The standardized residuals given in tech10 are not the Pearson residuals but are the standardized Pearson residuals. See Agresti's Categorical Data Analysis book, Sections 3.3.1 and 4.5.5. The original article on this topic is The Analysis of Residuals in Cross-Classified Tables, Shelby J. Haberman, Biometrics, Vol. 29, No. 1 (Mar., 1973), pp. 205-220.
The advantage of the standardized Pearson residuals is that they are just like standard normal residuals, while the unstandardized Pearson residuals are less variable. For the bivariate tables these are computed by (O-E)/[sqrt(E)*sqrt(1-E/n)].
QianLi Xue posted on Sunday, December 07, 2008 - 6:25 pm
Is it true that when sampling weights are used in LCA fitting, the Pearson Chi-square statistics for model fit in the general output and for response pattern frequencies in TECH10 are no longer valid? Is that why some of the Chi-square values in TECH10 become negative?
Hi, I am running a latent class analysis with 10 binary indicators and sample weights. If the tech10 output (the bivariate residuals) does not take account of the sample weights (as mentioned in the post above) how can I assess model fit? thanks
Hello. I am running a GMM negative binomial model and I would like to examine the standardized residuals; however, the section of the output where the tech 10 results should appear is blank (except for the heading "technical 10"). Is it possible to obtain standardized residuals with the negative binomial? Thank you.
Dear Dr. Muthen, I am running a CFA model using MLR for parameter estimation. The data structure includes a booklet design (i.e., data missing at random), and both 72 categorical indicators (binary responses indicating an ability factor) and 72 continuous indicators (log-transformed response times indicating a slowness factor). To obtain model fit information for the categorical dependent variables in the model, I wanted to look at TECH10 for the standardized residuals. However, with a higher number of indicators the TECH10 Output disappeared. When I used 6, 9, or 15 indicators per factor TECH10 was regularly provided. With about 30 Items per factor it wasn’t. How do I get TECH10 in this case? Thank you in advance! With kind regards, Annette
Daniel Lee posted on Friday, April 28, 2017 - 11:17 am
Hi I'm trying to estimate standardized residuals for a path model. The independent variable is a manifest variable, mediators are latent variables, and the dependent variable is a manifest dichotomous variable.
I was wondering if fit indicators were available when using ML as estimator (logit link) and realized that it was not.
I was told that you can compared observed vs. model estimated frequencies and standardized residuals. When I included tech10 in the output line of the .inp file, I received an error message:
TECH10 OUTPUT FOR CATEGORICAL VARIABLES IS NOT AVAILABLE FOR MODELS WITH COVARIATES.
The error would not go away when I removed the covariates from the model (so all that was left was the one IV, 2 latent variable mediators, and the binary dependent variable). I would appreciate assistance and getting tech10 to work. Thank you!
TECH10 would be relevant if you have more than one categorical DV. ML with categorical outcomes does not have fit indices. You can try Bayes which uses probit but the power of it test of model fit has low power. One approach in the ML context is to create neighboring models that are less restrictive than the model you consider and then check if the extra parameters are significant.
I am trying to compare a series of CFA models with 60 categorical indicators, and therefore using MLR estimation and montecarlo integration. The chi-square test cannot be computed because the frequency table for the categorical variables is too large. So, I included TECH10 in the output, but the TECH10 output for the categorical variables is not available because the frequency table is too large (but there is model fit information). Do you have any suggestions/recommendations for how to compare models in this case? Also, could you point me to resources to interpret the TECH10 output that is available? Thank you!
You want to look at the standardized residual column which gives z-tests, so you look for values greater than 1.96 in absolute value. This is for each response pattern, for the univariates, and for the bivariates.