RMP.Application:5.3 - Separate a delay forecast from an intervention effect

Suppose a separate set of 30 old quotations contains 20 late and 10 on-time quotations. The head approver was absent for 13 of the late quotations and three of the on-time quotations. The odds of lateness are initially 20/10 = 2. On learning that the approver was absent, the likelihood ratio is (13/20)/(3/10) = 13/6. Updated odds are 13/3, corresponding to probability 13/16.

This calculation describes the association in that table, under its use as the relevant probability model. It does not give the effect of appointing another approver. Absence may coincide with high workload or difficult cases; the proposed substitute may work differently. Even the forecast for future quotations needs the table to remain an appropriate basis.

To compare “approval delay”, “missing input” and “both”, an analyst needs the likelihood of the proposed records under those accounts, or a weaker contrast the available records can actually support. Accounts can allow the same observation with different probabilities. Conversely, available observations may fail to distinguish them. Do not fill “the hypothesis is false” with an arbitrary mixture of alternatives or force coexisting mechanisms into an exclusive pair.

Use statistical inference under an observation model (MMP.13), for the probability model and update; observations that distinguish model alternatives (MMP.16), for what an obtainable observation could discriminate; and intervention-effect identification (MMP.15), for whether the available comparison identifies the causal quantity. These are different supplied results. A plausible mechanism or larger probability is not yet a staffing decision.