Dear community,
I would like to know if you have ever came across a similar issue. I have an item that is negatively correlated with other 5 items that I am using to build a formative construct.
There is no multicollinearity problem (measured by VIF).
Is it correct to drop this item from the construct? To my understanding, this means that probably the items are measuring something different.
If this sounds feasible, do you know of any paper that could back up this approach?
Thank you in advance!
Andrea
Have you tried reverse coding the negative correlated item.? Best wishes David Booth
Hii Andrea...have you developed the scale or is it pre-established? If you've developed the scale then you can delete the item. But before doing so, check with the item's language, whether it is capturing the same meaning in the opposite way. For example, "I like the shopping" is a positively-keyed item, and "I find shopping is boring" is a negatively-keyed item, but both are measuring/capturing one's knack for shopping. If that is the case in your scale, you need to reverse code the item which is negatively correlated to others. If not, then delete the item, considering a justification such as respondents might not perceive the item correctly, etc... However, in case your scale is already-established in literature, I'd advise looking into the studies which have used the same scale and what are the measurement results (including their take on reverse coding). If you find similar issues in other studies, you can cite them accordingly.
- Badges
- Science topic
- Similar topics
- Correction