Examining DETECT"s Performance in finding Dimensionally Distinctive Clusters in Comparison with Other Clustering Methods

Examining DETECT"s Performance in finding Dimensionally Distinctive Clusters in Comparison with Other Clustering Methods

  A data dimension analysis tool, DETECT, was compared with factor analysis, discrimination and classification analysis, and cluster analysis in both simulation and real data study, especially for small size sample. In simulation DETECT"s recovering rate is slightly smaller than in factor analysis. But in real data study both results resemble each other much and have acceptable meanings. Discrimination and classification analysis performs well with simulated data, but it has limit that with real data because cluster structure is latent. Results from cluster analysis were not strong enough as in other three procedures. It is contemplated that cluster analysis is" very sensitive to noise, hence partial points might be considered as a noise not as information useful. Overall both factor analysis and DETECT procedure are compatible and reliable in educational/psychometric data scored in graded manner. It is useful to refer to both procedures" results together to make analysis more reliable and valid.