This paper presents a new approach to predictive analysis for social processes. A key element of the proposed methodology is proper characterization of the interplay between the intrinsic aspects of a social process (e.g., the persuasiveness of an argument) and the social dynamics which is its realization (e.g., the way the argument propagates through a segment of society). We show that this interplay can be modeled within a novel multi-scale framework that is sociologically and mathematically sensible, expressive, illuminating, and amenable to formal analysis. We then develop a scientifically rigorous, computationally tractable approach to predictive analysis. Among other capabilities, this analytic approach enables assessment of process predictability, identification of measurables which have predictive power, discovery of reliable early indicators for events of interest, and scalable, robust prediction. The potential of the proposed approach is illustrated through a case study involving early warning analysis for mobilization/protest events.