-

1. Trust and Recall of Information across Varying
Degrees of Title-Visualization Misalignment
Ha-Kyung Kong, Zhicheng Liu, Karrie Karahalios
roceedings of the 2019 CHI Conference on Human Factors in Computing Systems

2. 1
Main Contribution
• Previous work has revealed that visualization interpretation can be manipulated
through slanted titles [1]
• Effects of the greater degrees of misalignment between titles and visualizations
 Whether people would notice the discrepancy between title and data
visualization if there is misalignment
 How the misaligned titles would affect how people recall and trust the
information
[1] Ha-Kyung Kong, Zhicheng Liu, and Karrie Karahalios.2018.Frames and Slants in Titles of Visualizationson Controversial
Topics. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems. ACM, 438.

3. 2
Related Work
• Digital Misinformation
 Misleading information may have a detrimental influence on people’s beliefs and attitudes [2]
 Hemsley and Snyder began Exploring misinformation through data visualization [3]
• Def. of misinformation : information that is ‘objectively incorrect’
• In this paper, “information that leads to misperceptions about the facts
 Borkin et al. found that people fixated heavily on the title of a visualization [4]
 Kong et al. studied the influence of titles on the recall of the main message of a visualization and
found that slanted title can lead to take away the message opposite to visualization [1]
• Confirmation Bias and Visualization
 Confirmation bias is “the seeking or interpreting of evidence in ways that are partial to existing
beliefs, expectations, or a hypothesis in hand” [5]
 When people are asked to compose a neutral title for a visualization, some of the resulting titles
contained slants that gave undue weight to the side that matched their beliefs [1]
[1] Ha-Kyung Kong, Zhicheng Liu, and Karrie Karahalios.2018.Frames and Slants in Titles of Visualizationson Controversial Topics. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems. ACM, 438.
[2] Soroush Vosoughi, Deb Roy, and Sinan Aral. 2018.The spread of true and false news online. Science 359, 6380 (2018),1146–1151.
[3] JEFF HEMSLEY and JAIME SNYDER. 2018.Dimensionsof VisualMisinformation in the Emerging Media Landscape. Misinformation and Mass Audiences (2018), 91.
[4] Michelle A Borkin, Zoya Bylinskii,Nam Wook Kim, Constance May Bainbridge, Chelsea S Yeh, Daniel Borkin, Hanspeter Pfister, and Aude Oliva. 2016.Beyond memorability: Visualization recognition and recall. IEEE transactions on visualization and computer graphics 22, 1 (2016),519–528.
[5] Raymond S Nickerson. 1998.Confirmation bias: A ubiquitous phenomenon in many guises. Review of general psychology 2, 2 (1998),175.

4. 1
Research Questions
RQ1a) How does the message of the title influence people’s recall of the information?
RQ1b) How does the degree of misalignment between title and visualization influence
people’s recall of the information?
RQ2a) How does the misalignment of the title and the visualization influence people’s
perception of bias?
RQ2b) How does people’s perception of bias differ when the titles are consistent with
their prior attitudes on a topic verses when the titles are inconsistent with their
attitudes?
RQ3) How does the discrepancy between the message of the visualization and the
message of the title affect the perceived credibility of the data, visualization, and title?
Recall of information
Perception of bias
Perceived credibility

5. 1
3 Types of title slants
a) Title only covers one side of the story; visualization presents both [1]
b) Visualization emphasizes one side of the story; title covers the other
c) Title only covers one side; visualization only presents the other
[1] Ha-Kyung Kong, Zhicheng Liu, and Karrie Karahalios.2018.Frames and Slants in Titles of Visualizationson Controversial
Topics. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems. ACM, 438.

6. 5
Visualization-title pairs
• Miscued slant condition
 E.g. Visually cued increasing trend (decreasing trend is in a
dotted pattern) & title with decreasing trend highlighted
• Contradictory slant conditions
 E.g. Visualization only with increasing trend & title
mentioning decreasing trend
• Total number of pairs = 8
 2 visualizations (military, refugee) x 2 conditions (miscued,
contradictory) x 2 titles (supporting, non-supporting)
• Each condition with 100 participants
Supporting titles Non-supporting titles

7. 6
Experiment Procedure
1. Pre-experiment attitude and demographic surveys
- Attitudes on controversial issues
2. Preview of the visualizations and factual questions
- 3 factual questions for miscued condition
- 2 factual questions for contradictory condition (1 visualization)
- Purpose of factual questions is to test the comprehension of the visualization and as a distractor task before recall stage
3. Recall question on the main message
- Open-ended question on the main message of the information
4. Consistency and impartiality questions
- Post-experiment attitude questions
- Participants’ perceptions of bias
5. Title recall stage
6. Credibility stage
- Based on the Meyer modification of the Gaziano-McGrath scales [6]
- Level of agreeing (in 7 scale) on the statement regarding five factors -- accuracy, fairness, trustworthiness, bias, and completeness
[6] Cecilie Gaziano and Kristin McGrath. 1986.Measuring the concept of credibility. Journalism quarterly 63, 3 (1986),451–462.

8. 7
Results
• The general topic recalled more than specific messages in
title and visualization (figure 1)
• Titles influence recalled messages
• RQ1a) How does the message of the title influence people’s recall
of the information? (figure 2)
 In refugee issue with supporting title,
• category low is aligned with the message of the title
• Number is aligned with the message of the visualization
 In military issue,
• Recalled message is less dramatic
• Degree of misalignment influences recalled message
• RQ1b) How does the degree of misalignment between title and
visualization influence people’s recall of the information?
 Significant difference between two conditions
• More likely to refer to the general topic in miscued condition
• More likely to rely on the visualization for the contradictory condition
figure 1
figure 2
low more
decrease increase

9. 8
Results
• Low awareness of bias despite misalignment
• RQ2a) How does the misalignment of the title and the
visualization influence people’s perception of bias?
 Majority (72-87%) reported that the information was
neutral regardless of the blatant slant
• Possible reason : presenting facts and statistics / not having
enough prior knowledge
 In military issue (line chart), the number of “biased”
responses in contradictory slant condition (28%) is
significantly higher than in miscued slant condition
• Possible reason: people’s aptitude in perceiving the trend in a
line chart make then to detect easier than checking the
number in bar chart
 The importance of the issue significantly predicted
the perception of bias
• Attitude-consistency influences the perceived bias
• RQ2b) How is attitude-consistency of the title associated with
perception of bias?
 The perceived bias for refugee visualization was
influenced more by the attitude-consistency of the title
than its misalignment with the visualization
 People were more likely to report as very biased or
biased if the title was inconsistent with their belief than
when it was consistent

10. 9
Results
• Perceived credibility varies for data, visualization & title
• RQ3) How does the discrepancy between the message of the
visualization and the message of the title affect the perceived
credibility of the data, visualization, and title? How is attitude-
consistency associated with perception of credibility?
 Alignment of title and visualization has a significant
impact on the credibility for both topics
 Less credible when there was bigger visualization-title
misalignment (contradictory < miscued)
 Credibility scores were lower when the title was attitude-
inconsistent compared to when they were attitude-
consistent or irrelevant, but not statistically significant
 Credibility scores differed statistically significantly
between data, visualization, and title for the refugee
visualization and the military visualization
 Data was perceived as the most credible, followed by the
visualization then the title
 When misalignment increases, the credibility of the title
suffered the most
figure 3
figure 4

11. 10
Discussion
• Trust and Confirmation Bias
 Perceived credibility and impartiality of the information significantly decreased when the title was
misaligned with the participants’ existing attitudes in both conditions
 Reveals a confirmation bias in interpreting and receiving information from visualization
 People are more likely to dismiss the information as less credible when the title is attitude-inconsistent
 Confirmation bias can lead some people to cherry pick the part of the visualization that matches their
attitude when the visualization and title convey opposing message
 People’s trust and the impact of the title decrease when there is a great discrepancy between the title and
the visualization
• The role of text on visual-based misinformation
 Titles have a stronger manipulation effect
 Future works:
 Different ways social network platforms can detect and combat the spread of visual-based misinformation
• Detecting clickbait titles and misleading headlines
• Provide additional information for raising awareness
 Investigating the effect of text by removing the contextual information

12. 11
Limitation
• Time gap with the previous work (selective slant condition)
• The climate of digital information on controversial issues has changed
• Only two types of visualization
• No baseline that the title and the visualization are aligned
 Unable to measure the exact impact of visualization-title mismatch
• Seem redundant to separate bias and credibility check

13. 12
Future Work
• Influence of textual components varies by chart types and the visual complexity
 Check whether a confirmation bias can cause from the image of a figure
 Compare the impact of title with the impact of non-visualization image
• Good approach to use
 the discrepancy between title and visualization, and
 the attitude-consistency as parameters
• Contain baseline to measure the exact degree