PDF pending restoration
ERIC Number: ED191922
Record Type: Non-Journal
Publication Date: 1979-Jan
Reference Count: N/A
Some Supporting Evidence for Accurate Multivariate Perceptions with Chernoff Faces, Project 547.
A scheme, using features in a cartoon-like human face to represent variables, is tested as to its ability to graphically depict multivariate data. A factor analysis of Harman's "24 Psychological Tests" was performed and yielded four orthogonal factors. Nose width represented the loading on Factor 1; eye size on Factor 2; curve of mouth on Factor 3; and height of eyes on Factor 4. The communality (i.e. the proportions of the total variance accounted for by the factor space) was represented by the size (area) of the face. Ten "experts" who knew about Chernoff faces, factor analysis, and Harman's 24 Psychological Tests, and 32 randomly selected "novices" sorted 24 squares depicting "faces" into as many piles as they thought sensible. Each subject's responses were coded, and group similarity matrices were constructed for the experts and the novices by averaging across their individual matrices. Results indicated that, for a four dimensional structure, untrained human judges performed with astonishing accuracy. "Faces" not only allowed the clustering of multivariate data points perceptually, but the distances between points as well. (RL)
Descriptors: Cartoons, Factor Analysis, Higher Education, Multivariate Analysis, Perception Tests, Psychometrics, Visual Perception
Bureau of Social Science Research, Inc., 1990 M St., N.W., Washington, DC 20036 ($1.60).
Publication Type: Reports - Research
Education Level: N/A
Sponsor: National Science Foundation, Washington, DC.
Authoring Institution: Bureau of Social Science Research, Inc., Washington, DC.