Discussant's response to "Using regression analysis to assist audit judgments in substantive testing";

              Discussant's Response to "Using Regression Analysis to Assist Audit Judgments in Substantive Testing"
William R. Kinney, Jr.
University of Texas at Austin
I'm very pleased to have the opportunity to discuss the updated version of STAR. Many of my papers have addressed problems in analytical review in auditing and particularly regression analysis as a tool. Hearing about STAR's revision was like hearing that an old friend hadn't died after all. Thus, it was with some enthusiasm that I accepted Raj's invitation to discuss the updated, interactive version of STAR with its new bells and whistles.
My comments are divided into four basic areas and are generally favorable toward the software and the approach. Rather than being overly technical, I will try to stimulate your thinking about STAR, provide some perspective, and assess where we might go from here. First is a brief history of STAR and some STAR-related regression analysis research in auditing. Second is an analysis of what's good about STAR and what's new in the current version, and third will be some areas that need elaboration or additional thought. Finally, there is an overall evaluation of STAR and its impact.
History
As many of you know, STAR, dollar unit sampling, and the AICPA's audit risk model were developed by Ken Stringer of the former Deloitte Haskins & Sells. I began my research on regression in auditing after a 1977 conversation with Jim Loebbecke. We were discussing his research on "combined attributes and variables" sampling which was related to Stringer's "cumulative monetary amount" version of dollar unit sampling. Jim said that he had based his efforts on the presumption that Stringer was probably right, so Jim took what he knew about CMA and tried to derive what he didn't. I decided that I would try the same approach for STAR.
Using Stringer [1975], I set out to derive what must be in a regression pack-age that could satisfy the requirements for a substantive test. My primary prob-lem was determining what Stringer meant by the "most adverse distribution of error." Stringer [1975] gave no clues but said that STAR was designed to be effective even under that most feared of circumstances. I finally decided that that must mean that the procedure was based on the sum of estimated misstate-ments, and therefore it didn't matter how misstatements were distributed. My solution appeared in Kinney [1979]. At a conference sponsored by DHS, I found out that I had not guessed correctly about STAR but still had a useful result.
Both STAR and Kinney [1979] use an upper precision limit (UPL) on error
136