Transcript Computer-Mediated Communication

Computer-Mediated
Communication
Reputation (Part II)
Coye Cheshire
// April 8, 2016
How might different types of
reputation systems solve the
lemons problem?
4/8/2016
Computer-Mediated Communication
1
“Reputation” systems in further detail
Explicit
Ratings by others
Implicit
Derived from behavior
Join Date: Mar 2004
Posts: 22
Behavior
4/8/2016
Direct experience
Computer-Mediated Communication
2
Yamagishi’s Experiment
 All participants play role of buyer and
seller in an open market
 Participants create ‘goods’ of some
value (x), and sell at price (y).
 Production points and payouts come
from the experimenter
 Buyers receive 1.5 times the true Buyer gets
value of the good.
60 x 1.5 = 90 points
4/8/2016
Computer-Mediated Communication
3
4/8/2016
Computer-Mediated Communication
4
Yamagishi’s Experimental Conditions
 Anonymity
 Identity
 Reputation
4/8/2016
Computer-Mediated Communication
5
Yamagishi et al. (First Experiments)
Conditions of
Environment
Outcome
(Independent Variables)
(Dependent
Variable)
-Anonymity
-Price?
-Identity
-Quality level?
-Reputation (3rd party only)
-Probability of item
being sold?
-Time to sale?
4/8/2016
Computer-Mediated Communication
6
Quality Levels
4/8/2016
Computer-Mediated Communication
7
Levels of “dishonesty”
4/8/2016
Computer-Mediated Communication
8
Yamagishi et al (Second Set of
Experiments)
Reputation System
Outcome
(Independent
Variable)
(Dependent
Variable)
-Positive
-Price?
-Negative
-Quality level?
-Mixed
-Probability of item
being sold?
-Time to sale?
4/8/2016
Computer-Mediated Communication
9
Comparing Quality Levels
4/8/2016
Computer-Mediated Communication
10
Quality levels over time (3rd Expt, Mock Auction)
4/8/2016
Computer-Mediated Communication
11
Identity
I also found the description of negative reviews interesting. In
the experiment, and a basis for the experiment, they assumed
that users had the freedom to change identities; and yet we
just had a discussion about the persistence of information and
how hard it is to "forget" in the digital era. As more mash-ups
of data are created, it may eventually be possible to trace
sellers who "quit" and reemerge as "new" sellers, there are
also sites that restrict creation of separate accounts by IP. I am
not saying that going "incognito" is impossible, it will just be
harder.
-Adam J.
4/8/2016
Computer-Mediated Communication
12
Strategy and Reality
Being someone who steals other people's candy, I can say that
(1) I love candy, so watch out and (2) accountability is key. If
you aren't being held accountable for your actions, exploiting
others is a profitable strategy. I mean, look around you.
-Galen “you better hide your candy” P.
I wonder if that conclusion would continue to hold if the game continued
over a longer time frame. For instance, over the long term, do the
sellers with persistent identities and/or better reputations and greater
honesty ultimately profit more because trust in them only increases over
time and thus makes lower-priced but less reputable alternatives look
less appealing?
-Kimra M.
About whether honesty pays, I had the same
question as Kimra about whether the results
would change in the longer term.
-Tanushree J.
4/8/2016
Computer-Mediated Communication
13
Resnick et al. 2006
Effect of strong reputations
on revenues compared to
those without reputation
Effect of “negatives” in a
brief reputation on
outcome of revenue
4/8/2016
Computer-Mediated Communication
14
Matched Pairs by Different Sellers
4/8/2016
Computer-Mediated Communication
15
Reputation Effects
 Do strong reputations matter?
 What is the impact of negative reputation marks
in a mixed reputation system such as eBay?
4/8/2016
Computer-Mediated Communication
16
Complexities and Limits of Field
Experiments
First of all, they didn't believe in using the number of bids as a
control variable. They claimed that it is "an endogenous indicator of
the impact of reputation on price and should not be an independent
variable in a simple regression model." I completely disagree with
this idea. Personally (coming from a background in social
psychology), I would argue that the number of bids serves as a form
of social proof, which may very well be a moderating factor, thus
muddying the effects they found.
They also claim to have made "roughly similar web sites" for all
sellers. They address this as one of the threats to validity, but they
seem to discount the importance of the visual aesthetic for each
seller.
-Courtney C.
4/8/2016
Computer-Mediated Communication
17
Critiquing
Methodological
Approaches to Similar
Problems
4/8/2016
Computer-Mediated Communication
18
Two approaches to studying online
reputation system (Ebay)
 Why study it in a lab?
 Why study it in a field
experiment?
4/8/2016
Computer-Mediated Communication
19
Putting Experimental Work in Context
 Selection of subjects
(i.e., what do they value?)
 Task length and learning
 Accounting for time in statistical
analyses
 Do not assume that an
experiment is even trying to
‘recreate’ a specific real-life
situation unless they explicitly
say so– and even then you
should always consider the
scope of the experiment.
4/8/2016
Computer-Mediated Communication
20
The Validity Problem
Internal
Validity
4/8/2016
External
Validity
Computer-Mediated Communication
21
External Validity: Generalizing From
Experiments
?
4/8/2016
Computer-Mediated Communication
22
Ecological Validity: Approximation of
Real-Life Activity
Yamagishi et al.
4/8/2016
Resnick et al.
Computer-Mediated Communication
23
Resnick et al. 2006
4/8/2016
Computer-Mediated Communication
24
Other Implicit ‘Reputation’
Information in CMC?
4/8/2016
Computer-Mediated Communication
25
Another side of reputation…
“Internet markets also have significant
advantages in establishing reputations …
any information that is gleaned can be
near costlessly tallied on a continuing
basis … [and] that information can be
near costlessly transmitted to millions of
potential customers.
”
— Resnick et al. 2006, p. 80
4/8/2016
Computer-Mediated Communication
26
The CMC and ‘Offline’ Reputation Link:
Emergent Reputation Systems and Identity
4/8/2016
Computer-Mediated Communication
27
As designers, what can we learn
from all of this?
 What kind of community do you have (or are you trying to foster)?
 When and Why to use Pos/Neg/Mixed/Hybrid Reputation
Systems?
 What behavior(s) do you want to encourage, reward, punish?
 Consider the “unintended consequences” of implicit information
 Just because you build a system to be interpreted a certain way
doesn’t mean that the user will agree…
4/8/2016
Computer-Mediated Communication
28