In a resarh paper I read the following:
"We rescaled both variables from a 1 to 5 scale to
-2 to +2 scale by subtracting 3 from the data."
In the same paper I also found for another variable:
"We first rescaled the self-reported likelihood to buy from a 1 to 7 scale to a -3 to +3 scale"
What is the rationale behind?
I am guessing the point was just centering; see this if helpful: https://www.theanalysisfactor.com/centering-and-standardizing-predictors/ and https://www.theanalysisfactor.com/when-not-to-center-a-predictor-variable-in-regression/#:~:text=There%20are%20two%20reasons%20to,the%20ones%20that%20were%20multiplied.
Very strictly this isn't centering (though it will be similar for many data sets). In centering you subtract the mean of the data not the mean of the scale. In a -3 to +3 scale it is unlikely the mean is exactly zero every time.
Generally this is done for interpretability so the intercept in a regression model is at the midpoint of the scale. In this respect it is similar to centering.
It will reduce collinearity caused by interactions in the model, but might not eliminate it. (Note that this collinearity impacts on interpretability of individual effects but doesn't impact the interpretability of the model).
Finally centering can improve efficiency of estimation in complex models and this sort of pseudo-entering could also help.
Perhaps it had something to do with the scale's semantics or interpretation of some statistical parameters. Another thing is that this "centering" could have stemed from the manual itself.
- Badges
- Science topic
- Similar topics
- Biological Science
- Anatomy
- Tissue