Dear all,
I am conducting research on the impact of blockchain traceability for charitable donations on donation intentions (experimental design with multiple conditions, i.e., no traceability vs. blockchain traceability).
One scale/factor measures “likelihood to donate” consisting of 3 items (dependent variable).
Another ”trust” factor, consisting of 4 items (potential mediator).
And a scale “prior blockchain knowledge” consisting of 4 items (control).
All factors/items are taken from previous research.
The first two (hypothesized) factors are measured on a 7 point Likert scale, ranging from 1 (Strongly disagree) to 7 (Strongly agree). However, the “prior blockchain knowledge” contains of 4 items that are measured as a semantical differential from -3 to +3. Essentially, all items in my study thus have the same “scale length” of 7.
My question is: if I perform an EFA/PCA on this data, is this a problem? Is there any gain in recoding the semantical differential to 1-7 after conducting the survey but before EFA/CFA. Or is this unnecessary or even problematic?
Curios to read your ideas, thank you in advance!
It definitely isn’t problematic but doesn’t really matter either way. You can think of factor loadings like correlations of your variable with the latent factor (they are always between .0 and 1.0). The factor loading should be the same regardless of whether you recode it but why not experiment and see if it differs. Either way it’s just correlation (being higher on one thing predicts being higher or lower on another thing) so whether the scale ranges 1-100, 1-5, or -100000-3 doesn’t really matter. I think sometimes it’s more computationally efficient to put things on the same scale like standardising the variables but in your case I don’t think it matters. Having less possible responses (like a 3 point Likert instead of a 7-point) could attenuate correlation estimates but thankfully you don’t have that problem here
Thank you mr. Weiner for sharing, helps a lot! I would agree upon your statements on factor loadings and that different scale “length” should not give problems here since the analysis works with correlations. But what if after factor analysis you find two items that load on the same factor but have different measurement scale meaning and not so much length (thus no need for standardization here). For example, an item that is measured as 1 (Strongly disagree) - 7 (Strongly agree) and another item 1 (No confidence at all) - 7 (A great deal of confidence). What does it mean for computing a meaningful factor score if there is a linguistic difference in scores on individual items?
Well that sounds like more of a theoretical question but recoding the -3,3 scale to a 1,7 scale won’t change the interpretation. I don’t think there’s a problem with using EFA to measure a latent factor that causes people to associate the semantic differential item more or less with whatever the thing is. As long as you believe theoretically that there is a latent cause to how people respond to the 4 items. PCA is also fine if your goal is to reduce the information from the 4 items to a single dimension for example. So I think you can go ahead with your EFA or PCA as long as the usual empirical and theoretical assumptions are met
- Badges
- Science topic
- Similar topics
- Psychology