Regression Analyses and Discovery

Finally, keep in mind that the potential worth of any regression analysis is limited by the prospect of uncovered something new that is not already known. Regardless of the size of the sample from which data are collected, regardless of the complexity of the model(s) being evaluated, regardless of the level of sophistication of the regression technique, and regardless of the number or kind of inferential tests applied to the data, ask yourself whether the research report contains a legitimate major (or even minor) discovery--or is it instead simply a nicely packaged restatement of what everyone knew to be the case all along?

Perhaps you think this final "warning" is unnecessary because editorial boards supposedly function to screen out studies that either were poorly executed or never had a chance to show anything interesting because of the nature of the variables or hypotheses. If so, take another look at Excerpt 19.26 [presented earlier in Chapter 19 but not shown here]. In the study from which that excerpt was taken, the logistic equation was found to be statistically significant (with p = .003), as were two of the three odds ratios (with p = .01 in each case). And what did this regression analysis show? It revealed (?) that "families that provided higher levels of help [for their relative with Alzheimer's disease] were less likely to institutionalize their ill elders than family members who reported providing low levels of family help." Are you surprised by this "discovery"? Do you think anyone would be surprised?

(From Chapter 19, pp. 604-605)

Copyright © 2003 Schuyler W. Huck
All rights reserved.