There are 3 scales that my company is currently using to benchmark products against one another. Two of the scales measure the emotions of customers after using the product and the third measures how usable the customers felt it was.
My task is to figure out how to reduce the three scales, but an EFA doesn't seem appropriate since (a) the scales all use different rankings, (b) they're focusing on different constructs, (c) they're evaluating different content (i.e. one of the emotion scales asks "how did you feel while completing tasks with the product?" and the other asks "how do you feel about the overall product?")
Is there a way to reduce the scales? To remove items that aren't accounting for much unique variance in the data? I already ran a correlation matrix, but none of the items correlate enough to suggest that they're measuring the same thing. I also tried removing items that would improve the scale-level alphas, but that only cut a couple of items.
A regression seems like a good option to identify items with the most variance but there isn't really a DV I can analyze since all 3 of the scales are already supposed to be the DVs.
Hello Justin,
The key issue here is, what goal are you trying to achieve? I get that you wish to reduce the sheer number of items across the three scales, but how are the scores that you obtain to be used?
If the purpose was to identify the "best" item subset (of a given, reduced item number) that maximally distinguishes products known to be different (on the dimensions of interest), you might consider an approach like this:
1. Identify product pairs that judges agree are different (on the attributes of interest).
2. Take the scores from the items from all 3 scales as a set.
3. Develop all subsets of X items, X+1 items, X+2 items, ... up to your limit for a "reduced" item set.
4. The "best" ensemble of X items is that which yields product pair means that differ maximally in space.
This is analogous to discriminant analysis, albeit with differing numbers of variables.
5. You'll want to repeat the process for multiple product pairs, to give confidence that the subset of items you're choosing consistently works to distinguish the pairs.
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Psychology
- Emotion