Heteroskedasticity
Error variance that differs across observations, breaking the regression assumption of constant variance. It makes the reported standard errors unreliable, so significance tests can no longer be trusted.
Where you will meet it
Heteroskedasticity appears in 6 learning outcomes on the CFA syllabus. A few of them:
- Set out what a simple linear regression assumes
“Violations have names and consequences — heteroskedasticity is non-constant error variance, serial correlation is dependence between successive errors — and both can leave the fitted coefficients looking more precise than they are, which makes significance tests unreliable rather than merely imperfect.”
- Explain what a multiple linear regression takes for granted and read residual plots for signs that those conditions fail
“Plot residuals against each independent variable and against predicted values: a curved pattern suggests non-linearity, and a fan shape that widens suggests heteroskedasticity.”
- Explain how a wrongly specified model distorts regression output and how the usual specification errors can be prevented
“Inappropriate scaling can create heteroskedasticity or multicollinearity.”
- Set out the kinds of heteroskedasticity and their consequences for statistical inference
“Heteroskedasticity means the error variance is not constant.”
See also
Practise it
Practise heteroskedasticity until it is automatic: questions that explain every wrong option.
Take the free diagnostic →