping descriptors of the cosmetic ingredients were projected in the PCA of the marketed products using the same overlapping descriptor set. This gives a good indication how the ingredients t into the `market space'. Appendix: List of ingredients sensory attributes A.1. Before rubbing  Appearance: Look at product after applied to skin, then evaluate. Thickness: Impression of the thickness of the product ranging from thin, watery, runny, or to thick, gooey, drippy, sti, or glue or honey-like; forms peaks, sap-like. Cloudiness: Degree to which the product looks cloudy, ranging from clear, colourless, or transparent to cloudy. Shiny: Degree to which the product looks shiny, oily, glossy, or sheen (slightlyvery).  Sensation: Evaluate the eect of product on your skin before rubbing. Tingling: Degree to which the product causes the skin to feel tingly or prickly (slightly very). Cooling: Degree to which the product causes the skin to feel cool or cold (slightlyvery).

A.2. During rubbing  Appearance: Rub product 10 times in a circular motion over a 4 inch diameter on the forearm. Evaluate while rubbing. Cloudiness: Degree to which the product looks cloudy, ranging from clear, colourless, or transparent to cloudy. Shiny: Degree to which the product looks shiny, oily, glossy, or sheen.  Feel: Evaluate the feel of product while rubbing. Absorbency: Impression of the rate of absorption of the product into the skin. A quickly absorbing product will disappear, vanish, evaporate, or be invisible in a short amount of time (slowlyquickly). Thickness: Impression of the thickness of the product ranging (thin-thick) from thin, watery, runny, or drippy to thick, gooey, sti, glue, honey-like or saplike. A thicker product would be more dense on the ngers and skin. Oily: Degree to which the product feels oily or slippery (slightlyvery). Greasy: Degree to which the product feels greasy, like Vaseline (slightlyvery). Sticky: Degree to which the product sticks to the skin, is gooey or tacky. A very sticky product will stick to the hairs on the arm (slightlyvery). Silky: Degree to which the product feels smooth, silky, or soft on the ngers and skin (slightlyvery). Moisturising: Degree to which the product causes a moisturising feeling on the skin; soaks up moisture while rubbing (a littlea lot).  Sensation: Evaluate the eect of product on your skin while rubbing. Cooling: Degree to which the product causes the skin to feel cool or cold while rubbing (slightlyvery). Warming: Degree to which the product causes the skin to feel warm while rubbing (slightlyvery). Tingling: Degree to which the product causes the skin to feel tingly or prickly while rubbing (slightlyvery).  Lift arm toward face and smell, then evaluate: Odour Amount of any type of odour, like vegetable oil, lard, grease, or shortening, sweet or maple (weakstrong).  Rub product down arm toward wrist, then evaluate:

V.A.L. Wortel, J.W. Wiechers / Food Quality and Preference 11 (2000) 121127

127

Spreadability:

Impression of the area that the product will cover when spread over the skin. A product that spreads easily will cover a large area of the skin, feel concentrated, and go a long way (a littlea lot)

Amount of residue

 Sensation: Cooling:

A.3. After rubbing  Appearance: Look at product after rubbing into skin, then evaluate. Absorbency: Impression of the rate of absorption of the product into the skin. A quickly absorbing product will disappear, vanish, evaporate, or be invisible in a short amount of time (slowlyquickly). Sheen: Degree to which the product has an oily or sheen appearance, looks shiny, glossy, is iridescent or glittery after rubbing (a littlea lot).  Feel: Evaluate the feel of product on your ngers after rubbing: Oily residue: Degree to which the ngers feel oily, slippery or greasy after rubbing (slightlyvery). Sticky residue: Degree to which the product leaves the skin feeling sticky, gooey or tacky after rubbing (slightlyvery). Silky residue: Degree to which the product feels smooth, silky, or soft on ngers after rubbing (slightlyvery).

Warming: Moisturising: Sti skin: Tingling: Coating:

Amount of residue the product leaves on the ngers; if you touched something, the product would get on it; doesn't wipe o easily; oily or waxy (a littlea lot). Evaluate the eect of product on your skin after rubbing: Degree to which the product leaves a cool or cooling sensation on the skin after rubbing (slightlyvery). Degree to which the product leaves a warm or warming sensation on the skin after rubbing (slightlyvery). Degree to which the product leaves a moisturising a feeling on the skin; soaks up moisture after rubbing (littlea lot). Degree to which skin feels sti or tight after rubbing (slightlyvery). Amount of tingling felt on the skin from the product after rubbing. Degree to which the product leaves an oily or waxy lm or coating on the skin after rubbing; waxy like a protective coating (a littlea lot).

References
Meilgaard, M., Civille, G. C., & Carr, B.T. Sensory evaluation techniques (2nd ed). New York: CRC Press. Stone, H. (1992). Quantitative descriptive analysis. In R. C. Hootman, Manual on descriptive analysis testing for sensory evaluation (pp. 15 21). Baltimore: (ASTM Manual Series: MNL 13). Massart, D. L., Vandeginste, G. G. M., Buydens, L. M. C., de Jong S., Lewi, P. J., & Smeyers-Verbeke, J. (1997). Handbook of chemometrics and qualimetrics: part A. Amsterdam: Elsevier.

S-ar putea să vă placă și