IFI scaled

[car...@web.de]

I have a CFA with the following coefficients: chisq,scaled = 101.791, df = 84, baseline,chisq,scaled = 1525.369. If I simply use the scaled results in the usual formula to calculate the IFI, then I get: (1525.369 - 101.791) / (1525.369 - 84) = 0.988. This also corresponds approximately to the result that EQS produces. Lavaan calculates the following result: ifi,scaled = 0.933. How does this not insignificant difference arise?

[car...@web.de]

After looking at the source code, it's clear. IFI is calculated as follows:

t1 <- X2.null - X2

t2 <- X2.null - df

[...]

IFI scaled is calculated like this:

t1 <- X2.null.scaled - X2.scaled

t2 <- X2.null.scaled

[...]

The denominator therefore does not take df into account. I think that this is not correct. Best regards.

[Terrence Jorgensen]

The denominator therefore does not take df into account. I think that this is not correct.

Thanks for tracking down the problem.  I'll make a note on GitHub to make the change.

Terrence D. Jorgensen

Postdoctoral Researcher, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

UvA web page: http://www.uva.nl/profile/t.d.jorgensen