Braun Tibor, Glänzel Wolfgang, Schubert András: Országok, szakterületek, folyóiratok tudománymetriai mutatószámai 1981-1985 (A MTAK Informatikai És Tudományelemzési Sorozata 6., 1992)
Indicators - Statistical reliability of scientometric indicators
A. SCHUBERT, W. GLÄNZEL, T. BRAUN : SCIENTOMETRIC DATAFILES where N and M are the number of the country's publications and citations, respectively, in the given field, and S and T the number of the country's publications and citations, respectively, in all science fields. If the field in question represents only a small fraction of the scientific endeavour of the country (1JN > > 1/5), then the above relations reduce to A AI~AI*NXf l , A AAI - AAI * M 1^ 2 , and one can rely upon the rule of thumb that staying within 10% error bounds requires a sample of publications (resp. citations) of at least 100 items in the given field. A simple test statistic to decide whether an AI or AAI value differs significantly from 1 can be defined as t M = (AI-I)/A AI ÍAA J = (AAI-1)/A AAI These statistics are random variables of Student's /-distribution, which can be approximated by a standard normal distribution, provided that the indicators are based on a sample of some reasonable size. Thus, e.g., if / < 2, the indicator does not differ significantly from 1 at a significance level of 0.95 . Test statistics can also be used for assessing the reliability of crossnational comparisons of AI and AAI. The required test statistics are then constructed as follows: / = (Af -AIJ * [(Af - 7) 2//, 2 + (AI 2 - l) 2/t 2Y f t (for the Activity Index) t = (AAI x-AAI J * [(AAI X - 7) 2//, 2 + (AAI 2 - l) 2/t 2] A' 2 (for the Attractivity Index). These test statistics can again be considered random variables of standard normal distribution. 14 Scientonietrics 16 (1989)
