There is welcome news about an old problem. For years, we've been getting only part of the story on clinical drug trials. The successful ones get published and touted, but others that didn't work out so well may never see the light of day. New developments, however, promise a long-awaited exposure of the negative results. Most scientific studies that examine a possible threat or benefit to the public health are repeated, sometimes by several different investigators. When high economic stakes are involved, someone is usually interested enough to perform a meta-analysis, pooling the results of all the published studies to test for significance. That's true for clinical trials, toxicity tests, and other studies designed to assess human risks. So far, so good. But a thoughtful statistician can spoil the fun: "Look, journals and scientists like positive results and get disappointed by negative results. So there's a problem—all the unpublished negative results lurking in those file drawers!" Thus, the fly in the meta-analysis ointment: It's likely that aggregated results from published papers constitute a biased sample.
展开▼