In recent years much evaluation conversation has bemoaned the fact that we have been basing our work on linear models. We are dissatisfied with linear interpretations of program behavior and need to move toward nonlinear understanding. What I believe though is that almost none of our models have ever been linear. They are, however, ambiguous, both with respect to mathematical relationships and to other information that we could attach to causal paths but don’t. Often ambiguity and mathematical linearity overlap. Traditional models can show nonlinearity, but they cannot explain why it occurs or its implications for program theory. For that, an appreciation of the behavior of complex systems is needed.
