|
Message/Author |
|
|
I am using a data(TIMSS) set with 5 plausible values(test scores). Is advice to use all the 5 plausible values in your analysis. Is there a simple command in Mplus that will do all the five analysis and give the average. i Soo difficult to do five different analysis. |
|
|
Yes, you should use all 5. Look up TYPE=IMPUTATION in the UG. This is the way to handle multiple imputation data. |
|
|
I am using the PISA data set (N = ~5,000) and predicting to mathematics performance (measured through 5 plausible values) using SEM and replicate weights. 1. Mplus noted that no indirect effect estimates and no confidence intervals for the estimates are available with multiple imputations. Why is this? 2. My data would not converge when I predicted to the mathematics outcome (mean = 480, SD = 90). After examining the output, I noticed that the residual variance for mathematics, was over 4,000 times larger than any of the other variables (see excerpt below). I think this is because all of the indicators I am using as predictors are on a scale of 1-4 and the math outcome is on a different scale? Thus, I scaled down the math plausible values by dividing them by 100. After doing this, the model converged. Is this okay or does this sort of scaling (transforming of the original variable) create unbiased estimates? ST46Q35 0.288 ST46Q67 0.266 ST46Q89 0.353 PV1MATH 4526.233 INTENTION 1.000 BEHAVIOR 1.000 3. I noticed that no traditional model fit statistics are available with multiple imputations (e.g., RMSEA, CFI, etc). Why is this? How do I evaluate model fit? |
|
|
1. They are available in Mplus 8.1. 2. This is ok 3. The methodology is not available yet, mostly because of the replicate weights. Use SRMR. |
|
Back to top |
|
|