Content uploaded by Michael C. Seto
Author content
All content in this area was uploaded by Michael C. Seto on Mar 23, 2018
Content may be subject to copyright.
arXiv:1607.01952v1 [cs.SI] 7 Jul 2016
1
A First Look at User Activity on Tinder
Gareth Tyson,1Vasile C. Perta,2Hamed Haddadi,1Michael C. Seto,3
1Queen Mary University of London, 2Sapienza University of Rome, 3Royal Ottawa Health Care Group
Abstract—Mobile dating apps have become a popular means
to meet potential partners. Although several exist, one recent
addition stands out amongst all others. Tinder presents its users
with pictures of people geographically nearby, whom they can
either like or dislike based on first impressions. If two users like
each other, they are allowed to initiate a conversation via the
chat feature. In this paper we use a set of curated profiles to
explore the behaviour of men and women in Tinder. We reveal
differences between the way men and women interact with the
app, highlighting the strategies employed. Women attain large
numbers of matches rapidly, whilst men only slowly accumulate
matches. To expand on our findings, we collect survey data to
understand user intentions on Tinder. Most notably, our results
indicate that a little effort in grooming profiles, especially for
male users, goes a long way in attracting attention.
I. INTRODUCTION
Online dating has become extremely popular, with 38%
of American adults who are “single or looking” having
experimented with it [1]. Mobile dating apps have become
particularly prevalent in recent years [2]. The most noticeable
shift these apps enable is the ability to discover and interact
with nearby potential mates. Location-based services have
long been touted as a commercial revolution (e.g., hailing
taxis). However, only recently has the discovery of people
become popular. One app stands out in this regard. Tinder
presents users with pictures of other nearby people; users are
then allowed to either “like” or “dislike” the picture(s). If two
users like each other, they will be given the opportunity to
interact via text messaging: this is called a match. Whereas
previous dating services have aimed to match on interests,
Tinder, instead, matches on locality. Thus, by focusing on first
impressions, Tinder constitutes a cut-down version of online
dating, without any of the features that make it possible to
understand the deeper characteristics of potential mates.
Despite Tinder’s popularity, there is a distinct lack of
research on the app’s usage. As a primarily heterosexual dating
app, we are particularly curious to understand the interactions
between the two genders, as its novel style of matching and
ambiguity in user intent raises many questions that have not
been explored yet. In this paper, we pursue two key avenues
of study. First, we ask how gender impacts matching and
messaging rates for Tinder profiles. Past work has established
clear differences between how genders pursue romantic en-
gagement; as such, we are curious to understand how this
translates to the novel matching and usage style of Tinder.
Second, we ask, what profile characteristics are common in
Tinder, as well as which characteristics can impact matching
and messaging rates. Although past online dating research has
shed insight on user behaviour [3], these traditional services
differ greatly from how Tinder matches users.
To explore these questions, we have performed a measure-
ment campaign of Tinder. We have created a number of curated
profiles [4], which we have injected into London and New
York (§III). We have used these profiles to monitor the way
others react to them, specifically in terms of matches and
subsequent messaging. Through data on almost half a million
users, we show that the two genders exhibit quite different
matching and messaging trends (§IV). Women tend to be
highly selective in whom they like, leading to a starvation
of matches for men. Men, on the other hand, are more
accommodating in their practices, hitting like for a far larger
proportion of women. This mirrors many sociological obser-
vations about mating, although Tinder seems to enact quite
extreme examples of this. Our findings suggest a “feedback
loop”, whereby men are driven to be less selective in the hope
of attaining a match, whilst women are increasingly driven to
be more selective, safe in the knowledge that any profiles they
like will probably result in a match. This leads us to explore
key characteristics that impact user matching rates (§V). We
show that simple improvements (e.g., including more pictures)
can substantially increase popularity, particularly for men.
Interestingly, these increases come primarily from women who
react more strongly to these profile improvements. In addition,
we also perform a user questionnaire to validate and expand
our interpretations of the data (§VI).
II. BACKGROUND
A. A brief tour of Tinder
Tinder is a mobile dating app launched in 2012. Tinder
profiles are very limited, containing just a name, age, interests
and a short bio. Users stipulate their desired match by selecting
the age and gender range, as well as writing a short description
of themselves. When a user turns on the app, their location
is reported to Tinder’s server, which then returns a set of
profiles matching the user’s stipulated criteria within a given
range (maximum is 100 miles). The user is then presented
with a picture of a nearby user. This screen contains two large
buttons, labelled with a cross and a heart. These allow the user
to stipulate if they like (termed “swiping right”) or dislike
the profile (“swiping left”). If two users say that they like
each other, they each are notified (otherwise the two users are
not notified of anything). From this point forward, the two
users can interact via text message within the app. This is the
limit of the app’s functionality and, as such, it constitutes an
extremely cut-down version of an online dating experience.
In fact, there is no formal means of reporting what a user
desires from a match and, therefore, Tinder can even be used
for simply meeting new friends.
2
B. Related work
Despite Tinder’s growing popularity and unconventional
matching style, it has received limited attention from the
research community. Most related is a recent study into the
privacy of mobile dating apps like Tinder [5]. They observed
a range of potential privacy concerns, primarily relating to
the ability of attackers to track user location. This is perhaps
exacerbated by the frequent use of location-based dating apps
for immediate sexual encounters [6]. Beyond this, little is
known about the nature and usage of Tinder. That said, there is
a significant body of research looking more broadly at online
dating services that match people based on interests.
One of the largest studies was performed by Rudder [7].
A number of interesting insights were gained, building a
model of pictures that others are attracted to. These findings
are backed by decades of psychology research, showing that
initial physical attractiveness is often associated with other
positive attributes, e.g., intelligence [8]. Although laboratory
research provides greater control over study conditions, Tinder
provides interesting potential, as it can provide data on in-the-
wild first impressions on physical attractiveness. Fiore et al.
also found that free-text components can play an important
role in predicting attractiveness [9], [10]. In Tinder, this latter
component is heavily suppressed, favouring first impressions
based on pictures. There have also been a number of studies
into user interaction on dating websites. For instance, it has
been found that people use a variety of strategies to reduce un-
certainty when interacting with new people [11]. Disclosure of
information is an important part of this [12], as deception has
been found to be commonplace [13]. This issue is particularly
important in Tinder, as initial disclosure of information is very
limited (due to the simplicity of profiles), although past work
has found that direct conversation is much more important
than preferences stipulated through profile bios [14].
A key focus of this paper is gender. Several prominent
sociological studies have investigated the nature of online
dating across genders [15], [16], as well as the characteristics
of the people using such services [2]. For example, Hitsch et
al. found that, on average, women show greater preference
than men for income over physical attractiveness [3]. This
is consistent with more traditional social theory, which has
found that, on average, men pay more attention than women
to youth and physical attractiveness, whilst women place more
prominence on social status [17], [18]. Interestingly, the cut
down nature of Tinder means that users are forced into making
decisions on primarily aesthetic qualities. Bolig et al. found
that personal adverts left in a magazine’s lonely hearts column
tended to show that physical characteristics are most salient in
identifying potential mates [19]. This is powerful motivation
for an app like Tinder, however, it is unclear how textual
descriptions of one’s self in a magazine compare to pictures
in Tinder. Underpinning this, various works have investigated
what is considered attractive, such as a female preference for
taller men [3], and a male preference for larger eyes [20].
Perhaps most profound is the tendency towards homophily for
users of online dating services, i.e., the propensity to pursue
partners who are similar to themselves [21]. Interestingly,
however, Hitsch et al. noted that in online settings, users
tend to break away from traditional models of homophily
measured by attractiveness; instead, users pursue attractive
users regardless of their own appearance. Currently, how this
maps to Tinder is unknown. We are particularly curious to see
how this pursuit translates into matching and messaging trends
between men and women. For instance, it has been found that,
in traditional dating websites, men and women exhibit similar
messaging rates early on in their lifecycles, but women tend
to reduce the number of messages sent over time while men
continue at the same rate [22].
In this paper, we provide vantage on Tinder. As much
as possible, unlike the above studies, we avoid categorising
users into groups of attractiveness or desirability. Instead, we
simplify our analysis by solely investigating the differences
between male and female profiles. It has been long understood
that the genders exhibit quite different patterns of behaviour
in dating. However, in contrast to prior studies, we posit
that Tinder suffers from far greater ambiguity; whereas other
dating services make a particular effort to pair appropriate
people, Tinder makes no such allowances. There is even a
lack of guidelines that might create a commonality of intent
between the user base. Instead, Tinder’s purpose is left open
to interpretation, allowing the ecosystem to emerge from a
bottom-up perspective. Thus, profiling its usage is key for
understanding this recent social phenomenon. To the best of
our knowledge, this is the first measurement study of Tinder.
III. METHODOLOGY
A. Data Collection
To study Tinder we have performed a measurement cam-
paign similar to [4]. This has involved injecting new profiles
specifically designed to record interactions initiated by other
users. This allows us to (i) collect thousands of user profiles,
including the demographics, bios etc.; and (ii) see which of our
profiles gain the most likes from other users. Our methodology
involves two stages: (i) manually creating curated profiles; and
(ii) injecting the profiles into a locale to collect data. Table I
presents an overview of the 14 curated profiles we created. To
design these profiles, we have followed the methodologies of
past studies [4], [23]. We use simple profiles that reflect the
characteristics of an “average” user, as defined by preliminary
measurements. Our methodology is intended to reduce the
number of variables that could potentially conflate results.
With this in mind, we restrict profile pictures to a single facial
shot and do not include any biography. We use facial shots
to prevent users interpreting extended information from the
clothing or background setting, e.g., income, education [8].
We also exclusively use Caucasian pictures to avoid the
complexities introduced by racial homophily [21].
We next describe the pictures used for the 14 profiles
described in Table I.1All profiles (bar 2) were placed in
London, to remove the bias introduced by different cities. The
profiles entitled stock are generated using a set of copyright-
free stock photos (each profile has one picture). The profiles
1Pictures are available at: http://www.eecs.qmul.ac.uk/∼tysong/tinder/pics.html
3
Profile #Likes #Matches
male-stock-1 43290 238
male-stock-2 44543 309
male-stock-3 43255 338
female-stock-1 6631 682
female-stock-2 10550 1109
female-stock-3 5652 577
male-no-pic 47695 79
female-no-pic 9554 610
male-account-disabled 71207 109
female-account-disabled 10526 936
male-real-1pic 86440 234
male-real-3pic 79936 1568
female-real-1pic 9337 1681
female-real-3pic 10030 2319
TABLE I
OVERVIEW OF MEASUR EMENT PROFILES
entitled no-pic contain no pictures, whereas the account-
disabled photos contain a picture saying the account has been
disabled. These were used as a benchmark against which the
picture-enabled profiles can be compared. Finally, we also
created profiles of a male and a female volunteer to allow
us to request extra pictures; these were placed in New York
to avoid bias caused by potential real-world relationships in
London (note, we only compare across results for profiles in
the same city). In all cases, we set the profile age to 24 as this
was the most frequently observed age in our early data.
Once all profiles had been created, we wrote software
to automatically register the profiles as available in a given
location. The software then exhaustively retrieved all profiles
within a 100 mile radius and clicked like for each one. If
any profiles generated a match, we recorded the timestamps
of the match and messages sent. Metadata (i.e., age, bio,
number of pictures) was recorded for all profiles returned.
It is worth noting that our female profiles performed fewer
likes than their male counterparts. This occurred because
Tinder automatically limits users with excessive numbers of
matches (our female profiles accumulated matches faster than
our males, c.f., §IV). To mitigate the impact of this, we
separate all analysis into male and female groups, ensuring
only proportional comparisons are performed between groups.
In total, we collected 230k male profiles, and 250k female
profiles. 12% of male profiles were homosexual or bisexual,
whereas this was the case for only 0.01% of female profiles.2
As such, we focus primarily on heterosexual users. We empha-
sise that our study is not intended to measure attributes like
beauty or attraction. Instead, the above profiles are intended
to provide us with a vantage point into Tinder. We therefore
solely use them to trace behaviour, rather than attempting
to measure fine-grained concepts of attractiveness. We later
explore the importance of this distinction.
B. Ethical Considerations
We have gone through Institutional Review Board proce-
dures. During this process, we made a number of consider-
ations. A first key concern was that personally identifiable
2We categorise a user as homosexual/bisexual if the profile is returned to
one of our profiles of the same gender.
0
100
200
300
400
500
600
700
800
900
1000
00:00
01:00
02:00
03:00
04:00
05:00
06:00
07:00
08:00
09:00
10:00
11:00
12:00
13:00
14:00
15:00
16:00
17:00
18:00
19:00
20:00
21:00
22:00
23:00
#Matches
Time
Fig. 1. Match arrival distribution
information may be revealed during the data collection. We
therefore avoided recording user profile pictures or names.
Further, we did not collect message data sent by users. To
analyse user bios, it was necessary to collect the text data.
However, such data is publicly available and, as such, users
are already aware that unknown third parties will be access-
ing it. The second concern was the nature of participants’
interactions. Thousands of users matched with our curated
profiles, some of whom sent messages. In all cases, we did
not respond. This wasted the time of users and potentially
generated frustration. However, a critical fact is that Tinder
is highly ephemeral with few parallels to traditional dating
services. Users may ‘like’ hundreds of profiles per day and our
questionnaire revealed that not initiating conversations and/or
meet-ups are very common. Consequently, liking one of our
curated profiles will generate little disruption for users. As
such, due to the game-like nature of the process, the impact
is trivial; this is an observation we made through consultation
with many real Tinder users.
IV. A MATCH MADE IN HEAVEN?
First, we wish to explore the differences in matching
and messaging rates for male and female users on Tinder.
We hypothesise that, in-line with evolutionary social theory,
gender will have a significant impact on these measures. We
begin by simply inspecting when users choose to use Tinder.
Figure 1 presents the matching across the time of the day
for our London profiles. Clear diurnal patterns can be seen.
Although activity is observed across the entire day, users tend
to peak around 9:00 and 18:00. These are prime commuting
hours in London; clearly, users have a tendency to use Tinder
to pass the time during their commute. This is just one of
the many benefits (or side effects) of embedding dating into
mobile devices. Usage also continues into the evening, with
matches reducing in the later hours (21:00 onwards). These
patterns are consistent across both male and female users.
Next, we ask if there is a noticeable difference in the
popularity of our male and female profiles, as measured by
matching and messaging rates. Figure 2 shows the percentage
of matches obtained across our various stock profiles. This
confirms a stark contrast. Our male profiles like a large number
of other users, but only match with a small minority (0.6%).
The opposite can be seen for our female profiles, who attain
4
0
5
10
15
20
25
30
male-stock-1
male-stock-2
male-stock-3
male-no-pic
male-account-disabled
male-real-1pic
male-real-3pic
female-stock-1
female-stock-2
female-stock-3
female-no-pic
female-account-disabled
female-real-1pic
female-real-3pic
Matches(%)
Male Female
Fig. 2. The number of matches per profile
0
100
200
300
400
500
600
700
1 2 3 4
#Matches
Time(h)
female-stock-1
female-stock-2
female-stock-3
male-stock-1
male-stock-2
male-stock-3
Fig. 3. The number of matches over time per profile (only stock profiles)
a far higher matching rate (10.5%). Of course, we cannot
be sure that every user we liked was also presented with
our profile and, as such, these values offer a lower-bound
on potential matches. That said, the contrast between our
male and female profiles is stark. More remarkable is the
observation that nearly all matches received come from men.
This includes our male profiles, indicating that homosexual
men are far more active in liking than heterosexual women.
Even though the male:female ratio in our dataset is roughly
even, on average, 86% of all the matches our male profiles
receive come from other men. There is, however, a notable
outlier. Our male-account-disabled profile actually acquires all
of its matches from females. This profile simply has a picture
stating that Tinder has taken down the account.
These observations are also reflected in the temporal trends
of matching. Figure 3 shows how the matches occur over
time with our stock photo profiles. Quite distinct trends are
present that are common amongst all profiles belonging to each
gender. Male profiles slowly build up matches over time, with
a very shallow gradient of increase. In contrast, female profiles
gain rapid popularity, achieving in excess of 200 matches in
the first hour. The linear nature of these likes is particularly
curious with a constant probability of matching over time. As
such, algorithmic techniques for facilitating earlier matches
would be highly beneficial for men, but largely unnecessary
for women.
Once a profile has matched, the two users can exchange
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
10 20 30 40 50 60
CDF
Time(minutes)
Female
Male
Fig. 4. Delay between match and first message sent
1
10
100
1000
Male Female
#Characters
Fig. 5. Distribution of message lengths
messages. Again, differences can be seen. Overall, we find that
21% of female matches send a message, whereas only 7% of
male matches send a message. Thus, women who match with
us are 3 times more engaged than men. This is likely driven
by the sheer number of male matches. Overall, we received
8248 male matches, most of whom do not pursue interaction.
In contrast, we garnered only 532 female matches, suggesting
that they are more careful with whom they like and therefore
consider it more worthwhile to send a message [24]. This is
rather different to other online dating services, where mes-
sages are usually the initial means of establishing interaction
(without the prior need to “match”).
A further difference is the speed by which users pursue
interaction. Figure 4 presents the cumulative distribution func-
tion (CDF) of the delay between a match occurring and a
message being sent. This reveals a significantly faster pace
of interaction than that seen in traditional online dating [10].
63% of messages sent by men occur within 5 minutes of the
match taking place. This is only 18% for women, suggesting
that female users often wait to receive a message first. The
median delay for sending messages is just 2 minutes for men,
compared to 38 minutes for women. This could be driven
by several factors, but it is well known that men often have
to compete and differentiate themselves more as part of the
mating ritual [25]. Their efforts, however, are not always
particularly emphatic. Figure 5 shows the message length
distribution. The median message length sent by men is only
12 characters, compared to 122 from women. For men, 25%
of message are under 6 characters (presumably “hello” or
“hi”). Consequently, it is clear that little information is being
imparted in opening conversations.
The above results show that men are willing to like a larger
proportion of women. In the most extreme case, this could
5
0
100
200
300
400
500
600
700
800
900
234567
#Matches
Number of profiles
Male Female
Fig. 6. Number of our profiles other users have matched with
0
5000
10000
15000
20000
20 25 30 35 40 45 50
Male
0
5000
10000
15000
20000
20 25 30 35 40 45 50
Female
Fig. 7. Age distribution of male and female users
involve clicking like for all users encountered. In contrast,
women are far more discerning. We conjecture that this creates
a “feedback loop” in Tinder. Men see that they are matching
with few people, and therefore become even less discerning;
women, on the other hand, find that they match with most men,
and therefore become even more discerning. To briefly explore
this hypothesis, Figure 6 presents the number of our profiles
that each individual user matched with. Only 6% of women
liked multiple of our accounts, whilst 16% of male profiles
match with multiples of our profiles. This confirms that men
are more liberal in whom they like. 4% of male profiles even
match with in excess of three of our profiles.
V. A TALE OF TWO GENDERS
The previous section has measured the differing popularities
of our male and female profiles. We next take an exploratory
approach, characterising all Tinder profiles collected by their
age, profile pictures and bio. To investigate the importance of
these three attributes we also perform controlled experiments
to observe their impact on popularity.
A. Age
We begin by inspecting the age distribution of users. Fig-
ure 7 presents a histogram of the age distribution for both male
and female profiles. There are broad similarities across the two
genders, although women tend to be marginally younger. The
mean age for women is 25.2, compared to 25.7 for men. This
is somewhat different to more typical online dating platforms,
0
100
200
300
400
500
600
20 25 30 35 40 45 50
Male
0
10
20
30
40
50
60
20 25 30 35 40 45 50
Female
Fig. 8. Age distribution of (a)male users matching with our female profiles
and (b)female users matching with our male profiles
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 1 2 3 4 5 6
CDF
Number of profile photos
Women
Men
Fig. 9. CDF of number of pictures on male and female profiles
where there is a greater prominence of older women compared
to men [22]. This is potentially due to the reputation that Tin-
der has for facilitating immediate and short-term relationships,
which is often less desired by older women [25]. We can also
compare this age distribution with the matches received for
our profiles (all of which were set to 24 years old). Figures 8
plots the ages of other users matching with our profiles. It can
be seen that the distribution of male users matching with us is
consistent with the overall population (an average of 25.7 vs.
25.8), whereas it is different for female profiles. Across the
whole dataset, the average age of females is 25.2; for matches,
this drops to 24.3. This is caused by a lack of older women
matching with our 24 year old profile; a propensity that is
frequently observed in mating [25]. It is also mirrored in other
online dating environments; Fiore et al. found that women in
their 20s and 30s were more likely to interact with older men,
whereas men consistently sought younger women [10].
B. Profile pictures
It has been found that 77% of online dating profile views
are for profiles with at least one photo [3]. Hence, we next
explore the role of profile pictures in Tinder. Figure 9 presents
the CDF of the number of profile pictures for males and female
profiles. This reveals a relatively similar numbers of pictures
on male and female profiles: an average of 4.4 for men vs. 4.9
for women. The next question is therefore how do the genders
view the importance of the number of pictures?
6
0
200
400
600
800
1000
1200
1 2 3 4
#Matches
Time(h)
female-real-3pic
female-real-1pic
male-real-3pic
male-real-1pic
Fig. 10. The number of matches over time per profile (only real profiles)
0
50
100
150
200
250
300
350
400
450
500
male-stock-1
male-stock-2
male-stock-3
male-stock-1
male-stock-2
male-stock-3
#Matches
Without bio
With bio
Without bio
With bio
Female matchesMale matches
Fig. 11. Matches per profile: bio vs non-bio (only male stock profiles)
To explore this, Figure 10 shows the number of matches
over time for our two real profiles (see Table I) when using
differing numbers of profile pictures. Note that our real profiles
were those contributed by volunteers to allow us to request
multiple specific photos (see §III). It can be seen that the
number of pictures has a notable impact. When increasing
the female profile from 1 to 3 pictures, a 37% increase is seen
in matches. This is even more significant in male profiles.
With a single profile picture, after 4 hours, only 44 matches
were made, whereas this increased to 238 with three pictures.
Even more interesting is the observation that the fraction of
women liking our male profile increases most dramatically.
After four hours, only 14 of our male profile matches were
from women; with three pictures, this number increased to 65.
This is likely partly driven by the greater concern that women
have of deception [12], alongside their preference for deeper
information about mates [24]. Either way, the need for men
to have multiple pictures is far greater than for women. Those
attempting to improve their matching rates should consider
this closely.
C. Bios
Tinder allows users to provide a short biography about
themselves. Figure 12 presents the distribution of bio lengths
(number of characters) for both male and female profiles. In-
line with the core principles of Tinder, people do not provide
much information. 36% of accounts have no bio, with the
majority under 100 characters (maximum is 500). This is
particularly prevalent amongst women, 42% of whom have
blank bios. Considering that free-text within dating profiles
can heavily contribute to attractiveness [9], this tendency in
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 100 200 300 400 500
CDF
#Characters in bio
Male
Female
Fig. 12. CDF of the number of characters in user bios
Tinder could be seen as a negative one. Likely the ability
of female profiles to gain high matching rates regardless has
reduced the value of text. Their perceived importance is greater
for male profiles though [25], shown by the smaller proportion
of male profiles with blank bios (30%).
To investigate this hypthoesis we recreated our male stock
profiles, but with short bios simply saying hello and that they
are from London. We focus on male profiles as female profiles
do well regardless. Figure 11 presents the matches attained.
For each of our male profiles, we present the number of
matches they attain both with and without a bio. The blue bars
count the number of matches accumulated from other men,
whilst the pink bars count the number of matches accumulated
from other women. In all cases, the profiles with bios do
far better. This is particularly the case for acquiring female
matches. Without bios, our male stock profiles received an
average of 16 matches from women; this increases four-fold to
69 with a bio. The number of matches from men also increases,
but far less substantially (by 58% on average).
VI. EXPLORING USER PERSPECTIVES
We next seek to validate and further explore the meaning
behind our findings. To achieve this, we have performed a
user questionnaire. This is intended to provide context to our
earlier findings. The questionnaire was distributed via social
media and various mailing lists. Naturally, self-selection and
reporting could introduce bias and, thus, we use the survey
data primarily to drive discussion. Due to a lack of data on
homosexual users, we filter all entries to leave users seeking
heterosexual interactions; this leaves 131 responses, 90 from
men, and 41 from women. In-line with our earlier empirical
observations, nearly all participants (83%) were under the
age of 35. We primarily recruited frequent users, with 87%
reporting using Tinder at least once a week. Interestingly,
we also noted that many users (68%) stated they had used
other online dating services before, indicating that Tinder is
not necessarily the first (or only) port of call for new online
daters.
To support our earlier findings, we began by asking users
to estimate the percentage of profiles that they click like
for. Figure 13 presents the probability density function of
the responses for the two genders. In-line with expectations,
7
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
1-5
6-10
11-20
21-30
31-40
41-50
51-60
61-70
71-80
81-90
91-100
Fraction of Participants
Estimated % of Likes
Male
Female
Fig. 13. Probability density of estimated percentage of likes
Men Women KS
One night stand 3.2 1.8 0.00034
Chat 3.0 3.7 0.29122
Casual Dating 3.8 3.3 0.33535
Look at Profiles 3.5 3.8 0.35559
Meet Partner 3.5 3.3 0.99094
TABLE II
MEAN SCORE FOR EACH I NTENT WHEN USI NG TINDER (ORDERED BY THE
EMPI RICAL DISTANCE BETWEEN MALE AND FEM ALE AN SWERS)
women report that they are far more selective in whom they
like: 59% of women estimate that they like 10% or fewer of
all profiles they encounter. In contrast, only 9% of men report
such high selectivity. To explore what led to this, we went on to
ask participants what their intentions were when using Tinder.
Specifically, we asked “Please rank your intentions when
using Tinder”. We gave several options in the questionnaire:
(i) I use Tinder to look at profiles; (ii) I use Tinder to chat
with people online; (iii) I use Tinder to find a partner; (iv) I
use Tinder for casual dating; and (v) I use Tinder for one night
stands. We asked the participants to rank each option from 1
to 5 (with 5 being the most important). Table II presents the
mean score allocated to each intent by the two genders ordered
by the difference between the genders’ responses (measured
by the Kolmogorov-Smirnov (KS) p-value). The higher the p-
value, the more similar the responses from male and female
participants are. There are broadly similar priorities reported
for most intents across the two genders. For instance, both
genders frequently report using Tinder for finding a partner.
More interesting is the intention that the two genders report
profoundly different rankings on. The mode average score
given by men to one night stands is 5, compared to just 1 by
women (this polarisation is reflected in an extremely low KS
p-value). 49% of male respondents rated “I use Tinder for one
night stands” as 4 or 5, compared to just 15% of women.
Tinder’s inability to differentiate between the intentions of its
users therefore inevitably leads to many matchings between
parties who are looking for very different things. This is hinted
at by the extremely low rate at which people report meeting
up with their matches. 73% of respondents estimated that 10%
or fewer of their matches result in a real-world meet-up — as
one would expect, the distribution of meet-up estimates are
near identical across both male and female respondents (for
heterosexual users, these are interdependent).
To investigate how these intentions translate to liking strate-
gies, we next asked participants “Which liking strategies do
you most frequently employ?”. Multiple options could be
selected from: (i) I casually like most profiles; (ii) I only
like profiles that I’m attracted to; (iii) I adapt my selectivity
based on how many matches I am getting that day; (iv) I
adapt my selectivity based on what I am looking for at the
time (e.g., chat, date, relationship); and (v) Other. Table III
presents the results as the percentage of respondents stating
that they use a given strategy frequently. Clear differences
between the genders emerge. Most blatant is the “I casually
like most profiles” strategy. Whereas 33% of men report
using this strategy regularly, no women report ever using this.
Instead, 93% of women report exclusively liking profiles they
are explicitly attracted to (this is how Tinder is intended to be
used). Interestingly, 13% of men also state that they regularly
adapt their liking rate based on how many matches they are
receiving.
We can now combine the above observations to explore
how these liking strategies impact the matching rates of the
opposite gender. To do this, we asked respondents to estimate
the percentage of their likes that turn into matches. Figure 14
presents the results as a probability density function. In-line
with our earlier empirical findings, women report far higher
probabilities of matching, with 63% of women estimating that
over half of the profiles they like result in a match. With
such high matching rates, it is unsurprising that women use a
strategy whereby they only like profiles they are confidently
attracted to. In contrast, 59% of male respondents estimate
that 10% or under of their likes result in a match. A possible
interpretation of the above data is that women get higher
matching rates simply because they put more effort in, and
find better potential partners to like. However, the very low
messaging rates indicate that this is not the case: 49% of
women estimate that under 30% of their matches result in
a conversation.
These findings support our earlier suspicions, and partially
explain why our female profiles found it easier to obtain high
matching rates: a notable proportion of men casually like
most profiles. 80% of male respondents who use this strategy
report liking in excess of half of all women encountered. Some
male respondents explicitly stated that this was caused by the
rarity of matches they achieve, adding weight to our earlier
discussion of an intuitive “feedback loop”. This behaviour not
only undermines the functionality of Tinder, but also shows
that, contrary to intuition, Tinder cannot necessarily be used
as an accurate tool for measuring (female) attractiveness or
certain social phenomena (e.g., homophily).
VII. CONCLUSION
We have presented a study of user activity in Tinder,
exploring how it differs between the two genders. We began
by asking (i) what impact does gender have on matching and
messaging rates, and (ii) what profile characteristics are com-
mon across the two genders? We have shown that male users
like a far higher proportion of profiles than females. Women,
however, have a greater propensity to establish conversation
8
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
1-56-10
11-20
21-30
31-40
41-50
51-60
61-70
71-80
81-90
91-100
Fraction of Participants
Estimated % of Matches
Male
Female
Fig. 14. Probability density of estimated percentage of likes that result in a
match
Men Women
I casually like most profiles 35% 0%
I only like profiles that I’m attracted to 72% 91%
I adapt my selectivity based on how many
matches I am getting that day 13% 4%
I adapt my selectivity based on what I am
looking for at the time 16% 22%
TABLE III
PERCENTAG E OF PA RTICI PANTS W HO REP ORTED FREQUENT LY US ING
EACH STRATEGY (O RDER ED BY TH E EMPI RICA L DISTANCE B ETWE EN
MALE AND FEMALE ANSWERS)
via messaging, although they tend to leave a longer interval
between matching and messaging than male users. It therefore
seems that, rather than pre-filtering their mates via the like
feature, many male users like in a relatively non-selective way
and post-filter after a match has been obtained. This gaming
of the system undermines its operation and likely leads to
much frustration. Through controlled experiments, we have
also observed clear attributes that can increase a user’s ability
to match, e.g., bios and profile pictures. This is something that
can help inform Tinder users in how to improve their online
matching rates.
It is important to note that our measurement study has
limitations. Most notably, we only sample users from the
vantage of a relatively small set of accounts in two urban
areas. These do not necessarily cover the whole spectrum
of attractiveness and, therefore, they may miss out matches
that would have happened otherwise. So far, we have also
treated Tinder as a black-box; the exact mechanism by which
it selects profiles to present is unknown. Reverse-engineering
and improving this could be highly productive and could
add extra depth to our results (although note that all our
profiles were setup in the same manner, ensuring equivalent
comparisons). Beyond this, there is a wealth of exciting lines
of future work that remain. Most important is expanding to
more sophisticated setups, e.g., comparing profiles based on
physical characteristics such as height or race. Through this,
we are keen to further explore our hypotheses and confirm that
our observations are generalisable across more profile types.
REFERENCES
[1] A. Smith and M. Duggan, “Online dating & relationships,” Pew Internet
& American Life Project, 2013.
[2] P. M. Valkenburg and J. Peter, “Who visits online dating sites? exploring
some characteristics of online daters,” CyberPsychology & Behavior,
vol. 10, no. 6, 2007.
[3] G. J. Hitsch, A. Hortac¸su, and D. Ariely, “What makes you click? mate
preferences in online dating,” Quantitative marketing and Economics,
vol. 8, no. 4, pp. 393–427, 2010.
[4] S. Webb, J. Caverlee, and C. Pu, “Social honeypots: Making friends
with a spammer near you.,” in Proc. CEAS, 2008.
[5] E. Toch and I. Levi, “Locality and privacy in people-nearby applica-
tions,” in Proc. ACM UBICOMP, 2013.
[6] M. J. Handel and I. Shklovski, “Disclosure, ambiguity and risk reduction
in real-time dating sites,” in Proc. ACM GROUP, 2012.
[7] C. Rudder, Dataclysm: Who We Are (when we think no one’s looking).
Crown, 2014.
[8] J. H. Langlois, L. Kalakanis, A. J. Rubenstein, A. Larson, M. Hallam,
and M. Smoot, “Maxims or myths of beauty? a meta-analytic and
theoretical review.,” Psychological bulletin, vol. 126, no. 3, 2000.
[9] A. T. Fiore, L. S. Taylor, G. A. Mendelsohn, and M. Hearst, “Assessing
attractiveness in online dating profiles,” in Proc. ACM CHI, 2008.
[10] A. T. Fiore, L. S. Taylor, X. Zhong, G. A. Mendelsohn, and C. Cheshire,
“Who’s right and who writes: People, profiles, contacts, and replies in
online dating,” in Proc. HICSS, 2010.
[11] L. C. Tidwell and J. B. Walther, “Computer-mediated communication
effects on disclosure, impressions, and interpersonal evaluations: Getting
to know one another a bit at a time,” Human Communication Research,
vol. 28, no. 3, 2002.
[12] J. L. Gibbs, N. B. Ellison, and C.-H. Lai, “First comes love, then comes
google: An investigation of uncertainty reduction strategies and self-
disclosure in online dating,” Communication Research, vol. 38, no. 1,
2010.
[13] J. T. Hancock, C. Toma, and N. Ellison, “The truth about lying in online
dating profiles,” in Proc. ACM CHI, 2007.
[14] J. Akehurst, I. Koprinska, K. Yacef, L. Pizzato, J. Kay, and T. Rej, “Ex-
plicit and implicit user preferences in online dating,” in New Frontiers
in Applied Data Mining, Springer, 2012.
[15] C. L. Toma and J. T. Hancock, “Looks and lies: The role of physical
attractiveness in online dating self-presentation and deception,” Commu-
nication Research, vol. 37, no. 3, pp. 335–351, 2010.
[16] M. T. Whitty, “Cyber-flirting: an examination of men’s and women’s
flirting behaviour both offline and on the internet,” Behaviour Change,
vol. 21, no. 02, 2004.
[17] L. L. Lance, “Gender differences in heterosexual dating: A content
analysis of personal ads,” The Journal of Men’s Studies, vol. 6, no. 3,
pp. 297–305, 1998.
[18] R. Fisman, S. S. Iyengar, E. Kamenica, and I. Simonson, “Gender
differences in mate selection: Evidence from a speed dating experiment,”
The Quarterly Journal of Economics, vol. 121, no. 2,pp. 673–697, 2006.
[19] R. Bolig, P. J. Stein, and P. C. McKenry, “The self-advertisement ap-
proach to dating: Male-female differences,” Family Relations, pp. 587–
592, 1984.
[20] G. Rhodes, “The evolutionary psychology of facial beauty,” Annu. Rev.
Psychol., vol. 57, pp. 199–226, 2006.
[21] A. T. Fiore and J. S. Donath, “Homophily in online dating: when do
you like someone like yourself?,” in Proc. ACM CHI, 2005.
[22] P. Xia, B. Ribeiro, C. Chen, B. Liu, and D. Towsley, “A study of user
behavior on an online dating site,” in Proc. ASONAM, 2013.
[23] H. Haddadi and P. Hui, “To add or not to add: privacy and social
honeypots,” in IEEE ICC, 2010.
[24] R. Trivers, Parental investment and sexual selection. Harvard University,
1972.
[25] A. W. Kruglanski and W. Stroebe, Handbook of the history of social
psychology. Psychology Press, 2012.