Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
APPENDIX G COMMENTARY ON THE FEASIBILITY OF SAMPLING Several of the alternative inspection programs evaluated by the committee (alternatives 2 and 3, in particular) are based on the premise that sampling procedures can be used under which-within each category of equipment-only a representative sample of the items need to be tested. ills, Al ill~ll~, all ~ given pia~lorm wiln iou nign-level sensors, the current inspection practice requires that all 100 sensors be tested while the inspector observes. The alternative programs are based on the proposition that with statistical sampling techniques the inspector need witness only a limited number of these 100 sensors to establish the reliability of all the sensors. The result of this approach is to free inspectors' time, which could then be put to use making more unannounced inspections, following up on weak spots, and carrying out essential tasks such as those described in Chapter 7. These are perceived by the committee as more effective uses of the inspectors' time than repetitive witnessing of routine tests. Under the alternative inspection programs the platform operator would continue to be responsible for performing all the routine tests, on the schedule MMS has prescribed; the thing that would be new is that now he would be required on demand to send records of any of these test results to the MMS (probably, he also would have to accommodate an MMS prescribed report format). The nature of MMS inspections, however would change; it now would verify that the operator was performing the tests and that his reported results are accurate, rather than witnessing all the annual tests as they are conducted by the operator. The purpose of sampling then, would be to verify operator performance of mandated tests. Such verification can be done using "statistical inference" methods. ~1~..^ p~ :_ ~ A:_ ~ ~~.~ '.1_ ~ ^^ t ' _~ ~ _ ~ .~ . For example, in his report to MMS the operator states that in testing for the past year the failure rate of High Level Switches (LSHs) was 0.01; however, MMS is receiving reports from all operators and is maintaining a data base from which MMS knows that the industry average failure rate for LSHs is 0.02. Moreover, MMS knows that the reported number falls in the top (best) fifth percentile of all platforms. MMS would now ask "does the operator's record in past inspections, his overall safety performance, and everything else we know about him support the assertion that now he is among the top five percent performers?" If not, MMS would plan to witness more tests during its onsite inspection than it would normally witness, i.e., it would establish a somewhat larger sample size for his platform. Suppose the MMS inspector, now on the platform, witnesses tests of a sample of 10 LSHs, chosen at random but with preference toward the "hard to get to" units, and one out of these 10 fails the test. Referring to statistical tables prepared on the basis of MMS data, the inspector sees 106
107 that he has reason to suspect that the operator's reported data are not accurate., Again referring to statistical tables, he determines that further verificaton of the operations perfo~ance is indicated, and orders 20 more tests, during which he finds 2 additional failures. He now orders a 100 percent inspection of LSHs and all other items on the PING list. If the operator is found to have been falsifying records, appropriate penalties would be applied, ranging from more frequent unannounced inspections to shut-ins, to criminal charges. The committee believes that it is feasible to develop valid sampling schedules for the industry with inspection increments based on the comparability of the operator's reports with his operating history and those of other operators, as well as test failures encountered in the course of the onsite inspection. With limited retraining current MMS inspectors can be trained to use sampling schedules. A changeover to a sampling procedure cannot and should not, however, be executed overnight. The sampling procedure should be phased in gradually; moreover, the preparation and initial implementation of sampling procedures will require the services of persons skilled in this kind of statistical inference. Certain failures, such as out-of-tolerance readings, do occur routinely. If an MMS inspector examines an operator's records, he would expect to see failures and corrective action recorded at about the statistical average for the industry. If he does not, he would look for problems with the operator's testing, recordkeeping, and/or reporting practices.