According to the report, Swimming Against the Tide: The Poor in American Higher Education (Terenzini, Cabrera, and Bernal, 2001), compared to other students, low socioeconomic status (SES) students, are less likely to earn a degree from a four-year institution. Similarly, the National Center for Education Statistics (NCES) reports that students from families with low-income or with parents who did not have a bachelor’s degree were more likely to drop out of college compared to students from families with high income or with parents who had a bachelor’s degree or higher (NCES, 2003). Carnevale and Rose (2004) allude to higher college dropout rates among low SES students being related their disproportionately low enrollment in selective institutions.

Because college attendance at institutions is stratified by SES (Astin and Oseguera, 2004) and institutional selectivity may be positively related to an institution’s financial resources (Winston, 1999), examining the financial context of four-year institutions may help us to further understand how colleges and universities affect the chances of persistence to degree completion for low SES students, compared to other students. Using national survey data, multilevel modeling techniques, and descriptive statistics, this study makes an effort to understand the influence of the financial context of institutions on the chance of college completion for low SES students at four-year colleges and universities.

Low SES students, college completion, and institutions

Compared to other students, low SES students complete college at lower rates (Gladieux and Swail, 1998). According to Upcraft and Crissman (1999), low SES students who attend institutions with a higher proportion of low SES students graduate at lower rates than low SES students who attend institutions with a lower proportion of low SES students. Berger (2000) contends that more research is needed to understand how institutions affect college student persistence by social class.

Several researchers have addressed the influence of SES on college student access (e.g., Astin, 1993; Cabrera, Burkum, and La Nasa, 2003; Hossler, Braxton, and Coopersmith, 1989; Kane and Spizman, 1994), college choice (e.g., Baker and Vélez, 1996; Cabrera, Burkum, and La Nasa, 2003; McPherson and Schapiro, 1998; Perna and Titus, 2004), and persistence (e.g., Paulsen and St. John, 2002; Stampen and Cabrera, 1986, 1988; St. John, Paulsen, and Starkey, 1996).

Although a few studies (e.g., Kim, Rhoades, and Woodard, 2003; Ryan, 2004) have demonstrated that college completion is influenced by aspects of the financial context of colleges and universities, there is no known research that explores the extent to which the financial context of four-year institutions impacts college completion by social class.

Conceptual framework

This research extends the Berger-Milem (2000) college impact model to examine how financial aspects of a four-year institution influence college completion for low SES students, compared to other students. Drawing on organizational behavior theory (Bolman and Deal, 1991), the research literature on peer group effects (Astin, 1993; Weidman, 1989), and student involvement theory (Astin, 1984), the Berger-Milem college impact model posits that both institutional and student characteristics influence student outcomes, including persistence. Modifying Bolman and Deal’s (1991) typology, Berger and Milem characterize organizational behavior as systemic. Using the concept of systemic behavior based, they contend that institutional behavior can be viewed from several perspectives, including a resource dependency framework which explains organizational behavior in terms of those resources that an organization must have in order to function (e.g., Pfeffer, 1997; Scott, 1995). Resource dependency theory explains organizational behavior in terms of an organization’s internal adjustment to changes in the availability of such external resources as finances that an organization must have in order to function (e.g., Pfeffer, 1997; Scott, 1995).

Berger and Milem hypothesize that in addition to organizational behavior, structural-demographic features, student peer group climate, and student entry characteristics, such as SES, influence student outcomes, including college persistence. Using constructs from the Berger-Milem (2000) college impact model, Titus (2004) developed a conceptual framework to investigate the influence of the institutional context on college student persistence and demonstrated that persistence is positively related to such institutional characteristics as selectivity, an aspect of student peer climate, and size after controlling for student-level variables. Winston (1999) speculates that institutional selectivity, reflected by student peer climate, may be positively related to an institution’s financial resources.

Resource dependency theory (Hrebiniak and Joyce, 1985; Pfeffer and Salancik, 1978; Scott, 1995) provides a framework to examine the how an institution’s internal allocation of financial resources is influenced by the relative magnitude of the external financial resources. In response to changes in the external environment, organizations strategically make choices so as to manage their dependency on vital resources. With respect to higher education institutions in the not-for-profit sector, Leslie and Slaughter (1997) contend that as the state begins to link the rationing of scarce financial resources to performance accountability indicators, power becomes centralized at the campus level. This centralization of power is partly reflected in an increase in the percent of expenditures on administrative functions at the institutional level (Francis and Hampton, 1999; Leslie and Rhoades, 1995; Leslie and Slaughter, 1997). Other characteristics, such as its scale (Minzberg, 1983; Greening and Gray, 1994), mission (Pfeffer and Salancik, 1978), and extent to which its sub-units operate autonomously (Cohen and March, 1974) may also determine how a college or university is influenced by changes in the external environment.

Drawing from the concept of systemic organizational behavior via a resource dependency perspective, prior research by Titus (in press) investigated how college student persistence is influence of market-based revenues and internal allocation of expenditures. This study extends the work of Titus (in press) and explores the extent to which college degree completion by SES is influenced by an institution’s dependence on market-based revenues and internal allocation of expenditures, after taking other institution-level and student-level variables into account.

Research method

The study uses student- and institution-level data to address, in sequential order, the following research questions:

  1. 1.

    After taking other student- and institution-level variables into account, how does SES influence college completion?

  2. 2.

    After taking student- and institution-level variables into account, is there a difference by SES in the chance of college completion?

  3. 3.

    After taking into account student-level and other institution-level variables, do institutional revenue and expenditure patterns help to explain the difference by SES in college completion?

Data

Student-level data are drawn from the 1996–2001 Beginning Postsecondary Students (BPS:96/01) survey, a longitudinal database sponsored by the U.S. Department of Education’s National Center for Education Statistics (NCES). The BPS collects data, using the 1996 National Postsecondary Student Aid Study (NPSAS:96) as a sampling frame. The focus of the cross-sectional NPSAS is on how students and their families finance postsecondary education. Data for the NSPSAS:96 were collected from government data files, institutions, students, and students’ parents on undergraduates and the institutions they attended over a period of time so as to provide such information as student background characteristics, financial aid, educational experiences, persistence, postsecondary enrollment, employment, income, and career aspirations (Wine et al., 2000). The NPSAS sampling frame is based on a two-stage sampling design in which postsecondary institutions were selected at the first stage and postsecondary students within institutions were selected at the second stage (Wine et al., 2000).

The BPS:96/01 includes data on first-time beginning postsecondary students and the institutions they attended during 1996 (i.e., 1995–1996 academic year) with follow-up data collected in 1998 and 2001. The BPS collects data over time so as to provide such information on student-level background characteristics, financial aid, educational experiences, persistence, employment, income, and career aspirations (Wine et al., 2000).

Institution-level data are drawn from the NCES’ Integrated Postsecondary Education Data System (IPEDS) Fall 1995 Institutional Characteristics (IC) Enrollment survey and Fiscal Year (FY) 1996 Financial surveys. The IPEDS IC survey collects such institution-level data as characteristics, enrollment, and academic program offerings while the IPEDS FY survey collects revenue and expenditure data from all postsecondary institutions in the United States and outlying U.S. territories.

The analytic sample used in this research is limited to Fall 1995 first-time, full-time, degree-seeking undergraduate students attending four-year colleges and universities. This study also excludes institutions with missing data on finances and enrollment (N=5). Data from the IPEDS Fall 1995 and fiscal year 1995 surveys are used to be consistent with the data that are drawn from the BPS:96/98 survey that describes students who first enrolled in the fall of 1995. The final analytic sample is comprised of 5,776 students attending 400 four-year institutions.

Variables

The dependent variable in this study is defined as having completed a bachelor’s degree program within six years after first enrolling in the same four-year institution. Therefore, graduation from the initial college or university rather than graduation from the system of higher education is the focus of this research.

The independent student-level variables include measures of student background characteristics, college experiences, unmet financial need, hours worked, and working off campus. Student background characteristics include pre-college academic performance, gender, race/ethnicity, and socioeconomic status (SES). Pre-college academic performance is measured by standardized SAT score. Four racial/ethnic groups are included in the analyses: African American, Hispanic, Asian, and White (the reference group). Following the example of Ethington (1990), a composite measure of socioeconomic status is derived from the sum of standardized parental income and standardized parental educational attainment. SES is measured by a standardized composite of parents’ income and educational level. In this study, SES is measured as a continuous and categorical variable, with four quartiles from lowest SES to highest SES (the reference group) included in the analyses.

Student experiences are measured by four components: college academic performance, declaring a major, living on campus, and student involvement during the freshmen year. College academic performance is measured by college GPA. Declaring a major is measured by whether or not a student declared a major during the freshmen year. Living on campus is measured by whether or not a student lived on campus during the freshman year.

Student involvement is measured using a factor composite of survey items from the BPS:96/01 reflecting how often students reported participating in various activities. These activities include participating in school clubs, participating in fine arts activities, attending lectures, conventions, or field trips, going places with friends, having social contact with faculty, attending study groups, meeting with advisor about plans, and talking with faculty outside of class. The alpha reliability coefficient for this factor is .65, below the threshold of .70 recommended by Pedhazur and Schmelkin (1991). Therefore, the student involvement factor composite should be interpreted with caution.

Unmet need is measured by a continuous variable defined in the BPS:96/01 data set as the student’s total cost of attendance less the estimated family contribution and total amount of financial aid received. Because the data are non-normally distributed, the number of hours worked per week is recoded into four categories: zero hours, one to ten hours, eleven to twenty hours, and more than twenty hours. Worked zero hours is the reference category. Working off campus is measured by whether or not a student worked off campus during the freshmen year.

The independent institution-level variables include measures of student peer characteristics, structural-demographic characteristics, and financial aspects of a four-year institution. Student peer group characteristics are measured by the percent female, racial/ethnic diversity, and the average SES of full-time freshmen attending the same institution. Percent female is measured by the percent of first-time freshmen at the same institution that is female. Using an index developed by Chang (1999), institutional racial/ethnic diversity is measured by an index that reflects an institution’s heterogeneity with respect to the racial/ethnic composition of undergraduate student enrollment. This index measures the variance in the racial/ethnic composition of full-time freshmen enrollment at an institution. Using the student-level composite, institutional-level peer group SES is derived by computing the average SES for full-time freshmen attending the same institution.

The analyses also include variables reflecting such structural-demographic characteristics as institutional control (public versus private), enrollment size, and average SAT score of full-time freshmen attending the same institution, a measure of institutional selectivity. Using total student headcount enrollment, size is recoded into three groups: small (less than 4,133); large (more than 19,576); and medium (4,133 to 19,576). Medium is the reference group.

Following the recommendation by Kingston and Smart (1990), institutional selectivity is organized into three groups: high selectivity (the SAT score for students in the highest quartile); low selectivity (the SAT score for students in the lowest quartile); and middle selectivity (the SAT score for students in the second and third quartiles). Middle selectivity is the reference group.

The financial context of a four-year institution is measured by variables reflecting institutional revenues and expenditures. The revenue variables include the following: percent of total revenue from state appropriations; percent of total revenue from tuition; percent of total revenue from competitive grants and contracts; and percent of total revenue from endowments. The expenditure variables are as follows: percent of total education and general (E&G) expenditures on instruction; percent of total E&G expenditures on research; percent of total E&G expenditures on administration; percent of total E&G expenditures on student services; percent of total E&G expenditures on grants and scholarships; and total E&G expenditures per full-time equivalent (FTE) student. Table 1 lists all of the variables used in this study.

Table 1 Variables Used in the Analysis
Full size table

Analyses

Because the data used in this study are from sample with a complex design, hierarchical generalized linear modeling (HGLM) is used to address the research questions. Correcting for the BPS:96/01 survey design using the second-order correction of Rao and Scott (1984), one-way ANOVA and cross tabulations are used to explore differences among students by SES with regard to the independent variables in the study. In HGLM, a sampling model has to be developed and estimated. Because the outcome variable is binary, the sampling model is Bernoulli:

$$\hbox{Prob}(Y_{ij}=\vert\beta_{j})=\Phi_{ij}$$
(1)

Because the analyses show the variation in the estimated average chance of persistence between institutions, the study proceeds with the use of both within-institution and between-institution models in an effort to address the research questions. The within-institution model with student-level predictors is expressed as follows:

$$\eqalign{ \log\left[{{\phi_{ij}}\over{1-\phi_{ij}}}\right]&=\beta_{0{j}}+\beta_{1{j}}\ast (\hbox{STUDENT CHARACTERISTICS})_{ij} \cr &\quad+\beta_{2{j}}\ast (\hbox{STUDENT COLLEGE EXPERIENCES})_{ij} \cr &\quad+\beta_{3{j}}\hbox{missing\_data}_{ij}+\mu_{ij}}$$
(2)

where i denotes the student and j denotes the institution, STUDENT CHARACTERISTICS, and STUDENT EXPERIENCES are the vectors of variables described above.

The institution-level model takes into account the variance in the intercept across institutions and is expressed as follows:

$$\eqalign{\beta_{0j}&=\gamma00+01\ast(\hbox{INSTITUTIONAL CHARACTERISTICS})_{j} \cr &\quad+\gamma 02\ast(\hbox{INSTITUTIONAL REVENUE PATTERNS})_{j} \cr &\quad+\gamma 03\ast (\hbox{INSTITUTIONAL EXPENDITURE PATTERNS})_{j}}$$
(3)

where j denotes the institution, and INSTITUTIONAL CHARACTERISTICS, INSTITUTIONAL REVENUE PATTERNS, and INSTITUTIONAL EXPENITURE PATTERNS are vectors of institution-level variables as described above.

With the exception of the variables that are aggregated to the institution level, all student-level variables are grand-mean centered. Grand-mean centering controls for differences between institutions in student characteristics and experiences. As a result, the intercept term in equation (2) may be interpreted as the estimated odds of persisting in college for students having the average characteristics and college experiences for the entire sample (Rumberger, 1995; Rumberger and Thomas, 2000).

The variables are entered sequentially in the HGLM analyses in conceptually-related blocks, with only the statistically significant variables from the preceding step retained in the subsequent step, as recommended by Raudenbush and Bryk (2002). This study uses odds-ratios to facilitate the interpretation of results. The odds-ratio represents the change in the odds of persistence relative to not persisting that is associated with a one-unit change in a specific independent variable while holding all other variables constant (Peng et al., 2002). An odds-ratio greater than one represents an increase in the likelihood of persisting in a four-year institution relative to not persisting. An odds-ratio of less than one represents a decrease in the likelihood of persistence. Odds ratios are computed for only those variables which are statistically significant.

The beta coefficients in equation (2) reflect the distribution of college persistence in institution j given observable student characteristics and experiences. In this study, with the exception of the intercept, the analyses constrain the coefficients for all within-institution predictors to be the same for all institutions. Chi-square tests were conducted and revealed that none of the coefficients vary systematically across institutions.

Following the example of Rumberger and Thomas (2000), the population-average estimates with robust standard errors are utilized. The population-average estimates are used to examine the average effect across students in the study of a one-unit change in a given institution-level predictor on the odds of completing a degree (Raudenbush and Bryk, 2002; Raudenbush et al., 2000).

Limitations

This study is limited in at least three ways. First, like all secondary data analysis, this study is limited by the availability of variables in the datasets. Second, the analysis is limited by missing data at the student level. Missing data analyses and chi-square tests reveal that data are missing at random. Because listwise deletion of cases would have resulted in a substantial reduction in the number of cases from 5,776 to 3,722, this study follows the recommendation by Cohen and Cohen (1983) and includes an independent variable at the student level that reflects the “tendency to have missing data” (p. 296), calculated as the number of independent variables on which data are missing. Finally, the multilevel modeling software (HLM5) used in this study does not allow for the use of weights at the student level and HGLM analysis (Raudenbush et al., 2000). Although ignoring sample weights may result in biased estimation of parameters, in this study, the multilevel estimation methods combine maximum likelihood techniques with empirical Bayes (EB) techniques. EB techniques take into account data with complex sample designs, such as the BPS:96/01, and help to improve efficiency and reduce parameter uncertainty (Raudenbush and Bryk, 2002).

Results

As shown in Table 4, 46% first-time full-time freshmen in the lowest SES quartile completed college within six years after first enrolling in the same four-year institution, compared with 51% of those in the second quartile, 63% in the third quartile, and 71% of those in the highest quartile. These descriptive statistics are also consistent with observations by several researchers (Astin and Oseguera, 2004; McPherson and Schapiro, 2002) of the social stratification of colleges and universities.

Although not shown, Scheffe multiple-comparison tests reveal that the difference between the first SES quartile and the second SES quartile in the chance of college completion is not statistically significant.

Hierarchical Generalized Linear Modeling Analyses

The HGLM demonstrates that college degree completion is positively influenced by college academic performance, degree major certainty, campus residence, and student involvement and negatively influenced by being a member of an underrepresented minority group, unmet financial need, and working more than a certain number of hours per week. As shown in Tables 2 and 3 from the HGLM analyses, the chance of college degree completion is positively associated with college GPA (odds ratio= 2.128, p<.001 and odds ratio= 2.152, p<.001, respectively), declared a major (odds ratio= 1.164, p<.05 and odds ratio= 1.166, p<.05, respectively), and lived on campus (odds ratio= 1.289, p<.01 and odds ratio= 2.186, p<.01, respectively). The odds of college completion is negative related to unmet financial need (odds ratio= 0.925 p<.05 and odds ratio= 0.928, p<.05, respectively). Tables 2 and 3 also reveal that, compared to White students, African American (odds ratio= 0.702, p<.01 and odds ratio= 0.711, p<.01, respectively) and Hispanic (odds ratio=0.742, p<.05 and odds ratio= 0.741, p<.05, respectively) students have a lower chance of college completion.

Table 2 Student- and Institution-level Predictors of College Completion Six Years after First Enrolling at the Same Four-year Institutions among Fall 1995 First-time Full-time Degree-seeking Freshmen – SES Continuous
Full size table
Table 3 Student- and Institution-level Predictors of College Completion Six Years after First Enrolling at the same Four-year Institutions among Fall 1995 First-time Full-time Degree-seeking Freshmen – SES Quartiles
Full size table

Table 2 reveals that after taking other student-level predictors of college completion into account, the chance of college completion increases with SES (odds-ratio=1.082, p<.05). Table 3 demonstrates that compared to students in the fourth SES quartile, students in the first SES quartile (odds-ratio=0.757, p<.01) and second SES quartile (odds-ratio=0.782, p<.01) have a lower chance of completing college in six years, compared to students in the fourth SES quartile, even after other variables are taken into account.

Although not shown, a multivariate hypothesis test reveals that after other variables are taken in account, the institution average SES has a positive effect on college completion that is over and above the individual effect of SES (difference in log-odds is 0.164, χ2=23.327, p<.0001). The HGLM analyses reveal that net of student-level variables, college completion is positively influenced by attending a racially/ethnically or socio-economically diverse, private or selective institution. Tables 2 and 3 show that after taking into account other variables, the odds of completing college are positively influenced by being a part of a racially/ethnically diverse freshmen class (odds-ratio = 1.090, p<.01 and odds-ratio=1.087, p<.01, respectively). This finding appears to be inconsistent with the lower chance of college completion for African American and Hispanic students.

The results of this study suggest that the chance of college completion is positively influenced by enrolling in a private institution. Tables 2 and 3 reveal that the odds of college completion is higher for a student in a private institution (odds-ratio = 1.212, p<.05). This study also suggests that after taking into account other variables, the chance of college completion increases with institutional selectivity. Tables 2 and 3 show that compared to students who enroll at other four-year institutions, students who attend high selectivity institutions have a higher chance of college completion (odds-ratio = 1.281, p<.05 and odds-ratio = 1.249, p<.05, respectively).

The HGLM analyses reveal that the chance of college completion is influenced by aspects of an institution’s financial context. Table 2 shows that college completion is positively related to total education and general expenditures (E&G) per FTE student (odds-ratio = 1.161, p<.05). Table 3 reveals that college completion is positively influenced by tuition and fees as a percent of total revenue (odds-ratio = 1.041, p<.01) and total E&G expenditures per FTE student (odds-ratio = 1.146, p<.05).

The results of this research show that a student’s chance of college completion is not influenced by the percent of revenue derived from state appropriations, contracts and grants, and endowment income. This study also reveals that the odds of college completion are not associated with the percent of expenditures on administrations, instruction, research, student services, or scholarships and grants. Because the lack of significance among most of the institution-level variables may be a result of inflated standard errors due to multicollinearity, diagnostics were conducted in two stages. First, institution-level residuals from the multilevel model were used to generate empirical Bayes (EB) estimates of institutional graduation rates, which were adjusted for student-level predictors. The adjusted graduation rates were regressed on the institution-level variables, with the regression diagnostics including tolerances (TOLs) and variance inflation factors (VIFs) to detect problems with multicollinearity. Although the results of the diagnostics revealed that the TOL (>.10) and VIF (<10) values were at acceptable levels (Cohen et al., 2003), bivariate correlation analysis did, however, reveal that sources of market-based revenues such as tuition, endowment income, and grants and contracts are interrelated. Further research is needed to examine how the likelihood of college completion and the time to degree may be influenced by changes over time in different types of market-based revenue and expenditure patterns.

Descriptive Statistics

Consistent with the observations of other researchers (e.g., Walpole, 2003), in this study, the descriptive statistics show that low SES students have different values for student characteristics and experiences, compared to high SES students. Table 4 shows that, by the end of their freshmen year, students in the first SES quartile and second SES quartile earn a lower GPA, are less likely to declare a major, are less involved on campus, have higher unmet financial need, and more likely to work more hours than their counterparts in the third SES and fourth SES quartile. The descriptive statistics also show that, compared to high SES students, low SES students are more likely to be a member of an underrepresented minority group, either African American or Hispanic.

Table 4 Descriptive statistics of Fall 1996 First-time Full-time Freshmen by Socioeconomic Status Quartile, Graduation Status as Spring 2001, Student-level, and Institution-level Predictors of College Completion
Full size table

The descriptive statistics reveal that low SES students disproportionately attend low selectivity institutions. While the HGLM analyses suggest that compared to high SES students, for low SES students who enroll in a four-year institution, college completion is positively related with attending a diverse, private or high selectivity four-year institution, the descriptive statistics show that compared to high SES students, low SES students are more likely to be enrolled in public and lower selectivity institutions. While the HGLM analyses suggest that aspects of the institutional financial context influence the chance of college completion, the descriptive statistics reveal that, compared to high SES students, low SES students are enrolled in institutions with less available financial resources, one indicator of institutional quality.

While the HGLM analyses suggest that SES has a positive contextual effect on college completion, the descriptive statistics show that compared to high SES students, only a small percent of low SES students attend institutions with a high concentration of high SES students. Table 4 reveals that only 8% of students in the first SES quartile and 14% of students in the second SES quartile attend institutions in the 75th percentile with regard to the average SES of students. In contrast, only 7% of students in the fourth SES quartile attend institutions in the 25th percentile with regard to the average SES of students.

The descriptive statistics show that compared to high SES students, low SES are less likely to attend a private institution. Table 4 demonstrates that one-third of students in the first SES quartile and 35% of students in the second SES quartile attend private four-year institutions, compared to 46% of students in the fourth SES quartile.

The descriptive statistics show that compared to high SES students, low SES students tend to enroll in institutions that receive a high percentage of their revenues from tuition and fees. Table 4 reveals that compared to 20% of students in the fourth SES quartile, 31% of students in the first SES quartile enrolled in the top 75th percentile of institutions in terms of tuition as a percent of total revenue.

While the HGLM analyses show that college completion is positively related to E&G expenditures per student FTE, the descriptive statistics show that students from low SES are disproportionately enrolled in institutions with low E&G expenditures per FTE student. Table 4 reveals that almost a third of students in the two lowest SES quartiles attend institutions at the bottom 25th percentile in terms of E&G expenditures per FTE student, compared with 22% of students in the third SES quartile and 13% of students in the fourth SES quartile. In contrast, only 14% of students in the first SES quartile and 18% of students in the second SES quartile attend institutions in the top 75th percentile in terms E&G expenditures per FTE student, compared to 26% of students in the third SES quartile and 40% of students in the fourth SES quartile.

While the HGLM analyses suggest that college completion is positively related to institutional selectivity, the descriptive analyses reveal that compared to high SES students, a substantially low percent of low SES student attend highly selective institutions. Table 4 shows that 13% of students in the first SES quartile and 20% of students in the second SES quartile compared to 28% of students in the third quartile and 42% students in the fourth SES quartile attend institutions in the 75th percentile with regard to selectivity. Although not shown, average SES and institutional selectivity are moderately correlated (r=.40, p<.001).

Conclusions

Five conclusions may be drawn from this research. First, the results of this study suggest that, after taking into account other variables, college completion is positively influenced by SES. Compared to students from high SES groups, students from low SES groups are less likely to complete a bachelor’s degree. This finding is consistent with previous research on college retention (Paulsen and St. John, 2002) and four-year college degree completion (Terenzini, Cabrera, and Bernal, 2001; Walpole, 2003).

Second, this research finds that, even after taking individual SES and other variables into account, SES has a contextual and positive effect on college completion. Consistent with earlier (Astin, 1993) and more recent (Winston and Zimmerman, 2004) research, the results of this study suggest that certain norms and values associated with higher levels of SES may help to increase the odds of college completion for all students.

Third, the results of this study suggest that college completion is positively influenced by such demographic-structural characteristics as racial/ethnic diversity. Because this study finds that compared to White students, African American and Hispanic students have a lower chance of completing college, this research also suggests that structural racial/ethnic diversity may have positive spillover effects for Whites, but not for under-represented minorities. It may be that because low SES students are most likely to be either African American or Hispanic, for either of these groups, the chance of college completion is most likely to be negatively influenced by low parental income rather than race/ethnicity. In addition to income, family wealth, which can be transferred across generations, provides an added degree of economic security for families to make human capital investments (Becker, 1962) on behalf of their children, can be a source of unearned income and may be used to accumulate cultural and social capital (Orr, 2003). According to research by Conley (1999, 2001), educational attainment is positively influenced by wealth, even after parental income and other variables are taken into account. Using an expanded definition of class that includes measures of family wealth, more research is needed to further understand racial/ethnic gaps in college completion.

Fourth, even after controlling for structural racial/ethnic diversity, selectivity, the contextual effect of SES, and other variables, this study finds that college completion is positively related to such financial aspects of the institutional context as tuition revenue as a percent of total revenue and E&G expenditures per FTE student. This research provides further evidence that the source and level of financial resources contribute to an environment that fosters student success.

Fifth, this research shows that, compared to high SES students, low SES students are disproportionately enrolled in institutions with lower levels of financial resources and a higher dependence on tuition as a source of total revenue. While it does provide some evidence that a positive relationship exists between college completion and percent of revenue from tuition, this study also suggests that, compared to high SES students, low SES students are enrolled institutions that have fewer available financial resources to help students persist towards graduation.

Implications

The results of this study have several implications for policy and future research. First, because it finds that the odds of degree completion for low SES students is higher for those students who enroll in more socio-economically diverse institutions, this study suggests that policies should be developed to further encourage institutions to adopt class-based admission criteria, which could result in the enrollment of a larger proportion of low SES students in a wider range of four-year institutions. Public policy makers should consider class-based admission policies which, in addition to addressing access for low SES students (Brown, Butler, and Donahoo, 2004; Kahlenberg, 2003), may also help to increase the odds of college completion for low SES students.

Second, because it also finds that, in general, the odds of college completion are higher and low SES students are less likely to enroll in private institutions, this study also suggests that policies should be further developed and supported to provide low SES students an opportunity to enroll in such institutions. Such policies may include providing low SES high school students and their families with more information on private colleges and universities as a part of pre-college outreach programs (Swail, Redd, and Perna, 2003).

Third, this study found that increased reliance on market-based revenues, specifically tuition, may result in institutional behavior and activities that lead to improving college student completion rates, with positive spillover effects for low SES students. Because it also finds that college completion is negatively influenced by unmet financial need, this study also suggests that policies should be put in place to help target financial aid to low SES college students so as to reduce the possible negative effect of increases in unmet financial need which may accompany increases over time in tuition as an offset to decreases in state appropriations to four-year institutions. In addition to the efforts that will be needed to help insure access for low SES students (Heller, 2002; Spencer, 2002), federal, state and institutional aid, will also be needed to increase the chance of college completion for low SES students after they enter four-year institutions.

Fourth, because it finds that college completion is positively related to E&G expenditures per FTE student and, compared to high SES students, low SES students are more likely to be enrolled in four-year institutions with relatively low E&G expenditures per FTE student, this study also suggests that policies should be promoted to provide more general subsidies to institutions. Because high subsidies and subsequent institutional spending on general and education related activities may enable institutions to increase access for and the chance of persistence of all students, including those from low SES backgrounds, both an increase in equity and efficiency may be realized as a result increased funding for four-year institutions. Future research should include measures of general subsidies generated from physical assets and savings or institutional wealth (Carbone and Winston, 2004). Institutional wealth may be a source of institutional economic, cultural (Bourdieu, 1986), and social capital (Coleman, 1988), all of which may facilitate an institution’s future efforts to contribute to students’ degree completion (Berger, 2000). Future research should explore the extent to which changes over time in the distribution of institutional wealth and expenditures per FTE student influence changes in the distribution of students across institutions and college completion rates by social class.

Finally, by further demonstrating the need to conceptualize college persistence models to include constructs that reflect aspects of the financial institutional context in which four-year institutions operate, the results of this study have implications for future research. With regard to examining the SES gap in college persistence, future research should carefully examine the relationship between the chance of college completion by social class and a state’s higher education policy context. By carefully examining the influence of various contexts on college completion, researchers may be able to better identify and understand the forces that impact the chances of college completion for low SES students.