Previous Next

Regression Model Analysis - Constant Variance (3)

By assumption, the random error terms all have the same variance and are uncorrelated with each other.  This property is known as homoskedasticity.   When the residual error term changes for different values in the regression (the error is not constant), the error terms as said to be heteroskedastic.  Violation of this assumption is usually due to either the time effect or the megaphone effect.   When heterosckeasticity is present, the regression paramters are no longer the minimum variance estimators.  Further, it is generally in error to rely upon the estimated parameter variances as well as the residual sum of squares.  ORS detects this problem via White’s Test for Homoskedasticity for homoskedasticity.  This test is an asymptotic test; hence it gains power as the sample size increases.


Previous Next