College and Research Libraries Measuring Collections Use at Virginia Tech Paul Metz and Charles A. Litchfield This study of the Virginia Tech Library collections brings together data on circulation and in- house use, including that of current periodicals, to assess differences according to kinds of use, variations in measurement technique, and time period. Use of current periodicals is found to be both large and qualitatively different from other kinds of use. Differences between circulation and in-house use are discussed. Despite these differences, circulation and in-house use are gen- erally similar in their subject distribution. The subject distribution of circulation patterns is remarkably stable over time, and can be reliably assessed using short sampling periods. ver the past several decades, many authors have contributed to our understanding of how li- brary collections are used. They have identified the characteristics of books receiving disproportionately heavy use, specified the proportion of overall use a minority of books typically receives, clari- fied the relationship between patterns of in-library use and external circulation, and studied the stability of use patterns over time, especially the frequency with which once used books continue to be used. 1 While the total body of use studies has greatly increased our understanding, it has not always been easy to synthesize disparate studies7 conclusions into gener- alizations applicable to most large collec- tions. In particular, three limitations exist. First, differences in units of analysis have often meant that generalizations would require disconcerting leaps of faith-often we want to know how book use behaves according to subject and have to settle for knowing how different kinds of use re- semble one another at the level of the physical piece. Secondly, we must con- sider the differences between in-library use and external use while simultaneously tolerating subtle differences in the ways various studies measure each of these. A final problem is whether the results of any study would be stable over time. We know that use persists at the title level, but few studies have addressed the larger issue of whether general use patterns for a collec- tion are stable. If we knew at least that use is stable over time, we might be more apt to accept generalizations from one collec- tion to another of from one type of use to another. The current study, by bringing together a large number of use measures taken in one institution, attempts to fill some of these gaps in the literature and, in the pro- cess, to specify when generalizations may or may not be made with confidence. While use is measured mostly within a Paul Metz is a Reference Librarian and Bibliographer, and Charles A. Litchfield is Retrospective Conversion Coordinator at the University Libraries, Virginia Polytechnic Institute and State Univer- sity, Blacksburg, Virginia 24061. The authors gratefully acknowledge the assistance of Lisa Brown and William Dougherty of the Uni- versity Libraries in gathering the data; the helpful advice of Jesse Arnold of the Virginia Tech Depart- ment of Statistics; and the thoughtful comments of Robert Broadus of the School of Library Science of the University of North Carolina. 501 L-------------------------------------------------------------------------- 502 College & Research Libraries five-month period, the study addresses the stability issue by comparing two time periods five years apart. It brings together 1987 data on holdings, circulation, in- house use of bound materials, and use of current periodicals, along with baseline data on circulation drawn in 1982 to focus on the following specific questions: 1. How do various kinds of use differ at the subject level? Specifically, how similar are in-library use and circulation patterns? How different is the use of current period- icals from other use? 2. How stable are circulation patterns, by subject, across a time period as long as five years? 3. How large a sample is the minimum size required to yield reliable estimates of use? 4. To what extent do differences in li- brary holdings across subjects artificially affect the correlations among use mea- sures differing in kinds of use measured, technique of measurement, or time pe- riod? (As this question emerges from the other analyses, its significance should be- come more apparent to the reader.) METHODOLOGY Virginia Tech (or Virginia Polytechnic Institute and State University) is a com- prehensive land-grant university with an enrollment of 23,000. While the university offers degrees in all fields, its programs emphasize the applied disciplines of engi- neering, agriculture, and business. With 1.7 million volumes, the library holds membership in the Association of Re- search Libraries (ARL) and ranks between the middle and the third quartile on most ARL measures of size and activity. The li- brary system is highly centralized. The only branch facilities are in art and archi- tecture, geology, veterinary medicine, and a remote facility to support graduate programs in northern Virginia. The li- brary's catalog, holdings records, and cir- . culation traffic are maintained on the on- line Virginia Tech Library System (VTLS). In 1983, Paul Metz reported the results of a large-scale study of the use of the Vir- ginia Tech collections. While the study re- ported overall use by subject categories, its emphasis was on "who uses what" November 1988 and the implications of the borrower-book relationship both for library practice and for the sociology of science. 2 The main data for the 1983 study came from a "snapshot" of the 58,457 books charged to 10,126 active borrowers during May 24-25, 1982. This snapshot did notre- cord circulation transactions but identi- fied, by subject and borrower, the books in circulation on the evening that the data- gathering program was run. Eighty-one subject categories, defined by call-number range, were used in the study. In 1986, a management statistics pack- age was added to the VTLS software. Among its features was the ability to rec- ord circulation transactions across thirty call-number ranges. All checkouts and item renewals at all library sites are re- corded. Separate but similar statistics are kept for batch renewals, a feature that al- lows patrons to renew all their eligible books in a single transaction. The ready availability of these data natu- rally raised interest in how circulation pat- terns might have changed in the five years since the original study. The present au- thors were allowed to specify the call- number ranges to be used in the library's first application of the management statis- tics package and made a point of defining their thirty categories consistent with the eighty-one used in 1982. Because the outer boundaries of the thirty call-number ranges chosen correspond to the bound- aries of various ranges from the prior study, observations in the eighty-one cat- egories of the older study could be col- lapsed into observations across the thirty new categories. Table 1 shows the call- number ranges selected for use in the cur- rent study. Both the selection of subject ranges and some problems with the actual gathering of the data require explanation. The sub- . ject categories obviously vary widely in their size and significance, but the reasons for this are not so arbitrary as may first ap- pear. In some cases it was necessary to de- limit small subject categories of relatively little interest so that other categories could be defined. For example, the study team had limited interest in the use of materials in astronomy (QB), but had to define this Measuring Collections Use at Virginia Tech 503 TABLE 1 SUBJECT CATEGORIES Class Dewey, A B C-D E-F G H-HA HB-HJ HM-HX J K Unreclassified Dewey materials, general Philosophy, psychology, religion General and world history History of the Americas Geography, anthropology, recreation General social sciences, social statistics Economics and business Sociology Politicaf science Law Subject L M Education and Virginia Tech publications, including theses and dissertations Music P-PQ PR PS Miscellaneous language and literature English literature American literature PT-PZ Q Miscellaneous language and literature, juvenile General science QB QC QD QE QH-QR R S-SD SF-SK Astronomy Physics Chemistry Geolo~ Life sctences Medicine Agriculture Animal science T U-V Technology, engineering, home economics Military, naval science Bibliography, library science z range in order to delimit both mathemat- ics and computer science (QA) and phys- ics (QC). With a limit of thirty subject categories, not all parts of the collection could be mea- sured in sufficient detail to specify disci- plines with the desirable precision. The decision was made to sacrifice detail within class T by devoting only one cate- . gory to this large and important area. Since it had required eleven subject cate- gories defined by sixteen call-number ranges to specify the various fields of engi- neering plus home economics in the 1982 study, to preserve detail within class T for the present study would have required an enormous sacrifice of precision in other call-number areas. One consequence of the reasoning that guided the selection of call-number ranges was that the ranges chosen varied enor- mously in their size. At the extreme, thirty-three times as much of the collec- tion is in subject category T as is held in QB. Later portions of the analysis will ex- plore the methodological consequences of these disparities. Because the VTLS statistical package is somewhat less sophisticated than the pro- grams that gathered data for the 1982 study, some minor problems were en- countered. Unlike the 1982 program, the statistical package could recognize neither government documents nor unassociated records for first-time circulations. Conse- quently each use of a government docu- ment (which accounted for just under 1 percent of use in the 1982 study) incre- mented the call-number category begin- ning with the same letter as the SuDocs number of the book being circulated. This made it impossible to measure use of doc- uments and caused a small degree of error throughout the data. VTLS substitutes the character string "not yet entered" for the call number of freshly circulated items not yet associated with bibliographic records. Unfortunately the statistical package, with infamous computer literalness, incremented the use 504 College & Research Libraries counter in class N for each circulation transaction of these materials. Because ''not yet entered's'' could not be disaggre- gated from use of actual N materials, class N was thrown out of the study. Percent- age totals for all measures reported in this study, whether of circulation, other use, or holdings total to 100 percent because of the complete elimination of class N from the data set. Circulation measures were gathered separately for every month from January through May 1987. The May data actually report measures from May 1-16 and May · 20-29 only, as the statistical package was inoperative for five days. For April and May, butnottheothermonths, it was pos- sible to record data for batch renewal~. The various automated measures of cir- culation activity were supplemented by a variety of manually recorded measures. For May 1987, shelvers in all library loca- tions noted the call number of each of the 95,086 bound items shelved. Government documents were the only exceptions in the gathering of these data. "A total of 26,881 observations was made of current periodicals use by subject." During May the staff also tallied the use of the 9,468 current periodical titles on dis- play at all library locations. Branch librari- ans reported that patrons in their less for- mal settings reshelve most of their own current periodicals and that use of current periodicals in the branch literatures of art, architecture, geology, and veterinary medicine would therefore be underre- ported. The last measures taken to form the study data were holdings measurements. Linear inches of the shelflist were mea- sured as an index of the library's holdings in each call-number range. The final variable, in-library use of bound materials, represents in-house use of all materials other than current periodi- November 1988 cals, materials in class N, and documents. It includes all use of bound periodicals, which are not allowed to circulate at Vir- ginia Tech. The variable was created by subtracting the number of April checkouts in each subject range (which totaled 32,408) from the number of bound items reshelved in May. Without this or a similar corrective measure, the reshelving statis- tics would have reflected a large percent- age of returns from circulation and would not accurately represent in-library use. As will be shown later, one month is a good surrogate for another in the selection of circulation data. April was chosen because it is the month closest to May, for which data were not complete, and because many books charged out in April would have been returned and then reshelved in May. This means of transforming the data seemed the most logical way to derive an imperfect, but arguably very close, mea- surement in-library use of bound materi- als for May. Because of the problems encountered in measuring the use of government docu- ments, branch periodicals, and materials in class N, it is not possible to describe the distribution of kinds of use with perfect accuracy. However, the data in table 2 should quite closely represent the overall mixture of materials use for May 1987. They indicate that at Virginia Tech, circu- lation accounts for less than one third of overall use, while current periodicals ac- count for more than a fifth of use. As Rob- ert Broadus has noted, various studies have yielded highly disparate ratios of in- library use to circulation, depending on measurement techniques and details of the local setting. 3 The present data suggest a ratio of more than two to one between in- library use and circulation if browsing of current periodicals is included and less than two to one if such use is excluded. FINDINGS In-Library Use Versus Circulation Various studies of the relationship be- tween in-library use and circulating use differ in the degree of similarity that is found between these two kinds of use. This debate is not extreme, in that no one argues that in-house use and circulation Measuring Collections Use at Virginia Tech 505 TABLE2 DISTRIBUTION OF KINDS OF LffiRARY USE, MAY 1987 Type of Use % n Notes In-library use of bound materials In-library use of current periodicals Circulation 49.8 62,678 May 1987 shelving minus April1987 on- line circulation transactions 21.4 26,881 28.8 36,175 May 1987 circulation transactions x 31/26 to correct for missing data Excludes May 1987 reshelving of 1,458 government documents separately reported . are totally dissimilar, and no one claims that they are identical. Only the degree of similarity is at issue. Differences in the two kinds of use may be attributed either to differences in the information content of the materials themselves-novels will be checked out, amortization tables will not-or to differences among disciplines in the ways in which materials are used. Common wisdom holds that scientists rely disproportionately on a more recent literature that appears mainly in periodi- cals, especially recent periodicals, while the humanities place greater emphasis on retrospective materials in monographic form, though citation and use studies have yielded surprisingly little evidence on this point. " 4 One would therefore ex- pect to find a higher ratio of in-library use to circulation in the sciences and technol- ogy than in the humanities. The present data make it possible to cal- culate correlations helpful in specifying the relationship between in-library use and other use. Before entering that analy- sis, it seems appropriate to present the raw data and discuss the qualitative differ- ences in use patterns among disciplines that set an upper limit on the similarity be- tween circulation and various kinds of in..: ternal use. Table 3 gives the percentage of use attributable to each call number for current periodicals use, all reshelving, in- house use (reshelving minus April check- outs}, and circulation (including batch re- newals) for May 1987. A number of pronounced differences in the kinds of materials preferred in the var- ious disciplines are readily apparent. While these are generally in the direction that one might predict, the data help to specify these differences with some accu- racy. One obvious example is the extent of use in call-number ranges ''Dewey'' and A and Z. Although the library has modest holdings of books not reclassified from the Dewey system but no periodicals in Dewey, both the percent of current peri- odicals use and of all in-house use in De- wey and A greatly exceed circulation in this range. In the case of the A's, much of the difference is explained by reader inter- est in current, general periodicals. For both subject categories, much of the expla- nation for this difference stems from the high proportion of materials in classes A and Z that are of a reference nature and may be used only within the building. 11 Current periodicals in business and economics (HB = H]) account for about half again as much overall use as do bound materials in this area, re- flecting in part the importance of cur- rent awareness in business and fi- nance.'' Other differences in the use of current periodicals and bound materials are more revealing of intrinsic differences in disci- plines and their literatures. Current peri- odicals in business and economics (HB-HJ) account for about half again as much over- all use as do bound materials in this area, reflecting in part the importance of current awareness in business and finance. Cur- rent periodicals use is also disproportion- ately heavy in music (M}, though the ex- perience of the reference staff is that at least some of this arises from use attribut- able to students reading Rolling Stone, Gui- 506 College & Research Libraries November 1988 TABLE 3 SUBJECT DISTRIBUTION OF USE, BY KIND OF USE, MAY 1987 A . Current B. Bound C. House D. Circulating Dewey, A 8.29% 2.26% 3.27% .70% B 2.47 4.70 4.27 6.16 C-D 3.20 2.82 2.06 4.66 E-F 1.18 2.80 1.93 3.41 G 2.92 2.09 1.35 2.76 H-HA .72 1.10 1.42 .52 HB-HJ 18.56 11.58 12.71 10.48 HM-HX 3.66 5.72 5.26 6.50 J 1.18 1.56 1.72 1.45 K 1.21 2.19 2.77 .98 L 5.64 3.41 3.05 4.16 M 3.36 .97 .78 1.14 P-PQ 2.15 3.47 3.06 4.94 PR .20 3.68 3.06 5.66 PS .28 2.79 1.68 4.57 PT-PZ .05 1.16 .54 1.97 Q 3.42 1.92 2.52 .61 QA 4.83 5.51 5.38. 5.98 QB .39 .24 .16 .28 QC 1.57 2.75 3.02 2.49 QD 2.75 3.20 3.95 1.83 QE .25 .75 .44 1.20 QH-QR 3.49 4.85 5.10 4.24 R 3.38 4.84 5.29 4.21 S-SD 2.67 2.42 2.68 1.81 SF-SK 1.57 1.42 1.62 .87 T 18.08 16.80 16.94 15.25 U-V 1.28 .79 .81 .80 z 1.27 2.21 3.18 .38 n 26,881 95,086 62,678 41,581 A . Current = Shelving of current periodicals . B. Bound = All shelving of bound materials . C. House = In-house use of bound materials (May shelving from column B minus April online circulation transactions) . D. Circulating = All May 1987 circulation transactions, online and batch . tar Player, and similar publications. The use of current periodicals in litera- ture is at the opposite extreme from use in business. Use is very light and, in the case of call numbers from PR through PZ, vir- tually negligible. These data suggest that current awareness is a far less salient reader strategy in literature and literary criticism than in other fields. Differences between the internal use of current periodicals and of bound materials are especially noteworthy when we con- sider the large proportion of in-library use attributable to current periodicals usage. The current data suggest that with 26,881 recorded uses of current periodicals and 62,678 other kinds of in-library use, cur- rent periodicals account for about 30 per- cent of overall in-library use. Comparisons among columns B, C, and D support and help to specify other com- mon generalizations about literature use. Since Column B reflects both in-library use and circulation, it is hardly surprising that with one minor exception (U-V), the percentage in B are intermediate between those for column C (in-library use) and column D (circulation.) In his 1983 study, Metz found that hu- manities materials accounted for a greater proportion of circulation than of in-house use. In-house use was somewhat dispro- portionately high in science and technol- ogy, while in-house use and circulation in the social sciences were roughly equiva- lent. 5 This finding suggests the greater de- pendence of scientists on journals, which in the case of the Virginia Tech libraries do Measuring Collections Use at Virginia Tech 507 not circulate. Comparisons between the data on in-library use and circulation use in table 3 support the same conclusion Metz reached in 1983. For classes B-F and M-P, circulation percentages exceed per- centages of in-house use. by a considerable margin (though raw circulation is less than in-house use for virtually all catego- ries). For most fields of science and tech- nology, the in-house-circulation relation- ship is reversed in that in-library use percentages are higher than percentages for circulation. Overall, materials in classes Q through T accounted for 47.1 percent of in-house use of bound materi- als and 42.4 percent of current periodicals use, but only 38.8 percent of overall circu- lation. If there is any surprise in this compari- son, it may be that in-house use does not exceed circulation more dramatically than it does in science and technology, and es- pecially that current periodicals use in not higher. However, both Metz and Harry Kriz have found that, common assump- tions notwithstanding, scientists and en- gineers make substantial use of circulating materials. 6 While in-library use of bound materials and use through circulation differ in ways that reveal something about humanities, science, and other disciplines, table 3 shows that as a rule subjects tend to vary together in their tendency to receive lighter or heavier use, regardless of which kind of use is being considered. Generally use studies comparing in-house use with circulation have found a positive correla- tion. An early example is Herman Fussier and Julian Simon's conclusion that ''Books that develop little recorded use develop little browsing, and books that develop much recorded use develop much browsing, except for the highest use books, for which extrinsic factors distort the picture. " 7 Anthony Hindle and Mi- · chael Buckland's work supports this con- clusion. Their comparison of in-house use and circulation at the University of Lan- caster library led them to state that ''the conclusion is very clear. Books that circu- late little get relatively little in-house use and the higher the circulation the higher the level of in-library use."8 The most recent validation of the simi- larity of in-house use and circulation comes from the University of Pittsburgh study. Comparing data on the internal use of books gathered in 1975-76 and their main database reflecting circulation activ- ity, Stephen Bulick and others found that at least 75 percent of books used in-house had also circulated externally, and that books used in-house had· a higher mean number of external circulations. Their findings led them to embrace Fussier and Simon's earlier conclusion that recorded use is a reasonably good index of all use. 9 If individual books tend to be either heavily or lightly used both internally and through circulation, the overall correla- tions comparing these two kinds of use within subject categories will be positive. Because the data have already shown that in-library and circulation use are by no means proportionate across subjects, the correlations, at least in the present case, must be well short of unity. How strong are the correlations, and what do they tell us about the similarity of in-library use and circulation across subject categories? William McGrath's 1971 study comparing the two kinds of use at the University of Southwestern Louisiana yielded correla- tions across his subject categories of .84 and .86, depending on small methodolog- ical variations. He concluded from this that "circulation totals, when grouped into self-delineating spans, can be reliable indicators of the subjects being used within as well as out of the library.' ' 10 The present data support this conclu- sion and, indeed, provide higher esti- mates of the relationship between in- library use and circulation. By measuring both kinds of use in slightly different ways, they also help to identify the degree of difference among various kinds of use. Table 4 reports the Pearson Product Mo- ment Correlations among a variety of measures of external and internal use. The single best estimate of the relation- ship between in-library use and circula- tion is the correlation of .885 between the variables measuring all circulation trans- actions for May and the May reshelving of bound materials not returning from circu- lation. This correlation, which is consis- 508 College & Research Libraries November 1988 TABLE4 CORRELATION COEFFICIENTS AMONG MEASURES OF IN-LffiRARY USE AND CIRCULATION, MAY 1987 Online Batch .942 Alkire .995* .970* House .860 .911 .885 Current .733 .789 .758 Shelving .897* .933* .918* Online Batch Alkire Online - Circulation online checkouts and renewals Batch = Circulation batch renewals Alkire = Sum of online and batch circulation .902 .988* House .932* Current Shelving House = May shelving of bound materials used in-house (May bound shelving minus April online circulation transactions) Current = Current periodicals shelving Shelving = All in-house shelving, including returns from circulation and current periodicals. *Denotes correlation of two measures, one of which is a subset of the other. Note: All correlations reported in this study are statistically significant at p< .001. tent with McGrath's observed correla- tions of .84 and .86, would indicate that about 78 percent of the variance in each variable would be synchronous with vari- ation in the other. The differences in the various kinds of use observed in table 3 make it difficult to describe in-library use and circulation as identical. Still, the two kinds of use are very similar, and, after al- lowing for fairly predictable differences, one may generally use one kind of use as a surrogate for the other. "Not surprisingly, the correlations between circulation and the internal use of current periodicals are lowest of all those reported here.'' The data in table 4 show important dif- ferences in the kinds of in-library use re- corded in this study. The correlation of . 902 between the subject distribution of in- library use for bound materials and cur- rent periodicals is high, but does reveal a sufficient difference to indicate that the distinction between these kinds of inter- nal use deserves a degree of attention not hitherto received in use studies. Not sur- prisingly, the correlations between circu- lation and the internal use of current peri- odicals are lowest of all those reported here. Future statements about the similar- ity of circulation and in-house use should carefully specify whether the use of cur- rent periodicals has been counted. While the distinctions among kinds of internal use are important, those between circulation measures are not. Differences in the subject distribution of books being checked out or renewed at the item level and those being batch renewed are rela- tively unimportant ( r = . 942). Stability of Circulation Patterns An unusual feature of the present data is the time period covered. Few if any stud- ies have been able to compare use data drawn at points as far apart as five years. These data make it possible to investigate the stability of circulation measures, first across measurement periods close to- gether and then across a time interval of a half-decade. Table 5 reports the Pearson correlation coefficients for measures of circulation transactions across the twenty-nine sub- ject categories for the individual months January through May 1987 and for the sum of these periods. The lowest correla- tion between pairs of individual months is .967. The lowest correlation of any month with the total period is . 987. These data re- quire little interpretation. They clearly show that circulation data by subject for any month are essentially interchangeable with those for any other and would pro- vide an excellent measure of overall circu- Measuring Collections Use at Virginia Tech 509 TABLES CORRELATION COEFFICIENTS, JANUARY THROUGH MAY ONLINE CIRCULATION TRANSACTIONS JANTRAN FEBTRAN . 973 MARTRAN . 977 APRTRAN .987 MA YTRAN .967 ALLTRAN .987 .993 .989 .994 .996 .990 .990 .996 JANTRAN FEBTRAN MARTRAN .984 .996 APRTRAN .993 MAYTRAN ALLTRAN TABLE6 CORRELATION COEFFICIENTS, SELECTED 1982 AND 1987 CIRCULATION MEASURES ALLTRAN87 APMAYALL87 MAY82SNAP OCT82SNAP (.996) .968 .958 ALLTRAN87 .976 .971 APMAYALL87 (.991) MAY82SNAP OCT82SNAP ALLTRAN87 - All January-May 1987 circulation online transactions APMA YALL87 - All April and May 1987 circulation transactions, online and batch MA Y82SNAP - May 1982 snapshot of books in use OCT82SNAP = October 1982 snapshot of books in use lation. Provided that school is in session and use of a library is normal, the issue of variation in use patterns by subject through the academic year appears to be irrelevant. The data in table 6 represent the correla- tions among two pairs of circulation mea- sures taken five years apart. The variable ''ALL TRAN87'' represents all online cir- culation transactions for January through May 1987. "APMAYALL87" represents the sum of both online and batch circula- tion transactions in April and May 1987. ''MAY82SNAP'' and ''OCT82SNAP'' re- present the main 1982 data, a snapshot of books charged out in May and an October 1982 snapshot taken to assess the stability of the circulation data between spring and fall1982. In the table the correlations between measures taken in the same year are re- ported in parentheses so as to highlight the four correlation coefficients reporting use across the five-year interval. All four correlations are very high, despite both the five-year interval and differences in measurement technique among online transactions, the sum of online transac- tions and batch renewals, and snapshots of materials in use. These findings buttress the argument that circulation measures are relatively im- mune to differences in measurement tech- nique. More importantly, they reveal an extraordinary measure of stability, indicating that circulation patterns in an institution not undergoing dramatic cur- ricular change or extensive changes in the direction of library acquisitions are re- markably constant. Together with the data in table 5, these correlations suggest that a single measure of circulation use drawn at any normal time in a given school year may provide an academic li- brary with measures on which it can rely for a considerable period of time. Use of Small Samples to Assess Collection Use The knowledge that circulation patterns are as stable as the current data suggest may encourage more academic libraries to conduct circulation studies. While such studies are relatively simple to conduct in automated libraries, they may be labori- ous and expensive in libraries whose cir- culation procedures are manual. This raises the question of cost-benefit trade- offs in the selection of sampling periods, and specifically the question, What is the 510 College & Research Libraries shortest (hence cheapest) period that will yield reliable estimates of circulation activ- ity? The team of researchers that studied li- brary use at the University of Pittsburgh in 1975 anticipated this question and con- ducted their own research toward answer- ing it. By drawing ex post facto subsam- ples of their data and comparing these to the full data set, they were led to conclude that "random samples of as few as three days produced correlations as high as . 95-.98 with the total population in regard to percent of usage by LC class. " 11 To replicate the work of the Pittsburgh team, data were drawn from the present data set to provide subsamples of time pe- riods equal to and both longer and shorter than the three-day period found sufficient by the Pittsburgh team. The analysis has already shown that one-month samples more than adequately represent circula- tion activity for a longer time period. The subsamples chawn to refine this conclu- sion further consisted of two seven-day periods from April J987; two three-day periods from April; and two single days drawn from March. The correlations of these subsamples with the entire set of 160,141 online circulation transactions are shown in table 7. The results both replicate the Pittsburgh findings and indicate that the three-day period t~ey found adequate may well de- fine the lower limit for short sample pe- riods. The results for three-day subsam- ples from the full data were as strong as those found at Pittsburgh. Indeed, three- day samples yielded results as reliable as those from seven-day samples. However, November 1988 correlations noticeably dropped off when subsamples .of only one day were used. In interpreting these findings, it should be remembered that the issue of sample size is at least as important as that of sam- pling time periods. A small library with modest levels of circulation would risk ob- taining unreliable estimates from a three- day sample because of insufficient volume of circulation transactions and the possi- bility that one or two individual patrons with many books could unduly influence the entire data set . Use Measures and the Effects of Library Holdings To some degree the use of any library collection must reflect the self-fulfillment of prophecies made by those who built the collection. Stated negatively, patrons can- not use materials the library never ob- tained. Stated positively, patrons may find their reading interests influenced by the strengths of a given library collection. If the profile of a library collection does influence use, it must necessarily influ- ence all kinds of use over various time pe- riods. This raises the question of whether the correlations among various measures differing in kind of use, measurement technique, or time period may not be arti- ficially boosted by the artifact of an indi- vidual collection's profile. A more de- tailed discussion of this question is given in the appendix. While this question is troubling, it is not intuitively obvious that it is a serious one, or that one is not "borrowing trouble" even to raise it. However, a conscientious approach to the current analysis requires TABLE 7 Subsample April-7-A April-7-B April-3-A April-3-B March-1-A March-1-B CORRELATIONS OF SELECTED SUBSAMPLES OF ONLINE CIRCULATION TRANSACTIONS WITH ALL ONLINE CIRCULATION TRANSACTIONS Correlation Coefficient .989 .987 .989 .991 .922 .925 Variable names indicate month, number of days, and first (A) or second (B) randomly selected period. Measuring Collections Use at Virginia Tech 511 that the possibility of artifactual effects on the findings be explored. Three considerations in particular man- date the pursuit of this question. One is the independent findings by Thomas John Pierce and William E. McGrath that hold- ings statistically explain about 65 percent of the variation in use across subjects. u That is, correlations between holdings and use across subjects have been found to be about .8. The second consideration is the large disparities in size between hold- ings across the subject categories used in the current study-as mentioned, the sub- ject categories vary by a factor of thirty- three in the proportion of the collection for which they account. The final consideration requiring some investigation of this issue is the realization that the correlations reported in the study are consistently somewhat higher than those found elsewhere. As one example, the correlation of . 991 between the May and October 1982 circulation snapshots exceeds the correlation of . 982 Metz found for the same data when he originally com- pared them using his eighty-one subject categories, which did not differ so dramat- ically in size as those used here. 13 The relationship between use and hold- ings for the current study was found to be in the same range as was found by Pierce and McGrath, indicating that perhaps the larger size disparities among subjects in the current study are not an important consideration. The correlation of all April and May online and batch circulation transactions with holdings across the sub- jects was .784. The correlation of all online circulation transactions from January through May with holdings was .793. Both measures are essentially identical to the Pierce and McGrath findings. To assess the possible effects of holdings in artificially boosting the correlations among use measures, partial correlations were calculated among the use measures with statistical controls for holdings. The derivation of partial correlations repre- sents a. standard approach to identifying the relationship between two variables af- ter accounting for the effects on each of a third variable. The results indicated that allowing for holdings effects does considerably reduce correlations between in-library use and circulation, but that correlations among other measures of use or among similar measures taken at various time periods re- main extremely high despite these con- trols. Table 8 shows a representative selection of some of the more important correla- tions among use measures reported in this study. The drop-off in correlations be- tween in-house use and external circula- tion (see first two items in table 8) is sub- stantial, although the partial correlations of .730 and .664 remain quite healthy and explain about half the variance. Measures TABLES Measure 1 MAYALL MAYTRAN JANTRAN MAYTRAN APRIL-3-A APRIL-3-B MAY82SNAP CORRELATIONS, ZERO-ORDER AND WITH CONTROLS, AMONG SELECTED USE MEASURES Raw Measure2 Corr. MAYHOUSE .885 MAYHOUSE .860 ALLTRAN .987 ALLTRAN .993 ALLTRAN .989 ALLTRAN .991 APMAYALL87 .976 MAYALL = All May 1987 circulation transactions, both online and batch MA YHOUSE = All May 1987 in-house use of bound materials JANTRAN, MAYTRAN, ALLTRAN - Online circulation transactions for January, May, and January-May APRIL-3-AIB - Three-day subsamples of April circulation transactions MA Y82SNAP = May 1982 snapshot of books in use APMA YALL87 = All April and May 1987 circulation transactions, online and batch Partial r Controling for Holdings .730 .664 .975 .984 .970 .976 .937 su· · College & Research Libraries of the stability of circulation across time and across different techniques of mea- surement are much more robust and show less diminution from these statistical con- trols. For those correlations that are signifi- cantly reduced by controls for size of hold- ings, we should remember that such con- trols have not previously been introduced in the literature. The raw rank-order corre- lations remain the best single measure of the consonance between indicators of use and should be compared to similar corre- lations elsewhere in the literature. How- ever, their reductions with statistical con- trols for holdings do indicate that the effect of holdings on use is a valid concern and should raise the issue of what reduc- tion such controls would bring to the cor- relations reported by other studies. To the extent that the subject categories used in other studies may not vary in size so widely as those used here, there will be less need for concern about the spurious effects of holdings. Still, it would be pru- dent for others conducting research in this area to assess the effects of holdings within their own studies. For those correlations relatively im- mune to such statistical controls, our con- fidence should be increased in these find- ings about the stability of use measures over time and across various measure- ment techniques. CONCLUSION While the current study has investi- gated a broad range of issues, all the ques- tions it raises treat the compara~ility of use ·measures across one of three .dimensions: types of use, measurement techniques for assessing use, and time. We have essen- tially asked, How stable and how similar are measures of use? in a variety of ways. The following statements summarize our several responses to this question. 1. In-library use of bound materials and November 1988 circulation are highly correlated. The dif- ferences between them follow fairly pre- dictably from differences in the kinds of information found in various call-number ranges and from the disproportionate use of circulating materials in the humanities as opposed to the sciences. 2. In-house use of current periodicals is qualitatively different from other kinds of use, though it does correlate significantly with them. Future studies should devote more attention to this kind of use, which accounts for a large fraction of in-library use, and should carefully distinguish it when results are reported. 3. Circulation statistics are extraordi- narily stable over time and extraordinarily insensitive to minor differences in mea- surement techniques. 4. For libraries with large circulation volume, three-day samples of circulation activity yield highly satisfactory estimates of overall circulation volume. 5. Holdings influence library use and introduce a factor that future studies are obliged to consider. Holdings account for a substantial part of the observed relation- ship between in-library use and circula- tion. They do not appear to have so impor- tant a role in accounting for the observed relationships among other measures of use. It has become traditional to end studies of this nature with calls for replication. In- deed, replication of this study's novel findings, especially concerning the nature and extent of current periodicals use, is highly desirable. Still, in important re- spects this study replicates others. By showing the stability of circulation data, it also suggests that most use studies will be readily replicable within their own set- tings. Taken together, these findings may caution against a paralyzing overempha- sis on the importance of replication in studies of library use. REFERENCES AND NOTES 1. Allen Kent and others, Use of Library Materials; The University of Pittsburgh Study (New York: Marcel Dekker, 1979); Herman Fussier and Julian Simon, Patterns in the Use of Books in Large Research Li- Measuring Collections Use at Virginia Tech 513 braries (Chicago: Univ. of Chicago Pr., 1969); Michael K. Buckland, Book Availability and the Library User (New York: Pergamon, 1975); Richard W. Trueswell, "User Circulation Satisfaction vs. Size of Holdings at Three Academic Libraries,'' College & Research Libraries 30:204-13 (May 1969); Wil- liam E. McGrath, "Correlating the Subjects of Books Taken Out of and Books Used within an Open-Stack Library," College & Research Libraries 32:280-85 Ouly 1971). 2. Paul Metz, The Landscape of Literatures: Use of Subject Collections in a University Library (Chicago: American Library Assn., 1983). 3. Robert Broadus, "Use Studies of Library Collections," Library Resources & Technical Services 24:317-24 (Fall1980). , 4. Citation studies showing the greater dependence of the natural sciences than of other disciplines on recent publications include Derek J. deSolla Price, "Citation Measures of Hard Science, Soft Science, Technology, and Nonscience," in ed. Carnot E. Nelson and Donald K. Pollock, Communi- cations among Scientists and Engineers (Lexington, Mass.: D. C. Health, 1970), p.3-22; and Duncan Macrae, Jr., "Growth and Decay Curves in Scientific Citations," American Sociological Review 34:631-35 (Oct. 1969). 5. Metz, Landscape of Literatures, p.23. 6. Metz, Landscape of Literatures, p.14; Harry M. Kriz, "Subscriptions vs. Books in a Constant Dollar Budget," College & Research Libraries 39:105-109 (Mar. 1978). 7. Fussier and Simon, Patterns in the Use of Books, p.115. 8. Anthony Hindle and Michael K. Buckland, "In-Library Books Usage in Relation to Circulation," Collection Management 2:265.-77 (Winter 1978). 9. StephenBulick, WilliamN. Sabor, andRogerR. Flynn, "Circulation and In-House UseofBooks," in Kent and others, Use of Library Materials, p.9-55. 10. McGrath, "Correlating the Subjects," p.285. 11. Bulick and others, "Circulation and In-House Use of Books," p.44. 12. Thomas John Pierce, "The Economics of Library Acquisitions: A Book Budget Allocation Model for University Libraries," (Ph.D. diss., Univ. of Notre Dame, 1976), p.46; William E. McGrath, "Predicting Book Circulation by Subject in a University Library," Collection Management 1:7-23 (Fall/Winter 1976-77), p .12. 13. Metz, Landscape of Literatures, p.116. APPENDIX A. The Possible Effects of Holdings on Use This issue may be best clarified in the form of a debate. PRO: Collections will influence use in a way that artificially heightens correlations among use measures. If monkeys were turned loose on the Virginia Tech collection on Monday and gibbons roamed the stacks on Tuesday, both would disturb many more books in class Tthan in class QB simply because there are so many more of the former. This would yield high but mean- ingless correlations between Monday and Tuesday's use. Only those books actually in the library can be used, and therefore use is to some degree an outcome of holdings. CON: The influence of collections on use patterns does not introduce a significant meth- odological artifact. Holdings may "explain" about 65 percent of use in the narrow statistical sense, but do they substantively "explain" anything? Some subjects attract more interest than others. More books are published in them, libraries buy more books in them, patrons read more books in them. If the correlations between holdings and use were zero, library collection development activities would be condemned as random processes. Statistical measures aside, the issue of just what is explained by holdings is an open one. We should remember that we are not considering the use patterns of monkeys and gib- bons, but of students and faculty, who will select only those materials they find useful. The existence of materials in a collection does not make their use inevitable. If for some reason 10 percent of the Virginia Tech collection were in numismatics, would one expect that these materials would receive 10 percent of use?