Page 1
Cody Buntain@[email protected] of Maryland
Erin McGrath, Gary LaFree{ecmcgrath,lafree}@umd.eduSTART Center, UMD
Jennifer [email protected] of Maryland
Comparing Social Media and Traditional Surveys Around the Boston Marathon Bombing
1
#Microposts201611 April 2016
Montreal, Quebec CA
Page 2
Surveys are mature, ubiquitous, and
powerful instruments
2 - Introduction
Page 3
Social media platforms produce huge data sets and let users respond to events
instantly
3 - Introduction
0
0.4
0.8
1.2
1.6
Face
book*
China
USA
Twitte
r*
Popu
latio
n (B
illion
s)
Instagram
Twitter
Facebook
1 100 10,000 1,000,000 100,000,000
New posts
Since I’ve been talking
Page 5
5 - Introduction
Socialmediaisn’t
representa1ve!
Surveysarebe7er!Socialmediais
cheap,big,and
easy!
The Argument
Page 6
What is the point of this talk?
6 - Introduction
Page 7
- Introduction7
The Point
To compare social media and survey data
Compared across: • Costs • Data Types • Relevance • Validity
Page 8
- Introduction8
Grounding the Point
Page 9
9 - Case Study
What makes the Boston Marathon Bombing a good case study?
Page 10
The Bombing from Multiple Perspectives
• Three-wave survey on public perceptions of law enforcement
• The bombing happened to occur between waves
• Post hoc survey on information seeking/sharing during crises
• Included respondents exposed to Boston Marathon coverage
• Analysis of public response to the bombing on Twitter
• 134 million tweets from April 2013
10 - Case Study
Page 11
Why Twitter?
11 - Case Study
Page 12
12 - Results
What are the strengths and weaknesses of surveys and social media?
Page 13
Four Axes of Comparison
• Data types
• Relevance
• Cost
• Validity
13 - Results
Page 14
Comparing Data Types
14 - Results
Survey Data
Social Media
Data
• Attitudinal data
• Behavioral data
• Demographic data
Page 15
Comparing Data Types
15 - Results
Survey Data
Social Media
Data
• Attitudinal data
• Behavioral data
• Demographic data
Page 16
Comparing Data Types
16 - Results
Survey Data
Social Media
Data
• Attitudinal data
• Behavioral data
• Demographic data
Page 17
Comparing Data Types
17 - Results
Page 18
Comparing Data Types
18 - Results
Boston PD Follower CountsBoston-Related Twitter Activity
Page 19
Comparing Data Types
19 - Results
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
04-01
04-03
04-05
04-07
04-09
04-11
04-13
04-15
04-17
04-19
04-21
04-23
04-25
04-27
04-29
Percen
tageofT
weetsCon
tainingEm
o=on
Date
Anger Disgust Fear Happiness Sadness Surprise
Emotions in Boston-related Tweets from the US
Page 20
Comparing Data Types
20 - Results
Self-reported “willingness”
Sentiment Towards Police in the US
Page 21
Takeaway: Social media and survey data provide
different types of data
- Results
Comparing Data Types
21
With some overlapAnd both are useful for different purposes.
Page 22
Four Axes of Comparison
• Data types
• Relevance
• Cost
• Validity
22 - Results
Page 23
Comparing Relevance
23 - Results
• Temporal Relevance
• Topical Relevance• Geographic Relevance
Page 24
Comparing Relevance
24 - Results
Temporal Relevance
2 mon
ths
March ‘
13
April ‘13
May ‘1
3
June
‘13
July ‘
13
Augus
t ‘13
Page 25
Comparing Relevance
25 - Results
Topical Relevance
April 1 ‘
13
April 8 ‘
13
April 15
‘13
April 22
‘13
April 29
‘13
May 6
‘13
“police”
in
Twitte
r
#delhirape Protests
Page 26
Comparing Relevance
26 - Results
Geographic Relevance
Distribution of Twitter Users Target Country for Survey
Page 27
Comparing Relevance
27 - Results
Geographic Relevance
Distribution of Twitter Users Target Country for Survey
???
Page 28
Takeaway: Social media data is more timely than survey
data, but surveys can better target topical and geographic responses for higher quality.
- Results
Comparing Relevance
28
Page 29
Four Axes of Comparison
• Data types
• Relevance
• Cost
• Validity
29 - Results
Page 30
- Results
Comparing Cost
30
Page 31
- Results
Comparing Cost
31
$300/month
$11/month
+ $3,000/month
$3,311/month
Average one-time fee:
$70,000
Page 32
- Results
Comparing Cost
32
Page 33
Takeaway: Storing large sets of social media data,
contracting with a data reseller, and processing it at scale is significantly cheaper than
running nationally representative surveys.
- Results
Comparing Cost
33
Page 34
Four Axes of Comparison
• Data types
• Relevance
• Cost
• Validity
34 - Results
Page 35
- Results
Comparing Validity
35
Page 36
- Results
Comparing Validity
36
Page 37
Takeaway: The typical definition of validity is difficult (and perhaps undesirable) to apply to social media
data.
- Results
Comparing Validity
37
Page 38
What is the point of this talk?
38 - Conclusions
Page 39
Complements and Synergies
• Use both social media and surveys where possible
• Social media may provide first approximation when time and cost are factors
• Survey data can be of higher quality, more targeted, and can ask deeper questions
39 - Conclusions
Page 40
Cody Buntain@codybuntain
[email protected]
University of Maryland
Comparing Social Media and Traditional Surveys Around the Boston Marathon Bombing
40
#Microposts201611 April 2016
Montreal, Quebec CA
Thank you! Questions?