CICLO DI SEMINARI

Giovedì 28 e venerdì 29 aprile, presso il Dipartimento di Economia e Territorio dell’Università di Cassino, via S. Angelo località Folcara, aula 1.05 (I piano)

AB MOOIJAART

Department of Psychology, Faculty of Social Sciences, Leiden University, The Netherlands

terrà un ciclo di seminari dal titolo “STRUCTURAL EQUATION MODELS”, secondo il seguente calendario:

Giovedì 28 aprile (ore 10-12 e 14.30 - 16.30)

Structural Equation Models and model selection

ABSTRACT. In this talk the basic characteristics of Structural Equation Models (SEM) are presented. In fact, SEM is dealing with an extension of regression model. A very important feature of this extension of regression is that some variables may be latent, i.e. unobserved, variables. In general in SEM there are two parts: the first part consists of the relationship of the observed (manifest) variables and the latent variables. This part is called the “measurement” part of the model. The second part consists of the relationship between the latent variables. This latter part is called the “structural” part of the model. An important issue in this talk is the choice of a proper model. In some cases we can use statistical test statistics for selecting a model. In other cases we will show what kind of model fit indices there are to decide whether a model fits the data or not. We will present different SEM models like factor analysis models, models for multiple group data, latent growth models, and multi level models.

Venerdì 29 aprile (ore 10-12 e 14.30 - 16.30)

Structural Equation Models for non-normal variables

ABSTRACT. This talk is a follow-up of the talk “Structural Equation Models and model selection”. In the talk about SEM and model selection the emphasis was on models for variables which are statistical normally distributed. In practice, however, variables are (mostly) not normally distributed. Two common situations are: the variables are categorical, which are obviously not normally distributed; or the variables are continuous, but the distribution of the scores is far from normally distributed. Both situations will be discussed in this talk.

We will also discuss how to deal with situations in which the variables are non-normally distributed. We will show, e.g., what the consequences are for applying SEM in those situations. Also we will show in what kind of situations non-normality does not harm the solutions.

For situations where non-normality of the scores has a serious impact of the results of the analyses, we will show how to deal with those situations. Basically, there are two methods for dealing with these problems: using (asymptotic) distribution free method or using some resampling method. In this talk we will illustrate how to use the bootstrap method, either for investigating the stability of the estimates of the model parameters, or for making the decision whether a model fits the data or not.

Tutti gli interessati sono invitati a partecipare

Per informazioni: http://stat.unicas.it link “seminari, convegni e giornate di studio”.

Claudio Conversano: c.conversano@unicas.it (0776 – 2994665)

Giovanni Porzio: porzio@eco.unicas.it (0776 – 2993448)