Upload
macroinc
View
196
Download
0
Embed Size (px)
DESCRIPTION
Detailed description of vastly improved Brand Imagery measurement technique
Citation preview
MACRO Consulting, Inc.
A New Approach
Brand Imagery MEASUREMENT
Paul Richard McCullough
Sawtooth Software Conference 2013
MACRO Consulting, Inc.
3MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Ricky Odello, Survey Sampling International
Tom Eagle, Eagle Analytics of California
Kirill Zaitsev, MACRO Consulting, Inc.
Keith Chrzan, Sawtooth Software, Inc.
Christine Lafontaine, MACRO Consulting, Inc.YOUTHANK
4MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
AGENDA
MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Introduction New Approach Case Study
Brand Imagery Measurement
• Current Approach• Issues
• Research Objectives• Research Methodology• Summary of Findings
5MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
AGENDA
MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Introduction
Brand Imagery Measurement
• Current Approach• Issues
6MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Current Approach
Introduction
• Rating Scales
• Each brand is rated independently on each statement in an image battery, eg, 10 point rating scale.
7MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Current Approach Issues
Introduction
1
2
3
4
5
6
7
8
9
10
• Flat Responses Across Statements
• Flat Responses Across Brands
• Brand Halo
• Scale Usage Bias
Ratings Scales
Resulting data are typically non-discriminating and highly correlated.
8MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
AGENDA
MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
New Approach
Brand Imagery Measurement
9MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Brand-Anchored Max/Diff
New Approach
Brand-anchored Max/Diff removes brand halo, scale-usage bias and is more discriminating than rating scales.
10MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Brand-Anchored Max/Diff with Dual Response
New Approach
Dual Response Max/Diff allows for a zero point in Max/Diff utilities, making comparisons across studies (and brands) feasible.
11MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Modified Brand-Anchored Max/Diff
New Approach
Max/Diff takes longer than ratings scales. Modified brand-anchored Max/Diff hopes to decrease the interview time of the Max/Diff Tasks.
12MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Animated Modified Brand-Anchored Max/Diff
New Approach
Animated Modified Brand-Anchored Max/Diff hopes to hold the respondent’s attention longer than traditional Max/Diff.
13MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Direct Binary Response- Positive DBR
New Approach
Dual Response Max/Diff ALLOWS FOR A ZERO POINT in Max/Diff utilities, MAKING COMPARISONS ACROSS STUDIES feasible.
Direct Binary Response is a MORE TIME-EFFICIENT way to collect dual-response data.
However, Dual Response Max/Diff has been shown to RE-INTRODUCE SOME SCALE USAGE BIAS.
14MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Dual Direct Binary Response- Negative DBR
New Approach
By ADDING A SECOND, NEGATIVE DIRECT BINARY RESPONSE QUESTION, we hope to REMOVE or MINIMIZE scale usage bias.
As a FURTHER ATTEMPT to minimize scale use bias, half of respondents will be required to SELECT AS MANY NEGATIVE ATTRIBUTES AS POSITIVE.
15MACRO Consulting, Inc.
A n i m a t e d M o d i f i e d B r a n d - a n c h o r e d M a x / D i f f S c a l i n g w i t h P o s i t i v e a n d N e g a t i v e D i r e c t B i n a r y R e s p o n s e
Summary of New Approach
A n i m a t e d M o d i f i e d B r a n d - a n c h o r e d M a x / D i f f S c a l i n g w i t h P o s i t i v e a n d N e g a t i v e D i r e c t B i n a r y R e s p o n s e ( A M B A M B R )
16MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Analytics-Derived Parsimony
New Approach
Latent Class Choice Models
• With Large Sample• With Covariates
Hierarchical Bayes
• Covariates in upper model• Adjusted priors
2Alternatives
The goal of the above analytic approaches is to minimize the number of Max/Diff tasks each respondent must complete and still estimate disaggregate utilities with acceptable accuracy.
17MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
AGENDA
MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Case Study
Brand Imagery Measurement
• Research Objectives• Research Methodology• Summary of Findings
MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Research Objectives
• Compare two approaches to brand imagery measurement, ratings scales and max/diff, in terms of:- Inter-brand discrimination- Inter-item discrimination- Predictive validity
• Explore alternative methods of estimating max/diff utilities most accurately and most efficiently:- Standard HB - HB with positive Direct Binary Response- HB with positive DBR and unconstrained
negative DBR- HB with positive DBR and constrained
negative DBR- Latent Class Choice- Use of covariates- Tuned priors
MACRO Consulting, Inc. 18w w w . m a c r o i n c . c o m
MACRO Consulting, Inc.
Online Survey:
• Two cells- Rating Scales (n=436)- Max/Diff (n=2,605)
• Three brands
• 12 items• Questionnaire:
- Brand image measurement- Three dependent variables
* Item top 3 rank-order* Brand purchase likelihood* Brand forced-choice preference
- Demographics
Research Methodology
w w w . m a c r o i n c . c o m 19
MACRO Consulting, Inc. w w w . m a c r o i n c . c o m MACRO Consulting, Inc. 20
Summary of Findings
w w w . m a c r o i n c . c o m
In general, AMBAMBR is superior to ratings scales:
- Better inter-item discrimination- Better predictive validity- Fewer unacceptable respondents- Elimination of both brand halo and scale usage bias
Of the AMBAMBR methods tested, the two methods which included negative DBR were superior:
- Positive DBR reinserts brand halo into the data- Positive DBR has slightly weaker inter-item discrimination than either
Negative DBR
AMBAMBR takes longer to administer and has higher incompletion rates
Task set reduction could not be fully explored with these data
MACRO Consulting, Inc. w w w . m a c r o i n c . c o m21
Positive DBR Appears to Show Greater Inter-Item Discrimination Than Rating Scales
Rating Scales
Positive DBR
Item
1Ite
m2
Item
3Ite
m4
Item
5Ite
m6
Item
7Ite
m8
Item
9
Item
10
Item
11
Item
12
5.00
5.50
6.00
6.50
7.00
7.50
8.00
8.50
Brand#1
NewBrand
Brand#2
Item
1Ite
m2
Item
3Ite
m4
Item
5Ite
m6
Item
7Ite
m8
Item
9
Item
10
Item
11
Item
12
-3.00
-2.50
-2.00
-1.50
-1.00
-0.50
0.00
0.50
1.00
Brand#1
NewBrand
Brand#2
MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
22
Negative DBR Approaches Yield Similar Results
ConstrainedNegative DBR
UnconstrainedNegative DBR
Item
1Ite
m2
Item
3Ite
m4
Item
5Ite
m6
Item
7Ite
m8
Item
9
Item
10
Item
11
Item
12
-1.00
-0.50
0.00
0.50
1.00
1.50
2.00
Brand#1
NewBrand
Brand#2
Item
1Ite
m2
Item
3Ite
m4
Item
5Ite
m6
Item
7Ite
m8
Item
9
Item
10
Item
11
Item
12
-1.00
-0.50
0.00
0.50
1.00
1.50
2.00
2.50
Brand#1
NewBrand
Brand#2
23MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Item
1
Item
2
Item
3
Item
4
Item
5
Item
6
Item
7
Item
8
Item
9
Item
10
Item
11
Item
12
5.00
5.50
6.00
6.50
7.00
7.50
8.00
8.50
Brand#1
NewBrand
Brand#2
ConstrainedNegative DBR
UnconstrainedNegative DBR
Rating Scales
Positive DBR
Negative DBR Approaches Bring New Brand Closer
Item
1
Item
2
Item
3
Item
4
Item
5
Item
6
Item
7
Item
8
Item
9
Item
10
Item
11
Item
12
-3.00
-2.50
-2.00
-1.50
-1.00
-0.50
0.00
0.50
1.00
Brand#1
NewBrand
Brand#2
Item
1
Item
2
Item
3
Item
4
Item
5
Item
6
Item
7
Item
8
Item
9
Item
10
Item
11
Item
12
-1.00
-0.50
0.00
0.50
1.00
1.50
2.00
Brand#1
NewBrand
Brand#2
Item
1
Item
2
Item
3
Item
4
Item
5
Item
6
Item
7
Item
8
Item
9
Item
10
Item
11
Item
12
-1.00
-0.50
0.00
0.50
1.00
1.50
2.00
2.50
Brand#1
NewBrand
Brand#2
24MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Ratings No DBR Positive DBR Unconstrained Negative DBR
Constrained Negative DBR
BRAND#1 1.75 4.46 3.90 4.30 4.68
NEW BRAND 0 4.28 3.16 4.25 4.50
BRAND#2 1 4.69 3.78 4.48 4.70
Inter-Item Discrimination Greatest for Negative DBR
Average number of statistically significant differences across 12 items, within brand*
* 10 random draws of n=436 were pulled for all data sets except Ratings
25MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Random Numbers Ratings No DBR Positive DBR Unconstrained
Negative DBRConstrained
Negative DBR
1 OF 1 8% 14% 27% 28% 27% 26%
(1 OR 2) OF 2 32% 30% 62% 64% 62% 65%
(1, 2 OR 3) OF 3 61% 51% 86% 87% 86% 88%
Predictive Validity of AMBAMBR Superior to Rating Scales
Hit Rates for Top 3 Items Ranking
26MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
AMBAMBR Yielded More Valid Completes
Invalid Completes
Max/Diff 4%Ratings 32% 32%
27MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Brand Halo Was Measured Using Confirmatory Factor Analysis
If brand halo exists, halo latent will positively
influence scores on all items
28MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Brand Halo Latent
Ratings No DBR Positive DBR Unconstrained Negative DBR
Constrained Negative DBR
Std Beta Prob Std Beta Prob Std Beta Prob Std Beta Prob Std Beta Prob
ITEM 1 0.85 *** -0.14 *** 0.90 *** 0.44 *** 0.27 ***
ITEM 2 0.84 *** -0.38 *** 0.78 *** -0.56 *** -0.72 ***
ITEM 3 0.90 *** -0.20 *** 0.95 *** 0.42 *** 0.32 ***
ITEM 4 0.86 *** 0.10 *** 0.90 *** 0.30 *** 0.16 ***
ITEM 5 0.77 *** -0.68 *** 0.88 *** 0.03 0.25 0.01 0.78
ITEM 6 0.85 *** -0.82 *** 0.87 *** -0.21 *** -0.24 ***
ITEM 7 0.83 *** 0.69 *** 0.83 *** 0.42 *** 0.20 ***
ITEM 8 0.82 *** 0.24 *** 0.75 *** 0.01 0.87 -0.23 ***
ITEM 9 0.88 *** 0.58 *** 0.90 *** 0.77 *** 0.62 ***
ITEM 10 0.87 *** 0.42 *** 0.94 *** 0.86 *** 0.90 ***
ITEM 11 0.77 *** -0.05 0.015 0.85 *** 0.07 0.02 -0.12 ***
ITEM 12 0.88 na 0.26 na 0.91 na 0.69 na 0.53 na
Ratings and Positive DBR Reflect Strong Brand Halos
29MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Scale Usage Bias Was Measured Using Confirmatory Factor Analysis
Brand halo drives scores within brand. Scale
usage bias drives scores independent of brand.
If scale usage bias exists, the scale usage
latent should load positively on all items
30MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Scale Usage Latent
Ratings No DBR Positive DBR Unconstrained Negative DBR
Constrained Negative DBR
NUMBER OF NEGATIVE LOADINGS
0 14 5 10 15
NUMBER OF STATISTICALLY SIGNIFICANT
LOADINGS
35 30 28 32 29
Only Ratings Reflect Strong Scale Usage Bias
31MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
AMBAMBR Superior But SlowerAMBAMBR Has Higher Dropout Rates
RATINGS AMBAMBR
TOTAL INTERVIEW LENGTH 9.7 MINUTES 15.8 MINUTES
BRAND IMAGE MEASUREMENT 1.7 MINUTES 6 MINUTES
RATINGS AMBAMBR
INCOMPLETION RATE 9% 31%?
Can We Reduce the Number of Max/Diff Tasks to Shorten Interview
Length and Decrease Dropout Rates?
32MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
U N C O N S T R A I N E D Negative DBR C O N S T R A I N E D Negative DBR
8 Tasks 4 Tasks 2 Tasks 8 Tasks 4 Tasks 2 TasksHB LC HB LC HB LC HB LC HB LC HB LC
1 OF 1 27% 19% 21% 20% 20% 19% 26% 21% 24% 21% 22% 22%
(1 OR 2) OF 2 62% 54% 59% 57% 58% 56% 65% 61% 61% 59% 59% 56%
(1, 2 OR 3) OF 3 86% 81% 85% 82% 82% 83% 88% 84% 86% 84% 85% 82%
HB Models May Perform Slightly Better Than LC With Full Task SetsBut All Perform Well
33MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
All HB and LC Models Perform Very Well -
• Hit rates seem relatively unaffected by:
− Number of tasks
− Number of latent classes
− Tuned priors
− Covariates
− Sample size
What’s Going On?
Perhaps Too Well
34MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Perhaps These Data Have Little Heterogeneity
• Category is not emotionally engaging
• Brands are not differentiated− Commodity-like category
− No polarizing brands, eg, Microsoft, Apple or Donald Trump
− Brands with new technologies not yet established
35MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Aggregate Model Works as Well as Disaggregate
U N C O N S T R A I N E D Negative DBR
RandomHB
8 TasksN=1,324
HB2 TasksN=105
HB2 Tasks
Constant utilsN=105
1 OF 1 8% 27% 22% 25%
(1 OR 2) OF 2 32% 62% 59% 61%
(1, 2 OR 3) OF 3 61% 86% 82% 82%
This seems to suggest
there is little
heterogeneity to
capture
36MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Summary and Implications
• The forms of Max/Diff referred to here as AMBAMBR are superior to rating scales for measuring brand imagery:- Better inter-item discrimination- Better predictive validity- Elimination of brand halo- Elimination of scale usage bias- Fewer invalid completes
• Positive DBR alone reintroduces brand halo
• Positive DBR must be combined with some form of negative DBR
• For comparability across brands and time: - Raw utils must be used rather than rescaled utils- Some form of dual response must be used
• AMBAMBR takes longer to administer and has higher incompletion rates
37MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
Further Research
• Can we reduce the number of tasks when brands are heterogeneously perceived?
• How many brands and statements can realistically by accommodated?
• Is constrained or unconstrained negative DBR superior?
• How would traditional dual response format affect these results?
• Is there a better way to evaluate utility performance?
38MACRO Consulting, Inc. w w w . m a c r o i n c . c o m
And The BIG Question:
• Is there a shorter name for this technique than:
Animated Modif ied Brand-anchored Max/Diff Scal ing
with Posit ive and Negative Direct Binary Response?
39
Thank you