(1) Experiment often reveals well-defined interactive effects, phenomena we cannot trace to any linear composition of the known capacities of causal agents, and (2) To avoid presuppositionsabout what kind of causal agents are at work one can construct a model that mediates between: mathematical equations agnostic about the character of the phenomena described; composite, often analogical descriptions of the phenomena that can be tested to narrow down the causes; and experimental data. (1)