|
|
Scaling factors, MLR and Decimals |
|
Message/Author |
|
|
Hello, I'm having problems with decimals on the MLR scaling correction factor. I've got two models which are nested, and the output from Mplus gives: 1: Chi-sq = 1794.786, df= 1246, scf = 1.118 2: Chi-sq = 1802.923, df = 1247, scf = 1.117 When I calculate the difference between the chi-squares, using the formula which incorporates the scaling correction factors, I get, chi-square = -56 (yes, negative). However, this appears to be due to rounding error. If I change the scaling factor on the second model to 1.1174, I get a chi-square difference of 22 (positive), a massive change. For some of the models where I'm running this, I can calculate the scaling correction factor 'by hand', by running the ML models and dividing the chi-squares. In which case I find that they are actually 1.117423 and 1.11754, and when I plug those two numbers in, I get a chi-square of 9. However, I'm running these models with and without weights, and when I run with weights, I can't calculate by hand, I have to use MLR. So, two questions: 1) Is there a way to make Mplus give more decimal places. 2) One solution that I've tried is to use the average of the two scaling correction factors for both - this works (and in the case that I gave, the average - 1.1175 is very close to both). However, is this going to give me the wrong results on occasion? Thanks, Jeremy |
|
|
I have found a partial solution, which is to use the savedata: results are results.dat And then pulling the scaling correction factor from there. It's not especially easy (as it's not labelled), and it's not ideal, as I don't know when I need the extra precision, and when I don't need this extra precision. (Well, I kind of know that when my chi-square is high, I need lots of precision). Jeremy |
|
|
Yes, results are saved with more decimals. We will add more decimals to the scaling correction factor in the output in the next update. |
|
Dex posted on Friday, September 18, 2015 - 5:08 pm
|
|
|
Is there any way to ask for saving the Scaling Correction Factor for MLM under montecarlo? "savedata: results=*.dat;“seemed not to include the Scaling Correction Factor.Thanks. |
|
|
No, I don't think so. But you can use techniques described on our Using Mplus via R page: http://www.statmodel.com/usingmplusviar.shtml |
|
Back to top |
|
|