Constant Error Variance

The Gauss-Markov theorem assumes that the error variance should remain constant across observations. When this assumption is violated, we have nonconstant error variance, also referred to as heteroskedastic error variance. In the presence of nonconstant error variance, OLS parameter estimates remain unbiased and consistent, but the standard errors of regression coefficients are estimated incorrectly, and therefore, the t-tests using the standard error estimates become invalid.

Residuals versus fitted values plot