Message/Author 

roachyang posted on Wednesday, July 18, 2012  1:49 am



What model is the following code fitting? the variable z is an observed variable. Is this a normal SEM model? What is z on y1 y2 y3 fitting? Is it regress z on the factor scores of y1 y2 y3? Is it a logistic regression? title: sem; data: file is "data.txt"; variable: names are x1 x2 x3 x4 x5 x6 x7 x8 x9 z; usevar are x1 x2 x3 x4 x5 x6 x7 x8 x9 z; categorical are x1 x2 x3 x4 x5 x6 x7 x8 x9 z; missing are all (9999); analysis: estimator = wlsmv; model: y1 by x1 x2 x3; y2 by x4 x5 x6; y3 by x7 x8 x9; z on y1 y2 y3; output: standardized(STDY); 


What is the scale of z? 

roachyang posted on Thursday, July 19, 2012  4:14 am



it's a 2 to 2 categorical variable. 


If you are using the default WLSMV estimator, the regression is probit. If you are using ML with a logit link, it is logistic. If you are using ML with a probit link, it is probit. 

roachyang posted on Saturday, July 21, 2012  12:32 am



Thanks! So it's a probit regression of z on the factor scores of y1 y2 y3? 


If you are using WLSMV or the probit link of ML. 

roachyang posted on Sunday, July 29, 2012  12:33 am



If I want to do a logistic regression of z on the factor scores of y1 y2 y3, how can I do it? 


You would put z on the CATEGORICAL list and use the MLR estimator. The MODEL command would be: y1 BY ... y2 BY ... y3 BY ... z ON y1 y2 y3; I would use the factors in the model not the factor scores. The factor scores will not be the same as the factors unless factor determinacy is one. This is seldom the case. 

roachyang posted on Friday, August 03, 2012  11:57 am



I'm not sure about the difference between factors and factor scores. How can we obtain factors and then use factors for later regression analysis? Isn't factor scores the estimate of latent factors? Can you elaborate on this? Thanks! 

roachyang posted on Friday, August 03, 2012  12:09 pm



Also I'm baffled by the different results returned from the following two models, i.e. the coefficients of y on the x's: model: y1 by x1 x2 x3; y2 by x4 x5 x6; y3 by x7 x8 x9;  model: y1 by x1 x2 x3; y2 by x4 x5 x6; y3 by x7 x8 x9; z on y1 y2 y3; I thought the second model just add a new regression of z on y1 y2 y3 to the first model. What exactly is the difference between the two models? Sorry for my newbie questions. 


When you add z to the model, you say that z relates to the factor indicators only through the factors. This is a very strong statement. 

Back to top 