You are on page 1of 25

Dating 101

your guide to happiness in life


lesson 1. what are dating markets?
• men+women, women+women, men+men
(and other permutations, but we will only study
heterosexual couples today.)
• not much research done on dating so not
too many models…
(more marriage models)
• complex preferences
– as with all human-psych research:
do people know what they want?
lesson 2: girls are confusing
(guys are too)

identifying preferences is very important.

eliciting preference information from both


yourself and others is very difficult.

thus, we look at preferences from controlled


dating studies.
lesson 3: is there someone for me?

Choose a side

Looking for “the one” Looking for


someone nice

(super selective) (less selective)


How It Works
 Users provide information about
demographics, socioeconomic situation,
physical characteristics, personality,
lifestyle, views, and a photo
 Search for matches by age range and
geographic location, view profiles, and
send messages (if a paid member)
Motivations For Joining
40%
35%
30%
25%
20%
Men
15% Women
10%
5%
0%
Long-Term Just Looking, Casual Sex
Relationship Curious
Who Sends E-Mails?
60%

50%

40%

30%
Men
20% Women

10%

0%
Long Term Just Looking, Casual Sex
Relationship Curious
Physical Characteristics
 Self-Reported Attractiveness Ratings:
Very Above Average Below
Good Average Average
Men 19% 49% 29% 3%

Women 24% 48% 26% 2%

 27.5% of users posted photos


Modeling Framework
 Adachi Search Model:
 Threshold Rule: U(m,w) > v(m)
 A(m, w) = Indicator function of crossing the 
threshold
 Expected Utility of Man Meeting Woman = 
U(mutual match) + U(mismatch)
 EUM(m, w) = UM(m,w)AM(m, w)AW(m, w) + v(m)(1 − 
AW(m,w))AM(m,w) + v(m)(1 − AM(m,w)) 
 Solve for equilibrium reservation utilities
 More than one set of equilibrium results
Modeling Framework
 Gale-Shapley Model:
 A matching μ is stable if there is no man m 
and woman w such that UM(m,w) > UM(m, 
μ(m)) and UW(w,m) > UW(w, μ(w))
 2 extremes using the deferred­acceptance 
algorithm: men­optimal and women­optimal
Modeling Framework
 As search costs become negligible,
Adachi and Gale-Shapley Models provide
identical results
 Are there search costs and costs to
sending e-mails in online dating?
Partner Choice
 Regardless of
the browser’s
attractiveness,
the probability of
sending an e-
mail increases
with the
attractiveness of
the photo
Reply Rates
 71% of men’s e-mails
are rejected, 56% of
women’s
 Responses increase
with attractiveness of
sender
 More attractive men
and women are
pickier, but a man in
the lowest quintile can
get a response from a
woman in the highest
with 20% p
Outcome Regression Approach
 Assumptions:
 Homogeneous preferences: all men (women)
agree on women’s (men’s) rankings
 All profiles will be equally sampled during
searches
 Can segment the single index into groups, regress
number of e-mails a user receives from group g on all
user attributes
 Incidence rate ratio: premium or penalty for a specific
attribute
 exp(θj)= E(Y|xA) / E(Y|xB)
Outcome Regression: Results
 R2m=28%, R2w=44%
 Stated dating goals
affect men’s outcomes
but not women’s
 Looks have the strongest explanatory power
 Photos improve outcome
 Positive correlation between self-description & outcome
 Ideal height & BMI ranges, as well as hair color
 Income increases outcome for men (>$50,000) but
not for women
 Women especially prefer equivalent education levels
 Men in legal, military, fire fighting, or medical careers
receive more e-mails
Outcome Regression: Results
Discrete Choice Estimation
 Assume heterogeneous preferences
 UM(m,w) = UM(Xm,Xw;θM) + errormw 
= xβM + |xw−xm||+αγ+M +  |xw−xm||−αγ−M + ∑ ∏{dmk=1 and 
dwl=1}øklM + errormw
 Fixed­effects Logit Model:

 Random­effects Probit Model

 Logit assumes U is non­linear (α>1) while Probit 
assumes reservation values are independent of 
covariates
 Logit puts more weight on attribute levels, Probit on
heterogeneity
Discrete Choice: Results
 Results confirm Outcome Regression results

Regressor Men - Men - Women - Women -


Logit Probit Logit Probit
Looks Rating .5630 .2774 .5432 .2647
“Very Good” Looks Rating .5995 .3299 .5276 .2790
“Above Average” Looks Rating .3769 .1937 .1871 .0968

“Other Looks” Looks Rating .3718 .0312 .0903 .0941


Height Difference (+) .0013 -.1120 -.0100 .0309
Height Difference (-) -.0108 -.0154 -.0420 -.2155
BMI -.3826 -.2423 .1285 .1182
Income ($1000) .0068 .0033 .0171 .0071
Discrete Choice: Tradeoffs
Predicting Match Structures
 Is the structure of matches different in online
versus traditional dating?
– Dating vs Marriage
– Sample size not representative of population
 Gale-Shapley Predictions
– Men optimal and women optimal matchings are
equivalent
– Keyword matchings also provide similar results
– Thus, online dating is a close approximation for the
online dating website
“Unobservables”
 Observable attributes only account for
part of the preferences
 The unobservable could
 Represent noise in users’ behavior
(mistakes in sending e-mails)
 Represent shared interests
Ethnic Preferences

 All ethnicities discriminate against other


ethnicities, despite stated preferences
 When simulating matching without
ethnicity as a factor, there is still some
endogamy but not as high as initially
Online Dating
 Is it more efficient to appeal to a broad
membership, or to niche groups (ethnic,
religious, age, etc.)?
 Should matches be provided based on
interests and characteristics, or should
users search based on preferences?
Online Dating Sites
Membership Goals Communication
Size
eHarmony > 2 million Serious Guided or Open
Relationships Communication

Match.com > 15 million Dating & E-Mails, Instant


Relationships Messages, “Wink”

American > 3 million Dating, E-Mails, E-Cards, or


Singles Romance, “Flirt”
Friendship
LavaLife > 2 million Dating/Relation Messages, Instant
ships, Intimate Messages, “Smile”
Encounters
Online Dating Sites
Matching Specifications

eHarmony Matches based on Age, Ethnicity, Religion,


personality test & Education, Income,
preferences Height, Smoking, Drinking
Match.com Search (preferences & Age, Ethnicity, Education,
keywords), Mutual & Relationship Status, Body
Reverse Matches Type, Smoking Drinking
American Search (preferences) Age, Relationship Status,
Singles Ethnicity, Height,
Smoking
LavaLife Search (preferences) Age, Ethnicity, Religion,
Income, Children, Body
Type, Height, Smoking,
Drinking, Interests

You might also like