Casilli lseisp110510 light

ISP SSS - LSE
Lonfon, May 11th, 2010

Studying Eating Disorders
in the Social Web
New methods, new questions
Antonio A. Casilli, CEM, EHESS

P Tubaro & AA Casilli Promesses et limites des SMA

• Edgar Morin Centre, School for
Advanced Studies in Social
Sciences (EHESS, Paris)
http://en.wikipedia.org/wiki/Edgar
_Morin_Centre
• Interdisciplinary Institute for
Contemporary Anthropology
(IIAC, Paris)
• Research topics: CMC and
health - especially eating
behaviors
• Social network approach to
eating disorders (ANAMIA
project)


TOC

• Pro-ED online communities
• Our approach
• Methods
• Empirical data
• Agent-based simulations
• Conclusions


• Pro-ED online communities


• “Pro-ana”, “pro-mia”:
controversial subcultures
6
advocating anorexia and 5

bulimia nervosa on the

Search Volume Index
4
Ana_mia
web 3 Pro_ana
Pro_mia

• A “movement”? A “skill”? 2

• How to study this social 1

0
phenomenon?
Ja
M 22
Ju 27 5
Se 19 005
D 11 05

M 26 05

Ju 22 07

D 7 2 07
M 30 7
Ju 23 007
Se 15 08
N 7 2 08
Fe 4 2 005

Au 21 006
N 13 006
Ja 5 2 006
Ap 28 06

O 5 2 07

Fe 30 08
ec 2

ec 0

ov 0
ar 00

ay 2

ar 2
ov 2

ct 0
n

n 2

l 1 20

n 20
p 20

p 20
n 0
b 0

b 20
g 2

r 20

22 0
• How to devise suitable

20 8
0

09
public health tools and Evolution over four years of the search volume index for common
ana-mia queries

communication policies? (Source: Google Trends, March 12th 2009)


• Rough typology of
online pro-ED
websites:

1. Personal websites
2. Online social
media/forums
3. ‘Cloaked’ websites
4. ‘Universities’
• ‘Thinspiration’, self-
help, advice


• Our approach


•The pro-ED population is:
•relatively small;
•vulnerable (health risk; underage);
•partly hidden;
•frequent migrations.

•Large quantitative surveys & webcrawling possible only to
an extent
•Rely on smaller-scale, purposive samples for qualitative
enquiry


• Ongoing project
(ANAMIA) : a social
networks approach to
pro-ED sociability
• Social factors
influencing health
behaviours
• Computer use
influencing social
factors
• Focus on online/offine
personal networks


• Methods


•A wide array of methods
•Online ethnographies
•Online experiments
•Social network analysis
•Web-based in-depth interviews
•Multi-agent simulations
•General framework: Ethnographically-
informed social simulation (P Tubaro &
AA Casilli BMS, 2010)


• Combining qualitative data and agent-based computer
simulation:
• enriches model with insight into actors behavior and motivations;
• performs “thought experiments” to test consistency of theories;
• replicates and generalizes findings from fieldwork;
• supports cross-disciplinary validation of results.


• Starting point: an actual
social process;
• Qualitative sub-loop:
formulate hypotheses,
collect data,
• adjust categories, until a
theory is produced;
• Design, build, code and
de-bug an agent-based
model;
• Generate simulated data
and revise theory;
• This may direct back to
the field (resample and
re-start sub-loop).

• Empirical data


• Fieldwork with ana-mia subjects is currently at an early
stage.
• Data collected so far are exploratory, and include:
– At micro level: a qualitative study of network tie
formation on Facebook;
– At macro level: a web cartography of the pro ED-sphere
in France and UK
• Use of preliminary data to inform and validate a first
simulation.


• Facebook experiment
• Creation of social structures
(‘friendship’, ‘social capital’)
through self-disclosure and
adoption of cultural traits
• 50 days (Apr. 27th, - Jun.
15th, 09)
• Name generator and IOS to
select subjects
• An actual and a control
profile send 50 friend
requests
• Recognizable name and
picture
• 15 reply (14 for control)

• “Friends” provide feedback
on how to adapt profile
– Comments
– Private messages
– Like
– Share
• “Friends” appreciate
disclosure
• Actual profile contains more
personal information,
images, interactive data
• Control data contains
sparse information


•Pete Warden, Nicholas Christakis :
«harvesters» and «dataminers»

•«Mark Zuckerberg has built his
social networking empire on the
belief that "information wants to be
shared", a particular philosophy of
information that directly impacts the
values built into the design of
Facebook, ranging from its user
interface, privacy policies, terms of
service, and method of governance »
(Michael Zimmer)
•Social media is ethically puzzling


Actual Profile Control Profile

Starting point 11 05 09



Taste display 19 05 09



Personal information 22 05 09



Photo album online 13 06 09


•Insight from preliminary
qualitative study is that
online network formation
may depend upon:
•Privacy settings, i.e.
visibility of contents to
others;
•Self-display, i.e. personal
and cultural traits exhibited
and that traits may change
with network composition.
•The model aims to
problematize these factors
in simulated larger
networks.


• Agent-based simulations


•We focus on the impact of:
•tendency to conformism vs.
dissonance in cultural traits;
•preference for bonding vs.
bridging in tie formation;
•possibility to limit incoming
ties through privacy protection.
•We measure impact
through:
•number and size of
components;
•homogeneity of traits within
and between components;
•evolution of privacy settings
over time.


•At initialization, each actor is endowed with:
•a vector (several dimensions) of traits;
•a privacy setting (visible/invisible).
•Actors can be:
•isolates;
•connected;
•If connected:
•they share most traits with their contacts;
•but may dffier on one dimension;
•this depends on the ‘Dissonance’ parameter
•We test the following values of parameters:
Paramètres Valeurs
Dissonance 0,01 0,03 0,08
Seuil de bonding 0,20 0,50 0,70
Privacy On Off


SETUP

Tick t+1 Update z(Ai) ≠ µz (friendsi)

Pick up
random A

N
Is anom(A) < AnomThrshld?

Y

Create Tie Break Tie

N N
Create Break
Is µz(A)-µz (grp) < BBThrshd? Is µz(A)-µz (grp) ≥ BBThrshd?
Bridging Tie Bridging Tie

Y
Y

Create Break
Bonding Tie Bonding Tie


• Simulation results (20,000 ticks): three stable configurations

(1) Giant Component (2) Hegemony (3) Little Boxes


Number and size of components, varying Dissonance and Bonding Propensity, privacy protection on


Number and size of components, varying Dissonance and Bonding Propensity, privacy protection off


• Explain the effects of parameters
– With lower propensity to bonding (=greater openness to
– bridging), only one or few components emerge;
– This effect is stronger with higher Dissonance;
– With higher propensity to bonding, many small communities emerge;
– In this case, differences in Dissonance have little impact;
– With no privacy protection, these effects are slightly amplified, because
more ties can be formed.

• Explain changes in privacy over time
– Agents restrict access only when a giant component appears;
– This is the only case in which average privacy increases;
– Otherwise, average privacy diminishes until there are no more isolates,
then is stable.


Average Privacy over time, varying Dissonance and Bonding propensity


Structural signatures of learning dynamics

1. Inprinting 2. No learning

3. Mixed situation (no learning & continuous learning)


•Retrodictive validation
•Web cartography of the pro ED-
sphere in France and UK
•Close to configuration 2
‘Hegemony’
•Large component: personal pages
and blogs of teenagers and young
adults, strongly pro-ana;
•Smaller components: different age
groups and social positioning
(cultural variance)
•Homogeneity within components,
heterogeneity between them.
French pro-ED sphere (by Dr. Manuel Boutet)


•Retrodictive validation
•Web cartography of the pro ED-
sphere in France and UK
•Close to configuration 2
‘Hegemony’
•Large component: personal pages
and blogs of teenagers and young
adults, strongly pro-ana;
•Smaller components: different age
groups and social positioning
(cultural variance)
•Homogeneity within components,
heterogeneity between them.
British pro-ED sphere (by Dr. Manuel Boutet)


• Conclusions


• Preliminary results support the claim that agent-based
models can complement analyses based on small
qualitative fieldworks

• Combine insight into social phenomenon and
generalization of results

• These methods are particularly useful with sensitive
and hidden populations.

• These methods are particularly open to cross-
disciplinary validation


Thank you!

Contact me at:
antonio [dot] casilli [at] ehess [dot] fr
Find this presentation on my research blog:
http://www.bodyspacesociety.eu
Follow me on Twitter:
http://www.twitter.com/bodyspacesoc


Casilli lseisp110510 light

Recomendados

Recomendados

Más contenido relacionado

La actualidad más candente

La actualidad más candente (12)

Destacado

Destacado (10)

Similar a Casilli lseisp110510 light

Similar a Casilli lseisp110510 light (6)

Más de Bodyspacesociety Blog

Más de Bodyspacesociety Blog (20)

Último

Último (20)

Casilli lseisp110510 light