Agenda

1
(No Transcript)
2
Agenda
May 21, 2006
Introduction
3
May21, 2006
1
Introduction
4
Introduction

• What are multivariate methods?
• Multivariate
• Bivariate
• Univariate
• The concept of a linear combination lies at the
  heart of most commonly used statistical methods
• Benjamin Franklins moral algebra

5
A Letter from Benjamin Franklin to Joseph Prestly
(1772)
...my way is to divide half a sheet of paper by a
line into two columns writing over the one Pro,
and over the other Con...When I have thus got
them all together in one view, I endeavor to
estimate their respective weights...And, though
the weight of the reasons cannot be taken with
the precision of algebraic quantities, yet when
each is thus considered, separately and
comparatively, and the whole lies before me, I
think I can judge better, and am less liable to
make a rash step, and in fact I have found great
advantage from this kind of equation, and what
might be called moral or prudential algebra.
Decision (Weight1Var1)(Weight2Var2)(WeightK
VarK)
6
Introduction

• Franklin described what we now call linear
  regression.
• Y ß0 (ß1X1) (ßkXk)
• Similar process underlies factor analysis,
  discriminant analysis, conjoint analysis, etc
• The weights are determined by one of various
  statistical algorithms

7
May 21, 2006
2
Multivariate Methods
8
Multivariate Methods Underlying Common Market
Research Projects
9
May 21, 2006
3
Examples
10
May 21, 2006
a
Perceptual Mapping
11
Perceptual Mapping

• Companies use perceptual mapping to understand
  the space that decision makers carry in their
  heads vis a vie the competitive set of products
  in a given market
• This information is used to create a marketing
  strategy that places their products into a
  well-defined and differentiated space
• Natural positioning vs desired positioning
• Several approaches for developing perceptual maps
• Multidimensional Scaling
• Discriminant Analysis
• Factor Analysis
• Correspondence Analysis

12
Perceptual Mapping Example Multidimensional
Scaling. One way to understand how similar two
objects are is to estimate how close they fall to
one another in space.
13
Perceptual Mapping Similarity Matrix. The
distances among cities in miles is analogous to a
correlation matrix.
14
Perceptual Mapping Example Multidimensional
Scaling. Using just the distances given on the
previous slide, can you locate the cities on this
map?
15
Reading Perceptual Maps

• Attributes are plotted as vectors with the arrow
  showing the direction of stronger association.
• Angle between 2 vectors shows strength of
  relationship among the attributes
• Strongly correlated attributes
• No relationship between attributes
• Attributes are negatively correlated
• Vector length proportionate to brand
  differentiation on that attribute
• Brands further from the origin are more highly
  associated with attributes

16
Reading Perceptual Maps
17
May 21, 2006
b
Conjoint/Discrete Choice
18
Conjoint/Discrete Choice

• What are conjoint and discrete choice analysis?
• A series family of tradeoff technique that
  measure product choice/preference
• Research techniques used to model a large range
  of potential product configurations
• Tools to measure the contribution of individual
  product components
• Techniques to understand the underlying drivers
  of product choice

19
Conjoint/Discrete Choice

• Arguably the most popular quantitative technique
  in our industry, because
• High product development costs
• Unlike other industries, product ingredients
  cannot always be determined in advance
• Many variations to be explored among expensive
  research participants
• Key customer (physicians) are arguably quite
  rational decision makers
• Captures the value of future products
• Easily converted to financial measures

20
Choice Model Applications Through the Product
Life Cycle

• New product development
• Positioning
• Pricing
• Line extensions
• New indications
• Impact of competition
• Franchise optimization

21
Features of Choice Models

• Assumes products are bundles of attributes
• Assumes customers are rationale decision makers
• Provides a more realistic estimate of the buying
  decision
• Observe what they do versus what they say they
  will do
• Through experimental design, allows one to model
  very large numbers of variations

22
Choice Models A glossary of terms
Attributes A factor or product feature which
will vary in the model (e.g., efficacy, side
effects, dosing frequency) Levels The
individual value or set of values for that
attribute (e.g., for dosing frequency QD, BID,
TID) Utilities The value or contribution of an
individual level of an attribute Card A product
profile which consists of a set of attributes and
levels Card deck A group of product profiles
included in the questionnaire Scenarios A
product profile or set of profiles Allocation
exercise Questions to determine the extent to
which participants will choose one product or
others Choice Set or Competitive Set The list
of available products Simulator Software which
allows the user to generate share estimates for
any product configuration Share of Preference
Output of market simulations. Not market share.
23
Conjoint vs Discrete Choice

• Conjoint
• Traditionally individual level models
• Used when more is unknown
• More variables
• More levels
• Often requires self-explicated exercises or
  bridging techniques
• Used earlier in product life cycle

• Discrete Choice
• Traditionally aggregate or group level models
• Used when more is known about a product
• Fewer variables
• More is fixed
• Product is closer to launch
• Requires fewer scenarios

Most of the models we use are a hybrid of the two
24
Examples of Attributes and Levels
EFFICACY Efficacy rates equivalent to Product X
0.5 Efficacy rates 10 lower than Product X
0.5 Efficacy rates 15 lower than Product X 0.5
PRICE Price equivalent to Product X 0.5 Price
10 higher than Product X 0.5 Price 15 higher
than Product X 0.5 Price 10 lower than Product
X 0.5
DOSING    B.I.D. T.I.D.
AVAILABILITY OF GENERIC Yes No
SIDE EFFECTS Incidence of burning/stinging
equivalent to Product X 0.5 Incidence of
burning/stinging 20 less than Product X
0.5 Incidence of burning/stinging 30 less than
Product X 0.5 Incidence of burning/stinging 40
less than Product X 0.5
PREVENTION INDICATION Yes No
25
Selecting Attributes

• Attributes must be clearly defined in the
  language of the customer
• Be mindful of the number of attributes you
  include in your model
• Include only those that truly vary
• Fewer is better than more
• Avoid double digits
• Two many will create respondent burden and noise
  in the data
• Avoid attributes that you know will have little
  influence on choice (e.g., tablet color)

26
Types of Levels

• Numerical (e.g., incidence of 10, 2 per day)
• Comparative (10 better than Flovent)
• Presence or Absence (indicated, studied, no data
  available)
• Availability (generic available, no generic
  available)

27
Selecting Levels

• Selecting the range of the levels is a critical
  design component
• In choice models, attribute importance is a
  function of the sensitivity from the highest to
  the lowest level in the attribute
• Price range from 2 to 14,000 (high importance)
• Price range from 2 to 5 (less importance)
• Levels should be developed based on what you know
  about the attribute
• Do not select levels that are unachievable or
  impossible
• You can interpolate between levels you cannot
  extrapolate outside them.
• 2 4 6 10 interpolate 8
• 2 4 6 8 cannot extrapolate 10

28
Attributes and Levels Whats wrong with these?
29
Common Challenges

• Getting consensus on attributes and levels
• Limiting attributes and levels to only those most
  meaningful
• Ensuring a manageable respondent task
• Handling comparisons to existing products

30
Sample Size Considerations

• As in other quantitative studies, more is better
  to a point
• Sample size is often driven by budgetary
  constraints
• Sample size is contingent upon the number of
  attributes and levels which dictates the number
  of scenario cards
• Rule of thumb is that there should be absolutely
  no less than 35 respondents per cell
• Heterogeneous groups require larger samples
  (e.g., PCPs, Patients) than homogeneous ones
• 100-125 for PCPs
• 50-75 for Specialists

31
Choice Model Output
32
Choice Model Output

• Key market drivers
• Sensitivity to levels of product performance
• Patient share estimates given various product
  configurations and availability scenarios

33
Attribute Importance
34
Utility Values

• Measures sensitivity in changes from one level of
  an attribute to another
• Slope of curve illustrates the importance of the
  attribute
• Often shown on a rescaled basis to enable
  comparisons across attributes
• Zero does not equate to 0 value but to least
  desirable level

35
Utility Values
36
Sensitivity Analysis
Change from Base Case Share Points
20, 6.0
14 days, 1.3
7 days, 3.2
3 weeks, 1.9
40 more, -1.3
Base Case
37
Simulations
38
Product Simulations

• One of the most valuable output of a choice model
  is the ability to simulate vast numbers of
  alternative product configurations and market
  scenarios
• Simulations produce estimates of preference
  shares of patients for any combination of
  attributes and levels
• Simulations allow you to play endless what-if
  games
• A simulator makes a choice model a living
  resource which is useful well after the study is
  completed

39
Simulations
40
Simulations
41
Simulator

• Typically a set of Excel macros that converts
  individual or group utilities to share estimates
  based on results of the choice exercises
• Allows user to select product and market
  scenarios
• Allows you to look at results overall or by
  segment (e.g., patient type, physician type)
• Produces estimated shares for any product
  configuration as well as other products in the
  choice set

42
Simulator Output
43
Choice Models Do Not Produce Market Shares

• Assumes perfect knowledge
• Assumes 100 awareness
• Assumes full availability
• Assumes perfect performance
• Does not account for competitive response
• Does not account for promotional expenditure or
  effectiveness

Choice Models do produce share of mind or
preference shares. They are are input to the
forecast, not the forecast itself.
44
Choice Model Examples
45
Linked-Choice Models

• Goal To understand the impact of different
  customer groups on overall product success
• Separate models are built for each
  decision-making group
• Physicians, patients, payers
• Link results from one decision-making group to
  another
• Iteratively analyze the results

46
Choice Models for New Product Development

• Large numbers of attributes
• Testing various possible outcomes
• Hybrid conjoint approach
• Entire choice model may be a bundle of attributes
  rather than any fixed profile
• Helping to provide input to clinical trial
  designs
• May simulate clients product as well as other
  competitors

47
Choice Models for Pricing

• Typically conducted a few months prior to launch
• After positioning research is completed
• Present fixed profile of the product which
  includes some of the selling language
• Vary price and perhaps final labeling in the
  choice exercise
• Discrete Choice approach

48
Choice Models for Line Extensions and New
Indications

• Can use choice models to measure the impact of
  new data, new formulations, new indications or
  other changes to your product
• Useful to determine the value of changes to your
  product
• Valuable for assessing Phase IIIb studies
• Can help to determine if it is worth funding
  further product development for an existing brand
• Discrete Choice approach with variables such as
  new data, new forms, new indications, new dosing
  schedule

49
Choice Models for Franchise Optimization

• Can use choice models to measure the impact of
  your new products or follow up compounds
• Measure the effect that the new brand will have
  on your older brand
• Determine how to avoid cannibalizing older brand
• Identify ways to grow franchise overall
• Discrete Choice or Conjoint depending upon
  numbers of variables

50
May 21, 2006
c
Market Segmentation
51
Why Segment?

• Segmenting the marketplace is a proven way to
  enhance sales and marketing efficiency
• Customers have different needs
• Customers have varying levels of attractiveness
• Depending upon the market, segmentation can
  support sales and marketing strategies by
• Targeting Promoting to those customers where
  promotion will yield the greatest ROI
• Tailoring Delivering message to each customer
  selected to match that customers needs
• Tactical Implementation Developing optimal
  media mix at the level of the individual customer

52
Issues That Will Drive Segmentation Research
Design Considerations

• Blockbuster product or not?
• Consumer segmentation and/or physician?
• Availability depth of secondary data
• Stage in the product life cycle
• U.S. vs non-U.S.

53
Segmentation Research

• Compared to other quantitative research
  techniques
• More customization
• Not one solution
• Requires greater collaboration
• Design
• Analysis
• Presentation
• Implementation

54
Using Multiple Types of Data to Segment

• Demographic Who they are
• Behavioral What they do
• Attitudinal Why they do what they do

55
Segmenting Physicians Requires a Multi-Faceted
View
Behavior

• Needs
• Treatment attitudes
• Product perceptions
• Medical orientation

Attitudes
Demographics
Decision Style
56
Consumer Segmentation
Behavior

• Disease involvement
• Physician relationship
• Impact of disease
• Comfort w/Rx

Attitudes
Demographics
Information Gathering

• Media use
• Innovation/Adoption
• Rx request

57
Comprehensive Segmentation Objectives--Physician
s

• Quantify and characterize physician prescribing
  behavior within the market
• Segment physicians based on their prescribing
  behaviors, attitudes, needs, opinions and
  perceptions of my product and other key
  competitive products used in treating the
  condition
• Identify homogeneous and actionable physician
  segments with high potential
• Tie results from segmentation research to sales
  implementation

58
Comprehensive Segmentation Objectives--Consumers

• Quantify and characterize the consumer/patient
  market for my product
• Segment consumers based on their medication
  seeking and taking behaviors, attitudes, needs,
  opinions and perceptions
• Identify homogeneous and actionable consumer
  segments with high potential
• Prioritize the segments with the greatest
  likelihood to seek and accept my product
• Determine the best messages and media to reach
  these consumers

59
Steps in a Comprehensive Segmentation Study

• In-depth Qualitative Understanding
• Gathering Reviewing Available Data (Individual
  Level Rx data when available)
• Structured Survey Research
• Analysis
• Implementation

60
1. In-Depth Qualitative Understanding

• Actionable segmentation demands insightful
  qualitative analysis prior to initiating the
  quantitative stage
• Identify dimensions
• Understand terminology and language
• Develop hypotheses

61
2. Gathering Reviewing Individual Level Rx Data
(when available)

• Rx Profiling
• Analyze individual Physician Level Prescription
  (IPLRx) data (IMS, NDC)
• Performed separately for
• Specialty
• Prescribing segments
• Group Practice

62
3. Structured Survey Research

• Self reported current Prescribing/Practice
  Demographics
• Attitudinal battery
• Treatment of condition/disease
• Classes of medications used
• Importance of patient characteristics,
  demographics in choice of medication
• New product adoption
• Confidence and expertise in treating
• Unmet needs
• Clinical orientation
• Presentation/reactions to product profile
• Prescribing allocation

63
Sample Size and Selection

• Representative physician and consumer samples
• Sample sizes must be large enough to derive
  segments accurately and have confidence in their
  stability
• The larger the sample size, the greater the
  number of segments that can be reliably
  identified
• Physician minimums in the range of 200-300
• Consumer minimums of 400

64
4. Analysis

• Determining the links between attitudes, behavior
  and demographics and reducing the data set
  
  (Factor Analysis, Canonical
  Correlation)
• Grouping customers into like-minded segments
  (Cluster Analysis, Latent
  Class, Individual Level Choice Models)
• Profiling the segments/identifying differences
  among groups (Univariate and Bivariate Analysis,
  Analysis of Variance)
• Validating the Segments using a holdout
  sample(Discriminant Function Analysis, CHAID,
  CART)
• Scoring or tagging the universe

64
65
Sizing the Segments
Product X Segments
66
Profiling Segments
67
Describing the Segments
Big Guns (34)

• Less satisfied with Gorillamycin but very
  interested in a new alternative
• Primarily PCP
• More group practice
• Heavy patient load patients
• Concerned with overgrowth
• A new brand has to be reliable, safe.
• Price sensitive

68
5. Making it Actionable

• While all segmentation is extremely valuable for
  understanding the market and prioritizing
  targets, additional value comes in targeting
  individual customers.
• Developing an algorithm and tagging the call list
  or scorecard for reps to tag doctors
• Final challenge to segmentation research is
  implementation
• Getting the sales organization to use the data
• Identified doctors that should not be called on