I am currently working on a project regarding to the relationship between physical ability and levels of dementia. We are using different questionnaires to evaluate the levels of dementia. We want to compare the reliability between those questionnaires using cronbach's alpha and further dropping some redundant items.
For example, we have the cronbach's alpha increased by 0.1 if we dropping 1 item form the questionnaire. How can we implement the distribution of cronhach's alpha to conclude that the dropping is valid?
As far as I know, there is no statistical test for dropping items from Cronbach's alpha. Instead, your goal is to maximize the reliability of your scale, so the standard practice is to drop any item that is lowering the reliability.
You might also consider doing an Exploratory Factor Analysis on the items, to see if the one(s) that are lowering the alpha corresponding to items with notably lower factor loadings.
I have collected a number of resources on these issues (which are not limited to LIkert-scored items):
https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items
As far as I know, there is no statistical test for dropping items from Cronbach's alpha. Instead, your goal is to maximize the reliability of your scale, so the standard practice is to drop any item that is lowering the reliability.
You might also consider doing an Exploratory Factor Analysis on the items, to see if the one(s) that are lowering the alpha corresponding to items with notably lower factor loadings.
I have collected a number of resources on these issues (which are not limited to LIkert-scored items):
https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items
Thank you! Dr. Morgan
I also notice that that are a lot of literature discussing the exact distribution and approximate distribution of cronbach's alpha. What is the general motivation to find out the distribution of cronbach's alpha?
Can we use the distribution (exact or approximate) of cronbach's alpha to make statistical inference?
I have never seen anyone report an inference level associated with an alpha. So, there may be some theoretical interest in that issue, but it has not have any effect on practice.
I agree with David and recommend together with Cronbach alpha, Spearman–Brown and Sigma coefficients to be used, and factor analysis, also.
Dear All,
In my modest opinion and to answer " using cronbach's alpha and further dropping some redundant items." I would compare 2 columns in alpha:
1) The actual alpha
2) "alpha if item deleted"
For example: if your alpha is 0,8 and for an item A the column "alpha if item is deleted" became alha = 0,3, then you would know that "A" is a item tha lower so much the realibilty of your scale. Your scale really dependt a lot on that item "A". Perhaps it is the most important item to define your scale!
.But assume now that if you take out an item "B" your "alpha if item deleted" rise to 0,9. Perhaps you would keep "B" becasuse your realbility is higher whether you keep or not B".
In sum , the comparison betwenn "Alpha" and "alpha if item is deleted" let you analyse if all your scale have a nice structure or if your scale really depend so much on specific items. It is good to have a balanced scale where all items are needed...
Have a nice time
Helena
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics
- Survey Design
- Questionnaire