Sanjoy posted on Monday, October 10, 2005 - 4:31 pm
Dear professors ... this is my initial model (where each r's and b's are 5 point ordinal)
R by r1, r2, r3 B by b1, b2, b3
I had a hunch for the possibility of cross-loading, and also because of initial EFA results, I went for it completely ... like this way
R by r1, r2, r3, b1, b2, b3 B by r1, r2, r3, b1, b2, b3
However it can not run ... saying "THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR MODEL PROBLEM INVOLVING PARAMETER 12 " ...
I mean our total freely estimable parameters are 10 lambdas, 2 factor variances and 1 factor covariances ... that makes 13 ... and from sample correlation matrix (lower traingular elements), we have (6*7)/2=21 values ... I'm missing the non-identification point
Q1. I was wondering why?
thanks and regards
bmuthen posted on Monday, October 10, 2005 - 4:43 pm
Remember the basic rule for exploratory factor analysis - with 2 factors you need 4 restrictions on Lambda and Psi. You only have 2 restrictions - the two unit loadings. You can specify an "EFA within CFA model" which we teach how to do in the first day of our 5-day course in Alexandria; you can purchase the handout from this day using the Mplus web site.
Sanjoy posted on Wednesday, October 12, 2005 - 8:27 pm
Thank you Professor...I'm not very sure yet about the reason why u have said so, yesterday I found Prof. Joreskog's article "Addendum, page 40-43", Advances in Factor Analysis and SEM, 1979, I suppose I can site his proof as the reference to this identification issue ...
yes as u have advised, I'm planning to buy that handout ... I have one suggestion in this regard ...why don't you sell it as a “.pdf” document instead of sending via UPS, I mean u can charge a bit more on handling account, however selling it as a “.pdf” document will be 1. less time consuming, besides 2. we can save the amount we pay to UPS
Dear professors: I'm working with categorical data. I'm interested in defining the therholds and the intercepts too; but I have problems with the model. Could you tell mee what's is wrong? I've 10 categorical variables, and so I define 10 latent variables (one for each), also I define one new latent factor (f1). VARIABLE: NAMES ARE u1-u31; USEVARIABLES ARE u2 u3 u4 u6 u9 u10 u12 u19 u28 u31; CATEGORICAL ARE ALL; ANALYSIS: TYPE=MEANSTRUCTURE; PARAMETERIZATION=THETA; ESTIMATOR=ULS; MODEL: f2 by u2; f3 by u3; f4 by u4; f6 by u6; f9 by u9; f10 by u10; f12 by u12; f19 by u19; f28 by u28; f31 by u31; f2@1; f3@1; f4@1; f6@1; f9@1; f10@1; f12@1; f19@1; f28@1; f31@1; f2-f31*.5 f1 by f2* f3 f4 f6 f9 f10@1 f12 f19 f28 f31; f1@1; [f2 f4-f31] OUTPUT: TECH1; TECH2; modindices;
You can identify the thresholds and intercepts only with multiple group analysis or repeated measures data. So you can't do this for your example. If you had multiple groups or timepoints, then you would not be able to identify the variances in addition to the intercepts. So you would need to eliminate the statement f2-f31*.5.