*** Rest of World Probability of Area Estimations *** From: To: Cc: Subject: Modified Mattson needs revision Date: Thu, 11 Dec 1997 12:23:44 -0400 (AST) Quoting Mike Dolan (very much out of context) from the SAR 2000 discussion list: >I am very in favor of the "Modified >Mattson" and not the straight Mattson. I assume you mean the so-called O'Connor method. I criticized this approach at the Denver meeting of Response '96. It has a deadly flaw which I'm surprised was not detected long ago. Briefly, when using the A-H rating scale to assign POA, the resulting POA for the ROW (rest of the world -- i.e., the chance that your search area does not contain the subject) will be arbitrarily affected by the number of segments you decide to divide your search area into. Think about it. Given two search areas of equal size, the one having more segments will have a ROW with a smaller POA. This shouldn't be. Using the O'Conner method, you could effectively reduce the POA for the ROW to nearly zero by parcelling the search area into many tiny "segments." The solution is to assign a numerical (percentage) probability to the ROW before doing the rating of segments. It requires only a minor modification in the algorithm. I'm including this observation in a general discussion of the POA issue for Response magazine. An omnibus SAR management software program (CASIE IV?) should also provide the option of employing the "cumulative probability of success" model introduced by Don Cooper and Jack Frost at the recent Syrotuck Symposium. I very much disagree with this model -- because it abandons the inclusion of the ROW "segment" in POA computations -- but it's a viable approach which has been used by the Coast Guard for years. It's application to ground SAR remains to be seen. Regards to All. Ken Hill -------------------------------------------------------------------------- Kenneth A. Hill, Ph.D. email: khill@husky1.stmarys.ca Professor of Psychology Home URL: http://husky1.stmarys.ca/~khill Saint Mary's University SAR URL: http://www.sarbc.org/k-hill.html Halifax, Nova Scotia Canada B3H 3C3 -------------------------------------------------------------------------- Date: Thu, 11 Dec 1997 11:30:42 -0700 From: Ed Cornell Subject: Re: Modified Mattson needs revision To: Ken Hill Ken, You might suggest in your forthcoming article that search managers begin recording or examine records to obtain empirical estimates of the ROW. That is, we can estimate ROW based on the case histories of searches.  For each case history in his or her files, the search manager examines the area designated for search in the first operational period, then determines whether the subject was found in that area.  The search managers in wilderness areas in southwestern Alberta recently asked us to do an independent analysis of their files.  We found that in 131 of163 case histories the subject was found in the search area as defined in the first operational period. Hence, for search operations in these areas, I suggest that the appropriate starting value for the ROW is 20%. While some commentators have interpreted this as the search manager being wrong 20% of the time, my interpretation is that the search manager has shown a 80% hit rate on first response.       Of course, the analysis of actual occurrence of the subject in the ROW could be repeated for searches that included second and subsequent operational periods. These histories could provide some real-world guidance for the deployment of resources in more extended operations. The hit rates are important figures to have when arguments are being made for funding cuts. As you have pointed out in your talks, Ken, there are several tragic histories of operations that continued to search in initially defined segments without expansion into bordering areas.  Case histories of the frequency of locating subjects in the ROW may help to determine when search is inefficiently perseverative. Cheers, Ed Cornell ------------------------------------