Presented at ICWSM 2018, this research explores the intersection of computational linguistics and personality psychology. The study demonstrates that digital footprints—specifically Facebook status updates—can accurately predict an individual's "Locus of Control" (LoC) and reveal how this trait manifests through linguistic patterns.
The Psychology of Control
Psychologists define Locus of Control (LoC) as the degree to which people believe they have power over the events in their lives. "Internals" believe they are the masters of their own fate, while "Externals" attribute outcomes to outside forces like luck, fate, or powerful others. LoC is a significant predictor of health, stress, and job performance, but it has traditionally required intrusive surveys to measure.
Methodology: From Words to Models
The researchers employed a multi-step computational approach to bridge the gap between trait prediction and linguistic understanding:
- Feature Extraction: The team used three types of linguistic features: LIWC (dictionary-based psychological concepts), N-grams (word sequences), and LDA Topics (grouping words into 2,000 topics based on a corpus of 18 million updates).
- Prediction: A Gradient-Boosted Classifier was trained to distinguish between "Internals" and "Externals."
- Visualization: To interpret the language, the authors mapped LoC-linked topics onto a coordinate system where the axes represented Conscientiousness (Cognitive Control) and Emotional Stability (Emotional Control).
Predictive Performance
The study utilized a primary dataset of 2,348 Facebook users who provided access to their posts and completed standardized psychological surveys. The results showed that social media language is a powerful predictor of personality traits.
The best-performing model, utilizing a combination of N-grams and LDA Topics, achieved an F1-score of 0.82. This significantly outperformed models based solely on demographic data like age, gender, and income (F1=0.54).
Visualizing the Language of Control
By placing LoC-related language on the axes of Conscientiousness and Emotional Stability, the researchers validated long-standing psychological theories. Internal language dominates the quadrants associated with high cognitive and emotional control.
Key Linguistic Insights
- The Language of "Internals": People with an internal LoC use language emphasizing social support ("boyfriend," "fiance," "supportive"), gratitude ("blessed," "grateful"), and future planning ("saturday," "plan"). This suggests a proactive, socially connected worldview.
- The Language of "Externals": Those with an external LoC use language reflecting negative affect and emotional distress ("scars," "wounds," "confused"). They also discuss media consumption (TV, movies) more frequently, which may serve as a form of "mood management" to escape negative feelings.
- Health Correlations: LoC proved to be a better predictor of absenteeism, general health, and stress than traditional demographics. The correlation between LoC and self-reported stress was notably high at -0.53.
Implications and Future Work
While the study currently models LoC as a binary variable and relies on a specific Facebook-using population, the implications are broad. The ability to unobtrusively monitor well-being through language offers a cost-effective alternative to traditional surveys. Future work may explore clinical interventions where changing linguistic patterns—and the underlying mindsets they represent—could lead to improved health outcomes.
Summary updated