Modeling in Logistic Regression
In using logistic regression, researchers try to "model the data." This simply means that they are trying to find a "good" set of independent variables that can help predict or explain group membership on the dependent variable. For this modeling process to be successful, the researcher must possess a thorough understanding of his or her discipline in order to decide which of a vast array of variables should be measured. But even if the "right" variables are measured, logistic regression presents the researcher with options as to how the data can be analyzed. Sometimes, different options must be selected and compared (in a trial-and-error fashion) in order to identify the best way to model the data. If you take another look at Excerpt 19.25 [not shown here], you will see that the researchers refer to variables that were a part of their "final model."
If you now look at Excerpt 19.37 [not shown here], you will see how the researchers of this study decided when independent variables were allowed to enter or were forced into the developing model. This excerpt nicely illustrates an exceedingly important point: the availability of computer programs that perform logistic regression analyses does not eliminate the need for researchers who can think carefully about their studies. The computer helps immensely, of course. Nonetheless, it is the researcher, not the computer, who performs the modeling!
(From Chapter 19, pp. 600-601)
Copyright © 2012
Schuyler W. Huck