Stepwise regression PreviousNext
Mplus Discussion > Confirmatory Factor Analysis >
Message/Author
 Anonymous posted on Thursday, April 15, 2004 - 11:35 am
Is it in any way possible to run a stepwise regression on the between-level of a two-level model in mPlus? For example, I have six observed independent variables (I will specify measurement error in each of these), and one latent dependent variable. I would first like to see to what extent five of the independent variables explain variance in the dependent variable, and then add a sixth independent variable to see if it adds anything to the prediction. I read a paper by deJong (1999) about Cholesky factoring and phantom factors. Is it possible to use this procedure in mPlus? I would appreciate any suggestions.
 bmuthen posted on Thursday, April 15, 2004 - 11:42 am
Mplus does not provide stepwise regression. You can do Cholesky factoring with phantom factors in a latent variable framework such as Mplus.
 Anonymous posted on Friday, July 16, 2004 - 2:12 am
I have tried to run a Cholesky factoring with phantom factors (as described in deJong, 1999) without success. Could you please see whether there is anything wrong in the syntax I have specified below? I get an error message that there are insufficient data in the data file, but the file functions well doing regular regression. Another question: Bentler & Satorra has an alternative approach, where one use residuals as predictors in equations. Is this possible in mPlus? I am very grateful for all your help.

TITLE: CHOLESKY
DATA: FILE IS cholesky.DAT;
TYPE IS CORRELATION STDEVIATIONS;
NOBSERVATIONS ARE 95;
VARIABLE: NAMES ARE
v1-v12 PH1 PH2 PH3 PH4;
MODEL: AW BY V1-V3;
SYN BY V4-V6;
SRN BY V7-V8;
NI BY V9-V11;
WORDREC BY V12@1;
V12@0;
PH1@1;
PH2@1;
PH3@1;
PH4@1;
PH1 ON NI AW SYN SRN;
PH2 ON AW SYN SRN;
PH3 ON SYN SRN;
PH4 ON SRN;
AW@0;
SYN@0;
SRN@0;
NI@0;
PH1 WITH PH2@0 PH3@0 PH4@0;
PH2 WITH PH3@0 PH4@0;
PH3 WITH PH4@0;
WORDREC ON PH1 PH2 PH3 PH4;
OUTPUT: STAND;
 Anonymous posted on Friday, July 16, 2004 - 3:13 am
I just noticed an error in the VARIABLES part of the syntax I sent you, it should be:

VARIABLE: NAMES ARE
v1-v12;
USEVARIABLES v1-v12 ph1 ph2 ph3 ph4;

However, it still is not accepted, and the problem seem to be to get the program to accept phantom factors without indicators. Thanks for any suggestions about this.
 Linda K. Muthen posted on Friday, July 16, 2004 - 5:59 am
Please send your input and data to support@statmodel.com.
 Anonymous posted on Friday, July 16, 2004 - 8:34 am
Following is the model modification you need. Tihomir

-----

MODEL: AW BY V1-V3;
SYN BY V4-V6;
SRN BY V7-V8;
NI BY V9-V11;
WORDREC BY V12@1;
V12@0;

PH1@1;
PH2@1;
PH3@1;
PH4@1;

PH1 BY NI* AW SYN SRN;
PH2 BY AW* SYN SRN;
PH3 BY SYN* SRN;
PH4 BY SRN*;

AW@0;
SYN@0;
SRN@0;
NI@0;

PH1 WITH PH2@0 PH3@0 PH4@0;
PH2 WITH PH3@0 PH4@0;
PH3 WITH PH4@0;

WORDREC ON PH1 PH2 PH3 PH4;
 melissa posted on Monday, June 18, 2007 - 9:37 am
Hello,
Back in 2004, you mentioned that Mplus does not provide stepwise regression (see April 15, 2004 - 11:42 am above).
Is this still the case in 2007?

If so, could one simply enter the various "blocks" in a series of progressive but separate analyses?

Thank you.
 Linda K. Muthen posted on Tuesday, June 19, 2007 - 8:15 am
No, we do not provide stepwise regression in Mplus. I am not sure but I don't think entering the covariates in blocks alone is equivalent to stepwise regression.
 Martin Ratzmann posted on Tuesday, July 05, 2011 - 5:18 am
Hello,

how I can request a plot of residuals resulted from simple regression of y ON x?

Thank You!
 Linda K. Muthen posted on Tuesday, July 05, 2011 - 3:52 pm
If you mean the difference between the predicted and observed values of y, you would need to create this residual in DEFINE and then it could be plotted.
 Martin Ratzmann posted on Tuesday, July 05, 2011 - 11:33 pm
This is exact, what I mean!

DEFINE:
Res = y - (y ON x); ! <- How is the correctly input

Thank You!
 Linda K. Muthen posted on Wednesday, July 06, 2011 - 11:43 am
DEFINE:

ypred = alpha + beta*x;
res = y - ypred;

You need to replace alpha and beta with the numbers from your output.
 Martin Ratzmann posted on Wednesday, July 06, 2011 - 12:28 pm
Thank You!
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: