ERIC Number: ED210276
Record Type: RIE
Publication Date: 1981-Apr
Reference Count: 0
Empirical Identification of Hierarchies.
McCormick, Douglas; And Others
Outlining a cluster procedure which maximizes specific criteria while building scales from binary measures using a sequential, agglomerative, overlapping, non-hierarchic method results in indices giving truer results than exploratory facotr analyses or multidimensional scaling. In a series of eleven figures, patterns within cluster histories reveal the structure of the data. If true clusters exist in the data, one way they reveal themselves is by a sharp drop in the index values as an item outside the true cluster is added. In spatial terms, this represents a "moat" surrounding the cluster; a low region of density between regions of higher density which are the clusters themselves. A series of analyses were conducted using artificial data which had a known cluster structure. The Birnbaum test model was used to produce unidimensional scales of three sizes, which were combined with six outliners to make the raw data for analysis. Means, variances, and distribution shapes were varied for the Birnbaum parameters of difficulty, ability and discrimination. (Author/CE)
Publication Type: Speeches/Meeting Papers; Reports - Research; Reports - Descriptive
Education Level: N/A
Sponsor: Department of Justice, Washington, DC. National Inst. of Justice.
Authoring Institution: N/A
Identifiers: Empirical Analysis; Hierarchical Cluster Analysis; Matrix Operations; Order Analysis
Note: Paper presented at the Annual Meeting of the American Educational Research Association (65th, Los Angeles, CA, April 13-17, 1981). Small print in figures.