210219 SE Diss, H: Testing Causal Hypotheses in Social Sciences (2008W)

This course focuses on testing of causal hypotheses with the use of Structural equation modeling (SEM). First the basic reasoning and the connected statistical analyses will be discussed for experimental, quasi experimental and non-experimental designs. After that the course will concentrate on non-experimental research. In that part the transformation of verbal theories in testable propositions will be given a lot of attention as an essential part of non-experimental research. After that estimation, testing and correction of causal hypotheses will be discussed.
After this general introduction of the SEM approach we will introduce a general model for structural equations which is a combination of an econometric simultaneous equation model and a factor analysis model. This general model allows the specification, estimation and testing of all known linear structural equation models discussed in the context of multivariate statistics. Specifications of this general model are the following models: regression models, simultaneous equation models, the canonical correlation model or mimic model, longitudinal models like the test retest, or simplex and quasi simplex models, time series models, factor models, measurement models, general structural equation models for latent variables and multiple group models for cross cultural research and other applications.
Often the observed variables contain measurement errors and systematic errors because the observed variables are not only effected by the variables they are supposed to measure but also by other variables. Structural equation modeling is especially developed to cope with the problem of measurement error. Therefore special attention will be paid to estimation of measurement errors, index construction and the analysis of data correcting for measurement errors.