Righteous, Reveler, Achiever, Bored: A Latent Class Analysis of First-Year Student Involvement

Abstract

Using the Wabash National Study on Liberal Arts Education and a latent class analysis of 28 outside-the-classroom activities and behaviors, we developed a typology of outside-the-classroom student engagement during the first year of college. We find ten classes of student involvement: academic artist, party athlete, serious athlete, conventional non-worker, disengaged, maximizer, moderate worker, detached partier, involved partier, and religious. Next, we examine the relationship between latent classes and students’ characteristics through a multinomial logistic regression analysis. Students reporting as first-generation or racially minoritized are overrepresented in the disengaged and involved partier classes. We found an overrepresentation of White students across all party classes. Students reporting as female were likelier to be members of the religious, moderate worker, and disengaged classes and not to be members of the party classes. Federal grant recipients were likelier to be in the academic artist and moderate worker classes. We discuss other sociocultural, economic, and academic relationships in the paper. Next, we explore the relationship of latent class to academic and developmental outcomes. We find academic artists as the only class with a significant positive relationship across the seven dependent measures. Involved partier, moderate worker, and religious classes have positive relationships with at least five dependent measures. The detached partier and party athlete classes have the lowest first-year GPAs of all latent classes. Finally, we discuss the relationships of latent classes, related institutional policy implications, and directions for future research.

Introduction

A student with a 15-credit-hour course load has 153 hours throughout a seven-day week where they are not in a classroom setting. Suppose we were to believe the typical Hollywood zeitgeist of the collegiate tropes of Animal House, Old School, and the Neighbors. In that case, we might expect to find most students on a pathway to drinking, socializing, and relaxing.¹ Higher education research echoes this fear in titling books Paying for the Party and Academically Adrift (Armstrong & Hamilton, 2013; Arum & Roksa, 2011). Nevertheless, even in Paying for the Party, the authors include three other involvement pathways besides partying in their typology. Qualitative analyses of other higher education institutions (HEIs) highlight the activities which connect the on-campus and surrounding communities (Silver, 2020b; Stuber, 2011). Engaging in one form of outside-the-classroom activity (OCA) may lead to engagement in another OCA and vice versa.

How students spend or do not spend their time outside class is a concern due to its connection to academics and how it may influence access to internships, graduate school, and high-status entry-level employment (Posselt, 2016; Rivera, 2011, 2012). Interacting with peers, student affairs professionals, and faculty members remains a catalyst for student growth (Pike & Kuh, 2005). Others found that both involvement and interactions outside the classroom foster better outcomes for first-generation students (Pascarella et al., 2004) and marginalized groups (Strayhorn, 2018). Quantifying engagement in OCAs tends to be problematic since it tends to focus on singular activities such as specific best practices, the total number of drinks, or co-curricular involvement as a catch-all category of the total amount of hours spent in co-curricular activities or studying (An et al., 2017; Baker, 2008; Kilgo et al., 2015).

The fundamental problem with these kinds of assessments is that they ignore other confounding factors of other concurrent OCAs in which a student may be involved while in college. Ignoring other concurrent involvements in various OCAs belies the intermingling and interconnection of OCAs. For example, when assessing the positive effects of participating in faculty research, Kilgo et al.’s (2015) methods control for Greek membership and the hours spent working on- and off-campus. Suppose these students also spend additional time interacting with faculty and studying at high rates while spending little time socializing and relaxing as a trade-off. In that case, these concurrent OCAs also need to be controlled. Similarly, Pascarella et al. (2004) indicate the number of hours spent in co-curricular activities and their ties to positive academic outcomes. However, they need to control for other OCAs how students could spend that time in college.

Latent class analysis (LCA) and less mathematically complex cluster analyses circumvent these issues and can account for students who engage in overlapping activities (Magidson & Vermunt, 2002). These analyses allow researchers to input many OCAs and elucidate a typology of student groupings (i.e., classes) of similar involvements across many OCAs through complex algorithms. In doing so, LCAs can identify a class of students oriented to high involvement across academic OCAs or another class of students oriented towards social and leisurely OCAs. After assigning students to these predicted classes, researchers can apply other statistical techniques to understand the likelihood of students being in a latent class according to background characteristics. Finally, one can assess how being in these classes relates to student outcomes when controlling for those background characteristics.

In this study, we use over 8,200 students at 44 higher education institutions (HEIs) from the Wabash National Study on Liberal Arts Education (WNS) to answer the following questions:

RQ₁: What are the latent classes of student involvement in OCAs during the first year of college?

RQ₂: What sociocultural, economic, and academic relationships are associated with students sorting into different latent classes?

RQ₃: What are the latent classes’ associations with academic and developmental outcomes?

In answering these questions, we developed a typology that classified students according to ten different classes of student involvement during the first year of college. We document significant associations for students in each class by sociocultural, economic, and academic characteristics of students. Finally, we show how the differences relate to various first-year educational, social, and psychological outcomes.

Literature Review

Collegiate Outcomes and Student Involvement

College students’ time outside the classroom is an often-discussed theoretical feature of higher education (Astin, 1984; Strayhorn, 2018; Tinto, 1993). Central to the theories, campus involvement affects students by helping or hurting them. In particular, Astin (1984) argues a three-stage process in his theory of student involvement. First, a student’s sociodemographic, economic, academic, and cultural characteristics serve as inputs in determining how and why students become involved in the college environment. The second stage encompasses a student’s college experiences, which Astin refers to as the environment. The final stage includes the outcomes of a student’s college experience. The model has five assumptions. First, involvement requires both physical and psychological energy. Second, involvement is continuous, and the amount of energy put in by students varies. Third, aspects of the involvement can be both qualitative and quantitative. Fourth, any gains from involvement are proportional to the quality and quantity of involvement. Finally, involvement and academic performance are correlated.

When applying these theories, many quantitative researchers pick a particular activity or hours spent in an activity and perform quantitative analyses to understand this activity’s relationship with various academic, psychological, and behavioral outcomes (i.e., Chen & Lingo, 2020; Kilgo et al., 2015; Pascarella et al., 2004). A problem with this kind of quantitative analysis is that it needs to include concurrent involvement in the college environment. For example, a university could provide a student-athlete with intensive tutoring and insufficient time to socialize, relax, and drink. Suppose subsequent analyses indicate student-athletes have higher GPAs than other students. In that case, one may question whether the outcome is caused by the student being a student-athlete or because of the student-athletes intensive tutoring or the limited time for drinking or socializing.

Qualitative literature points to how this is indeed the case. In examining a large selective state school and a very selective small liberal arts college (SLAC), Stuber (2011) finds that many working-class students are involved in activities associated with work and sociocultural affinity groups. However, upper-middle-class students are more likely to pursue wide-scale OCA involvement and take on leadership positions. Although Jack’s (2016, 2019) work focuses on how selective institutions fail to meet the needs of traditionally marginalized African American/Black and Latino/a/x students, the scholarship showcases how students’ OCAs interweaved with one another. Finally, Silver’s (2020a, 2020b, 2020c) research series examines student OCA involvement at a large public university. The main focus of Silver’s series of manuscripts concerns race/ethnicity, gender, and class and how such constructs shape involvement in OCAs. Throughout the series, Silver’s analyses indicate that students are often involved in multiple OCAs that feed into the involvement of one another. Although neither of the three authors presents students’ typologies, they provide evidence of how activities are intermingled and intertwined throughout college. The push and pull of these activities and behaviors force students to make trade-offs on time spent studying, working, or socializing (Greene & Maggs, 2015).

Qualitative Typologies of Involvement

There is a history of qualitative and quantitative research concerning the discussion of student typologies going back to the 1960s (Hu et al., 2011; Kuh et al., 2000). The underlying idea behind this research is that students sort themselves into different involvement groups. A recent and well-known qualitative example of this type of research is Armstrong and Hamilton’s (2013) Paying for the Party. In the book, the authors provide a narrative of a dorm floor of first-year students at a Research 1 university in the Midwest. Their typology describes four pathways of involvement (social isolation, striving, achievement, and party) that emerge during the first year of college. Students on the party pathway spent time socializing, drinking, and in the Greek system. Party pathway students sacrificed their academics and original academic plans to maintain their social ties and beauty. Students on the achievement pathway were generally less affluent than those on the party pathway but more affluent than those on the uninvolved path. The achievers’ behaviors focused on academics and activities relevant to their career pathway. While a few strivers were cream skimmed by university programs to help disadvantaged students do well academically, most strivers either dropped out or transferred institutions. Finally, the authors maintained that socially isolated students did not fit into any pathway. The authors viewed social isolation among these students as a protective mechanism. Social class differences affected how students navigated and harnessed the advantages and disadvantages of being on a particular pathway. However, the book highlighted how OCAs interlink.

Cluster Analysis Derived Typologies of Student Involvement

In the quantitative realm of student typology research, some turn to cluster analyses to examine student involvement among college students (Hu & McCormick, 2012; Kuh et al., 2000). Researchers first use a factor analysis with many variables in these cluster analyses to create broad-based standardized factor variables of student involvement. Hu and McCormick (2012) merged 42 survey items related to student perceptions, behaviors, and experiences into five standardized factors, while Kuh et al. (2000) merged 126 survey items into eight standardized factors. Using this form of distance-based cluster analysis, researchers use the students’ scores across these factors to cluster students into groups of best fit (Magidson & Vermunt, 2002). The process creates clusters of students with similar involvement levels across the standardized factors. Kuh et al. (2000) found 10 clusters of involvement (disengaged, recreator, socializer, collegiate, scientist, individualist, artists, grind, intellectuals, and conventionalists) with students from the 1979 College Student Experiences survey. Hu and McCormick (2012) found seven clusters of involvement (academics, unconventionals, disengaged, collegiate, maximizers, grinds, and conventionals) with only the 2006 wave of the WNS. All three found different typologies in the various selected outcomes among the groupings.

An issue of the two k-means cluster analyses is the requirement of standardized variables to keep clusters dominated by variables with significant differences in variance from emerging (Magidson & Vermunt, 2002). The use of factors and the use of standardized variables can be problematic. For example, merging variables from many constructs can make it difficult to discern where differences in specific OCAs exist between classes or clusters. For example, there are 26 variables in the academics factor variable from Kuh et al. (2000). Latent class and profile analyses overcome the flaw of standardized variables because their “clustering solutions are invariant of transformation on the variables” (Magidson & Vermunt, 2002; p. 41).

LCA Created Typologies of Student Involvement

Besides allowing for categorical variables, LCAs are better because they use a probabilistic framework, while cluster analyses use a less complex distance-based methodology (Magidson & Vermunt, 2002). The LCA still allows us to meet the theoretical understanding that the activity involvement of college students is interrelated with one another. Because of this interconnected nature and complexity of behaviors, LCAs are one approach that allows researchers to help understand individuals from the totality of behaviors instead of the sum of the parts (Magnusson, 2000, 2003; Von Eye & Bergman, 2003). The method uses structural equation modeling to observe heterogeneity in a population on various activities and behaviors instead of describing the variability of a particular activity (Lanza & Cooper, 2016; Von Eye & Bergman, 2003). Latent class analyses act as a holistic, person-centered approach to classifying students into "mutually exclusive and exhaustive subgroups" that are comprised of individuals (students) engaging in similar behaviors (OCAs) (Lanza & Cooper, 2016, p. 59). As qualitative literature demonstrates how students engage in OCAs, students’ behaviors interact and overlap. The LCA models how student involvement is more a pathway than a singular construct of only one activity. Once the analysis is complete, researchers compare the classes’ mean levels of variables. Researchers assign each class a name through this comparison of the mean levels. For example, students with the highest mean levels spent in academic OCAs may be labeled academics.

There are several examples of researchers using LCAs to understand student involvement. Quadlin and Rudel (2015) use students’ time throughout the week (e.g., studying, sleeping, or co-curricular activities) in their LCA using the 1999 National Longitudinal Survey of Freshman. The researchers identified three involvement classes (serious students, inactive, and socially engaged). Finally, the authors focus on how these classes relate to receiving student loans and student loan debt, finding that socially engaged students were less likely to have loans and loan debt than the other classes. Fosnacht et al. (2018) also use time-use data from a 2014 and 2015 National Survey of Student Engagement sample to identify four classes (balanced, involved, partiers, and parents).² The researchers found significant differences in how students sort into these classes by gender, race/ethnicity, SAT score, age, academic major, Greek life membership, on-campus residence, and student-athlete. In subsequent regression analyses on developmental outcomes, authors found the involved class better off and the partier class worse off than balanced students at the end of the first year. Meanwhile, all four classes did not differ by GPA. Finally, Willoughby et al. (2020) used a LCA with a single institution dataset in Canada and found four classes (good, sensation seeking, struggling, and club involved). The struggling class had a variety of worse academic and employment outcomes, and the sensation class was more likely to drop out.

Collectively, the three studies are not without concerns. One significant issue of Willoughby et al. and’s (2020) LCA is its inclusion of variables for demographic, academic, and psychological measurements. Similar to the other cluster or LCAs detailed in this paper, we view demographic and educational characteristics as part of how students sort into these portfolios of involvement. An issue with both Quadlin and Rudel (2015) and Fosnacht et al. (2018) is that both rely on time-use variables. The time-use variables face a similar limitation to the factor variables in the cluster analyses in that they do not distinguish how students are involved in those hours. A student who spends twenty hours in co-curricular activities surrounding the arts may be distinctively different from a student who spends those co-curricular hours in athletics. Critical information may be lost in the LCA using aggregated coding. For example, past latent class analyses of high school students indicate activity-specific pathways involving athletics, performance arts, and religion (Feldman & Matjasko, 2007; Kort-Butler & Martin, 2015; Linver et al., 2009). Other research indicates that the groupings under a typology tend to follow students from high school into college (Crabbe et al., 2019). For within reason, using constructs surrounding athletics, performing arts, and religion with college OCAs may reveal similar latent classes heavily involved in these program areas. LCAs using hourly or factor variables may miss classes centered on OCAs involving athletics, performing arts, or relgion.

Importance of Typologies of Student Involvement

A few studies suggest that specific student typologies of OCA involvement are required to access elite internships, entry-level jobs, and graduate schools. Social fit, modeled through OCAs, provides a signal as to an ideal candidate. Rivera’s (2011, 2012) interview of 120 human resources directors from law, investment, and consulting firms found that the postsecondary institution prestige of applicants served as the first cut. Co-curricular participation is the second-cut point for determining whether to interview an applicant over GPA. The representatives found those without a robust portfolio of OCAs boring. Most firms viewed the pathway of a student-athlete as an ideal candidate for their firms. Posselt’s (2016) research of admissions committees at ten prestigious postsecondary institutions highlights how the graduate admission committees preferred a student on a research pathway with research-centered OCA involvement. Although the data focuses on elite areas of employment and elite graduate schools, the studies indicate that social identity fostered through OCA involvement can shape a student’s post-college trajectory.

Current Study

When researchers explore the associated effects of participating in a singular OCA in college, they often need to pay more attention to students’ concurrent involvement in other OCAs. Ignoring concurrent involvement is vital because qualitative researchers note the interdependent and interconnected nature of OCAs (Jack, 2016; Silver, 2020a, 2020b, 2020c; Stuber, 2011). One pair of researchers goes so far as to detail how overlapping involvements result in student involvement typology (Armstrong & Hamilton, 2013). A significant problem in this line of research is that the authors are limited to one or two often selective or highly selective institutions. Within research using latent class and cluster analyses, previous authors’ variables lack the specificity needed to draw inferences about student OCAs during the first year of college because their constructs focus on only the number of hours or a small set of activities. These narrow constructs may result in the analyses missing other latent classes specific to certain OCAs.

We are using a sample of 44 four-year institutions with a range of selectivity, research intensity, and enrollment patterns from the WNS. Drawing from various institutions ensures that we know different latent classes, which we may or may not find at other institutions. To overcome the issues of variable aggregation found in past LCAs and cluster analyses, we provide twenty-three activity-specific constructs (e.g., student-athlete, member of a religious congregation or student group, and participating in a faculty research project) alongside five constructs of hourly involvement (e.g., hours spent studying and hours spent working) in OCAs. After completing the LCA, we use the WNS’s rich student characteristics to understand various sociocultural, demographic, economic, and academic connections to OCA class membership. Finally, we assess the associations between being a latent class membership with various developmental and academic outcomes.

Data and Methods

We derive our analyses from Astin’s model of student involvement (see Fig. 1). To answer RQ1, we conducted a latent class analysis to determine the environments (e.g., classes) students sort into the first year. In RQ2, we examine the inputs and how they relate to student involvement. Finally, RQ3 examines how these latent classes (e.g., environments) may shape student outcomes.

Fig. 1

Adapted from Astin’s (1984) Theory of Student Involvement

Theoretical model.

Dataset

The WNS was a 49-campus longitudinal study of college students’ pre-college and college experiences that asked about their collegiate engagement and provided questionnaires to assess cognitive and non-cognitive outcomes. The original questionnaire included more than 15,000 students from three cohorts enrolling in college during the fall of 2006, 2007, and 2008. Although the colleges differed in selectivity, control, student enrollment, and geographic region, they were similar in a steadfast commitment to liberal arts education (i.e., University of Michigan-Ann Arbor, Vassar College, Delaware State University).³ The study assessed students during the fall of their first year, spring of their first year, and spring of their fourth year.

Due to this study’s focus on first-year involvement, the analytical sample only contains students who answered the first two surveys, fall and spring of first-year. The first survey included questions about students’ sociodemographic characteristics, attitudes and behaviors, social and psychological outcomes, and participation in various aspects of their high school careers. The second survey measured in-college behaviors and non-cognitive and cognitive tests and included transcripts from colleges and institutionally reported student retention and GPA. Because of community colleges’ small institutional sample size (three) and the desire to focus on four-year students for comparisons, we do not include the three community colleges. Two four-year institutions did not collect data on OCAs, so we did not have their students in the study.

For RQ₁, the doLCA program we use in the LCA includes any student with at least one completed data point and imputes missing values in the LCA. Once Stata conducts the analysis, we assign students their class based on their predicted probability. We do not include students without a predicted probability in the investigation. The method leaves the final sample size for RQ₁ at 8,291 students attending 44 institutions.⁴

For RQ₂, four students were missing the gender construct. We did not include the students in the RQ₂ analysis. A total of five variables were missing 1.93% to 3.34% of values. We ran Little’s (1988) missing completely at random (MCAR) test with the predictor variables to determine the need for multiple imputations. The MCAR test determines whether there is anything systematic in the missing and observed values in the dataset. If the data is MCAR, list-wise deletion is preferable to avoid bias in the results (Raghunathan, 2004). The test indicates that the data is not MCAR. Therefore, we follow the suggestion from guidance from Pedersen et al., (2017) and Sinharay et al., (2001) and use multiple imputations by chained equations (MICE) of missing independent variables for RQ₂ for a total of 20 imputations (Allison, 2001). Multiple imputations provide valid estimates of associations based on information from the data and handle that the data are both missing at random (MAR) and missing not at random (MNAR) (Pedersen et al., 2017; Sinharay et al., 2001). It is difficult to know whether the missingness mechanism is MAR or MNAR, although multiple imputations under the assumption of MAR when the actual missingness mechanism is MNAR may produce little biased but improved estimates (Sinharay et al., 2001). Pedersen et al. (2017) also suggest that multiple imputations are suitable for handling MCAR and provide accurate variability for each missing value. Given the advantages of multiple imputations, this study used multiple imputations rather than list-wise deletion. The final sample size for RQ₂ is 8,287 students attending 44 institutions.

Research Question 3 addresses seven dependent variables on students’ academic and developmental outcomes. We restrict our dataset to only those students with assigned classes and without missing values for the dependent variables. To determine whether the data was MCAR, we ran Little’s (1988) MCAR test with independent and control variables. The test indicated that the data was not MCAR, and we followed the suggestions from previous literature (Pedersen et al., 2017; Raghunathan, 2004; Sinharay et al., 2001) to impute the missing values with MICE for 20 imputations (Allison, 2001). The final sample size for RQ₃ is 6,697 students attending 39 institutions.

Research Question 1 Variables and Methods

The first research question uses LCA to construct and examine classes of student involvement during the first year of college. We chose any variable from the dataset that implied an activity that potentially occurred outside the collegiate classroom.⁵ The variables used in the LCA reflect potential outside-the-classroom activities or behaviors in the academic, co-curricular, leadership, social, and work experiences while attending college. Activities and behaviors reflecting academic experience include weekly hours studying and participating in an internship or field experience, a faculty research project, a first-year seminar, an honors program, a learning community, and a study group (see Appendix C for the coding variables mean-level and missingness).⁶ The co-curricular-related activities or behaviors include weekly hours of co-curricular involvement and participating or being in or at an arts event, a diversity workshop, a political or social justice event, spirituality-enhancing activities, a student-athlete, a volunteer, or a spiritual or religious group. Leadership activities or behaviors include participating in committee or organization work with faculty members, committee or organization work with student affairs personnel, a leadership position, leadership training, summer orientation staff position, and a non-academic peer educator position. The social activities include the total hours spent relaxing or socializing, weekly binge drinking, daily drinking, hourly weekly exercise, and intramural sports participation. Finally, the work constructs include the hours spent working on- or off-campus and work-study participation.

This study used a LCA to identify a typology of how students spend their time outside of the classroom. While a latent profile analysis uses continuous variables, an LCA uses categorical or binary variables in the study. LCA is a statistical method of using multiple variables to classify individuals from a heterogeneous population into relatively homogeneous and unobserved subgroups (Muthén & Muthén, 2000). An LCA uses observed values on a set of categorical variables. We use Stata 16.1 statistical software and the doLCA plugin to complete the analyses (Lanza et al., 2015; LCA Stata Plugin, 2015). Choosing the number of latent classes to represent the typology of involvement requires a combination of statistical model-fit indicators and theoretical frameworks related to the substantive interpretability of the classes. This study started with a single-class model and then increased latent classes by one each time. Each time we compared the Bayesian Information Criterion (BIC) and Akaike’s Information Criterion (AIC) values to find the model with the lowest BIC and AIC values (Nylund et al., 2007). Additionally, we studied the entropy of each model. Entropy estimates each subject’s probability of assignment to their most likely predicted class and then adds them into one number between 0 and 1. The closer the number is to 1, the better the model identifies predicted classes (Celeux & Soromenho, 1996).

We found several trends while consulting the test statistics (see Table 1). While AIC continued to decrease throughout the models, the BIC increased in the 11-class model. In the 11-class model, entropy decreased from 0.912 to 0.892. Because AIC will often be insensitive to model order and potentially overestimate the number of latent classes (Koehler & Murphree, 1988), both BIC and entropy indicate a wrong direction in the 11-class model. We decided on the 10-class model to best represent our data.

Table 1 Latent class analysis test statistics

After completing the LCA, we use doLCA to produce the predicted mean or probabilities of students in a class to be a part of an OCA. For example, the detached partier class had a 0.105 predicted mean or probability of being in an honors organization. The 0.105 indicates that the probability of the detached partier class being in an honors organization is 10.5%.

Research Question #2 Variables and Methods

This question explores the likelihood of students sorting into latent classes based on sociocultural and academic constructs. The dependent variable is the assigned latent class from the LCA. After completing the LCA, we use Stata to give each student a predicted probability of which class they would most likely be assigned. Following best practices, we assign each student to the profile they most likely belong to (Vermunt & Magidson, 2003). For example, if a student was at a 90% likelihood of being placed in Class A and a 10% likelihood of being placed in Class B, we assigned the student to Class A.

We base our choice of variables on the works of Stuber (2011), Armstrong and Hamilton (2013), Jack (2016), and Silver (2020b). All five authors offer rich examples of how race/ethnicity, gender, parental education, familial monetary wealth, human capital, college major, career goals, and the kind of institution attended may shape how students engage in behaviors and activities outside the classroom. We include variables for gender, race/ethnicity, the highest level of parental education, whether students had a dependent, international student status, English as a second language, student loan recipient, federal Grant recipient, high school GPA, intended major, type of institution attended, and advanced degree aspirations (see Appendix D for a complete list of variables, coding, mean-level, and missingness).

For the analysis, we use a multinomial logistic regression with standard errors clustered by the institution attended to understand how different student characteristics predicted students’ latent class. We then convert the coefficients and report them as average marginal effects (AME). The AMEs indicate the percentage point increase/decrease in the likelihood of a student being a member of a particular class given a one-unit increase of the independent variable relative to its reference variable. For example, the AME for African American/Black students and academic artists is 0.081, and the reference race/ethnicity is White. The results suggest that a student reporting as African American/Black is 8.1 percentage points more likely to be an academic artist than their White peers.

Research Question 3 Variables and Methods

Research Question 3 explores the relationship between students’ ascribed class and academic and developmental outcomes. Our dependent variables from the students’ institutional records are end-of-first-year GPA, second-year retention, and fourth-year retention (see Appendix E for the coding of variables, means, and missingness). The developmental outcomes we explore come from end-of-first-year surveys by the WNS and are the same outcomes as a past typology analysis (Hu & McCormick, 2012). The outcomes include the Socially Responsive Leadership Scale (SRLS), which measures a student’s motivations in leadership for social change (Tyree, 1998); the Ryff Scales of Psychological Well-Being (Ryff), which measures the social, psychological, and health measures of well-being (Ryff, 1989; Ryff & Keyes, 1995); the Need for Cognition Scale that measures a student’s tendency to both engage in and enjoy learning (Caciopppo, 1996); and the Milville-Guzman Universality-Diversity Scale—Short Form (M-GUDS) that measures a student’s awareness and acceptance of differences among people (Miville et al., 1999). All scales are z-scored.

Our independent variables are the ascribed classes of student involvement in RQ₁. Following past cluster and LCAs, the comparison’s disengaged pathway is the reference group. Each model controls for the sociodemographic characteristics of gender, race/ethnicity, parental level of education, student loan recipient, and student federal grant recipient status. We also control for the academic attributes of ACT/SAT score, choice of major, and advanced degree aspirations. Due to contextual differences in the campus setting, we control institutional type. All coding is the same as listed in RQ₂. Finally, we control for student fall scores as a baseline measurement for the analyses of SRLS, M-GUDS, Ryff, and Need for Cognition scales.

Because the WNS nests students within their collegiate institution, multilevel analyses are preferable in exploring outcomes (Luke, 2004). In single-level analyses, un-modeled group-level information would compile into the individual error term (Duncan et al., 1998). The group-level information in the individual error terms could cause a potential correlation between students’ error terms attending the same postsecondary institution. Multilevel analyses overcome this issue by accounting for the group-level differences by examining the variance in outcomes when the independent variables are at the student and institutional levels (Luke, 2004). We use multilevel linear regressions to explore the continuous outcomes of first-year GPA, SRLS, Need for Cognition scale, M-GUDS, and Ryff scale. For the outcome of student retention, we use a multilevel logistic regression. To ease the understanding of coefficients, we converted them to odds ratios.

Limitations

The dataset’s first limitation is sample attrition. Of the roughly 15,000 students sampled in the fall of their first year of college, approximately 8,500 students answered the follow-up survey in the spring. The attrition creates a student sample that is more White, higher in parental education, and better off academically (Bowman, 2012). Another limitation of the WNS is its age. The survey measures used in these analyses are from students interviewed in the fall of 2006, 2007, and 2008 and the spring of 2007, 2008, and 2009, which means administrators had students complete surveys 16 years ago. As colleges and universities continue to evolve in the era of COVID:19, there is a question of how the data would currently apply during and after the pandemic. The WNS’s focus on liberal arts-focused institutions is another issue with the data. With the preponderance of postsecondary institutions surveyed as SLACs, the results may indicate a primarily liberal arts college experience.

In any case, the WNS is the best data set for the analyses for three reasons. First, the WNS includes 28 different out-of-classroom involvement at the end of the first year of college. Unlike other typologies of college students, this allows for a rich set of activities to understand student engagement patterns. Secondly, the WNS supplies a rich set of student-level sociodemographic and academic variables at the onset of college. The richness of the data helps us understand the link between these characteristics and student engagement. A final limitation is not necessarily the dataset but the methodology for RQ₂ and RQ_3. Though the variables precede the dependent measurements by nine months, we cannot identify causal effects. This study aims not to argue a causal framework but to understand what characteristics of students relate to the latent classes or how these latent classes relate to outcomes. This analysis can offer insight into how student affairs professionals can support students navigating a new campus environment and how the classes in the typology relate to outcome measures.

Findings

Research Question 1

The LCA exposes ten distinct classes of student involvement. After examining the mean level differences in OCA involvement, we refer to the classes as academic artists, party athletes, serious athletes, conventional non-workers, disengaged, detached partiers, involved partiers, maximizers, moderate workers, and religious (see Table 2).

Table 2 Latent classes of first year involvement and predicted probabilities

Academic Artists (8.3% of the sample)-The class’s highest studying and study group usage levels are two defining characteristics. Though the class spends a moderate amount of time in co-curricular activities, the class has a predicted mean of nearly 75% for attending arts events often or very often. Additionally, the class showcases the second-highest levels of diversity workshops and political or social justice event participation. The class is second to only maximizers in many leadership variables, including orientation staff and peer educators. The class exhibits lower socializing or relaxing and drinking alcohol levels. The class works higher hours than others, and students have over a 50% likelihood of being a work-study.

Serious Athlete (6.1% of the sample)-The class has an 84% probability of being a student-athlete and a nearly 91% probability of exercising five or more hours a week. The class has the highest co-curricular involvement. Unlike the party athlete class, serious athletes exhibit low drinking and socializing or relaxing levels. The class shows higher than average levels of studying and study group usage.

Party Athlete (7.7% of the sample)-The class has an 80% probability of being a student-athlete and over a 95% probability of reporting exercising at least 3 hours a week. The class exhibits the second-highest predicted levels of co-curricular involvement and spends an average amount of time socializing or relaxing. The class has a 60% predicted level of drinking at least two days a week and a 97% probability of binge drinking at least once a week.

Conventional Non-Worker (12.0% of the sample)-Among all OCAs in the LCA, the class has middling involvement leading to the use of the term conventional. The class distinguishes itself in that it has a 79% predicted probability to report not working and a 5% likelihood of participating in work-study.

Disengaged (13.9% of the sample)-The class has some of the lowest predicted academic, co-curricular, and leadership involvement levels. The group is somewhat paradoxical in that it has the second highest probability of students reporting 26 or more hours socializing or relaxing and working 21 hours or more a week. Overall, the class is united in their low-level involvement in all other OCAs.

Maximizer (5.2% of the sample)-The class showcases the highest or second to highest predicted levels of academic, co-curricular, and leadership involvement among almost all the OCAs. Maximizers have over a 70% predicted likelihood of having a leadership position and under a 10% likelihood of reporting never working with faculty or student affairs staff. Additionally, maximizers have a 30% likelihood to report being a summer orientation staff member or peer educator. Though predicted to socialize or relax less on average, the maximizer class has some of the highest predicted levels of daily and binge drinking and intramural sports participation. Finally, maximizers showcase the third-highest predicted level of work-study and the highest predicted likelihood of working 21 or more hours a week.

Moderate Worker (7.6% of the sample)-The class has around a 92% probability of reporting working between one to twenty hours a week and an 87% probability of being a work-study. The class reports higher predicted levels of hourly studying but showcases low predicted levels of participation in academic, leadership, and socializing activities and moderate levels of involvement in co-curricular activities.

Detached Partier (14.2% of the sample)-The detached partier exhibits the lowest predicted levels of involvement in academic, co-curricular, leadership, and working OCAs. The class has over a 70% predicted likelihood of drinking at least two days a week and a nearly 99% likelihood of binge drinking at least one day a week.

Involved Partier (14.1% of the sample)-While the involved partier drinks at predicted levels like the detached partier, they spend less time socializing and relaxing. The involved partier appears to replace the detached partier’s predicted likelihood of time spent socializing and relaxing with moderate to high predicted probabilities of involvement in co-curricular, leadership, academic, and work OCAs.

Religious (10.9% of the sample)-The religious class has a nearly 100% likelihood of involvement in a campus religious group or congregation and sometimes participating in spirituality-enhancing activities. The class has a high predicted level of participation among most academic activities and lower predicted levels of involvement among most co-curricular, leadership, and socializing activities. One notable exception is the second-highest predicted level of attending art events. The class has around a 50% likelihood of working and a 30% likelihood of having a work-study position.

Research Question 2

Seven of the ten classes differ by gender (see Table 3). While students reported as female are less likely to be a/n serious athletes, party athletes, maximizers, or involved partiers, they are significantly more likely to be disengaged, moderate workers, or religious. Compared to White students, African American/Black students are significantly more likely to be in the academic artist, conventional non-worker, and disengaged classes. Meanwhile, they are significantly less likely to be party athletes, detached partiers, and involved partiers than their White peers. Students identified as Asian were more likely to be academic artists, conventional non-workers, disengaged, and less likely to be in either the serious athlete, party athlete, detached partier, or involved partier classes than their White peers. Students identified as Latino/a/x were significantly more likely to be academic artists or moderate workers and less likely to be party athletes than their White peers. Students identified as Other/Native American were more likely to be in the disengaged class.

Table 3 Average marginal effects on the likelihood of a latent class

Regarding parental education, first-generation students were more likely to be disengaged and less likely to be in the involved partier class. Those with a dependent were less likely to be in the detached partier class. Students with bachelor’s degree-educated parents were more likely to be in the religious class than first-generation students. Students receiving a student loan or federal grant were more likely to be moderate workers. Otherwise, federal grant-receiving students were more likely to be academic artists and less likely to be conventional non-workers. International students were more likely to be either academic artists, conventional non-workers, or moderate workers. The students were significantly less likely to be in the party athlete or detached partier classes. English as a second language had no significant bearing on predicted class.

A student’s reported high school GPA had a variety of associations with student class. Compared to students reporting a C+ or less in high school, B− to B+ students were less likely to be disengaged. Students reporting an A− to A+ GPA were significantly more likely to be in academic artists, serious athletes, and religious classes. These students were significantly less likely to be in the disengaged and detached partier classes.

Compared to peers attending SLACs, students at regional colleges and universities were more likely to be conventional non-workers, disengaged, and detached partier typologies. Additionally, regional college or university attendees were significantly less likely to be in either the academic artist, serious athlete, party athlete, or moderate worker classes. Students attending research universities were more likely to be members of the disengaged or detached partier classes and less likely to be members of the serious athlete, party athlete, and moderate worker classes than students attending a SLAC.

The maximizer and academic artist classes had no significant difference by major. Compared to a student intending to be a STEM major, liberal arts, education, and social science were less likely to be of the serious athlete class. Liberal arts majors were less likely, and business majors were more likely than their STEM peers to be party athletes. Business majors were less likely to be in the conventional non-workers than STEM majors, while liberal arts and education majors were more likely to be in the disengaged class. While liberal arts majors were more likely to be moderate workers than STEM majors, business majors were less likely. Students in the allied health and business fields were more likely to be on the detached partier pathway, while business social science majors were more likely to be involved partiers. Education majors were more likely to be in the religious class. Finally, students seeking an advanced degree were more likely to be academic artists, conventional non-workers, and maximizers. Additionally, they were significantly less likely to be in the disengaged or detached partier classes.

Research Question 3

This question explores the relationship of outcomes between the various latent classes. At the end of the first year, students from the academic artist, moderate worker, and religious classes had significantly higher GPAs than the disengaged class (see Table 4). The party athlete and detached partier classes had significantly lower GPAs than the disengaged class. Serious athletes, maximizers, and involved partiers were similar to the disengaged by their end-of-first-year GPA. Regarding second-year retention, involved partiers, religious, and maximizers were at 55–72% greater odds of retention than disengaged students. Party athletes, moderate workers, and academic artists had 2.5–3 times greater odds of second-year retention than disengaged students. Conventional non-workers and detached partiers had no significant difference from their disengaged peers. By the fourth year, differences in retention dropped. Conventional non-workers, maximizers, and detached partiers did not significantly differ from disengaged peers. Academic artists and party athletes had around two times the odds of retention compared to disengaged peers. Serious athletes, moderate workers, involved partiers, and religious students were between 53 and 82% greater odds of fourth-year retention.

Table 4 Latent class association with academic and developmental outcomes

After accounting for baseline SRLS, MGUDs, Need for Cognition, and Ryff scores from the beginning of the first year of college, academic artists show the highest increases in the outcomes at the end of the first year relative to the disengaged class. The academic artists exceed all groups by showcasing at least a 0.15 improvement in all scores over disengaged peers. Conventional non-workers and involved partiers were the only other two classes to improve across at least four survey measures relative to disengaged students. Serious athletes significantly improved their SRLS and Ryff scores over disengaged students. Moderate workers have associations with increasing their SRLS and M-GUDs scores, while religious class students significantly increased their SRLS and Ryff scores relative to the disengaged class. The maximizer and detached partier classes had no associated increase across all four scales compared to their disengaged peers. With a 0.068 standard deviation decrease in the Need for Cognition scale, the party athletes were the only group to test significantly lower than disengaged students on any scale.

Discussion

Racialized and Classed Differences Across Pathways

Access and acceptance into various campus clubs and organizations at predominantly White institutions (PWIs) have been long-standing problems for students of marginalized racial/ethnic identities and social classes (Jack, 2019; Means & Pyne, 2017; Silver, 2020b). Students who report being African American/Black, Asian, and Latino/a/x are significantly more likely to be in the academic artist class and less likely to be in three of the four-partying classes than their White peers. Given the associated advantages on outcomes among academic artists, this may be welcome news for those interested in research surrounding racially minoritized students. Additionally, African American/Black and Asian students’ greater likelihood of being conventional non-workers provides some evidence that some are more likely to be moderately involved in OCAs. Unfortunately, students identified as African American/Black, Asian, and Other/Native American are significantly more likely to be disengaged. Student programming for students from racially minoritized communities may play a role in acclimating some racially minoritized students to campus and pushing students toward leadership and campus involvement positions. However, a lack of knowledge of such programs results in some students falling through the cracks and disengaging (Jack, 2016; Silver, 2020b). Though there is research on the greater likelihood of racially minoritized students being more likely to disengage from OCAs (Ogunyemi et al., 2020), further qualitative research could help us understand the greater likelihood of racially minoritized groups being academic artists and conventional non-workers.

Although not as salient as differences by race/ethnicity, our results show several differences among variables related to social class. For example, the results confirmed reports that first-generation students are less likely to be engaged in campus activities than their peers (Armstrong & Hamilton, 2013; Silver, 2020b; Stuber, 2011). Furthermore, first-generation students were significantly more likely to be members of the disengaged class but were significantly less likely to be members of the involved partier class. With these two exceptions and the one significant difference in the religious pathway, there is little difference in parental education levels among students in seven of the ten classes. Even among loan and federal grant recipients, the only difference was that these students were more likely members of the moderate worker class. The outcomes of the relationship between the latent classes and social class constructs may surprise some. Although Armstrong and Hamilton (2013) maintain that social class bifurcation emerges in how students navigate pathways, our results suggest that other variables may play a more significant role in sorting students into many classes than parental education or having a loan or federal grant.

It is still important to note that the disengaged class may act as protection against the negative ramifications of partying. Armstrong and Hamilton (2013) note that the social isolation of being disengaged often served to protect these students. Although some may prefer that students not become uninvolved or participate primarily in activities associated with the alcohol-fueled latent classes, it must be acknowledged that not all students will have the resources or backgrounds to be associated with one of the pro-academic and pro-developmental latent classes. Sociocultural affinity groups may provide some remedy to help students from traditionally marginalized backgrounds to become more active on and around campus. Postsecondary institutions create these spaces for various marginalized racial/ethnic campus groups, and there has been a recent increase in campus groups for low-income, first-generation and working-class students. These spaces allow students to evade micro- and macro-aggressions (Lee & Harris, 2020; Means & Pyne, 2017). Campuses need to continue supporting and implementing new racial-ethnic and social-class-oriented affinity groups. Besides the fact that these organizations could help create the opportunity for increased connectivity and belonging, they can provide students the chance for leadership roles and support both new and returning students to navigate the predominantly White middle-class context that is higher education (Harper, 2013; Means & Pyne, 2017; Museus, 2008).

The Party Lifestyles

Due to its negative associations with academic outcomes and strong connections with sexual and physical violence throughout college campuses (Shorey et al., 2011; Trolian et al., 2016; Vander Ven, 2011), heavy drinking in college is a widely discussed dimension of college in both popular culture and widely covered in academia. Our results describe four classes of students that engage in higher levels of drinking and moderate to high levels of socializing and relaxing: party athlete, maximizer, detached partier, and involved partier. Many of these students are men, more likely to be business majors, and have lower GPAs in high school. Except for the involved partier class, our results indicate that these classes are associated with no academic and little developmental benefit. Increased retention among the party athletes and involved partiers is one of the few associated benefits of being in one of the four party classes. Although maximizers engage in many activities during their first year of college, it did not significantly influence fourth-year retention compared to the disengaged class. Academic artists and involved partiers are second only to the maximizer in involvement across all OCAs, but the two classes have many more relationships to positive outcomes than the maximizer. One reason the differences exist may be that the maximizer drinks and binge drinks are at a greater level than the involved partier and far exceeds the academic artist.

With the party classes representing over 40% of the students, the results and outcomes highlight the significance of promoting campus involvement without simultaneously increasing alcohol consumption. Some athletic coaches and administrators may want to pay attention to the bifurcation between serious athletes and party athletes to keep party athletes on the road to collegiate success. Administrators and student affairs professionals in business schools may want to design student development to keep students out of these party lifestyles, which is vital for producing students who are developmentally and academically ready for careers in business. Campus leaders may want to engage in targeted interventions to change campus alcohol culture and work with the local community on the availability and marketing of alcohol-fueled events (Wechsler & Nelson, 2008).

A Religiously Involved Pathway

First, it is essential to note that the variables in the WNS capturing religious involvement do not specifically mention the type of faith or religion. Although we found a highly religious class of students, we cannot say if they are a religious majority, minority, unaffiliated, or a mixture of the three groups. How one engages in religion and the potential outcomes from that engagement has been a widely explored topic in higher education literature (e.g., Bowman & Small, 2012; Rockenbach & Mayhew, 2013; Stoppa, 2017), but there is no reference to religion as a class of student involvement in higher education research. Our analysis suggests that religion is a primary method of the OCA involvement of many students. Other analyses show that religiously involved students, whether majority or minority, report feeling out of place and stigmatized on secular college campuses (Binder & Wood, 2013; Nielsen & Small, 2019). These findings are essential for student affairs personnel and those working to improve student feelings of belonging insofar as our LCA highlights the importance of religious involvement among college students. Faith-based organizations may protect students from feeling marginalized by their religion as they face micro- and macro-aggressions. Further qualitative research could help elucidate how those faith-based organizations help guide students through religious connections across their first year of college.

The SLAC Advantage?

Some studies report an involvement advantage for students who attend SLACs, and our results support these reports. A SLAC’s smaller settings allow students to engage in and form networks to learn about OCAs (Stuber, 2011), and students attending these institutions are much less likely to be disengaged. Although we cannot definitively state that SLACs create fewer disengaged students or that those disengaged students are less likely to attend SLACs, we must acknowledge their low prevalence at these institutions. Additionally, students at SLACs are more likely to be serious or party athletes. Although SLACs tend to have a greater student-athlete to student ratio, SLAC administrators and student affairs personnel may want to pay close attention to their party athletes. Since SLACs tend to have fewer detached partiers than their regional and research institution peers, the party athlete may be one of the most prominent partier latent classes central to the SLAC experience.

SLACs also tend to have more moderate workers than other HEIs. The distinction is important in positive outcomes among college students. As our results from RQ₃ show, these students have associated higher GPAs, better retention rates, and significantly higher SRLS and M-GUDs scores than their disengaged peers. Remember that 90% of these students work less than 20 hours weekly, and 87% are on work-study. Given the rate of work-study and positive outcomes related to the moderate worker class, larger regional and research institutions would do well to integrate work-study positions throughout their campuses more fully.

The Disengaged Class

One complicating factor for the disengaged class is that the class may encompass two classes. On average, students in the disengaged class work some of the least and most hours compared to the average student in the dataset. Additionally, the disengaged class had one of the highest percentages of students who socialized or relaxed for 26 or more hours a week. However, the average total number spent socializing and relaxing for the class was less than five hours per week, reflecting the dataset average. These descriptors may suggest that a small but sizable portion of the disengaged class socializes or relaxes greatly, but most do not. Furthermore, this class remains the lowest or second-lowest across all OCA involvement outside of working and relaxing. Thus, there may be two classes within the disengaged profile, one class working more hours and another socializing more hours. Potentially due to small subsamples, the LCA unites these two groups by the students’ low level of involvement of all other OCAs. For example, in the dataset, only 4.9% of students report working 21 or more hours a week, while the National Center for Education Statistics reports that 27% of students work 20 or more hours a week (NCES, 2020). Arguably, there is a reasonable probability that a "disengaged" group of students working long hours may exist in other datasets.

Nevertheless, our results are within the norm for LCAs. Neither Quadlin and Rudel’s (2015), Fosnact et al.’s (2018), or Willoughby et al.’s (2020) LCAs nor Hu et al.’s (2011) review of prior class analyses indicate a group of students that spend an overabundance of their hours working. Then again, because no previous studies have found a high work latent class does not negate its existence. Future quantitative datasets with less second-semester attrition may determine whether there is a group of disengaged students due to work. The disengaged group does reflect other past studies on student typologies (Hu et al., 2011) and may be an area of concern for college administrators and student affairs professionals. Although not necessarily any better than the disengaged class on academic outcomes, the conventional non-worker indicates that simply getting non-working disengaged students to have moderate levels of OCA involvement may bolster student developmental outcomes.

Future Research

Our study points to the need for quantitative analyses of OCA involvement to consider the interconnectivity of activities and behaviors. Those wishing to make causal claims require controlling for other potentially interconnected OCAs. Researchers developing their datasets concerning specific OCAs (i.e., athletics, research with faculty, religious participation) need to include questions about potential OCAs related to that particular activity. For example, if one were to study religious involvement and academics, they should include survey questions asking these students about behaviors surrounding studying, drinking, and socializing. These OCAs have a solid connection to religious participation and academics. As such, drinking, socializing, or studying could be a mediating force behind any relationship between involvement in religion and any positive academic outcome. Additionally, given the degree to which arts, athletics, work-study, and religion help sort students into classes, it stands to reason that including affinity and cultural groups may affect the number of latent classes and latent class structure. As the newer generation of college students includes more racially minoritized and first-generation students, campus organizations centered on the sociocultural characteristics of students could become a key component in how students sort into various classes.

Another avenue for future research is to create datasets that follow students throughout the calendar year and their time in college. Having multiple data points allows one to create a latent transition analysis. A latent transition analysis examines how and in which students change profiles throughout their time at a college. Furthermore, following OCA involvement across the first year will better understand the types of students that disengage in prosocial OCA involvement and at what point the disengagement begins. Because race/ethnicity, gender, and class are intersectional and formative for student OCA involvement (Silver, 2020b), future LCAs should test for differences by interacting with the three constructs. Future LCAs should not only look at how these classes affect collegiate and post-collegiate outcomes but explore how intersectionality may aid or inhibit the effects of the classes. One can imagine the utility of a one-time dataset at the end of the first semester of the first year to mitigate attrition by those who do not return for a second semester. Another avenue for typology research could be to present to hiring firms and human resource workers lists of the average involvement of students in various activities to understand which class of student would be the ideal worker at their place of employment.

Finally, our results may vary based on the types of institutions surveyed. As noted in the limitations, most students attend SLACs (51.2%), and only a handful are adult learners (0.6%). One example of how SLACs may have biased the dataset is the less than one to three percent chance of serious and party athletes attending regional and research institutions (see Table 3). These questions about the dataset and results of the LCA indicate that the number of classes and classes themselves may reflect the institutions involved. Furthermore, since only 4.4% of students work more than 20 hours a week, future LCAs of student involvement may need to be more specific to institutional sectors or types (e.g., HBCUs, regional or research institutions, and community colleges), which tend to have greater rates of students working 20 or more hours per week. Narrowly focusing on particular sectors may help student affairs personnel and college administrators better tailor to meet the needs of students within their institutions.

Conclusion

Student involvement in OCAs throughout college does not happen in a vacuum. Activities or behaviors that students engage in one day may affect the engagement or disengagement in other activities and behaviors the next day. Our study provides evidence of the interconnectivity of OCAs during the first year of college. In answering the first research question, the analyses indicate ten distinct classes of outside-the-classroom involvement that students sort themselves into during the first year of college. Like previous typology studies (see Hu et al., 2011 for a complete discussion), we found evidence of various groups based on academics, social involvement, leadership, and work. Nevertheless, contrary to these studies, we also found that specific areas of involvement (e.g., religious, arts, and athletic participation) help surface other student engagement pathways or classes. Our second research question addresses significant advantages and disadvantages for classes associated with various sociocultural, economic, and academic constructs. Students reported as African American/Black, Latino/a/x, and Other/Native American and first-generation students are disproportionately on the disengaged path. Finally, in addressing our third research question, we found that the academic artist class far exceeds their peers in academic and developmental outcomes. Meanwhile, classes with the highest levels of partying, detached partiers, and party athletes perform poorly and sometimes worse than disengaged class students.

Appendices

Appendix

Appendix A

Institutions covered for the research questions

RQ1 and RQ2- 44 institutions, including 6 research universities, 29 liberal arts college, and 9 regional universities.

RQ3- 39 institutions, including, including 5 research universities, 26 liberal arts college, and 8 regional universities.

RQ1 and RQ2	RQ3
Allegheny College	Allegheny College
Alma College	Alma College
Alverno College
Augustana College	Augustana College
Bard College	Bard College
Bennington College	Bennington College
Blackburn College	Blackburn College
Brandeis University
Butler University	Butler University
Carleton College	Carleton College
Coe College	Coe College
College of the Holy Cross	College of the Holy Cross
Columbia College of SC	Columbia College of SC
Connecticut College	Connecticut College
Delaware State	Delaware State
Drew University	Drew University
Fairfield University	Fairfield University
Franklin College	Franklin College
Gustavus Adolphus University	Gustavus Adolphus University
Hamilton College	Hamilton College
Hampshire College
Hobart and William Smith Colleges	Hobart and William Smith Colleges
Hope College	Hope College
Lasell College	Lasell College
Marlboro College	Marlboro College
Millersville University	Millersville University
New College of Florida
North Carolina A&T	North Carolina A&T
Oxford College of Emory University	Oxford College of Emory University
Prescott College
Ripon College	Ripon College
Salem State College	Salem State College
San Jose State University	San Jose State University
Univ. of North Carolina at Wilmington	Univ. of North Carolina at Wilmington
University of Kentucky	University of Kentucky
University of Michigan	University of Michigan
University of Notre Dame	University of Notre Dame
University of Rhode Island	University of Rhode Island
Vassar College	Vassar College
Wabash College	Wabash College
Warren Wilson College	Warren Wilson College
Wheelock College	Wheelock College
Whittier College	Whittier College
Worcester Polytechnic Institute	Worcester Polytechnic Institute

Appendix B

Recoding of Variables

Total number of hours worked on- or off campus-The original was coded in 2.5-hour intervals. To cut down on the number of categories, we made the decision to change the coding to five-hour intervals. Because 4,286 students worked zero hours, we made this our base-level. Students reported working as many as 65 hours. Because only 4.4% of students worked 21 or more hours a week, we used this level as our top code. The top-coding is a small cell size. However, much of the literature that surrounds working in college notes a deleterious effect of working while in college at the more than 20 hours a week level.

Total hours of studying a week-The original survey included 26 student reporting to study zero-hours. Because of the small cell size, we recoded these students to be included in the “1 to 5 Hours” category. To keep the proportion of students in the cell sizes from becoming to small, recoded the 26 to 30 hours and 30 or more hours of studying into one category.

Time Spent Socializing or Relaxing-Because only 47 students reported no hours spent socializing or relaxing, we recoded these students to be included with students reporting as “1 to 5 Hours” of socializing and relaxing a week. Due to small cell sizes, socializing 26 to 30 and 30 or more hours a week were recoded into one category.

Number of times binge drinking a week-The original variable had coding for “0-Never,” “1–1 Time,” “2–2 Times,” “3.5–3 or 4 Times,” and “5–5 Times.” We top-coded the number of days drinking a week at 3, because less than 3.5% of the sampled drank four or more days a week.

Number of times drinking in a week-Due to small cell sizes of students drinking five, six, seven, or seven or more times a week, we opted to top code all four together as “Five or more times a week.”

Hours of exercise in the week-The original coding was done as “0–0 Hours,” “1.5–1 or 2 Hours,” “3.5–3 or 4 Hours,” “5.5–5 or 6 Hours” and “7–7 or More Hours.” To make this categorical, we recoded the hour categories as 0, 1, 2, 3, and 4.

Hours spent in co-curricular activities-The original coding is in five hour intervals starting at 0 and moving onto 1 to 5 hours. Because the percentage of students at 21 to 25, 26 to 30, and more than 30 hours is between 1.3% to 3.1% of the dataset, we top-coded the variable at 21 or more hours.

Appendix C

Mean-values and missingness of Research Question 1 Variables.

Activity	Mean	% Missing		Mean (Stdev)	% Missing
Academic			Leadership
Studying—Weekly hours		1.35%	Committee/organization work with faculty members		0.80%
5 or less	10.8%		Never	49.1%
6 to 10	21.7%		Sometimes	33.6%
11 to 15	21.9%		Often	12.6%
16 to 20	18.8%		Very Often	4.8%
21 to 25	13.7%		Committee/organization work with student affairs personnel		3.21%
26 or more	13.2%		Never	42.1%
Completed internship or field experience	7.8%	1.28%	Rarely	21.8%
Faculty research project participation	5.2%	1.46%	Sometimes	20.3%
First-year seminar participant	66.8%	2.61%	Often	10.8%
Honors program participant	16.4%	2.42%	Very often	5.0%
Learning community participant	39.5%	2.27%	Leadership position	24.1%	2.38%
Study group usage		3.16%	Leadership training	13.3%	2.73%
Never	12.3%		Orientation staff member	3.4%	2.48%
Rarely	18.3%		Peer educator—Non-academic	4.1%	2.70%
Sometimes	27.3%		Socializing
Often	24.9%		Socializing or relaxing—Weekly hours	17.7%	1.45%
Very Often	17.2%		Less than 5	28.1%
Cocurriculars			6 to 10	22.0%
Cocurricular involvement—Weekly hours		1.27%	11 to 15	7.4%
0	23.3%		16 to 20	14.6%
1 to 5	37.5%		21 to 25	7.2%
6 to 10	18.4%		26 or more	10.4%
11 to 15	8.8%		Alcohol drinking—Days a week		3.03%
16 to 20	5.6%		0	47.5%
21 or more	6.4%		1	22.9%
Arts events attendance		1.12%	2	18.6%
Never	12.3%		3 or more	11.0%
Sometimes	49.8%		Binge Drinking—Days a week		3.05%
Often	23.9%		0	57.5%
Very Often	14.0%		1 or 2	32.8%
Diversity workshop participation		2.44%	3 or more	9.7%
Never	53.0%		Exercise—Weekly hours		2.97%
Rarely	25.7%		0	17.0%
Sometimes	14.9%		1 or 2	23.6%
Often	4.6%		3 or 4	26.1%
Very Often	1.7%		5 or 6	14.4%
Political or social justice event participation		2.20%	7 or more	18.9%
Never	33.1%		Intramural sport participation		2.22%
Rarely	29.5%		Never	51.8%
Sometimes	24.3%		Rarely	12.2%
Often	9.0%		Sometimes	15.2%
Very Often	4.1%		Often	11.9%
Religious group member	29.2%	2.54%	Very Often	8.9%
Spirituality enhancing activities		1.21%	Work
Never	41.3%		Working—Weekly hours		1.39%
Sometimes	30.6%		0	52.3%
Often	14.0%		1 to 10	29.8%
Very Often	14.1%		11 to 20	13.4%
Student athlete	16.1%	1.64%	21 or more	4.4%
Volunteered	51.5%	1.58%	Work study participant	32.4%	9.42%

Appendix D

Mean values and missingness of Research Question 2 variables

	Mean (std err)	% Missing		Mean (std err)	% Missing
Dependent variables
Academic artist	8.27%		Studious athlete	6.07%
Party athlete	7.70%		Conventional non-worker	12.03%
Uninvolved	13.90%		Maximizer	5.23%
Moderate worker	7.63%		Partier	14.18%
Involved partier	14.11%		Religious	10.90%
Independent variables
Gender			High School GPA		2.37%
Male	37.36%		Below C+	3.23%
Female	62.64%		B−  to B+	37.36%
Race			A−  to A+	57.04%
White	68.85%		Institutional Type
African American/Black	9.11%		Liberal arts college	51.14%
Asian	5.51%		Regional university	22.67%
Latino/a/x	5.20%		Research university	26.19%
Other/Native American	11.32%		Major		1.93%
Parents’ highest level of education		2.75%	STEM	21.00%
High School or Less	10.26%		Liberal arts	12.09%
Some college/AA	18.34%		Allied health	10.03%
BA	30.87%		Business	12.28%
Advanced Degree	37.78%		Education	8.21%
Loan recipient	54.43%		Social Sciences	10.41%
Federal grant recipient	15.10%		Other/Undecided	24.05%
Have dependents	1.65%		Advanced Degree Aspirations	77.31%	3.34%
International student	2.04%		Non-native English speaker	7.75%	2.04%

Appendix E

Mean values and missingness for Research Question 3

	Mean (std err)	% Missing		Mean (std err)	% Missing
Dependent Variables
Spring of first year GPA	3.182(0.666)		Universal-diverse orientation score (M-GUDS)	4.554(0.681)
Second year retention	92.79%		Need for cognition	3.435(0.620)
Fourth year retention (Fall)	83.72%		Ryff score	4.513(0.627)
Socially responsible leadership score (SRLS)	4.016(0.465)
Independent variables
Academic artist	8.42%		Studious athlete	6.29%
Party athlete	8.03%		Conventional non-worker	11.81%
Uninvolved	12.63%		Maximizer	4.93%
Moderate worker	7.23%		Partier	13.99%
Involved partier	14.93%		Religious	11.74%
Control variables
Gender			High School GPA		1.99%
Male	37.11%		Below C+	2.84%
Female	62.89%		B−  to B+	36.55%
Race			A−  to A+	58.62%
White	69.72%		Institutional Type
African American/Black	9.47%		Liberal arts college	51.90%
Asian	5.67%		Regional university	21.95%
Latino/a/x	4.79%		Research university	26.15%
Other/Native American	10.35%		Major		1.87%
Parents’ highest level of education		2.66%	STEM	20.22%
High School or Less	10.03%		Liberal arts	11.21%
Some college/AA	18.92%		Allied health	10.66%
BA	31.75%		Business	13.35%
Advanced Degree	36.64%		Education	8.53%
Loan recipient	56.38%		Social Sciences	10.27%
Federal grant recipient	15.41%		Other/Undecided	23.89%
Have dependents	1.25%		Advanced Degree Aspirations	77.59%	3.24%
International student	1.87%		Non-native English speaker	7.50%	2.42%
Fall of first year scores (used as control only for matching dependent variable)
Socially responsible leadership score (SRLS	3.400(0.430)		Universal-diverse orientation score (M-GUDS)	4.589(0.650)
Need for cognition	3.428(0.607)		Ryff score	4.524(0.587)