In this chapter, we extend the One-Way ANOVA to situations with two factors or categorical explanatory variables in a method that is generally called the Two-Way ANOVA. This allows researchers to simultaneously study two variables that might explain variability in the responses and explore whether the impacts of one explanatory variable change depending on the level of the other explanatory variable. In some situations, each observation is so expensive that researchers want to use a single study to explore two different sets of research questions in the same round of data collection. For example, a company might want to study factors that affect the number of defective products per day and are interested in the impacts of two different types of training programs and three different levels of production quotas. These methods would allow engineers to compare the training programs, production quotas, and see if the training programs “work differently” for different production quotas. In a clinical trials context, it is well known that certain factors can change the performance of certain drugs. For example, different dosages of a drug might have different benefits or side-effects on men, versus women or children or even for different age groups in adults. When the impact of one factor on the response changes depending on the level of another factor, we say that the two explanatory variables interact. It is also possible for both factors to be related to differences in the mean responses and not interact. For example, suppose there is a difference in the response variable means between young and old subjects and a difference in the responses among various dosages, but the effect of increasing the dosage is the same for both young and old subjects. This is an example of what is called an additive type of model. In general, the world is more complicated than the single factor models we considered in Chapter ?? can account for, especially in observational studies, so these models allow us to start to handle more realistic situations.
Consider the following “experiment” where we want to compare the strength of different brands of paper towels when they are wet. The response variable will be the time to failure in seconds (a continuous response variable) when a weight is placed on the towel held at the four corners. We are interested in studying the differences between brands and the impact of different amounts of water applied to the towels.
Predictors (Explanatory Variables): A: Brand (2 brands of interest,
named B1 and B2) and B: Number of Drops of water (10, 20, 30 drops).
Response: Time to failure (in seconds) of a towel (\(y\)) with a weight sitting in the middle of the towel.