

CFA. missing data by design (cat vari... 

Message/Author 

Julio Cesar posted on Wednesday, April 10, 2013  12:21 pm



Hi Dr. Muthen and Muthen, In an earlier posts: USER: "Is it correct that it is not possible to use the pattern statement for defining data missing by design when using categorical variables in an EFA or CFA? If so, is there a way to get around this other than treating the variables as interval variables? You had stated: "I am assuming that you must have tried this and gotten an error message to that effect. This seems possible given that missing was originally developed for continuous outcomes. You can get around this by doing your analysis in two steps. In the first step, do not treat the variables as categorical. Use the PATTERN option and do a TYPE = MISSING BASIC; in conjunction with the SAVEDATA command. In the second step, analyze the saved data while declaring the outcomes as categorical." By the SAVEDATA command, did you mean to use: SAVEDATA: FILE IS newdata.dat; I have used that and followed your second step where i no longer use the pattern variable (originally used in the first step to get the new data). And I am not getting some Model Fit Statistics (i.e., RMSEA, CFI, etc.), but i do get everything else including AIC, BIC, and parameter estimates. Do they just not print? Do you know if i can get them? 

Julio Cesar posted on Wednesday, April 10, 2013  1:02 pm



Sorry, I should have been a bit more clear as to what I am trying to accomplish. I am running a CFA with 80 dichotomous variables with missing data by design. Meaning that the first 20 variables are shared by the whole sample, the following 20 are given to just one group, the next 20 are given to the second group, and the last 20 are given to the third group. Each group has a total of 40 items, with 20 shared. I am running a one factor CFA with FIML (ML) estimation. I am unable to get Model Fit indices for the model and thus do not know if my categorical data are fitting the model or not. Thanks in advance for any recommendations. 


When means, variances, and covariances are not sufficient statistics for model estimation, chisquare and related fit statistics are not available. With categorical and maximum likelihood, this is the case. Please limit your future posts to one window. 

Julio Cesar posted on Wednesday, April 10, 2013  3:27 pm



Thank you for your response Dr. Muthen. I have followed additional posts (yours) that have provided with the needed information: RMSEA = sqrt((chi/(n*df))  (1/n)) CFI = 1  ((max(chitdft,0))/(max(chitdft,chibdfb,0))) TLI = (chib  (chit * (dfb/dft)))/(chibdfb) GFI = 1(chit/chib) (t = tested model; b = base model). I have moved on to do a CFA group comparison (male/female) of the above mentioned data (missing by design with categorical observed data) by looking at four diff types of invariance tests: 1) Configural Invariance; 2)Weak Invariance; 3) Strong Invariance; and 4) Strict Invariance. As an example, I have tried using the ML estimator (since i want to use the FIML estimator bec of the missing by design data) by using: DATA: FILE IS data.dat; VARIABLE: NAMES ARE y1y6 g; USEVARIABLES y1y6 g; GROUPING IS g (1=male 2=female); ANALYSIS: ESTIMATOR = ML; MODEL: f1 BY y1 y2 y3; MODEL female: f1 BY y2 y3; [f1@0]; [y1y3] But I get an error message that reads that group comparisons can not be done with the ALGORITHM=INTEGRATION. I tried changing it to ALGORITHM=EM, but without success. Do you have any suggestions as to how I ought to proceed? 


Please send the output and your license number to support@statmodel.com. 

Back to top 

