Assessing Attractiveness in Online Dating Profiles

1
Assessing Attractiveness in Online Dating
Profiles
Andrew T. Fiore Lindsay Shaw Taylor G.A.
Mendelsohn Marti Hearst
School of Information Department of
Psychology University of California, Berkeley
2
In the U.S. 63m know someone who has used a
dating site 16m have used a dating site
themselves 53m know someone who has gone on a
date 7m have gone on a date themselves 64
of online dating users think the large pool
helps people find a better date 47 of all
online adults concur Source Pew Internet and
American Life Project
3
Perception and attraction,offline and online
4
(No Transcript)
5
(No Transcript)
6
Performing perceiving self

• Performance of identity
• giving vs. giving off (Goffman 1959)
• Great capacity for control in online performance
• Signals convey information with varying degrees
  of certainty (Donath 1999)
• Conventional vs. assessment
• Social Information Processing, hyperpersonal
  comm. (Walther 1992, 1996)

7
Whats in a profile?

• Combination of fixed-choice categorical
  descriptors, free-text self-description, and
  photos
• Highly optimized self-presentations
• Carefully selected detail
• Unlimited time to craft
• Exaggerations? Lies?
• A lot of people mislead a little (Hancock et al.
  2007)
• Do they reflect actual self? Ideal self?

8
PHilton81 Age 27 Height 58 Weight 115
lbs Occupation Heiress ABOUT ME People say
they envy my lifestyle, but I'm convinced that
anyone with a little imagination can live The
Life.
Sources Wikipedia, Confessions of an Heiress,
Reuters
9
Perceptions of profiles

• Substantial inferences from small cues
   Walthers SIP (Ellison et al. 2006)
• Thin slices, big inferences from bits of
  Facebook profiles (Stecher Counts 2008)
• Fiore Donath (2005)
• Messages received as proxy for attractiveness
• Different predictors for men and women
• Norton, Frost, Ariely (2007)
• More information, less liking (better
  discrimination)

10
Norton, Frost, Ariely (2007)
11
Methodology
12
Profiles (rating targets)

• 50 Yahoo! Personals profiles with photos
• 25 men, 25 women, 20 to 30 years old
• 5 of each from Atlanta, Boston, San Diego,
  Seattle, and St. Louis (geographic diversity)
• One profile randomly chosen from each of the
  first five pages of search results
• Random sample of recently active users

13
(No Transcript)
14
Rating dimensions for profiles

• Attractive
• Genuine, trustworthy
• Masculine
• Feminine
• Warm, kind
• Self-esteem
• Extraverted
• Self-centered

15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
Procedure

• Participants provide information about age,
  gender, sexual preference.
• We provided only profiles and pieces of the
  appropriate target gender.
• Rate randomly ordered profiles and pieces through
  the Web application for 50 min.
• Indicate own self-esteem and attractiveness on
  Likert-type scale.
• Debriefing, payment.

20
Participants (raters)

• Recruited through UC-Berkeley Xlab
• 41 women, 23 men, heterosexual
• 66 Asian
• Between 19 and 25 years old (median 21)
• Self-reported attractiveness mean 2.8 on 0 to
  4 scale
• Self-reported self-esteem mean 2.7 on 0 to 4
  scale

21
Results
22
Raw data and standardization

• Each profile and profile component rated by
  multiple participants 29,120 total ratings
• Ipsatization standardize by each participant,
  for each dimension
• Scales are arbitrary  what is high or low
  for a given participant, for a given dimension?
• Averaged ratings of each profile and profile
  piece on each dimension
• Necessary because data are sparse few
  participants rated every piece of every profile

23
Checking for repetition effects

• Participants rated more than one piece from each
  profile  is this a problem?
• They never rated the exact same piece twice.
• Whole profiles generally presented after pieces.
• No systematic differences in ratings upon first
  exposure to a piece of a given profile and
  subsequent ratings of other pieces.
• Bottom line We can safely use all the ratings
  for our analysis.

24
Attractivenessof whole profiles
25
Dimensions of whole profiles
26
Whole profiles and pieces




27
Whole profiles and pieces
28
Attractivenessof profile pieces
29
Attractiveness of photos
30
Attractiveness of free text
31
Putting it together
32
The big picture Modeling whole-profile
attractiveness

• Mens profiles
• Photo attractive
• Free-text attractive
• Masculine
• Warm and kind in photo
• Genuine/trustworthy in photo
• Photo attractive x fixed-choice attractive
  x free-text attractive

• Womens profiles
• Photo attractive
• Free-text attractive
• Masculine
• Extraverted
• Self-esteem in photo
• Feminine in photo

33
Mens whole profile attractiveness
34
What wasnt associatedwith attractiveness

• Attractiveness of fixed-choice components(after
  adjusting for other component effects)
• Self-rated self-esteem or attractiveness of
  participants
• Length of text in free-text piece
• Use of positive or negative emotion words or
  self-references in profile text (measured with
  LIWCS)

35
Limitations

• Purely associational data, not causal
• Representativeness of participant sample
• Asians overrepresented among raters
   problematic for studying attractiveness
• What is good is beautiful what is beautiful is
  good (Dion et al. 1972)  a halo effect?
• But not all desirable dimensions were associated
  with attractiveness
• What do averages mean for dyadic phenomena?

36
Whats next?

• Systematically combine attractive and
  unattractive components what dominates?
• Examine deal-makers and deal-breakers
• What role do the categorical pieces play in the
  process of identifying potential dates?
• Identify pairs of users about to meet how do
  their perceptions based on profiles change when
  the meet face to face?

37
Thank you! Any questions?
Andrew T. Fiore Lindsay Shaw Taylor G.A.
Mendelsohn Marti Hearst For more
information http//www.ischool.berkeley.edu/atf/
atf_at_ischool.berkeley.edu Thanks to the National
Science Foundation and Microsoft Research for
sponsoring this work.
38
(No Transcript)