Ian Kudel posted on Thursday, August 30, 2007 - 5:29 am
Hello, A colleague and I are analyzing the same dataset and we used slightly different code that we believe should be equivalent (we are using the same estimator etc.), but we are getting different results (for example my x2=10.254, and his is 38.949. Mine:
ANALYSIS: Type=meanstructure; ESTIMATOR = WLSMV; MODEL: I S | h1@0 h2* h3* h4@1;
His MODEL: I BY h1-h4@1; C BY h1@0 h2*1 h3*2 h4@3; I C; [I@0 C]; I with C; [H1$1](1); [H1$2](2); [H1$3](3); [H1$4](4); [H1$5](5);
The best way to compare the two models is to compare the free parameters from TECH1 for the two models. If this does not help, then send the two outputs with TECH1 along with both of your license numbers to firstname.lastname@example.org.
Ian Kudel posted on Saturday, September 01, 2007 - 12:25 pm
Hello, Thanks for the quick reply. Perhaps some clarification is needed. This is for a paper and we want to know if we did something wrong, or if there are differences in the approaches. If there are differences, which is right? And why is one preferred over the other?
If the fit indices and other stats were close we would not bother with the question, but there seems to be a wide disparity and it has important implications because it will effect how we write-up our manuscript (including defending our approach), consider other studies etc.
Thanks and I hope you have an enjoyable labor Day.
It is not possible to answer your question without more information. There are many things you could be specifying differently in the full input that I can't see. You have different time scores in the two models. This can affect parameter values but should not change the fit statistics at least not for WLS or WLSM.
Note that the | language is just a shorthand form of specifying a growth model. Mplus creates the growth model language using BY behind the scenes. BY and | do not represent different approaches to growth modeling. The BY language behind various growth model specifications is shown in Chapter 16.
Ian Kudel posted on Wednesday, September 05, 2007 - 5:50 am
Hi, Thanks. We also analyzed the data using the same time scores. I found time scores did not change our own respective parameter values or fit indices. In other words the parameter values and fit indices for my approach to the analyses were the same regardless of the time values. There were still differences between our findings.
Our code was the same except with regard to the model.I tried to include the complete code in the original posting, but deleted some because it made the post too long. I will send you whatever you need?