THE PRECAUTIONARY PRINCIPLE PROJECT
ONLINE CONSULTATION
| E-Conference Homepage | Contents |
| Next comment | Previous comment | Previous comment in thread | Next comment in thread |
From: Daniel J. McGarvey
Affiliation: Ph.D. Candidate, The University of Alabama
Remote Name: dhcp-108.biology.as.ua.edu
Time: 03:53:48 +0100
With regard to “the decision to collect additional information”, Mr. Ryan Hill and Mr. Perran Ross each provide valuable insight to a crucial problem: the dearth of reliable, scientific information that tends to be available when important regulatory decisions must be made. Sometimes, when uncertainty abounds, “a little science” is no more valuable than “no science.” This is due to the unbalanced nature of conventional inference testing, which places the burden of proof upon the null hypothesis (i.e., the theory one is trying to prove, such as the deleterious effects of some human endeavor upon the environment); in many cases, it is impossible to “prove” an alternative hypothesis without a minimal amount of empirical data, regardless of how true that hypothesis may, in fact, be. Statistical power analysis provides one potential, completely objective tool for answering the all important question: when should we demand additional information before making a decision? Power analysis allows one to calculate the probability of failing to detect a significant effect. In other words, it is an indicator of the “quality” or “reliability” of one’s dataset; when statistical power is low, one cannot confidently conclude that an effect does not exist. This paradox is illustrated by the National Research Council’s (United States of America) recent assessment of two federally endangered fishes within Upper Klamath Lake (southern Oregon). When the National Research Council concluded that, based upon nine water quality data points, “there is presently no sound scientific basis” for modifying the status quo operations of a federal reclamation (i.e. irrigation) project, it failed to consider the fact that the data at hand provided very little statistical power. However, post-hoc power analysis of those same data demonstrates that one is approximately 20 times more likely to fail to detect a significant effect (of reclamation water demands upon the endangered fishes) than to successfully detect such an effect. (Complete details of this analysis are provided in: McGarvey, D. J. and B. Marshall. 2005. Making Sense of Scientists and “Sound Science:” Truth and Consequences for Endangered Species in the Klamath Basin and Beyond. The Ecology Law Quarterly 32(1) – in press.) I agree with Mr. Ross, in that an inflexible criterion for additional data, in the face of uncertainty, is not a tenable application of the Precautionary Principle. But neither is the unexamined and unchallenged assumption that additional data are not necessary. Therefore, consistent with Point #3 of the Draft Guidance, I believe that a case-by-case review of the adequacy of existing data is critical. Moreover, power analysis should become a standard protocol in the review of scientific studies.
[_borders/disc6_aftr.htm]