A classroom version of Nosofsky (1991), Experiment 1A. Learn two families of schematic faces, classify new faces, then see how the exemplar model (the generalized context model) predicts your choices from similarity.
These ten faces belong to two families. Look at them carefully: what makes a Family A face, and what makes a Family B face? Next you will see new faces without names and decide which family each belongs to.
The faces differ on four features: eye height, eye separation, nose length and mouth height. There is no single feature that perfectly separates the families.
You will see 34 faces, one at a time. Some you have seen, most are new. There is no feedback; go with your impression.
Click a button or press F for Family A and J for Family B.
Proportion of "Family A" responses for faces grouped by how similar they are to Family A, relative to Family B.
Studied faces are scored against their true family. New faces have no true family, so they are scored against the family they are more similar to.
Relative similarity = Σ similarity to Family A ÷ (Σ to A + Σ to B), computed by the exemplar model with parameters fitted to Nosofsky's participants. Gray points are his 1991 group data (Experiment 1A, all participants). Step 4 shows how this number is computed.
The exemplar model stores every studied face. A new face is compared with each one, feature by feature. Pick a face to follow the calculation.
The model uses psychological coordinates, derived by Nosofsky from people's similarity ratings of the faces (multidimensional scaling), not raw millimetres.
Attention weights are rescaled to sum to 1 (shown on the right).
For each of the 10 family members, take the difference on each feature. Click a row to see its calculation worked out in steps 2 and 3.
The differences are combined as a weighted Euclidean distance. Features that get more attention count more.
Similarity decays exponentially with distance. A larger c makes similarity fall off faster, so only very close family members count.
Each dot is one studied face, placed at its distance from the face you picked. Blue dots are Family A, orange dots are Family B.
Each segment is one family member's contribution.
Each point is one face: the model's P(A) with the current parameters against the proportion of Nosofsky's participants who chose Family A. Move the sliders and watch the fit change.
With the fitted parameters (c ≈ 3.2; attention mostly on eye height and nose length) the model explains about 95% of the variance in people's choices. Try putting all attention on mouth height to see the fit collapse.