forum.smartpls.com

Just a short question:

what information do I get from the p-value in the cb-sem model fit. I have 0,000 in the null and estimated model and it is coloured red. All the other values are ok up to quite good (green)

thx

Hupfman

The p-value in the fit table in CB-SEM is the chi-square test of model fit. This indicates that the model-implied and observed covariance/correlation matrix significantly differ.

Is this a problem? It depends on the viewpoint. Certainly, you want a good fitting model. However, this test is very sensitive and mostly significant indicating misfit. I have rarely seen practical models that have good fit without the researcher changing the model to make it fit.
One problem is that the more data you have the more likely you will find even small misfit significant. The more data you have the larger is the power on your tests. What is usually a good thing might be a bad thing here. Because models with small samples are more likely to produce an insignificant chi-square model fit test. Are these models better? No, we just have higher uncertainty and therefore we can neither conclude that they are fitting nor misfitting.
Another problem is, that these stringent fit test might lead researcher to tweak their models in undesired ways. At least, tweaking the models increases the chance of overfitting the model to the data making the model overall less generalizable.

Therefore, researchers have proposed other approximate fit indices that are less stringent on the fit requirements and most researchers focus only on these other fit measures. Nevertheless, there is some debate around which fit measures you want to use and when.

thx. Yes. My data sample is considerable large with over 700 cases. This might explain why everything else is good just not this criterion.