Social Learning and Consumer Demand

1
Social Learning and Consumer Demand

• Markus Mobius (Harvard University and NBER)
• Paul Niehaus (Harvard University)
• Tanya Rosenblat (Wesleyan University and IAS)
• CMPO, 2 June 2006

2
Motivation

• We want to study social learning in the context
  of how consumer preferences form.
• How strong are social learning effects absolutely
  and relatively compared to informative
  advertising?
• How strong are social influence effects (on
  valuations) absolutely and relatively compared to
  persuasive advertising?
• Which agents are influential?

3
Strong Social Learning
Agents communicate directly about the product,
sharing factual information I didnt buy it
because its not Mac compatible Ive heard Sony
makes the most reliable ones They have a lot of
vegetarian dishes on the menu
4
Strong Social Learning
Weak Social Learning
Agents observe their friends consumption
decisions and enjoyment of products and make
inferences about the products attributes. Greg
got one for Christmas and I know he really liked
it These inferences should be sharper when
friends know their friends preferences well.
5
Social Influence
Strong Social Learning
Weak Social Learning

• Agents observe their friends consumption
  decisions and....
• Their private tastes are altered
• The status value of consuming the product is
  altered

6
Social Influence
Strong Social Learning
Weak Social Learning
Persuasive Advertising
Informative Advertising
Agents observe advertising for the product. They
may learn about objective features of the product
or be persuaded to like it or be persuaded of its
prestige value.
7
Methodology basic paradigm

• Stage 1 Measure the network (Harvard
  Undergraduates)
• Stage 2 Distribute actual products and track
  social learning

8
Methodology

• Measuring the Social Network

9
Measuring the Network

• Rather than surveys, agents play in a trivia game
• Leveraged popularity of www.thefacebook.com
• Membership rate at Harvard College over 90
• 95 weekly return rate

Data provided by the founders of thefacebook.com
10

• Markus
• His Profile
• (Ad Space)
• His Friends

11
Trivia Game Recruitment

• On login, each Harvard undergraduate member of
  thefacebook.com saw an invitation to play in the
  trivia game.
• Subjects agree to an informed consent form now
  we can email them!
• Subjects list 10 friends about whom they want to
  answer trivia questions.
• This list of 10 people is what were interested
  in (not their performance in the trivia game)

12
(No Transcript)
13
Trivia Game Trivia Questions

• Subjects list 10 friends this creates 10N
  possible pairings.
• Every night, new pairs are randomly selected by
  the computer
• Example Suppose Markus listed Tanya as one of
  his 10 friends, and that this pairing gets
  picked.

14
Trivia Game Example

1. Tanya (subject) gets an email asking her to log
   in and answer a question about herself
2. Tanya logs in and answers, which of the
   following kinds of music do you prefer?

15
(No Transcript)
16
Trivia Game Example (cont.)

1. Once Tanya has answered, Markus gets an email
   inviting him to log in and answer a question
   about one of his friends.
2. After logging in, Markus has 20 seconds to answer
   which of the following kinds of music does Tanya
   prefer?

17
(No Transcript)
18
Trivia Game Example (cont.)

1. If Markus answer is correct, he and Tanya are
   entered together into a nightly drawing to win a
   prize.

19
Trivia Game Summary

• Subjects have incentives to list the 10 people
  they are most likely to be able to answer trivia
  questions about
• This is our (implicit) definition of a friend
• This definition is suited for measuring social
  learning about products.
• Answers to trivia questions are unimportant
• ok if people game the answers as long as the
  people its easiest to game with are the same as
  those they know best.
• Roommates were disallowed
• 20 second time limit to answer
• On average subjects got 50 of 4/5 answer
  multiple choice questions right and many were
  easy

20
Recruitment

• In addition to invitations on login,
• Posters in all hallways
• Workers in dining halls with laptops to step
  through signup
• Personalized snail mail to all upper-class
  students
• Article in The Crimson on first grand prize
  winner
• Average acquisition cost per subject 2.50

21
Network Data

• 23,600 links from participants
• 12,782 links between participants
• 6,880 of these symmetric (3,440 coordinated
  friendships)
• Similar to 2003 results
• Construct the network using or link definition
• 5576 out of 6389 undergraduates (87)
  participated or were named
• One giant cluster
• Average path length between participants 4.2
• Cluster coefficient for participants 17
• Lower than 2003 results because many named
  friends are in different houses

22
Number of Roommate links, friend (N1), indirect
friend (N2), and friends of distance 3 (N3) for
an average subject (OR network on all
participants of trivia game)
Type of Link Number of Links Ratio
Roommate .96 1
N1 7.68 8
N2 57.91 60.32
N3 347.14 361.6
23
Methods in Comparison

• 2003 House Experiment in 2 undergraduate houses
• Email-data Sacerdote and Marmaris (QJE 2006)
• Mutual-friend methods with facebook data?
  (Glaeser et al, QJE 2000)

24
Methodology

• Seeding Information

25
Seeding Information

• Elicit subjects initial valuations
• Center empirical estimates
• Decompose valuations (hedonics)
• Randomized treatments
• Distribute product samples
• Information / instructions
• Randomized advertising
• Print (Crimson) and online (thefacebook.com)
• Informative and persuasive
• Elicit subjects final valuations

26
Example

• A hypothetical subject Paul might be exposed to
  the following treatments
• A friend of Pauls of social distance 2 used a
  PDA
• The friend was told about the PDAs instant
  messenger capabilities
• Paul saw an advertisement for the PDA in the
  newspaper that emphasized its hip-ness
• Paul did not see online advertising for the PDA

27
Product Samples

• We want new products to maximize the potential
  for social learning.
• Want to vary products by
• Likely demographic appeal
• Potential for strong learning (need a manual?)
• Potential for weak learning and social influence
  the buzz factor

28
Durables
T-Mobile Sidekick II
Philips Key019 Digital Camcorder
Philips ShoqBox
29
Perishables
Student Advantage Discount Card
Baptiste Studios Yoga Vouchers
Qdoba Meal Vouchers
30
Step I Elicit Valuations

• We want to elicit valuations for a product
  without telling subjects what the product is.
• Our solution We treat a product as a vector of
  attributes which span a space containing the
  specific product.
• We can elicit valuations for each attribute
  without revealing product.

31
Step I Configurators

• Familiar examples with posted menus of prices
• many computer manufacturers (e.g. Dell)
• some car manufacturers
• Here, subjects bid for features
• Baseline bid for featureless product
• Incremental bids for distinct features

32
Constructed Bids

• Subjects told that either this bid or their bid
  in the followup will be entered into a
  uniform-price auction with equal probability
• Construction
• Incentives bid as accurately as possible
• Extension interactions between features

33
Feature descriptions
Feature bids
Baseline bid
34
(20)
(50)
(35)
(150)
(150)
(250)
(Price)
35
Distributions of Imputed Bids

• Results from configurators look sensible
• In each case, market prices lie between median
  bid and upper tail
• T-Mobile and Philips confirmed that demand curves
  for their products are similar to results from
  more traditional analysis

36
Step 2 Randomized Product Trials

• Perishables
• ½ year Student Advantage cards
• 5 yoga vouchers
• 5 meal vouchers
• Durables
• Try out for approximately 4 weeks during end of
  term

37
Randomization

• Blocked by year of graduation, gender, and
  residential house
• Email invitations to come pick up samples
• Invitation times varied to vary strength of
  exposure (April 26th May 3rd)

38
(No Transcript)
39
Info Treatments

• Varied information communicated verbally by
  workers doing distribution
• Information treatments correspond to product
  features in our configurators (5 or 6 features
  for each product).
• Reinforced this information treatment with
  reminder emails
• Each treatment given with 50 probability to each
  subject

40
(No Transcript)
41
Buzz Treatments

• Product-specific treatments without information
  content
• Intended to increase subjects enjoyment of the
  product
• Examples
• Subway tokens for yoga, Qdoba
• 5 free MP3s on ShoqBox
• Extra pre-paid balance on Sidekicks
• Special one-store subsidy on Student Advantage
  cards
• Given with 50 probability to each subject

42
Step 2 Advertising
Online Advertising

• Delivered via thefacebook.com
• Mixed in with normal paid advertising
• 65 of subjects saw ads
• 232,736 impressions (approx. 300 per treated
  subject)
• 136 clicks (in line with averages)

43
Advertising Content

• Content from sponsor companies
• Tweaked to vary informational content in line
  with product features
• Also non-informative versions

44
Step 2 Advertising
Print Advertising

• Inlets in The Crimson, Harvards student
  newspaper
• One of nations largest student papers, daily
  readership approx. 14,000
• Delivered to undergrad students rooms
• Inlets allow randomization across residential
  houses

45
(No Transcript)
46
(No Transcript)
47
All ads for a product has the same style and
differed only in the informational content.
48
Print advertising

• 4 inlets with two ads each.
• 3 ads emphasizing a single feature of a product.
• Residents in a house were exposed to either 2 or
  3 impressions of the same print ad.

49
Step 4 Final Valuations

• Subjects receive full product descriptions and
  submit a second round of bids, which go into the
  auctions with 50 probability
• Subjects also
• Predict what the average bid will be
• Predict what a sample of their friends will bid
  in the auction
• Answer factual questions about each product
• Indicate their confidence in these answers

50
(No Transcript)
51
(No Transcript)
52
Eliciting Confidence Levels

• Meet Bob the Robot and his clones Bob 1 Bob
  100
• Subjects are randomly paired with an (unknown)
  Bob
• Subjects indicated a cutoff Bob at which they
  are indifferent about who should answer the
  question
• If assigned Bob is better than the cutoff, Bob
  answers the question otherwise we use subjects
  answer
• Incentive-compatible mechanism to elicit
  subjects belief that he/she will get the
  question right

53
(No Transcript)
54
(No Transcript)
55
Analysis

• Measuring Learning

56
Analysis

• Stage I Check whether info and ad treatments
  affected a subjects knowledge.
• Stage II Use info treatments as instruments to
  measure social learning.

57
Analysis

• Stage I Check whether info and ad treatments
  affected a subjects knowledge.
• Product Group (PG) Likelihood of answering a
  question about a feature correctly if primed
  about that feature at distribution
• Non-Product Group (NPG) Likelihood of answering
  a question about a feature correctly if exposed
  to informative advertising about that feature

58
Stage I Effect of Info Treatments on Knowledge
(PG)
59
Stage I Effect of Info Treatments on Knowledge
(PG)
94.2
85.2
60
Stage I Effect of Info Treatments on Knowledge
(PG)
94.2
85.2
Subjects who received a product and were primed
on a Feature are about 9 more likely to answer
the question about the feature correctly.
61
Stage I Info-Treatments
FCONFIDENCE FCONFIDENCE FCONFIDENCE FCORRECTANSWER FCORRECTANSWER FCORRECTANSWER
(1) (2) (3) (4) (5) (6)
NUMTREATED .748 (.373) .766 (.505) .007 (.007) .007 (.007)
FTREATED 7.057 (.825) 7.087 (.825) 7.080 (.825) .082 (.015) .083 (.014) .085 (.014)
Intercept 85.468 (1.065) 85.361 (1.065) 85.645 (1.065) .838 (.019) .837 (.021) .856 (.010)
Fixed effects None RE FE None RE FE
N 1927 1927 1927 1930 1930 1930
R2 .054 .056 .058 .022 .023 .022
Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1
62
Stage I Info-Treatments
FCONFIDENCE FCONFIDENCE FCONFIDENCE FCORRECTANSWER FCORRECTANSWER FCORRECTANSWER
(1) (2) (3) (4) (5) (6)
NUMTREATED .748 (.373) .766 (.505) .007 (.007) .007 (.007)
FTREATED 7.057 (.825) 7.087 (.825) 7.080 (.825) .082 (.015) .083 (.014) .085 (.014)
Intercept 85.468 (1.065) 85.361 (1.065) 87.645 (1.065) .838 (.019) .837 (.021) .856 (.010)
Fixed effects None RE FE None RE FE
N 1927 1927 1927 1930 1930 1930
R2 .054 .056 .058 .022 .023 .022
Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1
Both confidence and knowledge increases with info
treatments.
63
Stage I Effect of Online Ad on Knowledge (NPG)
Effect of online ads on subjects who did not
receive products or print ads.
64
Stage I Effect of Online Ad on Knowledge (NPG)
73.5
71.0
64.7
Effect of online ads on subjects who did not
receive products or print ads.
65
Stage I Effect of Online Ad on Knowledge (NPG)
73.5
71.0
64.7
Subjects who received online ads are about 5-8
more likely to answer the question about the
feature correctly.
Effect of online ads on subjects who did not
receive products or print ads.
66
Stage I Effect of Print Ad on Knowledge (NPG)
Effect of print ads on subjects who did not
receive products or online ads.
67
Stage I Effect of Print Ad on Knowledge (NPG)
79.8
71.3
64.8
Effect of print ads on subjects who did not
receive products or online ads.
68
Stage I Effect of Print Ad on Knowledge (NPG)
79.8
71.3
64.8
Subjects who received print ads are about 8-15
more likely to answer the question about the
feature correctly. The effect is increasing in
intensity of exposure.
Effect of print ads on subjects who did not
receive products or online ads.
69
Stage I Ad-Treatments
FCONFIDENCE FCONFIDENCE FCONFIDENCE FCORRECTANSWER FCORRECTANSWER FCORRECTANSWER
(1) (2) (3) (4) (5) (6)
PIMPRESSIONS 1.108 (.698) 1.142 (1.133) -.022 (.012) -.022 (.014)
FIMPRESSIONS 2.278 (1.525) 2.198 (1.075) 2.182 (1.075) .121 (.026) .121 (.026) .120 (.025)
PCRIMSONNUMADS -.520 (.146) -.496 (.243) - .008 (.003) - .008 (.003)
FCRIMSONNUMADS 1.883 (.264) 1.659 (.187) 1.614 (.187) .052 (.005) .051 (.004) .048 (.004)
Intercept 63.496 (0.249) 63.509 (0.439) 63.144 (0.138) .650 (.004) .650 (.005) .640 (.003)
Fixed effects None RE FE None RE FE
N 22,959 22,959 22,959 22,995 22,995 22,995
R2 .003 .003 .004 .006 .007 .008
Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
70
Stage I Ad-Treatments
FCONFIDENCE FCONFIDENCE FCONFIDENCE FCORRECTANSWER FCORRECTANSWER FCORRECTANSWER
(1) (2) (3) (4) (5) (6)
PIMPRESSIONS 1.108 (.698) 1.142 (1.133) -.022 (.012) -.022 (.014)
FIMPRESSIONS 2.278 (1.525) 2.198 (1.075) 2.182 (1.075) .121 (.026) .121 (.026) .120 (.025)
PCRIMSONNUMADS -.520 (.146) -.496 (.243) - .008 (.003) - .008 (.003)
FCRIMSONNUMADS 1.883 (.264) 1.659 (.187) 1.614 (.187) .052 (.005) .051 (.004) .048 (.004)
Intercept 63.496 (0.249) 63.509 (0.439) 63.144 (0.138) .650 (.004) .650 (.005) .640 (.003)
Fixed effects None RE FE None RE FE
N 22,959 22,959 22,959 22,995 22,995 22,995
R2 .003 .003 .004 .006 .007 .008
Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
Both confidence and knowledge increases with ad
treatments.
71
Stage I Buzz-Treatments
BID BID BID
All Products Services Gadgets
BUZZ BUZZ 8.504 (4.206) 1.516 (1.561) 23.706 (9.176)
NUMTREATED NUMTREATED 3.780 (1.886) .822 (.669) 5.837 (4.526)

N 373 373 227 146
R2 .019 .019 .01 .048
Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1
72
Stage I Buzz-Treatments
BID BID BID
All Products Services Gadgets
BUZZ BUZZ 8.504 (4.206) 1.516 (1.561) 23.706 (9.176)
NUMTREATED NUMTREATED 3.780 (1.886) .822 (.669) 5.837 (4.526)

N 373 373 227 146
R2 .019 .019 .01 .048
Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1 Significance Levels 5 1
Buzz treatments raise valuations for gadgets.
73
Analysis stage II

• Use successful first stage as instruments for
  measuring the effects of social learning.
• Regress confidence or correct answers of every
  NPG member on sum friends knowledge (PG) at
  various social distance using sum of info
  treatments as instruments.

74
Confidence
FCONFIDENCE FCONFIDENCE
(1) (2) (1) (2)
PGFCONFIDENCE_R .064 (.029) .057 (.031)
PGFCONFIDENCE_NW1 .040 (.013) .034 (.014)
PGFCONFIDENCE_NW2 .005 (.005) .008 (.005)
PGFCONFIDENCE_NW3 .003 (.001) .009 (.001)
Control for of Eligible NO YES
Intercept 59.628 (.826) 67.870 (1.197)
N 8,982 8,982
R2 0.018 0.045
Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
75
FCONFIDENCE
FCONFIDENCE FCONFIDENCE
(1) (2) (1) (2)
PGFCONFIDENCE_R .064 (.029) .057 (.031)
PGFCONFIDENCE_NW1 .040 (.013) .034 (.014)
PGFCONFIDENCE_NW2 .005 (.005) .008 (.005)
PGFCONFIDENCE_NW3 .003 (.001) .009 (.001)
Control for of Eligible NO YES
Intercept 59.628 (.826) 67.870 (1.197)
N 8,982 8,982
R2 0.018 0.045
Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
76
FCONFIDENCE
FCONFIDENCE FCONFIDENCE
(1) (2) (1) (2)
PGFCONFIDENCE_R .064 (.029) .057 (.031)
PGFCONFIDENCE_NW1 .040 (.013) .034 (.014)
PGFCONFIDENCE_NW2 .005 (.005) .008 (.005)
PGFCONFIDENCE_NW3 .003 (.001) .009 (.001)
Control for of Eligible NO YES
Intercept 59.628 (.826) 67.870 (1.197)
N 8,982 8,982
R2 0.018 0.045
Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
Control for of subjects who were eligible to
receive products at distance R, NW1, NW2 and NW3.
77
FCORRECTANSWER
FCORRECTANSWER FCORRECTANSWER
(1) (2) (1) (2)
PGFCORRECTANSWER_R .108 (.026) .070 (.030)
PGFCORRECTANSWER_NW1 .041 (.013) .018 (.014)
PGFCORRECTANSWER_NW2 .019 (.005) .020 (.005)
PGFCORRECTANSWER_NW3 .007 (.001) .018 (.002)
Control for of Eligible NO YES
Intercept .567 (.010) 0.696 (0.014)
N 9,006 9,006
R2 0.033 0.064
Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
78
FCORRECTANSWER
FCORRECTANSWER FCORRECTANSWER
(1) (2) (1) (2)
PGFCORRECTANSWER_R .108 (.026) .070 (.030)
PGFCORRECTANSWER_NW1 .041 (.013) .018 (.014)
PGFCORRECTANSWER_NW2 .019 (.005) .020 (.005)
PGFCORRECTANSWER_NW3 .007 (.001) .018 (.002)
Control for of Eligible NO YES
Intercept .567 (.010) 0.696 (0.014)
N 9,006 9,006
R2 0.033 0.064
Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
79
FCORRECTANSWER
FCORRECTANSWER FCORRECTANSWER
(1) (2) (1) (2)
PGFCORRECTANSWER_R .108 (.026) .070 (.030)
PGFCORRECTANSWER_NW1 .041 (.013) .018 (.014)
PGFCORRECTANSWER_NW2 .019 (.005) .020 (.005)
PGFCORRECTANSWER_NW3 .007 (.001) .018 (.002)
Control for of Eligible NO YES
Intercept .567 (.010) 0.696 (0.014)
N 9,006 9,006
R2 0.033 0.064
Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
One standard deviation increase in each friends
knowledge (about 30) raises my knowledge by 1
to 2. The total effect is about 9 because
subjects are influenced by several treated
subjects on average.
80
Alternative approach

• Regressing knowledge on friends knowledge only
  measures average amount of social learning.
• We can instead measure social learning
  conditional on two subjects having reported to
  have talked to each other (collected during
  follow-up 350 NPG subjects listed specific PG
  subjects whom they had talked to).
• We exploit the fact that we both randomly
  distributed products and randomized information
  for each subject who received a product.
• We assume that a NPG-subjects pre-information is
  uncorrelated with the info treatment received by
  the PG-subject whom he or she talks to about the
  product.
• This excludes the following situation If I know
  that a Sidekick has AOL messenger I will
  specifically seek out subjects who received a
  product and whom we told about the AOL messenger
  capability of the Sidekick.

81
Effect of Info-Treated Friends on Knowledge (NPG)
Effect of PG-subjects info-treatment on
NPG-subjects knowledge (only for subjects
who Reported to have talked to specific PG
subject)
82
Effect of Info-Treated Friends on Knowledge (NPG)
74.3
68.4
Effect of PG-subjects info-treatment on
NPG-subjects knowledge (only for subjects
who Reported to have talked to specific PG
subject and seen PG subject with product)
83
Effect of Info-Treated Friends on Knowledge (NPG)
74.3
68.4
Subjects who reported to have talked to a friend
who had the product and whom they have seen use
the product are 6 more likely to correctly
answer a question about the feature if their
friend had received an info treatment.
Effect of PG-subjects info-treatment on
NPG-subjects knowledge (only for subjects
who Reported to have talked to specific PG
subject and seen PG subject with product)
84
IV-Regression confidence in answer
FCONFIDENCE FCONFIDENCE FCONFIDENCE FCONFIDENCE
Talked about OR seen (all) Talked about OR seen (services) Talked about OR seen (gadget) Talked about AND seen
FR_FCONFIDENCE .142 (.054) .124 (.100) .151 (.064) .184 (.074)
Intercept 61.617 (5.626) 67.697 (10.124) 59.495 (6.795) 57.503 (7.790)
N 1,912 400 1,511 1,207

Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
85
IV-Regression confidence in answer
FCONFIDENCE FCONFIDENCE FCONFIDENCE FCONFIDENCE
Talked about OR seen (all) Talked about OR seen (services) Talked about OR seen (gadget) Talked about AND seen
FR_FCONFIDENCE .142 (.054) .124 (.100) .151 (.064) .184 (.074)
Intercept 61.617 (5.626) 67.697 (10.124) 59.495 (6.795) 57.503 (7.790)
N 1,912 400 1,511 1,207

Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
86
IV-Regression - knowledge
FCORRECTANSWER FCORRECTANSWER FCORRECTANSWER FCORRECTANSWER
Talked about OR seen (all) Talked about OR seen (services) Talked about OR seen (gadget) Talked about AND seen
FR_FCORRECTANSWER .180 (.067) .011 (.106) .246 (.077) .325 (.112)
Intercept .567 (.068) .890 (.107) .461 (.077) .400 (.109)
N 1,919 400 1,519 1,209

Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
87
IV-Regression - knowledge
FCORRECTANSWER FCORRECTANSWER FCORRECTANSWER FCORRECTANSWER
Talked about OR seen (all) Talked about OR seen (services) Talked about OR seen (gadget) Talked about AND seen
FR_FCORRECTANSWER .180 (.067) .011 (.106) .246 (.077) .325 (.112)
Intercept .567 (.068) .890 (.107) .461 (.077) .400 (.109)
N 1,919 400 1,519 1,209

Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1 Significance Levels 10 5 1
Info-treatment of friend is used as instrument.
Estimated social-learning effects are about 3-15
times greater than the average effects estimated
across all subjects.
88
Observations

• Conditional on having communicated about the
  product social learning seems strongest for
  gadgets rather than services.
• This might indicate that visual observation is
  important for social learning.
• It is also possible that our feature set for
  gadgets provides a more natural decomposition of
  real-world communication than our feature set for
  services.

89
Analysis

• Alternative Model

90
Model

• An untreated (uninformed) subject has a
  probability p of interacting with some treated
  (informed) subject.
• The interaction probability p depends on the
  social distance between uninformed and informed
  subject.
• We distinguish three types of social distances
  room mates (M), direct friends (NW1) and indirect
  friends (NW2).

91
Model

• We define knowledge as the subjective or
  objective probability of answering a question
  about the product correctly.
• If an informed and uninformed subject interact
  the knowledge of the informed subject is
  transferred to the uninformed subject (informed
  treated with a product).

92
Model

• We define knowledge as the subjective or
  objective probability of answering a question
  about the product correctly.
• If an informed and uninformed subject interact
  the knowledge of the informed subject is
  transferred to the uninformed subject (informed
  treated with a product).

After interacting the uninformed subject has the
same probability of answering a question
correctly as the informed subject.
93
Model

• Assume that the knowledge of an informed subject
  is and the knowledge of an uninformed
  subject is .
• Assume that the uninformeds probability of
  interacting with some informed subject is X.
  Then we can express the final expected knowledge
  of the uninformed agent as

94
What is X?

• Assume that the uninformed agent has
  room mates who were offered a product,
  direct friends and indirect friends.
  Then we can express X as

95
What is X?

• Assume that the uninformed agent has
  room mates who were offered a product,
  direct friends and indirect friends.
  Then we can express X as

The probability of interacting with some informed
subject is 1 minus the probability of interacting
with none of them.
96
Model

• We obtain

• We observe and in the
  followup survey.

97
Model

• We obtain

• We observe and in the
  followup survey.
• We do not observe because we
  cannot do a baseline quiz without revealing the
  product.

98
Model

• We obtain expression ()

• We observe and in the
  followup survey.
• However, we do not observe
  because we cannot do a baseline quiz without
  revealing the product.
• Moreover, we expect the information of
  uninformed agents to vary with the number of
  eligible neighbors (and hence the number of
  neighbors who were offered a treatment) due to
  selection.

99
We instead compare agents in similar cells
100
We instead compare untreated agents in similar
cells
We say the green subject lives in a (1,4,4) cell
to indicate that she has one treated room-mate,
and four treated NW1 and NW2 friends AND she has
at least one more eligible (but non-treated) NW1
friend (indicated by plus sign).
101
For example, compare a (1,4,4) cell with a
(1,5,4) cell



102
For example, compare a (1,4,4) cell with a
(1,5,4) cell



The green agent on the right faces the same
neighborhood as the agent on the left but the
randomization turned one eligible, untreated
agent into a treated agent.
103
Model

• By dividing expression () for all agents in cell
  (1,5,4) by expression () for all agents in cell
  (1,4,4) we obtain the marginal impact of
  treating one more NW1 neighbor

104
Model

• By dividing expression () for all agents in cell
  (1,5,4) by expression () for all agents in cell
  (1,4,4) we obtain the marginal impact of
  treating one more NW1 neighbor

Since we only have finitely many observations per
cell we get an estimate for p. For each marginal
comparison between two neighboring cells we get a
new estimate. From this we can construct an
estimate for p and a confidence interval.
105
Model

• By dividing expression () for all agents in cell
  (1,5,4) by expression () for all agents in cell
  (1,4,4) we obtain the marginal impact of
  treating one more NW1 neighbor

By comparing neighboring cells we are essentially
differing out the unobserved knowledge of the
uninformed agent.
106
Analysis

• Results

107
Results

• We are estimating the interaction probabilities
  separately for each product.
• We use both subjective knowledge (What is the
  probability that you can answer a Yes/No question
  correctly?) and objective knowledge (Actual
  share of correctly answered questions in the
  quiz).

108
Results - Card
109
Results - Card
SE (0.16) (0.21)
(0.02) (0.04)
(0.09) (0.03)
110
Results - Yoga
SE (0.19) (0.23)
(0.04) (0.03)
(0.03) (0.05)
111
Results Restaurant
SE (0.03) (0.08)
(0.03) (0.04)
(0.02) (0.01)
112
Results Camcorder
SE (0.02) (0.02)
(0.02) (0.03)
(0.02) (0.02)
113
Results MP3
SE (0.06) (0.07)
(0.03) (0.04)
(0.02) (0.01)
114
Results PDA
SE (0.04) (0.07)
(0.03) (0.04)
(0.02) (0.02)
115
Results

• For private products the interaction
  probability for NW2 neighbors is usually
  insignificant.
• For public products the NW2 effect is small but
  significant.
• NW2 neighborhoods are also 7-times as large as
  NW1 neighborhoods! Therefore, the expected number
  of influenced NW2 agents can be large.

116
Who is influenced the most by social learning
(close or distant neighbors)?(expected number of
interactions taking Nhood size into account
subjective knowledge and significant
probabilities only)
M NW1 NW2 TOTAL
CARD 0.50 1.12 1.62
YOGA 0.60 1.60 1.20
FOOD 0.24 0.80 1.04
CAM. 0.65 1.12 2.85 4.62
SOUND 0.50 0.64 2.28 3.42
PDA 0.45 1.44 2.85 4.74
117
Who is influenced the most by social learning
(close or distant neighbors)?(expected number of
interactions taking Nhood size into account
subjective knowledge and significant
probabilities only)
M NW1 NW2 TOTAL
CARD 0.50 1.12 1.62
YOGA 0.60 1.60 1.20
FOOD 0.24 0.80 1.04
CAM. 0.65 1.12 2.85 4.62
SOUND 0.50 0.64 2.28 3.42
PDA 0.45 1.44 2.85 4.74
Although there is a greater probability to
interact with close agents the expected number of
interactions increases with distance.
118
Summary

• Three methodological contributions
• Applicationspecific measure of social
  connectedness
• Hedonic analysis using configurators
• Measure of confidence using the Bobs
• Advertising increases information.
• Social learning is as important as effects of
  advertising.
• Future work
• Disentangle weak and strong social learning
  channels
• Measure social influence.