wemba b, causal research, conjoint analysis entitle insurance

49
1 WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance Market Intelligence Julie Edell Britton Session 7 September 25, 2009

Upload: quincy

Post on 12-Jan-2016

34 views

Category:

Documents


0 download

DESCRIPTION

WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance. Market Intelligence Julie Edell Britton Session 7 September 25, 2009. Today’s Agenda. Announcements WEMBA A Causal Research – Experiments Pre-experimental Designs True Experiments Factorial Designs and Interaction Effects - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

1

WEMBA B, Causal Research,Conjoint AnalysisEntitle Insurance

Market IntelligenceJulie Edell Britton

Session 7September 25, 2009

Page 2: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

2

Today’s Agenda•Announcements

•WEMBA A

•Causal Research – Experiments

•Pre-experimental Designs

•True Experiments

•Factorial Designs and Interaction Effects

•Conjoint Analysis

Page 3: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

3

Announcements

• Submit IBM Global Mobile Computing slides by 10 pm tonight!

Page 4: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

44

WEMBA (A): School Choice Model

ValuesPerceptionsIndividual Differences & Constraints

Become a Duke MBA

Assumes that behavior is driven by differences in:Values (Importance of key attributes)Perceptions (Duke and Competition on key attributes)Individual Differences & Constraints (travel, cost, etc.)

Page 5: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

5

The Funnel

Matriculate

Admitted

Opt Out

Apply

SelectedOut

Attend Information SessionDo not attend Information Session

Do not apply Do not apply

Page 6: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

66

The Analysis Approach• Sample groups that differ in behavior

• Compare the groups on relevant dimensions:• Perceptions• Values• Individual Difference & Constraints

• Infer that any difference found between groups are partly responsible for differences in behavior

Page 7: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

77

WEMBA B

• What factors drive application?• Perception of Duke – Perception of Comp

• Individual difference measures (demos, % paid by company, etc.)

• Conditional on applying, what drives acceptance?

• How do info sessions alter perceptions of Duke?

• Who should Nagy target, and how can he reach target?

• What perceptions might Nagy try to alter with info sessions?

Page 8: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

8

Today’s Agenda•Announcements

•WEMBA A

•Causal Research – Experiments

•Pre-experimental Designs

•True Experiments

•Entitle Case

•Factorial Designs and Interaction Effects

•Conjoint Analysis

Page 9: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

9

Causal Research - ValidityThe strength of our conclusions

i.e., Is what we conclude from our experiment correct?

Threats to Validity

9

History: an event occurring around same time as treatment that has nothing to do with treatment

Maturation: people change pre to post

Testing: pretest causes change in response

Instrumentation: measures changed meaning

Statistical Regression: Original measure was due to a random peak or valley

Page 10: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

1010

Online Investor Performance

X = brick and mortar brokerage customer moves online to trade in 1999

O = Annualized turnover 1998 – 40% annualized turnover 2000 – 100% annualized turnover Did going online cause people to trade

more actively? Threats with one-group pre-post?

Page 11: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

1111

Quasi-Experimental Designs: Interrupted Time Series

Same as one-group pretest posttest, but observations at many points in time before and after key treatment for same people:

EG O1 O2 O3 X O4 O5 O6

Extra time periods help control for history, maturation, testing. “Quasi-experiment”

Page 12: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

1212

Online Investor Performance

-10

-5

0

5

10

15

-36

-33

-30

-27

-24

-21

-18

-15

-12 -9 -6 -3 0 3 6 9 12 15 18 21 24 27 30 33 36

Event Month (0 = month of first online trade)

Cu

mu

lati

ve M

arke

t A

dju

sted

Ret

urn

(%

)

Gross Returns

Net Returns

Page 13: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

1313

Portfolio Turnover

0%

20%

40%

60%

80%

100%

120%

140%

-24

-21

-18

-15

-12 -9 -6 -3 0 3 6 9 12 15 18 21 24

Event Month (0 is month of first online trade)

An

nu

aliz

ed

Tu

rno

ve

r (%

)

Size-Matched

Online

Page 14: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

141414

2 Groups: Unmatched Control Group(Effect of Prior Knowledge on Search)

Hypothesis: People with little knowledge about cars search less online

100 Durham residents who are in the market for a car

Experimental Group X1 (Auto Shop Course) O1 (6 hrs online)

---------------------------------------------------------

Control Group X2 (Electronics Course) O2 (3 hrs online)

Page 15: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

151515

2 Groups: Matched Control Group (True Experiment)

Experimental Group R X1 (Auto Shop Course) O1 (6 hrs) ---------------------------------------------------------

Control Group R X2 (Electronics Course) O2 (3 hrs)

Control for Selection Threat

Key Point: For causal research, chance (not respondent) must determine

respondent assignment to condition.

Page 16: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

1616

16

Breckenridge Brewery Ads Breckenridge Brewery wants to assess the efficacy of TV

ad spots for its new amber ale.

Time 1 (O1): Duke undergrads are brought to the lab and asked to rate their frequency of buying a series of brands in various categories over the past week. The list includes Breckenridge Amber Ale. Mean = 0.2 packs per week.

Time 2 (X): Two weeks of ads for Breckenridge Ale.

Time 3 (O2): Same Duke undergrads brought back to lab to rate frequency of buying same set of brands over past week. Mean = 1.3 packs per week.

1.3 - 0.2 = 1.1 increase in number of packs per week.

Page 17: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

17

2-group Before-After Design

• Now add a randomly assigned “Control” group with mean scores O1 = 0.3, O2 = 0.5.

O1 O2 O2 - O1Difference

ExperimentalO1 X O2 0.2 1.3 1.1

ControlO1 O2

0.3 0.5 0.2

Page 18: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

1818

Factorial Designs

Independent Variable: Factor manipulated by the researcher

Dependent Variable: Effect or response measured by researcher

Factorial Design: 2 or more independent variables, each with two or

more levels. All possible combinations of levels of A & levels of

B.

Page 19: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

19

Oreo Promotion Experiment

Kroger: Supporting a discount on Oreo cookies

Factor A: Ads in local paper a1 = no ads a2 = ad in Thursday local paper Factor B: Display location b1 = regular shelf b2 = end aisle

Page 20: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

20

Oreo Promotion Experiment(Expenditures/customer/2 wks)

a1 = no ads a2 = ads R ow A ve

b1 = regu lar shelf

.60 .90 .75

b2 = end a isle

.65 .95 .80

C o l. A ve .625 .925

Page 21: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

21

Sales of Oreos on Promotion as function of Local Advertising, Display Location

Sales of Oreos with Ads and Display

0

0.2

0.4

0.6

0.8

1

1 2

No Ads Ads

$/C

ust

om

er/2

wee

ks o

n

Ore

as Regular Shelf

End Display

Page 22: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

22

Oreo Example, No Interaction

Main Effect of A (Ads)? Main Effect of B (Display Location)? No AxB (say A by B) interaction. Effect of

changing A (Ads) is independent of level of B (Display Location). Sales go up by $0.30 when you advertise, regardless of location.

Implies that Ad & Display decisions can be decoupled…they influence sales additively.

Page 23: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

2323

Managerial Implications of Interactions

If two controllable marketing decision variables interact (e.g., advertising x display), implication is that you can’t decouple decisions; must coordinate.

If A is a controllable decision variable and B is a potential segmentation variable (e.g., ads x urban/suburban), interaction means that segments respond differently to this lever.

Page 24: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

24

Interactions and segmentation

c

Exposure, Attention, & PerceptionPsychology of Consumers

Page 25: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

25

Sales of Oreos on Promotion as function of Local Coupons, ay Location

Sales of Oreos with Coupon in Suburbs

and Urban Areas

0.00

0.20

0.40

0.60

0.80

1.00

No Coupon

$/C

ust

om

er/2

wee

ks

on

Ore

os

Suburbs

Urban

Coupon

Page 26: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

26

Analyzing Factorial Design in SPSS

1

5

10

AdtypeInformational Emotional Transformational

Exposures

n = 9 per cell

Page 27: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

27

Page 28: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

28

Page 29: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

29

Page 30: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

30

Page 31: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

31

Page 32: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

32

SPSS Output

Page 33: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

33

Estimated Means

Page 34: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

34

Page 35: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

35

Takeaways for Causal Research

Threats to validity in pre-experimental and quasi-experimental designs

Factorial Designs – Main effects and interactions 2 marketing tactics interact coordinate Marketing tactic interacts with customer classification

implies classification a potential basis for segmentation…different sensitivities to some marketing mix variable

Page 36: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

36

Today’s Agenda•Announcements

•WEMBA A

•Causal Research – Experiments

•Pre-experimental Designs

•True Experiments

•Factorial Designs and Interaction Effects

•Conjoint Analysis

Page 37: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

3737

•Conjoint analysis: family of techniques to measure customer preferences, tradeoffs.

CONJOINT ANALYSIS

Page 38: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

3838

Applications

•New product concept identification•Pricing•Benefit segmentation•Competitive analysis•Repositioning or modifying existing products

Page 39: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

3939

Modeling a Single Consumer

•Sysco wants to create first class lunch defined on:

•Appetizer• a1 = Mushroom tart•a2 = Shrimp cocktail

• Salad/Vegetable• b1 = Tossed salad• b2 = Fresh asparagus

• Entree• c1 = Fried grouper• c2 = Sole bonne femme

Page 40: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

40

• Goal•Find the combination of appetizer, salad/veggie, and entree that will be most attractive to customers who are buyers at major airlines

• Procedure•Customer evaluates subset of combos (15-pt scale)•Estimate “average liking” item effects •Forecast liking of all combos•Design optimal meal for that customer

Goal and Procedure

Page 41: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

41

Imagine a customer who obeys an additive model:

Overall Liking (ijk) = u a(i) + u b(j) + u c(k) =for Whole Meal

Utility / liking for Appetizer (i) + Utility / liking for Salad/Veg (j) + Utility / liking for Entrée (k)

And further, suppose:

Mushroom tart u (a1) = -2Shrimp cocktail u (a2) = +2Salad u (b1) = +1Asparagus u (b2) = +4Grouper u (c1) = +4Sole u (c2) = +6

Page 42: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

42

We cannot observe these true utilities (the u’s) directly, but we can observe the overall ratings R(ijk)

a1 = Tart a1 = Tart a2 = Shrimp Cocktail

a2 = Shrimp Cocktail

b1 = salad

b2 = asparagus

b1 = salad

b2 = asparagus

c1 = grouper

-2 + 1 + 4 = 3

-2 + 4 + 4 = 6

+2 +1 +4 = 7

+2 +4 +4 = 10

c2 = sole

-2 +1 +6 = 5

-2 +4 +6 = 8

+2 + 1 + 6 = 9

+2 +4 + 6 = 12

Page 43: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

4343

Notice there is no interaction of preferences across attributes. When this holds, we can get a separate interval scale of “part-utility” from the marginal means for each factor: a + b (part Util)

A: R(1..) = 5.5 B: R(.1.) = 6.0 C: R(..1) = 6.5R(2..) = 9.5 R(.2.) = 9.0 R(..2) = 8.5

1. Because these share a common unit, differences between two levels of factor A can be compared meaningfully to differences between two levels of B and C. Appetizer factor A twice as important as entrée factor C.

2. Because these scales have different and unknown intercepts, we cannot compare the absolute level of one level of factor A to that of a single level of factor B or C. e.g., Though R(2..)= 9.5 for shrimp > R(..2) = 8.5 for sole, u(a2) = +2 for shrimp < u(c2) = +6 for sole.

Page 44: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

4444

Imagine a customer who obeys an additive model:

Overall Liking (ijk) = u a(i) + u b(j) + u c(k) =for Whole Meal

Utility / liking for Appetizer (i) + Utility / liking for Salad/Veg (j) + Utility / liking for Entrée (k)

And further, suppose:

Mushroom tart u (a1) = -2 R(1..) = 5.5Shrimp cocktail u (a2) = +2 R(2..) = 9.5Salad u (b1) = +1 R(.1.) = 6.0Asparagus u (b2) = +4 R(.2.) = 9.0Grouper u (c1) = +4 R(..1) = 6.5Sole u (c2) = +6 R(..2) = 8.5

Page 45: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

45

Tradeoffs

Which meal would this guy prefer?

Option 1 Option 2Shrimp Cocktail Mushroom TartSalad AsparagusGrouper Sole

Page 46: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

46

Same Conclusions from Subset

Critically, we can get the same utility scales if we ask only for a specially chosen subset of all 8 possible combinations:

Combo Customer RatingMushroom tart, salad, grouper 3Mushroom tart, asparagus, sole 8Shrimp cocktail, salad, sole 9Shrimp cocktail, asparagus, grouper 10

Guess the average evaluation of untested combinations?

Page 47: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

47

a1 = tart a1 = tart a2 = shrimp cocktail

a2 = shrimp cocktail

b1 = salad

b2 = asparagus

b1 = salad

b2 = asparagus

c1 = grouper

3 10

c2 = sole

8 9

Goal: Compute expected evaluation of remaining four combos so we can pick the best out of 8.

Overall Average? = 7.5 Deviation from 7.5?a1=Mushroom tart Average = 5.5a2=Shrimp cocktail Average = 9.5b1=Salad Average = 6.0b2=Asparagus Average = 9.0c1=Grouper Average = 6.5c2=Sole Average = 8.5

Page 48: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

48

Now let’s consider how much of a bump up or down we get from the overall average (7.5) for each attribute level.Overall Average? = 7.5 Deviation from 7.5?a1= Mush. Tart Avg = 5.5 5.5 – 7.5 = -2a2= Shrimp Average = 9.5 9.5 – 7.5 = +2b1=Salad Average = 6.0 6.0 – 7.5 = -1.5b2=Asparagus Avg = 9.0 9.0 – 7.5 = +1.5c1=Grouper Average= 6.5 6.5 – 7.5 = -1c2=Sole Average = 8.5 8.5 – 7.5 = +1Compute predicted rating of missing cells by saying:Overall Average + Dev a(i) + Dev b(j) + Dev c(k)

e.g., Tart (a1), Salad (b1), Sole (c2) = 7.5 + (-2) + (-1.5) + (+1) = 5

a1 = Mushroom

a1 = Mushroom

a2 = Shrimp a2 = Shrimp

b1 = salad

b2 = asparagus

b1 = salad

b2 = asparagus

c1 = grouper

3 7.5- 2+1.5– 1 = 6

7.5+ 2–1.5–1 = 7

10

c2 = sole

7.5–2–1.5 + 1 = 5

8 9 7.5+2+1.5+1 = 12

Page 49: WEMBA B, Causal Research, Conjoint Analysis Entitle Insurance

49

a. Best meal?

b. If you now sell a1, b1, c1, what single change is best? What if you sell a2, b1, c1?

c. Most important attribute?

d. Can also cluster individual customers based on their part-utility differences for each attribute to get “benefit segments.”

e. Can make market share forecasts (next)

f. Can use for pricing, when price is an attribute

What can we conclude?