TY - THES
T1 - Permutation distribution clustering and structural equation model trees
A1 - Brandmaier,Andreas Markus
Y1 - 2012/01/24
N2 - The primary goal of this thesis is to present novel methodologies for the exploratory analysis of psychological data sets that support researchers in informed theory development. Psychological data analysis bears a long tradition of confirming hypotheses generated prior to data collection. However, in practical research, the following two situations are commonly observed: In the first instance, there are no initial hypotheses about the data. In that case, there is no model available and one has to resort to uninformed methods to reveal structure in the data. In the second instance, existing models that reflect prior hypotheses need to be extended and improved, thereby altering and renewing hypotheses about the data and refining descriptions of the observed phenomena. This dissertation introduces a novel method for the exploratory analysis of psychological data sets for each of the two situations. Both methods focus on time series analysis, which is particularly interesting for the analysis of psychophysiological data and longitudinal data typically collected by developmental psychologists. Nonetheless, the methods are generally applicable and useful for other fields that analyze time series data, e.g., sociology, economics, neuroscience, and genetics. The first part of the dissertation proposes a clustering method for time series. A dissimilarity measure of time series based on the permutation distribution is developed. Employing this measure in a hierarchical scheme allows for a novel clustering method for time series based on their relative complexity: Permutation Distribution Clustering (PDC). Two methods for the determination of the number of distinct clusters are discussed based on a statistical and an information-theoretic criterion. Structural Equation Models (SEMs) constitute a versatile modeling technique, which is frequently employed in psychological research. The second part of the dissertation introduces an extension of SEMs to Structural Equation Modeling Trees (SEM Trees). SEM Trees describe partitions of a covariate-space which explain differences in the model parameters. They can provide solutions in situations in which hypotheses in the form of a model exist but may potentially be refined by integrating other variables. By harnessing the full power of SEM, they represent a general data analysis technique that can be used for both time series and non-time series data. SEM Trees algorithmically refine initial models of the sample and thus support researchers in theory development. This thesis includes demonstrations of the methods on simulated as well as on real data sets, including applications of SEM Trees to longitudinal models of cognitive development and cross-sectional cognitive factor models, and applications of PDC on psychophysiological data, including electroencephalographic, electrocardiographic, and genetic data.
KW - Cluster
KW - Strukturgleichungsmodell
KW - Entscheidungsbaum
KW - Data Mining
KW - Zeitreihe
CY - Saarbrücken
PB - Saarländische Universitäts- und Landesbibliothek
AD - Postfach 151141, 66041 Saarbrücken
UR - http://scidok.sulb.uni-saarland.de/volltexte/2012/4545
ER -