An algorithm for the estimation of pseudocanonical factors
| Autor: | Ksenija Bosnar, Franjo Prot, Konstantin Momirović, Vesna Lužar, Vesna Dobrić |
|---|---|
| Jazyk: | chorvatština |
| Rok vydání: | 1982 |
| Zdroj: | Kinesiology Volume 13. Issue 1.-2. |
| ISSN: | 1848-638X 1331-1441 |
| Popis: | Predložen je algoritam i napisan program za faktorsku analizu nekog skupa varijabli koji se osniva na relacijama kanoničnog faktorskog model (Rao, 1955), image modela (Guttman, 1953) i modela sa univerzalnom metrikom (Harris, 1962). An algorithm has been proposed and a program written for factor analysis of set of variables is based on relations of the canonical factor model (Rao, 1955), image model (Guttman, 1953) and the model with universal matrics (Harris, 1962). The algorithm and its associated program CLIMAX perform the following operations: (1) analyse latent dimensions under the component model and fix the number of dimensions in accordance with PB criterion (Štalec and Momirović); (2) determine communalities of variables by iterative procedure for the thus fixed number of factors; (3) define uniquties on the basis of the thus estimated communalities and form the reduced matrix of variable covariances re-scaled to the inverse value of unique commponents; (4) determine principal axes of the re-scaled reduced covariance matrix, fix the number of factors in accordance with the DMEAN criterion (Momirović adn Štalec, 1973) and re-scale the factor to the metrics of standardized variables; (5) rotate factors into normal varimax position (Kaiser, 1958;) (6) transform factors into promax position (Hendrickson adn White, 1964); (7) for all three solutions estimate the values of variables and latent dimensions and estimate reliability of latent dimensions (Kaiser adn Caffrey, 1965); (8) for all three solutions estimate the values of entities on latent dimensions by regression (Thurstone, 1947) and Bartlett, 1937) procedures. It has been shown that the algorithm approiximates very well the actual canonical factors defined under the canonical model of factor analysis (Rao, 1955) or under the model of maximum likelihood (Lawely, 1940; 1949; Lawely and Maxwell, 1963; Jöreskog, 1967, 1969; Jöreskog and Lawley, 1968) with considerably greater efficiency and lower sensitivity to the generalized Heywood case. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|