An Automated Technology-Based Personality Classifier

University Of Texas At Austin, Austin TX

Investigators

Abstract

Individuals differ from each other in stable ways that have important implications for their prosperity, health, and welfare. Personality traits predict numerous consequential outcomes in the economic, social, and health domains, such as work performance, relationship quality, and the likelihood of getting sick. Despite personality's demonstrated importance, the predominant technology for assessing personality-- "self-report questionnaires"--has remained virtually unchanged over time. These self-reports are subject to an array of limitations, such as being disruptive, time consuming, and vulnerable to memory biases; these limitations potentially undermine the validity of personality questionnaires. Progress in the field of personality assessment has been constrained by the fact that the everyday behaviors and language through which personality is expressed have been challenging to measure directly in the natural stream of daily life. The few studies that have collected objective measures of behavior have typically done so on a very limited range of behaviors or within the artificial confines of a laboratory. However, the advent of smartphones and their ubiquity in modern life offer the promise of revolutionizing the field of personality assessment. The present research will use a smartphone that will automatically and unobtrusively measure personality as it is expressed in daily life. The app will use embedded sensors (e.g., accelerometer, light sensor, microphone, GPS) to gather behavioral (e.g., activity type, sleep patterns, sociability, location) data from participants. The research will collect data from up to 2000 participants that is broad-based (many kinds of behavior), fine-grained (many assessments per hour), longitudinal (many weeks of continuous behavioral data), and context tagged. These data will have two primary uses. First, they will provide large-scale objective records of how behavior unfolds in the context of everyday life, allowing researchers to learn what kinds of behavior tend to co-occur in every life and what kinds of temporal patterns they follow. Second, the data will be used to generate an unobtrusive automated method for measuring personality via smartphones in everyday lives. Using such data, the principal investigator will determine whether the standard five-factor personality model adequately captures the structure of real-world behaviors or whether a new bottom-up, empirically derived revision of personality structure is needed. Methodologically, the project will yield software, analysis tools, and classifiers that allow researchers to move beyond their reliance on self-reports and artificially constrained lab-based proxies of real-world behavior. The study will advance and validate sensing techniques used to infer complex behaviors (e.g., situations, conversation contribution) from continuous steams of sensor data, and the resulting software and analyses tools will be made available for researchers to use.