I am a doctoral student and my theses is to explore the relationship between hospital coding practice and data quality.
Hospital coding practice is a latent variable, measured by an ongoing national questionnaire survey of 930 hospitals. The questionnaire was developed based on findings from case study. Factor Analysis will be sued to analyze and transform 50 Likert-type questions to be a new scale for hospital coding practice construct.
Data quality is latent outcome variable. I will use the national audit database, which contains medical record audit results of almost 60,000 patients in 930 hospitals. The error code variable in that database is categorical (e.g. Code 1 = Incorrect Principal Diagnosis, Code 2, Incorrect Procedure Information) and each patient record can have many types of error. Perfect record will have Code 0 (no error). Latent Class Analysis will be used to classify hospitals into classes.
Also, I would like to account for multilevel nature of data (3 levels: patients in the same hospital and hospitals in the same province).
Is it ok to use SEM for this kind of multilevel analysis? If so, can M-Plus accommodate it?
As for the hospital coding, Mplus can handle 2-level factor analysis of Likert-type items.
As for the data quality, it sounds like you have nominal items. Mplus can handle 2-level latent class analysis of nominal items. The top level, province, can be handled either via dummy variables, multiple group analysis, or with many provinces, as Type = Complex Twolevel. There is a 2-level latent class analysis example on our web site under Paper, Multilevel Mixture Modeling:
Henry, K. & Muthén, B. (2009). Multilevel latent class analysis: An application of adolescent smoking typologies with individual and contextual predictors. Forthcoming in Structural Equation Modeling.