On Wed, Oct 8, 2008 at 7:57 AM, Valdez, Bill <bill.val...@science.doe.gov> wrote:
> the primary reason that I believe bibliometrics, innovation > indices, patent analysis and econometric modeling are flawed is that > they rely upon the counting of things (paper, money, people, etc.) > without understanding the underlying motivations of the actors within > the scientific ecosystem. There are two ways to evaluate: Subjectively (expert judgement, peer review, opinion polls) or Objectively: counting things The same is true of motives: you can assess them subjectively or objectively. If objectively, you have to count things. That's metrics. Philosophers say "Show me someone who wishes to discard metaphysics, and I'll show you a metaphysician with a rival (metaphysical) system." The metric equivalent is "Show me someone who wishes to discard metrics (counting things), and I'll show you a metrician with a rival (metric) system." Objective metrics, however, must be *validated*, and that usually begins by initializing their weights based on their correlation with existing (already validated, or face-valid) metrics and/or peer review (expert judgment). Note also that there are a-priori evaluations (research funding proposals, research findings submittedf or publication) and a-posteriori evaluations (research performance assessment). > what,,, motivates scientists to collaborate? You can ask them (subjective), or you can count things (co-authorships, co-citations, etc.) to infer what factors underlie collaboration (objective). > Second, what science policy makers want is a set of decision support > tools that supplement the existing gold standard (expert judgment) and > provide options for the future. New metrics need to be validated against existing, already validated (or face-valid) metrics which in turn have to be validated against the "gold standard" (expert judgment. Once shown to be reliable and valid, metrics can then predict on their own, especially jointly, with suitable weights: The UK RAE 2008 offers an ideal opportunity to validate a wide spectrum of old and new metrics, jointly, field by field, against expert judgment: Harnad, S. (2007) Open Access Scientometrics and the UK Research Assessment Exercise. In Proceedings of 11th Annual Meeting of the International Society for Scientometrics and Informetrics 11(1), pp. 27-33, Madrid, Spain. Torres-Salinas, D. and Moed, H. F., Eds. http://eprints.ecs.soton.ac.uk/13804/ Sample of candidate OA-era metrics: Citations (C) CiteRank Co-citations Downloads (D) C/D Correlations Hub/Authority index Chronometrics: Latency/Longevity Endogamy/Exogamy Book citation index Research funding Students Prizes h-index Co-authorships Number of articles Number of publishing years Semiometrics (latent semantic indexing, text overlap, etc.) > policy makers need to understand the benefits and effectiveness of their > investment decisions in R&D. Currently, policy makers rely on big > committee reviews, peer review, and their own best judgment to make > those decisions. The current set of tools available don't provide > policy makers with rigorous answers to the benefits/effectiveness > questions... and they are too difficult to use and/or > inexplicable to the normal policy maker. The result is the laundry list > of "metrics" or "indicators" that are contained in the "Gathering Storm" > or any of the innovation indices that I have seen to date. The difference between unvalidated and validated metrics is the difference between night and day. The role of expert judgment will obviously remain primary in the case of a-priori evaluations (specific research proposals and submissions for publication) and a-posteriori evaluations (research performance evaluation, impact studies) > Finally, I don't think we know enough about the functioning of the > innovation system to begin making judgments about which > metrics/indicators are reliable enough to provide guidance to policy > makers. I believe that we must move to an ecosystem model of innovation > and that if you do that, then non-obvious indicators (relative > competitiveness/openness of the system, embedded infrastructure, etc.) > become much more important than the traditional metrics used by NSF, > OECD, EU and others. In addition, the decision support tools will > gravitate away from the static (econometric modeling, > patent/bibliometric citations) and toward the dynamic (systems modeling, > visual analytics). I'm not sure what all these measures are, but assuming they are countale metrics, they all need prior validation against validated or face-valid criteria, fields by field, and preferably a large battery of candidate metrics, validated jointly, initializing the weights of each. OA will help provide us with a rich new spectrum of candidate metrics and an open means of monitoring, validating, and fine-tuning them. Stevan Harnad