Why your causal diagram should probably include sampling and measurement processes

Rob James Boyd, Rob Cooke, Gary D. Powney, Oliver L. Pescott 

Insect scientists are starting to use causal diagrams to display assumptions about causal relationships between variables that exist before any data have been collected. The perception appears to be that these assumptions are sufficient to determine whether observed associations between variables can be interpreted causally. But an observed association implies an observation process (sampling and measurement), and assumptions about that process are also required. We draw on the literature from other disciplines to explain how insect scientists can incorporate assumptions about sampling and measurement in causal diagrams. Making these assumptions explicit allows the investigator to reason more holistically about whether observed associations can be interpreted as causal effects. It also reveals that causal diagrams are not just a tool for causal inference. Assumptions about sampling and measurement are needed to answer descriptive and predictive questions as well. Hence, causal diagrams that incorporate these processes provide a general framework for displaying assumptions regardless of inferential goal.