As implied in the question, I have one bothersome variable called "merit value" which has different scales depending on which group the student was accepted from.

The groups are:

Grades             -     Scale: 10-22.5
SAT equivalent     -     Scale: 0-2
Community college  -     Scale: 1-4

So far I've been reluctant to use this as an independent variable in my regression analysis. Not just because of the different scales but because of their different interpretation. For instance, my material indicates that it's much more difficult for a student to achieve a high score on our SAT equivalent than it is to achieve a high grade.

My technical supervisor has suggested I use standardized beta coefficients in order to even out these differences and in order to convert the variable values to a common scale.

Is this the proper solution in this case or might there be some unforseen consequences?

  • 1
    $\begingroup$ Standardizing would still retain the structure you're referring to in this particular coefficient. How about modelling this as a three-level factor instead? Or even splitting it into three different variables all together (where students that don't have SAT scores have missing values)? $\endgroup$ – harisf Mar 23 at 10:55
  • $\begingroup$ I have a variable called "selection group" so that's already quite close to what you're describing. I could use the second solution but currently the variable measures the merit value of the selection group from which the student got accepted. I have redesigned the data extraction/data shaping processes in order to retain "all" merit values but sadly, I'll probably have to save that for a later project! $\endgroup$ – Magnus Mar 23 at 11:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.