Diagnostic Checking

When doing regression analysis, the possible model pitfalls are:

• The regression function is not linear. (Nonlinearity, Chapter ??)

• Error terms are not normally distributed. (Nonnormality, Chapter ?? )

• Error terms do not have constant variance. (Heteroskecasticity, Chapter ??)

• Error terms are not independent. (Autocorrelation, Chapter ??)

• There is linear dependency among the set of regressors. (Multicollinearity, Chapter ??)

• Model fits all but one or few observations (Existence of Outliers and Influential Observations, Chapter ??).

  Note: having no outliers or influential observations in the dataset is not a model assumption, however, it may affect the model fit in general.