ABSTRACT
Much like bivariate least squares, multiple least squares summarize linear relationships between independent variables and a single dependent variable. In multivariate models where we are dealing with nth dimensional relationships, we have to use a surrogate to evaluate heteroskedasticity. That surrogate is a plot of the residuals. Heteroskedasticity only nullifies the hypothesis tests of those coefficients. The downside to heteroskedasticity is that while we can model nonlinearity, heteroskedasticity is a function of the data—that is to say that data are naturally distributed with uneven variance, and often there is not much that can be done about it. This most commonly occurs in data that are collected across time as opposed to cross-sectional data; however, it may also occur in a spatial distribution as well. The correlation matrix will identify most potential problems with multicollinearity, while the bivariate scatter plots will help identify the functional forms of the relationships between the dependent variable and independent variables.
