Optimising sensory properties of products : 1. Conjoint Analysis

1
Optimising sensory properties of products 1.
Conjoint Analysis

• Agricultural Economics
• February 2003

2
Web site for courses http//www.halmacfie.com Ver
y informative food qual. and pref.web-site
http//www.elsevier.nl/locate/foodqual Join the
Sensometric Society for 100 and receive Food
Quality and Preference! Visit http//www.sensomet
ric.org
3
Conjoint Analysis

• Conjoint in Action
• Uses
• Theory
• How to do it
• Variations
• Worked example

4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
Conjoint Analysis

• Examines how consumers develop overall
  preferences for goods and services
• Assumes consumers take individual utilities of
  each attribute and sum them to give an overall
  utility value
• Requires consumers to perform a simple task
• eg rating, ranking, choice, pairwise preference
• Calculates a profile for each consumer
• Gives an overall profile or by segments
• Can be used to predict choice patterns

15
Managerial Uses of Conjoint Analysis

• Define product with optimum combination of
  features
• Indicate relative contributions of each attribute
  to overall evaluation
• Predict market share
• Identify market segments
• Identify marketing opportunities

16
Conjoint Analysis - Method
Designing Stimuli
Data collection
Analysis, evaluation and decision making
17
Conjoint Analysis - Method
Designing Stimuli
Lets look at this
Data collection
Analysis, evaluation and decision making
18
Designing the stimuli

• Specify the attributes that are likely to be
  important to consumers
• focus groups, one to one interviews
• Select factors (eg attributes) that you are going
  to vary in the trial
• eg colour of pack, quantity of information
• brand name, language of fruit description,
  fruit image
• Factors must be actionable and communicable not
  fuzzy

19
(No Transcript)
20
Designing the stimuli

• Select the levels of the factors you are going to
  use in the stimuli
• eg colour - red or white
• quantity of information -none or a lot
• brand - none, not well known, very well known
• Levels may be at the edge or a little extreme
  from current settings

21
(No Transcript)
22
Basic model form

• Additive
• Consumers simply add together the part worths to
  give an overall total across the attributes
• Compositional
• Consumers add their part-worths but in some cases
  the total may be more than their sum

23
(No Transcript)
24
Conjoint Analysis - Method
Designing Stimuli
Data collection
Lets look at this
Analysis, evaluation and decision making
25
Data collection Choosing a presentation
method What type of stimuli will be used
Full-profile (Traditional)
Trade-off matrix
Pairwise comparison
26
Trade Off Approach

• Subjects rank all combinations of pairs of
  attributes
• Simple to do and simple to administer
• large no of judgements, unrealistic, no pictures
• not used very much

27
Full-Profile Approach

• White package
• Southern delight
• Lots Information
• Photo of fruit
• Maracuja Juice
• How much would you like this product?

28
Full Profile approach

• Realistic presentation
• Simple to do
• Can reduce numbers with fractionals
• Subject has to make trade-offs in doing the task
• subjects can suffer from information overload
• order of presentation can influence results
• most common approach

29
Pairwise approach

• Which one do you prefer and by how much?

30
Pairwise approach

• Uses whole products so better than trade off
• Used in specialised conjoint designs
• eg adaptive conjoint
• Not particularly realistic to the food situation
• Never used it myself.

31
Data collection Choosing a presentation
method What type of stimuli will be used
Pairwise comparison
Trade-off matrix
Full-profile (Traditonal)
Data collection Creating the stimuli can
respondents assess all stimuli
Subset
All
Factorial design
Fractional factorial
32
Creating the stimuli

• Form an experimental design consisting of a
  number of treatment combinations
• Here is a full factorial design

33
A fractional design for the passion fruit trial
34
Data Collection Selecting a preference
measure rating or ranking (non-metric)
Data Collection Form of survey Personal
Interview Mail survey Phone surveys
35
(No Transcript)
36
Conjoint Analysis - Method
Designing Stimuli
Data collection
Lets look at this
Analysis, evaluation and decision making
37
Analysis

• Use Analysis of variance
• Inspect individual part worths
• Cluster individuals to find segments

38
Results after cluster analysis
39
Interpret and recommend

• Unpacked ham preferred
• Zwan and Stegeman liked by segment 1
• High price preferred by segment 2
• Sell unpacked Z and S at premium price, stressing
  the brand names would satisfy these segments.

40
Assessing Reliability and validity

• Use goodness of fit of model eg R2
• Take a few replicate judgements and test how well
  they correspond
• Hold out some stimuli, fit the data and see how
  well the predictions fit
• If using aggregate model, split popn into subsets
  and compare solutions.

41
Choice Simulator

• Take a set of stimuli (market place)
• For each individual or segment calculate
  preference for each stimulus
• Can estimate the proportion of choices made for
  each product
• can extend with probability and brand switching
  theory

42
Conjoint Analysis Example

• You have been asked to develop a new series of
  menus that will be suitable for use in the
  workers and for visitors that are being brought
  to the canteen
• You must choose from the following selections
• Starter Main dessert
• Fresh fruit beef with gorgonzola ice cream
• Tomato soup fish in cheese sauce sticky toffee
  pudding

43

• We construct a table like that shown before using
  the following factors
• starter, main course, dessert
• Select a low and a high level of each factor and
  write them in to the table
• Eg fresh fruit low and tomato soup high
• Factor
• Name Low level High Level
• A starter fruit soup
• B main beef fish
• C dessert ice cream toffee

44

• Form a set of treatment combinations in the table
  using the settings shown below
• Factor
• A B C Liking
• Low Low Low
• High Low Low
• Low High Low
• High High Low
• Low Low High
• High Low High
• Low High High
• High High High

45
Constructing a Design matrix

• starter main dessert liking liking
• employee visitor
• fruit beef ice cream
• fruit beef toffee
• soup beef toffee
• soup beef ice cream
• fruit fish toffee
• fruit fish ice cream
• soup fish ice cream
• soup fish toffee

46
Get an employee (or a group of employees) to
score for liking

• starter main dessert liking liking
• employee visitor
• fruit beef ice cream 4
• fruit beef toffee 5
• soup beef toffee 3
• soup beef ice cream 3
• fruit fish toffee 8
• fruit fish ice cream 7
• soup fish ice cream 5
• soup fish toffee 6

47
Then repeat exercise for visitors

• starter main dessert liking liking
• employee visitor
• fruit beef ice cream 4 7
• fruit beef toffee 5 8
• soup beef toffee 3 6
• soup beef ice cream 3 5
• fruit fish toffee 8 5
• fruit fish ice cream 7 4
• soup fish ice cream 5 3
• soup fish toffee 6 4

48
Calculating effects

• For the starter we can calculate the effect of
  changing from a soup to a fruit
• Mean liking scores for fruit Mean liking score
  for soup
• for an employee as (4587)/4 (3356)/4
• ( 24-17)/4 7/4
• Visitor as (7854)/4 (6534)/4 (24-18)/46/4

49
Calculating effects (main course)
50
Calculating effects dessert

• For the dessert we can calculate the effect of
  changing from an ice cream to a toffee pudding
• for an employee as one quarter of(5386)
  (4375) 20-19 1/4
• Visitor as one quarter of (8654)- (7543)
  23-19 4/4

51
Form table of effects of each level from zero
Note the effect from zero is half the difference
between the two levels
52
Form a bar chart of effects
53
Recent forms of conjoint analysis

• Hybrid
• Choice based
• Adaptive

54
Hybrid

• useful when large number of assessments
• each subject assesses a subset of the total
• each subject gives estimates of relative
  importance and desirability of attributes of
  attributes
• can get an aggregate model and still retain some
  individual differences

55
Choice based Conjoint Analysis

• Realistic task
• Can permit no choice option
• Can vary the number of stimuli presented
• Analysis usually at aggregate level
• Can lead to large number of choices
• restrict to 6 factors or less
• usage increasing

56
(No Transcript)
57
(No Transcript)
58
(No Transcript)
59
(No Transcript)
60
Adaptive Conjoint Analysis

• Used when the number of stimuli is large
• Recommended for more than 6 factors
• OK for paper concepts
• may be difficult for package image or product
  testing
• usage increasing as s/ware becomes available

61
Adaptive Conjoint Analysis

• scores to the relative importance of the factors
• pairwise decisions.
• Computer determines order of assessment
• uses previous scores to select choices that are
  difficult to predict

62
Example of a Concept test tool using conjoint
analysis
Web Site www.mji-designlab.com
63
5. Summary

• A powerful technique to measure decision making
• Widely used
• Analysis relatively easy
• Segmentation through cluster analysis very useful
• Dont be too ambitious in numbers of
  factors/levels

64
References

• Green, P. E. and Srinivasan, V. (1978) Conjoint
  Analysis in Consumer research Issues and
  Outlook. J. Consumer Research , 5, 103-123
• Haix, Anderson, Totham and Beack (1995)
  Multivariate Data Analysis, Prentice Hall, 4th
  edition, pp507-601
• Malhota, N. (1996) Marketing Research. An applied
  orientation. Prentice Hall, p693-727
• Steenkamp, J.E.M. (1987) Conjoint measurement in
  ham quality evaluation. Journal of Agricultural
  Economics, 38,473-480
• Louviere, J.J. and Woodworth, G. (1983) Design
  and Analysis of simulated consumer choice or
  allocation experiments an approach based on
  aggregate data Journal of Marketing Research,
  20,350-367

65
Conjoint Exercise You have been asked to develop
a new health bar product and you decide to
examine the effects of three important factors on
a group of consumers Following the example in
the lecture select three factors and then two
levels of each factor write them in to the
table Eg fresh fruit low and tomato soup high
from the previous example Factor Name Low
level High Level 1 2 3
66
The table below gives all possible sets of
treatment combinations as was done in the
example.
Factor A B C Liking Low Low Low High Low
Low Low High Low High High Low Low L
ow High High Low High Low High High Hig
h High High
67
Complete the table below using your factors.
Please follow the exact format of the low and
high settings given above. Then you and your
partner can complete the table using a nine point
scale.
68
Calculate the effects below. Note here we are
working with means and calculating effects from
zero Effect of factor 1 high
level 0.5(Mean High level scores -mean Low level
scores) Effect of factor 2 high
level 0.5(Mean High level scores -mean Low level
scores) Effect of factor 3 high
level 0.5(Mean High level scores -mean Low level
scores) Repeat for the
second person. Effect of factor 1 high
level 0.5(Mean High level scores -mean Low level
scores) Effect of factor 2 high
level 0.5(Mean High level scores -mean Low level
scores) Effect of factor 3 high
level 0.5(Mean High level scores -mean Low level
scores)
69
Draw up a histogram and select the combination of
levels that is optimum for each person. Did you
get differences between the two people?