On 5 Aug 2003 06:46:40 -0700, [EMAIL PROTECTED] (Bob Roberts) wrote: [ snip, a lot of detail, including sample data ]
> The lawsuits are settled now, and the first wood pile is being cleaned > up. The It is estimated to be 3,000 tons or 12,800 pallets/crates. > > It was decided to sample 12 pallets/crates for aresenic. In selecting > the 12 pallets/crates, an attempt was made to select different-looking > ones. Each selected pallet/crate separately was ground into chips, > stirred, spread out, and a composite sample was collected. The 12 > composites were tested by the TCLP protocol, which mimics the chemical > leaching that occurs in a landfill. (Assume that the grinding, > mixing, composite sampling, and TCLP analysis was done correctly.) Several distinct issues come to mind, not entirely statistical. 1) Okay, even assuming that the TCLP analysis was done correctly, how valid is that analysis? Mimicking the leaching of a landfill sounds like a damn-poor criterion to me. What is the maximum As that is present? - is that reflected directly in the TCLP analysis? 2) Here is a question easier than validity... How good is the statistical reliability of measurements? How similar are the results when you split one of those samples into two parts, and do two TCLP analyses? 3) 12 pallets were selected, out of 12 thousand, and they were "different looking" ones. That sounds like a noble intention, but using grossly non-random selection might raise more problems than it solve. Is that all that was done, or were the 12 considered as part of an "experiment"? - Was there an attempt, for instance, to rate the 12 as to (a) their mutual, expected similarities, or (b) their singular, expected Arsenic levels? Here is a gross problem. The 12 were not random, and you don't know what hidden agenda might have existed behind their selection; nor whether the *intentions* of a hidden agenda could have, even conceivable, mattered. 4) *Some* amount of random sampling surely needs to be done (as distinct from the stratified sample, so far). Some amount of knowledge is needed (Do you have it) about the quality of those assays. Some knowledge is needed about whether the "eyeballing" works at all, in classifying pallets into sets that are apt to be more or less hazardous -- else, there was no usefulness in doing the selection of 12, was there? - I think it should be good to find ways to categorize the pallets, and then to investigate "main effects" systematically. Will there be surprises in the sample? Or, What is the *size* of the surprise that might arise? I think that you can't make a very good guess if the process has never been attempted before. >From the pieces of experience of mine, half-related to that, I have some opinions based on the dozen "test results" that were presented. The 5 scores that were above 1.0 were high enough to be 'worrisome' when the criterion is 5.0, the relations seem logarithmic, and the range is already wide. If dozens of scores were selected on the same basis as those, some of the dozen probably would exceed 5; maybe even exceed 25. Now, some 'decision' questions: How serious is a single 5? single 25? How predictable are those high-scoring pallets, based on visible characteristics or pallet-history? To me as an environmentalist, it seems that you do have to distinguish between "good science" and "corporate science" for the basic facts. You have to fight further if you hope to get extrapolation. Good luck. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html "Taxes are the price we pay for civilization." Justice Holmes. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
