Aiding Consumer Decisions on the Web Gary McClelland University of Colorado @ Boulder with assistance from Barbara Fasolo & Katharine Lange Presented at.

Aiding Consumer Decisions on the Web

Gary McClelland

University of Colorado @ Boulder

with assistance from

Barbara Fasolo & Katharine Lange

Presented at The Wharton SchoolUniversity of Pennsylvania

26 February 2001

Today’s Tour

• Consumer-Aiding Websites• Winnowing via EBA, LEX, MCD,

WADD, MAUT, etc.• Why Attribute Correlations Matter• Typical Attribute Correlations• Effects of Attribute Correlations

Decision Guides

Decision Guides

MouseLab IDB (ca. 1990)

Consumer Reports

Consumer Reports Online

Paid Subscriptions:532,000 in Feb 2001

IDB—Options x Attributes

www.decide.com

IDB—Attributes x Options

www.point.com

IDB—46 x 5

www.activebuyersguide.com

IDB—Continued


IDB—Still More


IDB—More Yet!


IDB—The End


Many, Many Options

www.personalogic.com www.point.com

Aiding the Consumer

•Winnowing•Comparing•Evaluating•Recommending•Choosing

Company Sites:Not Much Help

www.panasonic.com

Compared to What?

www.panasonic.com

Additional Info

www.panasonic.com

Compared to Other Phone?

www.panasonic.com

Automobile Sites:This is Decision Help?

IDBs Designed for Decision MakingSort by Attribute, Eliminate Options, Choose

www.cdw.com

Decision Sites in Transition

Retail Sales 2000:4Q

• Online: Up 36% to $8.7 billion• Online: > 1.1 % of total retail• Total: Up 5.4%

Source: Reuters, 16 Feb 2001

Today’s Tour



Winnowing Options

• Setting Attribute Cutoffs (EBA)• Sorting along Attributes

(LEX,TB)• Weighting Attributes (WADD)• Measuring Tradeoffs (MAUT)

Winnowing:LexicographicO1 O2 O3 O4 O5

A1 ++ + + + 0A2 – 0 + – ++A3 – + ++ + ++A4 – 0 –– + ++A5 – 0 0 + 0

LEX

Winnowing:Elimination by Aspects

O1 O2 O3 O4 O5

A1 ++ + + + 0A2 – 0 + – ++A3 – + ++ + ++A4 – 0 –– + ++A5 – 0 0 + 0

EBA

Winnowing:Satisficing

O1 O2 O3 O4 O5

A1 ++ + + + 0A2 – 0 + – ++A3 – + ++ + ++A4 – 0 –– + ++A5 – 0 0 + 0

SAT

Winnowing: Most Confirming

DimensionsO1 O2 O3 O4 O5

A1 ++ + + + 0A2 – 0 + – ++A3 – + ++ + ++A4 – 0 –– + ++A5 – 0 0 + 0

MCD

Winnowing:Adding (Equal Wts)

O1 O2 O3 O4 O5

A1 ++ + + + 0A2 – 0 + – ++A3 – + ++ + ++A4 – 0 –– + ++A5 – 0 0 + 0

ADD

Winnowing:ImplicationsO1 O2 O3 O4 O5

A1 ++ + + + 0A2 – 0 + – ++A3 – + ++ + ++A4 – 0 –– + ++A5 – 0 0 + 0

LEX SAT EBA MCD ADD

Attribute Processing

Opt 1 Opt 2 Opt 3

Att A VA1 VA2 VA3

Att B VB1 VB2 VB3

Att C VC1 VC2 VC3

EBA orLEX orTakeBest

Elimination-by-Aspects

www.point.com

Elimination-by-Aspects

www.point.com

EBA—Are You Sure?


EBA & LEX

www.decide.com

Option Processing

Opt 1 Opt 2 Opt 3

Att A VA1 VA2 VA3

Att B VB1 VB2 VB3

Att C VC1 VC2 VC3

WADD or MAUT Score

WADD—Getting the Wts

www.personalogic.com

WADD—Weights & Values

mro.frictionless.com

Weight Profiles


WADD—Option Score


MAUT—Tradeoffs


MAUT—Global


Collaborative Filtering

• www.amazon.com

• movielens.umn.edu

• www.imdb.com

Today’s Tour


WADD, MAUT, etc.• Why Attribute Correlations

Matter• Typical Attribute Correlations• Effects of Attribute Correlations

Positive Correlation = “Friendly Decision”

r = .21

Weight Insensitivity3 X1 + X2X1 + 3 X2

r= .21

Attribute AgreementX1 X2B BA GG FC CE DF ED A

r = .21

= .24

Equal Weights History

• Wilks (1938)• Gulliksen (1950)• Dawes & Corrigan (1974)• Einhorn & Hogarth (1975)• Wainer (1976)• Meehl (1999)

Equal Wts Correlations

Markets -> Nondominated Options

r = -.87

Weight Sensitivity!

3 X1 + X2X1 + 3 X2

r= -.87

Nondominated Shapes

Equal Wts Value Loss

X1 X2A EB DC CD BE A

Attribute Disagreement

r = -.87

= -1.0

Today’s Tour



Real Choice Sets

• What are their attribute correlations?

• Are they approximately nondominated sets?

• Consumer Reports

Consumer Reports

Example CR Choice SetMtn Bikes

r = -.82

-P/Q Correlations from CR

-.17 Air Conditioners

-.27 Bike Helmets

+.16 Dishwashers

-.82 Mtn Bikes

-.74 Printers

-.49 Pro Ranges

-.28 Fridge

-.59 27” TVs

-.16 Vacuums

-.56 Wall Oven

-P/Q Correlations

Attribute Correlations from CR

+.35 Air Conditioners

+.07 Bike Helmets

-.03 Dishwashers

+.37 Mtn Bikes

+.18 Printers

+.47 Pro Ranges

-.06 Fridge

+.05 27” TVs

+.12 Vacuums

+.04 Wall Oven

Average r = –.05

O1 O2 O3 O4 O5

A1 ++ + + + 0A2 – 0 + – ++A3 – + ++ + ++A4 – 0 –– + ++A5 – 0 0 + 0

LEX SAT EBA MCD ADD

Today’s Tour


WADD, MAUT, etc.• Why Attribute Correlations Matter• Typical Attribute Correlations• Effects of Attribute

Correlations

WebIDB

Sample Data Streams

Att Opt Time0 1 276

0 0 15951 0 8510 0 8590 1 8360 0 5350 1 9750 2 6520 3 6520 4 543

Att Opt Time1 4 5571 3 1660 2 2341 2 4721 1 765

2 1 2111 2 2 2519 2 3 10312 4 448

Attribute Focus

WebIDB cf. MouseLab

• WebIDB replicated MouseLab results– Attribute Focus is the default strategy– Increasing Attributes -> Attribute Focus– Increasing Options -> Less info, more var.

• Different result– Somewhat more information viewed in

WebIDB– Somewhat greater attribute focus in

WebIDB

Prior Research on Correlation Effects

• Johnson, Meyer & Ghose (1989)

• Theory: Negative –> Attribute-based

• Results: Null

• Bettman, Johnson, Luce & Payne (1993)

• Theory: Negative –> Option-based

• Results: Negative –> Option-based

Experiments

• Study 1– 8 att x 5 opts– Attribute

Correlation: Pos (.5) vs. Neg (-.14)

– 8 Matrices– Between Subjects

• Study 2– Within Subjects:

Switch after 4 Matrices

Proportion of Cells Visited

Attribute Visit S.D.

EqualAttention

SelectiveAttention

Payne Index =(Opt-Att)/(Opt+Att)

AttributeFocus

OptionFocus

Self-Ratings

Switch Attribute Correlation

Results Summary

• Default Strategy is Attribute Processing

• Negative Correlation —> Option Processing

• Immediate Sensitivity to Correlation• Quickly Switch to Option Processing• Amount of Information Constant

Research Questions

• What winnowing strategies do consumers use?

• Attribute-based unless forced towards Option-based by negative attribute correlations

Research Questions

• What winnowing strategies might consumers be willing to use if aided?

• And how do attribute correlations affect the use of such aids?

Correlations and EBA

Correlations and WADD

End of Tour



Aiding Consumer Decisions on the Web Gary McClelland University of Colorado @ Boulder with assistance from Barbara Fasolo & Katharine Lange Presented at.

Documents

eba slide

lex slide

mcd slide

transition slide

takebest slide

lexsatebamcdadd slide

maut score slide

consumer winnowing