Food Quality and Preference

Food Quality and Preference 1989 1 (2) 69-73 (c) Longman Croup UK Ltd 1989 0950-3293/89/012040691/$03.50
Received 5 August 1988 Accepted 30 December 1988

C Guy, J R Piggott* and S Marie

University of Strathclyde, Food Science Division, Department of Bioscience and Biotechnology, 131 Albion Street, Glasgow G1 1SD, UK

Keywords: consumer; descriptive analysis; flavour profile generalised Procrustes analysis; sensory analysis; whisky

Consumer profiling of Scotch whisky

Abstract

Free-choice profiling has recently been used for collecting profile information on a variety of foods and beverages from laboratory panels of selected assessors. but the use of this method by consumers has not been reported. Free-choice profiling and subsequent generalised Procrustes analysis were therefore used for descriptive analysis of 8 whiskies by 100 consumers, to test the method under these conditions and to identify the dimensions used by consumers to discriminate between whiskies. The same whiskies were profiled by a trained panel using an agreed vocabulary. The consumer data showed that the panel could discriminate between the samples, and that the sample configuration could be interpreted in terms of the descriptors used by both panels and of the colour of the samples. Free-choice profiling was found to be a potentially useful method for consumer research, though difficulties were experienced in interpretation of the results.

Introduction

There are two widely used forms of conventional descriptive analysis, the Flavour Profile Method (Cairncross & Sjostrom 1950) and Quantitative Descriptive Analysis (Stone et al. 1974), but both have disadvantages especially for use with untrained consumers (Williams 1983; Powers, 1988). To overcome some of these problems, free choice profiling (FCP) was developed (Langron 1983; Arnold & Williams 1986). FCP assumes that subjects do not differ in the number and kind of sensory characteristics they perceive, but that they do differ in the way they label them. The assessors are therefore allowed to develop their own individual vocabularies, and use them to score a set of samples (Williams & Arnold 1985). Generalised Procrustes analysis (GPA) (Gower 1975; Langron 1983) followed by principal coordinate analyses (PCO) (Piggott & Sharman 1986) are then used to produce information on the inter-relationships between the samples and between the assessors, whereas in the case of QDA principal components analysis (PCA) (Piggott & Sharman 1986) is commonly used to produce a graphic representation of the samples.

GPA of a set of sensory data consists of three logical stages (Arnold & Williams 1986): firstly, the sample configurations are translated to zero means, to remove variations caused by assessors using different parts of the scales; secondly, the sample configurations are rotated or reflected to remove variations caused by different usages of the descriptive terms; thirdly, an isotropic scaling may be applied to remove variations caused by the assessors' use of different scale ranges. Finally, the sample configurations are commonly referred to principal axes by PCO. The results of this analysis can provide a consensus sample configuration, showing the relationships between the samples, and a table showing the residual distances of the assessors' transformed configurations from the consensus. Langron (1981) has shown how this can be used to provide an assessor plot, giving a graphic representation of the distances. Gains et al. (1988) have recently published a particularly clear explanation of GPA.

PCA calculates linear combinations of variables describing as much of the variance of the original data as possible, and so allows the original multidimensional matrix to be plotted in fewer dimensions without significant loss of information.

Very little training is required for FCP, and so to give satisfactory results assessors must simply be objective, be capable of using scales, and use the developed vocabulary consistently (Williams & Langron 1984). FCP has been used to study a variety of products (e.g. Williams & Langron 1984; Williams & Arnold 1985), but its use by a panel of consumers has not been reported. It should however be possible for consumers to use the method successfully, because of the absence of training required and the simplicity of the technique.

The aim of this experiment was two-fold: firstly, to determine whether FCP could be used by consumers in a consistent and meaningful way; secondly, to investigate the dimensions used by consumers to discriminate between whiskies and to identify consumer perceptions of smoothness and maturity in whisky, since these attributes are regarded as important for the marketing of whiskies (Morrice 1983). The conventionally trained whisky panel at Strathclyde University provided a trained panel analysis of the samples with which to compare the consumer results.

Method

Samples

The whisky brands used were: Claymore, Langs, Bells, Hundred Pipers, Bells 12 year old Connoisseur, Chivas Regal and Johnnie Walker Black Label. These whiskies were chosen because they are wellestablished brands in the UK, sell in large quantities and cover a range of sensory characteristics (Morrice 1983). An eighth sample was prepared by adding 0.05 % caramel (a colouring used in many blended whiskies) to Bells in order to determine how the appearance of the whisky affected perception.

Consumer panel procedure

One hundred consumers were recruited to the panel through personal contact by the authors and colleagues. This rather large number of assessors was used for two reasons. Firstly, previous reported applications of FCP have been by trained panels of typically 15 to 20 assessors, and it was not known how many untrained consumers would be required to provide a stable sample configuration. Secondly, a sufficiently large number of assessors was required to allow for a low return rate.

Assessment of the whiskies took place in two separate stages. At the first meetings of ten small groups in the homes of the authors and colleagues, the assessors were instructed to describe in their own words the appearance, aroma, and flavour of the eight samples. Samples were presented in 130 ml disposable plastic cups (Glacier, DRG Plastics) on white paper plates under domestic lighting to allow the colour of the samples to be assessed as consistently as possible. The assessors were allowed to add water to the whiskies throughout the procedure, but were asked to ensure that the same amount was added to each one. To investigate the consumers' perceptions of smoothness and maturity, these terms were added to each assessor's list where they were not already present. This is not a common procedure in FCP, but there is no fundamental reason why descriptive terms should not be suggested or provided. In this case, the interest was in the meaning the assessors attached to these terms. The ways in which the terms were used were considered to be satisfactory reflections of their meanings for the assessors. The eight whiskies were coded 1 to 8, and four samples selected from these were presented in the same way as above, and assessed by the consumers using their own vocabularies and 100 mm line scales (Land and Shepherd 1988) under test conditions. Thus the method for the future scoring of the whiskies was established and practised. The assessors were also asked to complete a short questionnaire to allow classification by socio-economic status (Chisnall 1985) and give details of their whisky preferences.

The second stage was carried out by the assessors individually in their own homes. Each assessor was provided with score sheets, disposable cups and 50 ml miniatures of each sample, coded A to H, and asked to assess the whiskies using their descriptive terms, in the same way as in the first stage. It was considered impracticable to attempt to control the conditions of assessment further. The score sheets were then returned to the University by post.

Trained panel procedure

The trained whisky panel at Strathclyde University assessed the same eight samples using an agreed vocabulary (Piggott and Canaway 1981; Piggott et al. 1985). The samples were diluted to 23 % (v/v) ethanol, coded with 3-digit random numbers, and presented to the assessors under red lighting in tulip-shaped nosing glasses similar to wine-tasting glasses (BS 5586:1978) but of 4 ounces nominal capacity, and covered by watch-glasses. These conditions were used to maximise the accuracy and precision of the assessment of odour, under laboratory conditions. The samples were assessed in duplicate on two consecutive days.

Data analysis

The FCP data for the complete panel were analysed by GPA using the GENSTAT macro described by Arnold (1987). Six further analyses were carried out of data from arbitrary subsets of 15 of the assessors, and two analyses of data from the 15 assessors at each extreme of the first axis of the assessor plot, using Procrustes-PC (see Research note). Finally, the first three axes of the sample configurations arising from the nine GPA of the consumer data were subjected to GPA by Procrustes-PC, to examine the stability of the configurations arising from the consumer panel. Panel means were calculated for the data from the trained panel, and the resulting matrix analysed by PCA.

Results

Consumer panel

Ninety-three usable sets of data were returned, and analysed by GPA. Some data sets were returned but were not usable either because they were incomplete or because the same scale had been used for more than one descriptive term. The response rate showed that the assessors did not find the task too difficult.

The first two axes of the assessor plot for all respondents are shown in Fig. 1. The group was homogeneous with no obvious clusters or outliers, showing that the consumers were perceiving broadly the same characteristics in the whiskies. No relationship was found between plot position and socio-economic status, age or stated preference. GPA of the nine sample configurations showed similar residual variances, and inspection of the sample configurations for the assessor groups at the extremes of the first axis in Fig. 1 showed considerable agreement. The sample configuration for all respondents was therefore examined without segmenting the panel. The first three principal axes accounted for 58 % of the variance, and are shown in Figs 2 and 3.

To investigate the terms used by the consumers and to aid interpretation of the axes, correlations (positive or negative) greater than 0.5 of all terms used by all assessors with the three axes were summarised and are shown in Table 1. A distinction was not made between positive and negative correlations because it was clear that some assessors were reversing some scales. This was especially evident in the case of 'negative' terms such as pale; some assessors were scoring these as the opposite 'positive' term, such as dark. Examination of Table 1 showed that terms related to colour or depth of colour were largely correlated with the first axis, though two other groups of terms were also related to this axis (flowery and warm). Many of the terms were related to the second axis, except the colour, flowery, warm and sweet groups. The third axis showed a broadly similar pattern to the second, though without the large number of correlations with maturity and including correlations with sweet.

As shown above, the first axis was apparently related to the colour of the samples, and was found to correlate (r = 0.994; p = < 0.001) with absorbance at 430 nm (Philp 1989). The second axis appeared to describe the malt content or maturity of the whiskies as it clearly separated the different types of blend. Deluxe blends (such as the Chivas Regal, Bells 12 year old and Black Label used here) tend to have a higher malt content and to contain older whiskies than standard blends (Morrice 1983). The meaning of the third axis was not clear from inspection of the samples, though it apparently represented another contrast between the deluxe blends and the other samples.

Trained panel

The first three components of a PCA of the trained panel data accounted for 74 % of variance, and are shown in Figs 4 and 5. The first component separated the deluxe and other whiskies, and was characterised by terms such as smooth, sweet, vanilla and malty having large positive loadings and terms such as soapy and oily having large negative loadings. The second component reflected a different contrast largely between the deluxe and other blends, and was characterised by terms such as fruity (eatery) and woody having positive loadings and buttery and grassy having negative loadings. The third component was characterised by solvent, sweet, floral and fruity (eatery) having large or moderate negative loadings.

Figure 1 (not available)

First (horizontal) and second (vertical) axes of principal coordinate analysis of assessor residual distances from consensus after GPA of consumer data (Langron, 1981)

Figure 2 (not available)

First (horizontal) and second (vertical) axes of sample configuration after GPA of consumer data

Figure 3 (not available)

Second (horizontal) and third (vertical) axes of sample configuration after GPA of consumer data

Table 1

Summary of terms correlated with each axis

Terms Axis 1 Axis 2 Axis 3

(a) (b) (c) (a) (b) (c) (a) (b) (c)
Maturity 12 5.15 28.97 11 8.87 49.89 5 3.76 21.15

Smooth 11 4.72 22.36 11 8.87 42.02 10 7.52 35.62

Mellow, velvety, rounded 0 0.00 0.00 8 6.45 55.08 7 5.26 44.92

Mild, creamy, rich, soft 1 0.43 4.37 7 5.65 57.42 5 3.76 38.20

Appearance terms (colour and depth) 108 46.35 82.02 7 5.65 10.00 6 4.51 7.98

Flowery, aromatic, fruity, estery and perfumed 9 3.86 41.82 2 1.61 17.40 5 3.76 40.74

Warm, fiery, burning 9 3.86 35.77 3 2.42 22.43 6 4.51 41.80

Harsh, rough, sharp, coarse, nippy, bitter 17 7.30 15.95 30 24.19 52.84 19 14.29 31.21

Malty 1 0.43 10.83 2 1.61 40.55 2 1.50 37.78

Nutty 1 0.43 21.61 1 0.81 40.70 1 0.75 37.69

Smoky, burnt wood, roasted 4 1.72 19.75 4 3.23 37.08 5 3.76 43.17

Peaty, heathery, earthy 5 2.15 19.71 9 7.26 66.54 2 1.50 13.75

Sweet, caramel, chocolate, treacle 9 3.86 25.23 3 2.42 15.82 12 9.02 58.95

(187) (98) (85)

Others 46 26 48

Total 233 124 133

Columns headed (a) show the frequency with which terms occurred; columns headed (b) show that frequency as a percentage of the axis total; and columns headed (c) show the percentage of the terms which were correlated with the axis.

Discussion

Inspection of Figs 3 and 4 shows obvious similarities between the second and third axes of the consumer panel sample configuration, and the first and second axes of the trained panel configuration. Correlation coefficients between the consumer panel's second axis and the trained panel's first axis, and the consumer panel's third axis and the trained panel's second axis, were both > 0.8. The sample of Bells with additional caramel clearly did not conform to this pattern of agreement. The trained panel assessed the samples under red lights to minimise colour differences, and so the sample relationships shown by the trained panel were due to odour only. The effect of adding caramel to the Bells sample therefore not only changed its appearance but also its perceived maturity, shown by its displacement on the second axis of the analysis of the consumer panel's data. There was therefore one pair of axes which the consumer panel characterised as maturity, and the trained panel as smooth, sweet, vanilla and malty. A further pair of axes was characterised by the consumer panel as smooth, mellow and sweet, and by the trained panel as estery and woody as opposed to buttery and grassy. The high degree of agreement between the sample spaces demonstrates that the panels perceived the major differences between the whiskies in a similar manner, despite the differences in conditions under which the assessments were carried out.

The difficulty found here in interpreting the results obtained by FCP has not been highlighted in the majority of published literature. Although FCP is claimed to reduce error by decreasing the influence of the panel leader and individual factors during the construction of a terms list, it would appear from the work reported here that this advantage is at least partially offset by the difficulty of interpretation. Clearly defined meanings of terms are not available and so the interpretation will depend to a large extent on the researcher's own interpretation of the terms. In other reports the range of words generated and the variety of word usage have not been so large and hence the difficulties may not have been so apparent. This may be because previous panels used a small number of participants who were familiar with sensory techniques, were trained in the generation of sensory terms, and could use scales logically. In this study, it was not practicable to examine individually the terms used by each assessor and their loadings or correlations with the axes as has been done in previously published work, because of the large number of assessors used. It appears that a smaller number of assessors could have been used, but care must be taken in reducing numbers in an untrained panel, because none of the subsets of 15 assessors gave the same sample configuration as the entire panel. When profile information is required directly from consumers, the use of a trained panel is clearly not an option. In this case, FCP can provide a viable method despite the difficulties experienced here.

Figure 4 (not available)

Trained panel sample scores plotted on first and second rotated principal components

Figure 5 (not available)

Trained panel sample scores plotted on second and third rotated principal components

Conclusion

The sample configuration obtained from free- choice profiling by consumers was very similar to that obtained from a trained consensus profiling panel. Therefore, with respect to the sample configuration, FCP by consumers has been shown to be capable of producing results of equivalent quality to consensus profiling by a trained panel. However, the large variety of terms generated and the diversity of their usage made interpretation of the consumers' word use extremely difficult and laborious. Interpretation of word usage in consensus profiling is considerably simpler. The sample plots and terms used by the consumers suggested that the dimensions used to discriminate between whiskies are colour, perceived maturity and smoothness, and smoothness and sweetness. Maturity was closely related to the probable age of the whiskies, and smoothness to characteristics described as mellow and sweet.

Acknowledgements

We would like to thank Chivas Brothers for supplying the samples, Gillian Arnold for the GPA, Maggie Sheen for consumer testing advice, and of course all the assessors who made the work possible.

Research note

PROCRUSTES-PC is a PC version of GPA and is available from Oliemans, Punter & Partners, Burgemeester Reigerstraat 89, 3581 KP Utrecht, The Netherlands.

References

ARNOLD, G M (1987)
A Generalised Procrustes macro for sensory analysis, Genstat Newsletter, No. 18, 61-80

ARNOLD, G M & WILLIAMS, A A (1986)
The use of generalised Procrustes analysis. In Statistical Procedures in Food Research (ea. J. R. Piggott). Elsevier Applied Science: London

CAIRNCROSS, S E & SJOSTROM, L B (1950)
Flavor profiles- a new approach to flavor problems. Food Technology, 4, 308-11

CHISNALL, P M (1985)
Marketing: A Behavioural Analysis, 2nd Edition. McGraw-Hill: London

GAINS, N, KRZANOWSKI, W J & THOMSON, D M H (1988)
A comparison of variable reduction techniques in an attitudinal investigation of meat products, Journal of Sensory Studies, 3, 37-48

GOWER, J C (1975)
Generalised Procrustes analysis. Psychometrika, 40, 33-52

LAND, D G & SHEPHERD, R (1988)
Scaling and ranking methods. In Sensory Analysis of Foods, 2nd Edition (ea. J R Piggott). Elsevier Applied Science: London

LANGRON, S P (1981)
The Statistical Treatment of Sensory Analysis Data. PhD Thesis. University of Bath

LANGRON, S P (1983)
The application of Procrustes statistics to sensory profiling. In Sensory Quality in Foods and Beverages: Definition, Measurement and Control (eds A A Williams & R K Atkin). Ellis Horwood: Chichester

MORRICE, P (1983)
The Schweppes Guide to Scotch. Alpha Books: Sherborne

PHILP, J M (1989)
Cask quality and warehouse conditions. In The Science and Technology of Whiskies (eds J R Piggott, R Sharp, R E B Duncan). Longman: London

PIGGOTT, J R & CANAWAY P R (1981)
Finding the word for it - methods and uses of descriptive sensory analysis. In Flavour '81 (ea. P Schreier). Walter de Grnyter: Berlin

PIGGOTT J R, CAREY, R G & CANAWAY, P R (1985)
Sensory analysis and evaluation of whisky. In Alcoholic Beverages (eds G G Birch & M G Lindley). Elsevier Applied Science: London

PIGGOTT, J R & SHARMAN, K (1986)
Methods to aid interpretation of multidimensional data. In Statistical Procedures in Food Research (ea. J R Piggott). Elsevier Applied Science: London

POWERS, J J (1988)
Current practices and application of descriptive methods. In Sensory Analysis of Foods, 2nd Edition (ea. J R Piggott). Elsevier Applied Science: London

STONE, H, SIDEL, J, OLIVER, S, WOOLSEY, A & SINGELTON, R C (1974)
Sensory evaluation by QDA, Food Technology, 28(11), 24-34

WILLIAMS, A A (1983)
Defining sensory quality in foods and beverages. Chemistry and Industry, 740-45

WILLIAMS, A A & ARNOLD G M (1985)
A comparison of the aromas of SLY coffees characterised by conventional profiling, freechoice profiling and similarity scaling methods, Journal of the Science of Food and Agriculture, 36, 204-14

WILLIAMS, A A & LANGRON, S P (1984)
The use of free-choice profiling for the evaluation of commercial ports, Journal of the Science of Food and Agriculture, 35, 558-68

Terms	Axis 1			Axis 2			Axis 3
	(a)	(b)	(c)	(a)	(b)	(c)	(a)	(b)	(c)
Maturity	12	5.15	28.97	11	8.87	49.89	5	3.76	21.15
Smooth	11	4.72	22.36	11	8.87	42.02	10	7.52	35.62
Mellow, velvety, rounded	0	0.00	0.00	8	6.45	55.08	7	5.26	44.92
Mild, creamy, rich, soft	1	0.43	4.37	7	5.65	57.42	5	3.76	38.20
Appearance terms (colour and depth)	108	46.35	82.02	7	5.65	10.00	6	4.51	7.98
Flowery, aromatic, fruity, estery and perfumed	9	3.86	41.82	2	1.61	17.40	5	3.76	40.74
Warm, fiery, burning	9	3.86	35.77	3	2.42	22.43	6	4.51	41.80
Harsh, rough, sharp, coarse, nippy, bitter	17	7.30	15.95	30	24.19	52.84	19	14.29	31.21
Malty	1	0.43	10.83	2	1.61	40.55	2	1.50	37.78
Nutty	1	0.43	21.61	1	0.81	40.70	1	0.75	37.69
Smoky, burnt wood, roasted	4	1.72	19.75	4	3.23	37.08	5	3.76	43.17
Peaty, heathery, earthy	5	2.15	19.71	9	7.26	66.54	2	1.50	13.75
Sweet, caramel, chocolate, treacle	9	3.86	25.23	3	2.42	15.82	12	9.02	58.95
	(187)			(98)			(85)
Others	46			26			48
Total	233			124			133