4

I am trying to use Cronbach's alpha in order to see if some variables are measuring the "same", but I have some doubts about the correct use:

  1. Can I compute alpha for numerical, ordinal or dichotomous variables? If it is correct to do that, then may I mix dichotomous, ordinal and numerical variables?

  2. If Cronbach's alpha cannot be used for numerical, should I turn these variables into ordinal, or not?

  3. If all my variables are ordinal, do they have to be on the same scale? I mean if variable 1 goes from 1 to 5, and variable 2 goes from 1 to 3, may I use Cronbach's alpha for these 2 variables?

Stephan Kolassa
  • 95,027
  • 13
  • 197
  • 357
eli
  • 43
  • 3
  • 1
    Possible duplicate http://stats.stackexchange.com/questions/30679/is-it-acceptable-to-use-cronbachs-alpha-to-assess-reliability-of-questionnaire – Momo Sep 04 '13 at 18:25

1 Answers1

3

First, alpha is a quantity for a scale (a set of items).

Ad 1: Strictly speaking, alpha only makes sense for metric items (which, I believe, you mean by numerical variables). However, it is often used on (sum) scales of ordinal items too (there the general rule of thumb is a sum of more than 7 items and more than 4 levels of the item). I believe this is bad practice though.

Ad 2: You can use alpha in this case. Generally, it is rarely a good idea to make a numeric variable ordinal.

Ad 3: I would not use it in this case (but mainly because they are ordinal).

Momo
  • 8,839
  • 3
  • 46
  • 59
  • Thank you for answering. For numerical variables I mean for example student grades, and ordinal if they are Likert sacale – eli Sep 04 '13 at 18:44