CRANV2 Supplementary tests and results chapter Cyril Cleverdon Michael Keen Cranfield An investigation supported by a grant to Aslib by the National Science Foundation. Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. - 239 - containing approximately similar numbers of questions; the results are presented in Fig. 6.22T. With some minor aberrations, these show, for any given coordination level, that the increase in total term postings results in a regular increase in recall ratio accompanied by a corresponding decrease in precision ratio. The next stage was to group the questions in relation to the average number of postings for each term. However, the preliminary stage of making up the groups of questions by tbis method showed that the groups differed little from those used in Fig. 6.22T, so no further work on this was done. Order of retrieval of relevant documents An analysis was made of the effect on retrieval of individual relevant documents in moving from one index language to another. The results in Chapter 4 show, for instance, that with Index Language I.l.a, at a coord- ination level of 5, there:were 94 relevant documents retrieved (see Fig. 4.200T). With Index Language 1.6.a,at a coordination level of 6, there were 87 relevant documents retrieved {see Fig. 4.203T). The question is whether the change of index language and the increase in coordination levels resulted in a different or a similar set of retrieved documents. To investigate this point, nine index languages were selected namely I.l.a, 1.5.a, 1.8.a, II.l.a, II.5.a, II.10.a, II.15.a, III. l.a and III.6.a. For the 42 questions, the records were checked to find the order of retrieval of the relevant documents for each of the nine languages. Some examples are given in Fig. 6.23T, which shows, for Questions 118, 170 and 250, the coord- ination levels at which the relevant documents were retrieved, and in Fig. 6.24T a ranked order of retrieval. From this type of data for the 42 questions, it was too involved to sort out what happened to each individual relevant document, but an analysis was made for each of the three main groups of index languages to find what happened to the relevant documents ranked first and last in the basic languages {i.e. I,l.a, II.l.a, and !tI. 1.a). While it was not possible to make a clear cut decision every time, Fig. 6.25T shows that in the very lar'ge majority of cases, the change from one language to another did not alter the retrieval rank of the first and last documents retrieved.