On Fri, 22 Nov 2002, David Goodman wrote: > I do not think that the comparison of the eventual value of the > different specialties of scientific research can be judged at the > time the research is being done.
"Can only be predicted with XX% reliability" is the statistically sound way of putting it. And both XX and the time-span will vary (with time, and field). Assessors for research funding don't ask for 100% predictive accuracy. They just want something like "Research/Researcher A is more likely than B" (when funds are finite). > That requires historical knowledge as well as scientometrics. Yes, but hindsight is not a predictor (unless it picks out a predictive pattern or index for the next time). > This does imply a certain humility about the ability to use current > knowledge as a valid basis for long term science policy. I don't know about long-term science policy. The RAE just wants some objective help in disbursing support for the next few years. http://www.ecs.soton.ac.uk/~harnad/Hypermail/Amsci/2373.html > Your second derivative technique, if the data are sufficiently > accurate to support it, sounds like an exceeding nice way of measuring the > potential short-term rise of a scientific field (or department). I would > be reluctant to extrapolate very far into the future with such methods. Extrapolate no further than your time-series correlations suggest you have a statistical basis for extrapolating. > For example, as judged by apparent current productivity, and its apparent > valuation by the scientific world in general, scientometrics does not > show very well. You and I know better, of course. :) The time-line for the betting on scientometrics is still very short, the field being new and its database growing. Its day is fast coming, though, and open-access (along with the scientometric analzers like http://citebase.eprints.org/ ) will help usher it in. Stevan Harnad
