Multiple Regression

Uses and Advantages of Multiple Regression

Researchers use multiple regression as a statistical procedure to analyze quantitative data with the goal of explaining relationships between variables. Researchers develop a hypothesis about how aspects of a particular phenomenon are related to one another and test those relationships by creating a model that explains the various relationships. Typically the data used for multiple regression are made up of continuous variables (e.g., interval-level measurements such as Likert scales or amounts of observable behavior), but it is also possible to use categorical data (e.g., demographic information such as gender or ethnicity). To use categorical data in multiple regression, one must employ a technique called dummy coding, which is discussed later in this entry. Unlike correlation, which shows the co-occurrence of variables (e.g., perceptions of liking), regression can be used for prediction and causal inference. For prediction, researchers create a multiple regression model containing multiple factors that will combine to predict an outcome, even if that outcome has not been directly observed (e.g., a certain amount of immediacy behavior will result in a certain amount of liking). In terms of causation, researchers use multiple regression to provide evidence that an increase or decrease in one or more variables creates a change in the outcome variable (e.g., more immediacy behaviors cause more liking). Of course, researchers can use simple linear regression to predict and show causal relationships, but there are several reasons why one might choose multiple regression over simple linear regression.

Researchers favor multiple regression over simple linear regression when they suspect that multiple factors contribute to explaining a particular phenomenon. Researchers in communication studies, like social scientists in other disciplines, often explore phenomena that likely have more than one factor that explain a particular phenomenon. Taken together, several variables working together can offer more predictive ability to explain the variable under investigation than a single predictor variable working by itself. For example, one immediacy behavior, such as smiling, might not predict liking, but combining several immediacy behaviors together, such as smiling, eye contact, and physical proximity, might have a greater ability to predict liking. In addition, multiple regression allows researchers to isolate the predictive ability of a particular variable by controlling for other, often related, variables. For example, if smiling and eye contact are held constant, does increasing physical proximity predict an increase in liking? Finally, researchers favor multiple regression because it allows them to see how predictor variables interact to explain the outcome. When researchers talk about mediation and moderation, they are describing an interaction between variables that can be found with a multiple regression analysis. Overall, multiple regression analysis is more robust than simple linear regression because it allows researchers to develop more complex models to investigate the relationship of several predictor variables with an outcome variable in combination or in isolation.

...