I want to adjust a normative set of cognitive data so that differences in scores related to age are removed. I thought that if I performed a linear regression of the cognitive data (as dependent variable) with age (as an independent variable) I could achieve this by saving the residuals. I could then transform the residuals to a more user-friendly format with, for example, a mean of 100 and a standard deviation of 15. If then any new score was put through the regression equation (based on the slope and intercept) and also the transformation, I would get its position relative to the mean (100) and standard deviation (15) of the original normative sample. And unlike if I just did this with the raw data, the effect of ageing would have been removed. Is this a reasonable strategy, or have I misunderstood something basic? Any advice or comments would be much appreciated.
While I've never done it, this strategy sounds reasonable to me. In fact we do something very similar in one of my research areas, reading. People who are analyzing the amount of time the eyes spend looking at a given word when reading a sentence will often calculate "residual reading times": they regress the reading time on word length and other predictors, and then use the residuals as the DV in their actual stats, to see how the experimental manipulation affects reading time after 'correcting' for word length and other nuisance variables that aren't of interest.
Yes it is valid and has been done before. Linear transformations are fine and don't affect your overall relationships. However, the interpretation of the regression coefficients will be different because they are in original units. Now, you have to keep in mind that your cognitive data are doubly-standardized units. Make sure you could explain what this means in results or discussion.
- Badges
- Science topic