

There are no standardized errors and ... 

Message/Author 

young park posted on Monday, January 07, 2013  1:46 pm



I am conducting EFA and CFA with categorical items (binary items) using large sample. I got two latent factors through two step. I made one path model. My simple syntax is as follow: usevariables are y3ses w wcluster x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 c1 c2 c3 c4; categorical are x1 x10; weight=w; type=complex; estimator=wlsmv; model pa_1 by x1 x2 x3 x4 x5; pa_2 by x6 x7 x8 x9 x10; read by c1 c2; narra by c3 c4; I got the good results about the above model; RMSEA=0.024, CFI/TLI=0.955/0.948, WRMR=1.048 Next, I would like to check the effect of SES (socioeconomic status) as a covariate on read and narra. I put it into the syntax; !covariate Read narra on SES Then, I did get part of results and model fits were acceptable (RMSEA=0.031, CFI/TLI=0.927/0.916, WRMR=1.202). But I did get the standardized coefficients, but did not get s.e. and pvalues. Here are my questions: Q1. In case of categorical and continuous variables, the WLSMV could be used in the model as an estimator. Is it right? Q2. In this case of WLSMV, what are model fit indices? Could we only check the pvalue, RMSEA, and CFI/TLI? Q3. In this case (there are no standardized errors and pvalues, but the standardized coefficients), what's wrong? Is it data or syntax problem? I really appreciate you in advance. 


1. Yes. 2. Yes. 3. With WLSMV and a model with covariates, standard errors are not available. 

young park posted on Wednesday, January 09, 2013  2:41 pm



Thank you so much. I have one more question. When a SES (socioeconomic status, a continuous variable, ranging from 2.25 to 3.00) variable was added to one of predictors (not a covariate), we only got std coefficients. There were no SE and pvalues. WLSMV was used as an estimator. After the SES variable was removed from the predictors, we could get STD coefficients, SE, and pvalues. What is the wrong? 


A predictor is a covariate. This is why you don't get standard errors and pvalues. 

young park posted on Monday, January 14, 2013  11:30 am



Hello Linda, I always appreciate you. I have questions about a model with covariates, using WLSMV. Although you gave me explanations, I am still confused. Sorry about it. My mplus syntax is as follow: Usevarables are y3ses w cluster x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 c1 c2 c3 z1 z2 z3 z4; Categorical are x1x10; Model !measurement model Pa_1 by x1 x2 x3 x4 x5; Pa_2 by x6 x7 x8 x9 x10; Lang by z1 z2 z3 z4; Read by c1 c2; Narra by c3 c4; !Path model Read Narra on pa_1 pa_2 Lang; !covariate Read Narra on y3ses; As I told you, I did not get any std errors and pvalues, adding ¡°y3ses (a continuous variable, ranging from 2.25 to 3.00) to a covariate in the model. Here are my questions: Q1. Is the reason we did not get std error estimates and pvalues that the ¡°y3ses¡± is a problematic variable? The y3ses ranges from 2.25 and 3.00 and is a negative and continuous variable. In our model, pa_1, pa_2, Lang, and y3ses are predictors and they are continuous. Q2. If not so, does not a model with covariates and WLSMV provide any std errors and pvalues regardless of types of variables (continuous, categorical, etc)? Q3. If an observed item as a predictor is not applied, can we convert the observed item into a factor with single indicator? Thank you so much. 


12. It is because the model has covariates. With WLSMV, standard errors and pvalues are not given for standardized results when the model has covariates. This does not indicate a problem. 3. A factor with a single indicator and residual variance of zero is identical to the observed variable so there is no reason to do this. 

Back to top 

