College and Research Libraries Research Notes Observations of Browsing Behavior in an Academic Library Johanna Ross The increasing costs of land and build- ings, coupled with inflationary pressures on operating budgets, are forcing library administrators to consider nontraditional methods in providing patron access to col- lections. Instead of building traditional library space, as buildings become crowded, warehouse-type structures are being constructed on less-expensive land. Items with low circulation records are be- ing stored on compact shelves to be paged as requested. Thus, traditional browsing access to large, open shelf library collec- tions may be substantially curtailed or even eliminated. This is particularly true at the Physical Sciences Library at the Uni- versity of California at Davis, where one volume will soon need to be moved off site for each new volume added. Volumes to be moved off site must be se- lected in a manner that minimizes incon- venience and provides for maximum doc- ument retrieval. The circulation record is an adequate measure of usage for those items actually borrowed, but it does not take into account information gathered in- house, i.e. those items that are used at the shelf and/or left for refiling. Browsing activity, or unrecorded use as it is sometimes called, has been of concern to librarians since the advent of the open shelf policy, but arriving at estimates of this activity has been most difficult. Some librarians have attempted to overcome this problem by counting refiles as brows- ing use. Although this may provide clues to the noncirculating use of the collection, refiles measure only a portion of in-house use. Taken alone, countingrefiles is not an adequate indication of browsing activity. Shelf consultations do not result in refiles or circulation, where statistics can be mea- sured, compared, collated, or classified. Patrons are free to remove books in an open stack situation, consult them briefly or at length, and then replace them. Only the patron is aware that the item was used and only he can assess the value of the in- formation acquired. Browsing research efforts have gener- ally focused on two general techniques. The first consists of surveying patrons en- tering or leaving the stack area. Question- naires designed to measure the nature and extent of the browsing activity are given to the patron to be filled out. 1 Certain meth- odological problems are associated with this technique. If patrons fill out informa- tion after the fact, knowledge that their ac- tivity is being recorded may influence memory. Then there is the problem of the definition of shelf consultation. Is it con- sidered a shelf consultation only when in- formation is actually found, or does con- sulting the table of contents or index of a book constitute a shelf consultation? If some patrons consider this trivial informa- tion not worthy of reporting, while others Johanna Ross is associate librarian, Physical Sciences Library, University of California at Davis. 269 270 College & Research Libraries do not, responses would not be consid- ered reliable. The second technique, pioneered in the work described by Simon and Fussier, 2 re- lies on questionnaires inserted into books that have been statistically selected. With this method it was possible to determine that books had been disturbed, even if the included questionnaires were not filled out. Pens were provided for a portion of the sample as an incentive tool and the pa- tron was instructed to keep the pen after filling out the questionnaire. A signifi- cantly greater number of the question- naires that had pens attached were filled out than those that did not. A wide range of shelf consultation has been reported in the literature. Slater and Fisher estimated that the average number of items consulted were 4.1 and the aver- age found useful were 2.4. 3 The study by Simon and Fussier estimated that three to nine books were consulted for every re- corded use. 4 C. Harris, 5 in an analysis of refiles, together with questionnaires in- serted in books statistically selected, esti- mated that as many as twenty times the number of books are used in-house than are checked out. With a low. figure of three consultations per recorded use and a high of twenty, these research results reveal a wide range in estimates of browsing activity. Some variation may be accounted for by the dif- ferent subject content of collections and different regulations concerning stack use. What have not been considered as a source of variation are the different inter- pretations of the meaning of the word browse. Does it mean extensive use at the shelf or at a study area? Does it mean a brief consultation? Different concepts of the meaning of the word by those filling out the questionnaires may account for some of these variations. If patrons had been actually observed, the word could have been standardized for the entire study. But researchers have not yet attempted to assess browsing ac- tivity in this manner. Upon examination, the methodological difficulties in assess- ing browsing use by observation did not appear any more difficult than doing it by inserting questionnaires into books statis- July 1983 tically selected. It was decided to under- take such a study where browsing would be defined as the examination of books at the shelf. Browsing intensity would be measured by recording the number of books removed and replaced and the time spent performing the activity. Those items that were taken from the area were not in- cluded as browsing, since these statistics would be recorded either as circulations or refiles. Although circulations and refiles were not considered as part of browsing activity, both were recorded and are given for comparison purposes (see appendix A). To be able to generalize about the meth- odology, observed patron activity must be selected at random. There are two aspects to the sample selection process-time and place. To select time, the hours of opening were divided into fifteen-minute inter- vals. Random tables were used to select the time interval, then to select the day of the week. To select starting places, the non-reference shelf ranges were num- bered and shelves were selected at ran- dom for each fifteen-minute interval. Forty fifteen-minute intervals were ob- served for each week of the study. Thir- teen weeks, one entire quarter and inter- session, were sampled in order that observed usage patterns would be repre- sentative and not influenced by a specific time during the instruction period. Since only one patron could be observed at a time and many individuals could be engaged in browsing activity, the entire li- brary was inspected five minutes before the selected time interval. The place of browsing and time of day were recorded for every patron observed browsing. When this was finished, the first patron located after the randomly selected start- ing point had browsing behavior re- corded. If it was the same person seen on the initial walk-through, the patron was not recounted. When a patron was observed standing at the shelf examining books, an incon- spicuous place providing full view of browsing activity was chosen. Using a clipboard with data forms attached, the observer would pretend to consult the shelves for needed information. To the pa- tron it would appear that the library staff member was gathering independent in- formation. In reality, patron activity at the shelf was being recorded. This unobtru- sive technique appears to have worked well. To our knowledge, none of the pa- trons were aware that their browsing be- havior was the object of study. The day of the week and time of day, as determined by random selection, were al- ready recorded on the form, together with starting location. The browsing area, given in LC class notation, time duration, and the number of books removed andre- placed were the remaining items recorded for each observation. Net browsing was defined as the number of books replaced. If a patron left the book-stack area with book in hand, it was assumed that the book would be counted either as circula- tion or refile. RESULTS Of the 520 fifteen-minute intervals se- lected, there were only 98 time periods when no one was browsing in the library at the selected time. In the remaining 422 sessions, 515 patrons were observed; 384 were first persons seen, while 131 obser- vations were the second or third persons seen during selected intervals. Circulation and refile counts were kept for the same time period to be compared with browsing data. Refiles are not in- cluded in the estimate of browsing use. Refiles are those items removed from the shelf but left on tables, in refile bins, or on refile shelves. Since circulations and re- files are actual counts and browse figures are based on a sample, the figures for all categories were first categorized by LC class number and then the numbers were reduced to percentages for meaningful comparisons. Thus, observed differences in use for the various classes can be seen more readily (see appendix A). Books Removed The number of books that patrons re- moved clusters at the low end of the distri- bution. Just over 50 percent of the patrons removed two books or less. Although the percentage of those who removed more than two books declines steadily, the Research Notes 271 number of books they removed rises. Nearly 25 percent inspected five or more books and approximately 4 percent re- moved ten or more (see appendix B). R2, or the coefficient of determination, indicates that time alone does not account for the number of items removed (R = .339; R2 = .1149). Only 11 percent (R2) of the total variation was explained by this relationship (see appendix C). Sample means were used in this study as point estimates of central tendency. However, because these estimates are based on survey sample data, the figures are subject to potential sampling error. Additional estimates commonly used to indicate the extent of statistical reliability of sample data are interval estimates or "confidence intervals." Confidence intervals specify ranges within which we can expect the actual value to be, with a certain level of confi- dence. The confidence interval (CI) of the sample means, X, as an estimate of the ac- tual mean, can be expressed as: CI = X ± ts/\((n); where X = the same mean; t = the confidence factor associated with the particular level of confidence desired; for the 95 percent confidence level, t = 1.96; s = the standard deviation of sample obser- vations about the mean; n = the sample size; s/\((n) = the standard error of the mean. In this study the observed mean number of books removed was 3.37 and the stan- dard deviation was 2.72 (see appendix D). Since it was assumed that patrons were observed halfway through the browse, the mean number of books removed, X, increases to 6.74 and the standard devia- tion to 5.44. For the 95 percent confidence level (t = 1.06), these figures applied to the above formula with 495 sample obser- vations (n): CI = 6.74 ± 1.96 (5.44)/\{(495) = 6.74 ± .48. At the 95 percent confidence level, the true number of books removed per patron was between 6.26 and 7.2 (6.74 ± .48). Books Replaced The mean number of books replaced was 2.26 (see appendix E). Time spent re- placing books had little to do with the number replaced. R2, or the coefficient of 272 College & Research Libraries determination, revealed that only 13 per- cent of the total variation was explained by this relationship (R2 = .1289; see appendix F). The number of books replaced corre- lates strongly with the number of books removed (R = .907; see appendix G). Again, applying the confidence interval we find: CI = 5.52 ± 1.96 (5.0)/y(495) = 5.52 ± .48. In short, at the 95 percent confidence level, the true number of books replaced per patron was between 5.02 and 6.0 (5.52 ± .48). Time Duration While the mean time patrons spent at the shelf was 6. 94 minutes (this becomes 13.88 when doubled because of being ob- served midway through the browse), ap- proximately one-third of the browsing sessions were for three minutes or less (see appendixe.s Hand 1). Most of the 138 patrons who removed only one book and the 8 patrons who never removed a single volume, but simply scanned spine titles, did not engage in extended browsing ( ap- pendix B). This may be an indication that these users were simply looking for a book by call number and may not have had browsing as the principal reason for com- ing to the shelf. Once at the shelf, how- ever, patrons usually examine more than one book before leaving, with or without one or more. A frequency time table is pro- vided in appendix J. After the first five minutes, the percent- age of patrons leaving begins to decline. If the dips at four, six, and fourteen minutes are ''seen'' as a human tendency to round off time (a stopwatch was not used since observers either used their own watch or July 1983 the clock on the wall), the percent of de- crease in the number of patrons with the increase of time appears smoother. Ap- proximately 20 percent of browsing activ- ity is for fifteen minutes or more (or as- suming they were caught at midpoint, thirty minutes). Though no attempt to qualify browsing was made, some patrons were obviously very engrossed in the ac- tivity (see appendix 1). CI = 13.88 ± 1.96 (10.06)/y(486) = 13.88 ± .89. At the 95 percent confidence level, the time spent at the shelf is between 12.99 and 14.77 minutes. Summary and Conclusion This study, based on observed browsing behavior, corresponds to the findings of the Simon and Fussier study. At the 95 percent confidence level, the number of books replaced per patron was between 5.02 and 6.0. This narrows the range of 3 to 9 books they reported. This estimate falls short of the estimate of 20 made by Harris, but the Harris study counted refiles as part of browsing. This study has employed the unobtru- sive observation technique to determine browsing activity. The technique is simple to use and can be replicated with only a modest budget. If the survey conducted at the Physical Sciences Library at the Uni- versity of California at Davis were repli- cated and extended to other university li- braries and different subject collections, a broader range of browsing behavior could be accumulated. This would enable librar- ians to have a firmer foundation on which to base decisions regarding the manage- ment and access to materials shelved in off-site locations. REFERENCES 1. The most thorough survey was done by Slater and Fisher when they surveyed 600 libraries; 104 cooperated and 6,300 usable questionnaires were analyzed. Margaret Slater and Pamela Fisher, Use Made of Technical Libraries (Aslib Occasional Pub., 2 [London: Aslib, 1969]), p .3. 2. Herman H. Fussier and Julian L. Simon, Patterns in the Use of Books in Large Research Libraries (Chi- cago: Univ. of Chicago Pr ., 1961). 3. Slater and Fisher, Use Made, p.3. 4. Fussier and Simon, Patterns in the Use, p.115. 5. C. Harris, "A Comparison oflssues and In-Library Use of Books," Aslib Proceedings 20:118-25 (Mar. 1977). APPENDIX A Total Library Book Books LC Total Total Total Observed Total Books Books Collection Circulation Refile Browsing Removed Replaced Class Books Circ Refile Browsers Browsers Removed Re£laced Percent Percent Percent Percent Percent Percent A-F 112 5 11 0 5 0 0 .08 .04 .13 .45 .00 .00 G-GZ 1,279 185 69 1 4 6 6 .96 1.55 .84 .36 .37 .56 H-P 1,404 146 91 6 12 17 12 1.06 1.22 1.11 1.08 1.06 1.12 Q 3,784 267 209 10 29 32 15 2.85 2.23 2.55 2.61 2.00 1.39 QA 16,769 2,069 1,316 78 166 302 221 12.64 17.30 16.03 14.94 18.86 20.54 QB 2,723 170 93 8 19 25 14 2.05 1.42 1.13 1.71 1.56 1.30 QC 18,522 1,519 950 60 140 205 134 13.96 12.70 11.57 12.61 12.80 12.45 QD 19,267 1,685 1,599 89 209 276 174 14.52 14.09 19.47 18.81 17.24 16.17 QE 13,839 1,260 398 27 74 99 56 10.43 10.54 4.85 6.66 6.18 5.20 QF-QZ 1,375 175 107 6 16 25 15 1.04 1.46 1.30 1.44 1.56 1.39 R 254 29 24 0 1 0 0 .19 .24 .29 .09 .00 .00 s 802 136 61 2 5 5 2 .60 1.14 .74 .45 .31 .19 T 3,693 206 97 3 12 11 4 2.78 1.72 1.18 1.08 .69 .37 TA 8,963 856 568 42 94 146 102 6.75 7.16 6.92 8.46 9.12 9.48 TC 2,667 222 165 16 31 67 46 2.01 1.86 2.01 2.79 4.19 4.27 TD 3.048 460 226 16 30 63 50 2.30 3.85 2.75 2.70 3.94 4.65 TE 710 78 23 0 7 0 0 .53 .65 .28 .63 .00 .00 TF 68 4 4 0 1 0 0 .05 .03 .05 .09 .00 .00 TG 167 6 8 0 1 0 0 .13 .05 .10 .09 .00 .00 TH 1,217 148 99 2 6 3 2 .92 1.24 1.21 .54 .19 .19 TJ 4,069 364 234 17 34 60 38 3.07 3.04 2.85 3.06 3.75 3.53 TK 7,456 759 746 26 78 72 49 5.62 6.35 9.08 7.02 4.50 4.55 TL 4,829 466 313 24 46 73 46 3.64 3.90 3.81 4.14 4.56 4.28 TN 5,234 331 201 17 37 66 58 3.94 2.77 2.45 3.33 4.12 5.39 ~ TP 5,444 326 428 20 44 41 27 4.10 2.72 5.21 3.96 2.56 2.51 (I) TR 520 30 29 1 3 1 0 .39 .25 .35 .27 .06 .00 ~ TS 1,253 35 68 2 3 3 2 .94 .29 .83 .27 .19 .19 e: n TT 11 1 1 0 0 0 0 .01 .01 .01 .18 .00 .00 ::r' TX-VM 189 9 19 1 2 3 3 .14 .07 .23 .18 .19 .28 z z 3,055 13 55 0 2 0 0 2.30 .11 .67 .00 .00 .00 0 -132,723 11,960 8,212 474* 1,111 1,601 * 1,076* 100.00 100.00 100.00 100.00 100.00 100.00 ~ (I) *Based on sample only . Note: There were 41 cases where LC classification was not recorded . ~ ~ w 274 College & Research Libraries July 1983 APPENDIXB Removed Number Books of Cases Percentase (n) 0 8 1.59 1 138 27.44 2 113 22.46 3 68 13.52 4 53 10.54 5 40 7.95 6 24 4.77 7 16 3.18 8 14 2.78 9 10 1.99 10 5 .99 11 7 1.39 12 3 .60 13 1 .20 14 1 .20 15 2 .40 16 0 0 17 1 .20 18 0 0 19 0 0 20 0 0 21 1 .20 503* 100 * 9 missing cases . t17 missing cases. APPENDIXC Chi Square = 246.37583 with 132 Degrees of Freedom Cramer's V = 0.21715 Contingency Coefficient = 0.58441 Pearson's R = 0.33933 APPENDIXD NUMBER OF BOOKS REMOVED Variable VB Mean 3.366 Std. Error 0.123 Variance 7.431 Minimum 1.000 Range 20.000 Sum 1666.000 Valid Observations= 495 Replaced Number of Cases (n) 130 111 91 60 26 29 14 8 8 5 5 4 2 1 0 1 0 1 0 0 0 0 495t Percentage 26.26 22.43 18.38 12.12 .25 5.86 2.83 1.62 1.62 1.01 1.01 .81 .40 .20 0 .20 0 .20 0 0 0 0 100 Std. Dev . 2.72 Maximum 21.00 APPENDIXE NUMBER OF BOOKS REPLACED Variable V9 Mean 2.26 Std. Error 0.131 Variance 6.232 Minimum 1.000 Range 16.000 Sum 1119.000 Valid Observations = 495 APPENDIXF "V7 Time Duration by Number of Books Replaced Chi Square = 203 .42319 Cramer's V = 0. 22986 Contingency Coefficient = 0.60628 Pearson's R = 0.35935 "V7 Number of Books Removed by V9 Number of Books Replaced Chi Square = 1634.47894 Cramer's V = 0.63804 Contingency Coefficient = 0.90413 Pearson' s R = 0.90698 Variable V7 APPENDIXG APPENDIXH Research Notes 275 Time Duration Std. Dev. 2.496 Maximum 17.000 Mean 6.944 Std. Error 0.228 Std. Dev. 5.031 Variance 25 .306 Minimum 1.000 Maximum 20.000 Range 19.000 Sum 3375 .000 Valid Observations 486 276 College & Research Libraries July 1983 Minutes 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Total Missing Data Time 8-9a.m. 9-10 10-11 11-12 12-1p.m. 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9 9-10 10-11 Missing Total APPENDIX I TIME DURATION Number of Cases 50 57 54 34 73 11 26 24 15 19 12 9 9 1 83 1 1 3 0 4 486 29 APPENDIXJ BROWSING FREQUENCY BY TIME OF THE DAY* Absolute Frequency 25 76 122 128 73 136 92 136 147 84 32 50 41 43 29 1 1,215 Browsing Rei. Freq . (percent) 2.1 6.3 10.0 10.5 6.0 11.2 7.6 11.2 12.1 6.9 2.6 4.1 3.4 3.5 2.4 .1 100.00 Percentages 10.29 11.73 11.11 7.00 15.02 2.27 5.35 4.94 3.09 3.91 2.47 1.85 1.85 .20 17.08 .20 .20 .62 0 .82 Incidence of No Browsing 15 6 9 3 6 0 6 3 1 8 12 13 8 6 5 0 101 *The figures given include patrons encountered on the initial walk-through that did not have browsing behavior recorded . The time of day, day of the week, and place of browsing are recorded for 703 patrons . Only 1 patron could be observed at a time so full data is available for 512 individual browsing sessions . There were a total of 1,215 instances of browsing during the time intervals chosen.