32 Relationships between Association of Research Libraries (ARL) Statistics and Bibliometric Indicators: A Principal Components Analysis Dean Hendrix Dean Hendrix is Coordinator of Education Services in the Health Sciences Library at The State University of New York at Buffalo. ©Dean Hendrix This study analyzed 2005–2006 Web of Science bibliometric data from institutions belonging to the Association of Research Libraries (ARL) and corresponding ARL statistics to find any associations between indicators from the two data sets. Principal components analysis on 36 variables from 103 universities revealed obvious associations between size-dependent variables, such as institution size, gross totals of library measures, and gross totals of articles and citations. However, size-inde- pendent library measures did not associate positively or negatively with any bibliometric indicator. More quantitative research must be done to authentically assess academic libraries’ influence on research outcomes. tatistical relationships be- t w e e n c o l l e c t e d l i b r a r y measures and bibliometric indicators offer a valuable perspective in viewing the research li- brary’s role in the broader context of their institution’s research. Macro analyses measure the overall strength in which a specific public services, collection devel- opment, or fiscal metric is linked with established indicators of research pro- ductivity and impact across institutions. Useful to librarians at major research institutions, these data analyses may in- form their decision-making processes in determining the focus of library services, collections, and staffing in order to affect institutional research outcomes. Using principal component analysis, this study’s objective was to discover if the most rec- ognized library measures, Association of Research Libraries (ARL) statistics, dem- onstrate statistical relationships with Web of Science (WOS) bibliometric indicators, well-regarded measures of an institution’s research productivity and impact. Using homegrown assessment statis- tics, several libraries have quantitatively illustrated the effect of library collections and services on users at the micro level, such as assessments of information literacy efforts on student learning and confidence. However, a paucity of quantitative analy- ses on the institutional effects of libraries’ services and collections exists on a macro level. Closest in scope to this study, John M. Budd studied the relationships between publication statistics at ARL and Asso- ciation of College and Research Libraries (ACRL) member institutions and ARL A Principal Components Analysis 33 statistics. Using rank-order correlations, Budd concluded that rank of total library materials budget, total volumes, and total PhDs awarded correlated positively with rank of total publications—an indicator of raw productivity. To a lesser degree, Budd correlated per capita ranks with total publications at ARL institutions.1 To gauge the adequacy of resource allocations to aca- demic libraries, Dickie and Allen created a library funding model using ARL data and institutional statistics.2 Employing ARL metrics, Mezick moderately correlated library expenditures, collection size, and total number of serials to undergraduate student persistence.3 Background Though controversial, bibliometric indica- tors have become increasingly important in academia. Employed by universities as evaluative tools, bibliometric measures may characterize aspects of research pro- ductivity and influence. Institutionally, university administrators use bibliometric indicators to assess their research groups, departments, and schools as well as their own university’s comparative standing domestically and internationally. These appraisals often assist in determining intra-university financial allocations4 and benchmarks5 in addition to identifying the extent of extramural collaboration,6 institutional strengths,7 and new and promising research fields.8 To individual university faculty, the use of bibliometric indicators—such as gross totals of publi- cations and citations, h-indices, and jour- nal impact factors in tenure, promotion, and reappointment decisions—can affect the direction of their career.9 Advocating for the interest of its 123 institutional members, ARL is one of the most influential organizations in the li- brary world. As a part of its mission, ARL annually compiles statistics that describe collections, services, human resources, and finances at the top research libraries in the United States and Canada. As a requirement of membership, ARL librar- ies “…must contribute the data necessary to establish the membership indices and to compile the annual ARL Statistics.”10 From these data, an ARL index score is calculated, and subsequently a rank is assigned for member libraries. The professional literature contains studies that discussed the meaning,11 temporal characteristics,12 internal library relation- ships,13 and limitations14 of ARL statistics, but very few explored connections with broader institutional measures. Methods The population for this study is academic institutions that have ARL libraries on their campuses (n=113). To retrieve bibliometric data, the author searched WOS using inclu- sive search strategies to retrieve institution- specific articles. Employing the “Analyze Results” feature in WOS, the author refined the results to include articles published in 2005 and 2006. The author refined by insti- tution and searched for the institutions and name variants. The author exercised due diligence in capturing all possible name variants of a university (for instance, Univ N Carolina, UNC, Univ North Carolina) within WOS. Bibliographic information from satellite campuses or other universi- ties comprising a larger university system were not included. From the WOS searches, the author retrieved four measures for each ARL academic institution: • Total articles published, 2005–2006 • Total citations to articles published in 2005–2006 • Institutional h-index, 2005–2006 • Total articles not cited, 2005–2006 At the time of data gathering, the lat- est published ARL statistics were from 2005–2006. The 2005–2006 ARL Statistics15 provided the following measures for most ARL libraries: • Total number of faculty • Total number of full-time students • Total number of library presentations • Total number of reference transac- tions • Total number of circulation transac- tions 34 College & Research Libraries January 2010 • Total number of interlibrary loan transactions • Total number of professional librar- ians • Total number of library staff • Total amount of library expendi- tures • Total amount of library expendi- tures on library materials • Total amount of library expendi- tures on monographs • Total amount of library expendi- tures on serials • Total amount of library expendi- tures on electronic resources • Total number of volumes • Total number of serials Ten libraries failed to disclose certain pieces of information to ARL; thus, the author could not use data from those universities during the principal compo- nents analysis. These libraries are: University of Cali- fornia, Berkeley; Dartmouth University; Georgia Institute of Technology; Harvard University; University of Michigan; Ohio State University; Oklahoma State Uni- versity; University of Pennsylvania; Rice University; and University of Wisconsin. All of the aforementioned measures were size-dependent indicators—mea- sures entirely based on the sum totals (size). Though they provide a picture of gross productivity and expenditures, size-dependent indicators do not measure institutional research impact, individual faculty productivity, individual library productivity, or expenditures on a stan- dardized scale. To address this problem and provide more robust data sets, the author synthesized 18 size-independent indicators from the size-dependent data. These included: • Citations per article • Impact index • Percentage of uncited articles • Articles per faculty member • Citations per faculty member • Library presentations per capita (faculty and students) • Reference transactions per capita • Circulation transactions per capita • Interlibrary loan transactions per capita • Professional librarians per capita • Library staff per capita • Library expenditures per capita • Percentage of total library expendi- tures spent on materials • Percentage of material expenditures spent on monographs • Percentage of material expenditures spent on serials • Percentage of material expenditures spent on electronic resources • Volumes per capita • Serials per capita Commonly used in the social sciences, principal component analysis (a mul- tivariate statistical technique) reduces the known variables to common hidden variables known as factors, which may reveal fundamental relationships between variables. To uncover the significant fac- tors, principal component analysis uses eigenvalues, a value that explains the vari- ance of the known variables linked with each factor. Rotational methods, such as Varimax, simplify the data interpretation by diversifying the variable loadings, thus showing which variables cluster together more clearly. Employing SPSS v.15 for statistical analysis, the author conducted a principal component analysis to examine which variables cluster together.16 Results of the Principal Components Analysis Variance For the 103 universities eligible for analy- sis, principal components analysis of the 37 variables sought to reveal significant relationships. Initially, the author ran a principal components analysis that ex- tracted all factors with eigenvalues over one, and subsequently rotated orthogo- nally using the Varimax technique. Eight factors with eigenvalues over one were extracted initially; however, review of the scree plot and the total explained variance led the author to recalculate the principal components analysis, limiting to five fac- A Principal Components Analysis 35 tors. The five extracted factors explained 75.6 percent of the total variance. Clustering of Variables Table 1 shows how the studied variables grouped together. The variables are sorted by the significance of their relationship to the factor. The first factor characterizes the size of an institution and its library as most of the size-dependent variables clustered together. Only two of the nineteen variables related to the first fac- tor—percentage of the materials budget spent on monographs and percentage of the materials budget spent on serials— were size-independent. Only one vari- able—percentage of the materials budget spent on serials—showed a negative and significant relationship. The second fac- tor illustrates the significant relationship between most of the size-independent library measures. The only anomaly, total number of students, demonstrated a negative and significant relationship with these per-capita measures due to its overwhelming influence in calculat- ing the per-capita denominator for these measures. The third factor grouped both size-dependent and size-independent bibliometric data together. The fourth factor was the most difficult to decipher. The factor describes a positive and sig- nificant relationship between reference service and library presentations, but it also indicates a negative and significant relationship between the aforementioned measures and two size-independent ex- penditure measures, percentage of total library expenditures spent on materials and percentage of material expenditures spent on electronic resources. The fifth factor expresses the link between the two interlibrary loan measures in the matrix. Table 2 illustrates statistically the clus- tering of factor loadings of essential fac- tors and the strength of their associations. Considering the size of the population (n=103), the threshold value for significant factor loadings was close to 0.512, accord- ing to Stevens.17 All factor loadings under 0.512 were excluded in the table. Discussion of the Results Principal components analysis confirmed some obvious assumptions regarding the size of an institution and its libraries. Described by the first component, gross productivity in terms of total articles and citations exhibited a significant relation- ship with all gross library expenditure measures, gross library and university staffing measures, and most gross library services measures. In the same vein, larger universities and libraries showed significant associations with larger total numbers of uncited articles. Inextricably linked, the percentage of a library’s material budget spent on mono- graphs and the percentage of a library’s material budget spent on serials clustered around the size-dependent measures, too. The data suggested that smaller universi- ties and libraries tended to spend a larger percentage of their materials budgets on serials. Consequently, gross productiv- ity in terms of total articles and citations showed significant statistical links with libraries that spent a larger percentage of their materials budgets on monographs. The author postulated that smaller librar- ies may have instituted more severe cuts in monographic acquisitions to keep pace with serials inflation. For size-independent bibliometric measures, no associations with any li- brary measures revealed themselves. Vol- ume counts, library services transactions, and budgets exhibited no measurable link to bibliometric measures that afford equitable institutional comparisons, such as citations per article, impact index, and articles per faculty member. Therefore, the author concluded that ARL measures do not demonstrate any association with the general impact and influence of an individual article or faculty member. Hardest to define, the fourth factor may indicate that libraries that perform more ref- erence and instruction services dedicate less of their budget toward materials purchases, especially electronic resources. Because of the strong negative association, the fourth factor also suggests that libraries that al- 36 College & Research Libraries January 2010 Table 1 Variable Clustering from the Principal Components analysis in Order of Significance Factor 1: Size of an institution and its library Factor 2: Per capita library data Factor 3: bibliometric data Factor 4: The relationship between library public services and library expenditures Factor 5: Interlibrary loan data Total amount of library expenditures on materials (SD) Total library expenditures per capita (SI) Citations per faculty member (SI) Reference transactions per capita (SI) Total number of interlibrary loan transactions (SD) Total amount of library expenditures (SD) Library staff per capita (SI) h-index (SD) Total number of reference transactions (SD) Interlibrary loan transactions per capita (SI) Total number of library staff (SD) Volumes per capita (SI) Citations per article (SI) Total number of library presentations (SD) Total amount of library expenditures on monographs (SD) Librarians per capita (SI) Total number of citations (SD) Percentage of material expenditures spent on electronic resources (SI) Total number of volumes (SD) Serials per capita (SI) Articles per faculty member (SI) Percentage of total library expenditures spent on materials (SI) Total amount of library expenditures on serials (SD) Total number of students (SD) (Negative association) Impact index (SI) Total number of librarians (SD) Library presentations per capita (SI) Total number of articles (SD) Total number of serials (SD) Interlibrary loan transactions per capita (SI) Not cited article percentage (SI) (Negative association) Total number of faculty (SD) Total number of articles with no citations (SD) A Principal Components Analysis 37 Table 1 Variable Clustering from the Principal Components analysis in Order of Significance Factor 1: Size of an institution and its library Factor 2: Per capita library data Factor 3: bibliometric data Factor 4: The relationship between library public services and library expenditures Factor 5: Interlibrary loan data Total number of circulation transactions (SD) Total number of students (SD) Total number of articles with no citations (SD) Total amount of library expenditures on electronic resources (SD) Percentage of the materials budget spent on serials (SI) (Negative association) Total number of articles (SD) Percentage of the materials budget spent on monographs (SI) Total number of reference transactions (SD) Total number of library presentations (SD) Total number of citations (SD) SD = Size Dependent Variable SI = Size Independent Variable 38 College & Research Libraries January 2010 Table 2 Rotated Component Matrix from Principal Components analysis, Varimax Rotation Factor loadings 1 2 3 4 5 Total amount of library expenditures on materials 0.867 Total amount of library expenditures 0.854 Total number of library staff 0.851 Total amount of library expenditures on monographs 0.827 Total number of volumes 0.805 Total amount of library expenditures on serials 0.735 Total number of librarians 0.713 Total number of serials 0.678 Total number of faculty 0.672 Total number of circulation transactions 0.653 Total amount of library expenditures on electronic resources 0.615 Percentage of the materials budget spent on serials –0.613 Percentage of the materials budget spent on monographs 0.591 Total library expenditures per capita 0.923 Library staff per capita 0.914 Volumes per capita 0.892 Librarians per capita 0.884 Serials per capita 0.805 Total number of students 0.643 –0.670 Library presentations per capita 0.657 Citations per faculty member 0.844 h-index 0.842 Citations per article 0.825 Total number of citations 0.513 0.793 Articles per faculty member 0.772 Impact index 0.738 Total number of articles 0.603 0.707 Not cited article percentage –0.704 Total number of articles with no citations 0.628 0.641 Circulation transactions per capita Percentage of total library expenditures spent on materials –0.681 A Principal Components Analysis 39 locate more of their budgets to collections and electronic resources field fewer refer- ence questions and perform fewer library presentations. This may point to an inherent tension between public services and col- lection development at ARL libraries. Fur- thermore, the inclusion of the percentage of material expenditures spent on electronic resources variable in this clustering raises related questions. Does the administration of electronic resources require so much more library staff and time that public services are scaled back to accommodate it? Do more expensive electronic resources also require more library resources that subsequently affect the provision of refer- ence and instruction services? Limitations Traditional ARL measures, such as vol- ume counts and total expenditures, have severe limitations qualitatively, as they only measure size and temporal growth. Assessing the effectiveness and quality of library services through traditional ARL statistics is not possible. Moreover, as libraries proffer more online services, the significance of evaluating electronic collections and services (also known as e-metrics) has increased. ARL measures used in this study did not include the raw numbers of database usage statistics, e-journal access statistics, e-book access statistics, and online tutorial statistics or any qualitative assessment of electronic collections and services. ARL is involved with several statistical initiatives to fill the void that address qualitative measures and electronic collections and services, such as LibQUAL+™,18 E-Metrics,19 and MINES for Libraries™ Project, 20 but they were not analyzed in this study. Other limitations include inherent prob- lems with the self-reporting methodology employed by ARL. Libraries may define the same measure differently, thus report- ing inconsistent numbers. Data omissions from some major universities required those libraries to be removed from the analysis. Part-time faculty members were not considered in the faculty counts. With ISI data, several limitations also emerge. First, citations errors exist within the Web of Science database. Moed stated that 7 percent of all cited references were erroneous.21 The author attempted to capture all the institutional name vari- ants while searching but acknowledges that records with unfamiliar institutional names and acronyms may have been missed. Another limitation with ISI data is selectivity. Proceedings, patents, technical reports, and many international journals are not indexed by the database. The au- Table 2 Rotated Component Matrix from Principal Components analysis, Varimax Rotation Factor loadings 1 2 3 4 5 Reference transactions per capita 0.641 Total number of reference transactions 0.556 0.584 Percentage of material expenditures spent on electronic resources –0.549 Total number of library presentations 0.516 0.534 Total number of interlibrary loan transactions 0.840 Interlibrary loan transactions per capita 0.595 0.682 Extraction Method: Principal Component Analysis. Rotation Method: Varimax with Kaiser Normalization. Factor Loading Threshold Value: 0.512 40 College & Research Libraries January 2010 thor did not apply fractional attribution methodologies in cases where multiple authors worked at different universities; all authors received equal treatment. Furthermore, self-citations were included, which may alter the data set. However, bibliometric analyses concluded that the effect of self-citations is inconsequential when performing macro level analyses.22 Conclusions This macro analysis demonstrated that tra- ditional ARL measures of library services, library expenditures, and library collections are not reliable predictors of research influ- ence or impact at the individual researcher and article levels. To be clear, this does not mean librarians and their work do not af- fect the research quality and productivity of the users they serve. Great numbers of case studies in the library and information science literature reflect that the opposite is true. In fact, micro level studies may be preferable in articulating the library’s influ- ence on individual researchers. As expected, the size of an institution demonstrates positive associations with li- brary metrics and with bibliometric totals. These raw data show that a larger library budget and more librarians are linked to more institutional citations, but these crude statistics are not the foundation for proving a library’s impact. It merely points out that larger universities are character- ized by larger faculties, which produce more scholarship as a whole, and larger libraries, which spend more money and offer more services as a whole. The afore- mentioned analysis of size-independent measures diminishes meaningful associa- tions between library service and collection metrics with bibliometric indicators. On an internal library level, this study discovered an oppositional relationship between some public services measures and collection development budget allo- cations. At an individual library level, this tension may or may not be readily appar- ent. Nonetheless, library administrators may want to keep this interrelatedness in mind to strike a healthy balance that works for their institutions. Assessing outcomes of library services and collections in relation to a university’s mission and objectives in a meaning- ful way proves difficult. Often, these missions and objectives are not clearly defined (or even quantifiable) and can change from university administration to university administration. For univer- sities whose missions include scholarly research, bibliometric indicators may of- fer standards from which to measure institutional outcomes. Further research using different data sets and a variety of statistical tests and measures—includ- ing canonical correlation and power law equations—must be performed to genu- inely quantify libraries’ influence on their parent universities at the macro level. Notes 1. John M. Budd, “Faculty Publishing Productivity: An Institutional Analysis and Comparison with Library and Other Measures,” College & Research Libraries 56 (Nov. 1995): 547–54; John M. Budd, “Increases in Faculty Publishing Activity: An Analysis of ARL and ACRL Institutions,” College & Research Libraries 60 (July 1999): 308–15; John M. Budd, “Faculty Publishing Productiv- ity: Comparisons over Time,” College & Research Libraries 67 (May 2006): 230–39. 2. Frank R. Allen and Mark Dickie, “Toward a Formula-Based Model for Academic Library Funding: Statistical Significance and Implications of a Model Based Upon Institutional Charac- teristics,” College & Research Libraries 68 (Mar. 2007): 170. 3. Elizabeth M. Mezick, “Return on Investment: Libraries and Student Retention,” Journal of Academic Librarianship 33 (Sept. 2007): 561–66. 4. Christine L. Borgman and Jonathan Furner, “Scholarly Communication and Bibliometrics,” Annual Review of Information Science and Technology 36 (2002): 3–72; Grant Lewison, Robert Cottrell, and Diane Dixon, “Bibliometric Indicators to Assist the Peer Review Process in Grant Decisions,” Research Evaluation 8 (Apr. 1999): 47–52; Penelope S. Murphy, “Journal Quality Assessment for Performance Based Funding,” Assessment & Evaluation in Higher Education 23 (Jan. 1998): 25–31 A Principal Components Analysis 41 5. Borgman and Furner, “Scholarly Communication”; E.C.M. Noyons, H.F. Moed, and M. Luwel, “Combining Mapping and Citation Analysis for Evaluative Bibliometric Purposes: A Bibliometric Study,” Journal of the American Society for Information Science 50 (1999): 115–31. 6. Borgman and Furner, “Scholarly Communication”; Eugene Garfield, “What Citations Tell Us About Canadian Research,” Canadian Journal of Information and Library Science-Revue Canadienne Des Sciences De L Information Et De Bibliotheconomie 18 (1993): 14–35. 7. Borgman and Furner, “Scholarly Communication”; Mu-Hsuan Huang, Han-wen Chang, and Dar-Zen Chen, “Research Evaluation of Research-Oriented Universities in Taiwan from 1993 to 2003,” Scientometrics 67 (June 2006): 419–35; Joachim Schummer, “The Global Institutionaliza- tion of Nanotechnology Research: A Bibliometric Approach to the Assessment of Science Policy,” Scientometrics 70 (Mar. 2007): 669–92. 8. Borgman and Furner, “Scholarly Communication”; Sybille Hinze, “Bibliographical Cartogra- phy of an Emerging Interdisciplinary Discipline: The Case of Bioelectronics,” Scientometrics 29 (Mar. 1994): 353–76; Loet Leydesdorff, Susan Cozzens, and Peter Van den Besselaar, “Tracking Areas of Strategic Importance Using Scientometric Journal Mappings,” Research Policy 23 (Mar. 1994): 217–29. 9. Borgman and Furner, “Scholarly Communication”; Blaise Cronin and Helen Barsky At- kins, “The Scholar’s Spoor,” in The Web of Knowledge: A Festschrift in Honor of Eugene Garfield, eds. Blaise Cronin and Helen Barsky Atkins (Medford, N.J.: Information Today, 2000), 1–7; Richard J. Epstein, “Journal Impact Factors Do Not Equitably Reflect Academic Staff Performance in Different Medical Subspecialties,” Journal of Investigative Medicine 52 (Dec. 2004): 531–36; Eugene Garfield, “How to Use Citation Analysis for Faculty Evaluations, and When Is It Relevant? Part 1,” Essays of an Information Scientist 6 (1983): 354–62; Eugene Garfield, “How to Use Citation Analysis for Faculty Evaluations, and When Is It Relevant? Part 2,” Essays of an information scientist 6 (1983): 363–72; Robert G. Maunder, “Using Publication Statistics for Evaluation in Academic Psychiatry,” Canadian Journal of Psychiatry-Revue Canadienne De Psychiatrie 52 (Dec. 2007): 790–97. 10. Association of Research Libraries, “Procedures for Membership in the Association of Research Libraries” (2008). Available online at www.arl.org/arl/membership/qualproc.shtml. [Accessed 28 July 2008]. 11. Martha Kyrillidou, “Reshaping ARL Statistics to Capture the New Environment,” ARL 256 (Feb. 2008): 9–11; Martha Kyrillidou, “To Describe and Measure the Performance of North American Research Libraries,” IFLA Journal 27 (2001): 257–63. 12. Kyrillidou, “Reshaping ARL Statistics”; Martha Kyrillidou, “Research Library Trends: ARL Statistics,” The Journal of Academic Librarianship 26 (Nov. 2000): 427–36; Martha Kyrillidou, “Serials Trends Reflected in the ARL Statistics 2002–03,” ARL 234 (June 2004): 14–15. 13. E. Stewart Saunders, “The Effect of Bibliographic Instruction on the Demand for Reference Services,” portal: Libraries and the Academy 3 (Jan. 2003): 35–39. 14. Johann van Reenen, “Library Budgets and Academic Library Rankings in Times of Transi- tion,” The Bottom Line: Managing Library Finances 14 (2001): 213–18; Kendon Stubbs, “Apples and Oranges and ARL Statistics,” Journal of Academic Librarianship 14 (Sept. 1988): 231–35. 15. Martha Kyrillidou, Mark Young, and the Association of Research Libraries, ARL Statistics, 2005–06: A Compilation of Statistics from the One Hundred and Twenty-Three Members of the Association of Research Libraries (Washington, D.C.: Association of Research Libraries, 2008). 16. Dennis Child, The Essentials of Factor Analysis (London: Continuum International Publish- ing Group, 2006); Paul Kline, An Easy Guide to Factor Analysis (London: Routledge, 1994); Andrew Laurence Comrey and Howard B. Lee, A First Course in Factor Analysis (Hillsdale, N.J.: Lawrence Erlbaum Associates, 1992). 17. James Stevens, Applied Multivariate Statistics for the Social Sciences, 4th ed. (Mahwah, N.J.: Lawrence Erlbaum Associates, 2002). 18. Association of Research Libraries, “LibQUAL+™: Charting Library Service Quality” (2008). Available online at www.libqual.org/. [Accessed 21 August 2008]. 19. “E-Metrics: Measures for Electronic Resources” (2007). Available online at www.arl.org/ stats/initiatives/emetrics/index.shtml [accessed 21 August 2008]; Measures for Electronic Resources (E-Metrics) (Washington, D.C.: Association of Research Libraries, 2002). 20. Martha Kyrillidou, Toni Olshen, Brinley Franklin, and Terry Plum, “MINES for Libraries™: Measuring the Impact of Networked Electronic Services and the Ontario Council of University Libraries’ Scholar Portal, Final Report” (Washington, D.C.: Association of Research Libraries, 2006). 21. Henk F. Moed, “The Impact Factors Debate: The ISI’s Uses and Limits,” Nature 415 (Feb. 2002): 731–32. 22. Wolfgang Glanzel, Koenraad Debackere, Bart Thijs, and András Schubert, “A Concise Review on the Role of Author Self-Citations in Information Science, Bibliometrics and Science Policy,” Scientometrics 67 (May 2006): 263–77; Bart Thijs and Wolfgang Glanzel, “The Influence of Author Self-Citations on Bibliometric Meso-Indicators: The Case of European Universities,” Scientometrics 66 (Dec. 2005): 71–80.