[Paper Review] Using Processing Fluency as a Metric of Trust in Scatterplot Visualizations

Hamza Elhamdadi, Lace Padilla+1 more|arXiv (Cornell University)|Sep 28, 2022

Data Visualization and Analytics4 citations

TL;DR

This paper proposes processing fluency—measured by speed and accuracy of perception—as a reliable proxy for trust in scatterplot visualizations. Through two empirical studies, it demonstrates that camouflaged visualizations (e.g., blurred, overlapped, or scaled marks) reduce processing fluency, leading to lower accuracy in data estimation and reduced trust in decision-making tasks, establishing fluency as a valid, objective metric for trust in data visualization.

ABSTRACT

Establishing trust with readers is an important first step in visual data communication. But what makes a visualization trustworthy? Psychology and behavioral economics research has found processing fluency (i.e., speed and accuracy of perceiving and processing a stimulus) is central to perceived trust. We examine the association between processing fluency and trust in visualizations through two empirical studies. In Experiment 1, we tested the effect of camouflaging a visualization on processing fluency. Participants estimated the proportion of data values within a specified range for six camouflaged visualizations and one non-camouflaged control; they also reported their perceived difficulty for each of the visualizations. Camouflaged visualizations produced less accurate estimations compared to the control. In Experiment 2, we created a decision task based on trust games adapted from behavioral economics. We asked participants to invest money in two hypothetical companies and report how much they trust each company. One company communicates its strategy with a camouflaged visualization, the other with a controlled visualization. Participants tended to invest less money in the company presenting a camouflaged visualization. Hence, we found support for the hypothesis that processing fluency is key to the perception of trust in visual data communication.

Motivation & Objective

• To investigate whether processing fluency—defined as the ease and accuracy of perceiving visualizations—can serve as a reliable proxy for trust in data visualization.
• To identify design manipulations that reduce perceptual clarity (i.e., camouflage techniques) and assess their impact on processing fluency and trust.
• To evaluate the validity of processing fluency as a more objective alternative to self-report trust metrics in visualization research.
• To explore the potential of using behavioral economics methods, such as trust games, to measure trust in visual data communication.
• To provide evidence-based design guidelines that enhance perceptual fluency and, by extension, trustworthiness in visualizations.

Proposed method

• Conducted Experiment 1 with six camouflaged scatterplots and one non-camouflaged control, measuring participants' accuracy in estimating data proportions within a range.
• Collected self-reported perceived difficulty to assess subjective processing fluency across visualization conditions.
• Designed a trust game-based decision task in Experiment 2, where participants invested money in two hypothetical companies—one using a camouflaged visualization, the other a clear control.
• Used investment amount as an objective behavioral measure of trust, with higher investment indicating greater perceived trust.
• Applied regression models to analyze the relationship between camouflage intensity, processing fluency, and trust outcomes.
• Evaluated the impact of individual differences (e.g., age, literacy) on trust perception through covariate-controlled analyses.

Experimental results

Research questions

• RQ1 Does camouflaging visual elements in scatterplots reduce processing fluency, as measured by estimation accuracy and perceived difficulty?
• RQ2 To what extent does reduced processing fluency lead to lower perceived trust in visualizations, as measured by behavioral decisions in a trust game?
• RQ3 Can processing fluency serve as a valid, objective proxy for trust in data visualization, surpassing self-report Likert scales?
• RQ4 How do specific camouflage techniques (e.g., blurring, overlapping, gridlines) differentially affect processing fluency and trust?
• RQ5 Do demographic factors such as age, literacy, or gender moderate the relationship between processing fluency and trust perception?

Key findings

• Camouflaged visualizations led to significantly less accurate data estimations compared to the non-camouflaged control in Experiment 1, indicating reduced processing fluency.
• Participants reported higher perceived difficulty for camouflaged visualizations, confirming that design manipulations impaired perceptual clarity.
• In the trust game, participants invested significantly less money in the company using a camouflaged visualization, demonstrating lower behavioral trust.
• The results support the hypothesis that processing fluency is a key driver of perceived trust in visual data communication.
• Camouflage techniques such as blurring, increased opacity, and scale manipulation were effective in reducing both fluency and trust.
• The study provides empirical support for using processing fluency as a more objective, behaviorally grounded metric of trust than self-report scales.