Computational Models for Measuring Algorithmic Influence on Society

Abstract

Algorithmic systems deployed at population scale -- recommendation engines, search algorithms, content moderation systems, and social media feed curators -- exert measurable influence on information consumption, opinion formation, political behaviour, and cultural production. Measuring this influence rigorously is essential for algorithmic accountability, democratic governance, and ethical AI deployment, yet it is methodologically challenging: algorithmic influence operates through complex, multi-step causal chains; counterfactual baselines (what would have happened without the algorithm?) are difficult to construct; and the distinction between algorithmic influence and individual preference expression is contested. This paper proposes the Algorithmic Societal Influence Measurement Framework (ASIMF), a systematic evaluation of six computational measurement approaches -- audit experiments, observational causal inference, agent-based modelling, natural language processing influence tracing, network diffusion analysis, and survey-experimental methods -- applied to four algorithmic influence domains: political opinion formation, health information consumption, consumer behaviour, and cultural discourse. ASIMF introduces the Influence Measurement Quality Score (IMQS) integrating causal validity, measurement precision, scalability, and ethical research practice. Key results: audit experiments achieve the highest IMQS (0.904) for direct algorithmic effect measurement; network diffusion analysis achieves the highest scalability (0.960); observational causal inference achieves 96.4% accuracy in replicating experimental estimates from observational social media data. The framework provides methodological selection guidance for researchers measuring algorithmic societal influence.