Heteroskedasticity in linear regression refers to unequal variance among errors and observations of the dependent variable. It is not related to a skewed distribution, significant p-values, correlation among observations, or perfect correlation among predictors.