meyer.p65 A Tool to Assess Journal Price Discrimination 269 A Tool to Assess Journal Price Discrimination1 Richard W. Meyer This econometric study tests pricing practices of publishers and their monopoly power. It suggests that traditional publishers will retain their market clout as they shift to offering electronic publications. Librarians’ common experience with price discrimination was corroborated by a powerful model comparing prices charged to institutions while holding constant for production costs, source of publication, discipline areas, and the availability of titles in electronic format. The model also provides a robust selection tool to compare actual prices to model-predicted prices among the subscriptions within any given collection and to predict those that, statistically, are significantly overpriced. The study results reveal that commercial publishers are not the only ones that appear to over price titles by a statistically significant amount. Campuses face contin ued increases in prices for traditional and electronic resources, but sta tistical modeling offers an opportunity for controlling costs. ibrarians decide whether to ac- sis of the prices of selected titles can be a quire or retain a subscription to powerful way to overcome the negative any given serial based on three impact of publisher monopoly power on variables: relevance, quality, the expenditure pattern of libraries. and price. Comparing its general theme to the academic program determines the relevance of a journal. A strong correlation between the subject of the journal and the disciplines covered by the relevant pro gram tends to support a favorable deci sion. The quality of a title likely will be determined subjectively or be dependent on survey information and reputation in the field. A determination of high quality also will tend to support acquisition. These two variables are not readily amenable to quantitative analysis, but price is. In fact, as this article shows, an econometric analy- Economic theory predicts that when barriers to entry diminish, monopoly power is eroded by the entry of new com petitive products. This means that as mo nopoly power diminishes, prices tend to be established at rates closer to the value placed on them by libraries. By compari son, the open systems architecture ap proach taken by IBM in the early days of the development of the personal com puter (PC) market provided opportuni ties to many upstart firms that success fully emerged with competitively priced computers. In contrast, the closed archi- Richard W. Meyer is Dean and Director of Libraries at the Georgia Institute of Technology; e-mail: richard.meyer@library.gatech.edu. Mr. Meyer was Director of the Library at Trinity University in San Antonio when this project was conducted. The author gratefully acknowledges the financial support of the Andrew W. Mellon Foundation for underwriting this study and for the work of Tanya Pinedo, who as sembled the data for the project. Any errors remain those of the author. 269 mailto:richard.meyer@library.gatech.edu 270 College & Research Libraries May 2001 tectural approach applied by Apple Com puter imposed a barrier to entry that, over time, dampened the growth of the mar ket and kept prices high for their alterna tive technology. Extending this theory to the journal lit erature suggests that perhaps the oppor tunity provided by electronic technology will lower the barriers to entry of new journals. In effect, because it might be easier for new scholarly journals to get started by using desktop publishing tech nology, the lowered barriers offered by this technology would erode the mo nopoly power held by the major print publishers. Indeed, desktop publication technology, along with major improve ments in digital storage capacity, has proved effective as a means of introduc ing new periodicals. However, new tech- Because most publishers offer some products in print only and others within the described electronic set, the prices of the electronic version might be expected to reflect an erosion of monopoly power. nology also has been more effective in allowing existing print journals to offer electronic counterparts. In less than a de cade, more than a thousand peer-re viewed electronic journals have been in troduced to the scholarly environment. ARL’s Directory of Electronic Journals, Newsletters, and Academic Discussion Lists listed 417 peer-reviewed titles in 1996.2 That figure grew to 1,049 in 1997, with hundreds of existing print titles expand ing their outreach in electronic versions. This rapid expansion of the e-journal market suggests that, indeed, there may be cost advantages in the electronic envi ronment that offer substance to the hope that periodical price inflation might come under control. Alternatively, expanded profit opportunity also might motivate a move to offer electronic periodicals among traditional publishers. With that in mind, the author designed an experi ment to determine whether price inflation might be dampened by electronic schol arship. The rest of this article describes the results of an econometric analysis of prices for 859 periodical titles for three consecutive years. The article concludes with a description of an analytical tool that may be used to assess journal prices. Economic Theory and Model To analyze the effects of electronic avail ability on journal price, a straightforward model was established that applied ordi nary least squares (OLS) regression on cross-sectional data similar to analyses reported by others.3 Earlier models typi cally regressed price on a number of vari ables to distinguish the statistical rel evance of publisher type in determining price.4 Not only do these studies confirm librarians’ belief that certain publishers practice price discrimination, but they also show that periodical prices are driven by other factors, as well. The costs of pro duction based on frequency of issue, number of pages, and presence of illus trations impact price. The availability of alternative revenue from advertising and the exchange rate risk for foreign publish ers also affect price. Quality measures on the content, such as number of times a periodical is cited, affect demand, which then impacts price. Production economies of scale available to some journals with large circulation also have been shown to affect price.5 Additional work using an alternative model examined the possibility that pub lishers exercise monopoly power in set ting prices. By substituting a measure of monopoly power in place of price, it has been shown that publishers have some ability to influence price.6 Theory predicts that in a competitive market, even when it is characterized as monopolistic com petition, the price offered to individuals will be elastic.7 Faced with a change in price of the subscriptions purchased from her or his own pocket, a scholar will act discriminably. Raise the price to individu als, and some will cancel their subscrip tions in favor of access to a library. Thus, the price of periodicals to individuals is a A Tool to Assess Journal Price Discrimination 271 determinant of demand for library ac cess.8 In contrast, the price to libraries, which is often much higher than the price to individuals, is set at a level intended to extract consumer surplus. The differ ence in these prices should offer a reason able measure of the extent of monopoly power, assuming that the individual sub scription price is an acceptable proxy for the marginal cost of production.9 Even if not completely true, some measure of monopoly power is represented by the difference in the prices. If a dummy vari able is then included in the data set for whether each journal is available elec tronically, this measure should vary sta tistically on that dummy. By modifying the earlier models, the analysis here seeks to determine whether monopoly power may be eroded in the electronic market. The methodology ap plied uses two specifications for an OLS regression model. The first regresses in stitutional price on the characteristics of a set of journal titles held by Trinity University’s library. The data set devel oped is considerably larger than those used in previous studies. Therefore, this study attempts to confirm the earlier works that concentrated on economics or chemistry journals across a larger set of disciplines. The specification includes the variables established earlier: frequency of publication, circulation, pages per year, and several dummy variables to control for whether the journals contain adver tising and for country of publication. Dummy variables are included for type of publisher, with the residual being com mercial. A second specification regresses the difference in price for libraries com pared to individuals on the same set of variables. Both specifications also include an additional dummy for electronic avail ability, which is intended to show whether price or monopoly power of given journals is driven in part by that publishing opportunity.10 It is anticipated that at the margin, the impact on publishers of Trinity canceling some of its print subscriptions would be trivial. However, national availability of the electronic versions could precipitate cancellations among many institutions in favor of electronic access. Prices then would be adjusted accordingly. Because most publishers offer some products in print only and others within the described electronic set, the prices of the electronic version might be expected to reflect an ero sion of monopoly power. Thus, the cross- sectional data would capture the effect of electronic availability on monopoly power. The larger data set used here, com posed of 859 periodical titles represent ing nearly all the academic disciplines, mitigates several concerns experienced by other investigators. The only study found in the literature thus far that looks at pub lishers from the standpoint of the exer cise of monopoly power focused on price discrimination.11 This project extends that analysis in two ways. First, a much broader database is used. Most of the pre vious work focused on limited data sets of fewer than a hundred titles in a single academic discipline. Second, the analysis is extended by assuming the existence of price discrimination given the difference in price to individuals versus libraries for most scholarly journals. With controls in the model for previous discoveries re garding price discrimination, this project attempted to test the null hypothesis that monopoly power will not increase in the electronic domain. It is impossible to distinguish the spe cific price of each journal from the elec tronic titles licensed from vendors such as Ebsco, UMI, and other aggregators be cause each is priced at a flat fee for the entire set of the aggregator. This pricing scheme may reflect an attempt by pub lishers to capture revenue lost to interli brary lending. However, it also may re flect publisher expectations that article demand will increase when user nondol lar costs decrease. Therefore, monopoly power will be reflected back on the sub scription price of print versions. As a re sult, the price of print copies is assumed as a proxy for the specific electronic price of each title if the copies could be made available on a title-by-title basis. http:discrimination.11 http:opportunity.10 272 College & Research Libraries May 2001 However, an alternative result could emerge. In monopolistic competition, anything that differentiates a product may increase its monopoly power. Firms that sell laundry detergent spend tremen dous amounts of money on advertising. They do so to create the impression that their product is qualitatively distinguish able from others. Similarly, it may be that electronic availability of specific titles will create an impression of superior quality. The general model of the first specifi cation is written: y = a + b LOWPRICE j 1 j + b EUROPE + b GRTBRIT + b OTHER + 2 j 3 j 4 j b RISK b ASSOC + b FOUND + 5 j + 6 j 7 j b GOVERN + b UNIV + b FREQ + 8 j 9 j 10 j b ARTPGS b PEER + b CCC + 11 j + 12 j 13 j b ARTILLUS + b ADVERT + 14 j 15 j b SUBFEE + b AGE + b BOOK + 16 j 17 j 18 j b TOTCITES b IMPACT b SCIENCE 19 j + 20 j + 21 j + b SOCSCI + b ELECTRNIC b CIRC 22 j 23 j + 24 j e where, y equals the library price + j ; (INSTIT) for journal j = 1, 2, 3, … n. The general model of the second specification is written the same way except that y equals an index of monopoly power (LERNER). The definitions of indepen dent variables are given in table 1, along with the expected signs on and calcula tions of the parameters b 1 through b 24 to be estimated by traditional single-regres sion techniques. It should be understood that most of the variables listed in table 1 were sug gested based on previous studies that have demonstrated their appropriateness. Testing with the regression model was required to determine the variables ulti mately useful to this study. Additional variables were introduced as the experi ments suggested them. A very brief ra tionale for the expected sign and the im portance of the variables is in order. If the difference in price between what publishers charge libraries versus indi viduals represents price discrimination, a variable for the individual price (LOWPRICE) will be a significant predic tor of price to institutions (INSTIT). As the individual experiences an increase in price, substitution of access to the library will take place. That is, higher individual prices will shift users toward use of the library, thus raising demand for library subscriptions, which will pull institu tional prices higher. The sign on this vari able is expected to be positive. One group of variables deals with the issue of price discrimination based on the monopoly power that can be exercised by foreign publishers. Publishers in Great Britain (GRTBRIT), Western Europe (EU ROPE), and other countries outside the United States (OTHER) may have enough market power to influence price. There fore, these variables will carry a positive sign if a sizeable market influence is ex erted. Some of these publishers also will be concerned with currency exchange risks (RISK), which they will adjust for in prices. However, because they offer discounts through vendors to libraries that prepay subscriptions, this variable will carry a negative sign if the price to individuals captures most of the financial burden of risk adjustment. On the basis of observations over time, it is expected that commercial publishers price discriminate more than their non profit counterparts do. Therefore, in com parison to the commercial residual, asso ciations and societies (ASSOC), govern ment agencies (GOVERN), university presses (UNIV), and foundations (FOUND) will capture the generally lower prices of the nonprofits. The signs on all these are expected to be negative. All the publishers will experience pro duction costs, processing and communi cation expenses, that can be exposed through variables that control for fre quency (FREQ), total pages of articles printed per year (ARTPGS), peer review (PEER), copyright clearance registration expenses (CCC), and the presence of graphics, maps, and illustrations (ARTILLUS). All of these costs will affect price positively to the extent that they are passed on to buyers. On the other hand, the inclusion of advertising (ADVERT) will provide additional revenue to that of sales, so this variable is expected to be negative because journals that include ads will have less incentive to extract revenue A Tool to Assess Journal Price Discrimination 273 TABLE 1 Regression Variables with Expected Signs Dependent Variables: INSTIT = The price for library subscriptions LERNER = Monopoly power as represented by INSTIT minus LOWPRICE Independent Variables: Dummies Expected Sign LOWPRICE = Price for individuals + EUROPE = 1 if the journal published in Europe, 0 otherwise � + GRTBRIT = 1 if the ournal published in Great Britain, � + o otherwise OTIER = 1 if the journal published outside U.S., Europe, and � + Great Britain, 0 otherwise RISK = Standard deviation of the monthly free-market exchange + rate between the currency of the home country of a foreign publisher to the U.S. dollar ASSOC = 1 if the ournal published by an association, � − o otherwise FOUND = 1 if the journal published by a foundation, � − o otherwise GOVERN = 1 if the journal published by a govt agency, � − o otherwise UNIV = 1 if the journal published by a university press, � − o otherwise FREQ = Number of issues per year + ARTPGS = Number of pages printed per year + PEER = 1 if article submissions are peer reviewed, 0 otherwise � + CCC = 1 if ournal is registered with the CCC, 0 otherwise � + ARTILLUS = 1 if the journal contains graphics or illustrations, � + o otherwise ADVERT = Number of pages of advertising in journal for the − year SUBFEE = 1 if journal requires authors to submit page charges, � − o otherwise AGE = Current year minus the date the journal first published − BOOK = 1 if the journal publishes book reviews, o otherwise � + TOTCITES = Sum of the ISI citation measures + IMPACT = Index of impact from the ISI citation studies + SCIENCE = 1 if the ournal is in the humanities, 0 otherwise � + SOCSCI = 1 if the ournal is in the social sciences, 0 otherwise � + ELECTRNIC = 1 if available in electronic form, 0 otherwise � − CIRC = Reported number of subscriptions to the journal + 274 College & Research Libraries May 2001 through sales. Similarly, control for jour nals receiving revenue from page charges or submission fees is captured by another dummy variable (SUBFEE), which should be significant and negative. New entries into the publishing arena are expected to experience advertising costs in order to increase awareness of their products, which will be partially passed on to con sumers. Therefore, age (AGE), which is the difference between the current date and the date the journal started, will be a negative predictor of price and monopoly power. A dummy variable for journals that publish book reviews (BOOK) is in cluded to control for publishing cost dif ferences in this type of material for those journals emphasizing reviews. Moreover, a pattern was noted of establishing individual prices that are almost exactly half the prices required of institutional purchasers. Previous studies have developed mea sures of quality based on rankings of pub lications compared to each other within a given discipline. Most of these compari sons work from information available from the Institute for Scientific Informa tion (ISI). Data acquired from this source showing the impact factor (IMPACT), immediacy index, half-life, total cites (TOTCITES), and cites per year will be tested by using two of these variables to capture quality of journals. These vari ables are expected to be positive with re gard to both price and monopoly power. The prices of journals across disciplines may be driven by different factors. In gen eral, prices are higher in the sciences and technical areas and lower in the humani ties. This is understandable when one considers that essentially no market ex ists for scholarly publications in the hu manities outside academe. In contrast, scientific publications are used heavily in corporate research by pharmaceutical firms and other industries highly depen dent on research. As a result, two addi tional dummies are included in the model to segment the specification along disci pline lines. Two dummy variables (SCI ENCE, SOCSCI) will control for differ ences in price among the science and so cial sciences as compared to the residual category of humanities. These variables are expected to be positive and strong predictors of price. Finally, a dummy variable is included to determine whether the availability of each journal electronically (ELCTRNIC) has a positive impact on ability to price discriminate. Because it has been hypoth esized that monopoly power will erode in the electronic arena, this variable should be a statistically significant and negative predictor of monopoly power. However, to the extent that a journal’s availability in an electronic format distin guishes it from print counterparts, there is some expectation that this variable could be positive. This would capture additional price discrimination by pub lishers that are able to capture lost rev enue in the electronic environment. As indicated in earlier studies, it could be expected that circulation (CIRC) would capture the effects of economies of scale, which those publications distributed in larger quantities will experience. Thus, this variable is expected to be negative. However, it introduces an econometric problem because it is likely to be endog enous to the determination of price (INSTIT) in the first specification and to the measure of monopoly power (LERNER) in the second. In effect, circu lation, which is a measure of demand, drives price and, in turn, changes in price drive circulation. Therefore, the specifi cation here is complicated by the likeli hood that a variable’s inclusion for circu lation implies simultaneous equations. This project attempted to overcome this limitation by considering another version of the specification with a multistage OLS regression. All the variables in the model were approximated with data collected from original sources. Database Development At the outset, it was the intention of this investigation to build the database for the A Tool to Assess Journal Price Discrimination 275 regression by enhancing the data on sub scriptions available from Trinity’s inte grated library system. To augment these statistics, the number of article pages, the number of advertisement pages, and pric ing information had to be obtained from the journals themselves. Additional infor mation such as impact measures, total ci tations, and related factors were acquired from ISI. Some data were obtained from a serials bibliography.12 Table 2 lists the nondummy variables with statistical averages, which are appro priate to the model specification as deter mined by theory and previous efforts by others. The data associated with those variables were obtained most readily by examining the journals and compiling the statistics from the actual items. Because literally millions of journal pages had to be examined, the investigation focused on the subset of the scholarly literature rep resented by the Trinity collection. Trinity gathers the current issues of every peri odical subscription in one location, orga nized according to classification number. Data were gathered by pulling every other current periodical from the display shelves. All titles that were obtained free of charge, were news oriented, or were dependent on advertising revenue were dropped from the data collection in favor of scholarly, academic titles. As might be expected with such a mas sive data-gathering project, several prob lems developed during the process. Ac curate, reliable circulation data proved to be difficult to acquire. Although Ulrich’s directory tries to provide these figures, they sometimes are misleading, missing, and/or not updated every year. Alterna tively, the circulation figures posted by the journals themselves in compliance with postal regulations are unavailable for foreign-based periodicals. Further more, an attempt to retrieve these data from the U.S. Post Office revealed that that agency does nothing to compile the information. Therefore, the figures were gathered from the journals themselves TABLE 2 Non-dummy Regression Variables with Minimum, Maximum, and Mean Values for 1997 Dependent Variables: Minimum Maximum Mean INSTIT = The price for library subscriptions 8.00 2,896.00 LERNER = Monopoly power as represented by 0.00 1,856.00 INSTIT minus LOWPRICE Independent Variables: Minimum Maximum 241.00 108.43 Mean LOWPRICE = Price for individuals 5.00 1,968.00 RISK = Standard deviation of the monthly free-market 0.00 184.70 exchange rate between the currency of the home country of a foreign publisher to the U.S. dollar FREQ = The number of issues per year 1.00 52.00 ARTPGS = Number of pages printed per year 14.00 3,209.00 ADVERT = Number of pages of advertising in journal 0.00 5,104.00 for the year AGE = Current year minus the date the journal first 0.00 182.00 published TOTCITES = Sum of the ISI citation measures 10.00 296,759.00 IMPACT = Index of impact from the ISI citation studies 0.00 37.29 CIRC = Reported number of subscriptions to the journal 249.00 252,573.00 96.93 0.40 5.90 956.38 28.08 42.95 4,330.54 1.30 6,033.23 http:bibliography.12 276 College & Research Libraries May 2001 and compared to the Ulrich’s information. When no figure was supplied in the jour nal, the Ulrich’s estimate was used. When both were available, the investigation re lied on the postal reporting of the jour nals. No efforts to acquire circulation counts from the foreign-based journals were made during this phase of research. Obtaining price information also proved challenging. Prices to libraries are usually indicated within each publication. Therefore, that source was relied on for institutional prices, which were then checked against actual Trinity invoices to verify accuracy. In many cases, the price to individuals was not indicated in the journal itself. In fact, this proved to be the general case for many large foreign-based publishers. The common supposition among librarians that individual sub scriptions to these titles have declined to zero seems to be borne out by answers to queries to the publishers requesting the price for individuals.13 Answers were in consistent and sometimes wildly differ ent from year to year. In other cases, the publisher responded as if it did not know the answer. Moreover, a pattern was noted of establishing individual prices that are almost exactly half the prices re quired of institutional purchasers. In some cases, individual prices were bound up in membership charges. For the purposes of this analysis, the information available was interpreted to parse out the lowest price that an individual would normally pay. Letters to a few societal publishers resolved some difficult ques tions. The issue of shadow prices associ ated with whether some subset of indi vidual subscribers would pay association dues for services not wanted in order to obtain a given periodical was ignored on the assumption that the effect on the model using aggregated statistics would be trivial. Likewise, the opportunity costs associated with substituting access in the library (transportation to the library, searching, photocopying, and so forth) for personal subscriptions also were ignored on the assumption that these would typi cally be borne by subordinates to the re searcher and, in the aggregate, would not substantially affect the outcome.14 Data gathering also sometimes ran into problems determining whether a pub lisher was commercial. In a number of cases, periodicals contained indications of both a commercial and a societal (or uni versity press) involvement. After care fully looking at pricing patterns, it became apparent that the price tended to be driven by commercial issues. In fact, dis cussion at a recent conference of univer sity presses substantiated that they are being pressed by parent institutions to stand on their own and produce a profit.15 Similarly, many societies indicate that their publications produce a return to equity that offsets dues payments by members.16 Therefore, each periodical where a combination of commercial and societal or university press was involved was coded as if it were commercial. Unfortunately, also, ISI does not index all the titles included in the sample gath ered at Trinity. Therefore, quality vari ables based on ISI statistics, such as im pact factor or total citations to each title, could not be included in the analysis for many of the observations. This informa tion was available for approximately 60 percent of the database. Also early on, it was difficult to deter mine reliably when a journal was actu ally available electronically. The journal itself was relied on to indicate whether it had an electronic counterpart. In most cases, this information was available and could be verified by other sources. How ever, in 1995, fewer than fifty titles that were part of the database had electronic counterparts. By 1997, this information could be corroborated and 294 titles for which the publishers were offering elec tronic versions were in the database. Model Results Generally, the results of the OLS regres sion tests of institutional price conformed to both theory and earlier studies. In fact, the sign on every variable that held up in the regressions conformed to expecta tions, with the exception of the variable http:members.16 http:profit.15 http:outcome.14 http:individuals.13 A Tool to Assess Journal Price Discrimination 277 TABLE 3 OLS and Three-stage Analysis of lnstitutional Price for 1997 (97lNSTlT) Dependent variable: 97INSTIT Independent Variables: OLS Model Three-stage Model Term Estimate t Ratio Prob>ltl Estimate t Ratio Prob>ltl Intercept -7.7295 -0.30 0.7679 -53.8460 -1.78 0.0750*** 97LOWPRICE 0.7345 17.69 <0.0001* 2.9445 14.56 <0.0001* EUROPE 96.5960 3.63 0.0003* 96.2500 2.95 0.0030* GRTBRIT 81.2282 5.45 <0.0001* 84.7660 4.53 <0.0001* OTHER 23.1433 0.87 0.3837 26.8190 0.82 0.4100 ASSOC -80.7462 -5.39 <0.0001* -91.0110 -4.80 <0.0000* FOUND -62.5348 -2.22 0.0266** -69.7290 -1.98 0.0490** GOVERN -54.7575 -0.67 0.5022 -56.1010 -0.56 0.5750 UNIV -63.9232 -4.59 <0.0001* -74.9410 -3.96 <0.0000* FREQ 21.9225 14.62 <0.0001* 97ARTPGS 0.0556 11.59 <0.0001* PEER 53.2032 4.81 <0.0001* CCC 3.7960 0.32 0.7463 ARTILLUST 17.7048 1.05 0.2937 97ADVERT -0.4852 -7.94 <0.0001* 97SUBFEE -55.2221 -2.99 0.0029* 97AGE -0.8906 -4.65 <0.0001* BOOK -23.2038 -1.93 0.0546*** SCIENCE 37.0897 1.95 0.0511*** 50.3920 1.78 0.0750*** SOCSCI -17.1302 -1.13 0.2573 -20.3560 -1.11 0.2680 97ELCTRNIC 61.0898 5.51 <0.0001* 57.3430 4.23 <0.0000* * Significant at the 0.01 level ** Significant at the 0.05 level *** Significant at the 0.10 level Summary of Fit RSquare 0.8105 RSquare .05249 RSquare Adj 0.8053 Durbin-Watson 1.8982 Root Mean Square Error 138.4037 Von Neumann ratio 1.9007 Mean of Response 202.0081 Observations (or Sum Wgts) 744 Analysis of Variance Source Model Error C total DF 20 723 743 Sum of Squares 59241126 13849486 73090612 Mean Square 2962056 19156 FRatio 154.63 Prob>F <.0001 278 College & Research Libraries May 2001 TABLE 4 OLS Analysis of Publisher Monopoly Power for 1997 (97LERNER) Dependent variable: 97LERNER Independent Variables: Term Estimate t Ratio Prob>ltl Intercept -27.2726 -1.02 0.3075 EUROPE 95.2394 3.48 0.0005* GRTBRIT 75.2878 4.93 <0.0001* OTHER 20.4782 0.75 0.4529 ASSOC -72.6558 -4.73 <0.0001* FOUND -52.6606 -1.82 0.0685*** GOVERN -47.7689 -0.57 0.5687 UNIV -53.0263 -3.73 0.0002* FREQ 20.3354 13.39 <0.0001* 97ARTPGS 0.0502 10.34 <0.0001* PEER 55.6431 4.90 <0.0001* CCC -0.8208 -0.07 0.9456 ARTILLUST 16.3332 0.94 0.3456 97ADVERT -0.4405 -7.06 <0.0001* 97SUBFEE -56.2801 -2.96 0.0031* 97AGE -0.8804 -4.47 <0.0001* BOOK -12.8829 -1.05 0.2940 SCIENCE 20.1348 1.04 0.2973 SOCSCI -16.2970 -1.05 0.2941 97ELCTRNIC 64.7188 5.69 <0.0001* * Significant at the 0.01 level ** Significant at the 0.05 level *** Significant at the 0.10 level Summary of Fit RSquare 0.6455 RSquare Adj 0.6362 Root Mean Square Error 142.1684 Mean of Response 110.5228 Observations (or Sum Wgts) 744 Analysis of Variance Source DF Sum of Squares Mean Square Model 19 26647512 1402501 Error 724 14633380 20212 C total 743 41280892 FRatio 69.39 Prob>F <.0001 for electronic availability (ELCTRNIC). For a few vari ables determined to be impor tant by earlier studies, the tests were statistically insig nificant. When their removal would have diminished the power of the model to predict by not holding constant for an important factor, the variables were left in the reported re sults. The basic models were run on data for 1995, 1996, and 1997. The results were nearly the same for all three years, with the exception of the vari able for electronic availability (ELCTRNIC). In both the model to predict institutional price and the model to predict monopoly power, the analy ses converged consistently with about a hundred obser vations lost because of null values. The results of the analyses are summarized in tables 3 and 4. Table 3 reports two versions of the results to accommodate a statistical complication introduced by potential endogenity of the price charged individuals (LOWPRICE). To explain these results, some caveats apply. First, as serting that these statistical models prove certain points can be misleading. By anal ogy, a Missouri road map is a model. It would be unreason able to try to use such a map to prove that St. Louis is lo cated at the conjunction of two major rivers on the eastern boundary of the state. How ever, used properly, the map makes highly reliable predic tions for any traveler. Pro vided a traveler in Missouri knows where he is to begin with, the map provides, with A Tool to Assess Journal Price Discrimination 279 a high degree of certainty, the direction to go in and the roads to take to get to St. Louis. Once in St. Louis, the traveler could corroborate the evidence provided by the map that, indeed, the city is located where the Missouri and Mississippi rivers join at the midpoint of the state’s eastern bor der. Similarly, it is better to understand that statistical models such as OLS regressions provide reliable support to make predic tions. In fact, these models can predict with a degree of confidence in excess of 90 percent, which is all that any statisti cal model can do. Therefore, although the results from the tests do not prove per se that certain publishers overprice their products or hold monopoly power, they do provide reliable predictors. In the case of the tests described here, the models predict that if given periodicals produced by certain publishers are examined, their price is likely to be too high compared to all the other periodicals examined in the tests. Furthermore, it can be confidently predicted that the outcome is accurate nine or more times out of ten. Second, the results here apply to a data set of periodicals subscribed to by Trinity University. The predictions they make would apply reasonably well to any school with nearly the same subscription list and perhaps as well to others with similar subscription lists. Although pre dicted prices for specific titles would vary considerably, the general outcomes of these tests could be expected to describe conditions for similar kinds of institutions with similar core collections. The reliabil ity of the predictions would decline if an attempt were made to apply them to sub stantially smaller or larger collections or to collections skewed heavily toward fewer disciplines. In terms of Trinity’s collection of peri odicals, the model predicts the price of academic periodicals with an R-squared statistic in excess of 0.81, indicating that over 80 percent in the variation of the de pendent variable library price (INSTIT) is explained by the model. Similarly, the R-squared statistic on the alternative model examining monopoly power (LERNER) exceeds 0.64 and thus explains over 64 percent of the variation in the variable for monopoly power. Both mod els have very large F-statistics; thus, both are statistically significant at better than the 0.01 level. That is, they predict cor rectly in excess of 99 percent of the time. The outcomes of the models in terms of specific independent variables also are very reliable. Nearly every independent variable conformed to theory, corrobo rated previously reported analyses, showed the expected sign, and was sta tistically significant. A few variables in the original function specified failed to be sta tistically significant. The variable separat ing the broad discipline of social science (SOCSCI) and those separating certain other factors (OTHER, FOUND, and CCC) were left in the model to control for important concerns that could contribute to price or monopoly power and to com plete the relevant dummy sets. The sub set of social science periodicals repre sented by a dummy variable (SOCSCI) was retained to distinguish those titles from science and the residual titles, which were all humanities. The variable for countries outside the United States, Great Britain, and Europe (OTHER) was left in to distinguish publications published outside this country from those published within. The costs associated with regis tering with the Copyright Clearance Cen ter (CCC) that could affect prices was left in to provide assurance that the study was controlling for this cost of production. The extra production costs associated with il lustrations (ARTILLUS) were left in to control for this issue. A few variables that were shown to be statistically significant in other studies were specified in the original function here but were dropped from the results when they proved to be very insignificant and when they were not associated with other dummy variables. Following the work of others studying revenue lost to photocopying, this study incorporated data from ISI on total citations (TOTCITES) and impact factor (IMPACT) as proxies for quality variations 280 College & Research Libraries May 2001 among titles.17 Neither of these variables held up in the final results, in part because their values could not be obtained for much more than about half the collection. Cita tion-based variables may never produce significant results because of the cross-dis ciplinary nature of this database. Citation rates, in general, vary considerably from the arts compared to the sciences. In an earlier study on economics titles, the exchange rate risk faced by foreign publishers with the decline of the dollar in world markets proved reliable as a pre dictor of price. In this study, this variable (RISK), as proxied by the standard devia tion of the annual exchange rate for the currency of each country with the dollar, was insignificant. It is likely that this vari able did not hold up in the model because the exchange rate of dollars for the for eign countries represented was reason ably steady within the 1995 to 1997 time frame of the analysis. Therefore, exchange rate risk may have played a minor role in setting prices for those years. Finally, the variable used to control for the circulation rates (subscriptions sold by the publishers) was not usable in the model. There is a statistical complication associated with this variable (CIRC) re lated to endogenity and with the variable for the price charged to individuals (LOWPRICE), which is closely related to circulation rates. Also, reliable circulation figures were available for too few obser vations to produce meaningful results using that variable. On the other hand, the price charged to individuals turned out to be statistically significant and posi tive. As this variable (LOWPRICE) is in creased, the model predicts and experi ence confirms a positive increase in the price charged to libraries (INSTIT). This is as expected because the loss of revenue caused by individual subscription cancel lations will be reflected in library prices as publishers attempt to make it up. How ever, this variable (LOWPRICE) may in troduce the same statistical problem that the statistic for circulation does. (The problem introduced by both variables is discussed in the next section.) All the other variables expected to con tribute to predictions of price and mo nopoly power held according to expected sign and statistical significance except for the dummy variable for electronic avail ability. Some variables included in the study that did not explicitly contribute to costs of production nevertheless contrib uted to variations in the dependent vari able. Science periodicals (SCIENCE) in the collection are more expensive compared to periodicals in social science and the humanities. Publications originating in Western Europe (EUROPE) and Great Britain (GRTBRIT) are priced higher, on average, than U.S. periodicals. Publica tions of associations and societies (ASSOC), university presses (UNIV), gov ernment publications (GOVERN), and foundations (FOUND) are all priced lower than commercial publications. Other variables that control for factors that could contribute to production costs were retained in the models because of their statistical significance. The costs of peer review (PEER), higher frequency of publication (FREQ), and higher number of pages of article content published (ARTPAGES) contribute positively to higher prices charged to libraries. Those higher prices are thus partially explained by higher production costs. Alternative sources of revenue from page charges or submission fees (SUBFEE) and advertis ing (ADVERT) help hold prices down, as expected. The models show that what li brarians believe to be predictable are in deed predictable with a high degree of certainty within the data set examined. One variable was used to extend this study beyond those reported earlier to determine the potential impact of elec tronic availability (ELECTRNIC). If bar riers to entry would be lowered by the opportunity to publish offered by elec tronic technology, this variable would be negative. Alternatively, if the electronic domain offered an opportunity for pub lishers to retain better control of their publications and possibly extend their pricing power, it would be positive. In the regression run on 1997 data where there http:titles.17 A Tool to Assess Journal Price Discrimination 281 was a large subset of electronically avail able periodicals, the results on this vari able were statistically significant and positive. As the primary outcome of the analy sis, it is possible to reject the null hypoth esis that monopoly power will not in crease in the electronic domain. Actually, the analysis indicates that the opportu nity to publish electronically neither low ers barriers to entry nor erodes the mo nopoly power of publishers. The dummy variable reflecting the availability of elec tronic versions of titles (ELECTRNIC) was statistically significant and positive in both specifications. Therefore, librarians can likely expect to see prices continue to increase and monopoly power extended as publishers introduce electronic ver sions of their products. Many variations of the model were submitted to OLS regression analysis to provide some assurance that as few con tributing factors as possible would be overlooked. In fact, nearly every issue that could be assessed quantitatively and in cluded was. There are two related factors for which a proxy could not be developed: economies of scope and economies of scale. Presumably, those publishers pub lishing greater numbers of journal titles could take advantage of sharing editorial and other production activities across journals by assigning employee slack time from one title to another. Economies of scope associated with this sharing should reduce production costs and, thus, price. In fact, some of those publishers, particularly Western European commer cial houses with large numbers of publi cations, actually appear to price higher.18 Related to this, there should be economies of scale associated with titles produced in large quantities for each issue. When the number of copies printed for a given issue is large, the fixed costs of production represented by factors such as editorial labor or building facilities are spread across more sales. Therefore, economies of scale come into play and prices might realistically be expected to be lower. Unfortunately, circulation fig ures (CIRC) capture the number of cop ies printed but also are linked closely to the price of and demand for given publi cations. Dealing with this issue in the re gression introduced complications. Similarly, there is a linkage between the price charged to individuals (LOWPRICE) and the price charged to libraries (INSTI). It expresses itself through demand reac tions. As the price to individuals increases, subscribers respond by canceling subscrip tions, producing a consequent loss of rev enue to publishers and an increase in de mand for access to library subscriptions. Publishers’ attempts to recover lost rev enue show up in higher prices charged to libraries, which also is a reaction to higher demand for library access by those indi viduals who still need the content but have cancelled personal subscriptions. As a fur ther consequence, total circulation de creases. The mathematical complications introduced into the study model by these interactions are discussed further below. The Circulation Issue Attempts to include a variable for circula tion figures, or in other terms the number of subscriptions sold, introduce problems in the regression analysis. Circulation fig ures reported to the U.S. Post Office, by law, are recorded within each publication at least once a year. Those reports indicate the total print run along with the number of copies sold, retained in inventory, and given away of the average issue for the given year. Circulation figures also are re ported by Ulrich’s Periodicals Directory. A comparison of these sources indicated that Ulrich’s numbers usually were close to the reported counts of subscriptions sold. Therefore, a circulation estimate was com piled using the reported numbers, when available, and Ulrich’s numbers when the journal did not report. The estimated cir culation was included as a variable in sev eral runs of the model. In most cases, the function converged with one-half of the observations lost due to null values. The variable for circulation (CIRC) carried a negative sign but did not hold up as sta tistically significant. http:higher.18 282 College & Research Libraries May 2001 The negative sign on the circulation variable corroborates the argument that there is an inverse relationship between circulation and price. That is, price in creases produce a response from subscrib ers, which results in cancellations. How ever, this relationship may not be mod eled meaningfully for several reasons. Previous work has shown that examina tions of this relationship across a broad range of titles produce scatter plots rather than systematic functions.19 In part, this is likely due to the aggregated nature of the circulation figures reported. For any given journal, neither the internal reports nor those given by Ulrich’s distinguish quantities sold to individuals from those sold to libraries. Because the prices of fered to these two groups of subscribers differ considerably, the resulting price- versus-demand relationship is unclear. It can be argued that the prices to indi viduals (LOWPRICE) and to libraries (INSTIT) are determined simultaneously and thus there is a danger of biased re gression coefficients. To address this con cern, the model displayed in the first col umn of table 3 was reestimated as a two- equation model. LOWPRICE was mod eled as a function of FREQ, ARTPGS, PEER, CCC, ARTILLUS, ADVERT, SUBFEE, AGE, and BOOK. Each of these variables is hypothesized to affect the net cost of producing a journal. The institu tional price (INSTIT) is then modeled as a function of the predicted values of LOWPRICE as well as EUROPE, GRTBRT, OTHER, ASSOC, FOUND, GOVERN, UNIV, SCIENCE, SOCSCI, and ELCTRNIC. A three-stage OLS procedure generated the results shown in the sec ond column of table 3. All of the coefficients retain their signs and are within a standard error of their OLS counterparts. The p-values also are roughly equivalent. The first-stage results are not presented here but continue to confirm the hypotheses concerning costs presented above. LOWPRICE is posi tively related to FREQ, ARTPGS, PEER, and CCC. ADVERT, SUBFEE, AGE, and BOOK all tend to reduce LOWPRICE. As was the case in the OLS results, ARTILLUS was positive, but statistically insignificant. Demand for individual and library subscriptions is more likely determined by the overall size of the market for re search in each given discipline. With some obvious exceptions, most journals serve a restricted audience. It is unlikely that any given journal in the humanities would be widely read by scholars work ing in the sciences. Each journal contains unique information not found in any other journal. Therefore, titles tend not to be in direct competition with each other. Also, when a scholarly discipline is stud ied by a relatively small contingent of scholars, the number of subscriptions is likely to be smaller in total. Overall, these factors suggest that the demand-versus price relationship has to be examined on a title-by-title basis or through an alter native model of some kind. The alternative model also must over come an endogenous relationship of cir culation in the model specified for this study. Placing a variable for circulation in the model introduces a classic demand and supply in the same equation prob lem.20 In effect, when an estimate for cir culation is included, the simpler model attempts to solve simultaneous equations. This is further complicated by the differ ence in demand by individuals and insti tutions. Therefore, there really are four simultaneous equations: demand and supply for individual subscriptions, and demand and supply for institutional sub scriptions. Solving four simultaneous equations implies a need for data on both subscription prices and both estimates of circulation. Because estimates of indi vidual-versus-institutional subscriptions could not be obtained, it was impossible to specify a reliable function to explain the relationship statistically. The complications introduced by the variables requiring a specification to solve four simultaneous equations take this analysis beyond the scope of this project. However, this complication leaves the proposed hypothesis test largely unaf http:functions.19 A Tool to Assess Journal Price Discrimination 283 fected. The specified model accommo dates every variable previously estab lished as exogenous and relevant except for a proxy of economies of scope and a metric of quality. In the first case, econo mies of scope should yield lower prices from those publishers producing a very large number of titles. Yet, the model dem onstrates a correlation between high prices and publishers of numerous titles. In the second case, a metric of quality can not be established for a cross-discipline data set. For example, it is meaningless to compare the academic quality of a music journal to one in physics or in busi ness. Furthermore, certain patterns showed up in the data. These may lead to addi tional fruitful analysis, which could ex tend the analysis reported here. Some titles in the data set are so high priced that they effectively have no market except to institutions. Where it was impossible to obtain the price charged to individuals, it was speculated that this was because the publisher was not selling to individu als. As this was examined, a pattern sug gested itself. These titles appeared to be very high priced, had low circulation, and often were described by users as second tier in quality. Other Statistical Considerations The analyses reported here are based on statistical procedures applied using the SAS Institute’s JMP™ software, which was developed to run on a PC. This par ticular version was chosen because, among other advantages, it automatically applies several standard statistical tests to confront the possibility of misleading results produced by problems typical of OLS regression analyses, such as hetereoskedasticy and collinearity. More over, the software plots the effects of each variable in a way that makes spotting sta tistical problems easy. As the various at tempts were run, which ultimately pro duced the results reported, a few variables introduced problems. Only one collinear ity issue emerged, as explained above, and the few variables that introduced ex ceptional leverage were excluded from the analysis. Fortunately, the only poten tially useful added variable, which was based on ISI quality measures, where the problems could not be corrected was eliminated for other reasons noted above. However, the specified model does not explain why prices change from year to year, especially at an inflation rate exceed ing the CPI. This project attempted to deal with this question by creating another specification to explain the price change from 1995 to 1997 based on the change in the variables over that same time frame. Delta values were calculated by subtract ing the 1995 values from the 1997 values for all of the variables listed in table 1 where they were meaningfully possible. Most of the dummy variables did not change because they were based on loca tion or other essentially unchangeable is sues. The dummy for electronic availabil ity was subject to change, so it was re tained along with the delta values. Table 5 displays the results of an OLS analysis of the change in institution price (L INSTIT) on the independent delta val ues. When this model was run using only the delta values, it converged with a sta tistically significant lack of fit. Because the lack of fit indicates that statistically sig nificant variables were missing from that model, the regression was run on the same function as described in table 3 but substituted delta values for those vari ables that changed over time. Two reports on this effort are included in table 5. As might be expected, the variables that predict the price to institutions in the basic model also tend to predict changes in price over time. In fact, the signs on the variables are the same and the statis tical significance of all but the one dummy variable for foundation publications (FOUND) remains. However, this regres sion continues to lack some variables that could affect price. As noted before, there is no reliable metric of quality across the data set and trying to provide one is en cumbered by inconsistent definitions of quality among a broad set of disciplines. In addition, it is impossible to control for 284 College & Research Libraries May 2001 TABLE 5 OLS Analysis of the Change in Institutional Price from 1995 to 1997 ( lNSTlT) Dependent variable: LINSTIT Independent Variables: Term Estimate t Ratio Prob>ltl Estimate t Ratio Prob>ltl Intercept -3.5292 -0.34 0.7310 -2.198156 -0.22 0.8246 LLOWPRICE 0.0761 3.33 0.0009* EUROPE 44.5040 3.87 0.0001* 54.7353 4.99 <0.0001* GRTBRIT 28.7676 4.84 <0.0001* 32.7037 5.10 <0.0001* OTHER 13.4242 1.25 0.2131 14.4774 1.17 0.2423 ASSOC 28.4983 4.82 <0.0001* 35.0194 5.36 <0.0001* FOUND 18.4451 1.63 0.1030 23.5840 1.82 0.0694 GOVERN 15.9448 0.50 0.6157 22.4261 0.60 0.5462 UNIV 17.7830 3.24 0.0013* 22.4015 3.60 0.0003* FREQ 6.9292 15.38 <0.0001* 7.8379 16.76 <0.0001* 97ARTPGS 0.0361 4.66 <0.0001* 0.0245 2.94 0.0033* PEER 16.0182 3.65 0.0003* 19.7601 4.06 <0.0001* CCC 2.2858 0.49 0.6234 ARTILLUST 4.2592 0.64 0.5212 LADVERT 0.4191 3.75 0.0002* 0.3635 2.80 0.0053* LSUBFEE 11.2972 0.74 0.4616 31.6983 1.96 0.0498** 97AGE 0.2830 3.75 0.0002* 0.3050 3.57 0.0004* BOOK 11.6550 2.47 0.0139** 13.6738 2.59 0.0097* SCIENCE 21.0204 2.79 0.0054* 33.6816 4.17 <0.0001* SOCSCI 0.1130 0.02 0.9849 0.2735 0.04 0.9672 LELCTRNIC 18.04703 3.9 0.0001* 27.5325 5.33 <0.0001* * Significant at the 0.01 level ** Significant at the 0.05 level *** Significant at the 0.10 level Summary of Fit RSquare RSquare Adj Root Mean Square Error Mean of Response Observations (or Sum Wgts) 0.4851 0.4701 53.8845 33.8169 710 0.5027 0.4914 63.2643 41.0601 765 Analysis of Variance Source DF Sum of Squares Model 20 1884567 Error 689 2000535 C total 709 3885102 Mean Square 94228.4 2903.5 DF Sum of Squares 17 3021962 747 2989770 764 6011731 Mean Square 177762 4002 FRatio 32.453 Prob>F <.0001 44.4143 Prob>F <0.0001 A Tool to Assess Journal Price Discrimination 285 economies of scale in this model, but if a variable were available, it would likely carry a negative sign. Therefore, in addi tion to these and the variables included in the model, it could be assumed that at least one other significant variable is miss ing. At this point, because quality and scale would not change dramatically in three years, it was further assumed that the missing variable would control for vagaries in publisher pricing. Isolating Specific Publisher Pricing The model specified and described above provided results that are not surprising to librarians. While holding constant for as many factors as possible that have been claimed to drive price increases by pub lishers, the model indicates that some publishers overprice their journals com pared to other titles. In fact, the statistical tool used in the analysis contains the means to look at every individual title compared to the whole data set to see which titles the model predicts will be the most overpriced. Table 6 isolates those titles priced at amounts statistically sig nificantly higher than the model predicts for them. This table holds the key to a very powerful tool that librarians can use in selecting journals to add to or deselect from their collections. Table 6 isolated the twenty titles for which the prices charged by the publish ers are statistically significantly higher than the model predicts for them within the data set. The software used for the analysis provides the means to save re siduals on every observation. Some of those residuals are statistically significant. Based on the model reported in tables 3 and 4, the investigation isolated the most overpriced journals using the residuals and sorted for those that were statistically significant. Therefore, this isolated sub set of the data set exposes the titles that are the best candidates for cancellation based on predicted subscription price for the most recent complete year. To use the model as a selection tool, Trinity would have to expand the data set to cover all subscriptions and enter the relevant data for every title. Although this undertaking may sound somewhat daunting, it does not add substantially to the record keeping already routine for most libraries. With the caveat that the information has to be entered into an au tomated record-keeping system, the stan dard bibliographic and check-in records already retained include nearly all the quantitative variables important to the specified model. Most of the dummy vari ables can be recorded once and then used indefinitely. The most difficult variable to obtain, and one that must wait for the completion of the subscription year to obtain accurately, is number of article pages. Although determining the number of pages committed to advertising and ar ticles was essential to the experiment con ducted here, it is not necessary for a slightly different model with equal pre dictive power. During the analyses, a dummy variable was submitted for whether each journal takes paid adver tisements and a quantitative variable for total pages to be used in lieu of number of article pages. Without changing any thing else, the regressions using these al ternatives yielded results closely approxi mating those reported in tables 3 and 4, based on the original specification. In fact, saving residuals on results iso lated all twenty of the titles listed in table 6. Two additional titles identified by the simpler model emerged as a result of slightly different residuals. For a library to use this tool, some simple additions need to be made to augment standard li brary record keeping for subscriptions. Those records need only be submitted to the model specified to isolate candidates for cancellation based on the extent of the difference between the price charged and the price predicted by the model. Table 6 makes it clear that aggressive price escalation may not be limited to com mercial publishers. Emerging in this table are five societal publishers that priced spe cific journals significantly in excess of the model predictions. Indeed, pressure from membership likely encourages societal TABLE 6 Comparison of Actual Institutional Price to Model-predicted Price, Ranked by Percent Difference Title Publisher Institutional Price in 1997 Model- Predicted Price for 1997 Excess of Actual Price over Predicted Price Actual Price As a Percent of Predicted Price Journal of Econometrics Review of Scientific Instruments Journal of Cell Science Geomorphology Veterinary Immunology and Immunopathology Developmental and Comparative Immunology Organometallics Personality and Individual Differences Vision Research Computers and Chemical Engineering Earth Science reviews Communications in Algebra Geophysical Journal International Cognition Journal of Molecular Spectroscopy Inorganic Chemistry Journal of Mathematical Analysis and Application Journal of Algebra International Journal of Energy Research Journal of Physical Chemistry A* Elsevier Science $1,798 American Institute of Physics $1,030 Company of Biologists, Ltd $1,195 Elsevier Science $1,185 Elsevier Science $1,463 Elsevier Science $876 American Chemical Society $1,340 Elsevier Science $1,029 Elsevier Science $1,895 Elsevier Science $1,403 Elsevier Science $648 Marcel Dekker Journals $1,975 Blackwell $1,043 Elsevier Science $983 Academic Press $1,603 American Chemical Society $1,395 Academic Press $2,725 Academic Press $2,475 Wiley and Sons $1,795 American Chemical Society $1,955 $609 $429 $537 $562 $713 $444 $694 $576 $1,066 $809 $379 $1,253 $677 $654 $1,124 $1,020 $2,094 $2,017 $1,496 $1,673 $1,189 $601 $658 $623 $750 $432 $646 $453 $829 $594 $269 $722 $366 $329 $479 $375 $631 $458 $299 $282 295� 240� 223� 211� 205� 198� 193� 179� 178� 173� 171� 158� 154� 150� 143� 137� 130� 123� 120� 117� *Indicated titles originated by a society or association that may still be involved with the publication 286 C ollege & R esearch L ib raries M ay 2001 A Tool to Assess Journal Price Discrimination 287 leadership to offset the extraction of rev enue from members’ dues by making a profit on scholarly publications. Conclusion This experiment yielded some very worthwhile results. First, the analysis pro vided statistically sound evidence that li brarians and scholars should place little hope in the expectation that the electronic era will readily introduce the kind of change needed to diminish publisher mo nopoly power. European-based commer cial publishers with a broad stable of titles are better positioned to introduce elec tronic counterparts to their print publi cations than scholars are to introduce new and competitive online titles. The former are aided by a well-funded infrastructure and an established editorial process; the latter must effectively start from scratch. Although not impossible, the commercial publishers have an edge.21 Second, it appears that publishers who blame exchange rate risk and production costs for the prices of their products may be using empty rhetoric. Although the study did not prove that a weakening dol lar could not be blamed for price in creases, it did show, within the short time frame covered, that risk is not a reliable predictor of price. In short, some price increases occurred despite the condition of the dollar. Third, the experiment offered a pow erful statistical tool for librarians to use to isolate the most egregious pricing as a selection criterion. Nearly all libraries base their selection of titles to acquire on a combination of three pieces of informa tion: price, quality, and potential level of use. Experience, intuition, and guesswork play major roles in deciding both esti mates. The model described here offers an opportunity to reduce the guesswork implicit in the price analysis. Notes 1. This article was excerpted from the author ’s report, Beating Publisher Price Discrimination (San Antonio, Tex., 2000). For a complete report on the project, address queries to the author. 2. See: Association of Research Libraries. Directory of Electronic Journals, Newsletters, and Aca demic Discussion Lists (Washington, D.C.: ARL, 1997). For excellent overviews on the develop ment of e-journals, see: Judy Luther, “Full-text Journal Subscriptions: An Evolutionary Process,” Against the Grain 9/3 (June 1997): 18, 20, 22, 24; Liza Chan, “Electronic Journals and Academic Libraries,” Library Hi Tech 17/1 (Jan. 1999): 10–16. 3. For a thorough overview of the economic framework, see: Richard W. Meyer, “Monopoly Power and Electronic Journals,” Library Quarterly 67/4 (Oct. 1997): 325–49. 4. See, especially: H. Craig Peterson, “Variations in Journal Prices: A Statistical Analysis,” Serials Librarian 17/1&2 (1989): 1–9; ———, “The Economics of Economics Journals: A Statistical Analysis of Pricing Practices by Publishers,” College and Research Libraries 53 (Mar. 1992): 176–81; George A. Chressanthis and June D. Chressanthis, “The Determinants of Library Subscription Prices of the Top-ranked Economics Journals: An Econometric Analysis,” Journal of Economic Education 25/4 (fall 1994): 367–82. 5. Roger Noll and W. Edward Steinmueller, “An Economic Analysis of Scientific Journal Prices: Preliminary Results,” Serials Review 18 (spring/summer 1992): 32–37. 6. George A. Chressanthis and June D. Chressanthis, “Publisher Monopoly Power and Third- degree Price Discrimination of Scholarly Journals,” Technical Services Quarterly 11/2 (1993): 13– 36. 7. Edward Chamberlin, The Theory of Monopolistic Competition (Cambridge, Mass.: Harvard Univ. Pr., 1935). 8. George A. Chressanthis and June D. Chressanthis, “A General Econometric Model of the Determinants of Library Subscription Prices of Scholarly Journals: The Role of Exchange Rate Risk and Other Factors,” Library Quarterly 64/3 (1994): 270–93. 9. This theory is based on the classic work: Abba Lerner, “The Concept of Monopoly and the Measurement of Monopoly Power,” Review of Economic Studies (June 1934): 157–75. 10. Two variations in the dependent variable were attempted, including the index of mo nopoly power based on the work of Lerner. This second version of the Lerner index provided no additional predictive power to the model. 288 College & Research Libraries May 2001 11. Chressanthis and Chressanthis, “Publisher Monopoly Power and Third-degree Price Dis crimination of Scholarly Journals.” 12. Ulrich’s Periodicals Directory (New York: R. R. Bowker, 1996–1998). 13. For a very good summary of trends that relate to this issue as well as general pricing, see: Carol Tenopir and Donald W. King, “Trends in Scientific Scholarly Publishing in the United States,” Journal of Scholarly Publishing 28/3 (Apr. 1997): 135–70. 14. Assumptions regarding these two issues were based on an earlier study of scholar de mands for journal content in: Charles River Associates, Development of a Model of the Demand for Scientific and Technical Information Services (Boston: Charles River Associates, 1979). 15. Association of American University Presses, annual meeting, Austin, Tex., June 20–12, 1999. 16. See, for example: Michael Barr, “ Where Does the Money Go?” Newsletter on Serials Pricing Issues 229 (July 13, 1999): 229.1. 17. For an explanation of price discrimination based on publisher effort to recover revenues lost to photocopying, see: S. J. Liebowitz, “Copying and Indirect Appropriability: Photocopying of Journals,” Journal of Political Economy 93/5 (1985): 945–57. 18. Mark J. McCabe, “Academic Journal Pricing and Market Power: A Portfolio Approach.” Revised Nov., 2000. Presented at the 2000 AEA meetings in Boston (under review at the AER). Available online at http://www.prism.gatech.edu/%7Emm284/JournPub.PDF. 19. Tenopir and Donald W. King, “Trends in Scientific Scholarly Publishing in the United States,” Journal of Scholarly Publishing 28/3 (Apr. 1997): 135–70. 20. Bruce Kingma, personal correspondence with the author. 21. Programs such as the SPARC initiative and discipline servers such as that at Los Alamos probably offer more viable opportunities for new approaches to compete with the commercial establishment. http://www.prism.gatech.edu/%7Emm284/JournPub.PDF