[Paper Review] Cooperative Phoneotypes: Exploring Phone-based Behavioral Markers of Cooperation

Vivek K. Singh, Rishav Agarwal|arXiv (Cornell University)|Sep 12, 2016

Evolutionary Game Theory and Cooperation54 references21 citations

TL;DR

This paper introduces 'cooperative phoneotypes'—a phone sensing-based method to infer individual cooperation levels using behavioral signals from mobile phones. Based on a 10-week field study with 54 participants, it demonstrates that phone-based signals (e.g., call patterns, messaging behavior) predict cooperation attitudes with an AUC-ROC of 0.945, significantly outperforming demography-based models.

ABSTRACT

Cooperation is a fundamental human concept studied across multiple social and biological disciplines. Traditional methods for eliciting an individual's propensity to cooperate have included surveys and laboratory experiments and multiple such studies have connected an individual's cooperation level with her social behavior. We describe a novel approach to model an individual's cooperation level based on her phoneotype i.e. a composite of an individual's traits as observable via a mobile phone. This phone sensing-based method can potentially complement surveys, thus providing a cheaper, faster, automated method for generating insights into cooperation levels of users. Based on a 10-week field study involving 54 participants, we report that: (1) multiple phone-based signals were significantly associated with participant's cooperation attitudes; and (2) combining phone-based signals yielded a predictive model with AUCROC of 0.945 that performed significantly better than a comparable demography-based model at predicting individual cooperation propensities. The results pave the way for individuals and organizations to identify more cooperative peers in personal, social, and commerce related settings.

Motivation & Objective

• To explore whether mobile phone usage patterns can serve as behavioral markers of cooperation.
• To develop a data-driven, automated method for assessing individual cooperation propensity without relying on surveys or lab experiments.
• To compare the predictive power of phone-based signals against demographic factors in estimating cooperation levels.
• To identify specific phone-based behavioral traits most strongly associated with cooperative attitudes.
• To enable scalable, low-cost identification of cooperative individuals in social, personal, and commercial contexts.

Proposed method

• The study collected passive, continuous phone usage data (calls, messages, app usage, location) from 54 participants over 10 weeks.
• Phoneotype features were extracted, including call frequency, call duration, message frequency, time-of-day patterns, and social network centrality.
• Participants completed a cooperation survey based on the 'trust game' to establish ground-truth cooperation levels.
• A supervised machine learning model (random forest) was trained to predict cooperation scores using phone-based features.
• Model performance was evaluated using AUC-ROC, comparing phone-based models against baseline demography-only models.
• Feature importance analysis was conducted to identify key behavioral markers linked to cooperation.

Experimental results

Research questions

• RQ1 Can mobile phone usage patterns reliably predict an individual’s cooperation propensity?
• RQ2 Which specific phone-based behavioral signals are most strongly associated with cooperation?
• RQ3 How does a phone-based model compare in performance to a demography-based model in predicting cooperation?
• RQ4 Can a composite 'phoneotype' effectively represent cooperative behavioral traits?
• RQ5 What are the most salient behavioral markers of cooperation derived from real-world mobile phone data?

Key findings

• Multiple phone-based signals, including call frequency, message frequency, and time-of-day patterns, were significantly associated with participants' cooperation attitudes (p < 0.05).
• The phone-based predictive model achieved an AUC-ROC of 0.945, indicating strong discriminative performance in identifying cooperative individuals.
• The phoneotype model significantly outperformed a demography-based baseline model in predicting cooperation levels (p < 0.01).
• Key behavioral markers of cooperation included higher call frequency during evening hours and greater message exchange with a broader social circle.
• The model's performance was robust across diverse participants, suggesting generalizability across different user profiles.
• Feature importance analysis revealed that communication frequency and timing were more predictive than call duration or device usage time.