Statistics for Sensory and Consumer Science E-Book

Tormod Næs

• Herausgeber: John Wiley & Sons
• Kategorie: Fachliteratur
• Sprache: Englisch

Seitenzahl: 564

Veröffentlichungsjahr: 2011

Beschreibung

As we move further into the 21st Century, sensory and consumer studies continue to develop, playing an important role in food science and industry. These studies are crucial for understanding the relation between food properties on one side and human liking and buying behaviour on the other. This book by a group of established scientists gives a comprehensive, up-to-date overview of the most common statistical methods for handling data from both trained sensory panels and consumer studies of food. It presents the topic in two distinct sections: problem-orientated (Part I) and method orientated (Part II), making it to appropriate for people at different levels with respect to their statistical skills. This book succesfully: * Makes a clear distinction between studies using a trained sensory panel and studies using consumers. * Concentrates on experimental studies with focus on how sensory assessors or consumers perceive and assess various product properties. * Focuses on relationships between methods and techniques and on considering all of them as special cases of more general statistical methodologies It is assumed that the reader has a basic knowledge of statistics and the most important data collection methods within sensory and consumer science. This text is aimed at food scientists and food engineers working in research and industry, as well as food science students at master and PhD level. In addition, applied statisticians with special interest in food science will also find relevant information within the book.

Leseprobe

Contents

Preface

Acknowledgments

1: Introduction

1.1 The Distinction between Trained Sensory Panels and Consumer Panels

1.2 The Need for Statistics in Experimental Planning and Analysis

1.3 Scales and Data Types

1.4 Organisation of the Book

2: Important Data Collection Techniques for Sensory and Consumer Studies

2.1 Sensory Panel Methodologies

2.2 Consumer Tests

3: Quality Control of Sensory Profile Data

3.1 General Introduction

3.2 Visual Inspection of Raw Data

3.3 Mixed Model ANOVA for Assessing the Importance of the Sensory Attributes

3.4 Overall Assessment of Assessor Differences Using All Variables Simultaneously

3.5 Methods for Detecting Differences in Use of the Scale

3.6 Comparing the Assessors’ Ability to Detect Differences between the Products

3.7 Relations between Individual Assessor Ratings and the Panel Average

3.8 Individual Line Plots for Detailed Inspection of Assessors

3.9 Miscellaneous Methods

4: Correction Methods and Other Remedies for Improving Sensory Profile Data

4.1 Introduction

4.2 Correcting for Different Use of the Scale

4.3 Computing Improved Panel Averages

4.4 Pre-processing of Data for Three-Way Analysis

5: Detecting and Studying Sensory Differences and Similarities between Products

5.1 Introduction

5.2 Analysing Sensory Profile Data: Univariate Case

5.3 Analysing Sensory Profile Data: Multivariate Case

6: Relating Sensory Data to Other Measurements

6.1 Introduction

6.2 Estimating Relations between Consensus Profiles and External Data

6.3 Estimating Relations between Individual Sensory Profiles and External Data

7: Discrimination and Similarity Testing

7.1 Introduction

7.2 Analysis of Data from Basic Sensory Discrimination Tests

7.3 Examples of Basic Discrimination Testing

7.4 Power Calculations in Discrimination Tests

7.5 Thurstonian Modelling: What Is It Really?

7.6 Similarity versus Difference Testing

7.7 Replications: What to Do?

7.8 Designed Experiments, Extended Analysis and Other Test Protocols

8: Investigating Important Factors Influencing Food Acceptance and Choice (Conjoint Analysis)

8.1 Introduction

8.2 Preliminary Analysis of Consumer Data Sets (Raw Data Overview)

8.3 Experimental Designs for Rating Based Consumer Studies

8.4 Analysis of Categorical Effect Variables

8.5 Incorporating Additional Information about Consumers

8.6 Modelling of Factors as Continuous Variables

8.7 Reliability/Validity Testing for Rating Based Methods

8.8 Rank Based Methodology

8.9 Choice Based Conjoint Analysis

8.10 Market Share Simulation

9: Preference Mapping for Understanding Relations between Sensory Product Attributes and Consumer Acceptance

9.1 Introduction

9.2 External and Internal Preference Mapping

9.3 Examples of Linear Preference Mapping

9.4 Ideal Point Preference Mapping

9.5 Selecting Samples for Preference Mapping

9.6 Incorporating Additional Consumer Attributes

9.7 Combining Preference Mapping with Additional Information about the Samples

10: Segmentation of Consumer Data

10.1 Introduction

10.2 Segmentation of Rating Data

10.3 Relating Segments to Consumer Attributes

11: Basic Statistics

11.1 Basic Concepts and Principles

11.2 Histogram, Frequency and Probability

11.3 Some Basic Properties of a Distribution (Mean, Variance and Standard Deviation)

11.4 Hypothesis Testing and Confidence Intervals for the Mean µ

11.5 Statistical Process Control

11.6 Relationships between Two or More Variables

11.7 Simple Linear Regression

11.8 Binomial Distribution and Tests

11.9 Contingency Tables and Homogeneity Testing

12: Design of Experiments for Sensory and Consumer Data

12.1 Introduction

12.2 Important Concepts and Distinctions

12.3 Full Factorial Designs

12.4 Fractional Factorial Designs: Screening Designs

12.5 Randomised Blocks and Incomplete Block Designs

12.6 Split-Plot and Nested Designs

12.7 Power of Experiments

13: ANOVA for Sensory and Consumer Data

13.1 Introduction

13.2 One-Way ANOVA

13.3 Single Replicate Two-Way ANOVA

13.4 Two-Way ANOVA with Randomised Replications

13.5 Multi-Way ANOVA

13.6 ANOVA for Fractional Factorial Designs

13.7 Fixed and Random Effects in ANOVA: Mixed Models

13.8 Nested and Split-Plot Models

13.9 Post Hoc Testing

14: Principal Component Analysis

14.1 Interpretation of Complex Data Sets by PCA

14.2 Data Structures for the PCA

14.3 PCA: Description of the Method

14.4 Projections and Linear Combinations

14.5 The Scores and Loadings Plots

14.6 Correlation Loadings Plot

14.7 Standardisation

14.8 Calculations and Missing Values

14.9 Validation

14.10 Outlier Diagnostics

14.11 Tucker-1

14.12 The Relation between PCA and Factor Analysis (FA)

15: Multiple Regression, Principal Components Regression and Partial Least Squares Regression

15.1 Introduction

15.2 Multivariate Linear Regression (MLR)

15.3 The Relation between ANOVA and Regression Analysis

15.4 Linear Regression Used for Estimating Polynomial Models

15.5 Combining Continuous and Categorical Variables

15.6 Variable Selection for Multiple Linear Regression

15.7 Principal Components Regression (PCR)

15.8 Partial Least Squares (PLS) Regression

15.9 Model Validation: Prediction Performance

15.10 Model Diagnostics and Outlier Detection

15.11 Discriminant Analysis

15.12 Generalised Linear Models, Logistic Regression and Multinomial Regression

16: Cluster Analysis: Unsupervised Classification

16.1 Introduction

16.2 Hierarchical Clustering

16.3 Partitioning Methods

16.4 Cluster Analysis for Matrices

17: Miscellaneous Methodologies

17.1 Three-Way Analysis of Sensory Data

17.2 Relating Three-Way Data to Two-Way Data

17.3 Path Modelling

17.4 MDS-Multidimensional Scaling

17.5 Analysing Rank Data

17.6 The L-PLS Method

17.7 Missing Value Estimation

Nomenclature, Symbols and Abbreviations

Nomenclature and Distinctions

Use of Lowercase, Italics, Boldface

Symbols

Abbreviations and Acronyms

Index

This edition first published 2010

© 2010 John Wiley & Sons Ltd

Registered office

John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, United Kingdom

For details of our global editorial offices, for customer services and for information about how to apply for permission to reuse the copyright material in this book please see our website at www.wiley.com.

The right of the author to be identified as the author of this work has been asserted in accordance with the Copyright, Designs and Patents Act 1988.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by the UK Copyright, Designs and Patents Act 1988, without the prior permission of the publisher.

Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic books.

Designations used by companies to distinguish their products are often claimed as trademarks. All brand names and product names used in this book are trade names, service marks, trademarks or registered trademarks of their respective owners. The publisher is not associated with any product or vendor mentioned in this book. This publication is designed to provide accurate and authoritative information in regard to the subject matter covered. It is sold on the understanding that the publisher is not engaged in rendering professional services. If professional advice or other expert assistance is required, the services of a competent professional should be sought.

The publisher and the author make no representations or warranties with respect to the accuracy or completeness of the contents of this work and specifically disclaim all warranties, including without limitation any implied warranties of fitness for a particular purpose. This work is sold with the understanding that the publisher is not engaged in rendering professional services. The advice and strategies contained herein may not be suitable for every situation. In view of ongoing research, equipment modifications, changes in governmental regulations, and the constant flow of information relating to the use of experimental reagents, equipment, and devices, the reader is urged to review and evaluate the information provided in the package insert or instructions for each chemical, piece of equipment, reagent, or device for, among other things, any changes in the instructions or indication of usage and for added warnings and precautions. The fact that an organization or Website is referred to in this work as a citation and/or a potential source of further information does not mean that the author or the publisher endorses the information the organization or Website may provide or recommendations it may make. Further, readers should be aware that Internet Websites listed in this work may have changed or disappeared between when this work was written and when it is read. No warranty may be created or extended by any promotional statements for this work. Neither the publisher nor the author shall be liable for any damages arising herefrom.

Library of Congress Cataloging-in-Publication Data

Næs, Tormod.

Statistics for sensory and consumer science/Tormod Næs and Per B. Brockhoff and Oliver Tomic.

p. cm.

Summary: “As we move further into the 21st Century, sensory and consumer studies continue to develop, playing an important role in food science and industry. These studies are crucial for understanding the relation between food properties on one side and human liking and buying behaviour on the other. This book by a group of established scientists gives a comprehensive, up-to-date overview of the most common statistical methods for handling data from both trained sensory panels and consumer studies of food. It presents the topic in two distinct sections: problem-orientated (Part I) and method orientated (Part II), making it to appropriate for people at different levels with respect to their statistical skills. This book succesfully makes a clear distinction between studies using a trained sensory panel and studies using consumers. Concentrates on experimental studies with focus on how sensory assessors or consumers perceive and assess various product properties. Focuses on relationships between methods and techniques and on considering all of them as special cases of more general statistical methodologies. It is assumed that the reader has a basic knowledge of statistics and the most important data collection methods within sensory and consumer science. This text is aimed at food scientists and food engineers working in research and industry, as well as food science students at master and PhD level. In addition, applied statisticians with special interest in food science will also find relevant information within the book”- Provided by publisher.

Summary: “This book will describe the most basic and used statistical methods for analysis of data from trained sensory panels and consumer panels with a focus on applications of the methods. It will start with a chapter discussing the differences and similarities between data from trained sensory and consumer tests”- Provided by publisher.

Includes bibliographical references and index.

ISBN 978-0-470-51821-2 (hardback)

1. Food-Sensory evaluation. 2. New products. I. Brockhoff, Per B. II. Tomic, Oliver. III. Title.

TX546.N34 2010

664′.07-dc22

2010016197

Preface

Sensory and consumer studies play an important role in food science and industry. They are both crucial for understanding the relation between food properties on one side and human liking and buying behaviour on the other. These studies produce large amounts of data and without proper data analysis techniques it is impossible to understand the data fully and to draw the best possible conclusions.

This book aims at presenting a comprehensive overview of the most important and frequently used statistical methods for handling data from both trained sensory panels and consumer studies of food. A major target group for this book is food scientists and food engineers working in research and industry. Another important target group is food science students at master and PhD level at universities and colleges. Applied statisticians with special interests in food science will also find important material in the book. The goal is to present the reader with enough material to make him/her able to use the most important methods and to interpret the results safely.

The book is organised in two main parts, Part I has a quite different focus to Part II. Part I has a problem oriented focus and presents a number of typical and important situations in which statistical methods are needed. Part II has a method oriented perspective and all the methods discussed in Part I are described more thoroughly here. There is a strong link between Part 1 and Part II through extensive cross-referencing. The structure of the book is also presented in Figure 1.1.

The book will have focus on relationships between methods and techniques and on considering all of them as special cases of more general statistical methodologies. In this way we will avoid, as far as possible, using “local dialect” for each of the themes discussed. Conjoint analysis is an example of an area which has developed into a separate discipline with a particular terminology and culture. In our approach conjoint analysis will be considered and presented as an example of an intelligent use of experimental design and analysis of variance which are both classical disciplines in statistics.

It will be assumed that the reader has a basic knowledge of statistics and also the most important data collection methods within sensory and consumer science. For some of the more advanced parts of Part II, an elementary knowledge of matrix algebra will make reading easier.

Tormod Næs, Per B. Brockhoff and Oliver Tomic

Acknowledgments

We would like to thank Nofima Mat and DTU for having made it possible for us to write this book. We would also like to thank colleagues for important discussions and support related to the examples used for illustration.

Tormod Næs: A large part of the book was written while I was visiting scientist at Dipartimento delle Biotechnologie Agrarie at University of Firenze (winter 2008-2009). The colleagues at University of Firenze are thanked for this opportunity. I would also like to thank my wife for staying faithfully and patiently with me during this period in our small house in Impruneta outside Firenze.

Oliver Tomic: I would like to dedicate this book to my daughter Nina, my wife Heidi and my parents Milica and Marko.

Per B. Brockhoff: My contribution to this book is dedicated to my wife Lene and my daughters Trine, Fie and Julie.

1

Introduction

Some of the most important aspects in food science and food industry today are related to human perception of the food and to enjoyment associated with food consumption. Therefore, very many activities in the food sector are devoted to improving already existing products and developing new products for the purpose of satisfying consumer preferences and needs. In order to achieve these goals one needs a number of experi- mental procedures, data collection techniques and data analysis methods.

1.1 The Distinction between Trained Sensory Panels and Consumer Panels

In this book we will, as usually done, make a clear distinction between studies using a trained sensory panel (Amerine et al., 1965; O’Mahony, 1986; Meilgaard et al., 1999) and studies using consumers (Lawless and Heyman, 1999). The former is either used for describing degree of product similarities and differences in terms of a set of sensory attributes, so- called sensory profiling, or for detecting differences between products, so-called sensory difference testing. For the various attributes, the measurement scale is calibrated and usually restricted to lie between a lower and an upper limit, for instance 1 and 9. A sensory panel will normally consist of between 10 and 15 trained assessors and be thought of as an analytical instrument. For consumer studies, however, the products are tested by a representative group of consumers who are asked to assess their degree of liking, their preference or their purchase intent for a number of products. These tests are often called hedonic or affective tests. While the trained sensory panel is only used for describing the products as objectively as possible, consumer studies are used for investigating what people or groups of people like or prefer. The number of consumers must be much higher than the number of assessors in a sensory panel in order to obtain useful and reliable information. Typically, one will use at least between 100 and 150 consumers in this type of studies.

Note that sometimes the term sensory science is used to comprise all types of tests where the human senses are used. This means that many consumer studies are also sensory tests. The difference between sensory consumer tests and sensory panel tests is the way they are used; sensory panels are used for describing the properties of products and sensory consumer tests are used for investigating the degree of liking. In this book it will be clear from the context and description of the situation which of these studies that is in focus.

Note also that one is often interested in relating the two types of data to each other for the purpose of understating which sensory attributes are important for liking. This is important both for product development studies, for developing good marketing strategies and also for the purpose of understanding more generally what are the opinions and trends in various consumer segments.

In this book, we will concentrate on experimental studies with focus on how sensory assessors or consumers perceive and assess various product properties. Consumer surveys of attitudes and habits will play a minor role here, even though some of the statistical methods treated may also be useful in such situations. In the examples presented main attention will be given to data that are collected by asking people about their opinion (stated acceptance or preference), but many of the same statistical methods can be used if data are obtained by monitoring of real behaviour (revealed acceptance or preference, see e.g. Jaeger and Rose, 2008).

For a broad discussion of possible problems and pitfalls when interpreting results from consumer studies we refer to Köster (2003).

1.2 The Need for Statistics in Experimental Planning and Analysis

In sensory and consumer science one is typically interested in identifying which of a num- ber of potential factors that have an influence on the sensory attributes and/or the consumer liking within a product category. For obtaining such information, the most efficient exper- imental procedures can be found within the area of statistical experimental design (Box et al., 1978). These are methods which describe how to combine the different factors and their levels in such a way that as much information as possible is obtained for the lowest possible cost. For optimising a product, one will typically need to work in sequence, starting out by eliminating uninteresting factors and ending up with optimising the most important ones in a limited experimental region. In all phases it is more efficient to consider series of experiments where all factors are varied instead of investigating one factor at a time. Im- portant building blocks in this tradition are the factorial designs, fractional factorial designs and central composite designs (Chapter 12). The concepts of randomisation and blocking, for systematically controlling uninteresting noise factors, are also important here. Another important point is representativity, which means that the objects and assessors are selected in such a way that they represent the situation one is interested in the best possible way. For instance, if a whole day’s production of a product is to be investigated, one should not investigate consecutive samples, but rather pick them at random throughout the day.

The data sets that emerge from sensory and consumer experiments are typically quite large and the amount of information available about relations between them is limited. It is therefore important to have data analysis methods that in a pragmatic way can handle such situations. Focus in the present book will be on analysis of variance (ANOVA) and regression based methods (Chapter 13, and 15) and methods based on PCA for data compression (Chapter 14, 16, 17). An important aspect of all these methods is that they are versatile and can be used in many practical applications. We will be interested in significance testing for indicating where the most important information is and plotting techniques for visual inspection of complex relations. Validation of models by the use of empirical data will also be important. The main philosophy in the exposition will be simplicity, transparency and practical usefulness.

It is important to emphasise that in order to get the most out of statistical design and analysis methods, one must use as much subject matter knowledge as possible. It is only when statistical and subject matter knowledge play well together that the best possible results can be obtained. It is also worth mentioning that although the book is primarily written with a focus on applications within food science, many of the same methods can be used also for other applications where sensory and consumer aspects are involved.

Other books that cover some of the same topics as discussed here are Gacula et al. (2009), Mazzocchi (2008), Næs and Risvik (1996), and Meullenet et al. (2007).

1.3 Scales and Data Types

In most of the book we will consider sensory panel data and also consumer rating data as continuous interval scale data. This means that the differences between two different values will be considered meaningful, not only the ordering of the data. One of the advantages of taking such a perspective is that a much larger set of methods which are easy to use and understand are made available. It is our general experience that such an approach is both reasonable and useful.

If the data are collected as rank data or as choice/preference data, it is necessary to use methods developed particularly for this purpose. For choice based conjoint experiments one will typically treat the data as nominal categorical data with a fixed set of outcomes and analyse with for instance generalised linear models (see e.g. Chapter 15). The same type of methods can also be used for rank data, but here other options are also available (see Chapter 17).

1.4 Organisation of the Book

This book is organised in two parts, Part I (Chapters 1-10) and Part II (Chapters 11-17). The first part is driven by applications and examples. The second part contains descriptions of a number of statistical methods that are relevant for the application in Part I. In Part I we will refer to the relevant methodologies presented in Part II for further details and discussion. In Part II we will refer to the different chapters in Part I for typical applications of the methods described. The more practically oriented reader may want to focus on Part I and look up the various specific methods in Part II when needed. The more statistically oriented reader may prefer to do it the other way round. The structure of the book is illustrated in Figure 1.1.

Figure 1.1 Description of how the book is organised.

References

Amerine, M.A., Pangborn, R.M., Roessler, E.B. (1965). Principles of Sensory Evaluation of Food. New York: Academic Press.

Box, G.E.P., Hunter, W., Hunter, S. (1978). Statistics for Experimenter. New York: John Wiley & Sons, Inc.

Gacula, M.C. Jr., Singh, J., Bi, J., Altan, S. (2009). Statistical Methods in Food and Consumer Science. Amsterdam, NL: Elsevier.

Jaeger, S.R., Rose, J.M. (2008). Stated choice experimentation, contextual influences and food choice. A case study. Food Quality and Preference 10, 539-64.

Köster, E.P. (2003). The psychology of food choice. Some often encountered fallacies. Food Quality and Preference 14, 359-73.

Lawless, H.T., Heymann, H. (1999). Sensory Evaluation of Food: Principles and Practices. New York: Chapman & Hall.

Mazzocchi, M. (2008). Statistics for Marketing and Consumer Research. Los Angeles: Sage Publications.

Meilgaard, M., Civille, G.V., Carr, B.T. (1999). Sensory Evaluation Techniques (2nd edn). Boca Raton, Florida: CRC Press, Inc.

Meullenet, J-F, Xiong, R., Findlay, C.J. (2007). Multivariate and Probabilistic Analysis of Sensory Science Problems. Ames, USA: Blackwell Publishing.

Næs, T., Risvik, E. (1996). Multivariate Analysis of Data in Sensory Science. Amsterdam: Elsevier.

O’Mahony, M. (1986), Sensory Evaluation of Food, Statistical Methods and Procedures. New York: Marcel Dekker, Inc.

2

Important Data Collection Techniques for Sensory and Consumer Studies

This chapter gives a brief description of some of the most important methodologies for collecting data in sensory and consumer science. For more detailed and comprehensive presentations of these and related methods we refer to Amerine et al. (1965), Lawless and Heyman (1999) and O’Mahony (1986).

2.1 Sensory Panel Methodologies

2.1.1 Descriptive Sensory Analysis

Descriptive sensory analysis or so-called sensory profiling is probably the most important method in sensory analysis and also the one that will be given the most attention here. This is a methodology which is used for describing products and differences between products by the use of trained sensory assessors. The main advantage of sensory analysis as compared to for instance chemical methods is that it describes the properties of a product in a language that is directly relevant for people’s perception, for instance degree of sweetness, hardness, colour intensity etc. The sensory panel used this way is thought of and used as an analytical instrument.

Typically, a sensory panel consists of between 10 and 15 trained assessors, usually recruited according to their ability to detect small differences in important product attributes. Before assessment of a series of products, the assessors gather to decide on the attributes to use for describing product differences. In some cases, certain attributes may also be given prior to this discussion. Usually, one will utilise some of the products for the purpose of calibrating the scale to be used, if not calibrated by other means. In some cases the assessors are allowed to use their own vocabulary (free choice profiling, FCP, see e.g. Arnold and Williams, 1987), but this type of analysis will not be given attention here. In most cases, the number of attributes will be between 10 and 20, but this depends on the complexity of the products and also the scope of the study. In the actual testing session, all assessors are given the products in random order and without knowing anything about the product differences, labelling, brands etc.; so-called blind tasting. The intensity scores for the different products are either given as numbers between a lower and an upper limit or as indications on a line, either on a computer screen or on paper. Descriptive sensory analysis produces data that can be presented in a three-way data structure as indicated in Figure 2.1. In most cases, each sample is tested in duplicate or triplicate. Within the three-way framework of Figure 2.1 replicates can be accommodated by representing them as new samples or products.

Figure 2.1 Three-way sensory data. This illustration depicts the general structure of sensory profiling data. For each assessor (i), there is a data table consisting of measurements (scores) for a number of attributes (k) and a number of products (j). If there are replicate (r) in the data set, these can be added as additional rows.

Different ways of presenting the samples to the assessors are possible. One possibility is to have standard sample(s) present during the whole session for the purpose of providing a stable calibration of the panel. In most cases, however, the samples are just presented to the assessor without any of these additional tools. For the purpose of the statistical analysis of the data, this aspect has little influence.

In this book sensory analysis data will be treated in several of the chapters. In some of the chapters the only focus is on sensory analysis itself and how it can be used to detect and understand sensory differences between products (for instance Chapter 5). In other cases, it will be used in combination with consumer data for the purpose of understanding better the consumer preferences and acceptances (Chapter 9). A separate chapter will be devoted to quality control of sensory panels (Chapter 3). This is of particular importance for the assessment of panel reliability and for panel improvement.

The main reasons for having several assessors, instead of one or only a few, in a panel are that this gives more precise assessments of product attributes, it provides an automatic internal quality control of the panel and that individual differences can be detected and analysed. The latter can in some cases be very important for assessing differences in use of the scale, for detecting differences in thresholds etc.

2.1.2 Discrimination Tests

Another type of sensory test is the class of so-called discrimination or difference tests, which are suitable for testing whether the assessors, either consumers or sensory panellists, are able to detect small differences between products tested. An important example is for recruiting assessors for a sensory panel. In such cases potential candidates are typically asked to distinguish between samples with very similar intensities of bitterness, sweetness, sourness, umami and saltiness. Another important example is for analysing product substitutability (Ennis, 1993a), which is typically of interest when introducing a new and cheaper ingredient. The triangle test is probably the most commonly used discrimination test. This is based on giving each assessor a so-called triangle consisting of two identical and one different sample. He/she is then asked to identify the one which is different. The random probability for this is equal to 1/3 if the products are identical. These standard tests for such hypotheses are usually based on the binomial distribution. More advanced methods based on Thurstonian modelling (see Chapter 7) can, however, be used to obtain more insight.

2.2 Consumer Tests

2.2.1 Hedonic or Affective Tests

Sensory analysis by trained sensory panels describes the products as objectively as possible, but in order to obtain information about what people like, various types of consumer studies are needed. Linking the two types of analyses is of particular importance since one is often interested in understanding what are the main drivers for food choice or liking. For instance, is the acceptance of a certain product related to sweetness or another sensory attribute or have extrinsic attributes like various types of information or packaging a larger impact? While a sensory panel is primarily selected based on the assessors ability to detect and measure sensory aspects of products, a consumer study will normally be based on results from consumers that are randomly drawn from a certain population. In some cases one may decide to select consumers that are consumers of a particular product or one may decide to do a more systematic sampling for the purpose of for instance ensuring a certain distribution of a demographic variable.

In this book, main emphasis will be given to experimental consumer studies with a hedo- nic response, which are typical and frequently used both in an industrial and in a research context. Both sensory product-related attributes as well as extrinsic attributes related to in- formation and packaging will be in focus. Important examples to be discussed are conjoint analysis and preference mapping studies. Large surveys based on questionnaires related to habits, attitudes etc. will only be touched upon briefly, but many of the methods considered in this book will also be useful in such a context. Other methods that will not be covered in this book are methods based on deeper interviews and discussions with consumers.

Experimental consumer tests may be conducted in central locations or labs, in homes or via internet. For the purpose of this book, which is mainly about statistical methods, a clear distinction between these techniques will not be made, because the data structures and statistical analysis techniques are usually the same for all. For a deeper discussion of all these aspects we refer to Lawless and Heymann (1999) and references therein.

2.2.2 Self-explicated Tests

The simplest types of tests that can be conducted are the self-explicated tests. These are tests in which the consumers are asked which of a series of attributes they put most emphasis on when they select food products or when they make choices. They may also be asked to rank the importance of the attributes. These tests can be useful and are not necessarily inferior to others (see Gustafsson et al. (2003)), but they also have a number of drawbacks. First of all they cannot be used for assessing interactions between the attributes, which can sometimes be a major concern. Secondly they require a mental processing which is not typical for a buying situation. Main emphases will therefore here be given to experimental strategies that combine various product attributes or contexts of interest using experimental design methodologies. The consumers are then asked to assess the different combinations of attributes varied.

2.2.3 Rating Based Studies

Rating based studies will be given main attention here. These are experimental studies where the consumers are asked to assess either their degree of liking, their degree of acceptance or their probability of buying the products for each of the combinations tested. In for instance purely sensory tests it is natural to ask about degree of liking while in more concept oriented tests including also several extrinsic attributes, purchase intent or purchase probability may sometimes be more relevant. Since the statistical analysis is usually the same regardless of which question is asked, we will not distinguish strictly between the different questions asked to the consumers, whether it is expected liking (Deliza and MacFie, 1996), actual liking of a real product or purchase probability. We refer to Lawless and Heymann (1999) and to Mela (2000) for discussion of various aspects related to the different types of consumer responses.

In some cases the different attributes tested are presented verbally or by using illustrations. Important examples are related to for instance health claims, brand name and packaging. In many cases, however, it is natural to bring in real products in order to assess the relative importance of sensory and extrinsic information as well as their possible interactions. The sensory perception is then brought in as an important aspect of the assessors’ rating of the products.

The consumers will typically, as for sensory analysis, give a score between a lower and an upper value for each of the products regardless of which question they are asked to respond to. The scale is in many cases anchored with for instance ‘like very little’ and ‘like very much’, but this is not always done.

2.2.4 Ranking Tests

Another type of tests is the ranking test. In this type of test all possibilities (samples) are presented simultaneously to the consumers and they are asked to rank them according to for instance liking or purchase intent. If there are many combinations to be tested, the sorting can be done in sequence, by first splitting in two, then in two again etc. until all have been ranked. Note that this type of test is impossible or at least very difficult to use in context studies where it is impossible to present all alternatives at the same time. Ranking tests are useful, but the main drawback lies in the analysis. Since the data are ordinal and the sum of all assessments for each assessor sum to a constant, this limits the number of possible analysis methods considerably. Some possibilities exist as will be discussed later in this book (Chapter 7 and Chapter 17), but the methods are often more difficult to understand and the number of possibilities is limited.

2.2.5 Choice Tests

Choice studies (Louviere et al., 2000) will also be discussed in this book (Chapter 8). These are tests for which the consumers are given a number of so-called choice sets and for each choice set they are asked to select the one they like best. A choice set is constructed by selecting systematically a number of products from the full design of possibilities. Some researchers claim that these tests are more realistic than rating and ranking tests since they fit better to a real buying situation. This is, however, questionable and studies exist which indicate that it does not always matter so much for the conclusions in which way the data are collected (see for instance Jaeger and Rose, 2008 for a discussion). As for the ranking studies, the number of analysis methods is more limited and they are more difficult to use and understand than for rating tests.

2.2.6 Auctions and Monitoring of Behaviour

Price is often an important factor to consider in probability-to-buy studies. A number of specific techniques have been developed for this purpose, for instance various types of auctions (Lange et al., 2002) and willingness to pay tests. These methods will not be given special attention here, although several of the standard statistical methods treated in Part II of the book can be used also for this purpose. Price is, however, also easy to incorporate as a separate factor in conjoint analysis and this will be the main approach taken here.

Real consumer behaviour in buying situations is, as mentioned above, more relevant than what they say, but also much more difficult to measure. Different methods have, however, been developed based on auctions or monitoring of actual choices made in a cafeteria or in a supermarket. Consumers are for instance given an amount of money before the test and they are asked to spend the money the way they like under certain restrictions on choice within product categories. Another possibility is to organise a set of different options in a cafeteria, with the possibility of repeated experimentation. The data can be analysed by the methods presented in this book.

2.2.7 Additional Consumer Attributes and Characteristics

When consumer acceptance data have been collected and analysed, one will typically also be interested in understanding the individual differences between consumers in a better way. This can be done if additional consumer attributes have been collected during or after the actual testing session. This can be done using a questionnaire with questions related to demographic variables such as gender, age, family size and variables related to attitudes and habits, either on a general basis or related to the actual product studied. These consumer attributes can be related to the individual differences observed using regression or tabulation methods (Chapter 8). Segmentation is an important concept which comes in here (Chapter 10). This can be done either a priori based on the consumer attributes themselves or based on the consumer liking pattern with subsequent analysis or relations to consumer attributes.

The data set based on a hedonic consumer test with additional consumer attributes can typically be represented in data tables with simple structure as indicated in Figure 2.2.

Figure 2.2 Two-way consumer data. This illustration simply depicts the structure of a matrix of consumer hedonic scores for a number of products.

References

Amerine, M.A., Pangborn, R.M., Roessler, E.B. (1965). Principles of Sensory Evaluation of Food. New York: Academic Press.

Arnold, G.M., Williams, A.A. (1987). The use of generalised procrustes techniques in sensory analysis. In J.R. Piggott (ed.), Statistical Procedures in Food Research. London: Elevier Science Publishers, 244-53.

Deliza, R., MacFie H.J.H. (1996). The generation of sensory expectation by external cues and its effect on sensory perception and hedonic ratings: A review. Journal of Sensory Studies 11, 103-28. Ennis, D.M. (1993a). The power of sensory discrimination methods. Journal of Sensory Studies 8, 353-70.

Gustafsson, A., Herrmann, A., Huber, F (2003). Conjoint Measurement. Methods and Applications. Berlin: Springer.

Jaeger, S.R., Rose, J.M. (2008). Stated choice experimentation, contextual influences and food choice. A case study. Food Quality and Preference 10, 539-64.

Lange, C. Martin, C., Chabanet, C., Combris, P., Issanchou, S. (2002). Impact of the information provided to consumers on their willingness to pay for Champagne: comparison with hedonic scores. Food Quality and Preference 13, 597-608.

Lawless, H.T., Heymann, H. (1999). Sensory Evaluation of Food: Principles and Practices. New York: Chapman & Hall.

Louviere, J.J., Hensher, D.A. and Swait, J.D. (2000). Stated Choice Methods: Analysis and Applications. Cambridge: Cambridge University Press.

Mela, D. J. (2000). Why do we like what we like? Journal of the Science of Food and Agriculture 81, 10-16.

O’Mahony, M. (1986). Sensory Evaluation of Food, Statistical Methods and Procedures. New York: Marcel Dekker, Inc.

3

Quality Control of Sensory Profile Data

This chapter focuses on quality control of trained sensory panels. This subject is important for determining panel reliability, for being able to improve the panel and for better handling of the actual data. A number of different methods will be proposed and discussed together with a strategy for how the different methods can be used together to identify different types of problems. Some of the methods are multivariate and may be used to obtain an overview of all the attributes. Other methods are univariate and therefore suitable for more detailed studies of specific attributes. A discussion is provided regarding advantages and disadvantages of the different techniques. An example from analysis of apples will be used to illustrate the different methods and how they can be used together to reveal a number of different problems. Most plots presented here are made in the open source software package PanelCheck available at