- An important concept in multiple regression is multicolinearity. Multiple regression is the extent to which predictors correlate with each other. This should be minimized. Multicolinearity can be assessed with VIF: the Variance Inflation Factor (that is, the extent to which predictors correlate).
- Compared to a standard regression, for multiple regression, we add one of two different kinds of assumptions. Which of these applies depends on the nature of the variables.
- For random variables: multivariate normality of the joint distribution of Y, X1, X2, ..., Xp.
- For fixed variables: the conditional distribution of Y is normally and independently distributed.
- The influence of a point can be determined by: (1) distance; (2) leverage; (3) influence. The most common measure of influence is Cook's D (rule of thumb: > 1 is high influence).
- Treatment of missing data is a big issue in statistics. Make sure to be familiar the three main methods of missing data treatment: (1) casewise/listwise deletion; (2) pairwise removal; (3) imputing. The latter is the most modern and promising method to deal with missing data.
- Another important topic that is discussed in this chapter concerns mediating and moderating relationships. To better understand these relationships, it is useful to draw diagrams. Therefore, we advise you to practice with drawing such diagrams.