When heteroscedasticity is detected, we sometimes use a weighted regression in which the dependent and independent variables are weighted by the variances of their error terms. Thus, the estimated regression becomes y_i/s_e = β₁x₁/s_e + β₂x₂/s_e + e_i/s_e, where se is the standard error of residuals. Explain intuitively why this may produce better regression results.