Many-Model Thinking

Definition

Many-model thinking is the practice of applying multiple, logically distinct formal models to the same complex phenomenon rather than relying on one. Each model is a deliberate simplification that highlights a different causal force; no single one is "correct," but together they triangulate a truer, less blind-spotted picture. As ecologist Richard Levins put it, in a line the book adopts as a motto, "our truth is the intersection of independent lies."

In the Book

Page opens the book by arguing that models are formal, tractable structures — built from mathematics and diagrams — that improve our ability to "reason, explain, design, communicate, act, predict, and explore" (Chapter 1). The case for using many of them, rather than the single best one, is made formally in Chapter 3 through two theorems: the Condorcet jury theorem, which shows that a majority vote among independently-correct classifiers approaches 100% accuracy as more classifiers are added, and the diversity prediction theorem, which decomposes crowd error into average individual error minus the diversity of predictions. Page grounds the practical case with real evidence: Google found that going from one job interviewer to four raises the odds of an above-average hire from 50% to 86%, and ensembles of economists forecasting unemployment and inflation consistently outperform even the single best-performing economist. He is also careful to show the limits — using categorization models (partitioning loan applicants by attributes like loan size and major) to demonstrate that the number of genuinely independent, accurate models you can construct is bounded by the dimensionality of your data, so "many" in practice means closer to five models than fifty.

Why It Matters

Any one model, however rigorous, embeds a specific set of simplifying assumptions and therefore has systematic blind spots — a policy built on a single economic model can miss income disparity, identity diversity, or interdependence with other systems entirely. Deliberately holding several models of the same situation side by side, and checking where their implications agree or diverge, converts modeling from a search for "the right answer" into a way of mapping the shape and size of your own uncertainty — useful whenever a decision (hiring, forecasting, policy design) can be viewed through more than one causal lens.