Contrast Model of Similarity

Definition

People assess similarity not by mapping objects into a multidimensional space and measuring distance, but by comparing their features. Similarity between object A and object B depends on three components: their common features (shared attributes), A's distinctive features (what A has that B lacks), and B's distinctive features (what B has that A lacks). Crucially, "A is like B" is not equivalent to "B is like A"—similarity is fundamentally directional and asymmetric.

In the Book

Tversky directly challenges the dominant geometric approach to similarity, which represented objects as points in coordinate space with distance corresponding to similarity. He demonstrates empirically that this model fails because it assumes symmetry (distance from A to B equals distance from B to A), but people systematically judge "North Korea is like Red China" as more similar than "Red China is like North Korea." The direction depends on salience: people use the more salient, prototypical, or prominent stimulus as the referent (the "B" in the comparison).

The contrast model explains this asymmetry through the diagnosticity principle: features matter more for similarity when they distinguish objects. In one striking experiment, Tversky showed that when Austria appears with Sweden and Hungary, it clusters with one group; when Poland (instead of Norway) appears in the set, Austria's similarity to Hungary increases substantially—simply because the context changed what features became diagnostic for classification. The model also explains the non-transitivity of similarity: Jamaica is similar to Cuba (geographic proximity), Cuba is similar to Russia (political affinity), yet Jamaica and Russia are not similar at all.

Why It Matters

This concept breaks apart the assumption that comparison is symmetric and context-independent, revealing that similarity judgments are constructed, not retrieved. The insight has immediate implications for how we classify things (similarity changes with what else is in the reference set), how metaphor works (direction matters—we say "this is like a war," not "a war is like this"), and how prototypes function in categories. It shows that psychological similarity doesn't follow the mathematical rules we often assume it should, making it crucial for understanding how people actually organize knowledge and make decisions.