Level 0 (L0): Collection of raw data sets accessible for each shared study. At this level there is no harmonization beyond the one produced by design, and data sets may be presented in different formats, files, schemas, and so on, as different software are used to collect and store the original data. This level contains the maximal information possible among the considered data sets. |
Level 1 (L1): This data set contains a combination of all the variables from the original data sets after formatting to a common data model (CDM). Harmonization is attempted at face value, aligning variable name of those variables and instruments that might be comparable by design. For instance, demographics and standardized measures are likely to be pooled at this level. At this stage, no variables are dropped, producing a sparse data set. This allows for new data sets to be incorporated in future efforts. Codifying for missingness such as ‘not collected’ would provide further information on potential harmonization steps in subsequent levels. This is the minimal level of harmonization feasible, with no loss of information. |
Level n (Ln): Subsequent levels build on the previous level (L2 from L1) by considering further transformations of the data such as aggregation to common denominator or statistical transformations, or deriving new harmonized variables as needed. As transformations are produced, the level of harmonization increases at the expense of losing information. |
Level final (Lf): The final level of harmonization is reached when no more transformations are possible and the nonharmonizable variables are dropped. Theoretically, there are different points at which the process can be considered at the maximal level of harmonization (i.e., new derivative variables are almost always possible). Practically, a researcher may want to set a criterion beyond which no further harmonization is attempted. The maximal harmonization level is achieved when only the equivalent variables are considered, producing the maximal loss of information. |