ABSTRACT
This chapter discusses: the commonality of ideal inference conditions shared by linear regression and ANOVA; general alternative approaches that may be considered when an ideal inference condition is not satisfied; reporting results when random sampling is not an option; violation of independence of errors; disadvantages of the single summary statistic approach; references for methods that take the correlation structure of errors into account; transformations of variables; interpretations of regression coefficients when X or Y or both X and Y are log-transformed; alternative approaches that may be considered in simple and multiple linear regression when the relationship between Y and a continuous predictor is nonlinear; approximate guidelines suggesting when unequal variances may cause serious harm; alternative approaches when harmful unequal variances occur; robust statistics methodology; bootstrapping, with an illustrative example where equal variances for a simple linear regression model may not be reasonable; and alternative approaches when harmful collinearity occurs.
