UC Berkeley

Through decades of research, scholars have illustrated the complexity of the relationship between unobservable personalities and observable behaviors. Yet, despite this complexity, the predominant empirical method of assessing personality remains the psychometric scale. In this dissertation, I extend past research on behavioral measures of personality by proposing a novel approach using natural language, and highlight at least three key ways it helps advance our understanding of personality and behavior. First, assessing personality from behavior directly bypasses potential cognitive and perceptual biases involved in the self- and peer-report of scales, allowing for a more objective account. Second, it uncovers temporal and situational variance largely obscured by scales, which capture mean tendencies and more stable, individual differences. Third, it has greater potential to shed light on the processes underlying documented correlations between personality and distal outcomes.

In Chapter 1, I review the personality and behavior literature and draw out two major conclusions, describing how they have led to three different methods of personality assessment. I discuss the limitations of these methods, explain how my approach overcomes these limitations, and introduce the personality construct of self-monitoring through which I illustrate my approach. In Chapter 2, I detail the construction of a behavioral measure of self-monitoring. I explain how characteristics of the technical apparatus map on to the general theory of self-monitoring, outline the design decisions I made in constructing the measure, and provide comparisons with the survey measure as well as robustness checks. In Chapter 3, I conclude with an empirical analysis of the measure, demonstrating that it is responsive to situations as expected. I additionally demonstrate its utility in its association with consequential individual outcomes such as salary bonus and network constraint.