Measuring Trust in Social Networks

1
Measuring Trust in Social Networks

• Tanya Rosenblat
• (Wesleyan University, IQSS and IAS)
• March 2, 2006

2
Motivation

• Trust game focuses on trust between strangers.
• We are interested in trust between agents in a
  social network.
• Specifically, we want to know how trust varies
  with social distance.

3
Trust Social Distance Channels

• Preferences We trust friends more because they
  like us more.
• Beliefs We trust friends more because we know
  their type (reliability for example).
• Enforcement We trust friends more because we
  interact more frequently with them and can punish
  them better.

4
Example 1

• Andy consider lending money to Guillaume.
• Preferences Andy thinks Guillaume likes him and
  wont inconvenience him by repaying late.
• Beliefs Andy knows that Guillaume is a reliable
  person he is less sure of the reliability of
  people he knows less well.
• Enforcement Andy sees Guillaume every day and
  will hide Guillaumes cigarettes or commit some
  other cruelty if he doesnt repay in time.

5
Example 2

• Muriel asks Tanya to look after her house and
  take care of financial matters while she travels.
• Preferences Muriel thinks Tanya likes her and
  will exert some effort to avoid penalties (from
  unpaid paying utility penalties etc.).
• Beliefs Muriel thinks Tanya is more reliable
  than Guillaume wholl set the house on fire.
• Enforcement Muriel sees Tanya often and can
  punish her if Tanya doesnt keep her promise to
  look after the house.

6
First Experiment Web-based

• Social networks in two student dorms (N569)
• Preferences use modified dictator games as in
  Andreoni-Miller (2002) to measure how altruistic
  we expect our friends to be and how altruistic
  they actually behave towards us (as compared to
  strangers).
• Enforcment Two within subject treatments to
  check for enforcement channel (T1) recipient
  finds out and (T2) recipient does not find out.

7
Second Experiment Field

• Two shantytowns in Lima, Peru (300 households
  each)
• Use a new microfinance experiment which requires
  clients to find sponsors who cosign their loan.
  Our experiment simulates the situation whom do I
  approach if I need money?
• We randomize interest rates to measure how much
  easier it is to ask a friend for money than a
  socially more distant neighbor.
• Clients choices reveal the sum of
  preferences/belief/enforcement channels.

8
House Experiment

• Methodology

9
House Experiment Methodology

• Stage I Network Elicitation Game
• Choose two student dorms (N802). About 50
  percent of friends inside dorm.
• 569 subjects complete baseline survey.
• Stage II Modified Dictator Games
• Half the subjects are allocators and play
  modified dictator games with 5 recipients of
  various social distance.
• The other half of subjects are recipients and are
  asked about beliefs of how 5 randomly chosen
  allocators at various social distance allocate
  tokens.

10
Stage I Network Elicitation

• Goal high participation rate to get as complete
  network as possible
• Web-based
• Use a novel coordination game with monetary
  payoffs to induce subjects to reveal their social
  network.
• Subjects name up to 10 friends and one attribute
  of their friendship (how much time they spend
  together during the week on average).
• Earnings participation fee plus experimental
  earnings

11
(No Transcript)
12
Network Elicitation Game
Tanya names Alain
Tanya
Alain
13
Network Elicitation Game
Tanya
Alain
Tanya and Alain get both 50 cents with 50
probability if they name each other.
Tanya
Alain
14
Network Elicitation Game
Tanya
Alain
Probability of receiving 50 cents increases to
75 if Tanya and Alain agree on attributes of
friendship as well (time spent together).
Tanya
Alain
15
(No Transcript)
16
(No Transcript)
17
Network Data

• In addition to the network game
• Know who the roommates are
• Geographical network (where rooms are located in
  the house)
• Data from the Registrars office
• Survey on lifestyle (clubs, sports) and
  socio-economic status

18
Network Data Statistics

• House1 - 46 (259) House2 - 54 (310)
• Sophomores - 31(174) Juniors - 30 (168)
  Seniors - 40 (227)
• Female - 51 (290) Male - 49 (279)
• 5690 one-way relationships in the dataset 4042
  excluding people from other houses
• 2086 symmetric relationships (1043 coordinated
  friendships)

19
Symmetric Friendships
20
Symmetric Friendships
The agreement rate on time spent together (/- 1
hour) is 80
21
Network description

• Cluster coefficient (probability that a friend of
  my friend is my friend) is 0.58
• The average path length is 6.57
• 1 giant cluster and 34 singletons
• If we ignore friends with less than 1 hr per
  week, many disjoint clusters (175).

22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
Stage II Game Phase

• Use Andreoni-Miller (Econometrica, 2002) GARP
  framework to measure altruistic types
• Modified dictator game in which the allocator
  divides tokens between herself and the recipient
  tokens can have different values to the allocator
  and the recipient.

Subjects divide 50 tokens which are worth 1
token to the allocator and 3 to the recipient 2
tokens to the allocator and 2 to the recipient 3
tokens to the allocator and 1 to the recipient
26
Stage II Game Phase

• Half the subjects have role of allocator and the
  other half are recipients.

27
Stage II Game Phase

• Half the subjects have role of allocator and the
  other half are recipients.
• Recipients are asked about their beliefs of how 7
  possible allocators split tokens in all three
  dictator game.
• Allocators are asked to allocate tokens between
  themselves and 5 possible recipients PLUS one
  anonymous recipient.

28
Stage II Game Phase

• Half the subjects have role of allocator and the
  other half are recipients.
• Recipients are asked about their beliefs of how 7
  possible allocators split tokens in all three
  dictator game.
• Allocators are asked to allocate tokens between
  themselves and 5 possible recipients PLUS one
  anonymous recipient.
• Two within treatments (all subjects) for each
  pair we ask about beliefs/allocations if the
  recipient (T1) does not find out who made the
  allocation and (T2) does find out.

29
Stage II Game Phase

• Half the subjects have role of allocator and the
  other half are recipients.
• Recipients are asked about their beliefs of how 7
  possible allocators split tokens in all three
  dictator game.
• Allocators are asked to allocate tokens between
  themselves and 5 possible recipients PLUS one
  anonymous recipient.
• Two within treatments (all subjects) for each
  pair we ask about beliefs/allocations if the
  recipient (T1) does not find out who made the
  allocation and (T2) does find out.
• Recipients and allocators are paid for one pair
  and one decision only.

30
Recipients are asked to make predictions in 7
situations (in random order) 1 direct friend 1
indirect friend of social distance 2 1 indirect
friend of social distance 3 1 person from the
same staircase 1 person from the same house 2
pairs chosen among direct and indirect friends
Recipients
Share staircase
Indirect Friend 2 links
Indirect Friend 3 links
Same house
31
Recipients are asked to make predictions in 7
situations (in random order) 1 direct friend 1
indirect friend of social distance 2 1 indirect
friend of social distance 3 1 person from the
same staircase 1 person from the same house 2
pairs chosen among direct and indirect friends
Recipients
A possible pair
Share staircase
Indirect Friend 2 links
Indirect Friend 3 links
Same house
32
Stage II Recipients

• Recipients make predictions about how much they
  will get from an allocator in a given situation
  and how much an allocator will give to another
  recipient that they know in a given situation.
• One decision is payoff-relevant
• gt The closer the estimate is to the actual
  number of tokens passed the higher are the
  earnings.

Incentive Compatible Mechanism to make good
predictions
Get 15 if predict exactly the number of tokens
that player 1 passed to player 2
For each mispredicted token 0.30 subtracted from
15. For example, if predict that player 1 passes
10 tokens and he actually passes 15 tokens then
receive 15-5 x 0.3013.50.
33
Allocators
For Allocator choose 5 Recipients (in random
order) 1 direct friend 1 indirect friend of
social distance 2 1 indirect friend of social
distance 3 1 person from the same staircase 1
person from the same house.
Share staircase
Indirect Friend 2 links
Indirect Friend 3 links
Same house
34
Stage II Allocators

• We also ask allocator to allocate tokens to an
  anonymous recipient.
• All together they make 6 times 3 allocation
  decisions in T1 treatment (recipient does not
  find out) and 6 times 3 allocation decisions in
  T2 treatment (recipient finds out).

35
Stage II Sample Screen Shots

• Allocator Screens

36
(No Transcript)
37
Stage II Sample Screen Shots

• Recipient Screens

38
(No Transcript)
39
(No Transcript)
40
House Experiment Analysis

• Identify Types

41
Analysis (AM)

• Selfish types take all tokens under all payrates.
• Leontieff (fair) types divide the surplus
  equally under all payrates.
• Social Maximizers keep everything if and only if
  a token is worth more to them.

42
Analysis (AM)

• About 50 of agents have pure types, the rest
  have weak types.
• Force weak types into selfish/fair/SM categories
  by looking at minimum Euclidean distance of
  actual decision vector from types decision.

43
(No Transcript)
44
Recipients think that friends are about 20 less
selfish under both treatments.
45
(No Transcript)
46
Allocators are only weakly less selfish towards
friends if the friends do NOT find out.
47
Allocators are 15 less selfish towards friends
if friends can find out.
48
House Experiment Summary

• Preferences some directed altruism but
  altruists tend to be altruistic to everybody and
  not just their friends.
• Enforcement strong evidence that enforcement
  makes people treat their friend a lot better.
• Recipients seem to find it difficult to
  distinguish the preference channel from
  enforcement channel they always expect friends
  to treat them more nicely than everybody else.

49
Field Experiment

• Location Urban shantytowns of Lima, Peru
• Trust Measurement Tool - a new microfinance
  program where borrowers can obtain loans at low
  interest by finding a sponsor from a
  predetermined group of people in the community
  who are willing to cosign the loan.

50
Types of Networks

• Which types of networks matter for trust?
• Survey work to identify
• Social
• Business
• Religious
• Kinship

51
Survey Work in Limas North Cone
52
Who is a sponsor?

• From surveys, select people who either have
  income or assets to serve as guarantors on other
  peoples loans.
• 25-30 for each community
• If join the program, allowed to take out personal
  loans (up to 30 of sponsor capacity).

53
(No Transcript)
54
(No Transcript)
55
Presenting Credit Program to Communities in
Limas North Cone
56
Experimental Design

• Three random variations
• Sponsor-specific interest rate
• Helps identify how trust varies with social
  distance (all channels)
• Sponsors liability for co-signed loan
• Helps separate out enforcement channel.
• Average Interest rate at community level
• Helps identify whether social networks are
  efficient at allocating resources

57
Random Variation 1
Sponsor-specific interest rate is randomized
Indirect Friend 2 links
Indirect Friend 3 links
58
Random Variation 1
Sponsor-specific interest rate is randomized
Sponsor 2 r2 lt r1
Indirect Friend 2 links
Indirect Friend 3 links
59
Random Variation 1
Sponsor-specific interest rate is randomized
Should I try to get sponsored by Sponsor1 or
Sponsor2?
Sponsor 2 r2 lt r1
Indirect Friend 2 links
Indirect Friend 3 links
The easier it is to substitute sponsors, the
higher is trust in the community.
60
Random Variation 1
Sponsor-specific interest rate is randomized
Should I try to get sponsored by Sponsor1 or
Sponsor2?
Sponsor 2 r2 lt r1
Indirect Friend 2 links
Indirect Friend 3 links
Measure the extent to which agents substitute
socially close but expensive sponsors for more
socially distant but cheaper sponsors.
61
Randomization of interest rates Decrease in
interest rate based on slope
Each client is randomly assigned a slope
(1,2,3,4)
SD1 SD2 SD3 SD4
Slope 1 0.125 0.250 0.375 0.500
Slope 2 0.250 0.500 0.750 1.000
Slope 3 0.500 1.000 1.500 2.000
Slope 4 0.750 1.500 2.250 3.000
Close friends generally provide the highest
interest rate and distant acquaintances the
lowest, but the decrease depends on SLOPE
62
Demand Effects

• The interest rate on the previous slide for 75
  of the sample and 0.5 percent higher for 25 of
  the sample to check for demand effects (people
  borrow more and for a different reason when
  interest rates are lower?).

63
Random Variation 2
Sponsors liability for the cosigned loan is
randomized (after borrower-sponsor pair is formed)
Sponsors liability might fall below 100
Indirect Friend 2 links
Indirect Friend 3 links
Measure the extent to which sponsors can control
ex-ante moral hazard. (can separate type trust
from enforcement trust by looking at repayment
rates).
64
Random Variation 3
Average interest rate at community level (to
measure cronyism)
Community 2 High r
Community 1 Low r
Under cronyism, the share of sponsored loans to
direct friends (insiders) increases as interest
rate is reduced.
65
Setting

• Urban Shantytowns in Limas North Cone.
• Some MFIs operate there, offering both individual
  and group lending, with varying levels of
  penetration but never very high.
• Work has been conducted in 2 communities in
  Limas North Cone.

66
Survey Work

• Household census
• Establish basic information on household assets
  and composition.
• Provides us with household roster for Social
  Mapping
• Provides us with starting point to identify
  potential sponsors
• Identify and sign-up sponsors through series of
  community meetings
• Offer lending product to community as a whole

67
Microlending Partner

• Alternativa, a Peruvian NGO
• Lending operation (both group and individual
  lending)
• Also engaged in plethora of community building,
  empowerment, information, education, etc.

68
Lending Product

• Community 300 households
• We identify 25-30 sponsors who have assets
  and/or stable income, sufficient to act as a
  guarantor on other peoples loans.
• A sponsor is given a capacity, the maximum
  amount of credit they can guarantee.
• A sponsor can borrow 30 of their capacity for
  themselves.
• Individuals in the community are each given a
  sponsor card which lists the sponsors in their
  community and their interest rate if they borrow
  from each sponsor.

69
(No Transcript)
70
(No Transcript)
71
Status

• So far work has been conducted in 2 communities
  in Limas North Cone.
• The first community has 240 households and the
  second community has 371 households.

72
Characteristics of Sponsored Loans

• The average size of a sponsored loan is 317
  Dollars or 1,040 soles.
• The average interest rate for sponsored loans is
  4.08

73
Social Distance of Actual Client-Sponsor by Slope
74
Social Distance of Actual Client-Sponsor by Slope
Greater slope makes distant neighbors more
attractive due to lower interest. We see
substitution away from expensive close neighbors.
75
Social Distance of Actual Client-Sponsor by Slope
Effect is mainly driven by clients substituting
SD1 for SD2 sponsors. There is less
substitution of SD2 sponsors for SD3,4
sponsors. Therefore, slope 2,3,4 look different
from slope 1 (where all interest rates are
essentially equal) but not so different from
each other.
76
(No Transcript)
77
(No Transcript)
78
Logistic regressions confirm earlier graphs and
quantify the size of the social distance/interest
rate tradeoff a direct link to a sponsor is
worth about 4 interest rate points. A link to a
neighbor at distance 2 is worth about half that
much.
79
Results using logistic regressions

• Direct social neighbor has the same effect as a
  3-4 percent decrease in interest rate
• Even acquaintance at social distance 3 is worth
  about as much as one percent decrease in interest
  rate
• Independent effect of geographic distance one
  standard deviation decrease in geographic
  distance is worth about as much as a one percent
  drop in interest rate

80
Repayment rates of clients and sponsors
81
Repayment rates of clients and sponsors
Repayment rates after n months (n1,2,..,12) are
similar for sponsors and non-sponsors in both
communities.
82
Effect of Second Randomization
83
Effect of Second Randomization
Higher sponsor responsibility increases
repayments rates of BAD clients (defined as
having paid back less than 50 percent after 6
months). No effect of repayment of high-quality
clients.
84
Effect of Second Randomization
Evidence for enforcement trust!
85
Peru Summary

• We develop a new microfinance program to measure
  trust within a social network.
• Preliminary evidence suggests that social
  networks can greatly reduce borrowing costs
  (measured in terms of interest rate on loan).
• Evidence that sponsors pick clients who are as
  likely to repay as they are (micro-finance
  organization is no better) (belief/type channel)
• Evidence that sponsors can enforce repayment for
  a chosen client (enforcement trust).

86
Trust Measures in Economics

• Surveys (General Social Survey and World Values
  Survey)
• Generally speaking, would you say that most
  people can be trusted or that you cant be too
  careful in dealing with people?
• Highest trust countries are in Scandinavia
  lowest trust in South America
• Some problems with GSS type questions
• What is the reference group?
• What is trust? (not defined in the question)
• Are participants truthful when answering this
  potentially sensitive question?

87
Trust Measures in Economics

• Trust (or Investment) Game

Player 1 Sender
Player 2 Receiver
S sends x to R R receives 3x
S
R
R keeps y and sends 3x y back to S
88
Trust Measures in Economics

• Trust (or Investment) Game

Interpretation S is trusting if he sends x gt0
R is trustworthy if she reciprocates by sending
ygt0
Player 1 Sender
Player 2 Receiver
S sends x to R R receives 3x
S
R
R sends y back to S and keeps 3x y
89
Trust Measures in Economics
Some reasons to be cautious

• Trust game and GSS answers dont coincide
• Trustworthy behavior predicts real life outcomes
  (e.g., repay loans)
• Trusting behavior possibly gambling