Data structure and related methods for pooled meta-analysis

Some questions in statistics can only be answered with a multi period setup (e.g. longitudinal studies). A widely used approach is called meta-analysis. This paper uses the variant pooled metaanalysis and shows data structures and methods for performing such an analysis. Pooled meta-analysis requires concatenating the primary data sets. If variables have different specifications over periods, then this concatenation is difficult. Therefore a data structure for describing the variables from different periods or studies and their codes is necessary. Methods for generating data sets complete these data structures. The shown data structure is founded on a description of each real variable. This description includes original question, coding and specification, range, measurement unit, categories and dependencies between the variables. The related method analyzes the dependencies between the variables and is able to generate a data set for further analysis which holds the data from selected variables from selected periods. Constraint violations such as missing value codes or values outside specification can be recognized. The shown algorithm allows easier handling of variables in complex data structures for pooled meta-analysis.