Robust Multiple Linear Regression.
AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH SCHOOL OF ENGINEERING
Pagination or Media Count:
An extensive Monte Carlo analysis is conducted to determine the performance of robust linear regression techniques with and without outliers. Thirteen methods of regression are compared including least squares and minimum absolute deviation. The classical robust techniques of Huber, Hampel were studied and robust techniques using the Q-statistic as a discriminant were introduced. The model studied contained eleven variables with 27 observations. The error distributions considered were uniformly normally, double exponentially distributed. Least squares gave the best fit without outliers. In the presence of gross outliers a rejection of outliers technique gave the best fit. Author
- Statistics and Probability