Summary.
We strongly advocate a multivariate approach in the
analysis of categorical times series data. Question arises if there
is such a possibility with this type of data. How can researchers go
beyond uni- and bivariate relationships among variables and include
three and more variables in analyses? We answer the question by
relying on configurations. We present in this section how
configurations are build and represented.
The building block of our strategy is the configuration. We define configurations as a multivariate combination of many variables having emergent properties [MorinMorin1977,HakenHaken1983]; that is, properties not found at the micro (univariate) level. Such configurations are usually composed of 2 to 8 categorical variables, so to keep representations and analyses manageable and understandable.
Although we developed a dynamical configural methodology on our own, investigation of literature shows that configurations is already an established concept. Configural frequency analysis [Krauth LienertKrauth Lienert1973,EyeEye1990,Eye, Spiel WoodEye 1996] is a method for analyzing the individual cells in contingency tables. Contrary to the usual contingency table analyses, which describe relationships between variables , Configural Frequency Analysis focuses on the particular cells , thus on the characteristics of subjects.
By concentrating on cells of contingency tables, Configural Frequency Analysis searches for types and antitypes . Types are cells exhibiting frequencies that are statistically more frequent from what would be expected given a particular model (usually a model of complete independence between variables). Antitypes are those cells that are statistically less frequent than expected (given the same model). Both types and antitypes bring to the researcher's attention exceptional combinations of variables, because they appear more or less frequently than expected.
One of the main advantage of the configural approach is that it makes no distinction between dependent and independent variables. Experience Sampling Method (ESM) researches seldom make such distinction [DelespaulDelespaul1995,BrandstätterBrandstätter1991]. Bio-psycho-social variables are so deeply intertwined, each of them influencing - and being influenced by - the other ones that it does not make sense to specify which variables come first in the global causal path. Configurations put them all on the same level.