Many models used in policy or systems analysis either cannot be validated in any fully adequate sense, such as by comparing them with actual data, or could adequately be validated but have not been. For example, in the area of combat analysis, the central models are arguably almost entirely unvalidated and most will never be susceptible to adequate validation. Nevertheless, such models are often used and can be used fruitfully, even though we have no theory for how to use them or how to interpret and place value on the results they produce. This paper takes a step toward providing such a theory by focusing on the logic that should govern the use of inadequately validated models and the costs and benefits of using them. To this end, it identifies and evaluates six legitimate uses to which such models can be put.
Hodges, James S., Six (or So) Things You Can Do with a Bad Model. Santa Monica, CA: RAND Corporation, 1991. https://www.rand.org/pubs/notes/N3381.html.
Hodges, James S., Six (or So) Things You Can Do with a Bad Model, Santa Monica, Calif.: RAND Corporation, N-3381-RC, 1991. As of June 15, 2021: https://www.rand.org/pubs/notes/N3381.html