search

melissagwolf / dynamic

Dynamic Fit Index Cutoffs For Latent Variable Models
GNU Affero General Public License v3.0
14 stars 2 forks source link

Model with reverse coded items produces different results than one including default coding #7

Open WillemSleegers opened 1 year ago

WillemSleegers commented 1 year ago

Hi,

I'm working on a project to develop a scale and I'm using DFIs to try and find a well-fitting model. In fitting the models, I've noticed that I get different results depending on whether I include the items as reverse coded items or their default coding. In the model where I've coded the items so that they all point in the same direction (all positive factor loadings), I can obtain level 1 DFIs, but when I run the same model with 2 of the items coded back to their default coding (negatively worded while the others are positively worded), I fail to obtain DFIs for level 1 (at least the 95/5 one) and the results are generally worse.

It was my understanding that the direction of the item coding should not matter. Is that correct?

melissagwolf commented 1 year ago

Hi,

Is this happening in the multi-factor model?

If so - yes, that's correct. I need to push an update that changes everything to absolute value. There's an error in the item selection algorithm (the abs() function was mistakenly omitted in one part). Please just use absolute values for everything as that is the update I will be pushing soon.