Effectively Maintaining Inequality in Toronto: Predicting Student Destinations in Ontario Universities


  • The authors thank Robert Brown from the Toronto District School Board for providing access to the data and running the models. We also thank the SSHRC funded Postsecondary Pathways Project team Paul Anisef, Chris Conley, Kristyn Frank, Maria Adamuti-Trache, Robert Sweet, David Walters, and Gillian Parekh for their feedback. Portions of this study have been presented at the 12th, 13th, and 14th National Metropolis Conference and Congress 2012.


L'accès aux universités prestigieuses, les mieux classées et dotées de ressources, quoique peu étudié, représente une dimension additionnelle des inégalités en éducation au Canada. La théorie de l'inégalité maintenue efficacement (IME) soutient que les groupes favorisés vont dominer l'accès aux institutions les mieux classées peu importe le palier scolaire. Cet article teste cette hypothèse en utilisant les données uniques de milliers d’élèves du Conseil Scolaire Public de Toronto (TDSB) qui ont été suivis à partir de la neuvième année jusqu’à leur entrée dans un établissement postsecondaire. Ces données ont ensuite été associées aux données de classement des universités, de leur revenu, de leurs dépenses et de leurs fonds de dotation. Une série de modèles statistiques à niveaux multiples indique que l'entrée dans la hiérarchie universitaire ontarienne tend à refléter les inégalités dans l'accès général aux universités. Les femmes, les étudiants d'origine asiatique, et les étudiants issus des quartiers ayant des statuts socio-économiques élevés sont plus susceptibles d'entrer dans les universités les mieux classées et dotées de ressources; tandis que les étudiants qui s'identifient comme Noirs et hommes, sont moins susceptibles d'entrer dans ces institutions. Les avantages du statut socio-économique élevé et de l'origine asiatique sont seulement partiellement expliqués par les variables académiques comme variables médiatrices. Ceci suggère que le statut culturel joue un rôle dans l’élaboration du choix universitaire, alors que le sexe ainsi que les autres inégalités raciales sont dus en grande partie aux processus du parcours académique.

Access to highly ranked, prestigious, and well-resourced universities represents an additional yet understudied dimension of educational inequality in Canada. The theory of effectively maintained inequality contends that advantaged groups will dominate access to the best-positioned institutions within any credential tier. This paper tests this hypothesis using unique data on thousands of Toronto District School Board students that were tracked from Grade 9 to their entry in Ontario postsecondary institutions, and then linked to data on university rankings, incomes, expenditures, and endowments. A series of multilevel models shows that entry into Ontario's university hierarchy tends to mirror inequalities in general access to universities. Female, Asian-origin, and students from higher socioeconomic neighborhoods are more likely to enter higher ranked and better resourced institutions, while students who self-identify as black and male are less likely to enter such institutions. High socioeconomic status and Asian-origin advantages are mediated only partly by academic variables, suggesting that status cultures play a role in shaping their university choices, while gender and other racial inequalities emerge largely through academic processes.


UNIVERSITY ENROLLMENTS AROUND the world have been expanding for half a century, and policy makers see this growth continuing into the foreseeable future. In most nations, including Canada, virtually all social groups have boosted their access to higher education over this period. But sociologists have questioned whether this expansion has meaningfully reduced inequalities in attainment (Goldthorpe and Jackson 2008; Lucas 2001; Shavit, Arum, and Gamoran 2007; Walters 2000). Sizable disparities in access to Canadian higher education across a variety of social groupings were initially documented in the 1970s when modern university expansion was still in its infancy (e.g., Anisef 1974; Porter, Porter, and Blishen 1979), and yet despite much further expansion over the next 40 years, studies reveal mixed improvements at best.

On the one hand, gender and racial patterns have changed somewhat in Canada. For the past quarter century, women have attended universities in greater proportions relative to men (Frenette and Zeman 2007), though gender segregation across fields of study continues to be marked (Andres and Adamuti-Trache 2007). Female advantages in enrollments appear to be mediated largely by academic mechanisms; male underachievement is evident from the earliest primary grades and is quite pronounced by high school (Anisef et al. 2010; Bowlby and McMullen 2002; Brown 2006). Disparities by race are mixed. Immigration has altered the face of the Canadian undergraduate population, with Asian-origin youth enjoying above-average rates of attendance, though black and especially aboriginal students have far lower rates (Anisef et al. 2010, 2011; Thiessen 2009). Some argue that Asian students’ higher academic achievement is mediated by their superior academic performance, which in turn is rooted in cultural factors such as their parents’ high expectations (Goyette and Xie 1999; Thiessen 2009).

On the other hand, disparities in access to higher education clearly persist by socioeconomic status (SES) (de Broucker 2005; Looker and Lowe 2001; Willms 2004). While Canadian postsecondary participation continues to rise and be relatively high in international comparison, the proportion of participants from lower income families has remained static for the past two decades (Deller and Oldford 2011). These SES inequalities appear to arise through inequalities in resources (e.g., parental financial support), academic processes (e.g., grades), and status culture practices (e.g., information gathering, planning, and encouragement for postsecondary study; Berger, Motte, and Parkin 2007; Frenette 2007).

Some theorists characterize this simultaneous process of ongoing expansion and persisting socioeconomic inequality as a form of “maximally maintained inequality” (MMI). MMI suggests that advantaged groups migrate to more advanced credential tiers when less advantaged groups are able to access lower tiers en masse (Raferty and Hout 1993). In the decades following WWII, the near-universal entry of youth into high schools prompted a great wave of higher education expansion. MMI theorists contend that when higher status groups sense that competition for better-paying jobs is intensifying, they pursue higher tiers of education to retain their advantages. Today, proponents of MMI would trace the expansion of professional and graduate tiers in part to widened access to university baccalaureate programs among lower status groups.1

But in today's context, an additional dimension of educational inequality may be emerging within higher education. Building on past Canadian research that linked social origins to postsecondary entry (e.g., Anisef 1974), newer work is premised on the notion that educational organizations within any tier are stratified by resources and prestige (Davies and Zarifa 2012). Well-positioned schools, institutions, fields of study, and programs can confer to their graduates more social recognition, cognitive resources, and/or labor market advantages than can other entities. When this stratification congeals to a certain level, higher education institutions and fields can form a hierarchy.2 Clearly arrayed hierarchies then encourage students to jockey for the most favorable positions within them.

Lucas (2001) describes EMI as a process in which widened access to a credential tier encourages privileged groups to migrate toward its most advantageous, selective, and prestigious sectors. If MMI describes how inequality is maintained through upward movements between credential tiers, EMI describes how inequality is maintained through lateral movements within a tier. For instance, wealthy Canadians may increasingly seek entry into elite private high schools in order to attain a more advantageous education than one obtained from a public high school.

This sorting process can produce disparities. Studies of America's highly stratified higher education system have examined race, class, and gender influences on access to and payoffs from institutions and fields of varying rank (Ayalon and Yogev 2005; Davies and Guppy 1997; Espenshade and Radford 2009; Gerber and Cheung 2008; Goyette and Mullen 2006; Grodsky and Jones 2007; Jacobs 1999; Mullen and Goyette 2010; Mullen, Goyette, and Soares 2003; Zarifa 2012a). Entrants into the top-ranked institutions tend to be elite in both academic and social terms (Espenshade, Chung, and Walling 2004), while female, black, and lower SES origin students tend to attend less elite institutions, though race and gender associations sometimes disappear after academic factors are controlled (Gerber and Cheung 2008:310).3

EMI may be intensified in today's academic climate. As national systems of higher education around the world continue to expand, many states are pushing them to undergo an additional transition under the banners of human capital formation and wealth creation. Policy makers are urging universities to become more differentiated, competitive, and entrepreneurial, whether by building partnerships with corporations, bidding for research contracts, winning massive private donations, or luring more and more tuition-paying students (for a Canadian discussion, see Clark et al. 2011). These competitive forces can combine with enrollment expansion to intensify processes of EMI in several ways.

First, some expansion occurs through the founding of new institutions, such as university colleges, for-profit and online universities, or new teaching-oriented campuses (Clark and Mayer 2011; Ruch 2001). Such institutions typically survive by digging niches in vocationally oriented programs that are used to court “nontraditional” students—older pupils with less salutary academic profiles. This process of “anticipatory subordination” allows new institutions to survive while minimizing direct competition with higher ranked institutions for top students (Brint and Karabel 1989). Second, these forces create incentives for universities to vie for status, resources, and high rankings, whether by recruiting high-profile researchers, erecting prestigious professional schools, or mounting ambitious fundraising campaigns (e.g., Brint, Riddle, and Hanneman 2006; Kirp 2004; Tuchman 2009). Third, once in motion, these processes can simultaneously encourage the most academically and socially advantaged students to self-select into the most resource-rich and reputation-rich universities, while at the same time encouraging those universities to become more selective in their admissions.4 These processes can further entrench institutional stratification and hierarchies.

In the United States, these mechanisms are generating a polar pattern of enrollment. While most of the growth in American universities has been in “broad access” institutions that are not particularly selective (Astin and Oseguera 2004; Kirst, Proctor, and Stevens 2011), students with the highest GPA's and SAT's are increasingly concentrating themselves in the most selective institutions (Hoxby 2009). Top-ranked Ivy League schools have become increasingly selective in recent years while many other institutions have become less selective than they were even 50 years ago (Hoxby 2009).


Originating in the United States, EMI theory assumes the existence of clear hierarchies of institutions and fields within any credential tier. At the K-12 level, American proponents point to elite private schools and newer public schools of choice that are better endowed, more socially exclusive, and offer surer routes to prestigious universities than standard public schools (Espenshade and Radford 2009; Mullen and Goyette 2010). A Canadian proponent could likewise point to this country's growing private sector in K-12 education, along with publicly subsidized independent schools and other schools of choice (Davies and Aurini 2011; Davies and Quirke 2007). Yet, applying EMI theory to Canadian higher education requires rethinking some important details.

Most importantly, Canadian universities are less hierarchically arrayed than American universities (Davies and Hammack 2005). Almost all Canadian universities are publicly-governed and, until recently, received provincial funding premised on norms of institutional parity. Moreover, Canada's undergraduate market is mostly local or provincial in scope, with relatively few students crossing provincial borders to attend baccalaureate programs (notwithstanding notable exceptions of Queens, McGill, and some Nova Scotia universities). And there is no Canadian equivalent to the world-renowned Ivy League, whose budgets can rival those of many entire nations, not to mention universities. As a result, Canadian universities are far less stratified by resources and selectivity than are American universities (Davies and Zarifa 2012). Canadian students enter a less-clearly established hierarchy of universities. Still, there are reasons to suspect EMI-like processes are operating in Canadian higher education, and perhaps increasingly so.

First, Canada may be a prime candidate for EMI in international perspective, given its relatively large proportion of university graduates and comparatively weak association between socioeconomic background and educational attainment (Beller and Hout 2006). According to Beller and Hout (2006), greater numbers in the baccalaureate tier serve to intensify competition, prompting higher SES students to seek advantages in lateral directions, perhaps by attending better-resourced institutions. Similarly, less-differentiated higher education systems tend to have higher rates of enrollment and weaker social exclusivity (Shavit, Arum, and Gamoran 2007). Canadian higher education is differentiated by several sectors—community colleges, private career colleges, and universities—but the latter has expanded the most. This combination of expansion and differentiation may produce less stratification in access to higher education (MMI) but may also trigger considerable amounts of EMI.

Second, while resources are distributed relatively evenly among Canadian universities compared to those in other nations, particularly the United States, Canada's stratification is nonetheless substantial and appears to be slowly growing (Davies and Zarifa 2012). The presidents of Canada's best-resourced research universities—the “U15”—have for 20 years pressured governments to fund them at levels higher than other universities, to permit them to raise their tuition to unprecedented levels, and/or to become semiprivate. In Ontario, policy makers are urging their universities to “differentiate” in ways that might further exacerbate these inequalities (Clark and Mayer 2011; Weingarten and Deller 2010). Also, Canadian universities are enjoying vastly differing successes in their private fundraising ventures, which tend to advantage older and more established institutions.5

Third, Canadian undergraduates, particularly from relatively privileged origins, may be increasingly attuned to these hierarchies. An Ontario study from 1979 found that established institutions such as the University of Toronto, McMaster, Waterloo, and Queens tended to recruit more students from higher status origins than did less-established institutions (Anisef 1982).6 But over the past 20 years, the rising visibility of the Maclean's rankings7 may have moulded a “rank consciousness” among some students. While those rankings are routinely criticized by professors and administrators, they correlate highly with university resources (discussed below). Further, Ontario institutions that are higher ranked by Maclean's, particularly in the medical-doctoral category, tend to receive more applications, as do institutions that raise their rank (Mueller and Rockerbie 2005), though this may occur only among smaller institutions (Drewes and Michael 2006). Canada-wide, Kong and Veall (2005) find some evidence that medical-doctoral universities that raise their rank then receive applicants with higher high school averages. Thus, if academically and/or socially elite students are particularly prone to consult rankings when choosing universities, they might increasingly concentrate themselves in better-resourced institutions.

Qualitative evidence from Ontario suggests affluent youth indeed perceive a hierarchy of universities in terms of their social caché and/or academic standing.8 Likewise, the geography of student choice of university across Toronto illustrates how affluent youth avoid lesser ranked institutions. Figure 1 maps the home location (using postal code) of students who eventually attended Queen's and York in 2006. Among the many Torontonians who chose York (signified by dots), relatively few lived in the city's most affluent neighborhoods, while the converse was true for those who attended Queen's (signified by triangles). The concentrations of Queen's attendees near the center, west end, and southeast correspond to the Yonge Street corridor, Bridal Path, Kingsway, and Beaches—all among the wealthiest neighborhoods in the Greater Toronto Area (GTA). In contrast, almost no Queen's attendees hailed from humbler areas like York, Rexdale, Downsview, and Scarborough. York attendees are scattered throughout the city, but are less concentrated in the aforementioned affluent areas.

Figure 1.

Students who accepted offers of admission to York University and Queens University.

Source: TDSB Research and Information Services, Facilities Planning.


We test EMI theory in a Canadian setting by asking the following questions: First, what kinds of students in Toronto attend Ontario's higher ranked universities? Are there differences by SES, gender, or race? Older American studies find that racial minorities and women are less likely to attend higher ranked universities (Gerber and Cheung 2008), suggesting that EMI tends to reinforce older forms of educational inequality. In contrast, a plausible counterhypothesis is that since today's MMI is mixed—women and some racial minorities are overrepresented in higher education, while SES disparities have remained stagnant over time—we may similarly observe greater proportions of women and minorities in higher ranked universities. Second, what variables mediate student entry into higher ranked universities? In particular, do academic measures such as high school grades predict such entry? If highly ranked universities require higher grade averages, we would expect grades to mediate some of the effects of social origins on entry. However, if those effects remain partly unexplained, they may also reflect status-group processes, that is, the tastes, expectations, and/or social pressures that shape high-SES groups’ preferences for universities, or shape mobility strategies for nontraditional university students.

Toronto provides an excellent testing ground for EMI theory. As Canada's largest school board, the Toronto District School Board (TDSB) offers the large numbers needed to examine subgroups by race, class, and gender. TDSB graduates have ample choice among universities, including 18 institutions within their province, four of these within their city, and several more within commuting distance. Among these universities there are striking differences in institutional age, ranging from the venerable University of Toronto (founded in 1850) to institutions that have acquired university status much more recently (Ryerson University in 1993, and the Ontario University Institute of Technology and the Ontario College of Art and Design University each in 2002), as well as sizable variations in research intensiveness and endowments. By comparing university destinations among students from a single city, we control for geographic proximity between their homes and their institution, and thus analyze how other factors shape outcomes.

Our design has some limits and inherent trade-offs. One limitation is its focus on one city. We do not know if we can generalize findings from Toronto to other Canadian regions. Similar processes are likely at work in other large cities, since there are no Canadian studies (to our knowledge) that show radically different patterns of higher education access. But different processes may be operating in less populated regions with fewer immediate universities. In such locales, student must weigh different issues when choosing, such as the psychic and financial costs of moving and switching communities (Looker and Naylor 2009). But EMI could be even stronger in outlying regions, at least among more affluent youth. Since those youth must leave home to attend any university, the certainty of those costs may motivate them more to attend a higher status institution. Future research is needed to verify either scenario. Also, Canadian provinces have different ways of governing colleges and universities: Ontario has a relatively rigid divide between its colleges and universities that restricts flows of students between those sectors, while other provinces such as British Columbia and Alberta have larger such flows (see Alberta Enterprise and Advanced Education 2012; Heslop 2012). These varying arrangements might complicate EMI processes, and empirical data from other provinces is needed to sort out their effects. We also lack information on the small percentage of TDSB graduates who applied to universities outside of Ontario, and for graduates from Toronto private and Catholic schools. Furthermore, we have to use a proxy for a crucial SES variable—family income—at the neighborhood level, which undoubtedly adds measurement error to our estimates. However, these limitations likely cause our models to underestimate the current extent of EMI, especially for SES. Students in Toronto private schools are likely to be far more affluent than those in the TDSB, given the cost of private tuition, which Ontario does not subsidize. And Toronto high school graduates who attend American universities, as well as out-of-province universities like McGill, are also likely to be affluent, due to costs of tuition and moving.9


Data Sources and Sample

We merged four data sets to track an entire cohort of TDSB students from the ninth grade to an Ontario postsecondary institution. The first database consisted of official records for all students in the TDSB who were 17 years of age in the fall of 2006 and attended a regular day school.10 This cohort totalled 19,082 students. It omitted students who earlier dropped out of school or transferred out of the TDSB later that year. This core data set was merged to a second data set from the Ontario Universities’ Application Centre, which records all applications and confirmed attendance for all Ontario postsecondary students in Ontario universities. Using student identification numbers, we linked any student from the TDSB cohort who confirmed attendance at an Ontario university in any one of three years (2007, 2008, and 2009). A third merged data set came from the TDSB “Student Census.” In the fall of 2006, all TDSB students in Grades 7 to 12 were surveyed about their demographics and school attitudes. Useable data were collected for approximately three-quarters of the 17-year-old cohort. And finally, to measure the hierarchy of Ontario universities, we merged data on each university's resources and ranking, the bulk of which was compiled by David Zarifa (2008). The institution-level data set contains measures of 2006 financial resources—annual income and expenditures—for almost all Ontario universities. These data come from several annual Statistics Canada surveys: the Financial Information of Universities and Colleges (FIUC), Tuition and Living Accommodation Costs Survey (TLAC), and the University Student Information System (USIS).11 These resources are critical for research, teaching, and operations, and can indirectly shape quality of education by affecting student-teacher ratios, physical plants, ranges of courses and programs, faculty salaries (which is correlated with productivity), and capacity to recruit students. We then supplemented these data with measures of university endowments in 2006 and with three years of Maclean's rankings.

This merged data set has several advantages for testing EMI theory. First, it uses a three-year time frame after graduation, thus capturing any students who did not apply to university immediately after graduation. Second, and most crucially, it identifies which Ontario university students eventually attended. Such data are rare in Canada. Available versions of the National Graduate Survey and Youth in Transition Survey, for instance, do not identify universities. Third, these data are also comprehensive. They contain an extensive set of measures from students’ official high school academic records, including their residential postal code (used to calculate distance between their home and their university of choice) and grades. Coming from administrative sources, these variables have far less measurement error than would those that relied on survey self-reports. Fourth, these variables have a host of demographic and attitudinal measures. And finally, by using multiple outcome measures, we were able to explore several related dimensions of institutional inequality, providing a robust test of the EMI thesis and enhancing the reliability of our conclusions.

Our analytic sample has several restrictions, however. Since we are testing EMI theory for the university tier, we exclude students who were admitted to Ontario community colleges or who did not apply to any postsecondary institution. Further, these data do not include any registrants at universities beyond Ontario. The TDSB estimates that 2 to 3 percent of its students attend postsecondary institutions outside Ontario; those students tend to be concentrated in a small number of socioeconomically advantaged schools (Brown 2009).12 After these restrictions, our original analytic sample consisted of 8,614 students who were 17 years old in 2006 and were admitted to an Ontario university within three years (2007–2009). Of these, approximately 75 percent wrote the student census. Thus, our analytic samples range from 6,479 to 6,420 students across our models. See Table 4 for frequencies of attendance at each Ontario university.

Our statistical analyses consist of multilevel regression models that estimate the effects of social background on student attendance within a hierarchy of universities. Our focal interest is on tracing the effects of student background (SES, gender, and race). Our models first estimate effects of SES, gender, and race, while controlling for other demographics. Model 1 includes controls for a series of additional demographic variables, namely whether or not students attended the same high school between Grades 9 and 12 and whether or not they live with two parents, and distance between their home in high school and their eventual university. These are good control variables because studies show that each is related to the likelihood of attending a postsecondary institution. Models 2 and 3 proceed to examine how those effects are mediated by measures of students’ academic experience and achievement. These are important mediating variables since both are related to social background and to postsecondary attendance. We have separate models for four related outcomes: average Maclean's rank over three years, expenditures per full-time equivalent (FTE) student, income per FTE student, and endowments per FTE student.13 In addition, we estimated multilevel logits to examine which students are more likely to attend one of two highly ranked universities outside of Toronto: Queen's and McMaster. These two institutions have the greatest resources per student among universities outside of Toronto.14


Table 1 presents descriptive statistics for all variables in our models. Our independent variables include gender, self-identified race, SES, and several key controls. SES is measured by two variables: whether or not at least one parent had completed university, and the median family income in each student's neighborhood, measured at the dissemination area level, using the student's postal code.15 Other demographic controls include a dummy variable for whether students had previously switched high schools and a continuous measure of the distance between students’ homes and the universities they attended.

Table 1. Descriptive Statistics of Study Variables
 (N = 8,614 students)  
Dependent variablesVariable descriptionMean/proportion (SD)


  1. Standard deviations are in parentheses.

Maclean's rankingsMaclean's rankings averaged for 2007, 2008, and 2009 (min = 2.00; max = 20.00)6.07(3.33)
Total income per full-time enrollment (FTE) studentThousands of dollars in income university could spend per student based on student FTE for 2006 (min = 13.9; max = 41.96)29.30(9.13)
Total expenditures per FTE studentThousands of dollars university spends per student based on FTE for 2006 (min = 13.75; max = 39.99)27.41(8.09)
Total endowments per FTE studentDollar value of university endowment per student for 2008 (min = 2,113.29; max = 36,300.32)16,822.73(12,973.81)
Level 1 variables   
MaleDummy variable = 1 if male0.46(0.50)
Same schoolDummy variable = 1 if student present in same school between 2003–2004 and 2006–20070.85(0.35)
Parent completed universityDummy variable = 1 if one or more parent completed university0.62(0.49)
Lives with two parentsDummy variable = 1 if both parents living in household0.82(0.39)
Self-identified raceDummy variables (reference = white)  
 Southeast Asian0.03(0.16)
 South Asian0.23(0.40)
 East Asian0.30(0.46)
 Middle eastern0.04(0.20)
DistanceDistance (km) from student residence to university institution (min = 0.25; max = 936.00)57.08(88.05)
Social engagement   
Social relationsScale: Chronbach's alpha = 0.73; Combination of three questions: (1) students getting along with other students in the school, (2) students feeling accepted by students in the school, (3) students feeling accepted by adults in the school. Unstandardized scale range 1–5; 5 = never4.13(0.76)
Student perceptions of instructionScale: Chronbach's alpha = 0.77; Combination of six questions: (1) teacher expects student to succeed in school, (2) satisfied with the way teachers teach, (3) feeling supported and encouraged by teachers, (4) feeling comfortable discussing problems with teachers, (5) school's staff respect background (e.g., cultural, racial, religious), (6) extra help is available when needed. Unstandardized scale range 1–5; 5 = none of them; for extra help 5 = never3.96(0.85)
School climateScale: Chronbach's alpha = 0.82; Combination of three questions: (1) enjoying school, (2) school is a friendly and welcoming place, (3) school building is an attractive and great place to learn. Unstandardized scale range 1–5; 5 = never3.47(0.87)
Average Grade 11/12 markAverage from all courses in 2006–2007 (min = 13.00; max = 99.00)76.81(9.55)
Grade 9 math markGrade 9 math or first math mark (min = 0.00; max = 100.00)75.30(15.11)
Level 2 variables   
School level   
Neighborhood family incomeMedian family income from 2001 census in thousands of dollars (min = 15.00; max = 110.00)61.49(24.14)
Social assistanceThe proportion of families in a student's neighborhood whose income comes from government sources (based on data about families with children) (min = 3.00; max = 109.00)76.79(23.55)
Lone parents familiesThe proportion of families in a neighborhood where the parent does not live with either a spouse or common law partner (based on data about families with children) (min = 1.00; max = 109.00)74.83(28.93)
LanguageProportion of students speaking English only (and a language other than English) in the school (min = 9.17; max = 92.24)58.11(20.77)

Distance is an important control variable. Given the high cost of attending an institution far from home, distance can affect whether or not students attend a university and, if so, which one they attend. Lopez-Turley (2009) found that the odds of applying to a U.S. four-year college increased significantly as the number of nearby colleges increased, even controlling for student and neighborhood-level factors. In Canada, Frenette (2009) examined the impact of newly created universities on postsecondary participation rates. The emergence of a new university served to increase university participation rates among local youth, though this trend occurred at the expense of college participation rates. Lower income families experienced the greatest increases in university participation, suggesting that greater distance does indeed represent a financial barrier. Other studies show that Canadians in rural areas not in close proximity to a university are less likely to attend a university, controlling for gender, family income, and parental education (Frenette 2003; Looker and Naylor 2009). Our expectation for TDSB students is somewhat different: Since all Toronto youth are in close proximity to several universities, advantaged students may be likelier to attend higher ranking universities outside of Toronto, due to their greater costs.

To investigate the extent to which student experiences mediate the effects of social background, three attitudinal scales about student experiences at school were created from survey items. Each has an acceptable level of internal consistency (Cronbach's alphas are reported in Table 1). Students’ social engagement at school was constructed from three survey questions about how well students get along, how accepting students are of each other, and how accepting adults in the school are of students. Another scale addresses students’ perceptions of instruction using items on whether students thought teachers expected them to succeed, their satisfaction with the teaching at their school, how much they felt supported and comfortable sharing with teachers, how well they felt staff respected their cultural, racial, or religious background, and the extent of extra help available. This variable was dropped from analyses because it did not improve model fit. A school climate scale was devised from measures of how much students enjoy school, whether school is a friendly and welcoming place, and whether the school building is an attractive place to learn.

The final model adds several measures of educational achievement, including students’ average grades from Grade 11 or 12, and their Grade 9 math mark. Grade 11 and 12 marks are central to students’ applications to university. Grade 9 math achievement has been found to be a fairly reliable predictor of later academic achievement (Brown 2010).

We use several related outcome variables: the dollar value of each university's income and expenditure per FTE in 2006, the financial endowment per FTE of each university in 2008, each university's Maclean's rank within its own category, averaged over three years (2007, 2008, and 2009), and whether or not students attended particular institutions.16 Table 2 shows that these measures are strongly intercorrelated; despite their small number of cases, all correlations are highly significant.

Table 2. Correlations among University-Level Measures (n = 18)
 Average Maclean's rankIncome (2006, $ thousands/FTE)Expenditures (2006, $ thousands/FTE)


  1. Pearson coefficients are reported; calculating Spearman coefficients yields similar results. Highly ranked universities are assigned low values, that is, top rank = 1.

  2. Two-tailed *p < .05; **p < .01; ***p < .001.

Average Maclean's rank   
Income (2006, $ thousands/FTE)−0.641**  
Expenditures (2006, $ thousands/FTE)−0.635**0.960*** 
Endowment (2006, $ thousands/FTE)−0.561*0.822***0.769***

Table 3 displays the Ontario hierarchy by cross-tabulating a variety of institutional-level measures for each university and converting them to ranks within the province. Looking across columns of Table 3, many universities have fairly consistent ranks across each measure. Queen's is highly ranked—either first or second on all measures —followed by U of T, which is ranked between second and fourth on each. McMaster, Western, and Waterloo are also ranked highly across these measures. In contrast, the lowest tiers of Ontario universities appear to be occupied by Nipissing and Brock.

Table 3. Rankings of University-Level Measures, Ontario Universities
Institution nameInstitution typeRank by total income per FTE studentRank by total expenditures per FTE studentRank by total income from endowments per FTE studentMaclean's ranking 2007Maclean's ranking 2008Maclean's ranking 2009Maclean's averaged rankings 2007, 2008, 2009
Queen'sMedical doctoral1112222
TorontoMedical doctoral3324423
TrentPrimarily undergraduate1514134465
Wilfrid LaurierPrimarily undergraduate1616186656
McMasterMedical doctoral2236666
WesternMedical doctoral45477108
OttawaMedical doctoral771188109
LakeheadPrimarily undergraduate1115511111111
LaurentianPrimarily undergraduate991510101712
RyersonPrimarily undergraduate13101711111312
BrockPrimarily undergraduate17171915151415
NipissingPrimarily undergraduate18181620202020
Table 4. Confirmed Attendance by Ontario University (n = 8,614)
InstitutionFrequencyPercentDistance from TDSB headquarters (5050 Yonge Street, Toronto) (km)


  1. Three years of postsecondary confirmations (2007, 2008, 2009). We do not have complete institutional-level data for the University of Ontario Institute of Technology and have none for the Ontario College of Art and Design. Neither of these institutions is included in the Maclean's rankings.

Wilfrid Laurier2262.698.8

Data Analysis

This study uses a series of multilevel models to examine the effects of the individual-level variables and school-level variables. These data have a hierarchical structure, with students nested in secondary schools, and some variables measured at the school level. Since the errors are likely to be correlated within schools, we use multilevel techniques (Raudenbush and Bryk 2002; Snijders and Bosker 2012). Our analysis begins by fitting null models to examine between-group effects for each dependent variable (not shown). This model can be expressed as the following:

display math

where the dependent variable Y for student i nested in school j is the sum of a general mean, γ00, U0j representing a random effect at the school level, and Rij representing a random effect at the student level (Snijders and Bosker 2012). We then proceed to estimate linear mixed-effects models with a random intercept by sequentially adding four successive blocks of variables.17 Thus, the null model can be extended by adding individual- and school-level covariates:

display math

where x1,…,xp represents independent variables at the individual level, and z1,…,zq represents independent variables at the school level (Snijders and Bosker 2012).

We also estimate an additional series of mixed-effects logistic regressions using the xtmelogit procedures in Stata 12 (StataCorp LP, College Station, TX). These models are appropriate when estimating multilevel models for dichotomous dependent variables. The logistic random intercept model for a dichotomous dependent variable for student i nested in school j can be expressed in the following way:

display math

where the logit of Pij is expressed as a sum of a linear function of the independent variables (x1,…,xr) and a random group-dependent component, U0j (Snijders and Bosker 2012).18


We begin by exploring simple counts and percentages. Table 4 reports the Ontario universities attended by TDSB graduates. The most striking pattern is that almost 62 percent chose to remain in Toronto to attend university. Their top choice was the University of Toronto, followed by York and Ryerson. This pattern is consistent with the contention that Canadian markets for undergraduate education tend to be relatively local and dominated by commuters, unlike those in some European nations and U.S. regions (Davies and Hammack 2005). The next most popular choices tend to have high status within province: Guelph, Waterloo, Western, Queen's, and McMaster, which combine for another 25 percent of choices. The two universities thought to be most preferred by students from more socioeconomically advantaged backgrounds—Western and Queen's—combine for 8 percent. About 13 percent of all TDSB graduates chose one of the remaining Ontario universities, and very few attended the lowest ranked institutions outside of Toronto: only 1 percent attended Nipissing, Lakehead, Laurentian, or Windsor. Toronto high school graduates prefer universities that are either nearby or of relatively high status.

Table 5 shows multilevel regression results for entering an Ontario university that has been scored by its Maclean's rank averaged over three years.19 Ranks are reverse-coded (bottom rank = 1 and the top rank = 18), so that positive coefficients signal that increases in that variable are associated with higher ranks. An analysis of the variance components in the null model (not shown) suggests only about 4 percent of the variance lies among schools, and the remainder lies among students. However, the intercept is highly statistically significant, suggesting there are considerable differences among schools in the rates that their graduates apply to top-ranked universities. Model 1 examines the effects of social background, controlling for distances between a student's home address and their university of choice. The significant and negative coefficient for the “male” variable (p < .01) signals that females are more likely to attend higher ranked universities. This finding differs from older American studies, which found that males were slightly more likely to attend highly selective institutions. Our finding may be tapping newer gender patterns in which EMI outcomes simply mirror those of MMI; a second possibility is that higher ranked Ontario universities, unlike those in the United States, tend to offer a general array of programs, including those with high female enrollments like nursing and education (Jacobs 1999). In terms of student SES, the strong, positive parent-education coefficient (p < .001) suggests that students with highly educated parents tend to attend higher ranked institutions. In terms of race and ethnicity, students who identify as South Asian or East Asian (p < .001) tend to attend higher ranked institutions, while those who self-identify as black (p < .001) are less likely to do so.

Table 5. Linear Mixed-Effects Models with Random Intercept of Maclean's Rankings on Student Characteristics, Social Engagement, and Grades
 Model 1Model 2Model 3


  1. N = 6,479 students (Model 1), 6,436 students (Model 2), and 6,420 students (Model 3) in 84 schools. Additional models (not shown) included census measures of proportions of families in students’ neighborhood on social assistance, proportions of single parents, and proportions speaking English. None of these predictors significantly improved overall model fit. Wald tests are reported adjacent to variable names for blocks of dummy regressors. The distance coefficient has been multiplied by 1,000.

  2. *p < .05; **p < .01; ***p < .001.

Fixed effects      
Student characteristics      
Same school0.027(0.118)0.060(0.119)−0.086(0.116)
Neighborhood family income0.001(0.002)0.002(0.002)0.002(0.002)
Parent completed university0.711(0.089)***0.697(0.089)***0.421(0.087)***
Lives with two parents0.155(0.108)***0.157(0.108)***0.002(0.105)***
Self-identified race (ref: white)      
Southeast Asian0.236(0.268)0.234(0.267)0.092(0.259)
South Asian0.521(0.135)***0.500(0.135)***0.538(0.131)***
East Asian1.335(0.120)***1.344(0.120)***0.989(0.120)***
Middle eastern0.047(0.223)0.050(0.224)0.273(0.218)
Social engagement      
Social relations  −0.050(0.063)−0.082(0.061)
School climate  0.238(0.055)***0.154(0.054)**
Rounded mark    0.073(0.005)***
Math    0.021(0.003)***
Random effects      
Pseudo R20.032 0.036 0.099 
Deviance33,627.28 33,370.88 32,865.90 
AIC33,661.29 33,408.88 32,907.89 
BIC33,776.49 33,537.50 33,050.00 

Model 2 adds measures of perceived school climate and social relations. Their inclusion has very little effect on the existing patterns of coefficients. School climate has an independent effect (p < .001), suggesting that students who have a more positive image of their school's climate are more likely to attend higher ranked universities. Model 3 adds measures of student academic achievement. As expected, students with higher grades are significantly more likely to attend higher ranked universities (p < .001), since those universities typically have higher entry requirements. The addition of these academic measures decreases the gender coefficient substantially so that it is no longer statistically significant. This pattern suggests that males’ disadvantage is largely explained by their academic achievement. By contrast, the parent education coefficient shrinks by only 40 percent, and remains statistically significant (p < .001). Thus, SES effects on entering the university hierarchy are not explained away by academic performance. Race continues to play a role in determining university selection, despite the addition of academic characteristics. While the negative effect for black students is no longer statistically significant, the effects for South Asians and East Asians remain significant (p < .001).20

Overall, Table 5 suggests that many of the disadvantages faced by male and black students are largely due to academics, while the advantages of higher SES and Asian students remain net of their academic performance alone. As further discussed in the conclusion, these net effects may be due to unmeasured status cultural processes among some dominant groups, and strategies for upward social mobility for others.

Table 6 shows regressions for income per FTE of each university in 2006. Positive coefficients signal that increases in independent variables are associated with attending a university with greater resources. In Model 1, a sizable and significant gender effect again emerges. Males are significantly less likely than females to attend a well-resourced university. Similarly, all three SES coefficients—neighborhood income, parental education, and family structure—are positive and significant. Students from higher SES families are more likely to enter well-resourced universities. As in Table 5, students who identify as South Asian and East Asian are more likely to enter universities that have higher incomes, while self-identified blacks are less likely. Model 2 that adds attitudinal measures has only one additional significant and independent effect—for school climate (p < .001). In Model 3, adding academic measures substantially shrinks the male coefficient by about 75 percent, making it no longer statistically significant. The effects for neighborhood income, family structure, and education are also reduced, though income and education remain significant. A similar pattern occurs for the Asian-ancestry coefficients, while the black disadvantage is no longer significant after controlling for grades.21

Table 6. Linear Mixed-Effects Models with Random Intercept of Institutional Income per FTE Student on Student Characteristics, Social Engagement, and Grades
 Model 1Model 2Model 3


  1. N = 6,479 (Model 1), 6,436 (Model 2), and 6,420 students (Model 3) in 84 schools. Additional models (not shown) included measures of social assistance, family structure, and language at school, but none of these additional predictors significantly improved the overall model fit. Wald tests are reported adjacent to variable names for blocks of dummy regressors. The distance coefficient has been multiplied by 1,000.

  2. *p < .05; **p < .01; ***p < .001.

Fixed effects      
Student characteristics      
Same school0.032(0.048)0.045(0.048)−0.010(0.047)
Neighborhood family income0.002(0.001)**0.002(0.001)**0.002(0.001)**
Parent completed university0.295(0.036)***0.288(0.036)***0.182(0.035)***
Lives with two parents0.076(0.043)***0.073(0.043)***0.014(0.042)***
Self-identified race (ref: white)      
Southeast Asian0.090(0.108)0.089(0.107)0.034(0.104)
South Asian0.291(0.054)***0.279(0.055)***0.296(0.053)***
East Asian0.498(0.049)***0.505(0.049)***0.360(0.048)***
Middle eastern0.074(0.090)0.070(0.090)0.169(0.088)
Social engagement      
Social relations  0.002(0.025)−0.009(0.024)
School climate  0.086(0.022)***0.051(0.022)*
Average Grade 11/12 mark    0.030(0.002)***
Grade 9 math mark    0.008(0.001)***
Random effects      
Pseudo R20.055 0.059 0.12 
Deviance21,805.12 21,639.80 21,148.52 
AIC21,839.12 21,677.80 21,190.53 
BIC21,954.31 21,806.42 21,332.64 

Table 7 investigates a measure of the wealth of each institution: their financial endowments per FTE. As in the previous sets of models, Model 1 shows males are also less likely to enter wealthy universities. That effect is reduced by almost one-half once academic variables are entered in Model 3. Model 1 also shows a statistically significant effect for parent education (p < .001). The effect shrinks by 40 percent but remains significant, once academic performance is taken into consideration (Model 3). In terms of race, Models 1 and 2 suggest blacks may be less likely to attend wealthy institution, yet this effect greatly weakens and is no longer statistically significant in Model 3. Students with Asian ancestry are more likely to enter wealthy universities, even controlling for academic performance.

Table 7. Linear Mixed-Effects Models with Random Intercept of Endowments per FTE Student on Student Characteristics, Social Engagement, and Grades
 Model 1Model 2Model 3


  1. N = 6,479 students (Model 1), 6,436 students (Model 2), and 6,420 students (Model 3) in 84 schools. Additional models (not shown) included measures of social assistance, family structure, and language at school, but none of these additional predictors significantly improved the overall model fit. Wald tests are reported adjacent to variable names for blocks of dummy regressors.

  2. *p < .05; **p < .01; ***p < .001.

Fixed effects      
Student characteristics      
Same school1.010(2.069)1.625(2.081)−0.767(2.046)
Neighborhood family income0.057(0.035)0.058(0.035)0.055(0.034)
Parent completed university9.814(1.548)***9.506(1.553)***5.338(1.534)***
Lives with two parents1.646(1.882)***1.339(1.890)***1.029(1.850)***
Self-identified race (ref: white)      
Southeast Asian4.366(4.674)4.336(4.667)2.380(4.560)
South Asian8.665(2.363)***8.234(2.366)***8.996(2.323)***
East Asian16.583(2.109)***16.815(2.112)***11.643(2.117)***
Middle eastern6.607(3.895)6.447(3.914)10.297(3.842)**
Social engagement      
Social relations  −0.170(1.094)−0.670(1.070)
School climate  4.001(0.963)***2.721(0.944)**
Average Grade 11/12 mark    1.194(0.093)***
Grade 9 math mark    0.264(0.057)***
Random effects      
Pseudo R20.027 0.031 0.077 
Deviance70,690.76 70,196.36 69,699.00 
AIC70,724.77 70,234.36 69,741.00 
BIC70,839.96 70,362.99 69,883.11 

Table 8 shows the results of several multilevel logistic regressions predicting the likelihood of a student attending one of two top-ranked Ontario universities outside of the GTA: Queen's and McMaster. These models are estimated on only those students who attended institutions outside of the GTA. Many of the previous results hold. Males are significantly less likely than females to attend those higher ranked universities (p < .01), even controlling for academic performance in Model 3, though the effect does shrink. Similarly, parental education initially has a strong and positive effect (p < .05), but it shrinks and is no longer significant once academic variables are included in Model 3. That is, for students who left the GTA, parent education played less of a direct role in shaping which university they attended, once academic factors were taken into consideration. South Asians are significantly more likely to attend a highly ranked school (p < .01) controlling for all other factors. Interestingly, neighborhood income is also positive and significant (p < .05) and appears to have an importance influence on whether students attend a highly ranked institution outside of the GTA.

Table 8. Mixed-Effects Logistic Regression with Random Intercept of Attending a Top-Ranked Ontario University Outside of the Greater Toronto Area on Student Characteristics, Social Engagement, and Grades
 Model 1Model 2Model 3


  1. N = 2,464 students (Model 1), 2,451 students (Model 2), and 2,441 students (Model 3) in 84 schools. Additional models (not shown) included measures of social assistance, family structure, and language at school, but none of these additional predictors significantly improved the overall model fit. Wald tests are reported adjacent to variable names for blocks of dummy regressors.

  2. *p < .05; **p < .01; ***p < .001.

Fixed effects      
Student characteristics      
Same school−0.007(0.148)0.003(0.150)−0.059(0.155)
Neighborhood family income0.004(0.002)0.004(0.002)0.005(0.002)*
Parent completed university0.282(0.120)*0.261(0.121)*0.120(0.125)
Lives with two parents0.046(0.141)0.029(0.142)0.001(0.146)
Self-identified race (ref: white)      
Southeast Asian−0.497(0.549)−0.515(0.549)−0.458(0.555)
South Asian0.380(0.172)*0.345(0.174)*0.471(0.177)**
East Asian0.232(0.139)0.237(0.140)0.029(0.147)
Middle eastern0.492(0.372)0.454(0.373)0.745(0.385)
Social engagement      
Social relations  0.051(0.079)0.017(0.082)
School climate  0.117(0.070)0.079(0.071)
Averaged Grade 11/12 mark    0.050(0.008)***
Grade 9 math mark    0.008(0.005)
Random effects      
Deviance2,538.00 2,520.80 2,417.57 
AIC2,570.00 2,556.80 2,457.57 
BIC2,662.95 2,661.28 2,573.57 

In summary, an array of statistical models for different measures—university rankings, income, expenditures, endowments, and attending a highly ranked institution outside of Toronto—yield quite similar patterns of effects. Males tend to attend lower status universities, but those effects are mostly mediated through academic processes. Self-identified blacks are similarly disadvantaged, but that effect too is largely mediated by academics. In contrast, students that have higher SES and Asian ancestry are more likely to attend higher status institutions, and their advantages are only partly mediated by their higher grades.


This paper offers a first Canadian look at EMI. It reveals an additional dimension of inequality in higher education. Advantaged youth in Toronto are not only overrepresented in Ontario universities, they are also overrepresented in higher ranked institutions. EMI thus tends to mirror MMI. Some of these patterns emerged largely through academic processes. But we found a more complicated link between EMI and broader processes of social reproduction and mobility. Much reproduction occurs in Toronto, as suggested by the robust SES effects on our outcomes, and these were only partly explained by academic variables. We interpreted these effects as reflecting the status cultures of higher SES youth. Our map of applicants to Queen's University and Baker's (forthcoming) ethnography of elite Toronto private schools both support that view. But we also found that some groups that were overrepresented in higher ranked institutions are, in historical perspective, relative newcomers to Canadian higher education—women and students of Asian ancestry. For them, EMI represents social mobility, not reproduction per se, at least in the realm of education. Their academic advantages may not necessarily translate into large advantages in the labor market, since women's growing educational attainment in Canada is yet to be matched by a comparable economic payoff (Davies and Guppy 2013). We have yet to see whether choice of institution offers an effective vehicle of upward mobility for women or Asian-origin minorities.

If these processes of EMI are robust within and beyond Toronto, they raise a key question for Canadian stratification research: How consequential are they for larger patterns of social inequality? It is well known that graduating from a university boosts a variety of life chances, and that a student's field of study has a powerful influence on their future earnings. But does a university's position in a hierarchy have comparable effects? Ontario's (and, by implication, Canada's) hierarchy may not be steep in international comparison, but it is nonetheless a further layer of stratification, and it appears to sort Toronto students by their SES, gender, and race. American research clearly shows that graduates from highly selective universities have the best incomes and access to elite corporate and government positions, though it is less clear whether these advantages persist net of students’ prior social and academic characteristics (Gerber and Cheung 2008). In Canada, comparable research is in its infancy, partly due to a lack of longitudinal data that identifies where students attend university. As a result, we do not know if attending one Canadian university or another has marked influences on students’ life chances. But EMI may at least have an experiential component. An additional American literature is describing stark variations across institutions of higher and lesser status. Quantitative research suggests that top-ranked universities are becoming more and more exclusive, both socially and academically, while “broad access” colleges are becoming less selective, and may even have falling academic standards (Arum and Roksa 2011; Espenshade and Radford 2009; Hoxby 2009). Qualitative research is describing how students’ experience of university life varies widely between institutions of higher and lesser status, and may be diverging with time (Mullen 2010; Stuber 2011). Top-ranked schools continue to be breeding grounds for elites (Karabel 2005; Stevens 2007) while many students elsewhere suffer from high rates of attrition and low levels of engagement (Professor X 2010). Might we need a similar literature for Canadian universities?

One could argue that EMI is an unsurprising by-product of expansion. In a continually growing region like the GTA, higher education has expanded massively over the past half century, with the building of York University, UT Mississauga, and UT Scarborough and, in the past 10 to 20 years, the granting of university status to Ryerson, OCAD University, and UOIT. This higher education expansion has surely offered much social mobility for hundreds of thousands of students. But it may have also forged new kinds of social inequality.

If EMI is consequential for higher education students’ life chances and/or experiences, our findings then have a key implication for current policies that call for expanded access and differentiation (Clark et al. 2011). While greater access will surely boost opportunities for many, including thousands of “first generation” students, a greater variety of institutional mandates, missions, funding, and/or program offerings may also generate disparate university experiences and later outcomes. If this scenario does emerge, our thinking about higher education inequality may need to change in step. In the 1950s, massive numbers of youth began to attend high schools for several years, so researchers and policy makers became increasingly sensitive to sorting processes within that credential tier, and paid more attention to unequal rates of streaming, dropping out, and myriad other factors that shaped student experiences and outcomes. Today, as near majorities of youth cohorts enter postsecondary education, we need to be similarly sensitized to sorting processes within that tier. Being placed into fields and institutions of varying rank is a kind of “streaming” that can affect student experiences and outcomes. Future research is needed to further uncover this additional layer of stratification.

  1. 1

    For a recent examination of MMI at graduate levels in Canada, see Zarifa (2012b).

  2. 2

    An additional dimension of effectively maintained inequality (EMI) is hierarchies among fields of study. We will explore that dimension in future research. For Canadian studies on access to various fields, see Walters and Frank (2010) and Zarifa (2012a).

  3. 3

    SES patterns are produced by both academic and status cultural processes. For instance, Mullen's (2009 2010) comparison of students at Yale and a nearby nonelite university identified among the former a status culture in which families, peers, and high schools long nurtured aspirations and competitive academic strategies. Many sociologists, including Suzanne Bianchi, Sean Reardon, Annette Lareau, and Peggy McDonough, to name a few, have highlighted how upper SES groups in recent decades have intensified their educationally oriented parenting by gathering information about school options and learning strategies (see Davies and Guppy 2013).

  4. 4

    Student self-selection can occur through two mechanisms. Academically ambitious students may perceive that well-resourced institutions offer superior educations, whether due to higher quality of instruction, smaller classes, or superior classmates and their peer effects. Or, status-conscious students may seek highly ranked, reputable, and resourced institutions not for educational benefits per se, but for their social recognition and exclusivity.

  5. 5

    Disparities in endowments, for instance, are huge. In 2010, the University of Toronto endowment was $1,538,820,000, UBC's was $1,045,829,000, and McGill's was $941,112,000, compared to Huntington University ($2,416,000) and Athabasca ($2,358,000; see CAUT 2012:51). Canadian universities are more unequal in endowments than in operating incomes and expenditures; thus, endowments represent a component of rising institutional stratification.

  6. 6

    That study lacked a sampling frame that could generate representative samples from each university, however, and therefore this statement was considered suggestive rather than definitive (Anisef, 1982:5).

  7. 7

    The Maclean's Guide to Canadian Universities (Maclean's 2013) provides annual rankings of Canada's universities in three categories: “medical doctoral,” “comprehensive,” and “primarily undergraduate.” Ranks are based on multiple indicators across six dimensions: (1) student characteristics (awards and scholarships) and classroom experiences; (2) faculty caliber (teaching fellowships, awards, and grants); (3) financial resources and expenditures; (4) student aid and services; (5) library holdings; and (6) reputation among university officials, high school guidance counselors, and heads of other organizations.

  8. 8

    Baker (forthcoming) found that students in two elite Toronto private schools were most impressed by the prospect of attending top American universities, though not all thought seriously about applying to those schools. Beyond the top U.S. universities, those students deemed McGill, Queen's, and Western to be far more desirable than other Ontario universities, including those in Toronto.

  9. 9

    A suggestive anecdote comes from the first author's research leave at Harvard University in 2006. When lunching with Harvard's Canadian Undergraduate club, a show of hands from the 12 students gathered revealed all were graduates of Canadian private high schools.

  10. 10

    Seventeen is the age when students are typically in Grade 12 and most frequently apply to postsecondary institutions, though some in this cohort were in Grade 11 that year.

  11. 11

    The FIUC provides annual information on income and expenditures for all universities and degree-granting colleges. The TLAC surveys provide annual financial information (e.g., tuition fees, living accommodation costs) on all degree-granting universities and colleges. The USIS provides annual information on student background (e.g., gender, citizenship, age) and level and type of education.

  12. 12

    TDSB data show that 743 students in an earlier cohort applied to universities outside Ontario, most of whom (466) applied to McGill. Another 264 did not later confirm admission to Ontario universities, likely meaning that they attended an institution outside the province. Among those 264 students, over three-quarters came from schools in neighborhoods that were in the two highest deciles of income, 57 percent were in the highest decile, and two-thirds were female.

  13. 13

    Results for income and expenditures were nearly identical, and so we present a table only for the former.

  14. 14

    We explore this outcome since the effects of social background may be especially strong for attending highly ranked universities beyond commuting distance.

  15. 15

    The latter measure is an aggregate-level proxy for students’ actual family income that undoubtedly introduces measurement error into our models.

  16. 16

    Due to moderate negative skew evident in graphical displays of the original distributions, we performed square root transformations on the institutional income, expenditures, and endowment variables to satisfy the assumption of normality (Tabachnick and Fidell 2007).

  17. 17

    The intraclass correlation coefficient for the models ranged between 4 and 5 percent across all unrestricted models, suggesting that roughly 5 percent of the variance may be attributable to school traits.

  18. 18

    Given the large sample size and large number of level 2 units, linear mixed-effects models were fitted using full maximum-likelihood estimation. The mixed-effects logit models were estimated using a penalized quasi-likelihood procedure. Therefore, the reported deviance, Akaike information criterion (AIC), and Bayesian information criterion (BIC) for these models should be interpreted with caution.

  19. 19

    Pseudo R2 values for random intercept models are reported for each model to assess the proportional reduction of prediction error (Snijders and Bosker 2012).

  20. 20

    For all outcomes, a fourth model (not shown) including several school-level measures of social assistance, family structure, and language at school was estimated. None of these variables were statistically significant, altered the broad patterns of coefficients described above, or improve the overall model fit.

  21. 21

    Additional models were also estimated for the total expenditures per FTE for each university in 2006. Their results were nearly identical to those for income, and so were not reproduced here, but are available upon request.