Big Data and the Global Gender Gap: The Promises and Perils of Digital Information

Rebecca Furst-NicholsDeputy Director, Data2X

Bapu VaitlaFellow, Data2X

UN System Staff College Knowledge Centre for Sustainable DevelopmentSD Talks Special Series on Data for Sustainable DevelopmentMarch 20th, 2018

Data2X: Our Mission

Data2X works to increase the availability and use of quality gender data. We:

• Promote expanded, unbiased, and innovative gender data collection.• Identify gender data gaps for priority attention.• Lead partnerships to close gender data gaps.• Advocate for better gender data and its use in decision making.• Educate about how to improve and use gender data to improve lives and

outcomes for all.

What is Gender Data?

• Data that is disaggregated by sex, such as primary school enrollment rates for girls and boys

• Data that pertains specifically to girls and women as a result of biology or their social roles, such as maternal mortality rates or unpaid care work

Where are the Gender Data gaps?

There are over 28 identified gaps in gender data based on need, population coverage, and policy relevance across five domains:

Defining Big Data

Partnerships: Big Data for Gender

Geospatial Data (Satellite Imagery)Flowminder Foundation

1. Obtain survey point data on well-being (e.g., DHS)

2. Obtain geospatial data at same locations: population density, infrastructure, vegetation type (satellite, etc.)

3. Correlate the two sets of info

4. Predict landscape of data on well-being

Geospatial Data (Satellite Imagery)Flowminder Foundation

1. Obtain survey point data on well-being (e.g., DHS)

2. Obtain geospatial data at same locations: population density, infrastructure, vegetation type (satellite, etc.)

3. Correlate the two sets of info

4. Predict landscape of data on well-being

Credit Card/Cell Phone Data: Economic LifestylesDi Clemente, Gonzalez, et al. (MIT)

- Anonymized credit card data from 150k users, with age, sex, location info

- Subset: cell phone data- Portraits of economic lifestyles:

mobility, access, preferences- Could illuminate how women

cope with economic/environmental shocks & stresses

Credit Card/Cell Phone Data: Economic LifestylesDi Clemente, Gonzalez, et al. (MIT)

- Anonymized credit card data from 150k users, with age, sex, location info

- Subset: cell phone data- Portraits of economic lifestyles:

mobility, access, preferences- Could illuminate how women

cope with economic/environmental shocks & stresses

Cell Phone Data: EpidemiologyWesolowski et al. (Carnegie Mellon, Harvard, etc.)

- Locations of ~15m cell phone subscribers; travel maps

- Malaria prevalence map based on existing data

- Source/sink maps based on human movement and parasite prevalence

- Allows precisely targeted interventions, in time and space

Cell Phone Data: EpidemiologyWesolowski et al. (Carnegie Mellon, Harvard, etc.)

- Locations of ~15m cell phone subscribers; travel maps

- Malaria prevalence map based on existing data

- Source/sink maps based on human movement and parasite prevalence

- Allows precisely targeted interventions, in time and space

Cell Phone Data: EpidemiologyWesolowski et al. (Carnegie Mellon, Harvard, etc.)

- Locations of ~15m cell phone subscribers; travel maps

- Malaria prevalence map based on existing data

- Source/sink maps based on human movement and parasite prevalence

- Allows precisely targeted interventions, in time and space

Cell Phone Data: EpidemiologyWesolowski et al. (Carnegie Mellon, Harvard, etc.)

- Locations of ~15m cell phone subscribers; travel maps

- Malaria prevalence map based on existing data

- Source/sink maps based on human movement and parasite prevalence

- Allows precisely targeted interventions, in time and space

Social Media: Patterns of Mental Health on TwitterDe Choudhury et al. (Georgia Tech)

1. Key signal phrases (validated with diagnostic test)

2. Analyze positive/negative affect, cognitive attributes, lexical density, social concerns, etc.

3. Compare men/women, control versus mental illness disclosure samples

Social Media: Patterns of Mental Health on TwitterDe Choudhury et al. (Georgia Tech)

1. Key signal phrases (validated with diagnostic test)

2. Analyze positive/negative affect, cognitive attributes, lexical density, social concerns, etc.

3. Compare men/women, control versus mental illness disclosure samples

IV. Internet Activity 26

We can further break down the attribute

categories. Female users in the MID sample

show 15.4% higher sadness and 10.7%

higher anxiety; prior literature indicates

that expression of these emotions is asso-

ciated with depression, mental instability,

and feelings of helplessness, loneliness, and

restlessness. However, female users also tend

to use 7.1% more positive af ect in their

content, perhaps to demonstrate a positive

outlook publicly despite the mental health

challenges they are facing. Male users, on

the other hand, express 2.6% more negative af ect overall, including 5.3% higher anger

and 9.5% more expressions with swearing. Females express fewer cognitive attributes on

social media than do males. Lower usage of words that denote certainty, for example,

may demonstrate heightened emotional instability. T ese dif erences in cognitive expres-

sion are not pronounced in the control

sample, however, suggesting that experience

of mental illness, not intrinsic dif erences

between the sexes, is responsible for the

observed gap.

We turn now to social/personal concerns

and interpersonal focus, both subtypes

within the linguistic style category. Male

MID users display an 8.1% lower sense

of achievement than women and girls,

a known signal of reduced self-esteem.22 Female MID users, meanwhile, express 6.0%

greater concern about their health and 2.7% greater concern about their body, which

may indicate a greater self-awareness about their health or, alternatively, more f xation

with social perceptions about their appearance. Another interesting

f nding is that male MID users exhibit lower use of words having to

do with social concerns, friends, or family. T eir female peers may be

using such language more frequently in their Twitter posts to explicitly

seek help from their social networks. T e interpersonal focus metrics

Figure 16. Differences in linguistic measures between female and male users, disaggregated by mental illness disclosure (MID) and control sample (CTL). Positive values indicate higher scores for female users.

“ #depression has invaded my peace and

#anxiety has exhausted my thoughts. Pain

isn’t always physical

– female user

“ why am I even here... No one needs or wants

me... I’m useless

– female user

Abso

lute

dif

fere

nce b

etw

een

fem

ale

an

d m

ale

use

rs (

%)

Aff

ective

att

rib

ute

s

Cogn

itiv

e

att

rib

ute

s

Lexic

al

den

sity

an

d a

ware

ness

Tem

po

ral

refe

ren

ces

So

cia

l/p

ers

on

al

con

cern

s

Inte

rpers

on

al

focu

s

9

8

7

6

5

4

3

2

1

0

CTL usersMID users

“ Over the past 2 years I have been hit with

physical and mental pain. The pain is real. It

is still ther e.

– female user

• Privacy

• Bias and access: Who does big data leave behind?

• Consider access, affordability, literacy, and other barriers

• Country context: One size doesn’t fit all

• Ground truth

• Digital data should enhance, not replace, information gathered from traditional sources like household surveys and censuses

Big Data: Risks and Considerations

What’s Next? Big Data for Gender Challenge

10 projects representing 29 researchers from 20 global institutions across 8 countries

Women in the Gig Economy: A Data Gap with Implications for Informal Work, Time Use, and PovertyLeads: Overseas Development Institute, Ulula, Data-Pop AllianceMethod: Mobile phone-based longitudinal survey

Gender and Urban Mobility: Addressing Unequal Access to Urban Transportation for Women and GirlsLeads: The GovLab, UNICEF, ISI Foundation, Universidad del Desarrollo, Telefónica, DigitalGlobeMethods: High-resolution satellite data; call detail records

Safety First: Perceived Risk of Street Harassment and Educational Choices of WomenLead: Girija Borker, PhD, Brown UniversityMethods: Student surveys, Google Maps travel route data, mobile application data

Learn more about Data2X at

www.data2x.org/big-data-challenge-awards/