Latent variable, but CTT.

Matt

EFA is based solely on correlations among variables, so it can produce factors that are hard to interpret. Even when are "easy" to interpret, it is up to you to create a name for what they measure.

The most likely alternative is to use Confirmatory Factor Analysis (CFA) where you begin by making explicit assignments of variables to factors that you believe capture what the items have in common.

Even if EFA present some disadvantages, several researchers consider this statistical method as the best for analyzing the structure of a measurement tool.

Look at this article that you can find on the web.

Best

CH

Hi all !

Hi Claude !

Now I only use an EFA for observation purpose when I don't have an idea about how the data behave, when I'm at the beginning of a mesurement instrument creation process. For example when I gather some affirmations during interviews of a small sample of teachers and when I then submit these affirmation to a larger sample of teacher asking them to give there agrement on a Lickert type scale. Brefly writing EFA has its name : exploration. I see also advantage in its capability to categorize data and in its ability to reduce data by removing information.

This leads to the disadvantage that it removes information ignoring maybe relevant information. One can choose a particular model with EFA that is just the reflection of noise in the data. I once ran an EFA removing items getting more and more unsatisfied of what happened and more and more surprised of the incoherent solution that was produced with weak model fit. Then I run a CFA based on the hypotesis I had "just for fun" as I am naturaly curious. I got better model fit, close to cut-of ; including some between item error correlation getting perfect fit. Another time I just had to include a general factor with all items with a 3 factor structure with each subgroups of the items.

To Resume, I now prefer to directly run CFA when I have a clear idea, mostly research informed or based on former validation, of data structure. I've posted a complementary question to yours on RG.

https://www.researchgate.net/post/CFA_without_EFA_in_a_questionnaire_validation_process

The answers are instructive.

All the best

Philippe

following

Advantage: resolve the common problem, in real contexts, of non-zero cross-loading.

Disadvantages: Fit indexes, data-drive structure without theory, problems with measurement errors, you can´t include common variance of the method and, most important, it can´t be used to test structural equation models.

Try to forget EFA and CFA and use ESEM... the best of both worlds.

Marsh, H. W., Morin, A. J., Parker, P. D., & Kaur, G. (2014). Exploratory structural equation modeling: An integration of the best features of exploratory and confirmatory factor analysis. Annual review of clinical psychology, 10, 85-110.