TL;DR
Asking survey participants for permission to track their mouse movements significantly reduces their willingness to participate. However, researchers face a strategic trade-off: "bundling" mouse-tracking into the main consent form yields more data overall, while making it a separate "opt-in" preserves higher general participation rates at the cost of less tracking data.
What Problem Does This Paper Address?
In the world of web surveys, researchers don't just care about what you answer, but how you answer. This "data about the data" is called paradata. Mouse-tracking is a powerful form of paradata that can reveal if a user is struggling with a question, hesitating between options, or finding the interface confusing.
The catch? Tracking a cursor feels significantly more invasive than simply recording a final click. It’s the digital equivalent of an interviewer watching over your shoulder as you fill out a form. While previous research has looked at attitudes toward recording keystrokes or timestamps, this paper addresses the specific privacy hurdles of mouse-tracking. Specifically, the authors investigate how the structure of the consent request and the stated motivation for tracking affect a person's willingness to join the study.
Methods
The researchers conducted a vignette experiment with a large sample of the German public. A vignette study asks participants to imagine a hypothetical scenario—in this case, receiving an email invitation to a 15-minute social/political survey for a 5€ reward.
The experiment manipulated two main factors across 10 different conditions:
- Consent Structure:
- Joint Consent: Mouse-tracking is presented as a mandatory part of the study from the start.
- Separate Consent: Participants first decide if they want to do the survey, and then (on a second screen) decide if they will allow mouse-tracking.
- Motivation/Justification: The researchers tested five levels of "why," ranging from no reason at all to specific scientific goals (e.g., "to better understand your responses" or "to improve the survey").
Willingness was measured on an 11-point scale (0–10), and the results were analyzed using linear and logistic regression models.
Datasets
The study utilized a sample of 1,498 members of the German public (750 female, 748 male, ages 18–69). The participants were recruited via the panel provider Respondi AG in July 2019. While this isn't a machine learning paper using a benchmark dataset, it provides a robust empirical snapshot of privacy attitudes in a major European demographic.
Key Insights and Ideas
- The "Privacy Tax" is Real: Simply mentioning mouse-tracking acts as a deterrent. Willingness to participate dropped significantly when tracking was introduced upfront.
- The Practitioner's Dilemma: There is no "perfect" way to ask.
- If you want the highest number of total participants, use a separate opt-in. You'll lose fewer people at the front door, but many will decline the tracking later.
- If you want the most tracking data possible, use joint consent. You'll scare more people away initially, but everyone who stays provides the data you need.
- The Justification Paradox: Intuitively, we think that explaining why we need data increases trust. This study found the opposite. Providing reasons for paradata collection did not help, and in some cases (like saying the data was for "improving the survey"), it actually decreased willingness. This suggests that detailed justifications might inadvertently signal to users that the data collection is more significant or "riskier" than they initially thought.
Key Results
- Participation Drop: In the joint consent condition, willingness to participate was 75%, compared to 86% in the control group (where mouse-tracking wasn't mentioned).
- Data Yield: When participants were given a separate opt-in for tracking, only 62% of the total sample ended up being willing to both take the survey and be tracked. This is significantly lower than the 75% yield from the joint consent approach.
- Motivation Impact: Descriptively, providing no reason for tracking actually led to the highest willingness to participate. Contrasting different reasons showed that "improving the survey" was viewed more negatively than "better understanding responses."
Limitations and Future Work
The primary limitation is the hypothetical nature of the vignette. People often say they care about privacy in surveys but then behave differently when faced with a real task and a real reward (the "Privacy Paradox"). The study also used a fixed 5€ incentive; it’s unclear if higher rewards would mitigate the privacy concerns.
For future work, the authors suggest exploring privacy-preserving paradata. Instead of recording and saving raw cursor coordinates to a server, researchers could use client-side processing to extract only specific features (like "total distance traveled") and delete the raw movement data immediately. This "privacy by design" approach might be the key to gathering rich insights without spooking the respondents.
Summary updated