The Review of Economics and Statistics

VOL. XCVII OCTOBER 2015 NUMBER 4

ARE TENURE TRACK PROFESSORS BETTER TEACHERS?

David N. Figlio, Morton O. Schapiro, and Kevin B. Soter*

Abstract—This study makes use of detailed student-level data from eight This trend has led some observers to lament the potential
cohorts of first-year students at Northwestern University to investigate the
relative effects of tenure track/tenured versus contingent faculty on stu- blow to academic freedom dealt by the decline of tenure
dent learning. We focus on classes taken during a student’s first term at and to focus on the often challenging employment condi-
Northwestern and employ an identification strategy in which we control tions under which many contingent faculty work (see, e.g.,
for both student-level fixed effects and next-class-taken fixed effects to
measure the degree to which contingent faculty contribute more or less to June, 2012, and Wilson, 2010). Further, McPherson and
lasting student learning than do other faculty. We find consistent evidence Schapiro (1999) point to efficiency gains from tenure; they
that students learn relatively more from contingent faculty in their first- outline its positive role in influencing the distribution of
term courses. This result is driven by the fact that the bottom quarter of
tenure track/tenured faculty (as indicated by our measure of teaching authority within colleges and universities.
effectiveness) has lower ‘‘value added’’ than their contingent counter- While those considerations certainly have relevance in
parts. Differences between contingent and tenure track/tenured faculty are evaluating the impact of the growing demise of the tenure
present across a wide variety of subject areas and are particularly pro-
nounced for Northwestern’s average and less-qualified students. system, there is an educational outcome that may be mea-
sured more directly: Do undergraduates taught by contin-
I. Introduction gent faculty members learn as much as those taught by
tenure track/tenured faculty?
There have been a number of attempts to answer this
T HE role of tenure in American higher education has
been reduced dramatically in recent decades. In 1975,
57% of all faculty (excluding graduate students) were in the
question. On a national level, Ehrenberg and Zhang (2005)
present evidence that hiring more part-time and contingent
tenure system; by 2011 that figure had been cut almost in faculty lowers institutional graduation rates. This result is
half, to 29%.1 Some observers predict that the share of bolstered by Bettinger and Long (2006), who find a simi-
tenure track/tenured faculty will bottom out at between larly negative effect on aggregate levels of persistence
15% to 20%, with tenure being largely limited to the flag- when they focus specifically on part-time adjuncts. These
ship public and private research universities and the wealth- types of results indicate that even if contingent faculty are
iest of the liberal arts colleges.2 more popular with students—perhaps because of classroom
There is evidence that this trend accelerated after January behaviors that maximize student evaluations but not student
1, 1994, when mandatory retirement for faculty was abol- learning—they nonetheless might not be successful in
ished by federal law. Ehrenberg (2012) reports that between improving students’ longer-term prospects.3 To date, how-
1995 and 2007, the share of part-time faculty rose at almost ever, there exists little evidence on the effects of faculty
all institutional types, while among full-time faculty, the tenure track status on genuine student learning.
movement away from the tenure system has quickened. The limited existing evidence on the relative perfor-
Especially notable is the rise of the full-time, contingent mance of tenure track/tenured professors versus faculty out-
faculty member at Ph.D.-granting universities. Their repre- side the tenure system makes it difficult for college and uni-
sentation within the entire group of full-time faculty went versity decision makers to determine the optimal staffing of
from 24% to 35% at public doctoral institutions and from their classrooms. This is particularly relevant for research
18% to 46% at private nonprofit doctoral institutions. universities, which face a multitasking problem of maxi-
mizing an objective function that includes both the produc-
Received for publication September 3, 2013. Revision accepted for pub- tion of cutting-edge research and the provision of outstand-
lication March 4, 2015. Editor: Asim Ijaz Khwaja. ing undergraduate teaching. While the paper closest to this
* Figlio: Northwestern University and NBER; Schapiro: Northwestern
University; Soter: Greatest Good. one in the literature, Bettinger and Long (2010), presents a
We are grateful to the Northwestern University Registrar’s Office, novel approach to measuring the effects of tenure line ver-
Office of Admissions, and Office of Human Resources for providing the sus other instruction, their analysis is largely centered on
data necessary to carry out this analysis, and to numerous colleagues,
seminar participants, and the editor and referees for many helpful sugges- institutions whose principal purpose is teaching. Further,
tions. Caitlin Ahearn, Rebecca Figlio, and Christine Mulhern provided their creative identification strategy uses short-term vacan-
exceptional research assistance. All opinions and errors are our own.
1
The American Association of University Professors website presents its
Contingent Faculty Index summarizing data from the IPEDS Fall Staff Survey.
2 3
There has been ongoing speculation about this topic in the educational Carrell and West (2010) show that instructors who have better student
press. Wilson (2010) is a good example. evaluations tend to produce lower levels of ‘‘deep learning.’’

The Review of Economics and Statistics, October 2015, 97(4): 715–724

Ó 2015 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology
doi:10.1162/REST_a_00529
716 THE REVIEW OF ECONOMICS AND STATISTICS

cies in departments, essentially analyzing the effects on Our identification strategy involves observing whether a
learning of transitory adjunct faculty. Their analysis there- student who takes, say, introductory economics with a con-
fore may not speak to the effects of part-time faculty with tingent faculty member and introductory political science
longer-term contracts and certainly does not address the with a tenure track/tenured professor in his or her first term
effects of full-time contingent faculty. In addition, while at Northwestern is (a) relatively more likely to take a sec-
they find some evidence that contingent faculty induce stu- ond political science class than another economics class
dent interest in a subject, as measured by the likelihood that and (b), conditional on taking more classes in both subjects,
students take additional courses in that subject, they are not more likely to perform better in the political science class
able to study how students perform in subsequent classes, than in the economics class.
an ideal way to see whether instructional quality has a last- The answers to these questions should shed light on one
ing impact. When one observes only student evaluations of of the most important outcomes relating to the dramatic
their instructors or the likelihood that students take more change in the professorate: its impact on student learning.
classes in the subject, it is difficult to judge whether one
type of instructor is genuinely better at education—that is,
whether they produce more ‘‘deep learning’’ in the words of II. Data and Methods
Carrell and West (2010)—or whether they are just more We make use of data on all Northwestern University
popular. Hoffmann and Oreopoulos (2009) evaluate teacher freshmen who entered between fall 2001 and fall 2008, a
quality in a Canadian research university setting, but, like total of 15,662 students.5 Our principal model for estimat-
Bettinger and Long (2010), they observe only the likelihood ing the relationship between the tenure track/tenured versus
that students take additional classes in the same subject contingent status of a student’s instructor and that student’s
rather than observe their academic performance in future level of learning in that subject is
classes.4 They find no evidence that contingent faculty
are either better or worse at inspiring students to take Gicstþ1 ¼ai þccstþ1 þbList þeist ;
more classes in their subjects. Carrell and West’s (2010)
analysis of professor quality examines follow-on classes and where, for student i taking a first-term class in subject s at
has outstanding internal validity as it relies on the random time t, L represents whether the class taken is taught by a
assignment of students to classes, but it is also based at an contingent faculty member and G represents that student’s
institution (the U.S. Air Force Academy) where teaching grade (on a four-point scale) the next time the student takes
rather than research is the dominant function. In addition, a class in subject s. The subscript c pertains to a specific
Carrell and West do not directly take on the question of instructor-class-term-year combination, so the inclusion of a
whether contingent faculty make better or worse instructors. fixed effect gcstþ1 means that we are comparing the relative
We bring to bear the first evidence within the research performance in subsequent classes in subjects A and B of a
university setting regarding the undergraduate learning student who took a class in subject A with a contingent
effects from different faculty types where we can observe faculty member and subject B with a tenure track/tenured
student performance in subsequent classes in the same sub- professor during his or her first term at Northwestern, hold-
ject. Specifically, we examine the initial classes taken by ing constant all of the specifics of the subsequent classes in
first-term freshmen in eight cohorts of undergraduates at subjects A and B. This means that we are obtaining our iden-
Northwestern University, a midsized research university tification from subjects where some first-term freshmen take
that is one of the twenty-six private universities among the classes from a tenure line professor and other first-term
sixty-two members of the Association of American Univer- freshmen take classes from a contingent faculty member.
sities and that consistently ranks among the most selective We also estimate linear probability models without the next-
undergraduate institutions in the United States. At North- class fixed effect but with student fixed effects where the
western, contingent faculty members tend to have stable, dependent variable is whether the student takes another class
long-term relationships with the university, and a substan- in subject s. We cluster standard errors at the instructor level
tial majority are full time. This allows us to study the to account for potential within-instructor error correlation.6
effects of contingent faculty at a major research university We obtained data from several offices at Northwestern
where these faculty members function as designated tea- University for the purposes of this analysis. The registrar’s
chers (both full time and part time) with long-term relation- office provided us with student transcript data, including
ships to the university. student grades, subjects, and instructor information; the
5
We limit our analysis to students who entered Northwestern in fall 2008
4
Hoffmann and Oreopoulos do study student performance in courses or before to give students sufficient time to complete their studies. Ninety-
with standardized tests shared across individual class sections in the same eight percent of students who ultimately earn an undergraduate degree at
term as their measure of instructor quality, but they do not perform a Northwestern do so within five years.
6
head-to-head comparison of lecturers versus tenure track/tenured faculty In a prior iteration of this paper, we clustered our standard errors at
members (their comparison controls for a measure of instructor quality) the individual student level (Figlio, Schapiro, & Soter, 2013). The stan-
and are not able to follow students into future classes to gauge ‘‘deep dard errors are larger in the case in which we cluster at the instructor
learning.’’ level, but the fundamental message is the same as before.
 ARE TENURE TRACK PROFESSORS BETTER TEACHERS? 717

TABLE 1.—DESCRIPTIVE STATISTICS, BY NUMBER OF CLASSES TAKEN WITH CONTINGENT FACULTY IN FALL QUARTER OF FRESHMAN YEAR
Highest Middle Lower Mean Took Another Mean Grade
Academic Academic Academic SAT Undeclared Class in Next Class
Count Preparation Preparation Preparation Score at Entry in Subject in Subject
Full sample 15,662 17% 57% 26% 1392 17% 74% 3.39
By number of contingent faculty classes
0 contingent 3,144 17% 56% 27% 1391 13% 72% 3.24
1 contingent 5,978 17% 58% 25% 1395 17% 72% 3.32
2 contingent 4,019 17% 56% 27% 1395 20% 74% 3.41
3 or more contingent 1,925 17% 57% 26% 1392 14% 82% 3.62
Only contingent 596 14% 51% 36% 1362 19% 77% 3.46
Data include all students who enrolled in their first quarter at Northwestern University during the fall terms between 2001 and 2008.

office of admissions provided information about the stu- tenure track/tenured versus contingent classes and initial prep-
dent’s initial intended major and academic qualifications; aration. All four groups have 17% with highest academic
and individual academic departments, as well as the office preparation, 25% to 27% with relatively low academic prep-
of human resources, confirmed the tenure track/tenured ver- aration, and average SAT scores between 1391 and 1395.
sus contingent status of all instructors.7 Table 1 presents Foreshadowing our results, the four groups differ sub-
some descriptive statistics of the population of Northwes- stantially in terms of their outcomes. The probability that a
tern students. We divide freshman students into groups student takes another class in the subject generally increases
based on academic preparation: 17% of entering freshman with the number of contingent faculty classes the student
have the highest preparation level, 57% have middle pre- takes in his or her first term at Northwestern, as does the
paration levels, and 26% have relatively low levels of pre- grade earned in the subsequent class. This latter pattern is
paration. The average SAT score (or converted ACT score) especially remarkable given that contingent faculty appear
for beginning freshmen was 1392, and 17% of entering to induce relatively marginal students, who might have
freshmen had not indicated an intended major at their time been expected to perform worse in subsequent classes, to
of entry to Northwestern. In 74% of cases, students took take those classes nonetheless. The bulk of this paper ex-
another class in a subject that they took during their first plores these relationships in a more systematic manner.
term of freshman year, and when they took the subsequent
class, they averaged a grade of 3.39 on a 4-point scale. III. Estimated Effects of Contingent Faculty on
We limit our analysis to first-term freshman students Subsequent Performance
because our identification assumption is that students select
their first classes with limited knowledge about instructor Table 2 presents our basic results. The unit of analysis is
quality or characteristics. We further condition on student the student-class pair for first-term freshmen at Northwes-
fixed effects because we are concerned that students who tern. To provide a basis for comparison, we report basic
take classes with one type of instructor versus another may OLS results in the first row of the table and then succes-
be relatively strong or weak students. The majority of stu- sively add layers of fixed effects. The left-most columns of
dents take at least one course with a contingent faculty the table are for all classes taken by all students, while the
member and at least one course with a tenure line professor second set of columns restricts the analysis to the 89.9% of
during their first term at Northwestern; 20.1% of students classes taken outside a student’s intended major.8
take classes only with tenure track/tenured professors, and As can be seen in the first row of table 2, the simple rela-
3.8% take classes only with contingent faculty members. tionships between contingent faculty status of the teacher of
Rows 2 through 6 of table 1 break down descriptive statis- a class and a student’s likelihood of taking another class
tics by the number of first-term classes taught by contingent and the grade obtained in that next class in a subject are
faculty. The small number of students who take only classes positive and strongly statistically significant. However,
taught by contingent faculty tend to be somewhat weaker because these relationships could reflect unmeasured stu-
than the other students; 36% come from the bottom ranks of dent characteristics, we compare subjects taken by the same
students (as opposed to 26% for the other 96.2 percent of stu- student and estimate student fixed-effects models. The esti-
dents), and their SAT scores average 1362 (as opposed to mated relationships remain reasonably large in magnitude:
1393 for the other students). But among the 96.2% of students a contingent faculty member increases the likelihood that a
who take at least one class from a tenure line faculty member, student will take another class in the subject by 7.3 percen-
there is no apparent relationship between the division of tage points (9.3 percentage points when limited to classes
outside the student’s intended major) and increases the
7
We exclude graduate students and visiting professors who hold faculty
8
appointments at other institutions from our analysis. Our results are fun- We show separate results for classes outside the student’s major to
damentally unchanged if we include these two groups, regardless of isolate the group of students for whom the choice to take additional
whether we assign them to the tenure track/tenured category or the contin- classes in the subject is most plausibly affected by the quality of the first
gent category of instructor. professor they encounter.
718 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 2.—ESTIMATED EFFECTS OF HAVING A CONTINGENT FACULTY MEMBER ON SUBSEQUENT COURSE TAKING AND PERFORMANCE IN THE SUBJECT:
LINEAR PROBABILITY MODELS
All Classes Classes Outside Intended Major (89.9% of Classes)
Probability of Grade in Next Probability of Grade in Next
Taking Next Class Class Taken Taking Next Class Class Taken
in Subject in Subject in Subject in Subject
OLS regression 0.077*** 0.185*** 0.085*** 0.218***
(0.030) (0.050) (0.032) (0.051)
Regressions with student fixed effects
All first-year fall classes 0.073*** 0.120*** 0.093*** 0.159***
(0.024) (0.037) (0.026) (0.041)
Students with no choice (35.2%) 0.108*** 0.095*** 0.139*** 0.148***
(0.034) (0.043) (0.038) (0.050)
Regressions with student fixed effects and next-class fixed effects
All first-year fall classes NA 0.060*** NA 0.079***
(0.008) (0.009)
Controlling for home country, experience in years NA 0.039*** NA 0.064***
(0.010) (0.010)
Controlling for home country, experience in six bins NA 0.042*** NA 0.066***
(0.010) (0.010)
Data include all students who enrolled in their first quarter at Northwestern University during the fall terms between 2001 and 2008. Each cell represents a different model specification. Standard errors adjusted for
clustering at the instructor level are reported in parentheses beneath coefficient estimates. Next-class information is recorded for the first time a student takes a second class in a given subject area. Intended majors
are recorded by the office of admissions. Coefficients are statistically distinct at ***1%, **5%, *10%. Data come from 15,662 students taking 56,599 first-quarter classes.

grade earned in that subsequent class by slightly more than that it is conceivable that a student might have a preference
one-tenth of a grade point (with a somewhat greater impact for either a tenure track/tenured or a contingent faculty
for classes outside the intended major).9 member, and conditional on deciding to take a given class,
We can further restrict our analysis to students with ‘‘no takes the class in the quarter in which the student’s pre-
choice’’—classes that are always taught either by tenure ferred faculty member’s status occurs. (This could happen
track/tenured faculty or by contingent faculty during the either across terms or within a term if the course is taught
entire time period considered. In this restriction, we expli- simultaneously by a tenure track/tenured faculty member
citly eliminate the possibility that a student is ‘‘shopping’’ and a contingent faculty member.) This specification helps
across instructors teaching a certain class. The idea here is to rule out, by construction, the possibility that our results
are driven by endogenous selection into sections of a given
course. In 35.2% of classes taken by first-term freshmen,
9
Almost all classes taught by contingent track faculty at Northwestern the student had no choice about the faculty status of an
are taught by those with a longer-term relationship with the university. instructor. The results are quite similar whether or not we
When we exclude the temporary lecturers and adjuncts, the estimates
barely change. Trimming the ‘‘one-off’’ lecturers and adjuncts, we find make the restriction to look only at students with no choice,
that for all students and all classes, a contingent faculty member is esti- as seen in the third row of table 2.10
mated to increase the likelihood that a student takes another class in the The available evidence presented thus far all suggests
subject by 7.5 percentage points and increases the grade by 0.12 grade
points. The results are similarly nearly identical for all other rows in the that there is no systematic sorting of students into classes
table. taught by tenure track/tenured versus contingent faculty,
Ten percent of students (19% of student-class pair observations) took but we can also investigate this question directly by esti-
multiple courses in the same subject in their first quarter, and in 2.3 per-
cent of cases, a student took classes taught by both tenure track/tenured mating sorting regressions in which we regress student attri-
and contingent faculty members in the same subject that term. If we butes against a contingent faculty dummy variable and
assign the next class to both tenure track/tenured and contingent faculty include fixed effects for the class-term-year combination.
members in the same subject, as we do in the specifications reported in
table 2, this has the effect of biasing our estimates toward 0. Indeed, if we We find no evidence of differential sorting by student char-
limit our analysis to students who took only one course in the subject dur- acteristics in the admissions data. In this model, contingent
ing their first quarter at Northwestern, we estimate that a contingent
faculty member increases the likelihood that a student takes another class
faculty attract students with a 0.0002 higher academic pre-
in the subject by 7.9 percentage points and increases the grade by 0.14
grade points. We continue to report the more conservative estimates in
10
the paper. We can also limit ourselves to the small number of cases—17% of
The results are also robust when we limit our analysis to each of the spe- students, 15% of student-class observations—in which a student indicates
cific colleges (there are six undergraduate colleges at Northwestern) no intended major preference at the time of entry to Northwestern. The
where the classes were offered. In student fixed-effects regressions, the results for this very restricted group are similar to those reported in the
relationship between contingent faculty status and the probability that a table. In models with student fixed effects, the coefficient on contingent
student will take another class in the subject is positive and statistically faculty is 0.120 (0.138 for students with no choice) when the dependent
significant in three of the four colleges (arts and sciences, music, and variable is the probability of taking another class in the subject and 0.089
engineering, but not communications) that teach almost all of the first- (0.088 for students with no choice) when the dependent variable is the
term freshman students, and the relationship between contingent faculty grade in the next class taken in the subject. All of these coefficient esti-
status and the grade earned in the next class in the subject is positive and mates are statistically significant at the 5% level or better. However, we
statistically significant in all four colleges. In section V of the paper, we do not have sufficient power to estimate our preferred specification—with
also break down our results by the grading standards of the subjects and both student fixed effects and next-class fixed effects—with the restricted
the qualifications of students who intend to major in those subjects. set of students who have no intended major at the time of entry.
 ARE TENURE TRACK PROFESSORS BETTER TEACHERS? 719

paration indicator (standard error of 0.0078) and with 0.886 FIGURE 1.—ESTIMATED EFFECTS OF HAVING A CONTINGENT FACULTY MEMBER ON
SUBSEQUENT PERFORMANCE IN THE SUBJECT, BY FRESHMAN YEAR COHORT
points lower SAT scores or SAT equivalent (standard error
of 1.259). Moreover, there is no evidence of differential sort-
ing by a broader set of student characteristics observed in
Northwestern’s administration of the freshman survey of the
Consortium on Financing Higher Education (COFHE): con-
tingent faculty draw students whose mothers are 0.5 percen-
tage points less likely to be college graduates (standard error
of 0.7), who are 0.01 percentage points more likely to have
named Northwestern as their first-choice school (standard
error of 1.04), who are 0.07 percentage points less likely to
have parents who are living together (standard error of
0.08), who are 0.1 percentage points more likely to be native These estimates are computed for the model with student fixed effects and next-course fixed effects.
English speakers (standard error of 0.5), who are 0.01 per- Data include all students who enrolled in their first quarter at Northwestern University during the fall
terms from 2001 to 2008. The solid line shows the estimated next-grade effect, with the dashed lines
centage points more likely to live within 10 miles of North- indicating a 90% confidence interval.
western (standard error of 0.35), and who are 1.0 percentage
points more likely to be female (standard error of 1.4). In is no evidence of a meaningful temporal pattern in these
summary, we could find no evidence that students differen- results, suggesting that any trends over time in the use or
tially self-select into taking courses by tenure track/tenured utility of contingent versus tenure line faculty members is
or contingent faculty along any observable dimension. not driving the findings that we report.12
We next move to our preferred model specification—one
with both student fixed effects and next-class fixed IV. Differences by Faculty Member Characteristics
effects—reported in the fourth row of table 2. In this model
we cannot study the relationship between contingent status Our results suggest that on average, first-term freshmen
and the likelihood of taking another class in the subject learn more from contingent faculty members than they do
because by default, all students have taken another class in from tenure track/tenured faculty. But are these differences
the subject. Moreover, we cannot limit ourselves to students constant across the entire spectrum, or is it the case that
with no choice because we must compare those who took most tenure track/tenured faculty members perform simi-
the introductory class in subject A with a contingent faculty larly to most contingent faculty members and the differ-
member to those who took that same class with a tenure line ences are due either to the best contingent faculty teachers
professor to have variation when we control for next-class substantially outperforming the best tenure track/tenured
fixed effects. When we estimate this highly parameterized teachers or to the worst tenure track/tenured teachers per-
model, we still find that having an initial experience in a forming considerably worse than the worst contingent
subject with a contingent faculty member increases a stu- faculty teachers? And if the difference is not constant across
dent’s performance in subsequent experiences with the sub- all faculty members, are our results driven by a handful of
ject. The point estimates are around half the size of those outliers or a by larger swath of the distribution?
found in the specification that includes only student fixed To explore this question, in figure 2 we compare the dis-
effects but are still statistically significant and sizable in tributions of value added of individual contingent faculty
magnitude, especially given that the typical student’s grade teachers and tenure track/tenured teachers, in which we plot
in the next class is a robust 3.39 out of 4. a variant of the cumulative density function where the per-
Because Northwestern relies somewhat more on contin- centile in each distribution is on the horizontal axis and the
gent faculty today than it did a decade ago11 and because corresponding value-added measure is on the vertical axis.
Northwestern freshman classes have become progressively (We choose to present the CDF in this manner because it
more qualified over time, we also investigate whether the makes clear exactly where in the distribution our results are
estimated impact of having a contingent faculty member is coming from.) An individual instructor’s value added is an
trending over time. (Of course, we have already ruled out instructor-specific fixed effect retained from our preferred
the primary effects of temporal trends by including both specification (in which we estimate instructors’ effect on
student fixed effects and next-class fixed effects.) As can be grade points earned in the next course, controlling for both
seen in figure 1, when we estimate our highly parameterized
model year by year, we still observe a positive relationship 12
One might also be concerned that changing grading standards over
between having a contingent faculty member and subse- time are potentially driving our results. However, grading standards have
remained quite flat at Northwestern during this time period. While the
quent grades in the subject in every year. In addition, there average next-course grade did rise modestly from 3.33 for fall 2001
entrants to 3.41 for fall 2008 entrants, student qualifications also rose dur-
ing this time period, with SATs increasing from 1376 for fall 2001
11
The typical freshman in fall 2001 took 38.9 percent of first-term entrants to 1421 for fall 2008 entrants, so that average qualifications-
courses from contingent faculty, as compared with 41.6 percent for the adjusted grades actually fell slightly over our time horizon. We describe
typical freshman in fall 2008. in section V our method for adjusting grades for student qualifications.
720 THE REVIEW OF ECONOMICS AND STATISTICS

FIGURE 2.—DISTRIBUTIONS OF ESTIMATED VALUE ADDED: CONTINGENT VERSUS TABLE 3.—ESTIMATED EFFECTS OF HAVING A CONTINGENT FACULTY
TENURE-TRACK FACULTY MEMBER ON SUBSEQUENT COURSE PERFORMANCE IN THE SUBJECT, BY
INSTRUCTOR EXPERIENCE: MODELS WITH STUDENT FIXED EFFECTS AND
NEXT-CLASS FIXED EFFECTS
Instructor All Classes Outside
Experience Classes Declared Major
5 years or fewer 0.042 0.048
(0.031) (0.033)
6–12 years 0.110*** 0.140***
(0.036) (0.040)
13 years or more 0.103*** 0.108***
(0.019) (0.018)
Data include all students who enrolled in their first quarter at Northwestern University during the fall
These estimates of instructor grade point effects are computed for the model with student fixed effects terms between 2001 and 2008. Each cell represents a different model specification. Standard errors
and next-course fixed effects. Data include all students who enrolled in their first quarter at Northwestern adjusted for clustering at the instructor level are reported in parentheses beneath coefficient estimates.
University during the fall terms from 2001 to 2008. This figure plots the distribution of instructor fixed Next-class information is recorded for the first time a student takes a second class in a given subject area.
effects, by faculty member type. Intended majors are recorded by the office of admissions. Coefficients are statistically distinct at ***1%,
**5%, *10%. Data come from 15,662 students taking 56,599 first-quarter classes.

student fixed effects and next-course fixed effects). As can

faculty are modestly more likely to have attended under-
be seen in the figure, the top three-quarters of the contingent
graduate institutions in English-speaking countries (86.3%
faculty and tenure track/tenured faculty distributions are
versus 79.2% for contingent faculty, p ¼ 0.123) and average
virtually perfectly overlapping, so the most outstanding
dramatically more experience (21.9 years versus 11.6 years
contingent faculty members and most outstanding tenure
for contingent faculty, p ¼ 0.000). Given these differences,
track/tenured faculty members perform essentially identi-
we consider two models in which we control for the country
cally, and the same is true at other points in the distribution
of undergraduate education (which we call ‘‘home country’’)
such as the median. But the bottom quarter of the tenure
and years of experience. In the fifth row, of table 2, we
track/tenured faculty have lower value added than the bot-
include years of experience as a linear control, and in the
tom quarter of the contingent faculty, and this difference is
sixth row, we categorize years of experience into six
substantial for the bottom 13% of the distribution (around
groups.15 While the magnitudes of coefficients are modestly
the weakest 150 instructors, by our definition). It is clear
lower (the point estimates are between 16% and 35% smal-
that our results are not being driven by a handful of outliers,
ler) in models in which we condition on home country and
but it is also clear that the difference in average outcomes
years of experience, we continue to find reasonably large
is due to the differences at the bottom of the value-added
and consistently robust evidence that on average, students
distribution.
learn more in first-term classes taught by contingent faculty
In some ways, this is exactly what we might have
members than they do when these classes are taught by
expected: contingent faculty members who are hired to
tenure track/tenured faculty members.
teach and who perform relatively poorly are less likely to
The (modest) differences that do exist between the results
be renewed than are those who perform well, while tenure
of models that do not control for experience and home coun-
track faculty who are relatively poor teachers may be pro-
try and those that do are driven by the control for experience
moted and retained for reasons other than their teaching
levels. Therefore, in table 3 we stratify faculty members into
ability. That said, the presence of these differences begs the
three groups based on measured experience: those with five
question of whether certain differences between contingent
or fewer years of experience, those with six to twelve years
faculty members and tenure track/tenured faculty members
of experience, and those with thirteen or more years of
can explain our findings. There are no administrative
experience.16 We observe that while students apparently
records that can address this question directly, so we col-
learn more on average from contingent faculty members
lected curriculum vitae available through extensive web
than from tenure track/tenured faculty in first-term courses,
searching in order to measure some differential attributes of
the results are concentrated in the two higher-experience
contingent and tenure track/tenured faculty.13 Two mea-
groups rather than in the low-experience group.17
sures that are the most directly observable are years of
Faculty members may perform differently depending on
experience (calculated based on time since Ph.D. and
their status at the university. We therefore divide tenure line
employment history) and native language (calculated based
on the country in which a faculty member earned his or her
bachelor’s degree or its equivalent).14 Tenure track/tenured
15
We also include missing data flags for the cases in which we are
missing either experience levels or home country. Our results are very
similar regardless of how we treat faculty members with missing data.
13 16
We found sufficiently complete curriculum vitae on the web for This is as close as we could come to constructing thirds of the contin-
87.1% of faculty members. gent faculty experience distribution.
14 17
We consider a faculty member to be a native English speaker if he or The reported results do not control for home country, but controlling
she attended an undergraduate institution in the United States, Canada, for home country barely changes the results. For example, the estimates
Great Britain, Ireland, Australia, New Zealand, or South Africa. All in the first column of table 3 would be 0.033 (standard error ¼ 0.037),
results are comparable if we look exclusively at those educated as under- 0.100 (standard error ¼ 0.037), and 0.104 (standard error ¼ 0.018) were
graduates in the United States or some subset of these countries. we to control for home country.
 ARE TENURE TRACK PROFESSORS BETTER TEACHERS? 721

TABLE 4.—ESTIMATED EFFECTS OF INSTRUCTOR TYPE ON SUBSEQUENT distinguished researchers. Northwestern has, since 1988,
COURSE PERFORMANCE IN THE SUBJECT, BY OTHER INSTRUCTOR ATTRIBUTES,
MODELS WITH STUDENT FIXED EFFECTS, AND NEXT-CLASS FIXED EFFECTS
annually recognized a set of scholars who have received
extraordinary honors for their scholarship.20 Around 40%
All Classes Outside
Faculty Typea Classes Intended Major of tenured faculty members at Northwestern have been
recognized at least once over the past 25 years as an extra-
Untenured professor 0.005 0.017
(29.7% of tenure-track) (0.010) (0.011) ordinary researcher. When we treat these faculty members
Part-time contingent faculty 0.052*** 0.060*** as a separate group, we find no difference in teaching out-
(23.1% of contingent) (0.012) (0.013) comes compared to tenured faculty who have not received
Full-time contingent faculty 0.062*** 0.081***
(76.9% of contingent) (0.009) (0.010) the recognition, and this does not significantly affect the
Data include all students who enrolled in their first quarter at Northwestern University during the fall other results in table 4.21
terms between 2001 and 2008. Each cell represents a different model specification. Standard errors
adjusted for clustering at the instructor level are reported in parentheses beneath coefficient estimates.
Next-class information is recorded for the first time a student takes a second class in a given subject area.
Intended majors are recorded by the office of admissions. Full-time contingent faculty members teach at
least four courses per year at Northwestern. Coefficients are statistically distinct at ***1%, **5%, *10%.
V. Differences by Subject and Student Qualifications
Data come from 15,662 students taking 56,599 first-quarter classes.
a
Comparison group is tenured professor. Are the results the same for all students and all subjects,
or are they present in some cases but not in others? In order
faculty into tenured professors versus untenured professors to investigate these questions, we next divide the course
(at the time of the course), and we divide contingent faculty subjects along two dimensions. First, we split the subjects
into full time (measured as teaching four or more courses into thirds based on the SAT scores of incoming students
per year at Northwestern) and part time (measured as teach- who intend to major in that discipline; we interpret this as a
ing three or fewer courses per year).18 We found that 29.7% measure of the perceived challenge of a subject by in-
of tenure track/tenured faculty in first-term courses are coming students. Second, we split the subjects into thirds
untenured, while 76.9% of contingent faculty teaching based on a measure of the grading standards of faculty
first-term courses are full time.19 Table 4 presents the esti- teaching that subject. We calculate grading standards by
mated effects of having a faculty member in one of these regressing grades against observed student qualifications;22
mutually exclusive groups—untenured professors, part-time we call the departments that award higher-than-predicted
contingent faculty, and full-time contingent faculty—with grades ‘‘higher-grading subjects.’’23 These two measures are
each group’s estimated effect compared to tenured faculty. negatively correlated: the correlation between the average
As can be seen in the table, tenure track professors with- SAT scores of intended majors in a department and the grades
out tenure have teaching outcomes that are on average equal that the department awards is 0.69. There is enough of a
to those of tenured professors, while both full-time and part- discordance between the two to make reporting both mea-
time contingent faculty members apparently outperform sures meaningful. For instance, though the highest-grading
tenured professors in the first-term classroom. The estimated subjects generally fall into the low-SAT subject group,
effect of full-time contingent faculty is modestly higher than 33.3% of the subjects with the highest grades are in the mid-
that for part-time contingent faculty, but the differences dle SAT group, and 4.1% are in the highest SAT group.
between these two groups are not statistically different from We report the results of these splits for our preferred
0. It is important to note that the overwhelming majority of model specification—with student fixed effects and next-
part-time contingent faculty at Northwestern still have long- class fixed effects—in table 5. As can be seen, the estimated
term relationships with the university, so we do not equate effects of contingent instructors for a first-term course
part-time faculty at Northwestern with ‘‘one-off’’ adjunct
20
instructors; rather, a large fraction of part-time contingent Reasons for being honored by Northwestern include recognition by
faculty members teach a course or two at Northwestern in the leading scholarly organization in their fields, receipt of prestigious fel-
lowships such as MacArthur or Guggenheim Fellowships, election to the
addition to their regular professional careers. This may be American Academy of Arts and Sciences, and comparable achievements.
the reason that we find a positive effect of part-time contin- We use this as a measure of research productivity because other potential
gent faculty at Northwestern, while previous research, such measures, such as the value of grants or number of citations, would likely
not have similar meaning across academic subjects.
as Bettinger and Long (2010), found less favorable effects: 21
Tenured faculty who were recognized between 1988 and 2013 for
part-time instructors at Northwestern almost always have research have an effect size of 0.001 (standard error ¼ 0.011) relative to
long-term relationships with the university, while adjuncts other tenured professors (estimate ¼ 0.006, standard error ¼ 0.011 for
classes outside of intended major). Treating only tenured faculty who
hired to fill temporary vacancies may not have the same have not received research recognition as the comparison group for table
sense of commitment to an institution and its students. 4 instead of all tenured faculty does not qualitatively affect the results
reported: point estimates for contingent faculty types remain positive and
We also sought to observe whether outstanding research- significant at the 1% level, and the magnitudes are within 7% of the
ers are better or worse introductory teachers than are less reported estimates, and untenured tenure-track faculty remain similar on
average to tenured faculty not recognized for research.
22
Betts and Grogger (2003) and Figlio and Lucas (2004) measure grad-
18
We have also estimated models in which we consider a three-course ing standards in similar ways by comparing grades awarded to some
teacher as a full-time contingent faculty member, and the results are very external benchmark of predicted grades.
23
similar. We are restricted by the registrar from identifying specific depart-
19
A larger fraction of arts and sciences contingent faculty are full time. ments in this paper.
722 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 5.—ESTIMATED EFFECTS OF HAVING A CONTINGENT FACULTY MEMBER ON TABLE 6.—PERCENTAGE OF STUDENTS EARNING A OR BETTER IN A CLASS,
SUBSEQUENT PERFORMANCE IN THE SUBJECT: BY SUBJECT AND ACADEMIC INDICATOR
DIFFERENTIAL EFFECTS BY MAJOR ‘‘CHALLENGE’’: STUDENT FIXED EFFECTS AND
NEXT-CLASS FIXED EFFECTS Relatively
Highest Middle Low
All Classes Outside Academic Academic Academic
Classes Declared Major Preparation Preparation Preparation
Subjects divided by SAT score of students (divided into thirds) All classes 67.2% 56.3% 46.3%
Highest SAT subjects 0.084*** 0.099*** Subjects divided by the average SAT scores of freshmen with
(0.013) (0.013) intended majors (divided into thirds)
Middle SAT subjects 0.063*** 0.091*** Highest SAT subjects 56.0% 38.1% 23.6%
(0.017) (0.016) Middle SAT subjects 76.2% 64.0% 47.6%
Lowest SAT subjects 0.030*** 0.043*** Lowest SAT subjects 82.7% 73.1% 61.5%
(0.014) (0.016) Subjects divided by typical grade in classes (divided into thirds)
Subjects divided by typical grade (divided into thirds) Lowest-grading subjects 54.5% 36.5% 21.8%
Lowest-grading subjects 0.077*** 0.097*** Middle-grading subjects 71.4% 57.5% 40.4%
(0.013) (0.013) Highest-grading subjects 87.4% 80.8% 71.5%
Middle-grading subjects 0.056*** 0.071***
(0.015) (0.015)
Highest-grading subjects 0.041*** 0.061***
(0.015) (0.018)
demic preparation.25 While there is certainly more room for
Data include all students who enrolled in their first quarter at Northwestern University during the fall
grade dispersion in the toughest-grading subjects, where
terms between 2001 and 2008. Each cell represents a different model specification. Standard errors
adjusted for clustering at the instructor level are reported in parentheses beneath coefficient estimates.
45% of students with the highest preparation earn a Bþ or
Next-class information is recorded for the first time a student takes a second class in a given subject area. lower and 78% of students with relatively low academic
Intended majors are recorded by the office of admissions. Coefficients are statistically distinct ***1%,
**5%, and *10%. Data come from 15,662 students taking 56,599 first-quarter classes. Grading levels of preparation do the same, the point is that in no subject and
subjects are determined by comparing average residuals of a regression of grades on observed student
characteristics; results are qualitatively unchanged if grading levels are unadjusted.
for no group of students is a grade of A or A a forgone
conclusion. Nonetheless, while we cannot say for certain
whether the stronger results for harder-grading subjects and
are positive for all sets of subjects, regardless of grading those attracting higher-rated students are due to the effects
standards or perceived challenge. That said, the estimated of faculty status truly being greater for those subjects or
effects are strongest for the subjects with tougher grading whether there is simply more room for grade dispersion in
standards (the relatively low-grading classes) and for those those subjects, the key finding is that we observe advantages
that attract the most qualified students. The estimated for contingent faculty across all subjects, regardless of our
effect of having a contingent faculty member on subse- measure of perceived challenge or grading standards.
quent grades is more than twice as large in the high-SAT In table 7 we split the population of students by student
subjects as in the low-SAT subjects, and is also substantial academic preparation and then split subjects by both SAT
when comparing the hardest-grading to the easiest-grading scores of incoming intended majors and by our measure of
subjects.24 faculty grading standards. The rationale for doing this is
This pattern of results could be due to the effects of being that students who are relatively less well prepared academi-
strongest for these groups of classes or because of ceiling cally are more representative of the college-going popula-
effects: perhaps in the easiest-grading subjects, most stu- tion in the United States as a whole and can therefore assist
dents earn top grades and there is little opportunity for dis- to some degree in assessing the potential external relevance
tinction. As can be seen in table 6, it is not the case that of our findings. While we find that the best-prepared stu-
everyone earns a top grade, even in the relatively easy- dents at Northwestern appear to perform about the same
grading subjects. This table presents the percentage of stu- regardless of whether their first class in the subject was
dents earning a grade of A or A in each group of subjects, taught by a contingent faculty member or a tenure track/
broken down by the student’s academic indicator (the tenured professor, the estimated positive effect of having a
admission office’s preenrollment prediction of a student’s contingent faculty member is present and strongly statisti-
academic success at the university). We see that in the cally significant for all other groups of students. Moreover,
easiest-grading third of subjects, 13% of students with the there appears to be an interaction between class type and
highest academic preparation earn grades of Bþ or lower student qualifications. While there is no apparent relation-
compared with 28% of students with relatively low aca- ship between instructor type and student outcomes for the
top-rated students, a clear pattern emerges for the other two
24
groups of students. For students with middle academic pre-
We have also divided the departments into those that are exception-
ally strong in research versus their competitors elsewhere and those that
paration, the estimated relationship between instructor type
are not. In 74.2% of cases, students take classes ranked in the top twenty and subsequent outcomes is about the same for the subjects
nationally in the most recent National Research Council rankings. If we attracting relatively low- and midlevel students but substan-
restrict our analyses to the 25.8% of classes in subjects ranked outside the
top twenty nationally, we continue to observe positive and significant esti-
25
mated effects of contingent faculty in our model with student and next- It is important to recall that even the relatively marginal students at
course fixed effects. The coefficient on contingent faculty is 0.049 (stan- Northwestern are still very highly qualified. The average SAT (or ACT
dard error of 0.016) for all students and 0.067 (standard error of 0.017) equivalent) among students with relatively low academic preparation is
for students taking courses outside their intended major. still a robust 1316.
 ARE TENURE TRACK PROFESSORS BETTER TEACHERS? 723

TABLE 7.—ESTIMATED EFFECTS OF HAVING A CONTINGENT FACULTY MEMBER ON gest gains to taking courses from contingent faculty were
SUBSEQUENT PERFORMANCE IN THE SUBJECT:
DIFFERENTIAL EFFECTS BY STUDENT ACADEMIC PREPARATION: STUDENT FIXED
for relatively weak students taking courses in the toughest-
EFFECTS AND NEXT-CLASS FIXED EFFECTS grading subjects.
Relatively
Highest Middle Low
Academic Academic Academic
VI. Conclusion
Preparation Preparation Preparation
Our findings suggest that contingent faculty at Northwes-
All classes 0.028 0.062*** 0.058*** tern not only induce first-term students to take more classes
(0.021) (0.011) (0.022)
Classes outside declared major 0.024 0.073*** 0.098*** in a given subject than do tenure line professors, but also
(0.023) (0.011) (0.026) lead the students to do better in subsequent course work
Subjects divided by the average SAT scores of freshmen with intended than do their tenure track/tenured colleagues. This result is
majors (divided into thirds)
Highest SAT subjects 0.013 0.099*** 0.168*** driven by the fact that those in the bottom quarter of tenure
(0.026) (0.018) (0.039) track/tenured faculty (as indicated by our measure of teach-
Middle SAT subjects 0.055 0.059*** 0.125*** ing effectiveness) have lower ‘‘value added’’ than their con-
(0.072) (0.022) (0.048)
Lowest SAT subjects 0.029 0.052*** 0.004 tingent counterparts.
(0.037) (0.019) (0.031) How generalizable are these results? Because a key part
Subjects divided by typical grade in classes (divided into thirds) of our identification strategy is to limit our analysis to first-
Lowest-grading subjects 0.007 0.094*** 0.175***
(0.026) (0.018) (0.038) term freshman undergraduates, the evidence that contingent
Middle-grading subjects 0.075 0.058*** 0.069* faculty produce better outcomes may not apply to more
(0.056) (0.020) (0.039) advanced courses. Further, Northwestern University is
Highest-grading subjects 0.013 0.059*** 0.027
0.050 0.023*** 0.040 among the most selective and highly ranked research uni-
Data include all students who enrolled in their first quarter at Northwestern University during the fall versities in the world, and its ability to attract first-class
terms between 2001 and 2008. Each cell represents a different model specification. Standard errors
adjusted for clustering at the instructor level are reported in parentheses beneath coefficient estimates.
contingent faculty may be different from that of most other
Next-class information is recorded for the first time a student takes a second class in a given subject area. institutions. Importantly, a substantial majority of contin-
Intended majors are recorded by the office of admissions. Coefficients are statistically distinct at ***1%,
**5%, *10%. Data come from 15,662 students taking 56,599 first-quarter classes. gent faculty at Northwestern are full-time faculty members
with long-term contracts and benefits, and therefore may
tially larger for the subjects attracting the most qualified stu- have a stronger commitment to the institution than some of
dents and about the same for easiest and middle-grading their contingent counterparts at other institutions. North-
subjects but considerably larger for the toughest-grading western’s tenure track/tenured faculty members may also
subjects. For students with relatively low academic prepara- have different classroom skills from those at other schools.
tion, the monotonic relationships are even more pronounced, Finally, Northwestern students come from a rarefied portion
with by far the strongest estimated results of all observed for of the preparation distribution and are far from reflective of
the relatively marginal students at Northwestern taking the the general student population in the United States. That
toughest-grading subjects and those attracting the most quali- said, our results are strongest for the students and subjects
fied students. Note from table 6 that the gap in the percentage that are most likely to generalize to a considerably wider
receiving an A or A between the toughest-grading and the range of institutions: the benefits of taking courses with
easiest-grading subjects is 49.7 percentage points for those contingent faculty appear to be stronger for the relatively
with relatively low academic preparation, but nearly as high marginal students at Northwestern (although they are still
(44.3 percentage points) for those with middle academic pre- very well-prepared students), and our results are similar for
paration and still quite high (32.9 percentage points) for those top-ranked departments as for lower-ranked departments.
with the highest academic preparation. Therefore, this pattern Our identification strategy and setting may help to
of estimated effects of instructor type broken down by stu- explain why our results find a more positive effect from
dent qualifications and subject type suggests that these find- contingent faculty members than the earlier literature does.
ings are likely due to genuinely differential effects of instruc- Because contingent faculty members at Northwestern tend
tor type across subject and student preparation rather than to have considerably different contracts than do contingent
just pure ceiling effects or the differential likelihood of earn- faculty members at many other institutions, our results may
ing higher grades in some subjects versus others. be better thought of as the effects of taking classes with
In sum, we found that the strong and significant effect of designated teachers, albeit a group of designated teachers
contingent faculty on our measure of deep learning held for who can be fired in the event of poor teaching, rather than
all subjects, regardless of grading standards or the qualifica- generalized results about contingent faculty. Our work also
tions of the students the subjects attracted. The apparent differs from the prior literature in that we look not only at
benefits of taking classes from contingent faculty were par- the likelihood that a student takes a subsequent class, but
ticularly strong for tougher-grading subjects and those that also at the likelihood of success in that class once the stu-
attracted the most qualified students, and the benefits were dent enrolls. It is therefore somewhat difficult to compare
enjoyed more by the less academically qualified students our results to previous findings because our treatment of
than by the more academically qualified students—the big- interest is considerably different, as is our outcome of inter-
724 THE REVIEW OF ECONOMICS AND STATISTICS

est: a measure of the effects of faculty of different types on Higher Education? The Shifting Financial Burden (Baltimore:
Johns Hopkins University Press, 2006).
students’ deep learning. In addition, our identification strat- ——— ‘‘Does Cheaper Mean Better? The Impact of Using Adjunct
egy differs from the closest paper in the literature to our Instructors on Student Outcomes,’’ this REVIEW 92, no. 3 (2010),
own, Bettinger and Long (2010), in ways that might par- 598–613.
Betts, Julian R., and Jeff Grogger, ‘‘The Impact of Grading Standards on
tially explain the differences. Because of their identification Student Achievement, Educational Attainment, and Entry-Level
strategy, Bettinger and Long are better able to identify the Earnings,’’ Economics of Education Review 22 (2003), 343–352.
effects of short-term transient adjuncts, while at Northwes- Carrell, Scott E., and James E. West, ‘‘Does Professor Quality Matter?
Evidence from Random Assignment of Students to Professors,’’
tern University, most of the contingent faculty are full time Journal of Political Economy 118 (2010), 409–432.
and even the part-time contingent faculty members gener- Ehrenberg, Ronald G., ‘‘American Higher Education in Transition,’’ Jour-
ally have long-term relationships with the university. nal of Economic Perspectives 26 (2012), 193–216.
Ehrenberg, Ronald G., and Liang Zhang, ‘‘Do Tenured and Tenure Track
There are many aspects relating to changes in the tenure Faculty Matter?’’ Journal of Human Resources 40 (2005), 647–
status of faculty, from the impact on research productivity 659.
to the protection of academic freedom. But certainly learn- Figlio, David N., and Maurice E. Lucas, ‘‘Do High Grading Standards
Affect Student Performance?’’ Journal of Public Economics 88
ing outcomes are an important consideration in evaluating (2004), 1815–1834.
whether the observed trend away from tenure track/tenured Figlio, David N., Morton O. Schapiro, and Kevin B. Soter, ‘‘Are Tenure
toward contingent faculty is good or bad. Our results pro- Track Professors Better Teachers?’’ NBER working paper 19406
(September 2013).
vide evidence that the rise of full-time designated teachers Hoffmann, Florian, and Philip Oreopoulos, ‘‘Professor Qualities and Stu-
at U.S. colleges and universities may be less of a cause for dent Achievement,’’ this REVIEW 91, no. 1 (2009), 83–92.
alarm than many assume. June, Audrey Williams, ‘‘Adjuncts Build Strength in Numbers: The New
Majority Generates a Shift in Academic Culture,’’ Chronicle of
Higher Education, November 5, 2012.
McPherson, Michael S., and Morton Owen Schapiro, ‘‘Tenure Issues in
REFERENCES Higher Education,’’ Journal of Economic Perspectives 13 (1999),
85–98.
Bettinger, Eric P., and Bridget T. Long, ‘‘The Increasing Use of Adjunct Wilson, Robin, ‘‘Tenure, RIP: What the Vanishing Status Means for the
Instructors at Public Institutions: Are We Hurting Students?’’ (pp. Future of Education,’’ Chronicle of Higher Education, July 4,
51–69), in Ronald G. Ehrenberg, ed., What’s Happening to Public 2010.