Behavioral consistency in the digital age: The activity profile of our smartphones reveals a lot about us

Behavioral Consistency in the Digital Age. Heather Shaw et al. Psychological Science, February 17, https://doi.org/10.1177/09567976211040491

Abstract: Efforts to infer personality from digital footprints have focused on behavioral stability at the trait level without considering situational dependency. We repeated a classic study of intraindividual consistency with secondary data (five data sets) containing 28,692 days of smartphone usage from 780 people. Using per-app measures of pickup frequency and usage duration, we found that profiles of daily smartphone usage were significantly more consistent when taken from the same user than from different users (d > 1.46). Random-forest models trained on 6 days of behavior identified each of the 780 users in test data with 35.8% accuracy for pickup frequency and 38.5% accuracy for duration frequency. This increased to 73.5% and 75.3%, respectively, when success was taken as the user appearing in the top 10 predictions (i.e., top 1%). Thus, situation-dependent stability in behavior is present in our digital lives, and its uniqueness provides both opportunities and risks to privacy.

Keywords: behavioral consistency, personality, digital footprint, intraindividual, open data, preregistered

Discussion

It has been almost five decades since Mischel (1973) outlined an interactionist conception of behavioral dispositions, yet most evidence for the theory comes from observations of off-line interactions. Here, we considered consistency in digital behaviors, through studying the variation of engagement (a behavior) across several nominal situations (apps), collected unobtrusively every second across several days. We found that smartphone users have unique patterns of behaviors for 21 different apps and the cues they present to the user. These usage profiles showed a degree of intraindividual consistency over repeated daily observations that was far greater than equivalent interindividual comparisons (e.g., a person consistently uses Facebook the most and Calculator the least every day). This was true for the daily duration of app use but also the simpler measure of daily app pickups—how many times you open each app per day. It was also true for profiles derived from individual days and profiles aggregated across multiple days. Therefore, by adopting an interactionist approach in personality research, we can predict a person’s future behavior from digital traces while mapping the unique characteristics of a particular individual. Research indicates that people spend on average 4 hr per day on their smartphone and pick up their smartphone on average 85 times per day (Ellis et al., 2019). It is important that theories can adapt to the way people behave presently in digital environments.

It may be considered a limitation that when examining if-then statements, we did not examine within-app behaviors (e.g., posts and comments) that result from experiencing the active ingredients of a particular digital situation. In future studies, researchers may wish to explore data that can be retrieved from different apps that share similar behaviors (e.g., posts across different social media sites). Instead, we examined the cross-situational engagement (a behavior) with each app (situation), which is a comparatively simple digital trace that can be collected easily and unobtrusively, to demonstrate that this alone has within-user consistency.

Consequently, the extent to which our daily smartphone use could act as a digital fingerprint, sufficient to betray our privacy in anonymized data or across devices (e.g., personal phone vs. work phone), is an increasing ethical concern. Our study adds value to the existing literature by illustrating how engagement with apps alone shows within-user consistency that can identify an individual. We modeled users’ unique behaviors by training random forests and then used their exported predictions to assign them to a top-10 candidate pool in separate data with 75.25% accuracy. Thus, an app that is granted access to a smartphone’s standard activity logging could render a reasonable prediction about a user’s identity even when they are logged out of their account. Similarly, if an app receives usage data from several third-party apps, our findings show that this can be used to profile a user and provide a signature that is separate from the device ID or username. So, for example, a law enforcement investigation to identify a criminal’s new phone from knowledge of their historic phone use could reduce a candidate pool of approximately 1,000 phones to 10 phones, with a 25% risk of missing them.

Pertinently, this identification is possible with no monitoring of the conversations or behaviors within the apps themselves and without triangulation of other data, such as geo-location. Perhaps this should come as no surprise. It is consistent with other research that shows how simple metadata can be used to make inferences about a particular user, such as assessing their personality from the smartphone operating system used (Shaw et al., 2016) and determining their home location from sparse call logs (Mayer et al., 2016), as well as identifying a particular user from installed apps (Tu et al., 2018). Given that many websites and apps collect these metadata from their users, it is important to acknowledge that usage alone can be sufficient to identify a user. It underscores the need for researchers collecting digital-trace data to ensure that usage profiles cannot be reverse engineered to determine participants’ identities, particularly if data are to be shared widely. Thus, context-dependent intraindividual stability in behavior extends into our digital lives, and its uniqueness affords both opportunities and risks.