VALIDATION OF A TARGET RANGE DECISION AID
Marvin S. Cohen · 1982
In passive target ranging aboard a nuclear attack submarine several analytical procedures exist f or estimating a single variable, and the e stimates which are obtained at a given time frequently do not agree. Typically, the Commanding Officer focuses on one or two of these estimates, thus failing to utilize additional information that is, in principle, available. To address this problem, an interactive Bayesian decision aid computes a single pooled estimate of target range, together with an assessment of its probable accuracy. The aid can operate either automatically or by incorporating subjective judgments concerning the validity of the e stimates being pooled. Quantitative tests with pre-recorded at-sea data (Rangex 1-78 and 1-79) evaluated the, aid with objective inputs, subjective i nputs, and a mix of the two. It was found that estimates computed by the pooling aid in all three of these modes were consistently more accurate than KAST, MATE, and Ekelund ranging procedures and the Command staff's unaided best guess as to target range. Maximum accuracy was obtained when Command staff subjective judgments were combined with objectively assessed parameters. Formal analysis of the sensitivity of pooling accuracy .to errors in inputs suggests that the advantage of pooling over reliance on a single estimate will be extremely robust over a wide range of conditions.