In my data analysis one construct has low alpha value and low composite reliability values that are 0.455 and 0.422 but AVE value is 0.6 and fulfill the convergent and discriminant validity criteria. Kindly tell me that can i keep the construct for the study.
In principle, the maximum possible correlation between two variables is the square root of the product of their reliabilities. So if you have one measure with reliability of about .45, it could correlate at .60 with a measure that enjoyed reliability of .90 or above - but just barely. The two measures would have to be accessing practically the same construct. In which case the question arises: why use the unreliable one when the more reliable one taps into the same underlying variable?
I can only think of one other possibility. Your low-reliability measure may not really be a unitary scale. There may be several different variables merged together in it, and it just so happens that every one of them correlates meaningfully with your criterion variable. In that case, you could "keep" the measure, but you really should try to analyze it into its constituent elements. If it's a multi-item scale, for example, you might try a factor analysis. Because it might really be an amalgam of several scales.
Just a thought.
As Prof. Joy commented, it is difficult to publish an article with a reliability with below .70 (which is commonly used cut-off criteria).
You can try to delete some items that cause the low reliability. You can check the inter-correlation among the items. Maybe there are several dimensions to be separated.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Values