95%IC for SRMR

[JB Lanfranchi]

Dear Heungsun,

I obtain theses results below with a short sample (N = 53) with this model :

Model fit measures               
    Estimate    SE    95%CI   
FIT    0.6524103    0.0299918    0.5903335    0.7075027
AFIT    0.6356654    0.0314367    0.5705981    0.6934118
FITs    0.4982611    0.0390896    0.4281782    0.5806402
FITm    0.6919358    0.0304619    0.6268094    0.7463656
GFI    0.9626313    0.0271979    0.85998            0.9650264
SRMR    0.0839326    0.0126935    0.0947164    0.1432927

Is it normal to have an 95%IC for SMR without the value of SRMR ?

Thanks a lot four your help and the soft !

Best,

Jean-Baptiste

[heungsun.hwang]

Dear Jean-Baptiste,

This is not likely to happen frequently yet can certainly happen. The percentile bootstrap method that the software uses does not construct confidence intervals based on the original sample's estimate, but based only on bootstrap samples. Thus, if the majority of the bootstrap samples is different from the original sample, the bootstrap CI may not contain the estimate. There are alternatives for adjusting for skewness in the bootstrap distribution, but they are not implemented yet. And it is not clear whether the alternatives are always preferred over the percentile method. I’d like to suggest increasing the number of bootstrap samples. Also, although we provide the bootstrap Cl of SRMR (and other fit indexes), you may consider reporting the original sample's SRMR only and applying its suggested rule of thumb cutoffs. 

Best,

Heungsun

[JB Lanfranchi]

Thanks a lot !

Very minor changes with 100 or 1000 or 5000 bootstrap samples...

With 100   : IC95% =  [ .0947930, .1434930 ]

With 1000 : IC95% = [ .0943903, .1424215 ]

With 5000 : IC95% = [ .0947164, .1432927 ]

Best,

Jean-Baptiste