7
$\begingroup$

I wonder if it is possible to have a poor model fit and at the same time strong significant and high path coefficient values among the latent variables?

How would be interpret this? Also, what if we obtained strong model fit but non-significant/low path coefficients. Can someone elaborate on this?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
7
$\begingroup$

Yes, it's easy.

Let's say that this is your population model:

+---+    0.5       +----+
| X +------------> | Y  |
+-+-+              +-+--+
  |     +----+       ^   
  +---->+ M  +-------+   
    0.5 +----+   0.5

And you fit this model: 

+---+              +----+
| X |              | Y  |
+-+-+              +-+--+
  |     +----+       ^   
  +---->+ M  +-------+   
        +----+

You've omitted the direct path from X to Y. This omission will make the model fit, very badly.

However, the parameters from X to M and M to Y will be high - higher than they should be, and (for any reasonable sample size) highly significant.

Model fit comes first. If your model doesn't fit, you don't trust the parameter estimates.

(That doesn't mean that if your model does fit, you do trust them.)

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.