Hi,
I'm using a self-reported and validated 9-item questionnaire. Items are presented in Lykert type format (1-5).
I calculated Cronbach's alpha (after reverse coding the relevant items) which was 0.61 (Low but acceptable) after omitting 4 items (44% of items).
1. Can I calculate scale score (mean score of the 5 items) or I can't because to many items were omitted?
2. Is there a maximum number/% of items that can be omitted?
If you can attach references it would be great.
Thank you!
Eran
Sure you can, but it seems that the alpha estimate is that only 61% of of true scores variance is accounted for by the metric. Although you say .61 is acceptable, I would be concerned.
Alpha is a function of the average intercorrelation (or covariances depending on how you scale it) among items and the number of items. So it isn't a matter (except for "validity drift") of how many items you delete, but how many are in the final scale and the degree to which they are related.
For a reference, see Nunnally's bible -- Psychometric Theory.
My best on your scale development,
John
I have never read an article that defines a percentage of the number of items omitted to prevent anything. I just do not recommend using the average of the 5 items as a score, since the responses of the items were collected via the Likert scale, so that the difference of one unit on the scale, in practice, is not the same for any point on the scale. I recommend using methods that generate weights for the items (Factor analysis or expert opinion, for example) or are able to generate a score by prediction or estimation via estimators.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Probability
- Probability Theory
- Reliability