There are a number of underlying assumptions about the dependent and independent
variables, and their relationship, which affect the statistical procedures such as LSE and significance
tests used for linear regression. These assumptions, listed here, need to be tested at the different
stages of the regression process.
Normality: Variables and their combination are assumed to follow the normal
distribution.
Linearity is assumed.
Homoscedasticity: Variance of dependent variable should not vary across range of
predictor variables.
Residuals (errors, i.e., predicted minus observed values) are assumed to be
independent. The prediction errors should be uncorrelated, otherwise it suggests some unexplained,
systematic relationship in the dependent variable.
