Different WRMR values with the same code PreviousNext
Mplus Discussion > Confirmatory Factor Analysis >
Message/Author
 Magnus Svensson posted on Sunday, October 14, 2007 - 10:17 am
Hi!
I have been using the Mplus 4.21 Demo just to get an idea of how it work and performs. I tested the script as follows,

VARIABLE:
NAMES ARE
Inr F91a F91c F91e F91g F91i vikt arbbef04 dik_F91a dik_F91c dik_F91g dik_F91e dik_F91i;
CATEGORICAL ARE
F91a F91c F91e F91g F91i;
IDVARIABLE IS
Inr;
USEVARIABLES ARE
F91a F91c F91e F91g F91i;
USEOBSERVATIONS ARE (arbbef04 EQ 1);
WEIGHT IS vikt;
MISSING ARE *;
ANALYSIS:
ESTIMATOR = WLSMV;
Model:
f1 BY F91a F91c F91e F91g F91i;

And got the result:
WRMR (Weighted Root Mean Square Residual)
Value 1.566
---
When I tested the above mentioned script with some extra script (in this case: SAVEDATA: ) as follows,

VARIABLE:
NAMES ARE
Inr F91a F91c F91e F91g F91i vikt arbbef04 dik_F91a dik_F91c dik_F91g dik_F91e dik_F91i;
CATEGORICAL ARE
F91a F91c F91e F91g F91i;
IDVARIABLE IS
Inr;
USEVARIABLES ARE
F91a F91c F91e F91g F91i;
USEOBSERVATIONS ARE (arbbef04 EQ 1);
WEIGHT IS vikt;
MISSING ARE *;
ANALYSIS:
ESTIMATOR = WLSMV;
Model:
f1 BY F91a F91c F91e F91g F91i;
SAVEDATA:
FILE IS C:\C:FSCORES.dat;
SAVE = FSCORES;

I got a different WRMR value,
WRMR (Weighted Root Mean Square Residual)
Value 1.279
 Magnus Svensson posted on Sunday, October 14, 2007 - 10:19 am
Both above mentioned scripts use the same input data and everything else expect the WRMR in the output give the same result. Why this inconsistency?

/Magnus
 Linda K. Muthen posted on Sunday, October 14, 2007 - 10:47 am
It's hard to imagine how this could occur. You would need to send both inputs, the data, and both outputs to support@statmodel.com for us to answer this.
 Linda K. Muthen posted on Sunday, October 14, 2007 - 3:03 pm
Thanks for sending the files. The reason this happens is that in the input with the SAVEDATA command, you ask for factor scores which causes the model to include thresholds whereas in the other analysis, the model did not include thresholds.
 Magnus Svensson posted on Tuesday, November 06, 2007 - 1:48 pm
Hi again!
another question regarding the above. Is it not contradictory that the WRMR produce a better (that is: lower) value when more parameters are estimated (as when the thresholds are estimated when the SAVEDATA command is used to ask for factor scores)? Should it not be the other way around? Should'nt I be "punished" by a higher WRMR value when I estimate more parameters? I have read what is in the Technical Appendices, but I still do not no know what this has to do with the thresholds. Are not the thresholds estimated either way, regardless if I ask for them or not?

And another question regarding WRMR. Could a high WRMR value indicate something particularly, especially in the analysis performed above.


BR,
Magnus
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: