文档介绍：Bootstrap Penalty
Analysis
Alternatives to distributions and
penalty analysis
Jean-Francois and Rui Xiong
Introduction
Hedonic and JAR (just-about-right) scales
are widely used together to provide directional
information for product reformulation or
optimization
Like extremely
Like very much Much tom Much
Like moderately Too much
Like slightly Just about right
Neither like nor dislike
JAR scale Too little
Dislike slightly Much too little
Hedonic scale
Dislike moderately
Dislike very much
Dislike extremely S
Introduction
Results from diagnostic attributes
are not always actionable
What is the percentage of consumers
required on the too little or too much
side to consider an attribute to be at a
inappropriate level?
If an attribute is not at its optimal
level, does that have an impact on
product liking?
S
Diagnostic results
100%
80%
Much Too Strong
60% Too Strong
JAR
40% Too Weak
Much Too Weak
PercentResponses
20%
0%
color nutty
banana
firmness
chocolate
sweetness S
Introduction
Simple graphical
method for assessing
the cost associated with
having an attribute not
at its optimum level Liking
A graphical technique,
understandable to
managers Sensory Level
Ignoring correlations
among attributes
Not a regression method
S
Penalty Analysis
Banana
50 Much Too
Consumers are split Weak
is 3 groups (TL, JAR, TM) 40 Too Weak
30
JAR
Penalties not 20
Frequency Too Strong
10
calculated if Much Too
0
proportion of Strong
consumers is less
than 20%
Liking Liking Liking
score score score
Penalty=Yjar-Y<or> for for for
this this this
group, group, group,
Y< Yjar Y> S
Penalty Analysis
The major limitations of penalty analysis are:
The fact that categories below and above JAR level are
collapsed (. because n is often not large enough
within a single category)
Collinearities be