Heteroscedastic data is a scary term when you first read it.
Yes, I know. It is an ugly word. But its meaning is far from ugly.
In essence, the concept means the data does not have equal variance. When the spread of the residuals in a regression model is not constant across different values of an independent variable (X), we will see the scatterplot of the data form a fan out, funnel in, or form a cone shape as X changes. This is heteroscedasticity.
Heteroscedasticity is when the values of the errors are not constant as the predictor (X) varies, thus the errors increase or decrease as X increases.
Example of heteroscedastic data. Image by the author.
OLS Regressions are built over some assumptions:
Normality of the residuals
Homoscedasticity of the error term
No multicollinearity
So, when the data has this kind of cone shape, running an OLS will not be the best option.…