In this approach, different data sets are merged into a mega-data set and then analyzed as if all data sets were derived from a single experiment. This approach ignores the inherent bias and heterogeneity of data sets from different sources.
Many other factors (experiment protocols, technical platforms, raw data processing procedures and so forth) can potentially contribute to the observed differences. This approach should only be used when data sets are very similar (i.e. from the same lab and same platform without batch effects).