College and Research Libraries WILLIAM E. McGRATH The Significance of Books Used According to a Classified Profile . of Academic Departments Classification Numbers of books were matched to a classified profile of the university teaching program descriptions. The profiles consisted of LC and DC numbers assigned to courses. It was found that book numbers which matched the course profiles were, (1) more likely to be charged out than not charged out, (2) after being removed from the shelves, more likely to be charged out than left on tables, (3) more likely to be taken off the shelves than left on. The differences between expected and actual proportions in these three situations are large enough to suggest that a precise, classified profile of the university program can be used successfully to select books and to predict cir- culation. THERE HAS BEEN much interest recent- ly in identifying factors which can be used to predict which books and how many will be most used and who will use them. If these factors were known and applied, librarians could be more confident that their book selections were appropriate and that they were build- ing collections relevant to their college or university programs. One recent study in this vein by G. Edward Evans concluded that books selected by librarians in the institutions · sampled were more likely to circulate than those selected either by faculty or by an on-approval method. 1 Even if Evans' findings are true in general, we still do not know what it is about a book that enables a librarian (or any one else) to identify it as more "circulatable" than any other book. Mr. McGrath iY director libraries and pro- fessor of library science at the University of Southwestern Louisiana, Lafayette, Louisiana. 212/ This study stems from the premise that reliable conclusions about book usage can be reached by an examina- tion of the characteristics of books them- selves rather than of how they were ac- quired or who selected them. If it could be shown that a highly used book or group of books possessed characteristics that little-used books did not possess, then these characteristics could be built into a book selection policy. And as the characteristics change so could the pol- icy. Fussier and Simon took this ap- proach and showed that of several vari- ables, immediate past use was the best single predictor of future use. 2 In this paper, the specific characteristic is the subject of books. In previous work the author devel- oped a technique for monitoring the col- lection by comparing book selection and circulation to a framework constructed from the university's catalog of courses -the classified course technique.3 The list of classification numbers generated· by the technique can be regarded as de- partmental subject profiles and hence as the university subject profile. Some doubts as to the effectiveness of the pro- file lingered, however, so a way to test it was sought. The question was simply: how accurate are the profiles; do they describe the departments and the uni- versity well enough to continue using them; are the profiles valid? A plan was formulated to measure what happens to books with classifica- tion numbers which match those in the profile as compared to those for which the numbers did not match. Which books were used and which were not used? The general collection of the Univer- sity of Southwestern Louisiana library is open stack, all students and faculty may remove books from the shelves and use them in the library. Thus many books each day are left on tables. THE HYPOTHESES The three conditions of use: (A) books charged out of the library; (B) books left on tables in the library; (C) books remaining on the shelves; are fundamental in formulating the hypoth- eses. Furthermore, we are interested in how the two contingencies, books whose classification numbers match or do not match those in the profile, affect the three conditions (A), (B), and (C). Thus the following basic hypotheses can be stated. For books whose numbers match the profile, there is no significant difference between the proportion of books which are I. (A) charged out vs. ( B ) and ( C ) not charged out, II. (A) charged out vs. (B) left on tables vs. (C) left on shelves, III. (A) charged out vs. (B) left on tables, IV. (A) and (B) taken off the shelves vs. (C) left on the shelves, V. (B) left on tables vs. (C) left on the shelves, Significance of Books Used I 213 VI. (A) charged out vs. (C) left on the shelves, VIII. (A) and ( C) charged out or left on shelves vs. (B) left on tables. If we found no significant differences among the conditions to be tested we would still not know for sure whether our profile was valid, and hence we would not know how much relevance the collection as developed over the years had to the university program, nor how much course content related to which books are used. On the other hand, if significant dif- ferences were measured, the profile might be considered valid to the extent indicated by the difference between ob- served and expected proportions of ac- tual use of the library. Furthermore, we would have some assurance that those parts of the collection in actual demand did reflect the university program, and that at least 'some of the demand stemmed directly from course content. The profile would thus be a valid tool for measuring these differences. METHOD Three independent data samples were collected by the author on three differ- ent occasions. The method for collecting the samples-counting the books falling within the LC or DC classification pro- file-is described in the author's paper on correlating books used in the library with those which go out. 4 The three samples were from ( 1) subject circula- tion of the University of Southwestern Louisiana books for nearly the entire academic year July 1969-May 1970; ( 2) one year's subject circulation from a study conducted by the author while at the South Dakota School of Mines and Technology in 1967 /68; and ( 3) the study on correlation cited above. 5 Each of these samples required a count of the shelflist. Since no actual counts were available, estimates were made by mea- suring the shelflists at each institution, 214 I College & Research Libraries • May 1972 counting 100 volumes to the inch. To match a book against the profile required no judgment, only a quick ob- servation by the person doing the count- ing. Biases of match or nonmatch would be introduced by the original construc- tion of the profile or by unknown fac- tors causing a book to be removed from the shelves. Since there are two contingencies of use, an appropriate design is the con- tingency figure, and an appropriate test is the chi-square test of independence. Only the first four hypotheses will be tested, since the last three are comple- ments of the first four and the results can be readily deduced from them. FIRST HYPOTHESIS Books Charged (A) vs. Those Not Charged (B) and (C) The counts in Tables 1 and 2 repre- sent books charged or not charged dur- ing the two annual counts. "Not charged" could include books left on the shelves or on tables; i.e., no separate count for books left on tables was made. The "charged out" counts are what remained after all nonclassified materials (e.g., current periodicals ) and permanent loans were removed. Charges did include books charged to the reserve reading room introducing a possible bias, since none of the samples considered the num- ber of times a book on reserve was used. Both figures then contain data for samples taken under essentially the same conditions but for two different institu- tions. Table 1 contains data from an eleven-month sample of the books charged from the library by undergrad- uates and graduate students at the Uni- versity of Southwestern Louisiana from July 1969 to May 1970. Table 2 contains data from the South Dakota School of Mines and Technology. TABLE 1 BooKs CHARGED OuT (A) vs. THOSE NoT CHARGED (B) AND (C) (U.S.L. Sample for 11 Months, 1969/70) Match Nonmatch Total Chi-square ( x 2 ) Roscoe's statistic ( C r ) Charged Out Act ual Expecte d 53,333 42,905 3,495 13,922 56,828 Not Charged Out Actual Expect ed 76,550 86,977 38,651 28,223 115,201 Total 129,883 42,146 172,029 15,448° .423 0 Extremely significant at a. oo;; ; v alue of x2 needed for rejection with ldf is 7.88; H o:¢ = 0 is thus rejected. TABLE 2 BooKs CHARGED (A) vs. THOSE NoT CHARGED (B) AND (C) FROM THE LIBRARY AT THE SouTH DAKOTA ScHOOL OF MrNEs AND TECHNOLOGY (Sample for the Year, 1967 /68) Match Nonmatch Total Chi-square ( x2 ) Roscoe's statistic ( C r ) Cha rged Out Actual Expected 6,056 5,109 1,640 2,587 7,696 Not Charged Out Actual Expected 25,894 26,841 14,539 13,592 40,433 Total 31,950 16,179 48,129 621.18° .159 0 Highly significant at a .oos ; value of x2 needed for rejection w ith ldf is 7.88 ; Ho:¢ = 0 is rejected. In both libraries, the chi-square statis- tic indicates that for the yearlong sam- ples, there is a statistically significant difference between the proportion of books charged and not charged. The chi- square statistic is highly significant in Table 2 and extremely signifi.cant in Ta- ble 1. In both samples, the hypotheses are overwhelmingly rejected. For ex- ample, in Table 1, at the University of Southwestern Louisiana, we expect the proportion of matching books charged out to be 42,905/129,883, or about 33 percent, but in actuality a larger pro- portion 53,222/129,883, or about 43 per- cent was charged. Conversely, we ex- pect the proportion of nonmatching books charged out to be 13,922/42,146 or again 33 percent, but instead 3,495/ 42,146, or about 8.3 percent were charged, a much smaller proportion. Similarly, in Table 2 for the South Da- kota School of Mines and Technology, we expect a proportion of matching books charged out to be 5,109/31,950, or about 16 percent, whereas in actuality a proportion of 6,056/31,950 or about 19 percent was charged. Again, the pro- portion of nonmatching books charged is smaller than expected. In both libraries, according to the sig- nificant value of x2 we can expect books with numbers that match the profile to have a greater chance of being charged out. Another way of looking at the ac- tual differences is through some statis- tic which measures the degree of ef- fect, or contingency. In this paper, we have used Roscoe's statistic, 6 where if x2 were 0, or small, Cr would be small, and if there were a perfect re- lationship between the contingencies- i.e., if all matching books circulated and all nonmatching books did not circulate, Cr would be large or near 1.0. Inter- pretation of Cr is subjective. In Table 1, Significance of Books Used I 215 Cr is substantial; in Table 2, the effect is much less. x2 statistic used by itself is misleading because significance is almost guaranteed with the large N' s used here. SECOND HYPOTHESIS Books Charged Out (A) or Left on Tables (B) or Left on Shelves (C) Librarians and faculty have long doubted that charge statistics reflect the true use of libraries. Charge statistics usually do not include counts of books in the library, which may reveal a dif- ferent use pattern. The two samples treated in Hypothesis 1 did not break down the data to include a count of in- library use. An in-library count, however, was included as part of the author's study on correlation. 7 A one-month count of all books left on tables, chairs, in restrooms, and other locations was ~ conducted. Books which were removed from the shelves by users were counted at the time of reshelving by student aids. The in-library count and the charge count were made during the same peri- od. The three-way count appears in Table 3. This hypothesis states that we expect no differences among the three propor- tions for matching books; those charged out, those left on tables, and those left on shelves. When the three conditions are evaluated together, the chi-square value of 443.13 indicates that there is a statistically significant difference among the three proportions. This statistic, how- ever, reflects the overall difference. · It does not tell us whether the individual differences are significantly larger or sig- nificantly smaller, and the effects of the contingencies are cancelled out as shown by the small Cr statistic. Dual comparisons must therefore be made, where the data for the three conditions are partitioned or combined so that two conditions are compared, as under Hy- pothesis 1. 216 I College & Research Libraries • May 1972 THIRD HYPOTHESIS Books Taken Out (A) vs. Those Left on Tables (B) Once a book is removed from the shelves, it can be placed in one of two categories. The book is ( 1) charged and taken out, or ( 2) the book is left on a table and/ or returned to the shelves. Since we are seeking to validate the profile, we are definitely interested in knowing whether the profile has any bearing on how a book is used-even af- ter it has been taken off the shelf. Here the hypothesis states that there is no difference between the proportion of books with matching numbers that are charged out or that are left on tables in the library. As with the first two hypotheses, a statistically significant difference be- tween the proportion of books charged out and those left on tables is found. With a chi-square value of 423.6, Hy- pothesis 3 is substantially rejected. We expect 6,900 matching books to have been charged out but instead 7,385 were charged; and whereas 2,052 nonmatch- ing books should have been charged, only 1,568 were charged. The signifi- cance, of course, is in the difference be- tween the actual and the expected pro- portions. In general, if a book's classifi- cation number matched a number in the profile, there was a greater probability that it would be charged out after having been removed from the shelves. If its number did not match, there was a great- er probability that it would be left on the tables. Roscoe's statistic shows that the profile has a moderate effect on the probabilities. FoURTH HYPOTHESIS Books Taken Off the Shelves (A) and (B) vs. Those Remaining On (C) This hypothesis is the complement of Hypothesis 3, where the total for books taken off the shelves (Table 5) equals TABLE 3 BooKS CHARGED OuT (A), LEFT ON TABLES (B), AND LEFT oN SHELVES (C) (One-Month Sample) Match Nonmatch Total Chi-square ( x 2 ) Roscoe's statistic ( Cr) Charged Out Left on Tables Actual Expected Actual Expected 7,385 6,900 2,989 3,473 1,568 2,052 1,518 1,033 8,953 4,507 Left on Shelves Actual Expected 119,509 119,720 39,060 38,849 158,569 Total 129,883 42,146 172,029 443.136° .072 0 Significant at a.oo5; value of x2 needed f<;>r rejection with 2df is 10.6; Ho:s>J = 0 is thus rejected. The x2 value of 443.136 equals the sum of x2 values 423.623 and 19.5128 in Tables 3 and 4. TABLE 4 BooKs CHARGED FROM THE LIBRARY (A) vs. THOSE LEFT oN TABLEs (B) (One-Month Sample) Match Nonmatch Total Chi-square ( x2 ) Roscoe's statistic ( Cr) Charged Out A ctual Expected 7,385 6,900 1,568 2,052 8,953 Left on Tables Actual Expected 2,989 3,473 1,518 1,033 4,507 0 Highly significant at a.ooo; value needed for rejection with ldf is 7.88; Ho:j'J = 0 is rejected. Total 10,374 3,086 13,460 423.623° .218 the sum of books charged out plus those left on tables as shown in Table 4. It is necessary to combine the two in this manner to . account for the degrees of freedom. The chi-square value of 19.5 is again significant, but as can be seen from the expected number of matching books removed from the shelves, and the small value of Cr the effect is not so readily apparent, thus pointing up the need for distinguishing between books actually charged and those left on tables. IMPLICATIONS With all four hypotheses substantially rejected, the conclusions are tempting: the profile does describe, within the limits of probability derived from the differences in actual and expected pro- portions, the books used; this usage is clearly related to the subjects embraced by the university's academic depart- ments; the profile is therefore a valid predictor of usage. These conclusions cannot be drawn unequivocally on the basis of only three samples. That is, for greater confidence, additional samples should be drawn from other libraries under carefully controlled conditions. The pattern of differences revealed by the three combinations of condi- tions discussed in hypotheses 1, 2, and 4 suggest the following. If a student or faculty member (we don't know which) removes a book from the shelves he is more likely to charge it out if its class Significance of Books Used I 217 number happens to match the profile (hypotheses 3 and 4). This is not to say that he examines the call number to ascertain the subject of the book. As a matter of speculation, the call number may be nothing more to the user than a location device. If a book is removed from the shelf and left on a table, there is a greater probability that its class number does not match the profile. That is, a person must remove the book from the shelf, and examine it before he knows whether he wants it. This is in- ""'dicated by the fact that he has taken the time to bring it to a table. The implica- tion here is that most books left on ta- bles may be those of which the users are unsure. So-called "in library" use, at least in an open stack library where users have a choice of taking books out or using them in the library may not constitute real use, at least in some subject areas. Such "use" may actually be "to see whether I want to use the book," and therefore should not be equated with out-of-library use. On the other hand, to draw a severe distinction between the two types of use may be stretching the point. Even though we can now de- scribe to a certain extent which sub- jects students will take out, we do not know for sure why they take them. We do not suggest that matching numbers "cause" a book to be taken out nor do we suggest that the matching number is the only, or the best, indica- tor. We do suggest, however, since the differences between expected and ac- TABLE 5 BooKs TAKEN OFF THE SHELVES (A) AND (B) vs. THOSE REMAINING ON (C) Match Nonmatch Total Chi-square ( X 2 ) · Roscoe's statistic ( C r ) Books Taken Off Actual Expected 10,37 4 10,162 3,086 3,298 13,460 Books Remaini ng On Actual Expecte d 119,509 119,720 39,060 38,848 158,569 0 Significant at ct