Logarithmic-linear model

from Wikipedia, the free encyclopedia

Logarithmic-linear models , or log-linear models for short, belong to the multivariate methods that are particularly used in statistics. Using log-linear models, nominally scaled data are analyzed. When analyzing multidimensional frequency tables, a logarithmic transformation can usually show the problem more clearly, for example in the sense that the main effects and interactions of a multidimensional frequency table can be put together linearly.

A distinction is made between different log-linear procedures:

  • As a general log-linear models is called method to investigate the non-directional connections between nominal scaled data.
  • Logit models investigate the directional relationship between a dependent nominally scaled variable and other independent variables.

Log-linear models offer the possibility of so-called saturated and unsaturated data analysis.

literature

  • A. Agresti: An introduction to categorial data analysis. Wiley, New York et al. 1996, ISBN 0-471-11338-7 , chapter 6.
  • YM Bishop, SE Fienberg, PW Holland: Discrete Multivariate Analysis. Theory and Practice. 12th edition. MIT Press, Cambridge 1995, ISBN 0-262-52040-0 .
  • B. Everitt: The Analysis of Contingency Tables. New York 1997, pp. 80-107.
  • L. Fahrmeir, A. Hamerle: Multivariate statistical methods. de Gruyter, Berlin et al. 1984, ISBN 3-11-008509-7 , chapter 10.