Hello,
I'm a student and for my last essay, I have to use structural equation modeling. I'm kind of lost here, and I don't really know what i'm doing on the software, and I have poor statistical knowledge in that field.
Context : I did a questionnary of 18 questions that measure perceived ease of use (3), perceived utility (3), attitude (3), self efficienty(3), social norms(4), and Easy access to the technology (2) according to a likert scale of 1 to 7.
Observed variables are answers to questions, each question helps to measure their latent variable. Hypothesis is that each latent variable also influence another latent variable wich is behavioral intent of use.
I did this diagram but the data won't run, it does the analysis but nothing seems to happen anywhere on the diagram (no numbers).
Does anyone know where did I go wrong?
Thank you
Did you look at the output? Even if your model does not run there should still have some output you can look at. The output will tell you if the model is unidentifiable which is my suspicion. It’s hard to tell without seeing the output but my guess is that you need to scale each latent in your model by setting one of the paths from each latent variable to one of its indicators to ‘1’ in the regression path text box. You need one of these for each latent variable - indicator relationship.
Also the way your model is specified is suggesting that the intent to use variable is a composite variable formed by the other latents in the model, was that you intention? Does that variable have any of its own indicators? Or is it fully a function of the other variables?
Hope that helps.
Dear Antoine
There are some problem.
1. Once you use AMOS and you have a construct(easy acces) with only two items, it would not work properly . you may put 1 on one of that constructs directions.
2. the construct (social norm) put the direction on last item.
Also I have question
does you endogenous variable also measured by likert scale?
I agree with Marianna that the model may not be identified. You have six latent variables that are measured by observed variables, and these need to be assigned a metric, either fixing the variance of the latent variable at 1 or fixing the first factor loading at 1. There also seems to be a missing factor loading for the social norms latent variable.
On top of that, you have the 6 latent variables predicting a BehavInt latent variable. The way you have the model set up means that the other latent variables are 'causing' the BehavInt latent variable. It's like a principal components model, rather than a factor analysis. There are many problems with this type of model, primarily identification, as Brown (1980s) showed that there has to be at least 'omitting' 2 paths from the formative latent variable. There are other equally serious issues in relation to the interpretation as these types of models, unlike factor analytic models, require that you have all the necessary causal indictors in the model.
This is a very interesting area (see Diamantopoulos et al for great review https://pdfs.semanticscholar.org/b89b/44c0b772200d86ff235d25ad625385d5ef7e.pdf) but you are entering a complex psychometric area. Remove the residual from the BehavInt latent variable, reverse the arrows and you have a 2nd order factor analysis. That is easier to fit and interpret.
Mark
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running