Rensvold & Cheung (1999), proposed a method involving Jacknifing to identify influencial cases (cases having an important impact on fit indices or coefficients) in the context of SEM (the model is re-estimated droping a single case at a time and repating the process for each case).
Is there a (simple) way to implement this in Mplus ? If not, can this be added to the "wish list" for the next version ?
Rensvold, Roger B. & Gordon W. Cheung (1999) “Analysis of Influential Cases in Structural Equation Model: A Methodological Note.” Organizational Research Methods, 2, 293-308.
Hi Linda, Thank you for this answer. I was not aware of the loglikelihood influence statistic. Very interesting.
The method proposed in the paper really consist of jacknifing the sample by taking a single case out each time and then to estimate the influence of each case on the various fit indices (can be extended to parameters) by looking at how much the indices change when the case is taken out. To implement it without doing it by hand, you need to be able to save these "influences" statistics (the fit indices from the sample without the case) in association wth each case. Someone on SEMNET suggest that it might be possible to implement it in Mplus trough the Monte Carlo function...
I can't think of a way that you could get the influence of other fit indices using Mplus. However, given that the fit measures are based on the loglikelihood, I would not imagine you would learn more than you would learn from the influence of the loglikelihood.