Role of "R Squared Change" in Multiple Regression Analysis (MRA)

R.W. writes, "In the PowerPoint walk through for Research Design Forum #3 (RDF3), at the point of the MRA configuration, you did not specify to check the "R squared change" box. It shows the statistics in the next slide and references to it, but I am not sure if it is really needed to complete the assignment. I wanted to let you know in case there was any confusion for others when doing the discussion post."

You raise a good question. MRA is largely concerned with identifying the amount of explained variation in the dependent variable (Y) from the linear contributions of all independent and control variables selected for inclusion in the equation on a range from 0 to 100 percent. The statistic that reveals this explained variation is R-Squared (R2), i.e., the multiple correlation coefficient (R) squared. Your question centers on how the researcher wants those Xs and control variables introduced into the regression model - forward, backward, stepwise, forced entry as a block, etc. This decision will depend on what is hoped to be accomplished through the MRA by the researcher.

Often there is a theoretical interest in first introducing/regressing a key independent variable (X1) in the regression model, via the main hypothesis, and assessing the simple regression output (i.e., the beta, R and R2). Here, the identified beta (a regression coefficient) and R (a multiple correlation coefficient) both equal the Pearson’s r (a zero-order correlation coefficient) between that X1 and Y since there’s only one independent variable in the equation. As I mentioned above, R2 provides the amount of explained variation in Y from X1 on a range from 0 percent to 100 percent. Armed with this important bivariate statistical information, the researcher can then move to expand or elaborate on the hypothesized bivariate relationship by adding key independent and control variables that are theoretically relevant to the research and then selecting the R-Squared Change option. This procedure will allow the researcher to identify change in the original R2 (as well as X1’s beta) based on the linear contribution of variables added into the regression model when those additonal variables are held constant. Note that the significance of the total equation can only be found through the F statistic and its level of significance as measured by the p value (conventionally, p must be < .05).  The F statistic and its level of significance (via the p value) also appear in the regression output.

In short, to your question and the class assignment, all variables were “forced” into the equation at once using the default METHOD=ENTER option. In this case, Robert, there is no R Squared Change to be reported. If we ran a different MRA method, such as that described above, then the "R-Squared Change" option would be appropriate and highly useful.

I hope this helps!

Professor Ziner