What integrity-test evidence can and cannot establish

Research context

The updated integrity meta-analysis reported different relationships depending on criterion source and study authorship. A subsequent commentary highlighted incomplete documentation in reconciling competing syntheses. A small observational study also illustrates that a particular integrity measure need not predict a particular honesty behavior.

Machiavelli development analysis

The most important distinction for MIO is between evidence about a class of instruments and evidence about this instrument. A review may combine several established tests, participant groups, settings, and outcomes. Its average relationship is not a property that transfers to any new questionnaire carrying the same label. MIO uses different wording, a different purpose, and an explicitly voluntary setting. Its predictive relationships are unknown. Research citations should help readers understand the questions that matter, rather than function as borrowed credentials for an untested product.

A second distinction concerns the outcome. Self-reported conduct, a supervisor’s impression, and an administrative record each capture different information. A record may miss unobserved events; an impression may reflect opportunity to observe; a self-report may share response tendencies with the predictor. None is automatically a perfect benchmark. The proposed validation program would define a narrow outcome before collecting data and explain why it is relevant to reflection. For example, independent clarity ratings of a project handover are more closely aligned with disclosure practice than an unrelated global performance score.

Disagreement among reviews is scientifically useful when the sources of disagreement are traceable. The MIO evidence register should therefore identify inclusion criteria, sample characteristics, score versions, correction choices, and uncertainty. A future technical report should show observed relationships before any statistical corrections and explain the assumptions behind corrected estimates. Selectively presenting the largest coefficient would hide information a careful reader needs. If an expected relationship fails to appear, that result belongs in the record alongside findings that favor the framework.

An original illustration is a pilot in which participants complete MIO and later rate the clarity of their own promises. A strong association might look impressive, yet both variables could reflect a general tendency to describe oneself positively. A stronger design would ask a consenting collaborator to evaluate a specific agreement without seeing the MIO report. Even then, the result would concern that agreement and sample. It would not establish that the score detects dishonesty elsewhere, and participants should not face employment consequences for joining the study.

Until such studies exist, the honest product description remains development framework. Reliability coefficients, validity estimates, and claims of prediction must not be invented from the length of the item bank or the stature of its references. The scientific contribution at this stage is explicit specification: what is proposed, why it might be useful, what alternatives could explain the answers, and what observations would count against it. That transparency gives professional readers something substantive to evaluate while preserving the difference between an informed hypothesis and a supported assessment claim.