Be Jar Whistleblower Documents

Internal Instagram and Facebook Documents
These documents were created during my time working at Instagram as a

consultant from 2019 to 2021. I had earlier worked at Facebook as a Director of
Engineering from 2009 to 2015. I was the senior engineering and product leader
for efforts to keep users safe and supported. I returned to Instagram in 2019 to
work exclusively on user experience and well-being.
The documents here in this Google drive are in chronological order. I’ve
also attached two folders. One contains the earlier work Facebook published to
the public on the subject matter of teens. It includes findings from research and
product development from 2011 to 2015. The second folder contains, for your
background, a document from the Facebook Files that discusses a similar
pre-existing survey that is mentioned in some of the other documents.
The following notes serve as a guide or crib sheet to help you make sense of the
documents.
Document labeled ‘1 - “Bad Experiences” Measurement - Plan for a Plan -

Nov 19 2020 WB review.pdf’
The first document is a set of slides prepared by me and members of the

Instagram Well-being team in November 2020, after I had been at Instagram for
about a year. My team and I had come up with a new framework with which to
measure what we started to call “bad experiences” for users. Our hope was to let
users tell us about those experiences and then develop tools to support them.
Along with product managers, researchers, and others in the company, I
prepared this slide set for the Well-Being leadership team at Instagram. We were
proposing to formally set goals for the reduction of bad experiences, as defined
by the users themselves, as well as measuring the effectiveness of the support
tools we planned to introduce. Did users think the tools were helping them deal
with the issues? Since data drives almost all work at Instagram and Facebook,
we were arguing for the creation of data sets that could be measured and
reported.
Note especially the slide called “Examples of bad experiences often in
policy gaps,” which helps explain why so many bad experiences did not violate
existing company policies.
1
(Some slides appear twice so the full slide can be visible as well as
comments various company employees made on top of them.)
Document labeled ‘2 - AI + FAI Workshop_ bad experiences.pdf’
By early 2021 the idea of working on “bad experiences” had begun

spreading to other parts of the company. This is a presentation prepared by
Facebook Research, pertaining to its own work on related issues. I had nothing
to do with preparing this document.
Note in particular, though, the slide labeled “bad experiences are common
and frequent,” which indicates that two out of five users on Facebook had an
experience in any given week that they considered “bad.” Another important slide
is the one labeled “Users in this study rated borderline content as harmful as
violating content.” Note that this data set was compiled in 2018. This shows that
the company had known for a long time that what users were experiencing as
harmful did not match the company’s definition of content that violated policy.
I don’t know if this work at Facebook continued. However, it is my
understanding that approximately half of the Facebook Research Team that did
this kind of work was eliminated in 2023 as part of the “year of efficiency.”
Notes on the following three documents, all with the term BEEF in their title
While our recommendations for regular measurement, reporting, and goal

setting were not all adopted following our initial presentation at Instagram, we did
get some traction with the “bad experiences” model. We were given more
resources and created BEEF—the Bad Experiences and Encounters Framework.
The first two documents illustrate the thought process behind BEEF, which
was intended to help employees better understand the bad experiences people
were having inside Instagram. In the surveys conducted as part of this work, we
began asking people about unwanted sexual advances.
These documents explain the research plan—how many people got the
surveys, the methodologies, etc. They demonstrate that the company was
running a very thorough and methodical survey program around this work.
The thumbnail slide presentation is the only form I have that in—this was
the full presentation of the results of the BEEF work. Apologies that it is partly
illegible.
2
Document labeled ‘7b - BEEF by Age (attachment to Gap in understanding

e-mail).png’
This table is the one piece of detail I have from the full BEEF presentation. It
comes from slide 19, entitled “Issues by all age groups.” I attached this chart in
the email I sent to Mark Zuckerberg, which is included in this drive.
Note: the data for all users is an accurate representation of what people
reported experiencing. The data in the columns listed by age groups, however, is
unadjusted. That’s important to recognize. The survey was conducted in two
parts, and only those users who indicated they had had some sort of bad
experience in the past seven days were asked to continue, including by
indicating their age. So these age-based tables are the percentage of all users
that age who had one or more bad experiences, who had this exact bad
experience. In other words, the percentages in the age-based tables do not
represent the total number of people that age who had that experience in the
past week.
For the correct adjusted numbers for some extremely important categories
affecting teens, see the final document—my email to Adam Mosseri. The
numbers in that email do reflect the total percentage of users that age who had
that bad experience.
Email to Mark Zuckerberg and M-team ‘7a - Gap in our understanding of

harm and bad experiences.pdf’
I sent this to express my concern about what we had been learning as well as to
make suggestions for steps the company could take to respond on behalf of
users. Before I sent it, I vetted it carefully with multiple people inside the
company, including some who were quite senior. In other words, I followed the
normal procedure for flagging issues to executive leadership in a technology
company, consistent with communications I participated in during my earlier stint
at Facebook, 2009-2015.
Regarding the statistics in the email to Zuckerberg for bullying,
experienced negative comparisons, and sexual harassment: these represent the
unadjusted figures (see note above), which had been given to me at the time by
company researchers. The adjusted figures for these categories, which I
obtained later, are all in the next document, the email I sent to Adam Mosseri.
3
Email to Adam Mosseri ‘8 - WSJ published Mosseri Pre-Read

ffpreread110223.pdf’
I later had a meeting with Adam to discuss these findings and recommendations.
This is the email I sent him in advance of our meeting, to “pre-brief” him. Note
that this email was published by the Wall Street Journal on November 2. The
statistics in this email are all the adjusted numbers.
Folder ‘2011-2015 Published work by Facebook on Teen Bullying’
This folder has presentations that were made available to the public by Facebook
between 2011 and 2015. The data was the result of close collaboration with Marc
Brackett and Robin Stern from the Yale Center for Emotional Intelligence. Of
special note is slide 3, titled ‘Why are we here’ in ‘CRD2_Yale
Team_Compassion Day 2 Presentation_FINAL copy.pptx’.
The document ‘CRD3_Yale_Team_Compassion Research Day

3__1_23_2013_FINAL.pptx’ shows the results of a product development process
for helping teens with the issues they experience on social media.
Folder ‘From Facebook Files’
For background, TRIPS was a pre-existing survey in which users told Facebook
about their bad experiences. The historical numbers are within range of the
BEEF findings.
4
From: Arturo
Subject: Gap in our understanding of harm and bad experiences
Date: October 5, 2021 at 9:37:59 PM PDT
To: Mark Zuckerberg
Cc: Sheryl Sandberg , Chris Cox , Adam Mosseri
, Mark Zuckerberg
Dear Mark,
I saw the note you shared today after the testimony, and I wanted to bring to your attention what
I believe is a critical gap in how we as a company approach harm, and how the people we serve
experience it. I've raised this to Chris, Sheryl, and Adam in the last couple of weeks.
I want to start by saying that my personal experience, and what I believe, is that you and m-team
care deeply about everyone we serve, and my goal in sending this is to be of service to that. It's
been 2 years since I've been back part-time.
51% of Instagram users say 'yes' to having had a bad or harmful experience in the last 7 days.
Out of those 1% of report and of those 2% have the content taken down (i.e. 0.02%). The
numbers are probably similar on Facebook.
Two weeks ago my daughter , 16, and an experimenting creator on Instagram, made a
post about cars, and someone commented 'Get back to the kitchen.' It was deeply upsetting to
her. At the same time the comment is far from being policy violating, and our tools of blocking or
deleting mean that this person will go to other profiles and continue to spread misogyny. I don't
think policy/reporting or having more content review are the solutions.
There is detailed data about what people experience in TRIPS, a statistically significant survey.
We ran a more detailed survey, I've attached the full age breakdown below, but here are some
key numbers (these questions are in the last 7 days):
21.8% of 13-15 year olds said they were the target of bullying.
39.4% of 13-15 year olds said they experienced negative comparison.
24.4% of 13-15 year old responded said they received unwanted advances.
Why does someone think it is ok to post 'get back to the kitchen' or harass someone? I believe it is
because it doesn't violate policy, and other than deleting or blocking, there is no feature that helps
people know that kind of behavior is not ok. Another example, is unsolicited penis pictures.
has received those from boys too since the age of 14, and her tool is to block them. I
asked her why boys keep doing that? She said if the only thing that happens is they get blocked,
why wouldn't they?
Why the gap between Prevalence and TRIPS? Today we don't don't know what % of content
people experience as misinformation, harassment, or racism is policy violating. We have done
great work in driving down prevalence, and there will always be more to do, but what if policy
based solutions only cover a single digit percentage of what is harming people?
1 of 2
Policy is necessary when the content is unambiguously inappropriate, yet it has many limitations.
It trails behavior, the interventions are heavy and risk over-enforcement and getting the border line
right is extraordinarily difficult. Policy enforcement is analogous to the police, it is necessary to
prevent crime, but it is not what makes a space feel safe.
What makes a workplace, or a school, or a dinner table feel safe is social norms.
If someone goes around telling women to 'get back to the kitchen', and the only thing that happens
is their content is deleted or they get blocked, don't we run the risk of normalizing bad behavior?
How are people to learn to be members of a safe and supportive community without visible
interventions that help set the social norms for the environment? I believe social norms also
protect speech.
At dinner tonight said: my car videos are getting 100,000 views, it's natural that I'm going
to get a lot of hate with that. Is it? Why is it acceptable for someone to harass someone on their
surface? The most powerful solution for the integrity and safety space is to affect the supply of bad
experiences via the actors creating them.
I might be wrong about my assessment, and welcome feedback about any effort or data that l'm
missing. I believe that it is important to get the following efforts well-funded and prioritized:
• What is the content that is causing bad experiences for our users? How intense is the
experience?
• What % of that content is policy violating? (i.e. how much of TRIPS is driven by content other
than what drives Prevalence?)
• What are visible product solutions that make the community better over time? e.g. actor
feedback, comment covers, pinned comments, etc.
The solutions we create I believe should have the following properties:
• The person who has the negative experience should feel heard, you don't 'perceive' racism or
harassment, you experience it, and you are the source of truth for that. The feedback flow
should not be just about filing a report, but about understanding the experience the person is
having so we can give them the right solution.
• We should empower creators, communities, and Instagram, in setting the social norms for the
spaces the are a part of.
• Where appropriate we should give feedback to actors, in the belief that they are acting with
good intention and might have caused unintentional harm. There can be a range of
interventions that start with 'nudges' that assume positive intention. This will allow us to
separate the people who would behave differently given feedback, from the ones who are
intentionally causing harm. We can then approach people who are intentionally malicious with
the integrity tools.
If you would like I can give more details or specifics on this. I am appealing to you because I
believe that working this way will require a culture shift. I know that everyone in m-team team
deeply cares about the people we serve, and the communities we are trying to nurture, and I
believe that this work will be of service to that.
Arturo
2 of 2
TRIPS (TrackingReachof lnteg
Survey)
What is TRIPS?
The "Tracking Reach of Integrity Problems Survey" (TRIPS), also known as the Percept
measures global perceptions of Integrity problems on Facebook, lnstagram, and Me
integrity problems relevant to Central Integrity, FB App Integrity, lnstagram Well-Being,
compare user perceptions across problems, estimating:
1. Perceived reach of a given integrity problem

2. Reputational reach (i.e., for those who have heard of others experiencingthis
3. Perceived intensity of experiencing this problem
How can I stay updated about TRIPS?

• Please join the TRIPS FYI and Updates group
• H2 2020 Roadmap
• Want to see the data? Please see Show me the data!
Why should we care about ·perception?

Sometimes we define Integrity problemsdifferently than
definitions to measure prevalence in VPVs. TRIPS, on t
nitions. This prevalence-reach distinction is i
users; for example, what is nud'
• eventsthat use
To measure intensity, we must rely on user perception. Furthermore, the
perceived intensity without relying on logged behaviors, which are noisy and
context. By gathering user perceptions of intensity, we can better establish persoma
thresholds for what constitutes a bad experience; these data also inform,parallel
Severity Framework.
Perception allows us to verify whether product interventions actually work. Integrity

and minimize harm on our platforms; accordingly, our product teams frequently launch ln"i
whether our solutions are indeed reducing the spread of harm on Facebook and other sw
however, is ensuring that our users actually perceive a reduction in harm in their day-t
products. TRIPSenables product teams to validate experiments and in-product solutions
real-time, ensuring that our Integrity efforts yield noticeably better experiences for the
What is TRIPS' mission?
Our mission: Enable the systematic, rigorous, and reliable measu.rementof people's
FACEBOOK,tracking integrity problems across the family of apps.
TRIPS was founded as a way to listen to users' voices in real time in the immediate
platforms. The data we collect capture the extent to which group-level prefe~
country or group, and differ from our policy classifications of harm. We aim to •
proactively detect sudden changes in perceived reach or intensity, guiding
harm on a global scale.
We combine rigorous methods with subject matter expertise. The li

social and cognitive psychologists, and quantitative specialists with d
rigorously develop and test questions to ensure ease of comp
comparability across problem types.
Broadly, we serve as a mouthpiece for the people who h

voices by surfacing patterns, inconsistencies, and u
product teams to guide priortr12ation,Jrafema
·-o •• , ,... ----- ---···- ·- ,:;,---- ,... ·-•·•·--··-
What problemsdoes TRIPScover?
Problemaru FB 10 MSGR
Nudity ,,,, ,,,, 2020
Fake accounts ,,,, ,,,, 2020 Impersonation
False / misleading content ,,,, ,,,, 2020
Bullying & harassment (target) ,,,, ,,,, 2020
Hate speech & discrimination ,,,, ,,,, 2020
(target)
SRG: Animal sales ,,,, ,,,, 2020

Clickbait ,,,, N/A 2020 Profanity
Off-site landing pages: Too ,,,, N/A 2020
many ads
Off-site landing pages: Low ,,,, N/A 2020

quality
Civic inflammatory ,,,, ,,,, 2020

Civic false/ misleading ,,,, ,,,, 2020
Political Affective Polarization ,,,, ,,,, 2020
What is the future of TRIPS?

WhatdoesmaturitylookIlka?AsTRIPSexpandsto newlwfaaa
aim to inform org-widestrategicdiscussions
bued on1M
operational, fully automatedmeasurementsystemthat:
1 -bles producttaamsto l1IOfllproactlvelv

2 drivesinmghtsacross problarna,~
S. lnllOences campanyS1r8l9fW
wtlln
C Ii internalfb.com ntern wk CI_Research/TRIPS_(Tracking_ Reach_of_Integrity _Problems._Survey)/
~ Se ~ F~ i-', Trncki Flea of lrlegnty Pro lems Surveyl

*
+NewPage My User Page My Activity
!'.El Wik1 ) Central Integrity Research TRIPS(Tracking Reachof Integrity ProblemsSurvey)

~ EditPage
un-stte Ianamg pages: Low
quality
.,
N/A 2020 Civic speech by fake account ., ., 2020
) Sur,,ey Methodology
Civic inflammatory ., ., 2020 Civic bullying ., .,
) Show Me the Data'
Civic false/ misleading ., ., 2020 Civic online discouragement ., .,
2020
) Exper ment1ng with TRIPS

Political Affective Polarization ., ., 2020 Civic demobilization ., .,
2020
2020
So You Want to Goal on TRIPS
What is the future of TRIPS?
Frequently Asked Questions
What does maturity look like? As TRIPS expands to new surfaces (e.g., WhatsApp, Messenger) and problemtypes w
C- TR PS {Comparat1\le TRIPS)
aim to inform org-wide strategic discussions based on the insights we gather. For this team, matunty entaHsan
Con•act the TRIPS Team operational, fully automated measurement system that:
1. enables product teams to more proactively detect and respond to harm,

What does Onboard1ng to TRIPS Lo
2. drives insights across problems, surfaces, and markets to inform Integrity priorities, and
3. influences company strategy when determining when, where, and how to minimize risk on our platforms
What are our goals in the near future? In the next two years, we aim to:
1. broaden coverage across more Integrity problems and more surfaces;

2. further expand our understanding of the difference between Fact and Perception Frameworks (i e , the Pen:8pllan
Framework/ Fact Framework Understand Taskforce);
3. establish a way to attribute user responses to spec1f1cexperiences (e.g., entities, p,eces of content), and
4. develop an experimentation framework to enable product teams to test the effect of product expet._,.. CIII_.
perception.
Along the way, we will continue to drive more rigorous, reliable, and representative insights to ourpartner teana ID ...
progress, highlight new concerns, and drive down bad experiences for the 2.8 billion people - -
• Central ln<ege!y R_,a,c;h
ontens11y PerceplionFramework Community

1ntegr1ty PAC TRIPfl r-=tl ~ ,.......
Show Me the Data·!
Where to find the data
There are two dashboards through which to explore TRIPS data: TRIPSteam da$11i
HOW TO INTERPRET THE VARIABLES
• Perceivedreach: % of people who have seen/experienced the integrity problem

• Reputational reach: % of people who heard about the integrity problem duriAt ffi
measure of reputational hearsay - something heard from friends or in the news.fl
someone saw personally on our platform.
• Intensity: % of people who said seeing/experiencing the integrity pro~lem was 11
intensity question is only administered to people who said yes to the perceived r.ea
question is not administered as a follow-up to the reputational reach question.
Note: We only report reach estimates for "Yes, during the last ! days" on the TRIPS
below, you may notice users can also answer "Yes, but more than 7 days ago"to the
response option to reduce over-reporting: if we asked users only about the last 7
about bad experiences that happened more than 7 days ago, and thus add emtJr-
You can find the full verbatim wording of all questions here.
PataalNd l'NCh
~--.........
_,__...,..__
-----·-
1. TRIPS Team Dashboard

lRIPS SWAMP Dashboard
PS D;rta Sour9es.
,.,_,,,O Perceived R-.ch 0
Reput•tJon.J R.ach 0
lnten11ty0
■ T•t,,e .•
16.55 Data V1S\Jabze
~
20
24.50
~ ..,.
43.15 50
~
♦-024% J Query 1 •NewQ ery
25
.. 5()% W/w 0 00% did
·155% did
15 ·1 SO-,,. W/w Dalq~
Nov '20 Dec '20 .4 41• ovm
·581% W/w ◄O II
Jan '21 Feb '21
Nov '20 Dec '20 ovm
Jan '21 Feb '21
Nov '20 Dec '20 Jan "2 Feb • I.
Data Transforms O
8.71 21.42 Range Filter, O
~ ~
49.54 + Add
~
♦-011% did
8
--0.28% did 20 Category Filters 0
♦-0 68% Wlw -tl.0~ did
-+-094% wJw
40 • Add
+223% m/m -+1.37% W/w
Jan '20 Apr '<O Jul '20 -009% mlm
ColumnsO
Oct '20 Jan '21 -+123% m/m
Jan '20 May '20 Sep '20 Jan '21
Jan '20 May "20 Sep '20 Jan ·21
GroupBy 0
20.00
~
21.42
~
20
57.83
~
60
-258% did PivotsO
--0.28% did 20
·3 65% wlw -+-057% did
-+-0.94% wlw 50
,-050% mlm 15 •1.03% wlw
--0.09% m/m
Jan '20 May '20 Sep '20 Jan '21 -3.15% m/m MetncsO
Jan '20 May'20 Sep '20 Jan '21 Jan '20 May'ZO Sep '20 Jan 2!
7.68 22.22 25
~ ~
44.93
116%
179% "
.,..
75
--0.27% did
20 -tl 63% did 40 ~
--076% wlw
-318% nvm
·279% .,.. Column Filters 0
May '20 Sep '20 Jan '21 -598% mlm
May'20 Sep '20 Jan '21 May'20 Sep '20 Jan ·21
Display Top N 0
+' Add
File Edit View History Bookmarks People Tab Window Help
Trendsin perceivedreach Reputational reach and intensity for each Integrity Problem
♦ Problem 0 Percefved AHch C)

Aeputallonar Reach O
lntensltyO ■,-
2S
21.48 o...
~
Vttuafia
....."
-
22.22 2S
+O 09% did
20
~ 55.99
~,,,.._,-r-M
-
+051% wlw -027% did
20 ·33CN
-2.<l1% m'm
May '20 Sep '20 Jan '21
-0 76% wlw
-3 18% m/m
May '20 Sep '20 Jan ·21
01
-t7<t% mlm
.. 1
•
38.49 •
~
Civic 40
Consumption +1 08'lr.
-2 06,
"
wlw No data
No.1.ata
c..ego,y,.....
•2~ nvm
May '20 Sep '20 Jan '21
13.31 14.48
~
46.96
~
Civic 15
-1.<l1'% Old
Demobilization +0.56'!1. did 15
-·
-3 20% Wfw ♦1 5,4-..
-136% wlw
-452'o ~20'II..,,,,
Nov'20 ·3 79% rTVm
Dec '20 Jan '21 Feb '21 +<t75-i. 111"'1
Nov'20 Dec'20 W,'21 Feb 21
13.58 15
16.04 20
000%
~ +165%
~
49.43
..,.. ~'N~
...... -
15
•12 w""
+1 3'%
May '20 Sep '20 Jan '21 May'20 Sep 20 j,n
~ cp,..
eop Tab Window Help
Perc•ived Reach 0
Reputationat Reach 0 • hble
lntenaityO
0,1a V1Sllaliz@
~
22.13
~
21.89 47.72
Query1 t New Query
~
25
25 50
-045% dlo
-135% did Oalq~
+3 90% wtw 20 -185% did
+-41◄% wlw 20 ◄O
-0 85% mtm -425% wtw
-1 00% m/m
May'20 Sep '20 Jan '21 -0.31% m/m
May '20 Sep '20 Jan '21 Data Transforms O
May'20 Sep '20 Jan '21
Range Filters O
+ Add
12.12
~
9.10
~
15
56.07 Category Fitters O
~
-0.16% did 10 + Add
Inflammatory -119% did 60
+289% w!w -216% d/d
-3 58% m/m 10
+111% w!w
7.5 ·• 70'
.,. ColumnsO
·900' mlm
May'20 Sep '20 Jan '21 •974 mlm
May'20 Sep '20 Jan '21 May'20 Sep '20 Jan '21
Group By 0
21.31
~
25
20.70
~
40.24
~
PivotsO
-0 75% did +-083"-d/d

Discouragement 20 -164% did
20 40
-2.83% mlm
,..
-0.29% W/W
mm ~
-3.64% wtw
-0.07% mlm
Metrics 0
May'20 Sep '20 Jan '21 May'20 Sep ·20 Jan ·z~
25.45
~
23.77 50
~
46.84
+-051% did
-035% wlw
25
+-0.93%d/d
•182% wtw 22.5
25
-1 64% did
40
~ ColUmn Frlters 0
+4 sa, wtw
+1 60% mlm -206% m/m
J•n '20 May'20 Sep '20 Jan '21 +252% mlm Orsplay Top N0
Jan '20 May'20 Sep '20 Jan '2 l Jan ·20 M,11y'20 Sep 20 Jan ,U
+ Add
,.,_,,,o Pe1>c.lvedRuch ft
RepuUltlonal Reach O ■ Table
Intensity C)
Data Vrsuallz@
16.55 24.50
~
20
~
43.15 so
~
Queryl + New Query
+02"'"'· 2S
0.00% did
.4 50"t,, wlw -155% did
-1 ~ wlw Delquitry
1S -581% w/w 40
Nov'20 Dec '20 Jan '21 Feb '21
Nov '20 Dec '20 Jan '21 Feb '21 Nov '20 Dec '20 Jan '2 Feb 21
Data Transforms 0
Range Fitters 0
8.71
~
21.42
~
49.54 Add
+o 11% :lid
+o 69'% wlw
8
-0.28% did 20
+I 02% d/d
~ CalegOfy Fitters
. Add
0
+o.94% wlw 40
+2 23% rnlm +1.37% w!w
-009% rnlm Columns O
Jan '20 N,r '20 Jul '20 Oct '20 Jan '21 +123% m1m
Jan '20 May'20 Sep '20 Jan '21 Jan '20 May'20 Sep '20 Jan '21
GroupBy 0
20.00
~ ~
21.42 57.83
~
20 60
·268% did
-385'"4 w/w
+0.50% m/m IS
-028% d/d
+094% wlw
20
+0.57% did
-103% wlw
so ...
Pivots
Metrics
0
0
-009% m/m -315% m/m
Jan '20 May'20 Sep '20 Jan '21 Jan '20 May'20 Sep '20 Jan '21 Jan '20 May'20 Sep '20 Jan '2 ~
7.68 2S
~
22.22 44.93
116%
I""-
""
wcw
7S
~ -027% did
-0.76% wlw
20 +163% d/cJ
•279% W/w
40 ~ Column FIiters 0
-318% m/m -5.98% rMn
May '20 Sep '20 Jan '21 May'20 Sep '20 Jan '21 May'20 Sep '20 Jan'21
Display Top N 0
+ Add
Chrome ~ e Edrt View History Bookmarks People Tab Window Help
• Problem 0 Perceived Re.ch 0 Repulltlonal Reach O lntensltyO

·--
....
Anger Towards
Political Out-
16.55
+0.24% did
-450%
8.71
WIW
20
IS
~
Nov'20 Dec '20 Jan'21 Feb '21
24.50
000% s
• series_l
-'-'---
1V,J.1.3. 202
22.64
43.15
-1.55% did
-5.81% W/w
50
40
~
Nov ·20 ~ 20 Jan '2 ,,. I
-
OalaT-·
Range-·
~
21.42 49.54 Add
~
-·
8 c...tor,-·
+0.11% did 20
-028% did
+0.69"4 WIW
+102"' did Add
+o.94% w/w +1.37% WIW
40
+223'11. mtm
-009% mlm +l.23'11. mlm
Jan '20 Apr ·20 Jul '20 Oct '20 Jan '21 Jan '20 May'20 Sep '20 Jan '21 Jan '20 May '20 Seo 10 .tn 21
20.00
-2.58%
-3.85%
did
WIW
+-0.50% m/m
20
1S
~
J.tn '20 May'20 Sep'20 J.tn'21
21.42
-0.28%
-009%
did
+0.94% w/w
mlm
20
Jan '20
~
May'20 Sep'20 Jan '21
57.83
+057%
-I 03%
-315%
d/CI
W/"w
rnlm
60~
50
J.in '20 May '20 Seg 20 Ja 21

-·
-·
7.68 22.22 2S
~ ~
44.93
-1.16%
.,,...
~, 7S
-0.27% did
20 +163'- did
40~
Cobnn-·
0.
"" -076%
-318%
w/w
mtm
-279% w'w
May'20 Sep '20 Jan'21 -5-~ mm
May'20 Sep '20 Jan '21 May "20 Sep l µ,, 21
--
•
-
16.55 24.50 43.15
~ •.
...,
8.71 21.42 49.54

~,Vv~
.......
...... ...., ... •ll7'
·•-
__ r~
20.00 21.42 57.83
.. - ....-
......
,
,.... .
...,..,.
7.68 22.22
V\J.. 44.93 .~
.....
-
• , 81
• '"".,.
..., ...., NO
--·
11.41
--·
22.22 .,-1-~~
55.911
~'(V
..-- - .. .. ·-
- ,,_
'W~i
- 4U6
14.48
-_-
.....
--
......
4U3
16.04
-
... ~
TrendsIn pen:elved reachReputationalreach and intensity for each Integrity Problem
·-· ..,_AoKhO
21.48
25
Reputational Reach O
25
~
22.22
~
55.99 60
+0.09% did -0.27% did

20 20 -3.30% did 50
+0.51% w/w -0.76% wlw ·6.01"o w/w
-2.41% mlm -3.18' m/m
May '20 Sep '20 ·1.74% m/m
Jan '21 May '20 Sep '20 Jan '21 May '20 Sep '20 Jan 21
~
38.49
40
+1.09% did No data No-
4.08'11, w/W
·l!.M% - Mav ·20 Sep '20 Jan '21
13.31 14.48 46.96
........... ~
-1.41"' did
-a..-. w/w
IS
+O.sn. did
•1.38% wlw
-3.79% m/m
15
~ +1.54% did
+e-20% wlw
+4.7ft mhn
Nov '20 Dec '20 Jan '21 Feb '21 Nov '20 Dec '20 Jan '21 Feb '21
20
15
16.04 49.43
~ ~
-
12.5 +1.116% did -1.57% cW
15
+12 25% wlw
+a.ao,r,
May '20 Sep '20 Jan '21 May '20 Sep '20 Jan '21
-·--·-
•-o P9fc.t¥ecl Reach 0
AepvtaUonal Reach C,
fnlen1ityC,
-...__
25
21.48
~ ..,,.,
22.22 25
~
55.99
~
60 au.i,,
+0()9"11. wi
+051"11, w'W
•2411, mlm
20
-07&"11,.,,,.
20 -3.30% did
•01 ... 10
--.
M,1y'ZO St'p '20 J•n 21
-3 18'1. IMn
-t.7-1% mtrn
May 10 Sep 20 ,. I
_,
~
38.49
+1.01'11,,
•2.06"11, w,W
·2M"lii. m1m
(I (1
,o
M,11y20 ~l'I) ·20 J•n 21

...... Noclata.
-
c_,.-·
----l.
13.31 46.96 so
........... _
•141'11i, c1#d
......
IS
__ ~
..
.........
+t.14'!1, did •o
Nov 20 DK 20 J.1r. 11 r,b '21
-3.79%f'I\.IIT'J
Nov '20 Ott ·20 Jan "21 Feb 21 N011'l0 De< l ,..
13.58 IS
16.04 20
49.43
...... ~
60~
. ..... ~
""'
,.,.
+1.3".mlm
12'
May'20 Sf,p'20 J.a.n'21

+1.8&%
......
+1225"4 W/w
did
-
15
M;r,y'20 ~p'20 Jan ·21 • -

·157"11,(Ud
"'
- ...
•-o Perceived Rnch
21.48
+0.09% did
+0.51% wlw
O
25
20 ~
Aeputatlonel Ruch
22.22
-0.27% did
0
25
20
lntenelty
55.99
0
-3.30'% did
60~
50
·-
Query1 ♦ -Query
-0.76% wlw
•2.41% m/m -601 W/w
-3.18% m/m
May'20 Sep '20 Jan '21 -1.74% m/m
May '20 Sep '20 Jan '21
May '20 Stp '20 Jan '21
DolllT-·
Rongo-·
~
38.49 Add
40
♦ 1.08% d/d Caoga,y-·
Conaumptlon Nodala Noda!a
•2.06% WIW
-2.88% mlm
13.31 14.48
........... ~
M(l~d.ly_ l>l<IY
15 • series_! 49.79 '
., ... , .. did 15
..0.58'W.dld • senes_I_d (r,1119e)46.7910 Sl.79
-a....- •1.311"1f.W/w
... ,.,. mlm
.....,..,.
Nov '20 Dec '20 Jan '21 Feb '21 Nov '20 Dec '20 Jan'21 feb'21 ♦◄-- - Nov ·20
13.58 15
16.04 20
~
49.43
..,
~
-
o..... 12.5
-
+185% did -t..57"4 cw
+7.88"ll. w/w 15
+1.34% IM1\
M;ay'20 Sep '20 J,1n'21
........
+1225'%. 'tllw
May '20 Sep '20 ~" '21

..
7 40
....,
Trends In perceivedruc:h Reputatlonalreachand intensity for eachIntegrity Problem
•-o .........,....,.0 X .,.,~

lnten•ttyO
25
21.48
~
22.22 25
~
55.99
~
Civic B & H 60
+0.09%. did
Wltneaa 20 -0.27% did
+0.51% w/w 20 -3.30% did 50
-0.76% W/w
-2.◄ 1% m/m -8 01~ wtw
May '20
-3.18% m/m
Sep '20 Jan '21 -1.7◄% m/m
May '20 Sep '20 Jan '21 May '20 Sep'20 Jan'21
~
38.49
◄O
+1.08% did
No data
-2.06% WIW No data
·2.118% mlm
May '20 Sep '20 Jan '21
13.31 14.48 46.96

~
50
15
-1.41% did
+0.56% did +1.54% did ◄O
15~
-3.20% Wfw -1.36% w/w +8.20% WIW
-4.52% nvm -3.79% m/m
Nov '20 Dec '20 Jan '21 Feb '21 +4.75% mlm
Nov '20 Dec '20 Jan '21 Feb '21
13.58 15
16.04 20
49.43
~
o.on. .., 12.5
◄O
May '20 Sep '20 Jan '21

TIWIClaIn perceived reach Reputational reach and Intensity for each Integrity Problem
·-· --"· Reputatlonal Reech 0 lntan■lty0
~
22.13
~
25
21.89 47.72 50
Civic Faleeand 25
.0.45% did
MlalNcllng •1.35% did
·l.85% did
+3.90% wlw 20 20
+4.14% wlw -4.25% wtw
40
--0.85% m/m
•1.ClO"!i(. mlm
May '20 Sep ·20 Jan '21 -0.31% m/m
May '20 Sep '20 Jan '21 M,1y'20 Sep '20 Jan 21
12.12
~
9.10
~
15
10
56.07
.Q.1n. did
-1.19'4 did 60
•2.18% cUd
+2.81% w/w +1.11% w/w
10 .fl~ WIW
·U""''""" -90()<' .. m/m 7.5
May '20 Sep '20 Jan '21
·9 74"4 mhn
May '20 Sep '20 Jan '21
21.31
~
20.70
~
25
40.24
..,
-0.75'11, did
w,.
•2...,,.., mhn
20
+0.83% did
-0.29% w/w
20 ., 84% did
-3.&4% WIW
40
.4 78~ m/m .Q.07"' m/m

May '20 Sep '20 Jan '21 May'20 Sep '20 Ja.n'21
25.45 23.n
~ ~ .....,._
25
46.84
·--
25
...... did
Jan '20 May '20 Sep '20 Jan '21

•1.1112%WIW
•2.CJII%fMQ
22.\
......
Jan '20 May '20 Sep '20 Jan '21 Jan 20
so
,., A ,II 15.92 I .. J 46.25
·-· Percetwct RNCh C)
22.13
-0.,46% did
25
~
Mr,;.,
"''"
~c;>n•LB••
Mu'
,., ....
202!)
serlH_l_ci (range): 27.64 10 29.t ◄

lntensltyC,
47.72 50
~
•1.35% did
+3.80% wtw ,o +4.14% W/w ZO
-1.85% did
◄.25% W/w
40
-0.116% mlm
·1.00% m/rn
May '20 Sep '20 Jan '21 -0.31% m/m
May'20 Sep'20 Jan '21 May'20 Sep '20 Jan'Zl
12.12
~
9.10
~
15
10
56.07
-0.11% did 60
·1.Ht% did -2.18% did
+2.89'% w/w
+1.11% W/w -87°' wtw
-3.58% min,
10 7.5
-9-~ rrvm -9.74, m1m
May '20 Sep '20 Jan '21 May '20 Sep'20 Jan ·21
21.31
~
20.70
~
25
40.24
-0.75% did +0.83% did
20 •1,84% did ◄O
-4 10-,. wfw 20 -0.29% wfw -3.64% w/w
-2.83% mlm -478% m..'m
May '20 Sep '20 -0.07% mlrn
Jan '21 May'20 Sep '20 Jan '21
25.45 2a.n
............ ~ ~
25
46.84
25
..0.51% did +0.93% did -1.64"4 CW
22.5 ◄O
+1.~ mhn
Jan'20 May'20 Sep '20 J;m'21
·1.112% W/w
-2.08% m/m
Jan '20 May'20 Sep '20 Jan '21
.....,._
....... wlw
Jan '20
11.37 ,., A ,r, 15.92 I ,. J 46.25

50
Trends in perceived reach Reputational reach and intensity for each Integrity Problem
·-· p.,~RNChO
+3.90% wlw
-0.85% mlm
20 w
May '20 Sep '20
Y.,,V
Jan '21
Reputation.I Reach O
+414% wtw
·1.00% m/m
20 V"'
Ncll::ofllmnto
~
V'I/ V
lnten■fty0
-4.25% W/w
--0.31% m/m
40
-
S...turday. J~n 6 2020
• series_!. 11.55
12.12
~
15
56.07
...,.,,._
-0.11% did 10
·1.19% did 60
·--
+2..,., Wlw -2.19% did
10 +1.11% w/w
-3.58% rnlm 7.5
May '20 Sep '20 Jan '21 -9 74
21.31
~
25
20.70
~
40.24
-0.75% did
Dlocouregement +0.83% did
-41()'",, w/w 20 20 •1.64% did 40
-0.29% W/w
·2.83% m/m -3.&1% W/w
-4 78~ m/m
May '20 Sep '20 Jan '21 -0.07% m/m
May '20 Sep '20 Jan '21
25.45 2a.n
..0.51% did
-0.36% WNI
25
~ .0.93% did
·US2% w/w
25
22.5
~
46.84
•1.84% did
40
+1.80% rn1m
Jan '20 May '20 Sep '20 Jan '21
-2.08% mlm
Jan '20 May '20 Sep '20 Jan '21
.....,._
+45h. wlw
Jan '20
11.37 15.92
~ ~
15
46.25
.o.- did
•1,22% w/w
10
--0.13% did
•3.34% w/w
.,..... did
12.5 +2..14% wlw
TrendsIn perceivedreachReputatlonal
reachandIntensity for each Integrity Problem
onro,:iot~
•--o PwetttV9c:I
Ruch 0
lntenelty 0
+3.90% wtw 20 w Y.,-V
+4.1"4% w/w 20 V"' ~
V"/ V
"4.25% Wlw
40
-0.85% m/m
May '20 ·1 00% m/m
Sep '20 Jan 'll -0.31% mlm
May '20 Sep '20 Jan '21 May 20 Sep '20 Jon
9.10
~
56.07
~
-0.18% did 10
·1.19""' did 60
.
+2.88'JI. w/w -2.18% d/d
+1.11% wlw
.a.58,li, nwn 10 7.5
wlW
May '20 Sep '20 Jan '21
900 mlm
May '20 Sep '20 Jan '21 ' M•y '20 Sep 20 Jon I
21.31
~
25
20.70
~
ClvlcOnllne
40.24
-0.75"4 did +0.83% did
DIIICOun,gement 20 20 -1.84% did 40
•◄ 100,.. w'W -0.211% WIW -3.84% Wlw
·2.83% m/m •4 78° m/m
May '20 -0.07% mlm
Sep '20 Jan '21 May '20 Sep '20 Jan '21
25.45 2a.n
~ ~ ....,._
25
46.84
25
..0.&1.. did +0.93% did ., ....... d,ld
-0.36'1, wlw
+1.flO-. mhn
Jan '20 May '20 Sep '20 Jan '21
-1.82% wlw
-2.0S,C. mlm
22.S
...........
Jan '20 May '20 Sep '20 Jan '21
.......... ~
11.37 15.92 46.25
•1.ll'JI. w/w
10 -0.13% did
-3.34% wlW 12.S
"
~ .,.......,
...,... -
l~Hil El l!Hil El GIEl El II ;;'
Trends in perceived reach Reputational reach and intensity for each Integrity Problem
·-· Percei-...d Ruch O
25.65
-086% did
•2.25% w/w
30~
20
Aeputational Reach O
24.08
+o.04% did
•E,I' Wlw
30~
20
lntenaltyC,
44.25
-0.74% did
50
45~
-
·-
Quory1
VIMlllt
♦ -Quory
·1.SN nvm •2.62% wlw

·2.86% m1m 40
Jan '20 May '20 Sep '20 Jan '21 -0.90'- mlm
Jan '20 May '20 Sep '20 Jan '21
'20
J.11n Moily'20 $.tp '20 J«n '21
~
11.58
~
IS 12.29 15 56.90
Graphic 60~
+1.14% did
Violence
+10.71% W/W
10
"'·"°"' did
+12.55% wtw
-0.25% did
•3.61% w/w 50
+6.24% mlm 10
+2.93% m/m +2.81% m/m
Jan '20 May '20 Sep '20 Jan '21 Jan '20 May '20 Sep '20 Jan '21 J.tn '20 M.ly '20 Sep '20 Jan 71
6.06
~
17.48
~
6
20 52.95
+3.24% did
+1,(1()% w/w
+o.23% did
-0.83% w/w 15
----
..O.~w/W
.., 40
+2.1n. nvm +0.81% m/m
Jan '20 Apr'20 Jut'20 Oct'20 Jan'21 +2.oe,&, mlm
Jan '20 May '20 Sep '20 Jan '21 Jan '20
~
16.38
~
20
17.48 20 62.32
.., ..,...,..., ............ 60
°"""
.a.ta w1w 15
Jan '20 May '20 Sep '20 Jan '21

-0.83'JI, w/w
+0.11% mlm
15
Jan '20 May '20 Sep '20 Jan '21 -----

.,,_ 50
J«n'20
JS '
Trends in perceivedreach Reputatlonal reach and intensity for each Integrity Problem
Clckontopofuehcotumntoraru- req ryp b lom

1' Problem 0 Perceived Reach O
ReputatlonalReach 0
lntenalty0
25.65 24.08 ~~- 44.25

so
Falaeand
Mlsleadlng
•0.66% did
-2.25% wfw
30~
20
+0.04°k did
-3.68° wlw
,,r. ~·
30 ., .
-0.74% did
-2.62% wlw
45
40
•1.57% mlm -2.86% mlm
Jan '20 May '20 Sep '20 Jan '21 -0.90% mlm
Jan '20 May '20 Sep '20 Jan '21 Jan '20 May '20 Sep 20 Jan 21
~
11.58
~
1S 12.29 15 56.90
Graphic
+1.14% did +0.90% did -0.25% did
Violence
+10.71% wfw +12.55% WIW -3.81% w/w
10
-H>.24% mlm 10
+2.93% mlm +281% m/m
Jan '20. May'20 Sep '20 Jan '21 Jan '20 May'20 Sep '20 Jan'21
6.06 6 17.48 52.95

20~
+3..24% did +0.23% did -2.88% did
+1.00% wlw -0.63% w/w 15 +0.40% w/w
+2.19% mlm +0.81% mlm +2.oe. rnl'm
Jan ,20 Apr '20 Jul '20 Oct '20 Jan '21 Jan '20 May '20 Sep '20 Jan '21
1638
20~
..0.23% did
15 -083% w/w
..0.81% mlm
Jan '20 May '20 Sep '20 Jan '21
,15
Trends In perceived reach Reputatlonal reach and Intensity for each Integrity Problem
·-· P•cetv.d Ruch 0
15
ReputaUonal Ruch 0
.,,,,_
lnten■fty0
~
10.63 11.07
~
12.5
53.57 60
Nudity
~
·266% did 10 ·1.95% did
•2.30% WIW +0.21% did 50
-2.81% wtw 10
+0.47% m/m •2.72% w/W
Jan '20 +5.53% m/m
May '20 Sep '20 Jan '21 +7.55% mlm
Jan '20 May '20 Sep '20 Jan '21 Jan '20 May '20 Sep '20 J.an'21
4.57 13.36
15
~
57.80
~ ~
60
+3"""' did
...
-0.15% did 12.S
·2.28% did
-
+1.11% wlw ·837" +1001% Wlw 40
•2.58% m/m -2.05% mlm
Jan '20 Apr '20 Jul '20 Oct '20 Jan '21 -155%
Jan '20 May '20 Sep '20 Jan '21 Jan '20 May'20 Sep '20 Jan 21
25.53
~
23.60
~ •-- ...
25 37.90 40
-1.50% cW 25 -0.21% did

22.5
+041% w/w +3.St% w!w •1.04% W1W
.... ,.,., mlm +2.82"ll. rn/m 30
Jan '20 Sep ·20 +3.81% m1rn
May '20 Jan '21 Jan '20 May '20 Sep '20 Jan '21 Jan '20
-
3.84 • 4.69 57.10
...
......,. ~ ~
.......
+1.96"4 did -t2.11% did
-2.90% "'"' +115% wJw
+11.30% IM1I +7.82"11, mlm +7.Dl'lomlrn
.)An '20 Apr'20 Jul '20 Oct '20 Jan '21 J.iin '20 May '20 Sep '20 Jan'21
.. •e- ...... ........

TrendsIn perceivedreachReputational reach and intensity for each Integrity Problem
Cl/dc1 I ofuch
•-1emo P•celved Ruch O
Friday, Apr 3, 2020 lntenalty 0
•seriesll429 ~
10.63 15
kAJ. l • serles=l_ci (range) 13 80 to 14 78
53.57 60
-2.66% did
-2.30% w/w
10 ~ I(~ ·195% did
..0.21% did 50~
2 81% w/w 10 -2.72% wlw
+0.47% m/m
+5.53% m/m +755% m/m
Jan '20 May '20 Sep '20 Jan '21
Jan '20 May'20 Sep '20 Jan '21 Jan '20 May '20 Sep '20 Jan '21
15
4.57 13.36 57.80
Impersonation
+3.39% did ~ -0.15% did 12.5
~
-
·2.28% did
+1.11% wlw -8.37% w/w +10.01% wlw 40
•2.56% mlm •2.05% m/m -1.55%
Jan '20 Apr '20 Jul '20 Oct '20 Jan '21 Jan '20 May '20 Sep '20 Jan '21 Jan'20
25.53
~
23.60 25 37.90 40
......
~--
-1.50% did -0.21% did did
25~ 22.5
..0.43% wlw +3.51% wlw •1.04" w1w
+2.92'11,, m/m ..S.81% ml'm 30
Jan '20 May '20 Sep '20 Jan '21 Jan '20 May '20 Sep '20 Jatl '21 Jan'20
3.84 4.69 57.10

~- did
+1.98% did
-2.oo,r. w/w
~ +2.11""dA::t
+1.96% w/w
..0.08% W/W
+1130% mhn +7.82"}1, m/m +7.Gft.. m1m
Jan '20 Apr '20 Jul ·20 • Oct '20 Jan '21 Jan'20 M.ay'20 Sep '20 J.an'21
..... " .... .......

Trends In perceived reach Reputatlonal reach and intensity for each Integrity Problem
•-o Percetv.ct Reach O

.kor,topofuch
Reputatlonal Reach O
+11.:.ilMo mm, lntenalty 0
+r.ts~ rrvm
Jan '20 Apr '20 Jul '20 Oct '20 Jan '21 +l,lr.,.,,. mm,
Jan '20 May '20 Sep '20 Jan '21
Jan '20
3.45
~
~
4.38 47.98 60
4
+1.17% did
·1.79% did
+8.49% wtw +2.65% did
-t-3.06% W/w
+6.48% m/m 3 -7.86% wlw
Jan '20 Apr '20 Jul '20 Oct '20 +5.04% m/m
Jan '21 ◄.78% mlm
Jan '20 May '20 Sep '20 Jan '21 Jan'20
21.20 20.94
~
22.S 22.5 50
46.56
WebAdload +0.14% did
0.00% w,.
+0.52% m/m
20
+0.19% did
-2.20% w!w
·1.78% m/m
20 ~ ~)..47"' did
-0.70% wlw
Jan '20 May '20 Sep '20 Jan '21 +8.04% mfm
Jan '20 May '20 Sep '20 Jan '21
14.74 14.99
·--
-•--..,
15
~ -1..38% did
-519% w/w
15
~
+1.49911ralm 12.5 ~.93% nvm
Jan '20 May '20 Sep '20 Jan '21 Jan ·20 y'20 Sep '20 Jan '21
6.93 9.08 51.46

~ ~
7.5 10
4.113%..,
-----
•2.99% did ·111"-cW
-8.10% wlw -9.47' w!w
+17", mlm +2.83% mlm
7.5 ..,__.,..
Ja,n '20 May'20 Sep '20 Jan '21 Jan '20 May'20 Sep '20 Jan '21
r - - - - I - - - - - , , -~~.
-•.1; ----.,;;-_<= - - -
1
I~ == I "" - ;-
Trends in perceivedreach Reputationalreachand intensityfor each IntegrityProblem
't ProblemO Perceived Reech O Clickonropofeachcolumn,orsnklnt,gntyp bf,m fromhgh ttolow tperc v«Jru r-,,utatloMlfNdl,orilln,ry
Reputatlonal Reach 0
+11.auu;. m,m lntenaltyC,
Jan '20 Apr '20 Jul '20 +I.Ill% mtm
Oct '20 Jan '21 +I.m,% mtm
Jan '20 May'20 Sep '20 Jan '21 Jan '20 May'20 Sep '20 Jin'21
3.45
~
4.38 47.98 60
SRG Drugs 4
+1.17% did
•1.79% did
+8.49% w/w +2.85% did
+3.06% W/w 3
+6.48% m/m -7.85% w/w
Jan '20 Apr '20 Jul '20 +5.04% mlm
Oct '20 Jan '21 -4.78% m/m
Jan '20 May '20 Sep '20 Jan '21 Jan '20 May'20 Sep '20 Jan'21
21.20 20.94
~
22.5 22.5 50
46.56
+o.14% did
0.00% w/w
+-0.52% m/m
20
+o.19% did
-2.20% w/w
20 ~ -0.47% did
-0.70% w/w 40
~
-1.78% mlm +6.04% m/m
Jan '20 May '20 Sep '20 Jan '21 Jan '20 May '20 Sep '20 Jan '21 Jan '20
14.74 14.99
~
41.34
-1.119%did
-1..80% w/w
15
-1.38% did
-5.19% w/w
15
~ ... ..,. did
~wJtJ(
12.5
+1.4S'll, mfm -0.93% m/m ..o.e1,r,mm
Jan '20 May '20 Sep '20 Jan '21 Jan '20 ~ay '20 Sep '20 Jan '21
6.93 9.08 51.46

~
7.5 10
-2.53% did -2.99% did -1.68')(,dfd

-6.10'JI, w/w -9.47% w/w -3.25% w/w
7.5
+1.78"' m/m +2.83% m/m -0.58% mhn
Jari '20 May '20 Sep '20 Jan '21 Jan '20 May'20 Sep '20 Jan '21
Manage • Help •
-...ur
Q.
....,_, -,,pmteon plot (NpUlallon, perception and lnt11n11tyIn the Int 7 day1) A .. ExploN
l7
Glllde
~......,Deel,
2S
Star!
22.5
.......,~ s:
:;:
,... ~
l..
20
End
• --
~
..
1G
~ 17.5
15
CDo,..I 11111
10
7.5
Settings • Manage • Help •
am
......... camparlaonplot (l9plllallon, pen:eptlon- lntanalty In the last 7 daya) A • • • Explore
27.5
-~Guide
•~lnlogrltyDllah ...
25
•lnlagrllr--..-n!Plal .. .
·~ ClaNlllerEell--•I ...
22.5
Start
CivicStart
:c
• lnlogrltycau...lla ii
~ 20
.., End
.~
•
~
l 17.5 RollingAverage
Window
PB
15
7.5' 10 12.5 15
• 17.5
" Heard of Problem (Reputatlonal Reach)
r Towards Pollllcal OUt-group

lm1111rsonatlon
• SRC.Animals
• Ovlc Fakeacct Contact
20 22.5 25
Nullity • Civic Online Discouragement

CiDrugs Civic Damoblllzatlon
B&HTar!lat • WebAdload
and Mlsludlng • Graphic Violence
B&HWltnus • Olckbalt
Civic False and Misleading
• Fakeacct Contact
.._. Help •
-..,-----··· ..................
27
.....-.. ............ -_.., ..... _1.,.,
Selectors
_.........
.......... 25
Start !HJ
0ec ,. 2018
22 S
Civ,cStart !HJ
Jan 8, 2020
......,_ ..
~
20
TIIIP9 "' End !HJ
Feb 10. 2021
l
•
~
l
E
J!
17 5 RolingA-
Window ~·
1 15 Show
-.-- •
-- .
'2
I!!
t
.!l
12 5 ~By
l
lit 10
7S
• Anter Towards Polltlcal 0ut-9roup

e Prlvatalmpe..-tlon
-Nudity
10 ••
12 5 15
" Heard of Problem(ReputatlonalReach)
• SAGAnimals
17.5
• Ovlc Fakeacct Contact

20
• Ovlc Online Dl1courqement

22 5
• SRCDrvg1 Ovlc Demobilization

e Civic I 6 H Target • WebAdload
FalH and Mis-Ing • Graphic Vlolence
86H-• • Olckbalt
Ovlc FalH and Ml.-..
• FakeKct Contact
Pn1111en1
__._, p1o1(NpUtallon, perception- lntanaHy In the lat 7 dayol Explore
......, __ I, lnol. ..
27 5
lot10elNloerds 0.-
• ~ lnlilgrlty 0..h .. .
25
.
•lnllgrlly-Pllt
• ClaolllllrEaUmawd-•I.
.. .
..
22.5 38
Start
Civic Start
@Dec 1, 2018
fBlJan 8, 2020
• lnllgrlly GuMmllll {
..,..
• 1RIP8
~
.,
"0
-~
20 End fBlFeb 10, 2021
•
~ Rolling Average
~ 17.5 Window
IGTRIPS
OE1RIP8 l'B
15 Show Wolghled -
OETRIPS 1G
...... (MVP){DlpNc:ad) Group By

12.5
10
7.5
10 ••
12.5 15
" Heard of Problem(Reputat1ona
17.
ge Help •
am
.......... cmnpa,INn plot (raputallon, ...._uon and lntenalty In the lut 7 daya) A '. Explore
27.5
NI Dlll'1boerde
Gulde
• Cennllad lnliog,ttyO.h .. .
25
·~
tlnlag,!lrllw_,,,,,,.Plat .. .
Start mlDec I, 2018
22.5
ClaNlllerEatl- Prwal...
:i:
CivicStart mlJen 8, 2020
• lnlagrllyGuanltalla :;l
/!}_ 20
l End
•
i 17.5
RollingAverage
Window
15
• ~ Towards
7.S, 10 12.5 15
• 17.5
" Heard of Problem (Reputational Reach)
Polltlcal OUt-group
,- PdvaM Impersonation
• SRG Anlmals
• Ovlc Fakucct Contact
20 22.5 25
we Nudity • Civic Onllne Discouragement

• SRGDrugs Civic Demoblllzatlon
■ Civic B & H Target a WebAdload .......,Pnlblom
FalN and Mlsludlng • Graphic Vlolence
Civic B & H Menus • Olckbalt
• ---It; • Civic FalH and Mlsludlng
• B&HMtnus • Fakeacct Contact
,....._ aa,...i-.p1a11--- ,..,..._,, __...._
,--,.-, - Intensity In Iha lat 7 days) A •• Explore
27.5
25
Start fiilDec 1, 2018
22.5
CivicStart fiil""'.'8, 2020
:2
:,;:
.,
..,
"'., 20
End fiilFeb 10,2021
.,
-~
•
~
i 17.5 RotllngAverage
Window
E
.!!
.c
£ 15 Show
12.5 Group By
10
•
75
7.5, 10 12.5 15 17.5 20 22.5 25

" Heard of Problem (Reputat,onal Reach}
• Angar Towards Polltlcal Out-group • SRGAnlmals

• l'dvateImpersonation • Ovlc Fakucct Contact
Web Nudity • Ovlc Onllne Discouragement
• SRGDrugs Civic Demoblllzatlon
• Ovlc I • H Target • WebAdload
FalH and Mlsludlng • Graphic Violence
Clvk:l&HWhnass • Cllckbalt
e Najllty • CIVlcFalse and Misleading
• Fakaacct Contact
• l&HWI-•
Settings • Manage • Help •
am
- -,,porteon plot (n,putlllon, parceptlon ■nd lntanolty In the lut 7 d■yo) • Explore
• lnlegrtty--&lnal. ..
IMl~Gulde
27.S Selectors
► c.ntral- lntegrtty Dash ...
25
Slart mlDae 1, 2018
• Acauracy
22.S
• CIMelfler Elllmated Preval... Civic Start mlJan 8, 2020
.c
► lntagrlty Guanlrallo u
~
~
20
• TRIPS "'
.., End mlFeb 10, 2021
·!
•
nTRIPS ~
~ 17.S RollingAverage
IGTRIPS Window
~
:is
OETRIPS-FB
~
.., IS Show Wolght■d-., •
OETRIPS-IG
,...,. (MVP)(Deprecatad)
i
·~
C. 12.S
GroupBy --
TRIPSMVP(Depfecated)
~
1G Rllnce T~ metric
FB & Civic Flnanoe Toplinem...

..
3: 10
U8_....,.TRIPS
7.5
7.5-
• Anger Towards Political Out-group

• Private Impersonation
Web Nudity
10 ••
12.S IS 17.S
"Heard of Problem (Reputational Reach)
• SRCiAnimals
• Civic Fakeacct Contact
20
• Civic Online Discouragement

22.S 25
• SRC Drugs Civic Demoblllzatlon

• Civic B & H Target • WebAdload
False and Misleading • Graphic Violence
• Clvk B & H Witness • Cllckbalt
• Nudity Civic False and Mlsleadlng
• B & H Witness Fakeacct Contact
This post is a problem
Arturo, Jake, Pete, Emma, Andy, Diane, Marc A. Brackett, Robin S. Stern, & Andrés Richner
Josh, Charles, Travis, Tijana Health, Emotion, & Behavior Laboratory
Yale University
Relationships Matter!
Why are we here?
Bullying is a real problem!
• Definition: An intentional act of aggression, based on an imbalance
of power, that is meant to harm a victim either psychologically or
physically. Bullying usually occurs repeatedly and over time, but
sometimes can be identified in a single event.
• Over 50% of kids say someone has said mean or hurtful things to
them online.
• Over 50% of kids admit saying something mean to another person
online.
We have a responsibility to provide students with tools so they can be

both psychologically and physically safe online and in everyday life.
Overview
• Forming the Facebook/Yale relationship

• The original report flow
• Ideas for improving the report flow
• Methods for developing new flow
• The new report flow (v1.1)
• What do the data say?
• Next steps
Facebook / Yale
• Compassion Day 1
• Initial conversations about Facebook’s needs

and what the Yale team could provide
Original Report Flow
Ideas for improving report flows
• Infuse developmental science
▪ 13/14 year olds are different from high school and college students
• Use more kid-friendly language

▪ “Report” vs. “This post is a problem”
• Enhance logic of the flow

▪ ‘What happened?’ to ‘how are you feeling?’ to ‘what can you do?’
• Differentiate the experience so we could tailor support

▪ Move from just “harassing me” to real experiences of this age group
• Empower youth to take a positive and safe action

▪ Provide simple, effective guidance (e.g., “don’t be alone with this person”)
• Help youth to get more help from their community
▪ Encourage kids to reach out to a trusted adult
Methods for developing new flows
• Iterative process between Yale Team and Facebook Team:

▪ Review of existing research
▪ Focus groups with diverse students in public and private schools
▪ Interviews with children who experienced cyberbullying
▪ Interviews with parents, school principals, teachers, and counselors
▪ Integration of best clinical practices
▪ Taking Facebook design into consideration (e.g., writing, editing, and making sure we
got everything we possibly could into the limited space).
Focus Groups and Interviews
▪ Participants
▪ Public and private school students (N = 50; 13 to 15 year olds; 8 groups total) from diverse
backgrounds (east and west coast)
• Takeaways from focus groups and interviews

• Kids were particular about the language we used
• E.g., report – meant ‘authority’ or ‘trouble’ or ‘evaluated,’ ‘get help’ suggested ‘technical
problem’
• Kids helped us to differentiate the bullying experiences

▪
• Kids wanted Facebook to do something about it, but were not sure what
• If questions were meaningful and kids believed they would be helpful, they would be more
motivated to complete the flow
• Kids said they wanted help crafting messages
• Kids said not everything needed to be reported b/c they would just tell their friend…
Focus Groups and Interviews
▪ Takeaways from interviews with parents
▪ Parents were mixed on whether they should be the trusted adults
▪ (Some) parents enabled kids to fake their age
▪ Parents wanted more resources for their kids to get help
▪ Our own takeaway
▪ Had to be a balance between what kids wanted and what we believed they need
▪ Threatened – may not want to tell trusted adult, but they need help
▪ We needed to provide children with more direct help

The new Report Flow (v 1.1)
If “makes me uncomfortable,” go into Community Standards flow
However, if
“about me” or
“someone I know,”
go into…
If “threatening,” we lead to social resolution with extra messaging about safety:
If “posted something I don’t like,” send message with pre-populated text:
Other options (e.g.,“uncomfortable” / “said mean things” / “won’t leave me alone”) lead
to emotions slide:
After identifying emotion, lead to social resolution with text/options that vary as
a function of the situation and intensity of emotion:
Similarly, the option to message someone you trust is pre-populated with text
that also varies as a function of the situation and intensity of emotion:
Thank you slides are differentiated by experience
What do the data say?
“The post is a problem” 5/25-6/03
23600 (unique users)
(Female/male) 66%/34%
‘Me’ ‘Uncomfortable’ ‘Someone I know’

62% 17% 6%
70%/30% 62%/38% 62%/38%
Don’t like Photo Said mean Won’t leave alone Threat

53% 13% 4% 7% 3%
72%/28% 70%/30% 60%/40% 60%/40% 55%/45%
Msg Unfriend Block Tr Msg

59% 2% 6% 10%
Category breakdown
Area Sub-categories
Just don’t like (77%) • Photos: Awk pics, screenshots, vs photos, tag besties, spam
• Text: call out person, relationship post, tag besties
Posted a photo that makes me • Mostly bad (candids, funny face)

uncomfortable (17%) • Screenshots, porn, relationship, making fun
Said mean things (5%) • Photos: Tag besties, political, vs photos, screenshots,
joking/mean comments
• Text: family conflict, fights, passive aggressive posts
Won’t leave me alone (9%) • Screenshots, vs photos, spam
• ‘Pestering’ as opposed to ‘stalking’
Threatening (4%) • Photos: Bad photos, vs photos, screenshots, spam
• Text: rating girls (top ten), harassment
For those who picked ‘posted

something that I just don’t like’
▪ 60% send msg
For those who pick ‘threatened to

hurt me’:
▪ What % block: 6%
▪ What % unfriend: 3%
▪ What % choose trusted msg: 11%

▪ What % end up sending msg: 14% (2% overall)
▪ What % cancel: 22%
▪ What % choose no option: 27%
▪ What % navigate away: 31%

• For users who picked ‘said mean

things to me’ / ‘won’t leave me
alone’ / ‘posted a photo that
makes me uncomfortable’
▪ What % completed overall: 85%
▪ What % completed that were forced: 96%

▪ What % completed that were unforced: 73%
Distribution of emotions
Sad Nervous Afraid Angry Embarrassed
No answer 27% 30% 30% 24% 25%
Not at all 42% 42% 46% 35% 30%
A little 11% 10% 9% 12% 16%
Very 5% 5% 4% 10% 8%
Extremely 13% 12% 10% 19% 21%

‘said mean things’
No answer 28% 30% 30% 24% 26%
Not at all 26% 35% 36% 19% 24%
A little 14% 12% 11% 8% 11%
Very 9% 7% 6% 12% 10%
Extremely 24% 17% 16% 36% 30%

‘won’t leave me alone’
No answer 34% 40% 37% 32% 37%
Not at all 41% 40% 41% 33% 37%
A little 9% 7% 8% 11% 9%
Very 5% 3% 4% 8% 4%
Extremely 10% 9% 9% 16% 13%

‘posted a photo that makes me
uncomfortable’
No answer 24% 26% 26% 22% 19%
Not at all 46% 44% 50% 40% 27%
A little 12% 12% 10% 14% 20%
Very 5% 5% 4% 7% 10%
Extremely 12% 12% 10% 17% 24%

For those who complete emotion slide:
▪ What % block: 7%
▪ What % unfriend: 6%
▪ What % choose trusted msg: 14%
▪ What % end up sending msg: 24% (3% overall)
▪ What % cancel: 9%
▪ What % choose no option: 11%
▪ What % navigate away: 53%

Comparing Old and New Flows
• Were users more or less satisfied with the new report flow?
▪ One concern was that kids would be less satisfied with the new flow compared to the
old flow because the new flow was longer
▪ There were no significant differences
New Flow Old Flow

How easy? 1.89 1.92
How helpful? 2.23 2.18
How comfortable? 2.23 2.17
How satisfied? 2.19 2.22
• Did we change actual behavior? YES!
▪ Of those who completed the report (for more extreme instances), a greater number of
users in the new flow reached out to a trusted adult
New Flow Old Flow
Reaching out to 43% 19%

trusted adult
Blocking 28% 44%
• Two-days later: Was the trusted adult helpful?
• Trusted adults were perceived as being more helpful in the new versus old flows
(Ns are small; more data necessary)
New Flow Old Flow

Was the trusted adult 2.08 2.77
helpful?
(lower number means more helpful 1-4 scale)
Next steps
• Tweak v1.1 and release v2.
§ Have fewer cancelations
§ Get more children to reach out to trusted adult (or friend)
§ Provide even more specialized help to youth who are in danger
• Analyze data more carefully; publish findings

▪ Categorical analysis: Do reported posts map onto categories
▪ Dive deeper into each category. Some numbers are alarming (e.g., physical threats). More
categories likely are necessary
▪ Learn more from kids about what they need to navigate their lives online
• Start helping older age groups

• Build a comprehensive help center for teens and parents
• Prevention is the key!
Let’s Imagine…
Thank you!
Today we are faced with the preeminent fact that, if civilization is to survive,
we must cultivate the science of human relationships... the ability of all
peoples, of all kinds, to live together, in the same world, at peace.
Franklin D. Roosevelt
1945
Relationships Matter!
Emotionally Intelligent
Bullying Prevention
The 3rd Compassion Research Day
Yale Center for Emotional Intelligence

Arturo, Jake, Pete, Charles, Marc Brackett, Robin Stern,
Emma, Josh, Diane, Dan, Andy, Zorana Ivcevic-Pringle, Andrés Richner,
Tijana, & Aileen & Diana Divecha
Our Team
Cyberbullying
• Cyberbullying and “traditional” bullying are similar in many ways:
▪ An intentional act of aggression, based on an imbalance of power, that is meant to
harm the victim (physically or psychologically).
▪ Tends to occur repeatedly and over time, but sometimes can be identified in a single
event.
• But Cyberbullying also has unique characteristics:

▪ It’s more easily replicated
▪ It has limitless scalability
▪ It’s permanent
Prevalence of Cyberbullying
• 50% of middle and high school students say they have been cyberbullied
and 33% report bullying someone online (Mishna et al. 2010).
• Adolescents report that cyberbullying spills into ‘real life’

▪ 25% had experiences on SNS that lead to a face-to-face argument,
▪ 22% had an experience that ended a friendship,
▪ 13% got in trouble with parents, and
▪ 6% got in trouble at school (Lenhart, 2012).
▪ Why study cyberbullying?

§ 20% of teens think that people their age are mostly unkind on SNS (Lenhart, 2012)
§ Adolescents say cyberbullying is more serious than face-to-face bullying (Mishna et al. 2009)
§ Cyberbullying is related to higher anxiety and depression, lower grades (Tokunaga, 2010) and
higher rates of suicidal thinking and suicide attempts in adolescents (Hinduja, & Patchin, 2010)
§ 80% of US teens use social networking sites; 93% of them have Facebook accounts (Rideout,
Foehr, & Roberts, 2010).
▪
Two (seemingly) disparate fields
Emotional Intelligence Technology/Social media

§ EI introduced to psychology in 1990; reaches ▪ Internet reaches the public in 1994
public in 1995
▪ Social media evolves out of the chat room
§ EI is the ability to reason with and about emotions to and into popular networks
enhance decision making and promote both personal
growth and pro-social behavior. ▪ Internet keeps getting blamed for social and
psychological problems that are not new
§ Hundreds of studies demonstrating that EI is
associated with positive outcomes for young ▪ Facebook recognizes the potential power of
adolescents integrating emotional intelligence principles
into reporting systems
§ Our EI program, RULER, has demonstrated
positive results in shifting school climate and
children’s prosocial behavior
The life of a 13-14 year old
Young Adolescent Development
Biological Changes
§ Onset of puberty leads to hormonal instability
§ Executive network that allows self-regulation, planning, and overall monitoring, are “under
development”
§ Social excitement literally overwhelms the ability to control behavior.
Cognitive Changes
§ Improvements in thought complexity makes kids more vulnerable to what others think.
“Imaginary audience” (thinking that everyone sees them) makes them especially self-conscious
and vulnerable to embarrassment.
Self and Identity

§ Separation/individuation from parents; peer group offers temporary identity so they can become
“autonomous”
§ Young adolescents are especially sensitive to peer relationships – power dynamics and increased
risk-taking especially in presence of peers.
Overview
• The original report flows (13-14 year olds)

• Infusing emotional intelligence
• What we learned from v1.1
• Version v2.0
• What the data reveal
• What’s next?
The original report flows
Infusing emotional intelligence
• Takeaways from initial focus groups and interviews

• Kids were particular about the language we used
• E.g., report – meant ‘authority’ or ‘trouble’ or ‘evaluated,’ whereas ‘get help’

suggested ‘technical problem’
• Kids helped us to differentiate bullying and non-bullying experiences
• Kids wanted Facebook to do something about it, but were not sure what that was;
wanted a ‘conversation’
• If questions were meaningful, specific, and helpful, they would be more motivated
to complete the flow
• Kids said they wanted help crafting messages
• Kids did not believe everything needs to be reported b/c they would just tell (call,
text) someone they trusted
▪ Takeaways from interviews with parents
▪ Parents were mixed on whether they should be the ‘trusted’ adults
▪ Some parents enabled kids to fake their age
▪ If their child was threatened, they wanted to know
▪ Parents wanted more resources for their kids
▪ Our own takeaways

▪ Had to be a balance between what kids wanted and what we believed they need
▪ E.g., Threatened – may not want to tell trusted adult, but they need help
▪ A conversational approach was ideal
▪ We needed to provide children, parents, and educators with more direct help
• Infuse developmental emotion science – more adolescent-friendly

language, enhanced logic, more relevant)
▪ 13/14 year olds prefer “This post is a problem” to “Report”
▪ ‘What happened?’ to ‘how are you feeling?’ to ‘what can you do?
▪ Move from just “harassing me” to “saying mean things to me”
• Integrate emotional intelligence

▪ How did the post/photo make you feel? (both emotion and intensity)
• Empower youth to take a positive, safe action both on- and off-line
▪ Provide simple, effective guidance for less versus more threatening posts
▪ Develop positive pre-populated messages to content creator/trusted adults or friends

The Present Study
Version 2.0
DEMOGRAPHICS
What we learned from v1.1
• Most reports were about ‘self’ as opposed to others
• Most kids just want to be ‘untagged’ from posts/photos
• Photo and post report systems needed to be separated
• We wanted to increase messaging to content creator and trusted
friends/adults and decrease blocking/unfriending
• We needed to improve pre-populated messages to help teens
communicate with content creators and trusted friends and adults,
• We also wanted to help trusted friends and adults communicate with
the reporter
• We wanted to increase completion rates
• Gender was a variable that needed to be explored
Descriptive Statistics
▪ Reporter Information
▪ N = 402,269 13-14 year olds (distinct users)
▪ Girls = 68%: Boys = 32%
▪ All reports are between 9/1/12 to 12/31/12
▪ Median # friends = 295; Girls = 332; Boys = 229
▪ 1.5 reports, on average
▪ Reporters were assigned randomly to old versus new flows
▪ Based on approximately 4,000 follow up surveys:
▪ Reports completed mostly by kids (85%), although some were
completed by kids with their parents and parents alone (15%).
Descriptive Statistics
▪ Content Creator Information

▪ Girls = 70%; Boys = 30%
▪ Median # friends = 405; Girls = 438; Boys = 351
▪ Reporter/Content Creator Mix

▪ Boy Reporters – Content Creators are: 55% (girl), 45% (boy)
▪ Girl Reporters – Content Creators are 75% (girl), 25% (boy)
Version 2.0
PHOTOS
Photo Report Flow 2.0
“I just don’t like it”
“I just don’t like it”
“It’s harmful and might affect my reputation”
• Messages are tailored to
emotion intensity
• Can also send message via
email
- spreading rumors -
Low Intensity High Intensity

Jake, I don’t appreciate the rumors Jake, I really don’t appreciate the
being spread about me. They make me rumors being spread about me. They
uncomfortable. Please stop and take make me very uncomfortable. Please
this post down. stop and take this post down.
“This PHOTO is a problem”
721,670
(Girl = 73%/Boy = 27%)
I just want to untag I would like this PHOTO I want to help

myself/spam removed from Facebook someone else
71% 15.5% 0.4%
77%/23% 62%/38% 65%/35%
I just don’t like it It’s harmful and might It shouldn’t be on

71% affect my reputation Facebook (TOS)
65%/35% 16% 13%
56%/44% 60%/40%
I just don’t like it
80,054
65%/35%
Bad photo Embarrassing Inappropriate offensive other

35% 15% behavior 6% 27%
65%/35% 68%/32% 4% 57%/43% 64%/36%
55%45%
• On average, 58% of kids send messages to content creator

• Girls are more likely than boys to send messages for bad or
embarrassing photos
It’s harmful and might affect my
reputation
18,642
56%/44%
Afraid Angry Embarrassed Sad None

7% 16% 28% 6% 10%
57%/43% 52%/48% 61%39% 53%/47% 53%/47%
• Embarrassment is the most frequently experienced emotion

• Embarrassment results in more messaging (18%) compared to all other emotions
• Girls are more likely than boys to send messages when embarrassed or afraid (7:3)
Message CC Rate by Intensity of Emotion
25%
20%
15%
Boys
10% Girls
5%
0%
Not at all A little Somewhat Very Extremely
• Emotion intensity is correlated with messaging, especially for girls

• Importantly, 84% of kids use our prepopulated (positive) messages
Reporting photos: Summary
• Most adolescents simply wish to ‘untag’ themselves
• When wanting to remove photo, most adolescents ‘just

don’t like it’ because ‘it’s a bad photo’
• Harmful photos are largely associated with embarrassment
• Stronger emotions result in greater likelihood of sending

messages
• Gender differences are noteworthy

Version 2.0
(POSTS)
Post Report Flow 2.0
“I just don’t like what it says”
“Someone is bothering or bullying me”
• Messages are tailored to emotion
intensity
• Can also send message via email
“The POST is a problem”
61,305
(Girl = 61%/Boy = 39%)
I just want to untag I would like this POST removed I want to help
myself/spam from Facebook someone else
39% 27% 1.4%
66%/34% 61%/39% 60%/30%
Someone is It shouldn’t be on
I just don’t like what bothering or
it says Facebook (TOS)
bullying me 19%
59% 22%
60%/40% 65%/35%
60%/40%
Message CC = 63%
57%/43%
Someone is bothering or bullying me
6,499
60%/40%
Posted mean Won’t leave me Is spreading Threatened to

things alone rumors hurt me
20% 10% 12% 6%
64%/36% 60%/40% 55%45% 60%/40%
Msg Other = 8%
Block = 4.4%
Unfriend = 12.5%
• Anger is the most experienced emotion across all categories
• Non-significant gender difference on emotion “pick”
• Girls report having more intense emotions than boys
• On average, 10% of kids send messages to content creator and 3% to trusted
friends or adults
Area Low intensity High intensity
Said mean • Mocking reporter for over engagement • Negative post about unnamed
things with FB individual
• Accusing reporter of being fake • Targeted insults (e.g. fat, gay, slut)
Won’t leave • Mocking reporter for over engagement • Re-sharing reporter’s content
me alone with FB • Top 10 lists
• Jokes about appearance
Spreading • Negative post about unnamed • Slurs

rumors individual • Top 10 lists
• Relationship gossip • Sexually derogatory comments
Threatening • Aggressive
(emotion not • Name calling
asked) • References to offline activity and
situations
• Said mean things

▪ “[He] is gay as hell !!!! Dont be his friend !!!!”
• Won’t leave me alone

▪ “Get some proactive , and a better attitude , THEN we'll talk . (;”
• Spreading rumors
▪ “[She] is such a whore.. she's told me she slept with 5 different guys and she's willing to do
more. What a whore.”
• Threatening
▪ “Watch your back you little bitch (; your going to wish you never fucked with me.”
Reporting Posts: Summary
• Similar to photo reports, most young adolescents simply wish to

‘untag’ themselves from posts
• When wanting to remove the post, most young adolescents ‘just

don’t like what it says’
• When being bothered or bullied, most report ‘mean things’ being

posted, resulting in anger
• We need to unpack more what’s happening for kids who report

that someone is threatening to hurt them
• Again, there are noteworthy gender differences

Experimental Findings:
Original vs. v2.0
Old Flow vs. 2.0 Flows
Old flow New flow

19 sec
80%
77%
15 sec
Completion rate Time spent in flow (not just untag)

Old Flow vs. Photo 2.0 Flow
71%
71%
Old Flow New Flow
21%
14%
10%
3%
2%
0.1%
Untag
Blocking Report Content Message Content Creator
Old Flow vs. Post 2.0 Flow
Old Flow New Flow
15%
11%
8%
3.9% 4%
1%
0.2% 0%
Blocking Unfriends Report Content Message Content

Creator
Discussion
• Gender matters
▪ Reporting behavior – girls report more than boys
▪ Bullying behavior – girls are more likely than boys to be the ‘content creators’
• Embarrassment is most frequent emotion associated with photos

▪ Kids are self-conscious about the way they look
• Anger is most frequent emotion associated with posts

▪ Kids “say mean things” which is perceived of as an injustice
• Emotion intensity is associated with behavior (messaging)

▪ Emotions drive decision making and action
• Providing kids with a more emotionally intelligent report flow helps to have
more positive interactions
▪ Kids are more likely to “stay in the relationship” and make constructive decisions like sending
positive messages as opposed to blocking
▪ In essence, we have eliminated ‘blocking’ – likely an ineffective coping strategy
Limitations and Next steps
• These data only represent kids who reported; many kids do not know about the
reporting system so there is a need to get the word out.
• We need to try new methods for follow-up survey data – satisfaction, resolution?
Follow up on content creator, trusted friends/adults
• Confirm findings in a fresh sample with some tweaking to the flow
• Qualitative Analyses
▪ Gender differences
▪ Mapping posts onto categories
▪ Examine posts preceding and following report
• Help Center for kids, parents, and educators
• 15-16 Year old flows coming soon

Thank you!
Emotionally Intelligent
Bullying Prevention
The 4th Compassion Research Day
December 5, 2013
Yale Team
Marc Brackett Robin Stern Zorana Ivcevic
Mrinalini Rao Diana Divecha Cynthia Dickason-Scott
Charlie Sherman
Facebook Team
Bhal Agashe Arturo Bejar Emiliana Simon-Thomas
Rob Boyle Tessa Cafiero Pete Fleming
Charles Gorintin Samantha Gruskin Jennifer Guagagno
Cheryl Lowry Mojtaba Mehrara Dan Muriello
Diane Murphy Mamal Poladia Nikki Staubli
Dave Steer Siqui Yan
Creating Evidence-Based Tools for Teens
• Part I – Social resolution flows for teens (ages 13-16)

– Provide kids with tools to help them manage unpleasant
experiences
• Part II – Bullying Prevention Hub

– Provide kids, parents, and educators with high-quality
resources to manage and prevent bullying
Adolescence and Social Media
Adolescence and Peer Relations
• Peer relationships are a central focus for teens
• Creating and maintaining positive relationships doesn’t

happen automatically
• The adolescent brain is different
• Emotion skills matter

Emotionally Intelligent Bullying Prevention
• Infused a developmental framework
• Incorporated age appropriate/conversational language
• Integrated emotional intelligence skills
• Empowered youth to take positive action

The Resolution Tools…
The Present Sample
60%
50%
• N = all 13 -16 year olds in
40% U.S. who entered resolution
Girls
tool within a 30-day period
30%
Boys • Older girls use the tool the
20%
most and also are reported
10% more
0%
Younger Older
What are the resolution tools being used for?
Posts
24% Posts:
Photos:
“It’s harmful or might hurt my reputation”
Photos
76%
*Of all teens entering the flows, 15% select

‘bullying.’ Most (66%) select ‘annoying.’
“What happened?”
45%
40%
35% • No gender differences
30%
• No age differences
25%
20% • Younger boys report more
15% threats. Older boys report more
mean things.
10%
5%
0%
Won't Mean Rumors Threats Other
leave things
alone
*Breakdown of 15% who select ‘bullying.’
“How does this post/photo make you feel?”
Posts Photos
50%
• Girls report more sadness and
45% embarrassment than boys
40%
35% • Boys use “none” more than girls
30%
25% • No age differences
20%
15% • Younger boys report being more
10%
afraid and more threats than
older boys
5%
0%
d
d
d
ry
e
ov
Sa
se
ai
ng
fr
ab
as
A
rr
e
ba
on
Em
N
Sometimes it’s clear why teens label posts as bullying
He was crying today lol You better watch yo back I'm
- Reported by 14 year old girl going to knock you out
tomorrow.
- Reported by 14 year old boy
I feel like he just used me! But I also

thought he loved me. I should have
known better. Maybe one day we'll And sometimes it’s not
get back together!
So ready to go home and go to bed!
- Reported by 17 year old boy - Reported by 13 year old boy
I got contacts. No more glasses.

- Reported by 16 year old boy
What actions do teens take?
• 25% of teens message - person who posted the content (90%) or a

trusted adult/friend (10%)
• 75% of teens use the pre-populated (positive) messages
• Younger teens message more. However, younger boys who report

‘afraid’ send more reports to Facebook.
• “Won’t leave me alone” è use pre-populated messages; “rumors” è

tailor messages
What happens next?
• Content creator behavior:

• 75% reply to the message
• 37% delete content
• Parental involvement:
• 38% of younger teens vs. 23% of older teens’ have parent
involvement
Summary of findings
• Like face-to-face bullying, online bullying results in a range of emotional
experiences for both boys and girls (embarrassment/anger are dominant)
• Teens’ online lives look similar to their offline lives:

- More girls than boys report being sad and embarrassed
- More boys report ‘threats’ than girls
- Boys are less willing to disclose feelings
• Age differences in the ‘content’ of online bullying are consistent with face-to-face
bullying (e.g., homophobic bullying)
• When given effective “tools” teens appear to send messages – and when they
learn have done something ‘wrong,’ they tend to respond
• These results helped to inform us about other tools kids and adults needed
Part 2: The Bullying Prevention Hub:
A day in the life of

father and son
Insights
Methodology Learnings
• Focus groups with teens, • Awareness of bullying, but low
educators, parents comprehension of what to do
• In depth summit with nonprofits • All stakeholders want guidance
• Data from social resolution • It’s about bullying intervention and

flows prevention wherever it occurs –
focus on the behavior, not the
location or platform
Our goal was to develop emotionally intelligent
bullying prevention resources
Resources for all stakeholders:

• Parents, educators, and teens
• Bullies, targets, and bystanders
Knowledge and skills content:

• Resources which build self-awareness, self-regulation, problem
solving, and healthy communication.
Safety is a Conversation
• Provide the right advice to the right user at the right time.
• Expand our bullying prevention campaign and the Family

Safety Center.
• Help on the other side of the reporting button.

• Showcase resources from dozens of organizations
Let’s go back to the role play
Charlie was accused (and is guilty) of posting

something inappropriate – a photo that was mean and
hurtful – about his classmate. It was a picture of his
classmate Jamie at a sleepover party. The photo
showed her drinking a beer.
Marc, his father, got the call about this from the
school principal.
STEP 1
Set yourself up for a successful conversation

with your child.
– Find the best space to have the conversation.
– Check in with and manage your own feelings (before)
– Remember, you are the role model.
– Support and listen.
STEP 2
Talk with your child about the problem.

– Find out what happened.
– Communicate your family’s values (e.g., respect, kindness).
– Use a calm and steady voice; avoid making empty promises.
Sample Conversation Starter:

“I got a call from your teacher today who told me that you have been posted
a offensive photo of Jamie. I need to know what happened so we can
decide what action needs to be taken.”
•
STEP 3
Work with your child on an action plan.

– Solve the problem together.
– Ask fact-finding/open-ended questions to help your child generate
solutions
– Decide on an appropriate action plan (e.g., apologize)
Sample Conversation Starter:

“What do believe are some appropriate ways to handle this situation?”
STEPS 4 & 5
Be clear about consequences, follow through,

and follow-up
– Be firm and consistent, taking into consideration your values and
severity of incident.
More opportunities to help your child…

– Pay closer attention to your child’s Internet and cell phone activity.
– Advocate for an evidence-based social and emotional learning
program for your child’s school.
– Consider counseling for your child and/or family.
Back to Charlie and Marc
• Set yourself up for a successful conversation with

your child.
• Talk with your child about the problem.
• Work with your child on an action plan.
• Be clear about consequences, follow through, and

follow-up
• Explore more opportunities to help your child

The future of emotionally intelligent bullying
prevention
• Social Resolution Tools

• Examine role of gender and age in more detail
• Conduct qualitative analysis of posts and photos
• Run longitudinal studies on teens online behavior, including follow-up survyes
• Begin cultural adaptations
• Share findings in peer-reviewed journals
• Bullying Prevention Hub

• Study the use and impact of hub
• Create more interactive tools for all stakeholders (e.g., videos)
• Build bully education center
Emotions
Without
Borders
Supporting Teens Across
the World on Facebook
Marc Brackett, Mrinalini Rao,

Robin Stern, & Zorana Ivcevic
Yale Center for Emotional Intelligence
Facebook’s Protect and Care Team

Vision
To use the power of emotional intelligence
to create a more healthy, effective, and
compassionate society.
Mission
To conduct rigorous research and develop
innovative educational approaches to empower
people of all ages with the emotional intelligence
skills they need to succeed.
Emotions Matter
A rollercoaster of emotions
Emotions Matter
Emotions drive:
• Attention, memory, and learning
• Decision making and judgment
• Relationship quality
• Physical and mental health
• Everyday effectiveness
Emotions Matter for Teenagers
To many, adolescents appear to be illogical, irrational,

and invincible, but…
• Puberty introduces hormonal changes
• Emotion and cognitive systems are not harmonized
• Separation and individuation from parents is
• Peers have a strong influence
Emotions Matter for Teenagers
• Seeking easier means to gain rewards

• Increased risk taking (e.g., driving, risky sexual behavior)
• Delinquent behaviors
• Substance abuse
• Psychiatric diagnoses
• Suicidality
Emotional Intelligence
• Recognizing emotions
• Understanding emotions
• Labeling emotions
• Expressing emotions
• Regulating emotions
How Emotional Intelligence Develops
Developing Emotional Intelligence
“Between stimulus and response, there

is a space. In that space lies our
freedom and power to choose our
response. In our response lies our
growth and freedom.”
VIKTOR E. FRANKL
Developing Emotional Intelligence
Moving from automatic to intentional ways of behaving

• Yelling to deep breathing
• Negative self-talk to positive self-talk
• Impulsivity to reframing
• Rumination to positive visualizations
• Avoidance to finding support from others
Facebook – Yale Collaboration
Applying EI to Facebook
• Initial focus was on building social resolution tools -

Helping youth manage unpleasant experiences
• Began working with teens (13-18) in the U.S.
• Consulted with teens and other stake holders

• Used a developmental framework
• Infused age-appropriate language
• Incorporated emotion science
Social Resolution Tools
Research and Evaluation
17.2 million events

(50-day period)
904,000 teens (5.2%)
59,311(6.6% or 0.3%)
Bullying Behaviors
25
20 Photo
Text
15
Percent
10
0
Won't leave Mean Rumors Threatening Other
me alone
Emotional Reactions to Feeling Bullied
50%
Photos
40% Text
30%
20%
10%
0%
Emotional Reactions to Feeling Bullied
50%
Photos
40% Text
30%
20%
10%
0%
Influence on Behavior
Old Flow New Flow
21%
14%
10%
3%
2%
0.1%
Blocking Report Content Message Content

Creator
Untag *These data are from earlier pilot data, comparing

adult flows to the revised flows
Facebook – A Global Company
How can we support teens in a developmentally and

culturally responsive manner?
Culture and Emotions
• Individual differences
• Social norms
• Culture
Cultural Display Rules
Culturally prescribed rules that govern how universal emotions

can be expressed.
• Rules of social appropriateness
• Learned early in life
• Automatic practice by adulthood

Cross-cultural differences
• Acceptable behavior • Experience of emotion
• Unwanted behavior • Social resolution

Understanding the Role of Culture and Language
United
Kingdom
United Ireland
States
Mexico
Vietnam
India
Indonesia
Brazil
Why don’t you want to see this?
100%
80% Annoying
60% Bullying
Don't like
40%
Inapproriate
20%
Spam
0%
co
a
nd
m
S
i
az
si
di
U
na
i
la
ne
ex
In
Br
I re
et
M
do
Vi
In
What is happening in this post?
50
45
US
40
India
35
30
25
20
15
10
5
0
Harassing Mean Rumors Threatening Something else
What is happening in this post?
“Oh yes, we bully each other all the time. It’s fun, it’s drama.”
“I bully my boyfriend’s ex. Then I found out my boyfriend was

still friends with her on Facebook, so I bully him about it too.”
“Bullying is entertaining, I love it. I’m so “kepo” – I’ll go their

profile to see it”
- High school students in Indonesia

How does this make you feel?
Challenges interpreting cross-cultural findings
• What does bullying mean?
• How do behaviors like bullying impact the individual?
• What do people do offline when they are offended?
• What is the best way to facilitate resolution on Facebook?

Supporting Teens Across the World
• Aligning with teens’ lived

experiences.
• Learning from teens: What is

going on?
• Offering online support that

parallels offline cultural norms
In Conclusion
• Emotional experiences around “meanness” or bullying are

universal. Behaviors that elicit the emotions vary culture to
culture.
• There is a universal need to be seen, heard, and met
• The ways in which people desire to be seen, heard, and

met vary as a function of culture
• Our goals are to investigate ways to promote both

universal and culturally specific respectful, compassionate
interactions online
Thank you

Be Jar Whistleblower Documents

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Be Jar Whistleblower Documents

Uploaded by

Copyright:

Available Formats

Internal Instagram and Facebook Documents

These documents were created during my time working at Instagram as a

Document labeled ‘1 - “Bad Experiences” Measurement - Plan for a Plan -

The first document is a set of slides prepared by me and members of the

Document labeled ‘2 - AI + FAI Workshop_ bad experiences.pdf’

By early 2021 the idea of working on “bad experiences” had begun

While our recommendations for regular measurement, reporting, and goal

Document labeled ‘7b - BEEF by Age (attachment to Gap in understanding

Email to Mark Zuckerberg and M-team ‘7a - Gap in our understanding of

Email to Adam Mosseri ‘8 - WSJ published Mosseri Pre-Read

Folder ‘2011-2015 Published work by Facebook on Teen Bullying’

The document ‘​​CRD3_Yale_Team_Compassion Research Day

Folder ‘From Facebook Files’

The solutions we create I believe should have the following properties:

compare user perceptions across problems, estimating:

1. Perceived reach of a given integrity problem

How can I stay updated about TRIPS?

Why should we care about ·perception?

Perception allows us to verify whether product interventions actually work. Integrity

What is TRIPS' mission?

We combine rigorous methods with subject matter expertise. The li

Broadly, we serve as a mouthpiece for the people who h

What problemsdoes TRIPScover?

SRG: Animal sales ,,,, ,,,, 2020

Off-site landing pages: Low ,,,, N/A 2020

Civic inflammatory ,,,, ,,,, 2020

What is the future of TRIPS?

1 -bles producttaamsto l1IOfllproactlvelv

~ Se ~ F~ i-', Trncki Flea of lrlegnty Pro lems Surveyl

!'.El Wik1 ) Central Integrity Research TRIPS(Tracking Reachof Integrity ProblemsSurvey)

) Exper ment1ng with TRIPS

1. enables product teams to more proactively detect and respond to harm,

1. broaden coverage across more Integrity problems and more surfaces;

• Central ln<ege!y R_,a,c;h

ontens11y PerceplionFramework Community

HOW TO INTERPRET THE VARIABLES

• Perceivedreach: % of people who have seen/experienced the integrity problem

1. TRIPS Team Dashboard

♦ Problem 0 Percefved AHch C)

-0 75% did +-083"-d/d

• Problem 0 Perceived Re.ch 0 Repulltlonal Reach O lntensltyO

J.in '20 May '20 Seg 20 Ja 21

8.71 21.42 49.54

+0.09% did -0.27% did

·l!.M% - Mav ·20 Sep '20 Jan '21

13.31 14.48 46.96

M,11y20 ~l'I) ·20 J•n 21

May'20 Sf,p'20 J.a.n'21

M;r,y'20 ~p'20 Jan ·21 • -

May '20 Sep '20 ~" '21

•-o .........,....,.0 X .,.,~

13.31 14.48 46.96

May '20 Sep '20 Jan '21

·-· --"· Reputatlonal Reech 0 lntan■lty0

.4 78~ m/m .Q.07"' m/m

Jan '20 May '20 Sep '20 Jan '21

serlH_l_ci (range): 27.64 10 29.t ◄

11.37 ,., A ,r, 15.92 I ,. J 46.25

·-· Percei-...d Ruch O

·1.SN nvm •2.62% wlw

Jan '20 May '20 Sep '20 Jan '21

Jan '20 May '20 Sep '20 Jan '21 -----

Clckontopofuehcotumntoraru- req ryp b lom

The document ‘CRD3_Yale_Team_Compassion Research Day