Match the proposed action and the evidence to the question

Turning a blank field green changes an indication. Correcting and applying the identifier correspondence changes how information is handed over. C.28.MR can derive their different consequences once the mechanisms and intervention are specified. C.28 separately asks what supports the resulting causal claim. A calculated consequence does not establish that the chosen mechanism describes this workplace.

A qualitative distinction may suffice. If the decision needs a quantitative causal effect, MMP.15 asks whether the effect is identified under the model and available information. Its adjustment, front-door and transport examples develop different ways to obtain or limit an answer.

An externally assigned offer of help is not the same variable as help actually used. For an instrumental-variable argument, MMP.15:4.3.1 derives the binary local effect and :5.4 works through the difference between offer, local-use and population effects. Hernán and Robins, Causal Inference: What If, Chapter 16, §§16.1–16.6, provide a fuller discussion of the assumptions and limits. Return with the target effect, the proposed instrument, the grounds for its assumptions, the identifying expression or remaining bounds, and the uncertainty relevant to this decision. For a binary offer and binary use, the difference in mean outcomes between offer groups divided by the difference in use rates is the Wald ratio. Relevance, exclusion and instrument exchangeability are necessary for that argument; those conditions alone do not identify the population average effect. With monotonicity and the chapter’s consistency and interference conditions, the ratio identifies the average effect among people whose use changes because of the offer. These people are not simply everyone who accepted help, and that local effect need not answer what universal use would achieve. An invitation that teaches the work independently of use breaks exclusion; a weak change in use makes the ratio unstable. Neither an external-looking event nor four clarified rows supplies the required premises or adequate precision. If this conditional result does not answer the decision, return to the target, information or feasible intervention instead of relabelling it as a population effect.

If an unresolved question still warrants research, RMP.1 frames the exact question and current basis. RMP.2 compares a useful, feasible design with using the evidence already available and selecting no new study. For a rule comparison, the design must address the load, information availability, timing and selection differences that could defeat its inference. A plan for such a comparison establishes no result of performing it.