Education work

American Economic Association

Do Students Go to Class? Should They? Author(s): David Romer Source: The Journal of Economic Perspectives, Vol. 7, No. 3 (Summer, 1993), pp. 167-174 Published by: American Economic Association Stable URL: http://www.jstor.org/stable/2138449 Accessed: 28-11-2016 17:26 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted

digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about

JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at

http://about.jstor.org/terms

American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The Journal of Economic Perspectives

This content downloaded from 155.247.238.90 on Mon, 28 Nov 2016 17:26:17 UTC All use subject to http://about.jstor.org/terms

Journal of Economic Perspectives- Volume 7, Number 3-Summer 1993-Pages 167-174

Do Students Go to Class? Should They?

David Romer

L ectures and other class meetings are a primary means of instruction in almost all undergraduate courses. Yet almost everyone who has taught an undergraduate course has probably noticed that attendance at these

meetings is far from perfect. There is surprisingly little systematic evidence, however, about attendance and its effects. There are three natural questions.

What is the extent of absenteeism? How much, if at all, does absenteeism affect learning? Should anything be done about absenteeism?

This article presents quantitative evidence on the first two of these ques- tions, and speculative comments on the third. First, attendance counts in economics courses at three relatively elite universities indicate that absenteeism is rampant: usually about one-third of students are not at class. Second, regression estimates of the relation between attendance and performance in one large lecture course suggest that attendance may substantially affect learn- ing: considering only students who do all of the problem sets and controlling

for prior grade point average, the difference in performance between a student who attends regularly and one who attends sporadically is about a full letter

grade. In light of these results, steps to increase attendance, including making attendance mandatory, may deserve serious consideration.'

I Ihave been unable to find any previous investigations of the extent of absenteeism. There have been a few other studies of the relation between attendance and performance (for example, Schmidt, 1983; Park and Kerr, 1990). These studies generally confirm the findings here that attendance and performance are related even when a variety of student characteristics are controlled for. The present study differs from this earlier work in focusing on the quantitative magnitude of the relationship and on the issue of the extent to which the relationship reflects a genuine effect of attendance.

* David Romer is Professor of Economics, University of California, Berkeley, California.

This content downloaded from 155.247.238.90 on Mon, 28 Nov 2016 17:26:17 UTC All use subject to http://about.jstor.org/terms

168 Journal of Economic Perspectives

Do Students Attend Class?

Counts were made of the number of students attending one meeting of

every undergraduate economics class during a "typical" week of the spring

1992 semester at three schools. School A is a medium-sized (6000 undergradu-

ates) private university; School B is a large (20,000 undergraduates) public

university; and School C is a small (2500 undergraduates) liberal arts college.

The schools are intended to be representative of the upper echelons of Ameri-

can colleges and universities. All three are classified by Barron's Profiles of

American Colleges (1991 edition) as "highly competitive," the second highest of

their six categories.2 The attendance counts were made a few weeks before the end of the

semester at each school. This choice avoided both times when attendance is

generally thought to be unusually low (such as just after exams and immedii-

ately before and after vacations) and times when it is generally thought to be

unusually high (such as just before exams). Individuals at all three schools

independently suggested that attendance a few weeks before the end of the

semester was likely to be representative of average attendance. Attendance was

taken at one meeting of each class during the sample week. Current enrollment

figures were obtained from departmental offices.

Table 1 reports the results. The first row shows the overall absenteeism

rates at each school. At School A, 34 percent of students were absent; at School

B, 40 percent; and at School C, 25 percent.3 In short, on a typical day at a

typical elite American university, roughly one-third of the students in eco-

nomics courses are not attending class.

The remaining rows of the table break down the overall figures along

various dimensions. Course size appears to have an important effect on absen-

teeism. At all three schools, absenteeism is considerably lower in the smallest

third of classes than in the largest third. In addition, average class size is lowest

at School C and highest at School B, which is consistent with the fact that

absenteeism is lowest at C and highest at B. Absenteeism is also lower in courses

with a significant mathematical component (such as econometrics, honors

sections of intermediate theory, and field courses in theory). This pattern holds

2A total of 127 schools, enrolling about 675,000 undergraduates, are classified by Barron's as "most competitive" or "highly competitive." Of these, 24 schools, with 140,000 undergraduates, are private universities with between 4000 and 10,000 undergraduates; 14, with 270,000 undergraduates, are public universities with over 10,000 undergraduates; and 56, with

100,000 undergraduates, are colleges or universities with minimal graduate programs and fewer than 3500 undergraduates. The remaining schools are small and medium-sized public colleges and universities (15 schools, with 70,000 undergraduates), large private universities (5 schools with 70,000 undergraduates), and small private universities (13 schools with 30,000 undergraduates). 3Attendance counts were inadvertently not made in a handful of classes at School C. These classes do not appear to differ in any systematic way from the classes at which attendance was taken.

This content downloaded from 155.247.238.90 on Mon, 28 Nov 2016 17:26:17 UTC All use subject to http://about.jstor.org/terms

David Romer 169

Table 1

Absenteeism Rates in Economics Classes

School A School B School C

All Economics Courses 34.0% 39.7% 24.8%

By Size of Course:

Small (bottom 33%) 27.0 37.7 21.5 Large(top 33%) 38.8 42.9 30.4

By Mathematical Content:

Mathematical 10.0 17.6 16.7 Non-Mathematical 34.3 41.5 25.5

By Type of Course:

Principles & 37.3 40.5 29.7 Intermediate Theory

Upper Level, Only 26.4 41.1 17.7 Principles Required

Upper Level, 33.3 35.2 20.7 Additional Requirements

at all three schools.4 Similarly, at all three schools absenteeism is somewhat

higher in core courses than in field courses.

Finally, it is generally perceived, not surprisingly, that students attend class

more often when the quality of instruction is higher. At School B, for example,

absenteeism is 34 percent for courses taught by regular faculty and 47 percent for courses with other instructors. To investigate this issue more systematically,

course evaluation data for all undergraduate economics courses for one term

were obtained from a fourth school, School D. This school, like School B, is a

large public university. The two variables of interest are students' average

rating of the overall effectiveness of the instructor and the fraction of the

students enrolled in the course who returned the course evaluation form

(which is a reasonably good measure of attendance at one of the last class meetings of the term). The point estimates from a simple regression of the fraction of students attending the class on the average rating imply that raising

the average rating from the 25th percentile to the 75th lowers absenteeism by

10 percentage points; the t-statistic on the rating variable is 3.4. Thus the

quality of instruction (or at least students' perception of that quality) appears to have an important impact on attendance.

4The figures for mathematical courses at School A are based on only one course. Thus this figure should be given little weight.

This content downloaded from 155.247.238.90 on Mon, 28 Nov 2016 17:26:17 UTC All use subject to http://about.jstor.org/terms

170 Journal of Economic Perspectives

Other features of the data from School D generally confirm the findings

for the other schools.5 Absenteeism is high (45 percent across all courses), and

lower in small courses than in large (31 percent in the smallest third of courses

and 54 percent in the largest third). Again, absenteeism is lower in courses with

a mathematical emphasis (39 percent, versus 47 percent for other courses), and

higher in core courses (52 percent, versus 31 percent in field courses that only require principles and 37 percent in advanced field courses).

A straightforward regression confirms these patterns of differences in

absenteeism across different types of courses. Specifically, using the data from

all four schools, I ran a regression (across courses) with the fraction of students

absent as the dependent variable, and a constant, the log of enrollment, and

dummies for mathematical content, for the two types of upper level courses,

and for three of the four schools as independent variables. The resulting

estimates imply that a doubling of enrollment is associated with a rise in

absenteeism of 4 percentage points; that mathematical content is associated

with a fall in absenteeism of 3 percentage points; and that moving from a core

course to either type of field course is associated with a fall in absenteeism of 5

to 7 percentage points. The coefficient on the enrollment variable is highly

statistically significant; those on the field course dummies are marginally so;

and those on the dummy for mathematical content and the three school

dummies are insignificant.

Should Students Attend Class?

These findings raise the question of whether absenteeism has a substantial

effect on learning. It is possible that students do not attend class because they

would learn relatively little if they did-because the instruction is of low quality, or because they have already mastered the material, or because they can learn

the material better by spending the same time studying in other ways. Alterna-

tively, it is possible that learning is severely adversely affected by absences, but

that many students are absent anyway-because they have genuinely better

uses of their time, or because they mistakenly believe that attendance is not

important to learning, or because they attach relatively little importance to

learning.

Because student attendance is not exogenous-students choose whether to

attend class-it is not possible to isolate definitively the impact of attendance on

5The data from School D are not strictly comparable with those from the other schools, because they reflect class meetings at the end of the term and because a few students are present but do not

return the evaluation form. It seems unlikely that these differences have any substantial effect on

the results.

This content downloaded from 155.247.238.90 on Mon, 28 Nov 2016 17:26:17 UTC All use subject to http://about.jstor.org/terms

Do Students Go To Class? Should They? 171

learning. But this section presents some suggestive evidence. In the fall 1990

semester, I took attendance at six meetings of my large intermediate macroeco-

nomics course. The resulting data can be used to investigate the relation

between attendance and performance.

As in other courses, overall absenteeism was high (25 percent). Twelve

percent of the students missed four or more of the meetings where attendance

was taken; 28 percent missed two or three; and 59 percent missed none or one.

Thus, absenteeism appears to be a mixture of some students missing most

classes and many students missing a smaller number of classes.

Student performance is measured as the overall score on the three exams

in the course. For ease of interpretation, the scores are converted to the usual

4-point grading scale: 3.84 and above represents an A; 3.50 to 3.83 an A-;

and so on down to 1.50 to 1.83 for a C- . Because no D+ 's or D- 's were

assigned, 1.17-1.49 represents a D and 1.16 and below an F. The first column of Table 2 reports the results of a simple regression of

performance on the fraction of lectures attended.6 The regression reveals a statistically significant and quantitatively large relation between attendance and

performance. The t-statistic on attendance is 6.2; the point estimates imply that a student who attends only a quarter of the lectures on average earns a 1.79

(C - ), while a student who attends all of the lectures on average earns a 3.44 (B + ). Attendance alone accounts for 31 percent of the variance in

performance.

Students who are more interested in the material, or more skilled academi-

cally, or more focused on academics are almost certain to attend class more often than students who are less interested, less skilled, or less focused (other

factors held constant). If this is the case, then the results in Column 1 of Table 2 to some extent reflect a general impact of motivation on performance rather

than a true effect of attendance.

I attempt to addi-ess this problem in three ways. First, I restrict the sample

to the 60 percent of the students who did all nine problem sets. It seems likely

that most of the students who were not devoting serious effort to the course did

not complete all of the problem sets. In addition, the lowest problem set score

was dropped in computing the course grade; thus the students who completed

all nine may have been especially motivated. On both grounds, this restricted

6There is one econometric complication worth mentioning: because attendance was not taken at every class meeting, some of the variation across students in measured attendance is due to measurement error rather than to true differences in attendance over the whole semester. If the class meetings at which attendance was taken were a random sample of all the meetings-which appears to be a good approximation-it is straightforward to estimate the size of the measurement error. This procedure implies that 28 percent of the variation in measured attendance represents measurement error. This estimate can be used to correct the regression coefficients, standard errors, and R2's for the bias that would otherwise be introduced by the measurement error. All of the results reported in Table 2 have been corrected in this way.

This content downloaded from 155.247.238.90 on Mon, 28 Nov 2016 17:26:17 UTC All use subject to http://about.jstor.org/terms

172 Journal of Economic Perspectives

Table 2

The Relationship Between Attendance and Performance

(1) (2) (3) (4) (5)

Sample: Full Restricted Full Full Restricted

Constant 1.25 1.02 1.07 - 0.67 -0.78 (0.27) (0.58) (0.23) (0.32) (0.43)

Fraction of 2.19 2.47 1.74 1.52 1.38

Lectures (0.35) (0.70) (0.46) (0.32) (0.58) Attended

Fraction of 0.60

Problem Sets (0.32)

Completed

Prior GPA 0.78 0.86

(0.12) (0.14)

Sample Size 195 116 195 195 116

R 2 0.31 0.26 0.33 0.47 0.48

Standard errors are in parentheses. The restricted sample consists of the students who completed

all of the problem sets.

sample may be more homogeneous in terms of general motivation than the full class. But as the second column of Table 2 shows, the relation between

attendance and performance in this sample is actually slightly stronger than for the class as a whole.

Second, doing the problem sets is arguably as good a proxy as attending the lectures for motivation. But Column 3 shows that there is a much stronger

relation between attendance and performance than between doing the problem

sets and performance: when both variables are entered in the regression, the coefficient on the fraction of lectures attended is almost three times as large as

the coefficient on the fraction of problem sets completed. Thus, either atten-

dance is a much better proxy than completing the problem sets for motivation, or attendance has a large additional impact on performance.

Third, data were obtained on students' grade point averages as of the

beginning of the semester. Including GPA as a control variable in the regres- sion serves to control for some of the differences across students in general

ability and motivation. In fact, because students' academic performance in previous classes depends in part on their attendance in those classes, the

coefficient on prior GPA will capture some of the effect of attendance on performance; as a result, including GPA as a control variable could cause the coefficient on attendance to understate the true impact of attendance on

performance.

This content downloaded from 155.247.238.90 on Mon, 28 Nov 2016 17:26:17 UTC All use subject to http://about.jstor.org/terms

David Romer 173

Column 4 of Table 2 shows the effects of including grade point average in

the regression. Prior GPA has an extremely strong relation with performance.

But the inclusion of GPA has little impact on the relation between attendance

and performance. The coefficient on attendance is two-thirds as large as it is in

the basic regression in Column 1, and it remains highly significant. The point

estimates imply that a student with the mean prior GPA earns on average a

2.13 (C) if he or she attends a quarter of the lectures but a 3.27 (B + ) if his or

her attendance is perfect.

Finally, Column 5 shows the results of both restricting the sample to the

students who did all nine problem sets and controlling for prior GPA. Even in

this case, the relation between attendance and performance remains large and

significant. The estimates imply that a student with the mean prior GPA earns

on average a C + if he or she attends only a quarter of the classes, compared to

a B + if attendance is perfect.

None of these ways of attempting to address the problem that attendance is

not exogenous is definitive. Nonetheless, they all give similar results: simple

ways of controlling for motivation and other omitted factors have only a

moderate impact on the relationship between attendance and performance.

Thus, although the possibility that the relationship reflects the impact of

omitted factors rather than a true effect cannot be ruled out, it seems likely that

an important part of the relationship reflects a genuine effect of attendance.

Should Attendance Be Mandatory?

Absenteeism is rampant in undergraduate economics courses at major

American universities. In addition, there is a very strong statistical relationship

between absenteeism and performance, and the evidence is consistent with the

view that this relationship has an important causal component.

These results raise the question of whether measures should be taken to

combat absenteeism. At the very least, exhortations to attend class seem called

for, and those exhortations can be backed up with data. But stronger measures

might be preferable. A generation ago, both in principle and in practice,

attendance at class was not optional. Today, often in principle and almost

always in practice, it is. Perhaps a return to the old system would make a large

difference to learning. There is no way to find out but to try. I believe that the

results here both about the extent of absenteeism and its relation to perfor-

mance are suggestive enough to warrant experimenting with making class

attendance mandatory in some undergraduate lecture courses.

One could also use mandatory attendance to perform a genuine controlled

experiment that could isolate the true impact of attendance on mastery of the

material. Specifically, one could randomly divide the students in a course into

two groups, an experimental group whose grading was based in part on

This content downloaded from 155.247.238.90 on Mon, 28 Nov 2016 17:26:17 UTC All use subject to http://about.jstor.org/terms

174 Journal of Economic Perspectives

attendance and a control group whose grading was not. By comparing the

attendance and the performance of the two groups, one could learn both the

impact of mandatory attendance on absenteeism and the impact of attendance

on performance.7 Unless either the impact of mandatory attendance on absen-

teeism or the size of the class were very large, the results of carrying out this

experiment for a single class would not allow one to estimate the impact of

attendance on performance with much precision. But the pooled results from

several such experiments could.

* I am grateful to Caroline Fohlin, Matthew Jones, and Costas Tsatsaronis for excellent

research assistance, and to Robert Cox, Roger Farmer, Steven Fazzari, Alan Krueger,

Christina Romer, Paul Ruud, Joseph Stiglitz, Timothy Taylor, and Robert Turner for

useful comments.

7Such an experiment would presumably require appropriate approval. Students could be given the right to opt out of the experiment by being allowed to choose (before the class is divided into the experimental and control groups) to have their grade based on a formula that gave attendance half the weight used in the grading formula for the experimental group. Fairness could be ensured by assigning grades to all students using all three procedures (experimental, control, and opting out), and making the mean grade for the full class the same under all three procedures, before the allocation of the students to the three groups was known to the person assigning grades.

References

Park, Kang H., and Peter M. Kerr, "De- terminants of Academic Performance: A Multinomial Logit Approach," Journal of Eco- nomic Education, Spring 1990, 21, 101-1 1.

Schmidt, Robert M., "Who Maximizes What? A Study in Student Time Allocation," American Economic Review, May 1983, 73:2, 23-28.

This content downloaded from 155.247.238.90 on Mon, 28 Nov 2016 17:26:17 UTC All use subject to http://about.jstor.org/terms