International Gender Differences and Gaps inOnline Social Networks
Gabriel Magno Ingmar Weber
This work was done while the first author was at QCRI
The Global GenderGap Report
Global Gender Gap Report● Introduced in 2006● Captures the magnitude
and scope of gender-based disparities and tracks their progress
● Designed to create awareness of the challenges posed by gender gaps
Global Gender Gap Index - Variables
● Social variables related to basic rights● Four categories (sub-indexes):
– Economy: wage, income, # managers, etc
– Education: literacy rate, educ. levels enrollment
– Health: births, life expectancy
– Politics: # seats in parliament, # ministers, etc
Global Gender Gap Index - Algorithm
1. Calculate the female by male ratio of the variables;
2. Truncate the ratios at a certain level;
3. Calculate sub-indexes for each category;
4. Calculate the average of the four sub-indexes to create the overall index.
Scores: 0.0 (total inequality)→ 1.0 (total equality)→
Google+
Google+ Dataset● Date: 1st semester of 2012● Extracted all IDs from
Google+'s sitemap– 193 million IDs
● Parsed profile and graph information
– 160 million profiles– 61 million nodes– 1 billion edges
Google+ User Information ● Country: last location from
the "Places lived" field.– 22 million users
● Gender: self-declared field.– 34.4% female– 63.8% male – 1.8% other
Google+ Variables - Network● In-degree: number of followers.
● Out-degree: number of friends.
● Reciprocity: fraction of reciprocal links.
● Clustering coefficient: probability of any two neighbors being neighbors.
● PageRank: relative importance of a user in the network.
Google+ Variables - Assortativity● Assortativity: fraction of links to the same gender.
– High value strong same-gender linkage, cross-gender links →are less likely to happen.
● Differential assortativity: "lift" of the fraction of users of the same gender followed by a particular user.– Example: computer science students (males linked to male)– High value the user is more likely than by random chance to →
follow other users of his/her same gender.
Methodology
Dataset Selection
● Users we know both gender and country
● Countries with at least 5,000 females and males
● Countries that are in the Gender Gap Report
73 countries 17 million users
Gender Ratio Algorithm1. Calculate metric
for each user
12
10
17
11
11
16
15
12
10
17
11
11
16
15
12
14
15
11
10
16
12
14
15
11
10
16
0.9
1.6
0.7
2. Group users by country and gender
● Calculate average of the metric
3. Group values by country
● Calculate gender ratio (f/m)
A
B
C
D
E
F
G
A
C
D
F
B
E
G
Gender Differences
Gender Ratio
Female predominance for Reciprocity and
Clust. Coeff.
Male predominance for # followees
Differences among countries for #
followers and PR
Online vs. Offline Gender Gaps
Gender Gap vs. # users
Countries with lower gender equality more →men than women online
Gender Gap vs. # followers
Countries with low offline equality women are, →surprisingly, followed more than men
Gender Gap vs. Assortativity
Countries with high gender equality →higher assortativity
Countries with low gender equality →women have higher assortativity
Countries with high gender equality →no difference between genders
Discussion
The Jackie Robinson Effect● Jackie Robinson: 1st African-
american baseball player to play in Major League Baseball (1947)
● Probably, only had the chance to play because he was really good
● Women who decided to go online in a country such as Pakistan are likely to be more self-confident and tech-savvy than random male counterparts
Online Stalking● “Stalking”: women attracting follow links from
men● In countries with low gender equality, this effect
might be stronger, so women have more followers than men
● In countries with low gender equality, women might shy away from cross-gender links, so female assortativity is higher than men
Conclusion
Concluding Remarks● Large-scale study of gender differences and
gender gaps around the world in Google+
● Online indicators can capture the offline gender gap trend among countries– # users positive correlation→
– # followers negative correlation→
Thank You!
http://www.dcc.ufmg.br/~magno
@GabrielMagno
Backup Slides
Correlations