Diagnostics validate the model assumptions and prevent wrong conclusions when interpreting the result and conducting statistical inference. They include measures of goodness of fit and various tests of the assumptions about the error term, including how closely the residuals match a normal distribution. Furthermore, diagnostics test whether the residual variance is indeed constant or exhibits heteroskedasticity, and if the errors are conditionally uncorrelated or exhibit serial correlation, that is, if knowing one error helps to predict consecutive errors.

In addition to the tests outlined as follows, it is always important to visually inspect the residuals to detect whether there are systematic patterns because these indicate that the model is missing one or more factors that drive the outcome.