Transcript citris-personals

Studying Computer-Mediated
Communication via Online Personals
Andrew Fiore, Marti Hearst, SIMS
Lindsay Shaw, Jerry Mendelsohn, Psychology
Computer-Mediated Communication

People now work and play together
at a distance




Students get degrees via distance courses
International teams write software and design
products together
Groups write position papers and organize
political activities
People provide advice and other services
Computer-Mediated Communication

How does online communication
differ from face-to-face?



People are more likable? Less?
Communication is better? Worse?
Different?
How to design CMC systems to best
promote positive relationships?
How is this studied?


To date, mainly by small controlled
studies
Example: Walther et al. 96, 01



Studied online workgroups
Pairs of students working on class
projects
2x2 design (short- or long-term
interaction, presence/absence of photos)
Walther et al. on CMC and Affinity
Found that users experienced
affection and social attraction:
1. Most of all in long-term online groups Hyperpersonal
without photographs.
2. Less so in long-term online groups
with photographs and short-term online
groups with photographs.
3. Least of all in short-term online
groups without photographs.
Personal
(offline norm)
Impersonal
Hyperpersonal interaction: accelerated affinity via wishful
thinking in the absence of strong social cues.
Scaling Up the Studies


Controlled studies are very useful,
but they are necessarily small
Millions of people are interacting
online, so


how can we leverage this massive-scale
interaction for study?
Idea: Study online personals
Online Personals

A HUGE socio-technical phenomenon



US has ~80 Million single adults
In 2003, ~40 Million UNIQUE visitors to
online personals websites
A virtually untapped data source for
studying technology-mediated interactions

Virtually untapped
Example (Fiore & Donath ’05):
Data from an online personals site

Anonymized eight-month snapshot



June 2002 to February 2003
153,942 completed user profiles
Messaging: who contacted whom, when, how much,
and who replied.

29,687 users sent 236,930 messages


110,722 distinct contacts



51,348 distinct recipients
One or more msgs sent between two users
Only 21.8 percent were reciprocated
Question: Does Homophily hold?

How similar are people to those whom they contact, and
on which features?
Method of analysis
1.
2.
3.
Calculate percent of dyads we would expect
to be the same on a given dimension if they
consisted of randomly selected men and
women.
Calculate actual percent of dyads the same
on that dimension from dating site data.
Compare actual and expected percentages.
Is actual similarity greater than we’d expect
by chance?
Characteristic
Marital status
Wants children
Num. of children
Physical build
Smoking
Phys. appear.
Educational level
Religion
Race
Drinking habits
Pet preferences
Pets owned
Exp. % same Actual % same
31.6
56.0 (1.77x)
25.1
40.5 (1.61x)
27.8
38.6 (1.39x)
19.2
25.6 (1.33x)
40.5
54.0 (1.33x)
37.6
49.2 (1.31x)
23.6
29.3 (1.24x)
42.4
52.6 (1.24x)
71.1
85.9 (1.21x)
61.2
73.4 (1.20x)
34.7
39.9 (1.15x)
21.8
24.0 (1.10x)
t stat.
76.00
48.55
34.35
22.44
41.98
35.89
19.36
31.59
65.81
42.69
16.43
8.04
Women
Widowed
Separated
Divorced
Married
Married
In relationship
In relationship
Never married
(Invalid)
No answer
Men
Studying Hyperpersonal Interaction in
the Online Personals Context

Issue: people disappointed with
face-to-face meetings based on
profiles and earlier interactions


Are people lying on their profiles?
Or … are people experiencing inflated
expectations caused by the CMC?
Hyperpersonal interactions:
 Accelerated affinity via wishful thinking in
the absence of strong social cues.

Near-Term Plans

Test the inflated expectations theory



Conduct a survey to determine
expectations before and after F2F
Analyze results with respect to a wide
range of factors
Use analysis to determine how to
better align expectations.
Longer-Term Plans

Use social psych research to

Understand problems with current CMC
systems
People are poor at self-description
 -> How to improve descriptions?


Understand what makes for good
matches
Complementarity vs. Compatibility
 Translate this into CMC representation


Design better systems