categorical data analysis for r-avoiders精品.pdfVIP

Categorical Data Analysis (for R-avoiders) Christoph Scheepers Glasgow Intro Linear Mixed Models (LME)  Discussed in two (orthogonal!) contexts  Simultaneous generalisation of effects across subjects and items (better alternative to calculating min. F’ from F1 and F2, cf Clark 1973)  Providing adjustments (i.e. distribution and link functions) for a wider range of analysis problems, including categorical data analysis (cf. Jaeger, 2008, JML) For the categorical data problem, alternatives to LME are available!  Hierarchical log-linear models  Generalized Estimating Equations  Logistic Regression  Etc. Overview Block 1  Categorical data analysis using hierarchical log-linear models in SPSS/PASW  Advantages/disadvantages using this approach  Examples/Demos Block 2  Categorical data analysis using Generalized Estimating Equations (GEE) in SPSS/PASW  Very similar to LME, actually  distribution link functions  Inclusion of continuous predictors (covariates)  No „random effects‟, but instead classical distinction between “within” and “between” factors (mixed designs are no problem)  Examples/Demos Overview Block 3  Linear Mixed Effects Models (LME) in SPSS/PASW  In SPSS/PASW, no distribution/link functions as yet (hence, can only be applied to normally distributed continuous data)  BUT, can be used to address the subject/item generalization issue  Specifying „proper‟ LME random models  Comparison to classical F1, F2, min. F‟ approach  Potential problems Log-linear Analysis Christoph Scheepers The Problem Measurement at nominal scale level  e.g.,