REGRESSION SIGNIFICANCE REFINING BY USING FICTIONAL REGRESSORS, THAT RESULT FROM WINSORIZATION OF CONDITIONAL VARIABLES Cover Image

ПРЕЦИЗИРАНЕ НА РЕГРЕСИОННАТА ЗНАЧИМОСТ ЧРЕЗ ФИКТИВНИ РЕГРЕСОРИ, ПРОДУКТ НА УИНСОРИЗАЦИЯ НА УСЛОВНИ ПРОМЕНЛИВИ
REGRESSION SIGNIFICANCE REFINING BY USING FICTIONAL REGRESSORS, THAT RESULT FROM WINSORIZATION OF CONDITIONAL VARIABLES

Author(s): Nikola Iliev
Subject(s): Economy
Published by: Стопанска академия »Д. А. Ценов«
Keywords: Confidence Region;Principal Component Analysis;Eigen-Values;Eigen-Vectors;Ellipsoid equation

Summary/Abstract: The regression relationship often faces a number of problems, including lack of statistical significance, presence of high statistical error and result uncertainty. To overcome this problems one can complicate the model, make new restrictive assumptions, each one restricting the degree of model application freedom. But the situation is different when there is a presence of extreme values, as they do not allow complication of the model or the making of assumptions. The said extreme values demand the use of an alternative methodology, like the winsorization. Developed for the needs of the preliminary data analysis, the winsorization is also applicable when using a regression relationship. Independently thereof the winsorization is characterized with a major deficiency – its use on data, assumed to be unconditional, even if said data is included in a conditional regression relationship. There’s a need for a model, describing said regression relationship, that not only considers the presence of extreme values, but also overcomes them, which in this case is through the use of fictional variables.

  • Issue Year: 2016
  • Issue No: 12
  • Page Range: 3-27
  • Page Count: 28
  • Language: Bulgarian
Toggle Accessibility Mode