This document discusses how evaluators are not taking full advantage of digital media analytics to inform program evaluations. It outlines how social media and online data can provide insights into public perceptions, behaviors, and opinions. Specifically, digital analytics can help measure outcomes, gather hard to reach stakeholder views, establish baselines, and provide context. However, there are also challenges like misinformation and privacy concerns. The document argues digital analytics could strengthen evaluations by supplementing traditional methods and calls for further exploring opportunities for its use.
Human Factors of XR: Using Human Factors to Design XR Systems
Why aren't Evaluators using Digital Media Analytics?
1. Why aren’t
evaluators taking
advantage of Digital
Media Analytics?
from social media to open web sources
Giles Crouch, CEO
giles@mediabadger.com
Tasha Truant, Consultant Manager
ttruant@ggi.ca
2. agenda
The Internet | Breaking Assumptions
What is Cyber Analytics?
Examples
Strengths and Challenges
Applications and Opportunities for Evaluation
Case Studies
MediaBadger + Goss Gilroy
4. dispelling some myths
Average age of social media users is 36
USA, Canada, UK, EU
The 55+ demographic is fastest growing
25% of all content uploaded is now mobile device
Connecting to social media & the web is not done
at home from a desktop anymore
Approximately 40% of the developing world is on
social media
5.
6.
7. what social media can tell us
Sentiment on issues (societal, political etc.)
Key influencers and their networks
Cultural behaviors and actions
(e.g. Kenya; local social media versus Facebook)
Myths and narratives on issues
e.g. Tar sands are acidic (in reality bitumen is less
acidic than traditional oil and gas)
Important to understand myths and how they turn
into narratives – very difficult to change narratives
8. what social media can tell us
Demographics
All social media apps record certain “meta data”
such as location (geo) time and IP address
Keywords
When gender isn’t identified, use face recognition in
from profile pictures
Over 60% of posts contain a person mentioning
where they are from and items such as tribe (i.e. for
First Nations) and often city/province
Combination of identifiers are used to create a user
profile (like the Target example from yesterday)
Margin of error is between 4-6%
11. 2010 New
Brunswick Flood
Event
Historical Trends | Communications Patterns | Citizen Behaviours
• Examined how people used social media immediately before, during and after the
flood event
• Findings: Citizens paid very little attention to crisis communications from authorities
• Instead, relied on each other, posting pictures and videos, warning people in
forums etc. of where to go or not go
• Result of research let to overhaul of emergency comms and use of social media
14. difference of approach
Traditional (survey)
Limited Data
Time Snapshot
Specific Point
Curated & Pristine
Must be clean
n=xxxx
Defined, Limited
active bias (bias in asking the
questions)
cyber research
Massive Data
Historical Data
1985 Onward
Chaotic Accuracy
Messiness is OK
N=ALL
Exponential
passive bias (listening bias)
15. premise for use in evaluations
Evaluators spend a lot of time and money trying to
find out how people perceive, and are affected by,
programs, policies, and organizations
World is increasingly voicing their opinions online
Methods to extract insights from the large and
complex collections of digital data openly available
exist
16. possible applications
Impact of public policy programmes + projects
Large data sets, near real-time
Historical trending & sentiment
Inform research design
Supports field research, focus groups & interviews
Using alongside traditional lines of evidence
Establish baselines for analysis & future monitoring
17. opportunities: measurement & context
Measure the outcomes of public outreach, communication,
advocacy, and information sharing programs
Create detailed stakeholder maps
Maps of social networks that capture relationships,
provide insight into their nature, and identify unknown
stakeholders and influencers
Provide context and insight to inform further data collection
E.g. Country profiles of internet/social media usage
Historical trends
Take a snapshot of these data at several points in time
18. opportunities: respondent groups
Traditional methodologies have a hard time reaching
certain respondent groups
Useful in gathering unvarnished views from groups that
have access/means to get online, but benefit from
anonymity. Useful for:
Data collection in sensitive environments (e.g. post-
conflict zones);
Obtaining views on issues people are quiet about in
person (e.g. racism)
Gathering perceptions from beneficiaries adverse to
authority (vulnerable and marginalized
populations, criminal offenders, youth)
19. When there could have been Sentiment
Analysis in Evaluation
Arts Promotion
Program aims to build stronger citizen
engagement in communities through
the performing and visual arts and in
the expression, celebration and
preservation of local historical
heritage.
“Limited evidence gathered to fully assess the
ultimate outcomes of the program, stemming
from … the fact that the evaluation team could
not gather direct views from a representative
sample of volunteers and the general public. “
Strengthen Civil Society
Program’s goal is to promote resilient,
healthy and just communities and
support processes that strengthen civil
society.
“Determining the extent to which [the program]
has increased the knowledge or actions taken
by Canadians in food, environmental and
biodiversity issues was not possible within the
scope and budget of this evaluation in the
absence of survey data or a baseline”.
20. When there could have been Sentiment
Analysis in Evaluation
Canadian Culture
Program aims to develop Canadian
writers, and to publish and
disseminate their books effectively in
Canada and abroad. The ultimate
program outcome is “Increased
access to a diverse range of
Canadian-authored books in Canada
and abroad”.
“Certain program outcomes could not be
directly measured during the evaluation…For
example, in order to measure indicators such
as “Increased Awareness” and “Increased
Access”, the evaluation has had to use proxy
indicators (e.g. sales) to infer awareness and
access.
Immigration/Settlement
Program aims to contribute to
improving labour market integration
outcomes of foreign-trained
individuals in targeted occupations
and sectors.
In most areas of anticipated outcomes, there
was not a baseline measure of these outcomes
at the point of implementation of the program.
As a result, measurements of change or
improvements rely on the recall and opinion
of current respondents.
21. biases + challenges
survey
Trying to locate users/beneficiaries
With small amounts of data,
accuracy is key
Locating hard to reach populations
Sampling Biases (Whom to survey?)
Selection biases (people that
respond to surveys)
sentiment analysis
Production of intentionally
misleading content (i.e. astro-
turfing)
Overly aggressive behaviours
(troll tendencies)
Difficulty in detecting sarcasm,
irony
Selection biases (people that
post opinions online)
22. Privacy and Ethics
All data collected are publicly accessible; no private data are
accessed.
MediaBadger is in full compliance with Canada’s privacy law
(PIPEDA) at all times.
MediaBadger is reviewing membership in industry associations that
provide clear, ethical guidelines in line with our values.
Digital Media Analytics and Sentiment Analysis maintain the
confidentiality of all respondents; responses are aggregated in ways
similar to traditional lines of evidence (i.e. key informant interviews).
23. Exploring opportunities for digital
media analytics as a line of evidence
in evaluation
Difficult when ToRs are established and
budgets are fixed
Outreach to the designers of
evaluations (many of you!)
Looking for opportunities to
further the field of evaluation
through this workBringing advanced cyber
analytics to program
evaluation.
partnering