Table 1.

Terminology for key steps in Diagram-based Analysis of Causal Systems

Type of diagram	Definition	Use
Conceptual causal diagram	A causal diagram developed using a priori knowledge, and supplemented by assumptions where necessary. This describes the whole system, and includes all potentially important proximal and distal determinants of a health outcome, potential pathways connecting them, and potential confounders and/or selection effects. All variables are in their conceptual or ideal form.	A conceptual causal diagram provides the underlying conceptual framework for statistical analysis, independent of the specific statistical approach or approaches chosen.
Operational causal diagram	A causal diagram derived from the conceptual diagram, and informed by data availability as determined by one data source. All variables are in their actually measured form, which may include proxies depending on the data source. Pathways connecting them may be testable (where all relevant variables are measured) or conceptual (where some or all relevant variables are unmeasured). Depending on the available data and approach to analysis, distinct versions of the operational causal diagrams may be developed for different data sources or datasets* (e.g. for different countries, settings or years).	An operational causal diagram provides the basis for statistical analysis, shedding light on specific statistical approaches that may be applied to a given data source.
Integrated causal diagram	A causal diagram derived from the conceptual diagram, as informed by empirical testing across more than one data source. As for operational causal diagrams, variables and the pathways connecting them may be in their actual or conceptual form; actual variables and the pathways connecting them derive from more than one data source.	An integrated causal diagram provides a current summary of knowledge about the whole system, illustrating causal pathways that are well-supported versus causal pathways where evidence is lacking.

* We use “data source” to refer to different types of data (e.g. DHS vs. WHS data), and “dataset” to refer to the same type of data being available for different settings (e.g. DHS data for Benin vs. DHS data for Ethiopia).