Investigating How Music Affects Persuasion, Engagement, and Emotion in Data Videos

TL;DR

The paper tackles whether background music modulates viewers' experience of data videos along three dimensions: persuasion, engagement, and emotion. Using a preregistered between-subjects design with three conditions (no music, default music, custom music) and a single data video about the global water crisis, it finds robust evidence that default music increases attitudinal persuasion, while custom music yields mixed results and does not enhance engagement; surprisingly, no-music conditions can induce stronger self-reported emotions. Methodologically, it combines structured data-video selection, careful measurement scales, and preregistered analyses, contributing new insights into audio-visual design for data-driven narratives. Practically, the results suggest designers may favor generic default music to boost persuasion, but must contend with potential engagement and emotional effects and the variability introduced by bespoke music. Overall, the work advances understanding of audio-visual interactions in data visualization and sets the stage for more nuanced investigations into music’s role in data storytelling.

Abstract

Data videos have become a prominent vessel for communicating data to broad audiences, and a common object of study in information visualization. Many of these videos include music, yet the impact of music on how people experience data videos remains largely unexplored. We conducted a preregistered study into the effect of music across three dimensions: persuasion, engagement, and emotion. We showed online participants an existing data video (1) without any music, (2) with its generic default music, and (3) with custom music designed by a professional composer. We found that the default music helped make the data video more persuasive. However, the effects of custom music were more mixed, and we did not find that music increased engagement. In addition, and contrary to our expectations, our participants reported more intense emotions without music. Our study contributes new insights into the intersection of music and data visualization and is a first step toward guiding designers in creating impactful data-driven narratives.

Figures (9)

• Figure 1: Emotion categorization in different messages
• Figure 2: (a) Mean attitude change for each music condition. Attitude change is normalized such that a score of 0 indicates no change and a score of -1 or 1 indicates the maximum change possible. (b) Pairwise differences in mean attitude change between conditions (right condition minus left condition). Error bars are 95% confidence intervals.
• Figure 3: (a) Mean engagement score for each music condition. Engagement scores are normalized between -1 and 1, with a score of 0 indicating a neutral response to the engagement questions. (b) Pairwise differences in mean engagement between conditions (right condition minus left condition). Error bars are 95% confidence intervals.
• Figure 4: (a) Mean emotion intensity for each music condition. Emotion intensities are normalized between 0 and 1, with a score of 0 indicating no emotion, a score of 0.5 indicating moderate intensity, and a score of 1 indicating extreme intensity. (b) Pairwise differences (right condition minus left condition). Error bars are 95% confidence intervals.
• Figure 5: Proportion of participants who found that a particular message (among M1--M5) was the best communicated. Error bars are 95% CIs.