In the press: Science and Public Policy

We advocate a scientometric, top-down, and institution-based research assessment methodology that is based on total citations accumulated from all publications associated-with a specific university during the survey period. The exercise could be done every year using a rolling 7 year retrospective sample and should be performed by at least two independent auditors. Identification of elite ‘revolutionary science’ institutions could be accomplished using a metric derived from the distribution of science Nobel Prizes. *** INTRODUCTION The purpose of the UK Research Assessment Exercise (RAE) is to measure the quality of research in UK universities, with the aim of providing central government funding to support the long-term research capability of an institution. At present, the UK RAE is a 'bottom-up' and discipline-based expert review process determined using a common set of information provided by each disciplinary unit within each university. By contrast, we advocate a 'top-down' institution-based research assessment methodology based on total citations accumulated from all publications associatedwith a specific university during the survey period. Such a survey could be done every year using a rolling 7 year retrospective sample and performed by at least two independent auditors. Identification of elite ‘revolutionary science’ institutions could be based-on a metric derived from the distribution of science Nobel Prizes. THE SCIENTOMETRIC TOP-DOWN APPROACH TO METRICS-BASED RESEARCH EVALUATION Our suggested usage of a RAE based on a metric of total citations from all publications associated-with a specific university is an example of top-down research evaluation, using a single 'macro'-level variable. By contrast, the current UK RAE is implicitly a bottom-up approach to research evaluation using an accumulation and average of many 'micro'-level evaluations. This top-down approach to research evaluation derives from the discipline of scientometrics in which the evaluation of research is seen as a science in its own right. Scientometric analysis is therefore typically performed by observers 'outside' the system being evaluated; in contrast to the bottom-up evaluations which are typically performed by 'peer review' of individuals with expertise in the discipline being evaluated. Since its origins in the 1940s, scientometrics has developed its own ‘system language’ including information selections and a distinctive lexicon and grammar. For this reason, the top-down scientometric procedures of research 1