| ||||

SMABS 2004 Home Organization About Jena Sponsors Links Imprint / Contact SMABS Home |
## Contributions: Abstract## Methods of the analysis of data files obtained in social studies which have dimension greater than two
Standard methods of multivariate statistics provide data analysis for two-dimensional data files: raw data or correlations calculated from raw data. These are quite adequate when each case corresponds to a one-dimensional data file whose length is equal to the number of variables. However, questionnaires in the social sciences often have a more complex structure, i.e. a two-dimensional data file. Examples of such structures can be Osgood's semantic differential, Kelly's repertoire grids or matrices of similarity. As a result, there is the problem of the adequate analysis of a cube of data, or files of even higher dimensions. In this research we present various ways of three-dimensional data files analysis, pointing out their advantages and disadvantages. The most common approach is to average the whole cube of the data in any dimension, i.e. mathematically it is a projection of a three-dimensional array to a two-dimensional plane. However this results in a loss of information. Another approach is data presentation obtained by treating each case as a one-dimensional file, and then reducing data analysis to a standard method. The more complex two-dimensional structure of data is broken up in this case. Yet another way is layer-wise analysis of data file. The data filled by all subjects on all variables on the same object is regarded as one group. Other possibilities are to use multi group comparisons in SEM, or hierarchical data analysis. A further approach is to "split" the sample according to the similarities in responses, and then average the individual data matrices within the obtained groupings. For the analysis of data obtained as a result of longitudinal research by multivariate questionnaires it is recommended to use dynamic factorial analysis and LGM. |