Electronic academic libraries services valuation: a case study of the Portuguese electronic scientific information consortium b-on

Electronic academic libraries

services valuation:

a case study of the Portuguese electronic

scientific information consortium b-on

Luiza Baptista Melo1 and Cesaltina Pires2

1CIDEHUS-UE - University of Évora, Portugal

e-mail: lbmelo@fc.up.pt

2CEFAGE-UE and Management Department - University of Évora, Portugal.

e-mail: cpires@uevora.pt

Abstract: This paper investigates the factors that influence the value for the users of the Portuguese electronic scientific information consortium b-on (Biblioteca do Conhecimento Online). In order to be able to estimate this value in monetary terms we used the contingent valuation method based on a willingness to pay scenario. Data was collected through an e-survey sent to the whole Portuguese academic users.

The main aims of this study are: (i) to investigate how the Portuguese academic community values b-on; (ii) to investigate the set of factors that determine whether the user is willing to pay the b-on services which allows us to estimate the demand function of b-on services. In order to achieve these objectives we use several regression analysis techniques – linear probability model (LPM), Logit and Probit models.

The results show that the demand for b-on services is quite sensitive to the «price», frequency of use, whether the user knew previously b-on or not, the type of the user and the scientific area of the user.

Keywords: Academic libraries, Electronic sources, Impact evaluation, Logit, Probit

1. Introduction

At present, the world is in a global economic crisis, in all developed countries there are several kinds of economic problems. Most public institutions are under increasing pressure on your own budgets. Portuguese researcher and academic institutions are no exception and it is urgent to study the problem of the academic libraries, the cost and the benefits of its services.

Nowadays the Portuguese electronic scientific information consortium b-on – Biblioteca do Conhecimento Online provides unlimited access to research and academic institutions to the full texts of more than 16,750 scientific publications, via Internet, at the national level (FCCN, 2009).

The Portuguese Government has been investing to improve the access to the production of knowledge so as to develop the country. It is important to know the return on the investments in university libraries. The purpose of this study is to identify the impact of the electronic sources in the Portuguese academic libraries.

This paper explores some issues, they are the following: (i) to investigate how the Portuguese academic community values b-on; (ii) to investigate the set of factors that determine whether the user is willing to pay the b-on services which allows us to estimate the demand function of b-on services. We consider the following factors: the price, the frequency of use, whether the user knew previously b-on or not, the type of the user, the scientific area of the user, and the institution of the use. In order to investigate these problems we use regression analysis techniques - Linear probability model (LPM), Logit and Probit models.The remainder of this paper is organized as follows: in Section 2 we present the methodology used. Section 3 presents and discusses the results. The final section concludes the paper.

2. Methodology

Murgai and Ahmadi (2007) refer the studies of Tenopir, from 1991 to 2000, and argue that “her findings indicate that libraries have adopted digital information sources and services at an accelerated rate due to availability of internet, in particular the World Wide Web.” The digital resources have changed the services of the academic libraries. These authors emphasise the need to keep and report statistics for accreditation and comparison purpose still remains.

According to Noonan (2003) “contingent valuation methodology (CVM) has been increasingly applied to cultural resources” and thus it is a natural candidate to estimate the monetary value of library services. In fact, in the last decade, some authors have used the contingent valuation method for monetary valuations of libraries’ services (Holt, G.E.; Elliott, D. and Moore, A., 1999; McDermott Miller, 2002; Morris, A.; Sumsion, J. and Hawkins, M., 2002; Holt, G.E. and Elliott, D.S., 2003; British Library, 2004; Barron et al., 2005; Aabø, 2005a; Morris; Ayre and Jones, 2006; Elliott, D. S. et al., 2007: Hider, 2008) e.g. of public libraries, (Harless and Frank, 1999) academic libraries, (Chung, 2007) and special libraries.

In the last decade, some of the libraries evaluation has applied regression models for demonstrating the impact of the academic libraries services. Logistic regression analysis has been used to study some libraries issues, for instances: to predict the relevance of a library catalogue search in University of Californias’ Melvyl online catalogue (Cooper and Chen, 2001); to reflect the relationship between library collections and the prestige of universities (Liu, 2003); to study users’ emotional and material satisfaction at micro/macro levels in an academic library (Yu, 2006); to predict the number of patrons that seek assistance at the reference desk of a library from the University of Tennessee (Murgai, Ahmadi, 2007) and to study the use and non-use of public libraries in the information age (Sin and Kim, 2008). We emphasise the studies of Aabø (2005b) in a research of valuing the benefits of the Norway public libraries, based in a contingent valuation, studied how citizens value these services and presented several measures of average valuation including estimations with the standard logistic distribution function.

In a previous research (Melo and Pires, 2009) we estimated the economic value in monetary terms for end-users of the services provided by the electronic scientific information consortium b-on. In order to estimate the monetary value for the end-users of the services provided by b-on we use two alternative valuation methods. On the one hand we estimate the value of the time saved by using this resource. On the other hand, we use the contingent valuation method (CVM) to estimate how much the user is willing to pay for the service. The data was collected through study an e-survey of Portuguese academic community. In this study we are going on our investigation with the same data.

In our study we would like to explain patterns behaviours based on regression models. We want to study the willingness to pay (WTP) of the b-on users as a function of the frequency of use, whether the user knew previously b-on or not, the type of the user, the scientific area of the user, and the institution of the user. Hence, the dependent variable, the willingness to pay, can take only two values: 1 if the user wants to pay some per month to continue having access to the services of b-on and 0 if the user does not want to pay to continue having access to the services of b-on. A feature of this dependent variable is the type that elicits a “yes” or “no” response. In this case the dependent variable is dichotomous. There are three most commonly used approaches to involve dichotomous response variables: Linear probability model (LPM); Logit model and Probit model.

2.1. Regression models for the probability of paying to access the services of b-on

Linear Probability Model (LPM)

We can equate Linear probability model (LPM) considering the following (Wooldridge, 518:2007):

(1)

Where:

The dependent variable y is dichotomous (y =1 if the user want to pay something to access to the b-on service and y =0 if the user does not want to pay to access to the b-on service);

The independent variables are , such as; the frequency of use, whether the user knew previously b-on or not, the type of the user, the scientific area of the user, and the institution of the user; and,

The function G must lie between 0 and 1, i.e. 0<G(z)<1 and .

Gujarati (576:1995) affirms that Linear probability model (LPM) is simple and it has several limitations, for instances, this model assumes that the conditional probabilities increase linearly with the values of the explanatory variables, for instances, the probabilities will tend to tapper off as the values of the explanatory variables increase or decrease indefinitely. This author says that “in group data, Logit and Probit estimates are fairly straightforward.”

Logit model

Wooldridge (518:2007) explains that the standard logistic distribution function G is the following:

(2)

Where G is distribution function, when is, for example, the value that a person places on b-on (Portuguese electronic scientific information consortium), might contain various individual characteristics, such as, frequency of use, whether the user knew previously b-on or not, the type of the user, the scientific area of the user, and the institution of the user, are parameter estimates to be calculated. The logistic regression is a useful way of describing the relationship between one or more outcomes expressed as a probability, that has only two possible values such as (yes=1 and no=0).

In the Logit model the magnitudes of the regressors on the dependent variable are difficulty interpretations and they are not useful (comparing to the linear probability model). Many researchers prefer to estimate the effect of on the probability of success of . To compute this effect we must obtain the partial derivate (Wooldridge, 2003).

Li and Mahendra (2009) explain that the “marginal effects measure the expected instantaneous change in the dependent variable as a function of a change in a certain explanatory variable while keeping all the other covariates constant.”

To find the marginal effects of roughly variables on the response probability we must calculate the partial derivate (for more details see Wooldridge (556:2003)), for instance:

(3)

Probit model

Wooldridge (519:2007) refers that in the Probit model the standard normal cumulative distribution function G is:

(4)

Where (z) is the standard normal density:

(5)

And 0< G(z)<1, when G(z) 0 as and G(z) 1 as z .

As we mentioned above, for the Logit model, it is similar for this model - the magnitudes of the regressors on the dependent variable are difficulty interpretations.

The marginal effects of roughly variables on the response probability is the partial derivate:

(6)

For large samples the coefficients are similar when we use models Logit and Probit to estimate them. The Probit model is specific for a binary response, for example problems that have only two possible outcomes which we will denote as yes=1 and no=0.

2.2 Data Collection

The data was collected through an electronic questionnaire. The questionnaire was based on the International Standard performance indicators and the contingent valuation method (CVM) to assess academic library electronic services. The questions were based on the International Standards ISO 11620:1998, 1:2003 Amendment, ISO 2789:2006 performance indicators: percentage of the target population using traditional library, percentage of the target population using digital library, percentage of the target population using both libraries, preferred location of use of the electronic services, service used (data bases, electronic collections, pay e-journals or Open Access journals). The scenario designed in this research to valuate electronic services of the Portuguese academic libraries is based on a hypothetical idea. The WTP scenario describes an economic situation which forces the Portuguese electronic scientific information consortium, b-on, to stop. It is suggested that the consortium b-on will continue if the users maintain the cost of these services paying a monthly tax with values between 5 to 50 Euros (Melo and Pires, 2009).

This questionnaire (paper version) was piloted with fifty academic library users. Then it was send by e-mail to thirty three Public Portuguese Universities (professors, researchers, students, administrative staff and everyone who usually uses academic services). The answers were received from 15th January to 15th May 2009. During these four months we received 1930 answers. The composition of the respondents sample is the following: professors 28.0%, PhD students/researchers 13.9%, master students 19.3%, undergraduate students 31.1% and others 7.7% (administrative and library staff). We can verify that the professors and the PhD students/researchers are overrepresented in the sample whereas the undergraduate students are underrepresented. This is an expected result. These two groups of users are the ones who are more directly involved in research and who are more aware about the scientific information consortium b-on and thus it is likely that they were more interested in answering.

2.3 Data Analysis

The first step of the analysis was to summarize the data collected. The next step involved the comparison between different types of users and between various scientific disciplines. We considered five groups of users: Professor, PhD Student/ Researcher, Master Students, Undergraduate Students and Others. The scientific disciplines were aggregated in the following six groups: Physiques and Chemistry Sciences; Humanities and Social Sciences; Earth and Planetary Sciences; Life and Health Sciences, Engineering; Mathematics and Computer Sciences.

As we referred above in order to estimate same coefficients to describe several user behaviours we use statistic and probabilistic data analysis - Linear probability model (LPM), Logit and Probit models for binary response. Our statistical and probability analysis was performed with the STATA – Data Analysis Statistic Software.

3. Analysis of the Results

We start by analyzing the results of user e-surveys. The user e-surveys involved 1930 answers. Table 1 presents the frequency of maximum willingness to pay (WTP) to continue to access Portuguese electronic scientific information consortium b-on (Melo and Pires, 2009). The two F statistics and the corresponding p-values, lead us to conclude that there exist statistically significant differences across groups. An overview shows us that the professors and the PhD students/researchers on average are willing to pay much higher amounts than the master and undergraduate students.

Table 1: Frequency of maximum willingness to pay (WTP) to continue to access Portuguese electronic scientific information consortium b-on (Melo and Pires, 2009).

Table 2 summarizes the data to estimate, by regression analysis, the probability of being willing to pay to continue to access the Portuguese electronic scientific information consortium, b-on. We estimated the coefficients and their p-values, in parenthesis, for Linear probability, Logit and Probit models. We decided to use these three models because they give us a more consistent analyses data. We ran regressions including the observations with null values (1930 observations) and regressions not including those observations (1157 observations).The signs of the coefficients are the same across the models and the same independent variables are statistically significant in each model.

Linear probability model

Table 2 first item presents the data accepting null values. The bottom of the column shows the value of F statistics that tests the null hypothesis that all explanatory variables coefficients are equal to zero. The F-statistics is equal to 21.67, with a p-value of 0.000. This means that the null hypothesis should be rejected, telling us that our model as a whole is statistically significant. The R-squared is equal to 0.19 which means than 19% of the total variation is explained by the model. This value seems low, however it is quite normal in regressions using cross-sectional data and in earlier studies in the library and information literature (see Aabø (2005) and references therein).

Let us now interpret each one of the coefficients in the regressions. The coefficient associated with a given explanatory variable “gives the change in the conditional probability of the event occurring for a given change in the value of the explanatory variable” (Gujarati, 550:1995). In what follows we only interpret the coefficients which are statistically significant (meaning that we can reject the hypothesis that they are equal to zero).

The coefficient -0.0098 attached to the variable “Bid” means that, holding all the other factors constant, the probability of being willing to pay to access b-on decreases by a factor of 0.0098 or 0.98 percent. Thus the higher the «price» of the b-on service, the lower the probability of a given user «buying» the service. In other words, the demand for the service is negatively related to its price.

The coefficient 0.0051 attached to the variable “Frequency digital library use” means that, assuming all other factors constant, the probability of being willing to pay the service of b-on increases 0.0051 or 0.5 percent for each additional day of digital libraries use. The coefficients of 0.1032 and 0.1061 attached, respectively, to the variables “Professor” and “Other” (mainly library staff) mean that, assuming all other factors constant, the probability of being willing to pay to access b-on by the “Professor” and “Other” categories are higher about 10 percent and 11 percent as compared with the base category of the “Master student”.

Similarly, the coefficients of 0.0403 and 0.0512 linked, respectively, to the variables “Life and Health Sciences” and “Humanities and Social Science” mean that, assuming all other factors constant, the probability of being willing to pay to access b-on by the users of these two scientific areas are higher about, respectively, 4 and 5 percent as compared with the base scientific area of “Engineering”.

We also can look to the coefficient 0.0588 linked to the institution “Universidade Nova de Lisboa” and say that the probability of being willing to pay to access b-on by the users of this university are higher about, 0.6 percent as compared with the base category “Polytechnic Institutes”.

The fifth column in Table 2 shows the results of the linear probability model without including the observations with null values (1157 observations). Considering the F-statistics we can conclude that the model as whole is statistically significant. In this regression, the explained variance (R-squared) is 35.5%.

The variables that are statistically significant and the signs of the coefficients are the same than in the regression including all the observations, except for the institution “Universidade Nova de Lisboa” and the area of “Humanities and Social Sciences”. The magnitude of the coefficients is similar in the two regressions but with some interesting differences. For instance, in this regression demand is more sensitive to the price and there is a larger difference between users from “Life and Health Sciences” and “Engineering”.

Table 2: Determinants of the probability of being willing to pay to access b-on, with all observations (columns 2,3 and 4) and excluding null values (columns 5,6 and 7).

Logit and Probit models

As previously mentioned our main aim is to study the set of factors that influence the probability of an academic user being willing to pay the b-on services. Ordinary least square regression (OLS) is a very common estimator, but presents some problems when using a binary response variable. The adequate models for binary dependent variables are Logit and Probit models.

The bottom of Table 2 shows the likelihood ratio chi-squared (LRchi2(21)) of the Logit an Probit models with the p-values, respectively, 471.10 (0.000), 465.56 (0.000), 551.03 (0.000) and 536.64 (0.000), tell us that the models as a whole fit significantly well. The pseudo-R-squared is also given. It is a pseudo-R-squared because there is no direct equivalent of an R-squared (from OLS) in non-linear models. The explained variance in the three models are about 19-26 percent (0.1926, 0.2631), for the data accepting null values and 36-39 percent (0.3551, 0.3942) to data not accepting null values. These values are low and higher values indicating better model fit. However they cannot be interpreted as one would interpret an OLS R-squared. Aabø (2005b) explains that more important than the high of the pseudo R2 are the signs of the significant explanatory factors.

The values of log likelihood can be used in comparisons of the Logit and Probit models. The higher values, -659.8280 and -423.3755, mean the Logit model is the best one to explain the impact of the different variables.

In the Table 2 we see the coefficients and the associated p-values. This table shows that these two models, which consider the maximum likelihood estimation (MLE), produce similar results. We obtained statistically significant values for the independent variables “Bid”, “Frequency digital library use”, “Professor”, “Other” and “University of Coimbra” (this last variable is not significant when null values are excluded). We emphasise that the signs of the coefficients are the same in the Logit and the Probit regressions.

The interpretation of the coefficients in the Logit and Probit models can be difficulty. For this reason, as many other researchers, we prefer to compute the more intuitive "marginal effects" of the independent variables on the probability. For continuous independent variables, to find the marginal effects on the probability we must calculate the partial derivate. Equations (3 and 6) give us these values. We computed these data with STATA.

Let us interpret marginal effects for Logit and Probit models show in Table 3 second ant third columns.

Table 3: Marginal effects of LPM, Logit and Probit models explaining the probability of being willing to pay to access the b-on services, with all observations (columns 2,3 and 4) and excluding null values observations (columns 5,6 and 7)

We obtained statistically significant values for the independent variables “Bid”, “Frequency digital library use”, “Professor”, “Other” for observations accepting null values and observations not accepting null value. The signs of the coefficients are the same and the magnitudes are similar, exception for “Other”. Now we can say for the results, of the Table 3 third item, that when “Bid” increases by one euro, the probability of buying the b-on services decreases by a factor 0.0086 or 0.9 percent in the academic population. On the other hand, the probability of being willing to pay to access b-on increases by a factor of 0.0003 or 0.03 percent for each additional day of digital library use. We can also tell that the probability of a Professor buying the b-on services is 8 percent higher than the probability of a master student doing so.

For Probit models, we say that the probability of a user, from area of “Humanities and Social Science”, paying the b-on services increase by a factor 0.0495 or 0.5 percent comparing user from the “Engineering” area.

We can make the other comparisons very easily with all data reported in Table 3 with statistically significant. For instance, looking the data of Logit model and not accepting null values, we can say the probability of a Professor paying the b-on services increase by a factor 0.1127 or 11 percent comparing with the category of Master student, and so on.

4. Demand curve of the b-on services

The demand curve is defined as the relationship between the price of the good and the amount or quantity the consumer is willing to purchase in a specified time period, maintaining constant the other determinants of demand (frequency of use, the type of the user, the scientific area and the institution of the user, in our case).

We analyse demand curve of the Portuguese electronic scientific information consortium b-on – Biblioteca do Conhecimento Online. Figure 1 shows the demand curves, respectively, to Professors and Master students, of the area of Humanities and Social Sciences, from University Nova de Lisboa, with an average frequency digital library use of about 53 days per year. The demand curves shows how the probability of being willing to pay to access b-on varies with the price. The curves were drawn considering the results of the Logit model as it describes better the behaviour of the academic Portuguese users. In our computations we took into account only the statistically significant variables, since for the remaining variables we cannot reject the null hypothesis of being equal to zero.

Figure 1: Demand curves for the Portuguese electronic scientific information consortium b-on – Biblioteca do Conhecimento Online, respectively, to Professors and Master students of the area of Humanities and Social Sciences, from University Nova de Lisboa, with an average frequency digital library use of about 53 days per year

The demand curves show that, for a given price, the probability of a Professor buying the b-on services is higher than the corresponding probability for a Master student. For instance, if the price is 15 Euros per month, the probability of being willing to buy is 63% for a Professor and 41% for a Master student.

5. Conclusions

The levels of importance of access and cost of e-resources in academic environment have quickly increased; WWW, e-books, e-journals and other e-resources have become important sources in libraries. However, there are not enough studies of the impact of the e-resources in the Portuguese academic libraries. This study aimed to improve the knowledge of the value in monetary terms of the scientific e-resources. In order to conduct an accurate assessment we used the contingent valuation method (CVM) to estimate the maximum willingness to pay (WTP) and regression analysis techniques - Linear probability model (LPM), Logit and Probit models – to identify the factors that influence the probability of an user being willing to pay to access the b-on services. The Logit and Probit models are used when the dependent variable is a dichotomous qualitative variable (in our case the categories are: willing to buy, not willing to buy). A qualitative variable can be transformed into a binary variable that takes the value 1 if the respondent answers «yes» and takes the value 0 if he answers «no».

In this paper we studied the factors that influence the probability of a user being willing to pay to access the b-on services. The results show that the probability of buying the b-on services is decreasing with the price charged for the service, showing that the demand for b-on services is quite sensitive to the «price». On the other hand, the users who use more frequently the digital library services present a higher probability of paying to access b-on. Among the various types of users, the Professor category is the one that shows higher probability of buying the services of b-on.

References

Aabø S., (2005). Valuing the benefits of public libraries, Information Economics and Policy, Vol.17, No.2, 175-198.

Barron, D., Williams, R.V., Bajjaly, S., Arns, J. and Wilson, S., (2005). The economic impact of public libraries on South Carolina. University of South Carolina, Columbia. Retrieved, June, 3, 2009. Available at: http://www.libsci.sc.edu/SCEIS/home.htm

British Library, (2004). Measuring Our Values. British Library, London. Retrieved January, 4, 2007. Available at: http://www.bl.uk/pdf/measuring.pdf

Brown, B., Found, C. and McConnell, M., (2007). Federal Science eLibrary Pilot: Seamless, equitable desktop access for Canadian government researchers, The Electronic Library, Vol.25, No.1, 8-17.

Chung, H.-K., (2007). Measuring the economic value of special libraries. The Bottom Line: Managing Library Finances, Vol.20, No.1, 30-44.

Cooper, M. D. and Chen, H.-M. (2001). Predicting the relevance of library catalog search, Journal of the American Society for Information Science and Technology, Vol.52, No.10, 813-827.

Elliott, D.S.; Holt, G.E.; Hayden, S.W. and Holt, L.E., (2007). Measuring your library’s value: how to do a cost-benefit analysis for public library. ALA, Chicago.

Harless, D. W. and Allen, F. R., (1999), Using the Contingent Valuation Method to Measure Patron Benefits of Reference Desk Service in an Academic Library, College & Research Libraries, Vol.60, No.1, 56-69.

FCCN – Fundação para a Computação Científica Nacional, (2009). b-on Biblioteca do Conhecimento Online. Retrieved December, 10, 2009. Available at: http://www.b-on.pt/index.php?option=com_content&view=article&id=116&Itemid=34〈=en.

Gujarati, D.N., (1995). Basic Econometrics, 3rd ed. McGraw-Hill, New York.

Hider, P., (2008). Using the contingent valuation method for dollar valuations of library services, Library Quarterly, Vol.78, No.4, 137-158.

Holt, G.E., Elliott, D. and Moore, A., (1999). Placing a value on public library services, Public Libraries, Vol.38, No.2, 98-108.

Holt, G.E. and Elliott, D., (2003). Measuring outcomes: Applying cost-benefit analysis to middle-sized and smaller public libraries, Library Trends, Vol.51, No. 3, 424-440.

Kerins, G., Madden, R. and Fulton, C., (2004). Information seeking and students studying for professional careers: the cases of engineering and law students in Ireland, Information Research, Vol.10, No.1, paper 208. Retrieved November , 20, 2009. Available at: http://InformationR.net/ir/10-1/paper208.html

Koenig, M.E.D., (1992). The importance of services for productivity: under-recognized and under-invested. Special Libraries, Vol. 83, No. 4, 199-210.

Li, Z. and Mahendra, G., (2009). Using “Recycled Predictions” for Computing Marginal effects s. Retrieved January, 20, 2010. Available at: http://www.lexjansen.com/wuss/2009/hor/HOR-Li.pdf

Liu, L. G., (2003). The economic behaviour of academic research libraries: toward a theory, Library Trends, Vol.51, No.3, 277-292.

Manduca, C.A., Fox, S., et al., (2006). Digital library as network and community center, D-Lib, Vol.12, No.12. Retrieved May, 16, 2009. Available at: http://www.dlib.org/dlib/december06/manduca/12manduca.html

McDermott, M., (2002). National Bibliographic Database and National Union Catalogue: Economic Evaluation for the National Library of New Zealand. National Library of New Zealand. Wellington. Retrieved December, 3 2006. Available at: http://www.natlib.gov.nz/catalogues/library-documents/economic-valuation-nbd-nuc.pdf

McMartin, F., Iverson E., Wolf, A., Morrill, J., Morgan, G. and Manduca, C., (2008). The use of online digital resources and educational digital libraries in higher education, International Journal of Digital Libraries, Vol.9, 65-79.

Melo, L.B., Pires. C., (2009). Measuring the economic value of the electronic scientific information services in Portuguese academic libraries. Journal of Librarianship and Information Science, [accept to publication December 2009].

Missingham, R., (2005). Libraries and economic value: a review of recent studies, Performance Measurement and Metrics, Vol.6, No.3, 142-158.

Morris, A., Ayres C. and Jones A., (2006). Audiovisual materials in UK public libraries: economic sense?, Journal of Documentation, Vol.62, No.5, 555-569.

Morris, A., Sumsion, J. and Hawkins, M., (2002). Economic value of public libraries in the UK, Libri, Vol.52, 78-87.

Murgai, S. R. and Ahmadi, M., (2007). A multiple regression model for predicting reference desk staffing requirements. The Bottom Line: Managing Library Finances, Vol.20, Nº2, 69-76.

Noonan, D. S., (2003). Contingent Valuation and Cultural Resources: A Meta-analytic Review of Literature, Journal of Cultural Economics, Vol.27, 159-176.

Plott, C. R.; Zeiler, K., (2005). The willingness to pay-willingness to accept gap, the “endowment effect”, subject misconceptions, and experimental procedures for eliciting valuations, American Economics Review, Vol.95, No.3, 530-545.

Sin, J. S.-C. and Kim, K.-S., (2008). Use and non-use of public libraries in the information age: a logistic regression analysis of household characteristics and library services variables, Library & Information Science Research, Vol.30, No.3, 207-215.

Tenopir, C.; Read, E., (2000). Patterns of database use in academic libraries. College & Research Libraries, Vol.61, No.May, 234-246.

Tenopir, C., (2003). Use and users of electronic library resources: an overview and analysis of recent research studies. Council on Library and Information Resources, Washington, D.C. Retrieved November, 6, 2009. Available at: http://www.clir.org/pubs/reports/pub120/pub120.pdf

Troll, D.A., (2001). How and why are libraries changing?, Digital Library Federation. Retrieved November, 5, 2009. Available at: http://www.diglib.org/use/whitepaper.htm

Wooldridge, J., (2003). Introductory econometrics: a modern approach. Thomson, South-Western.

Wooldridge, J., (2007). Introdução à Econometria: uma abordagem moderna. Thomson Learning, São Paulo.

Yu, F., (2006). Users’ Emotional and Material Satisfaction at the Micro/Macro Levels in na Academic Library. Thesis submitted for to the Graduate Faculty of School of Information Science for the degree of Doctor of Philosophy, University of Pittsburgh, Pittsburgh. Retrieved December, 6, 2009. Available at: http://etd.library.pitt.edu/ETD/available/etd-07242006160531/unrestricted/FEIYU_ETDDISSERTATION.pdf