Consensus Points on best reporting practices in ML/MLR, CFA and SEM |
The best guide to assessing model fit is strong substantive theory. |
The chi-square test statistic should not be the sole basis for determining model fit |
The use of multiple fit statistics promotes more reliable and conservative evaluations |
We should not ignore the fit of the components of a model. The researchers should always examine the components of fit along with the overall fit measures. |
it is better to examine several alternative models than only a single model. |
Recommendations |
Outliers and influential cases should be traced and the distributional assumptions of an estimator should be satisfied as a perquisite. Next steps follow. |
When reporting fit indices, choose ones that represent different families of measures or tap different aspects of the model. |
Choose fit indices with sampling distribution means that are not or are only weakly related to the sample size. |
Choose fit indices that take into consideration the model degrees of freedom |
Evaluate the model adequacy based on prior studies. Decide on the optimal model on the basis of comparison |
The objective of fitting SEMs is to understand a substantive area, not simply to obtain an adequate fit (e.g local optimum) |
The test statistics and fit indices are useful, but they cannot replace sound judgment and Expertise |
As proposed by |