516 Longitudinal Associations between Online Usage of Library-Licensed Content and Undergraduate Student Performance Felichism Kabo, Annaliese Paulson, Doreen Bradley, Ken Varnum, Stephanie Teasley* Seeking to better understand the longitudinal association between online usage of library-licensed content and short- and long-term student performance, we linked EZproxy logs to institutional university data to study how library usage impacts se- mester and cumulative GPAs. Panel linear mixed effects regression models indicate online library usage is significantly associated with both semester and cumulative GPAs. The library usage effect is larger for semester GPA, and varies by on- and off- campus residency. The effect on semester GPA is larger for off-campus students, while for cumulative GPA the effect is larger for on-campus students. Longitudinally linked library-institutional data offers key insights on the library’s value. Introduction Library usage is correlated with important undergraduate student outcomes including academic performance and retention. However, the relationship between library usage and academic performance is better understood over the short term, and for specific subsets of students, such as first-year undergraduate students.1 We need to develop a better understanding of this relationship both over the long term, and for all undergraduate students. One reason for our currently limited understanding of this relationship is that, in most universities—owing to privacy concerns—libraries either do not collect or retain user data with identifiers. This makes it impossible to link library usage data with other institutional or administrative data from the university, including data regarding academic success and retention. Another limitation is that library usage data are often collected as very large logs (millions and billions of records) that may require the application of methodological approaches, such as Big Data techniques, to structure and store in ways that make them more amenable to analysis. Therefore, there *  Felichism Kabo is Director of Research, CannonDesign, email: fkabo@cannondesign.com and Research Fellow, Zell Lurie Institute, Ross School of Business, University of Michigan, email: fkabo@umich.edu; Annaliese Paulson is a PhD student at the School of Education, University of Michigan, email: annamp@umich.edu; Doreen Bradley is the Director of Learning Programs and Initiatives at University of Michigan Library, email: dbradley@umich. edu; Ken Varnum is a Senior Program Manager and Discovery Strategist at University of Michigan Library, email: varnum@umich.edu; Stephanie Teasley is a Research Professor at the School of Information, University of Michigan, email: steasley@umich.edu. ©2024 Felichism Kabo, Annaliese Paulson, Doreen Bradley, Ken Varnum, Stephanie Teasley, Attribution-NonCommercial (https://creativecommons.org/licenses/by-nc/4.0/) CC BY-NC. mailto:fkabo@cannondesign.com mailto:fkabo@umich.edu mailto:annamp@umich.edu mailto:dbradley@umich.edu mailto:dbradley@umich.edu mailto:varnum@umich.edu mailto:steasley@umich.edu https://creativecommons.org/licenses/by-nc/4.0/ Longitudinal Associations 517 is a need for empirical, longitudinal studies that not only use identifiable library data, but also employ Big Data and statistical methods to advance our understanding of the library’s contribution to student success. In this paper, we present the results of a longitudinal study of the association between online library resource usage and student performance for the entire population of undergraduates enrolled at the University of Michigan (U-M) between 2016 and 2019. The privacy concerns described above are valid; however, other research domains—for which the potential risk of unintended exposure is higher than those of library usage data, such as the type of patient health information covered by the Health Insurance Portability and Accountability Act of 1996 (HIPAA) —have found ways to successfully handle data while maintaining privacy. Yet, these advances in the biomedical and social sciences, which would better serve the privacy requirements of library professional ethics, are still not widely known in libraries. Fortunately, many libraries now adopt the best privacy practices from the social and biomedical sciences. These initiatives make it possible to employ Big Data methods in longitudinal studies of the links from library usage to academic outcomes for the entire stu- dent body. There are two such initiatives critical to the work described in this paper: first, after a multi-year process of engaging with a diverse set of stakeholders including the U-M Learning Analytics Task Force, the U-M Library revised its privacy policy in 2016 to allow the collection and retention of identifiable library usage data;2 second, the Library Learning Analytics Project (LLAP)—funded by the Institute of Museum and Library Services (IMLS)—examined how libraries impact learning outcomes including in course instruction. Learning processes require that members of the university community engage in activities such as accessing digital data and publication repositories, conducting literature reviews, managing citations, and creating data management plans. These activities often entail interacting with the library virtually, such as when accessing and retrieving library licensed content through the proxy server. This paper reports on analyses performed on the links between off-campus, or off-network, electronic usage of library resources, as well as undergraduate academic performance over the short- and long-term. The best context for work of this nature is one in which library users have agency with how they engage with the library services in question. For library licensed content, individuals can access these resources via computers that are on-campus (physically located in the library or elsewhere in the university), or virtually via the proxy server should they choose to use these resources when off-campus. For this reason, the authors limited the analysis to the relationship between online library usage and student outcomes to the time before the COVID-19 pandemic. That is, the study focuses on when students had the choice of accessing library licensed content through on- or off-campus means. Literature Review This work is informed by models of information behavior,3 which describes how individuals seek and utilize information.4 Information behavior is contingent on factors such as social contexts, socio-demographics, individual expertise, as well as access to, and ease of use of, technology.5 The work also builds on two lines of inquiry: 1) research into the associations between college residence and academic performance; 2) work on digital inequalities or the digital divide. We examine the link from library usage to student outcomes in two ways: first, defining library usage in terms of use of licensed online content provided by the library, 518 College & Research Libraries May 2024 and second, evaluating the impacts of on-campus residency for access to library and other resources and reliable internet. Research on campus residency has examined the issue of whether there are gains in learn- ing and academic performance from living on- versus off-campus. A study of nearly 95,000 first year students in the United States found living on-campus was significantly associated with a range of learning variables, even though the residency effect size was small to medium.6 An earlier study of first-year students found that the benefits of on-campus residency on academic performance were different across, and within, racial groups. For example, Black students who lived on-campus had significantly higher grade point averages (GPAs) than Black students who lived off-campus.7 Approaching the issue from a different angle, a study of the causal link between campus residency and academic outcomes found living in university-owned housing had a positive association with student retention.8 This finding was in line with prior analysis that established an association between on-campus living and academic performance and student retention for first-year students.9 However, an important caveat is that students who were better prepared academically were more likely to live on-campus as opposed to off-campus.10 Most studies of the link between on-campus residence and student persistence are based on four-year institutions. One exception is a quasi-experimental analysis of com- munity college students that found that living on-campus was associated with a significant increase in upward transfer (to a four-year institution) and, subsequently, bachelor’s degree completion rates.11 However, the association between on-campus residence and academic outcomes is not always positive. A study conducted at a public four-year university in the southeast United States found that commuter or off-campus students had higher GPAs than residential or on-campus students.12 Demographic, geographic, and economic factors all help shape digital disparities in American K-16 education. These disparities are commonly referred to as the “digital divide,” or the gap between those who have access to the internet and other information and commu- nication technologies (ICT), and those who do not. Digital inequalities and disparities affect a broad range of life opportunities and outcomes beyond education, such as economic activity and health care.13 In education, digital inequalities and disparities are a life-course issue and affect disadvantaged students. Their effects are felt from early14 to late in the K-16 pipeline.15 The increasing use of technology inside and outside the classroom has significant ramifica- tions for the digital divide and its effect on student performance. Importantly, some groups of students are systematically more likely to experience digital disparities than others. For example, in 2015, higher percentages of students who were White (66%) used the internet at home compared to Black (53%), Hispanic (52%), and American Indian/Alaska Native (49%) students.16 American Indian/Alaska Native students are more likely than other racial groups to have no internet access, or to have only dial-up internet access at home.17 The interaction of demography and geography disadvantages some students further. While 18 percent of all students in remote rural areas did not have internet access, or had only dial-up access in 2015, a much larger percentage of Black (41%) students in remote rural areas did not have internet access compared to White (13%) and Asian (11%) students. Having no or low-bandwidth in- ternet is detrimental to any form of online learning. For example, students cannot participate in classes offered via video meeting systems that rely on high-speed internet.18 The COVID-19 pandemic worsened the effects of the digital divide, such as for rural students.19 Students of color have been especially impacted by the pandemic and, as noted earlier, are more likely to Longitudinal Associations 519 lack access to reliable broadband internet, and even computers. The pandemic exacerbated existing educational disparities for minority students and likely widened the achievement gap for students of low socioeconomic status.20 In the United States, the effects of the pandemic on the digital divide have demonstrably impacted the entire K-16 pipeline. There were varied institutional responses across the Ameri- can higher education landscape. Perversely, these varied responses present opportunities for “quasi-experimental” observations regarding the impact of the digital divide on amplifying disparities in student performance. For example, where many colleges and universities stipu- lated that students residing on-campus leave these residences, some made allowances for students who could not return home, which thus allowed them to still have access to reliable broadband internet via the institution.21 What was fairly universal, however, was the extent and speed with which university libraries adapted to offering primarily online resources,22 which can only meaningfully be accessed via reliable internet connections. Thus, not only were students no longer able to access the library’s physical collections, but they also no longer had access to the library as a study space, including for group or collaborative activi- ties.23 By examining how “regular” (pre-pandemic) electronic library usage is associated with academic performance, this study may therefore help us better understand the likely impacts of the worsening of the digital divide during the pandemic. Based on evidence that the digital divide has worsened during the pandemic,24 we can reasonably assume that the importance of the relationship between online library usage and academic performance has only increased. The literature also indicates that models of student performance need to account for other demographic, socioeconomic, and academic factors, including include gender, first-generation status, family or household income, high school GPA, and academic class level. Across na- tional contexts in developed countries, female students are more likely to have both higher work ethics and GPAs than males.25 First-generation students are more likely to contend with barriers to academic success—such as job and family responsibilities and/or inadequate study skills26—and thus tend to have poorer academic outcomes.27 Students who enter college with higher family or household incomes have significantly higher GPAs than those from lower socioeconomic backgrounds.28 High school GPA is a strong predictor of college or university GPA as well, especially in the first year.29 Academic class level is correlated with GPA, as up- per class students (e.g. seniors) are more likely to have higher grades, especially in classes that also have lower class students, such as sophomores.30 Theoretical Framework Building on models of information-seeking behavior, we developed a theoretical framework (Figure 1) that correlates student performance with library usage as captured by EZproxy sessions, controlling for factors like socio-demographics and academic background.31 A key strength of the framework is that it presents testable relationships among demographic and contextual factors, information-seeking behaviors, and academic outcomes. This paper examines the association between information-seeking behavior (off-campus or off-network electronic library resource use), and both semester and cumulative GPA. However, this relationship must also be understood in the context of contextual factors (“in- tervening” variables), which contribute to disparities in access to the digital resources that are needed to make effective use of electronic library licensed content. Research shows that access to, and proper use of, digital technology generally has a positive correlation with academic 520 College & Research Libraries May 2024 performance; this finding is robust across regional and national settings.33 Based on these findings, we hypothesize that students identified as accessing online library licensed content will have better academic outcomes than those students with no evidence of digital access to these resources. However, there is also evidence that our hypothesized relationship has both short- and long-term implications. While not specific to electronic resources, studies suggest that library usage is positively correlated with student performance both in the short-term,34 and in the long-term.35 Therefore: H1: Students who electronically access library licensed content will have higher semester GPAs. H2: Students who electronically access library licensed content will have higher cumula- tive GPAs. Methodology The study sample is all undergraduate students (N = 45,254) who were enrolled at the Uni- versity of Michigan (U-M) from fall 2016 through winter 2019 (or September 2016 through April 2019). We focus on these six semesters before the pandemic because students had more agency with respect to their usage of electronic library licensed content. That is, students could choose to access materials using computers that are physically on-campus, or off-campus access via the proxy server. We sourced library usage data from EZproxy logs (690,300,076 records) stored in a secure repository that the U-M Library managed. We obtained student demographic and outcome data (GPAs) from the research-focused Learning Analytics Data Architecture (LARC) data set maintained by the U-M Office of Enrollment Management. The project team implemented several measures to protect the privacy and confidentiality of the FIGURE 1 Theoretical Framework for Associations between Library Usage and Student Outcomes Adapted from Models of Information Behavior32* *These models draw on research from multiple fields including information science, psychology, decision-making, innovation, health communication, and consumer research. Longitudinal Associations 521 individuals in the library and LARC data. For example, the library data were classified at the “Restricted” level of data security. This is the highest classification or sensitivity level for U-M institutional data, has the most stringent legal or regulatory requirements, and has the most prescriptive security controls. These controls included restricting access to only two members of the project team, storing and curating the data on a secure enclave, setting up access to the enclave via a terminal in a locked and restricted data room, and requiring that all analyses be performed on the enclave. Our primary interest in this paper is the relationship between information-seeking behavior (EZproxy sessions) and student performance. EZproxy is proxy server software that many academic libraries use to give authenticated off-campus users access to electronic resources licensed by the library as if they were on campus. After authenticating to a campus system, off-campus users receive an on-campus IP address and are then considered to be a member of the campus community by the information provider. The authors cleaned and normalized raw, unstructured EZproxy logs using Python scripts and regular expressions, and then entered the data into a relational database using structured query language (SQL) scripts. Over 80 percent of the EZproxy data have strong university identifiers which enables merges with other administrative data, such as LARC. It is critical to note that EZproxy logs available to the study: a) did not include any on-campus usage, and b) did not include anyone who used the university’s virtual private networks (VPN). Using SQL and R scripts, we merged the data and exported the resultant data set into Stata 16 statistical software for modeling and analysis.36 The theoretical framework shown in Figure 1 suggests that student outcomes are a func- tion of factors, such as race and gender, that apply to all the students in the study (“fixed effects”), and factors, such as academic units or schools, that cluster student behaviors and outcomes (“random effects”). We also accounted for student random effects for unobserved, time invariant factors, such as motivation or grit. Thus, we ran panel linear mixed effects (LME) regression models of the association between library usage and student GPA, contingent on students being enrolled in at least four semesters over the study period. Variables The two continuous dependent variables are semester GPA (“SEM_GPA”) and cumulative GPA (“CUM_GPA”). While SEM_GPA is on a 0 – 4.4 scale and CUM_GPA is on a 0 – 4.314 scale, fewer than 0.5 percent of students have a semester or cumulative GPA that is higher than 4.0. The dichotomous independent variable “EZproxy Session in Term” is coded one if a student is associated with one or more EZproxy sessions during an academic term, and is coded zero otherwise. We also account, or control, for potential “intervening” variables as follows: the dichoto- mous variable “On-campus Residence” is coded one if a student was residing in a university residence, and zero otherwise; the variable “High School GPA” is on a continuous 0 – 4 scale and captures a student’s academic performance before enrollment at the university; gender is captured by the dichotomous variable “GENDER” (1 = Female, 2 = Male). Note that the LARC data set used for the study does not account for non-binary options. The effects of race, first generation status, family income, and class level were controlled for using the categorical variables “RACE” (1 = White, 2 = Asian, 3 = Black, 4 = Hispanic, 5 = Two or More, 6 = Other, 7 = Not Indicated), “FIRST GENERATION” (1 = First Gen, 2 = Not First Gen, 3 = Don’t Know), 522 College & Research Libraries May 2024 “FAMILY INCOME” (1 = More than $100,000; 2 = Less than $25,000; 3 = $25,000 - $49,999; 4 = $50,000 - $74,999; 5 = $75,000 - $99,999; 6 = Don’t Know; 7 = Missing), and “CLASS LEVEL” (1 = Freshman, 2 = Sophomore, 3 = Junior, 4 = Senior), respectively. Statistical Modeling We ran panel LME regression models with random effects for individuals, as well as by school or academic unit (see Table A.7 in the appendix for a list of the 15 schools that undergraduate students were affiliated with). LME models, an extension of simple linear models, are use- ful when there is non-independence in the data. This arises from, for example, a hierarchical structure in the data, such as when students are sampled from within academic units. Panel regression approaches are necessary when working with longitudinal study designs, where multiple observations are made on each individual subject. LME models have both fixed ef- fects, which are directly estimated and are analogous to standard regression coefficients, and random effects, which in our case take the form of random intercepts. The fixed effects in our LME models correspond to the “intervening” variables. The random effects account for the fact that student behaviors and outcomes may, instead of being uniform across all undergraduates, be grouped by academic units which map onto disciplinary boundaries that likely affect library usage. The random effects also enable us to account for unobserved, time-invariant individual- level factors, such as motivation or grit. Table A.7 in the appendix shows that there are notable differences across schools with respect to the percentage of students who have at least one EZproxy session during an academic term. After each LME model, we ran a likelihood-ratio comparing this model with a one-level ordinary linear regression. This test was highly signifi- cant for each of the LME models in our study, supporting the decision to use the LME model. Findings and Discussion Descriptive Statistics Over half of enrolled undergraduates had at least one EZproxy session during an academic term over the study period (Table 1). There are some notable differences in library usage among enrolled undergradu- ates. Table 2 below illustrates differences in library usage by demographic, academic, and residency factors for the winter 2019 term (see the appendix for similar statistics on all se- mesters). Off-campus students are more likely to have at least one EZproxy session in the academic term than are on-campus students. This makes sense because students who are on-campus are more likely to access electronic library resources on the university’s network, in which case authentication is not required. Recall that students are identifiable in the EZproxy logs only when authentication is re- quired. An example of this is when a student accesses electronic library resources outside TABLE 1 Percentage of Students Associated with EZproxy Sessions by Semester, Fall 2016 – Winter 2019 Academic Term Enrolled Students EZproxy Session % ≥ 1 EZproxy Session FA 2016 28,682 16,605 58% WN 2017 27,408 13,434 49% FA 2017 29,161 16,034 55% WN 2018 27,852 14,855 53% FA 2018 29,726 16,191 54% WN 2019 28,355 16,299 57% TOTAL* 171,184 94,418 55% *This is a tally of unique student-term combinations, as there were 45,254 enrolled undergraduates over the study period. Longitudinal Associations 523 TABLE 2 Percentage of Undergraduate Students Associated with EZproxy Sessions by Socio- Demographics and Academic Background, Winter 2019 Variable Category Enrolled Students EZproxy Session % ≥ 1 EZproxy Session First Gen Status First Gen 3,890 2,310 59% Not First Gen 24,418 13,957 57% Don’t Know 47 32 68% Family Income Less than $25,000 1,507 923 61% $25,000–$49,999 2,212 1,269 57% $50,000–$74,999 2,009 1,217 61% $75,000–$99,999 2,074 1,213 58% More than $100,000 13,951 7,892 57% Don’t Know 515 278 54% Missing Income Information 6,087 3,507 58% Class Level Freshman 2,557 1,300 51% Sophomore 6,397 3,373 53% Junior 7,132 4,114 58% Senior 12,269 7,512 61% Race Asian 5,829 3,137 54% Black 1,268 766 60% Hispanic 1,899 1,099 58% White 16,604 9,738 59% 2 or More 1,302 745 57% Other 46 22 48% Not Indic 1,407 792 56% Gender Female 14,204 9,219 65% Male 14,151 7,080 50% Residency On-campus 9,261 4,540 49% Off-campus 19,110 11,765 62% Academic Unit Architecture 181 124 69% Art and Design 524 381 73% Business Administration 1,799 740 41% Dental Hygiene 101 70 69% Education 126 54 43% Engineering 6,313 2,847 45% Information 260 122 47% Joined Degree Program 10 7 70% Kinesiology 954 678 71% Literature, Science and the Arts 16,409 10,030 61% Music, Theare, & Dance 717 515 72% Nursing 607 475 78% Pharmacy 55 36 65% Public Health 157 116 74% Public Policy 142 104 73% 524 College & Research Libraries May 2024 the university’s network such as from an off-campus residence, coffee shop, etc. There is a significant gender difference, with females much more likely than males to have an EZproxy session, despite more males (69%) than females (66%) residing off-campus in winter 2019. Note that the likelihood of having at least one EZproxy session increases with each class level. Perhaps this is because students are more likely to move or reside off-campus as they progress from freshman to seniors. However, a factor that weakens this explanation is U-M does not require freshmen and sophomores to live on-campus, as is the case in some colleges and uni- versities. An alternative explanation is that lower-level classes are less research-intensive and students may not need library-provided resources to complete research and writing projects. Finally, there are noteworthy differences between academic units. Additional work would be needed to clarify the factors that account for these differences. For example, 45 percent of engineering undergraduates had at least one EZproxy session compared to 73 percent of art and design undergraduates, even though both academic units are co-located at the university. A potential explanation could be that these differences reflect disciplinary differences (STEM versus arts and humanities). Another plausible explanation could be that the differences re- flect gaps in technological expertise between the two groups of students, with engineering students being more likely to access electronic library resources using the university’s VPN which bypasses the authentication process on the library’s proxy server. We should also keep in mind factors such as the interplay between residency and socioeconomic statuses. It is more expensive to live on- rather than off-campus, implying that students in the former group may tend to be from wealthier families. For example, 78 percent of nursing undergraduates had at least one EZproxy session, compared to 41 percent of business administration undergradu- ates. Tabulations of residency for the two academic units showed that 32 percent of business undergraduates resided on-campus in winter 2019, compared to 20 percent of nursing under- graduates. Similarly, tabulations of family income for these two academic units showed that 58 percent of business undergraduates had a family income of more than $100,000, compared to 48 percent of nursing undergraduates. These findings suggest that library usage data have the potential to reveal disparities and inequalities, and could therefore help libraries make significant analytical contributions of interest to their institutions. Regression Models The results from the regression modeling are summarized in Tables 3 (semester GPA) and 4 (cumulative GPA). The regression models showed positive and statistically significant associa- tions between having at least one EZproxy session in an academic term, and both semester and cumulative GPAs, controlling or accounting for residency, race, gender, high school GPA, family income, first generation status, and class level. Overall, the results from the regression models for semester GPA provide strong sup- port for hypothesis H1. That is, students that use electronic library licensed content have higher semester GPAs. Having an EZproxy session during an academic term was correlated with a 0.14 point increase in semester GPA (model 1). To further examine the impact of campus residency, considering the link between authentication requirements and a stu- dent’s presence in the EZproxy logs, we ran separate models for on-campus (model 2) and off-campus (model 3) students. For off-campus students, having an EZproxy session in an academic term is correlated with a 0.17 point increase in semester GPA. In comparison, for on-campus students, having an EZproxy session in an academic term is correlated with a Longitudinal Associations 525 TABLE 3 Panel LME Regressions for Association between Library Usage and Semester GPA, FA 2016–WN 2019 (Four or More Semesters) (1: All Students) (2: On-campus) (3: Off-Campus) VARIABLES SEM_GPA SEM_GPA SEM_GPA EZproxy Session in Term 0.138*** 0.0837*** 0.171*** (0.00304) (0.00415) (0.00419) On-campus Residence 0.0967*** (0.00471) High School GPA 0.0273*** 0.0435*** 0.0211*** (0.00194) (0.00345) (0.00235) GENDER (Reference = Female) Male –0.0908*** –0.0616*** –0.108*** (0.00529) (0.00662) (0.00685) RACE (reference = White) Asian 0.0499*** 0.0534*** 0.0404*** (0.00660) (0.00838) (0.00851) Black –0.376*** –0.374*** –0.400*** (0.0128) (0.0145) (0.0181) Hispanic –0.164*** –0.181*** –0.143*** (0.0107) (0.0126) (0.0145) Two or More –0.101*** –0.0812*** –0.121*** (0.0126) (0.0150) (0.0167) Other –0.239*** –0.209** –0.255** (0.0631) (0.0781) (0.0784) Not Indic –0.00568 0.0168 –0.0188 (0.0121) (0.0160) (0.0155) FIRST GENERATION (reference = First Gen) Not First Gen 0.119*** 0.138*** 0.112*** (0.00851) (0.0106) (0.0112) Don’t Know –0.166** –0.0157 –0.202** (0.0525) (0.0845) (0.0640) FAMILY INCOME (reference = More than $100,000) Less than $25,000 –0.150*** –0.129*** –0.166*** (0.0127) (0.0159) (0.0167) $25,000 – $49,999 –0.101*** –0.115*** –0.102*** (0.0106) (0.0131) (0.0141) $50,000 – $74,999 –0.0557*** –0.0719*** –0.0581*** (0.0104) (0.0133) (0.0134) $75,000 – $99,999 –0.0545*** –0.0528*** –0.0572*** (0.0100) (0.0129) (0.0128) Don’t Know –0.0505* –0.0385 –0.0688** (0.0196) (0.0238) (0.0260) Missing Income Information –0.00505 –0.0117 –0.00127 (0.00652) (0.00827) (0.00831) 526 College & Research Libraries May 2024 0.09 point increase in semester GPA. For the other “intervening” variables, it is noteworthy that the GPA gender gap in favor of females is smaller for on-campus students compared to their off-campus peers. Interestingly, notwithstanding the small sizes of the effects, the first-generation disadvantage of lower GPAs is more pronounced for on-campus students relative to their off-campus peers. Overall, the results from the regression models for cumulative GPA provide strong sup- port for hypothesis H2. That is, students that use electronic library licensed content have higher cumulative GPAs. However, the effect of having at least one EZproxy session in an academic term is smaller for cumulative GPA than it is for semester GPA. Model 4 shows that having an EZproxy session in an academic term was correlated with a 0.02 point increase in cumulative GPA. To examine the effect of being on- or off-campus, we ran separate models for on- (model 5) and off-campus (model 6) students, which show differences between the two groups of students—although in ways that are opposite to semester GPA. Having an EZproxy session in an academic term has a larger effect on cumulative GPA for on-campus students TABLE 3 Panel LME Regressions for Association between Library Usage and Semester GPA, FA 2016–WN 2019 (Four or More Semesters) (1: All Students) (2: On-campus) (3: Off-Campus) VARIABLES SEM_GPA SEM_GPA SEM_GPA CLASS LEVEL (reference = Freshman) Sophomore 0.0176*** 0.0237*** 0.0184 (0.00498) (0.00455) (0.0229) Junior 0.0326*** 0.00259 0.0704** (0.00605) (0.00680) (0.0229) Senior 0.0815*** 0.0403*** 0.116*** (0.00662) (0.0112) (0.0230) Constant 3.207*** 3.242*** 3.174*** (0.0357) (0.0444) (0.0448) Observations 151,049 53,896 97,153 Standard errors in parentheses *** p<0.001, ** p<0.01, * p<0.05 TABLE 4 Panel LME Regressions for Association between Library Usage and Cumulative GPA, FA 2016–WN 2019 (Four or More Semesters) (4: All Students) (5: On-Campus) (6: Off-Campus) VARIABLES CUM_GPA CUM_GPA CUM_GPA EZproxy Session in Term 0.0201*** 0.0242*** 0.0144*** (0.000896) (0.00190) (0.000871) On-campus Residence 0.0216*** (0.00149) High School GPA 0.0222*** 0.0364*** 0.0141*** (0.00162) (0.00313) (0.00182) Longitudinal Associations 527 TABLE 4 Panel LME Regressions for Association between Library Usage and Cumulative GPA, FA 2016–WN 2019 (Four or More Semesters) (4: All Students) (5: On-Campus) (6: Off-Campus) VARIABLES CUM_GPA CUM_GPA CUM_GPA GENDER (Reference = Female) Male –0.0735*** –0.0573*** –0.0841*** (0.00447) (0.00603) (0.00528) RACE (reference = White) Asian 0.0655*** 0.0654*** 0.0559*** (0.00558) (0.00763) (0.00658) Black –0.330*** –0.328*** –0.364*** (0.0108) (0.0134) (0.0139) Hispanic –0.157*** –0.168*** –0.150*** (0.00904) (0.0116) (0.0112) Two or More –0.0769*** –0.0648*** –0.0885*** (0.0107) (0.0137) (0.0129) Other –0.197*** –0.159* –0.192** (0.0549) (0.0717) (0.0611) Not Indic 0.0120 0.0296* –0.00456 (0.0102) (0.0145) (0.0121) FIRST GENERATION (reference = First Gen) Not First Gen 0.105*** 0.118*** 0.102*** (0.00721) (0.00971) (0.00867) Don’t Know –0.209*** –0.0901 –0.233*** (0.0451) (0.0786) (0.0503) FAMILY INCOME (reference = More than $100,000) Less than $25,000 –0.113*** –0.101*** –0.126*** (0.0107) (0.0147) (0.0129) $25,000 – $49,999 –0.0806*** –0.0963*** –0.0837*** (0.00901) (0.0120) (0.0109) $50,000 – $74,999 –0.0342*** –0.0543*** –0.0364*** (0.00883) (0.0122) (0.0104) $75,000 – $99,999 –0.0438*** –0.0416*** –0.0460*** (0.00850) (0.0118) (0.00990) Don’t Know –0.0326* –0.0317 –0.0454* (0.0165) (0.0217) (0.0200) Missing Income Information –0.00391 –0.00968 –0.00127 (0.00553) (0.00753) (0.00644) CLASS LEVEL (reference = Freshman) Sophomore –0.00343* –0.00615** 0.00310 (0.00150) (0.00209) (0.00513) Junior –0.00137 –0.0235*** 0.0217*** (0.00187) (0.00322) (0.00517) Senior 0.0241*** –0.0114* 0.0483*** (0.00209) (0.00538) (0.00520) 528 College & Research Libraries May 2024 compared to their off-campus peers. However, the magnitude of both effects is very small. Also note that, like semester GPA, the female advantage in cumulative GPA was smaller for on-campus students relative to off-campus students. The first-generation disadvantage with respect to lower cumulative GPAs is more pronounced for on-campus students compared to those that are off-campus. The study findings suggest that using library resources positively effects academic performance. These effects were larger in magnitude for semester GPA relative to cumula- tive GPA. For example, regarding semester GPA, first-generation students had a lower GPA (-0.119) than non-first-generation students. Further, males had a lower semester GPA (-0.091) than females. Thus, the impacts of gender and first-generation status on semester GPA were smaller in magnitude than the impact of having at least one EZproxy session during an academic term. Conclusion Because library data are often not integrated into other university data, there are major ob- stacles in demonstrating the richness and complexity of the value of academic library usage for the students who use these resources. We show that merging library usage and student outcome data yields valuable insights on the value of the academic library. Understand- ing patterns of off-campus use of library resources offers an additional point of insight into potential gaps in use by certain groups of students, such as those living off campus, which may correlate with lower academic success and retention. If students in particular programs tend to live off campus, yet their programs are library-research intensive, what could this mean for those students? For example, 80 percent of undergraduate nursing students live off campus, yet the nursing program integrates the library heavily in its curriculum. We could explore off-campus use by students in this program to potentially identify students at risk of lower academic performance, or to provide indicators to faculty advisors if a student’s GPA in research-intensive courses falls below a certain threshold. As additional data from other library services is collected in the future, libraries can develop models to explore other questions around library usage, student success, and curricular integration. Libraries could use the work by the LLAP and allied initiatives to identify opportunities for mitigating educational disparities. Library usage data adds depth of perspective of the student experi- ence, and student engagement broadly, during undergraduate study, and can therefore be a valuable addition to institutions of higher education as they continue to make data-informed TABLE 4 Panel LME Regressions for Association between Library Usage and Cumulative GPA, FA 2016–WN 2019 (Four or More Semesters) (4: All Students) (5: On-Campus) (6: Off-Campus) VARIABLES CUM_GPA CUM_GPA CUM_GPA Constant 3.430*** 3.376*** 3.456*** (0.0275) (0.0358) (0.0304) Observations 151,049 53,896 97,153 Standard errors in parentheses *** p<0.001, ** p<0.01, * p<0.05 Longitudinal Associations 529 decisions to improve undergraduate education. Further, in the process of doing this work, we have created shareable scripts and tools that could be used to replicate our work in other institutional settings. These and other resources can be downloaded for free from the LLAP project’s GitHub site (https://github.com/Learning-Library-Analytics-Project) and website (https://libraryanalytics.org/). Libraries are often new participants within campus learning analytics efforts. The research described here could lead to new partnerships between libraries and other institutional or- ganizations. Much as traditional academic advisors and partners have great insight into the specific needs and capabilities of their students, so could libraries better tailor their services to those needs. By being better informed about both the kinds of assignments and the needs of the individual students, along with a more granular conceptualization of the technologies they have access to, library staff could be better situated to deliver information services tailored to individual needs. As noted by researcher Megan Oakleaf, designing library services and instruction for the average student harms almost everyone (Oakleaf et al. 2020).37 Future work could build on our findings by disentangling the effects of students who are off-campus and not using the VPN (and thus need authentication), versus those who are on-campus but choose to access library licensed content via non-university devices, and hence the library proxy server. Undoubtedly there are economic, technical, and experiential factors contributing to these types of differences in accessing library licensed content. Unfortunately, we were not able to capture them in our study. In addition to multiple socioeconomic factors that could impact student use of library licensed content, there are other factors that could account for these differences, such as the varying nature and demands of curricula across programs and colleges. While there is a healthy demand for library curriculum-integrated instruction (CII) at U-M, programs and instructors may require CII at different times in the progression of a student’s academic career. For example, some programs require library CII in first-year experience courses, while other programs may only require CII in the third- or fourth-year. This suggests several lines of future inquiry, such as how course selection affects the need and motivation to use library-licensed resources, or even how the level of study (such as first-year, third-year, and so on) correlates to use of licensed resources and, subsequently, to academic outcomes. Acknowledgement The work described in this paper is primarily supported by funding from the Institute of Museum and Library Services (IMLS, LG-96-18-0040-18), and secondarily by the University of Michigan Library. https://github.com/Learning-Library-Analytics-Project https://libraryanalytics.org/ 530 College & Research Libraries May 2024 Appendix Tables A.1 – A.7 show the percentages of students who had at least one EZproxy session in an academic term by various sociodemographic and academic factors. TABLE A.1 Percentage of Undergraduate Students Associated with EZproxy Sessions by First-Gen Status, FA16–WN19 Academic Term First-Gen Status Enrolled Students EZproxy Session % ≥ 1 EZproxy Session FA 2016 First-Gen 3,520 2,062 59% Not First-Gen 24,903 14,372 58% Don’t Know 259 171 66% WN 2017 First-Gen 3,364 1,664 49% Not First-Gen 23,818 11,631 49% Don’t Know 226 139 62% FA 2017 First-Gen 3,753 2,054 55% Not First-Gen 25,316 13,928 55% Don’t Know 92 52 57% WN 2018 First-Gen 3,605 2,025 56% Not First-Gen 24,162 12,788 53% Don’t Know 85 42 49% FA 2018 First-Gen 4,091 2,308 56% Not First-Gen 25,582 13,855 54% Don’t Know 53 28 53% WN 2019 First-Gen 3,890 2,310 59% Not First-Gen 24,418 13,957 57% Don’t Know 47 32 68% TABLE A.2 Percentage of Undergraduate Students Associated with EZproxy Sessions by On-Campus, FA16–WN19 Academic Term Residency Enrolled Students EZproxy Session % ≥ 1 EZproxy Session FA 2016 Off-campus 19,130 11,554 60% On-campus 9,552 5,051 53% WN 2017 Off-campus 17,971 10,353 58% On-campus 9,437 3,081 33% FA 2017 Off-campus 19,993 12,049 60% On-campus 9,168 3,985 43% WN 2018 Off-campus 18,793 11,043 59% On-campus 9,059 3,812 42% FA 2018 Off-campus 20,357 12,014 59% On-campus 9,386 4,187 45% WN 2019 Off-campus 19,110 11,765 62% On-campus 9,261 4,540 49% Longitudinal Associations 531 TABLE A.3 Percentage of Undergraduate Students Associated with EZproxy Sessions by Gender, FA16–WN19 Academic Term Gender Enrolled Students EZproxy Session % ≥ 1 EZproxy Session FA 2016 Female 14,296 9,510 67% Male 14,386 7,095 49% WN 2017 Female 13,630 7,817 57% Male 13,778 5,617 41% FA 2017 Female 14,599 9,227 63% Male 14,562 6,807 47% WN 2018 Female 13,910 8,589 62% Male 13,942 6,266 45% FA 2018 Female 14,833 9,304 63% Male 14,893 6,887 46% WN 2019 Female 14,204 9,219 65% Male 14,151 7,080 50% TABLE A.4 Percentage of Undergraduate Students Associated with Ezproxy Sessions by Class Level, FA16–WN19 Academic Term Class Level Enrolled Students EZproxy Session % ≥ 1 EZproxy Session FA 2016 Freshman 5,665 2,982 53% Sophomore 6,621 3,724 56% Junior 7,035 3,979 57% Senior 9,361 5,920 63% WN 2017 Freshman 2,727 874 32% Sophomore 6,296 2,383 38% Junior 6,489 3,291 51% Senior 11,896 6,886 58% FA 2017 Freshman 5,387 2,391 44% Sophomore 7,043 3,704 53% Junior 7,084 3,918 55% Senior 9,647 6,021 62% WN 2018 Freshman 2,511 1,088 43% Sophomore 6,407 2,911 45% Junior 6,949 3,785 54% Senior 11,985 7,071 59% FA 2018 Freshman 5,440 2,477 46% Sophomore 6,957 3,601 52% Junior 7,666 4,257 56% Senior 9,663 5,856 61% WN 2019 Freshman 2,557 1,300 51% Sophomore 6,397 3,373 53% Junior 7,132 4,114 58% Senior 12,269 7,512 61% 532 College & Research Libraries May 2024 TABLE A.5 Percentage of Undergraduate Students Associated with EZproxy Sessions by Family Income, FA16–WN19 Academic Term Family Income Enrolled Students EZproxy Session % ≥ 1 EZproxy Session FA 2016 Less than $25,000 1,470 896 61% $25,000 – $49,999 2,073 1,206 58% $50,000 – $74,999 2,190 1,294 59% $75,000 – $99,999 2,356 1,372 58% More than $100,000 14,246 8,256 58% Don’t Know 935 558 60% Missing Income Information 5,412 3,023 56% WN 2017 Less than $25,000 1,417 724 51% $25,000 – $49,999 1,973 953 48% $50,000 – $74,999 2,114 1,069 51% $75,000 – $99,999 2,249 1,145 51% More than $100,000 13,636 6,683 49% Don’t Know 851 435 51% Missing Income Information 5,168 2,425 47% FA 2017 Less than $25,000 1,486 855 58% $25,000 – $49,999 2,091 1,139 54% $50,000 – $74,999 2,090 1,134 54% $75,000 – $99,999 2,210 1,263 57% More than $100,000 14,336 7,749 54% Don’t Know 476 263 55% Missing Income Information 6,472 3,631 56% WN 2018 Less than $25,000 1,441 795 55% $25,000 – $49,999 2,026 1,119 55% $50,000 – $74,999 2,024 1,123 55% $75,000 – $99,999 2,080 1,167 56% More than $100,000 13,689 7,095 52% Don’t Know 430 222 52% Missing Income Information 6,162 3,334 54% FA 2018 Less than $25,000 1,586 911 57% $25,000 – $49,999 2,307 1,299 56% $50,000 – $74,999 2,066 1,150 56% $75,000 – $99,999 2,161 1,204 56% More than $100,000 14,632 7,760 53% Don’t Know 540 285 53% Missing Income Information 6,434 3,582 56% WN 2019 Less than $25,000 1,507 923 61% $25,000 – $49,999 2,212 1,269 57% $50,000 – $74,999 2,009 1,217 61% $75,000 – $99,999 2,074 1,213 58% More than $100,000 13,951 7,892 57% Don’t Know 515 278 54% Missing Income Information 6,087 3,507 58% Longitudinal Associations 533 TABLE A.6 Percentage of Undergraduate Students Associated with EZproxy Sessions by Race, FA16 – WN19 Academic Term Race Enrolled Students EZproxy Session % ≥ 1 EZproxy Session FA 2016 Asian 5,460 3,019 55% Black 1,268 730 58% Hispanic 1,564 916 59% White 17,743 10,439 59% 2 or More 1,111 642 58% Other 53 30 57% Not Indic 1,483 829 56% WN 2017 Asian 5,282 2,425 46% Black 1,213 574 47% Hispanic 1,500 747 50% White 16,876 8,438 50% 2 or More 1,084 515 48% Other 53 23 43% Not Indic 1,400 712 51% FA 2017 Asian 5,685 2,941 52% Black 1,291 698 54% Hispanic 1,762 955 54% White 17,803 10,053 56% 2 or More 1,206 631 52% Other 56 29 52% Not Indic 1,358 727 54% WN 2018 Asian 5,501 2,746 50% Black 1,252 683 55% Hispanic 1,698 908 53% White 16,924 9,220 54% 2 or More 1,155 599 52% Other 54 26 48% Not Indic 1,268 673 53% FA 2018 Asian 6,047 3,063 51% Black 1,315 748 57% Hispanic 1,972 1,051 53% White 17,525 9,794 56% 2 or More 1,346 702 52% Other 49 23 47% Not Indic 1,472 810 55% WN 2019 Asian 5,829 3,137 54% Black 1,268 766 60% Hispanic 1,899 1,099 58% White 16,604 9,738 59% 2 or More 1,302 745 57% Other 46 22 48% Not Indic 1,407 792 56% 534 College & Research Libraries May 2024 TABLE A.7 Percentage of Undergraduate Students Associated with EZproxy Sessions by School, FA16–WN19 Academic Term School Enrolled Students EZproxy Session % ≥ 1 EZproxy Session FA 2016 Architecture 145 65 45% Art and Design 495 356 72% Business Administration 1,673 890 53% Dental Hygiene 111 77 69% Education 112 66 59% Engineering 6,078 2,736 45% Information 208 123 59% Joined Degree Program 10 7 70% Kinesiology 946 698 74% Literature, Science & the Arts 17,306 10,395 60% Music, Theater & Dance 732 447 61% Nursing 705 626 89% Pharmacy 14 11 79% Public Policy 147 108 73% WN 2017 Architecture 140 71 51% Art and Design 462 249 54% Business Administration 1,639 746 46% Dental Hygiene 107 63 59% Education 112 53 47% Engineering 5,909 1,958 33% Information 186 89 48% Joined Degree Program 8 5 63% Kinesiology 918 576 63% Literature, Science & the Arts 16,400 8,614 53% Music, Theater & Dance 700 402 57% Nursing 685 512 75% Pharmacy 14 7 50% Public Policy 128 89 70% FA 2017 Architecture 155 82 53% Art and Design 497 363 73% Business Administration 1,773 869 49% Dental Hygiene 112 79 71% Education 120 46 38% Engineering 6,409 2,666 42% Information 253 147 58% Joined Degree Program 12 8 67% Kinesiology 976 627 64% Longitudinal Associations 535 TABLE A.7 Percentage of Undergraduate Students Associated with EZproxy Sessions by School, FA16–WN19 Academic Term School Enrolled Students EZproxy Session % ≥ 1 EZproxy Session Literature, Science & the Arts 17,160 9,942 58% Music, Theater & Dance 747 495 66% Nursing 667 516 77% Pharmacy 42 19 45% Public Health 85 72 85% Public Policy 153 103 67% WN 2018 Architecture 153 107 70% Art and Design 481 356 74% Business Administration 1,757 760 43% Dental Hygiene 109 82 75% Education 118 40 34% Engineering 6,150 2,571 42% Information 214 122 57% Joined Degree Program 12 7 58% Kinesiology 951 594 62% Literature, Science & the Arts 16,294 9,034 55% Music, Theater & Dance 715 509 71% Nursing 636 487 77% Pharmacy 42 25 60% Public Health 84 72 86% Public Policy 136 89 65% FA 2018 Architecture 181 119 66% Art and Design 556 396 71% Business Administration 1,826 753 41% Dental Hygiene 103 71 69% Education 131 60 46% Engineering 6,649 2,755 41% Information 302 135 45% Joined Degree Program 11 9 82% Kinesiology 962 617 64% Literature, Science & the Arts 17,262 9,918 57% Music, Theater & Dance 743 524 71% Nursing 632 543 86% Pharmacy 56 33 59% Public Health 158 130 82% Public Policy 154 128 83% WN 2019 Architecture 181 124 69% 536 College & Research Libraries May 2024 Notes 1. Krista M. Soria, Jan Fransen, and Shane Nackerud, “Library Use and Undergraduate Student Outcomes: New Evidence for Students’ Retention and Academic Success,” portal: Libraries and the Academy 13, no. 2 (2013): 147–64, https://doi.org/10.1353/pla.2013.0010. 2. Laurie Alexander, Doreen R. Bradley, and Kenneth J. Varnum, “On the Road to Learning Analytics: The University of Michigan Library’s Experience with Privacy and Library Data,” in Using Digital Analytics for Smart Assessment, ed. Tabatha Farney (Chicago: American Library Association, 2018), 83–93. 3. T. D. Wilson, “Models in Information Behaviour Research,” Journal of Documentation 55, no. 3 (1999): 249–70, https://doi.org/10.1108/EUM0000000007145; J. David Johnson, “Cancer-related information seeking.” Health Communication (Cresskill, N.J.: Hampton Press, 1997); Thomas D. Wilson, “Information Behavior Models,” in Encyclopedia of Library and Information Science, 4th Edition (CRC Press, 2017), 2086-2093. 4. Marcia J. Bates, “Information Behavior,” in Encyclopedia of Library and Information Science, 4th Edition (CRC Press, 2017), 2074-2085. 5. Bates, “Information Behavior,”; Lotta Haglund and Per Olsson, “The Impact on University Libraries of Changes in Information Behavior among Academic Researchers: A Multiple Case Study,” The Journal of Academic Librarianship 34, no. 1 (2008): 52–9, https://doi.org/10.1016/j.acalib.2007.11.010; Xi Niu and Bradley M. Hemminger. “A Study of Factors that Affect the Information-seeking Behavior of Academic Scientists,” Journal of the American Society for Information Science and Technology 63, no. 2 (2012): 336–53, https://doi.org/10.1002/asi.21669. 6. Polly A. Graham, Sarah Socorro Hurtado, and Robert M. Gonyea, “The Benefits of Living on Campus: Do Residence Halls Provide Distinctive Environments of Engagement?” Journal of Student Affairs Research and Practice 55, no. 3 (2018): 255–69, https://doi.org//10.1080/19496591.2018.1474752. 7. Ruth N. López Turley and Geoffrey Wodtke. “College Residence and Academic Performance: Who Benefits from Living on Campus?” Urban Education 45, no. 4 (2010): 506–32, https://doi.org/10.1177/0042085910372351. 8. Lauren T. Schudde, “The Causal Effect of Campus Residency on College Student Retention,” Review of Higher Education 34, no. 4 (2011): 581–610, https://doi.org/10.1353/rhe.2011.0023. 9. Clare Huhn, “The “Housing Effect” on First-Year Outcomes,” (Madison, WI: Academic Planning and Analysis, Office of the Provost, University of Wisconsin-Madison, 2006). 10. Ibid. 11. Jonathan M. Turk and Manuel S. González Canché, “On-Campus Housing’s Impact on Degree Comple- TABLE A.7 Percentage of Undergraduate Students Associated with EZproxy Sessions by School, FA16–WN19 Academic Term School Enrolled Students EZproxy Session % ≥ 1 EZproxy Session Art and Design 524 381 73% Business Administration 1,799 740 41% Dental Hygiene 101 70 69% Education 126 54 43% Engineering 6,313 2,847 45% Information 260 122 47% Joined Degree Program 10 7 70% Kinesiology 954 678 71% Literature, Science & the Arts 16,409 10,030 61% Music, Theater & Dance 717 515 72% Nursing 607 475 78% Pharmacy 55 36 65% Public Health 157 116 74% Public Policy 142 104 73% https://doi.org/10.1353/pla.2013.0010 https://doi.org/10.1108/EUM0000000007145 https://doi.org/10.1016/j.acalib.2007.11.010 https://doi.org/10.1002/asi.21669 https://doi.org/10.1080/19496591.2018.1474752 https://doi.org/10.1177/0042085910372351 https://doi.org/10.1353/rhe.2011.0023 Longitudinal Associations 537 tion and Upward Transfer in the Community College Sector: A Comprehensive Quasi-Experimental Analysis,” The Journal of Higher Education 90, no. 2 (2019): 244–71, https://doi.org/10.1080/00221546.2018.1487755. 12. Denise Balfour Simpson and Dana Burnett, “Commuters Versus Residents: The Effects of Living Arrange- ment and Student Engagement on Academic Performance,” Journal of College Student Retention: Research, Theory & Practice 21, no. 3 (2017): 286–304, https://doi.org/10.1177/1521025117707516. 13. Laura Robinson et al., “Digital Inequalities and Why They Matter.” Information, Communication & Society 18, no. 5 (2015): 569–82, https://doi.org/10.1080/1369118X.2015.1012532; Xinzhi Zhang et al. “Big Data Science: Op- portunities and Challenges to Address Minority Health and Health Disparities in the 21st Century,” Ethnicity & disease 27, no. 2 (2017): 95–106, https://doi.org/10.18865/ed.27.2.95. 14. Paul F. Cleary, Glenn Pierce, and Eileen M. Trauth, “Closing the Digital Divide: Understanding Racial, Ethnic, Social Class, Gender and Geographic Disparities in Internet Use among School Age Children in the United States,” Universal Access in the Information Society 4, no. 4 (2006): 354–73, https://doi.org/10.1007/s10209-005- 0001-0; Linda A. Jackson et al., “Does Home Internet Use Influence the Academic Performance of Low-income Children?” Developmental Psychology 42, no. 3 (2006): 429–35, https://doi.org/10.1037/0012-1649.42.3.429. 15. Elizabeth F. Farrell, “Among Freshmen, a Growing Digital Divide,” The Chronicle of Higher Education Feb- ruary 2, 2005, A32-; Steve Jones, Camille Johnson-Yale, Sarah Millermaier, and Francisco Seoane Pérez, “U.S. College Students’ Internet Use: Race, Gender and Digital Divides,” Journal of Computer-Mediated Communication 14, no. 2 (2009): 244–64, https://doi.org/10.1111/j.1083-6101.2009.01439.x. 16. Angelina KewalRamani et al., Student Access to Digital Learning Resources outside of the Classroom. NCES 2017-098. Edited by American Institutes for Research. National Center for Education Statistics. Washington, DC: U.S. Department of Education, 2018. 17. Lauren Musu, “The Digital Divide: Differences in Home Internet Access,” in NCES Blog, edited by NCES Blog Editor. Washington, DC: National Center for Education Statistics, October 18, 2018, https://nces.ed.gov/blogs/ nces/post/the-digital-divide-differences-in-home-internet-access. 18. Ana-Paula Correia, “Healing the Digital Divide During the COVID-19 Pandemic,” Quarterly Review of Distance Education 21, no. 1 (2020): 13–21. 19. John Lai and Nicole O. Widmar, “Revisiting the Digital Divide in the COVID-19 Era,” Applied Economic Perspectives and Policy 43, no. 1 (2021): 458–64, https://doi.org/10.1002/aepp.13104. 20. Fawzia Reza, “COVID-19 and Disparities in Education: Collective Responsibility Can Address Inequities,” Knowledge Cultures 8, no. 3 (2020): 68–75, https://doi.org/10.22381/KC83202010; Megan Kuhfeld et al., “Projecting the Potential Impact of COVID-19 School Closures on Academic Achievement,” Educational Researcher 49, no. 8 (2020): 549–65, https://doi.org/10.3102/0013189X20965918. 21. Terence Day et al. “The Immediate Impact of COVID-19 on Postsecondary Teaching and Learning,” The Professional Geographer 73, no. 1 (2021): 1–13, https://doi.org/10.1080/00330124.2020.1823864. 22. Ibid. 23. Ibid.; Dipti Mehta and Xiaocan Wang, “COVID-19 and Digital Library Services – A Case Study of a Uni- versity Library,” Digital Library Perspectives 36, no. 4 (2020): 351–63, https://doi.org/10.1108/DLP-05-2020-0030. 24. Ana-Paula Correia, “Healing the Digital Divide During the COVID-19 Pandemic,” Quarterly Review of Distance Education 21, no. 1 (2020): 13–21. 25. Kyong Hee Chee, Nathan W. Pino, and William L. Smith, “Gender Differences in the Academic Ethic and Academic Achievement *,” College Student Journal 39, no. 3 (September 2005): 604+; Michael Sheard, “Hardiness Commitment, Gender, and Age Differentiate University Academic Performance,” British Journal of Educational Psychology 79, no. 1 (2009): 189–204. https://doi.org/10.1348/000709908X304406; Arna Kristín Harðardóttir, Sigurður Guðjónsson, Inga Minelgaite, and Kári Kristinsson, “Ethics as Usual? Gender Differences in Work Ethic and Grades,” Management: Journal of Contemporary Management Issues 24, no. 2 (2019): 11–21, https://doi.org/10.30924/ mjcmi.24.2.2. 26. Michael J. Stebleton and Krista M. Soria, “Breaking Down Barriers: Academic Obstacles of First Genera- tion Students at Research Universities,” The Learning Assistance Review 17, no. 2 (2012): 7+. 27. Xi Wang, Minhao Dai, and Robin Mathis, “The Influences of Student- and School-level Factors on Engi- neering Undergraduate Student Success Outcomes: A Multi-level Multi-school Study,” International Journal of STEM Education 9, no. 1 (2022): 23, https://doi.org/10.1186/s40594-022-00338-y. 28. John M. Trussel and Lisa Burke-Smalley, “Demography and Student Success: Early Warning Tools to Drive Intervention,” Journal of Education for Business 93, no. 8 (2018): 363–72, https://doi.org/10.1080/08832323.201 8.1496893; Julian R. Betts and Darlene Morell, “The Determinants of Undergraduate Grade Point Average: The Relative Importance of Family Background, High School Resources, and Peer Group Effects,” The Journal of Hu- man Resources 34, no. 2 (1999): 268–93, https://doi.org/10.2307/146346. 29. Siu-Man Raymond Ting and Tracy L. Robinson, “First-year Academic Success: A Prediction Combining https://doi.org/10.1080/00221546.2018.1487755 https://doi.org/10.1177/1521025117707516 https://doi.org/10.1080/1369118X.2015.1012532 https://doi.org/10.18865/ed.27.2.95 https://doi.org/10.1007/s10209-005-0001-0 https://doi.org/10.1007/s10209-005-0001-0 https://doi.org/10.1037/0012-1649.42.3.429 https://doi.org/10.1111/j.1083-6101.2009.01439.x https://nces.ed.gov/blogs/nces/post/the-digital-divide-differences-in-home-internet-access https://nces.ed.gov/blogs/nces/post/the-digital-divide-differences-in-home-internet-access https://doi.org/10.1002/aepp.13104 https://doi.org/10.22381/KC83202010 https://doi.org/10.3102/0013189X20965918 https://doi.org/10.1080/00330124.2020.1823864 https://doi.org/10.1108/DLP-05-2020-0030 https://doi.org/10.1348/000709908X304406 https://doi.org/10.30924/mjcmi.24.2.2 https://doi.org/10.30924/mjcmi.24.2.2 https://doi.org/10.1186/s40594-022-00338-y https://doi.org/10.1080/08832323.2018.1496893 https://doi.org/10.1080/08832323.2018.1496893 https://doi.org/10.2307/146346 538 College & Research Libraries May 2024 Cognitive and Psychosocial Variables for Caucasian and African American Students,” Journal of College Student Development 39, no. 6 (1998): 599–610; Betts and Morell, “The Determiniants of Undergraduate Grade Point Average”; Andrew P. Barkley and Jerry J. Forst, “The Determinants of First-Year Academic Performance in the College of Agriculture at Kansas State University, 1990–1999,” Journal of Agricultural and Applied Economics 36, no. 2 (2004): 437–48, https://doi.org/10.1017/S1074070800026729. 30. Forrest E. Huffman, “Student Performance in an Undergraduate Advanced Real Estate Course: Real Estate Majors vs. Finance Majors,” Journal of Real Estate Practice and Education 14, no. 2 (2011): 111–23, https://doi.org/10. 1080/10835547.2011.12091693. 31. Wilson, “Models in Information Behaviour”; Johnson, Cancer-related information seeking. 32. Ibid.; Johnson, J. David. Cancer-related information seeking. Cresskill, N.J.: Health Communication. Cresskill, N.J.: Hampton Press, 1997. 33. Jerry Chih-Yuan Sun and Susan E. Metros, “The Digital Divide and Its Impact on Academic Perfor- mance,” US-China Education Review A 1, no. 2 (2011): 153–61; Joanna Goode, “The Digital Identity Divide: How Technology Knowledge Impacts College Students,” New Media & Society 12, no. 3 (2010): 497–513, https://doi. org/10.1177/1461444809343560; Laura Robinson, Øyvind Wiborg, and Jeremy Schulz, “Interlocking Inequalities: Digital Stratification Meets Academic Stratification,” American Behavioral Scientist 62, no. 9 (2018): 1251–72, https:// doi.org/10.1177/0002764218773826; KewalRamani et al., Student Access to Digital Learning Resources. 34. Penny Beile, Kanak Choudhury, Rachel Mulvihill, and Morgan Wang, “Aligning Library Assessment with Institutional Priorities: A Study of Student Academic Performance and Use of Five Library Services,” Col- lege & Research Libraries 81, no. 3 (2020): 435–58, https://doi.org/10.5860/crl.81.3.435; Soria, Fransen, and Nackerud, “Library Use and Undergraduate Student Outcomes.” 35. (Soria, Fransen, and Nackerud 2014, 2016; Wong and Webb 2011); Krista M. Soria, Jan Fransen, and Shane Nackerud, “Stacks, Serials, Search Engines, and Students’ Success: First-Year Undergraduate Students’ Library Use, Academic Achievement, and Retention,” The Journal of Academic Librarianship 40, no. 1 (2014): 84–91, https:// doi.org/10.1016/j.acalib.2013.12.002; Krista M. Soria, Jan Fransen, and Shane Nackerud, “Beyond Books: The Ex- tended Academic Benefits of Library Use for First-Year College Students,” College & Research Libraries 78, no. 1 (2016): 8–22, https://doi.org/10.5860/crl.78.1.8; Shun Han Rebekah Wong and T. D. Webb, “Uncovering Meaningful Correlation between Student Academic Performance and Library Material Usage,” College & Research Libraries 72, no. 4 (2011): 361–70, https://doi.org/10.5860/crl-129. 36. Stata Statistical Software: Release 16. StataCorp LLC, College Station, TX. 37. Megan Oakleaf et al., Connecting Libraries and Learning Analytics for Student Success, (Syracuse, NY: Syracuse University, 2020), https://library.educause.edu/-/media/files/library/2020/12/cllassfinalwhitepaper.pdf. https://doi.org/10.1017/S1074070800026729 https://doi.org/10.1080/10835547.2011.12091693 https://doi.org/10.1080/10835547.2011.12091693 https://doi.org/10.1177/1461444809343560 https://doi.org/10.1177/1461444809343560 https://doi.org/10.1177/0002764218773826 https://doi.org/10.1177/0002764218773826 https://doi.org/10.5860/crl.81.3.435 https://doi.org/10.1016/j.acalib.2013.12.002 https://doi.org/10.1016/j.acalib.2013.12.002 https://doi.org/10.5860/crl.78.1.8 https://doi.org/10.5860/crl-129 https://library.educause.edu/-/media/files/library/2020/12/cllassfinalwhitepaper.pdf