Does Attendance Matter? Evidence from an Ontario ITALeconomics.ca/2005/papers/0483.pdf · Does Attendance Matter? Evidence from an Ontario ITAL Pierre-Pascal Gendron and Paul Pieper

Does Attendance Matter? Evidence from an Ontario ITAL

Pierre-Pascal Gendron and Paul Pieper

The Business School Humber Institute of Technology & Advanced Learning

Toronto, Canada

Draft for Discussion

May 15, 2005

Abstract

Academic administrators and faculty often take for granted the positive impact of classroom attendance on student achievement. The evidence is usually anecdotal but nevertheless well accepted. As a result, there have been few studies focussing on the separate effect of attendance on achievement. This study examines the separate impact of classroom attendance on student achievement using a repeated cross section of student- and classroom-specific data. Attendance is measured very precisely and the data come from an introductory microeconomics course we taught at the Humber Institute of Technology & Advanced Learning in Toronto, Ontario, Canada. Preliminary OLS results using our baseline model shows that attendance has a strong positive impact on final course grade. Our fixed-effects model (with class-specific effects) shows again a strong impact of attendance on final grade. In both models, the relationship is non-linear in a way that suggests diminishing returns to attendance. Unlike some of the previous literature, however, we do not find in our sample a threshold beyond which attendance would negatively affect achievement. Key words: student attendance, achievement, performance

JEL Classification: I21

Acknowledgements: We wish to thank several members of Humber’s Office of Research: Dr. Peter Dietsche for comments, and Gillian Brenning and Chali Chen for expert research assistance. We gratefully acknowledge financial assistance through an Institutional Research Grant from the Office of Research.

i

Contents

1. Introduction..................................................................................................................... 1 2. The Literature.................................................................................................................. 3

2.1. Non-Behavioural Determinants of Student Performance ................................... 4 2.2. Student Absences and Academic Performance................................................... 6 2.3 Synthesis ............................................................................................................. 9

3. The Institutional Environment ...................................................................................... 10 3.1. Institution and Student Body ............................................................................ 10 3.2. The Business School ......................................................................................... 11

4. The Data........................................................................................................................ 12 5. Empirical Results .......................................................................................................... 16

5.1. Baseline Model ................................................................................................. 17 5.2. Fixed-Effects Model ......................................................................................... 18 5.3. Sensitivity Analysis .......................................................................................... 19

6. Conclusions................................................................................................................... 20 Appendix A: Scatter Plot of Grade on Attendance Ratio ................................................. 21 Appendix B: Tabular Summary of Literature Review...................................................... 22 References......................................................................................................................... 35

1

1. Introduction

Academic institution administrators and faculty intuitively believe that student

attendance in the classroom under either the lecture, tutorial, or discussion group delivery

systems matters for student achievement. In most institutions casual observation or

common sense supports this belief. Unfortunately, the evidence remains anecdotal unless

the impact of attendance on achievement can be quantified.

The examination of the association between attendance and achievement brings

forth additional benefits besides understanding the relationship between the two in

isolation. Other student and classroom characteristics may also be examined in the

process, especially in the context of their relationship with student achievement.

Understanding the impact of attendance has implications for faculty and

administrators. Instructors want learning to take place. Since the classroom environment

evolves slowly and the lecture delivery method remains dominant, instructors may want

to maximize attendance within this system using the various methods at their disposal,

such as class discussions, class exercises, rewards, and so on. Administrators also have a

stake in this: in order to minimize student attrition, they may find it desirable to clarify or

strengthen existing academic rules in favour of attendance. The problem is generally

more confined in programs where attendance is compulsory.

The paper reviews studies of attendance and various other determinants of student

performance from a variety of academic disciplines. It then presents the results of an

empirical investigation of the relationship between final course grades and attendance as

well as other student and classroom characteristics. We conducted the investigation at

the Humber Institute of Technology & Advanced Learning (hereafter “Humber”), a large

community college in Toronto, Ontario, Canada. Moreover, we relied exclusively on

primary student- and classroom-based data.

2

Section 2 reviews and summarizes the relevant literature, and outlines some of the

directions for research that emerge from the literature. Section 3 describes Humber’s

institutional environment and student body, with an emphasis on The Business School,

within which the study was conducted. Section 4 describes the data and collection

methods, and provides results from univariate and bivariate statistical analyses. Section 5

describes the models used, the empirical analyses carried out, and their results. We

conclude in Section 6.

3

2. The Literature

The literature is consistent overall in showing a link between classroom

attendance and academic performance (usually measured as a continuous percentage

grade). In order to better understand the context of student performance, the first section

of the literature review examines possible non-behavioural determinants (e.g. external

learning environment effects) and student socio-demographic characteristics and their

relation to academic performance. Next, we present a review of the literature as it

pertains to classroom attendance and the impact of non-attendance on grades as an

outcome of academic performance. We mention other determinants of academic

performance when relevant.

Not all studies have been able to find a statistically significant relationship

between attendance and grades. In nursing, where classroom attendance was very much

part of the program culture, the relatively low number of absences (an average of 1.4

classes missed by students with an 80 percent plus grade, and an average of 4.4 classes

missed by students in the 50 to 60 percent) even for the weaker students might account

for the mixed results by Brown et al. (1999) when correlating attendance with absences.

Of the nine class sections studied, only four were found to display any significant

negative correlation. Although the research population was drawn from the same college

as the present study, the students under review in Brown et al. (1999) represent a unique

population with a significantly different program-induced attitude toward student

responsibility with respect to absenteeism.

Although Thompson and Plummer (1979) noted a significant difference between

successful and unsuccessful remedial English students, they were unable to show that

attendance had any impact on the letter grades achieved by the successful students. Since

Berenson et al. (1992) dealt with remedial students in mathematics, additional underlying

factors other than attendance may explain their failure to find a positive correlation

between attendance and grades. Hyde and Flournoy (1986) surprisingly concluded that

mandatory classroom attendance policies could impede learning. However, the majority

4

of the literature reviewed would seem to support the intuitive and reasonable notion that

attendance does improve academic performance.

2.1. Non-Behavioural Determinants of Student Performance

In order to deal with the context of student absenteeism many studies focused on

researching the potential link between the characteristics of the classroom, or schools,

and student performance, whereby class attendance was only one of those characteristics.

Besides measuring individual student characteristics such as gender, race, parental

income, and other individual student characteristics, studies such as Betts and Morell

(1999), Hanushek et al. (2003), Nichols (2003), and Sosin et al. (2004) all examined

various student populations, ranging from kindergarten to university economics students,

to determine which environmental (i.e. classroom or school) or student (socio-

demographic background) characteristics best explained the variance in educational

performance.

Betts and Morell (1999) found significant differences in grades, as one outcome

of educational performance, between different programs. Specifically, they found the

GPAs were lowest in science and engineering and highest in the arts and sciences.

Hanushek et al. (2003) were able to establish that the peer experience in the classroom

had a significant effect. Because members of peer groups tend to have similar

experiences over time through systematic neighbourhood and school choice, they argued,

omitted peer characteristic factors would be common to the entire peer group, including

contemporaneous inputs.

Supporting the findings of Hanushek et al. (2003), Nichols (2003) observed in a

large empirical study of high school graduation student data in Indiana that lower income

students consistently had greater failure rates, sometimes double that of higher income

student graduate groups, and that ethnic majority (i.e. white) females tend to have higher

GPAs than other comparator groups. Interestingly, the results were not equally strong

across all disciplines. Correlation analysis found a consistently strong negative

5

correlation among language, mathematics, and reading scores in relation to yearly

average absenteeism rates. The relation between absences and reading was less clear,

while for mathematics and English it was very clear. Sosin et al. (2004) also found the

technology used in teaching, students’ overall GPA, and whether the student had taken a

prior mathematics course, to have a significant impact on course performance.

Didia and Hasnat (1998) modeled a standard production function approach using

both questionnaire and institutional data. Among the 210 SUNY students enrolled in

introductory financial management courses taught by four different faculty members, a

strong positive statistical relation was found between students’ cumulative GPA score

and their current course grade. That students’ overall average performance should be a

good predictor of academic achievement in any one course is not surprising. What was

unique was their finding that there appeared to be a marginally significant negative

relationship between the number of hours of study and course grades. Gender or age (i.e.

maturity) did not contribute to predicting grades.

Emerson and Taylor (2004) also used a standard production function approach,

but used it to determine whether how a Principles of Microeconomics was taught had an

impact on outcome scores. They specifically investigated whether students learned better

in a traditional lecture course or in a course with an emphasis on classroom experiments.

Students in experiment-based classes seemed to achieve better TUCE and other exam

scores than students following the more traditional pedagogy, but there were potential

problems with positive selection, and teacher effects. Females seemed to benefit most,

but not non-whites.

In an econometric analysis of a large-scale experiment on the placement of

kindergarten students, Krueger (1999) analyzed the impact of randomizing student

placements into small and regular sized classrooms with or without teaching aides. He

found that the average performance on standardized tests increased by four percentile

points the first year that students attended a small class. This could be important for the

present research in that our average college classroom size is significantly smaller than

6

that typically found in freshman introductory university courses. He also found that the

class size had a larger grade impact on minority students, which again is a concern at the

current college study site. Teacher characteristics (e.g. Master’s versus Bachelor’s

degree or length of teaching experience) did not have much of an effect.

Using an education production function, Krieg and Uyar (1997) found that

university students enrolled in introductory business/economics statistics courses

achieved student performance that was significantly better in the spring than fall

semesters. Similar to other studies, they noted that the students’ overall GPA,

mathematics scores, and a parental income surrogate measure were positive and

significant factors in predicting course grades. On the other hand, student gender, being a

transfer student, and examination schedules were insignificant. Living in a dorm, hours

spent working and the percent of Friday classes missed also tended to lower course

grades.

Linking student postal codes to census data describing various socio-economic

characteristics of the communities in which the students lived allowed Johnson (2005) to

estimate that only 40 to 50 percent of the variation in school success rates can be ascribed

to socio-economic characteristics.

2.2. Student Absences and Academic Performance

Although background characteristics may have an influence on academic

performance, as measured by grades, other studies developed research designs to directly

examine the impact of individual student behaviour, specifically students’ decisions not

to attend class. This work is close in spirit to optimization models that incorporate time

allocation decisions.

Siegfried and Walstad (1998) argued that student effort, study time and

attendance have an important influence on students achieving higher grades. While study

time was measured, neither the quality nor the intensity of the learning could be

7

quantified. Johnson et al. (2002) used ordinary least squares (OLS) regression to assess

more directly the relation between effort and student performance. Instead of self-

reported survey data on effort, the authors measured the number of attempts made on

practice quizzes and the time spent on them. They found a positive and significant

relationship between both measures of effort and grades. Again, they did not find gender

to be significant but they did find a positive relation between course grades in an

introductory financial management course and overall GPA.

However imperfect attendance measurement data are as a proxy for effort or

quality of time spent in the classroom, they can be viewed as a distinct input variable that

at a minimum represents a level of effort and perhaps motivation by the student. In that

vein, several studies specifically examined the relationship between class attendance (or

lack thereof) and final course grades. Often these studies were set up as economic

production function models.

Empirical evidence of an important relationship between classroom attendance

and academic performance has been established in many different learning and teaching

environments. Durden and Ellis (1995) found that the attendance effect was non-linear

and mattered only after a student missed more than four classes, with the size of the

negative impact increasing with each additional absence beyond that threshold. They

also found that in addition to absenteeism, students’ GPA and college entrance

examination scores (MSAT, VSAT) are among the most important determinants of

student academic performance. Having taken a calculus course had a significant positive

effect on outcome grades. The finding of a non-linear relationship with a threshold effect

before it negatively affects student performance was also the conclusion reached by Arce

et al. (1996).

Buckles and McMahon (1971) conducted one of the few random experiments that

looked at the differential impact of having only a programmed text but not attending

economics lectures versus a control group that had students attending lectures with the

programmed text. They found that students gained little from attending lectures if the

8

lectures did no more than explain the material covered in the assigned reading. This

study suggests that there must be a benefit to students’ learning if students are to be

expected to attend classes and that it may not be enough to only measure attendance and

not the impact of a lack of attendance.

Marburger (2001) made an interesting observation on why the absences, or lack

of attendance, may not be as easily detected. Consistent with Romer (1993), he

concluded that students quickly determine which teacher is the better instructor from their

perspective, and therefore provides the greatest value-added in terms of attendance. To

the extent that Marburger (2001) found examination scores of students with relatively

high attendance patterns to have a significant positive effect, it follows that classes

missed in which quality of instruction was poor would have less of an impact.

Moore et al. (2003) performed OLS regression on historical data and class

attendance data collected by various science instructors and found a significant positive

association between attendance and course grades. Shimoff and Catania (2001) explored

the potential impact of encouraging attendance through having introductory psychology

students sign-in. They found that having students sign-in did produce a significant

impact on student outcomes, as measured by quiz marks. They also found that students

in the B and C grade range experienced the most pronounced positive effect. The latter

result may demonstrate that students do not always accurately estimate the extent of their

absences.

Van Blerkom (1992) noted that undergraduate psychology students self-reported

that they attended classes more regularly early in the semester and less frequently later in

the semester. Even with these self-reported data, the correlation between class attendance

and course grades was significant. Sophomores reported missing more classes than

freshmen or senior students, and gender was not significant.

9

In another health educational setting, Newman et al. (1981) found that third-year

dental students did demonstrate a significant and positive correlation between the number

of classes missed and numerical final grades. What surprises here is that the authors

could establish a correlation since these senior professional students knew that attendance

was required and it was recorded.

2.3 Synthesis

As the review of the last few studies cited suggests, even in situations where

students are aware of attendance expectations and the implications of absenteeism, some

students still choose to miss classes. With few exceptions, most studies have been able to

show that student class absenteeism has a predictable negative impact on academic

performance as measured by course grades. Despite several attempts, research has not

conclusively established yet the impact on grades of the following variables, among

others: student motivation, student innate abilities, and the amount studying time spent by

students. The potential difference in quality of instruction across teachers has also not

been sufficiently examined in the literature. Since the correct model is not yet known, it

appears important for pedagogical reasons and academic policy and to determine in each

unique learning environment the extent to which student absenteeism is to be acceptable

or accepted.

10

3. The Institutional Environment

We briefly describe below the environment at the Humber Institute of Technology

& Advanced Learning (“Humber”), its student body, and the profile of students from The

Business School who are the subjects of this study.

3.1. Institution and Student Body

In April 2003, the institution then known as Humber College obtained a

designation as an Institute of Technology & Advanced Learning (ITAL). This

designation was bestowed on only three of the 22 community colleges in Ontario by the

provincial government. The designation was in recognition of the advanced level and

high quality of academic programming that Humber College has provided since its

creation in 1967.

Humber is one of the largest colleges in Ontario, with over 16,000 full-time

students, 1,200 of which are enrolled in the University of Guelph-Humber. The latter is a

unique joint venture between the University of Guelph and Humber, whereby students

graduate with both a degree and a diploma from the two respective educational

institutions. Humber’s students are enrolled in over 160 full-time programs including:

diploma, certificate, apprenticeship, bachelor degree, and post graduate certificate

programs. In addition to full-time offerings, Humber provides part-time courses to

approximately 65,000 students who are enrolled in traditional evening courses, on-line

courses, and off-site industry purchased programming that is customized to meet their

specific needs. Finally, Humber is one of only 12 Vanguard Learning Colleges in North

America—and the only one in Canada—selected for their excellence in education and

training.

11

3.2. The Business School

The Business School is the largest of six schools at Humber, with approximately

30 percent of Humber’s total enrolment. The Business School mirrors Humber in that it

offers high-quality advanced programming, both full- and part-time, and at the certificate,

diploma, bachelor degree, and post graduate certificate programs. The School has several

partnerships with various industries and is actively involved in several international

projects, notably with Ningbo University (China), Université de Lyon (France), and

projects in St. Vincent, Tanzania, and Zimbabwe to name a few.

Within the Business School there are several distinct post-secondary program

areas that include Accounting, Business Administration, Business Management, Court

and Tribunal Agent, Fashion Arts, Law Clerk, Logistics, and Marketing programs. At the

post-graduate level, the School offers Human Resource Management, International

Marketing, International Project Management, Professional Golf Management, and

Public Administration certificate programs. The School offers a Bachelor of Applied

Business in e-Business as well as a Bachelor of Applied Arts in Paralegal Studies.

The students included in this study come from the following post-secondary

programs: four- and six-semester Accounting diploma programs, a six-semester Business

Administration diploma program, and two four-semester diploma programs in Business

Management and Marketing. Accounting students take the introductory Microeconomics

course in their third semester. The other three Business diploma programs have a

common first year and it is during the second semester of that first year that students are

scheduled to take the introductory Microeconomics course. In all four programs,

Microeconomics and all other economics course are one-semester course. Only

Accounting and Business Administration students are scheduled to take additional

economics courses: Macroeconomics for the former, and Macroeconomics, Labour

Economics, and Money, Banking and Finance for the latter.

12

4. The Data

We collected the data for this study in the courses we taught during the normal

day-time teaching activities at Humber in the 2003-2004 academic year. The data come

from 13 sections of the Microeconomics principles course for full-time students. The

data were collected over the three semesters (fall, winter, and summer) over which the

course is offered. We collected the data for almost all the sections we taught and focused

on the Microeconomics course only since it is a mandatory course in all business diploma

programs. As noted earlier, students usually take the course in their first year of study.

We collected the data on student attendance throughout the semester, and compiled and

coded student characteristics (including final course grade) and classroom characteristics

at the conclusion of each semester.

The review of the literature suggests that several factors affect student

achievement beyond contemporaneous student and classroom characteristics. This is no

doubt true, as the variety of data and methods surveyed in the previous section shows.

We made a deliberate choice to limit ourselves to contemporaneous student- and

classroom-specific data for several reasons. Firstly, we had complete control over the

data collection and coding and, as a result, were able to avoid measurement error almost

entirely. Secondly, we wanted to use all of the information available from the classroom

at the individual section level, thus naturally setting the stage for a fixed (section) effects

model. Finally, we wanted to understand all usable data prior to enriching the model

with other entry characteristics such as high-school grade-point average. As a result of

our choices, the data set consists solely of primary data.

We note at the outset that attendance is not compulsory in Business School

programs. We recorded attendance in each class and here we measure attendance for

each student as a weighted average ratio by dividing the student’s number of classroom

hours attended over the entire semester by total possible hours over the semester.1

1 In fact, each period lasts for 50 minutes with a five-minute break between each period. The approximation is made to simplify exposition only and does not affect the accuracy of the calculations.

13

Classes amount to a total of three hours per week (four in the summer semester) and

those can be delivered in block times of one hour plus two hours, or 3 straight hours (two

plus two hours in the summer). Our weighted ratio uses this extra information in that it

accounts for each hour attended and hence acknowledges that missing a straight three-

hour block is more serious than missing a single period or a two-hour block. The

measure also picks up variation across sections by accounting for classes not taking place

due to holidays, teacher absences, or two-hour tests or exams taking place during a three-

hour block, for example. This measure thus maximizes the variation in attendance at the

student level by picking up more variability than measures normally used in the literature.

One further advantage is that the absence ratio is simply one minus the attendance ratio.

The class sizes varied between 27 and 39 students per class and so we knew the students

well and therefore had a high degree of confidence in the accuracy of these ratios. To our

knowledge, this way of measuring attendance has not been used yet in the literature.

Table 1 provides descriptions of the variables employed in the baseline OLS

model, along with their means, standard deviations (Std Dev), and minimum and

maximum values.

Table 1—Description of Variables (n = 429)

Variable Description Mean Std Dev Min Max

GRADE Student course final grade 0.642 0.183 0.030 0.980ARATIO Student attendance ratio 0.691 0.221 0.048 1.000ARATIO2 Student attendance ratio squared 0.527 0.278 0.002 1.000DGR Dummy: 1 if female, 0 if male 0.490 0.501 0.000 1.000DPG Dummy: 1 if Business

Administration program, 0 otherwise

0.478 0.500 0.000 1.000

CSZ Class size 33.776 3.350 27.000 39.000D30 Dummy: 1 if three-hour block, 0

otherwise 0.324 0.469 0.000 1.000

DSEM Dummy: 1 if fall semester, 0 otherwise

0.375 0.485 0.000 1.000

DPR Dummy: 1 if professor A, 0 otherwise

0.685 0.465 0.000 1.000

14

The scatter diagram shown in Appendix A allows for a quick view of the key

relationship here. It suggests a fairly strong positive association between GRADE and

ARATIO, with a mass of observations near the upper right-hand corner of the quadrant

and a fairly steep and sparse relationship near the origin. The scatter plot suggests a non-

linear relationship which justifies our addition of the square of ARATIO to allow for a

quadratic curve shape.

We coded student and classroom characteristics based primarily on data

availability, and secondarily on the literature. We comment below on each variable and

our expectations, if any. We refer to Table 1 for variable descriptions.

GRADE, ARATIO, and ARATIO2. The foregoing discussion and the scatter diagram in

Appendix A point to a positive and non-linear association between attendance and

achievement.

DGR. We do not have a precise expectation with respect to the association of gender with

achievement. Many studies have found it not to be a significant factor either way.

DPG. The Business Administration diploma program is the most comprehensive diploma

program offered by The Business School in that it lasts for six semesters and features

significant breadth and depth. We would expect the program to attract the strongest

students and hence being registered in the program to be associated with higher

achievement, everything else equal.

CSZ. Class size is believed by many to affect achievement; witness current elementary

school reforms in Ontario where reductions in class size have been the centerpiece of the

reformers’ arguments. The small variation in class sizes in our sample makes any

expectation unlikely to be fulfilled.

15

D30. Students and teachers frequently express the view that consecutive three-hour

blocks are less conducive to good learning than broken blocks. The reasons commonly

cited include fatigue and saturation with material. We therefore expect that variable to be

associated with lower achievement, everything else equal.

DSEM. Stronger students usually take the Microeconomics course in the fall so we

expect fall results to be stronger than for other semesters, everything else equal.

DPR. The professor (or teacher) dummy accounts for the fact that the authors each taught

a share of the sections in the sample. Given that there are only two teachers in the

sample, the variable is of limited interest at this time. In a similar study with more

instructors, however, one could effectively model different instruction methods and their

effectiveness and examine their impact on grades as well as attendance. This is left for

future work.

16

5. Empirical Results

Table 2 reports the results of two OLS regressions with GRADE used as the

measure of student performance and hence as dependent variable. All the other variables

listed in Table 1 are used as independent variables.

Table 2—Determinants of Student Performance in Introductory Microeconomics (Dependent Variable: GRADE; n = 429)

Independent variable

Model 1 Coefficient

Model 1 t value

Model 1 P value

Model 2 Coefficient

Model 2 t value

Model 2 P value

Constant 0.033 0.40 0.693 0.158 3.07 0.002ARATIO 1.241 7.98 <0.0001 1.146 7.51 <0.0001ARATIO2 -0.531 -4.30 <0.0001 -0.432 -3.56 <0.001DGR -0.017 -1.31 0.190 -0.017 -1.34 0.182DPG 0.005 0.36 0.719 0.007 0.52 0.600CSZ <0.001 0.15 0.880 D30 -0.018 -1.16 0.246 DSEM -0.014 -1.04 0.301 DPR 0.062 4.24 <0.0001 SECTION1 -0.087 -2.66 0.008SECTION2 -0.051 -1.57 0.117SECTION3 -0.164 -4.99 <0.0001SECTION4 -0.121 -3.84 0.0001SECTION5 -0.009 -0.29 0.772SECTION6 -0.040 -1.18 0.240SECTION7 -0.040 -1.16 0.246SECTION8 -0.152 -4.74 <0.0001SECTION9 -0.140 -4.30 <0.0001SECTION10 -0.100 -3.10 0.002SECTION11 -0.041 -1.23 0.219SECTION12 -0.043 -1.37 0.172Adjusted R2 0.459 0.498 ESSb 7.640 6.950 Notes: a The P value is the probability, if the test statistic really were distributed as it would be under the null hypothesis, of observing a test statistic no less extreme than the one actually observed. If it is less than α [say 5%], then we would reject [the null hypothesis that the coefficient is zero] at the α [=5%] level. See Davidson and MacKinnon (1993), pp. 80-81; b Estimated Sum of Squares.

17

5.1. Baseline Model

Model 1 is estimated by regressing GRADE on all the independent variables

listed in Table 1. This is our baseline model (Model 1). We comment on the SECTIONi

variables shown in Table 1 further below. The estimates from Model 1 reveal a

statistically significant and quantitatively large relation between attendance and

performance, even when taking other student and classroom characteristics into account:

the t value on attendance is almost eight. As suspected, the coefficient on ARATIO2 is

negative and statistically significant so the relationship is non-linear. This confirms our

initial interpretation of the scatter diagram in Appendix A: the impact of attendance is

strongest for students who are close to the origin but becomes weaker as ARATIO

increases. This suggests the presence of diminishing returns to attendance. Turning the

result around, this suggests that missing a few hours will have a small negative effect on

achievement while missing a significant number of hours will have a negative and

significantly larger effect on achievement. This is consistent with the findings of Durden

and Ellis (1995), and Romer (1993), for example, in the context of economics classes.

The impact of the professor could not be determined reliably at this point, as there

were too few professor-sections. The Microeconomics course has a common course

outline, outlining the evaluation scheme and schedule to be used so each course should be

very similar in content and evaluation. However, the two professors did use slightly

different approaches to classroom attendance management, and so this aspect of the study

will be examined further as more data are collected.

One of the weaknesses of the baseline model is that it is the restricted version of a

more comprehensive model. Firstly, it is likely that there exist other characteristics at the

section level that influence success in the course. Those variables are not included in the

equation due to data unavailability and hence represent omitted variables. Secondly, the

sample used in estimating Model 1 is a pooled set of repeated cross sections with each

class section (there are 13 of them) representing a cross-sectional unit. The assumption

implicit to the pooling procedure is that both the intercept and slope are constant across

18

sections; that is not necessarily reasonable. A typical procedure with cross-section

models is to specify cross-section fixed effects whereby there are 12 section (=13 – 1)

dummies that take the value 1 for the individual section, 0 otherwise. Given that the data

are repeated cross sections without a time-series component, there is no reason to believe

that random effects would serve any purpose. We turn to that model next.

5.2. Fixed-Effects Model

Results from the estimation of the fixed-effects model (Model 2) are shown in the

columns labelled Model 2 in Table 2. As is often the case in models where dummy

variables account for a significant share of the independent variables, complex patterns of

multicollinearity can emerge. We performed sensitivity analyses on the models and

based on the results, determined that the four independent variables CSZ, D30, DSEM,

and DPR jointly caused the problem in the presence of the section dummies. The first

variable is continuous but the remaining three are dichotomous. We eliminated the

problem by dropping those variables from the fixed-effects model. Results for Model 2

in Table 2 reflect this elimination.

As shown in Table 2, section dummies are statistically significant at conventional

levels in half the sections. This is not a bad result considering the relatively small total

number of sections (13). Even with the presence of fixed effects, ARATIO continues to

be large and highly significant: its t value is 7.51, slightly below the value of 7.98 in the

baseline model. The quadratic term ARATIO2 is still negative and significant, implying

a non-linear relationship between GRADE and ARATIO. As in the baseline model,

DGR and DPG are not significant.

Finally, we perform an F test to evaluate the restrictions imposed by the baseline

model in comparison to the fixed-effects model. The baseline OLS model is the

restricted model while the fixed-effects model is unrestricted. The test evaluates whether

the difference in the error sums of squares (ESS) is sufficiently large. The test statistic is

equal to [(ESSrestricted – ESSunrestricted)/r] / [ESSunrestricted)/(n – k)] with r = 12 restrictions or

19

degrees of freedom on the numerator, and (n – k) = 429 – 17 = 412 degrees of freedom

on the denominator. Using the sums of squares reported at the bottom of Table 2, the F

test statistic is 3.41, which exceeds the critical value of 2.18 with an α = 1% size test.

The test rejects the null hypothesis that the equal-intercept restrictions are correct and

thus supports the fixed-effects model. Correcting the test statistic to account for the fact

that four variables (CSZ, D30, DSEM, and DPR) were dropped in the unrestricted model

does not change the test’s conclusion.

5.3. Sensitivity Analysis

The results are in agreement with some parts of the literature in that student

characteristics such as gender do not exhibit a significant relationship with student

achievement. What is perhaps more unusual is the strength of the attendance effect even

when taking account of available student and classroom characteristics and class section

effects.

In one of our sensitivity analyses, whose results are not detailed here, we

estimated restricted and fixed-effects versions of a model where the dependent variable is

instead ARATIO and the independent variables consist of the variables listed in Table 1,

excluding ARATIO and ARATIO2. The qualitative results are rather similar to those

obtained from Model 1 and Model 2 with the exception of gender, which is significant.

The results suggest that females attend more. We strongly qualify this by noting that the

fit in models using ARATIO as a dependent variable is very poor relative to that of the

models using GRADE as dependent variable. One explanation for those results is that

while contemporaneous student and classroom characteristics fail to capture the

determinants of attendance per se, they capture a significantly larger share of the

determinants of achievement. Even though the models are not strictly comparable due to

the different dependent and independent variables, we note that the fixed effects model

with GRADE as dependent variable explains 2.7 times the variance of the fixed effects

model with ARATIO as dependent variable.

20

6. Conclusions

The literature generally suggests that attendance matters for academic

achievement. There is less unanimity with respect to the other variables that contribute to

achievement, however. The identity of such variables depends on studies’ features such

as samples, levels, academic disciplines, and so on. In this study, we chose to restrict

ourselves to data specific to student activity in introductory Microeconomics classes.

The results of this study indicate that attendance does matter for academic

achievement in the Microeconomics course, even after considering other student- and

classroom-specific characteristics. The evidence suggests that the effect is non-linear: the

effect of attendance on the grade is stronger at lower levels of attendance but levels off at

higher levels of attendance.

There are several potential extensions to this work. Firstly, we are currently in the

process of adding data for the 2004-2005 academic year. In addition to increasing the

reliability of the results, this will allow us to test a (double-cohort) year effect. Secondly,

it is straightforward to convert our continuous attendance ratio into discrete absence

categories following Durden and Ellis (1995) to establish the threshold past which lack of

attendance begins affecting achievement in a negative and significant fashion. Finally,

we intend in later work to match student characteristics at entry (e.g. prior achievement

variables such high school GPA) and possibly socio-economic characteristics with the

student records that currently make up our expanding data set to see how the model and

results change. This will also allow us to examine the effect of prior characteristics on

grades as well as attendance itself.

21

Appendix A: Scatter Plot of Grade on Attendance Ratio Plot of GRADE*ARATIO. Legend: A = 1 obs, B = 2 obs, etc. ‚ ‚ 1.0 ˆ ‚ A A A ‚ A A B BA A ‚ AA C B B A A ‚ A A AA B ‚ A A A A A AD A ‚ A A A A AA B B C B 0.8 ˆ A AA A A A B A ABAC E CA C F ‚ A A A B A A AA A ACA BDAA F A A i ‚ A AB A AB AACDA AA A AC C B n ‚ A A A D E A A E B B AB AB I B A a ‚ A A A A A A AA AA B AAB B l ‚ AA A A C CAABAABA C AB B CBA AAAA DAB B ‚ A ABAA AAB AAA A CCA B C A A AAAB B C 0.6 ˆ B A A A C AA A B BE CA A BAA B o ‚ B A A A A u ‚ A B EA AB B B A AAA A ABAA B A r ‚ A A B A B B BA A A AA A A s ‚ AA A AA AA A A e ‚ B A B A ‚ AA A G 0.4 ˆ A A A A A r ‚ A a ‚ A A d ‚ A B A A A e ‚ A A A A ‚ A A ‚ A A 0.2 ˆ A B A A ‚ AAB A A A ‚ A A ‚ A ‚ ‚ A A ‚ A AA A A 0.0 ˆ ‚ Šƒƒˆƒƒƒƒƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒƒƒƒƒƒˆƒƒ 0.0 0.2 0.4 0.6 0.8 1.0 Attendance Ratio

22

Appendix B: Tabular Summary of Literature Review

23

Table B1: Summary of Literature Documenting Association between Attendance and Grades

Study/ Details

Type of Study Study Population Dependent Variable(s)

Independent Variable(s)

Methodological Issues Key Relevant Findings

Arce et al. (1996)

Empirical controlled experimental study

Conscientious students are more likely to attend classes but would do well even if they did not attend

Attendance conferred a positive effect on learning. Relationship between attendance and performance is probably non-linear and has a threshold before it negatively affects student performance.

Berenson et al. (1992)

Tinto theoretical predictive model measuring the strength of both students’ goal and the respective institutional commitments to understanding the factors influencing failure

First semester freshman college students enrolled in a remedial math course during Fall 1988, N=263

Final grade SAT-Verbal and –Math scores, high school grade point average, Group Assessment of Logical Thinking (GALT), an attitude to math score, and attendance

Stepwise regression of predictor variables for two years, where attendance was required in 1988 and not required in 1987

Results indicated no differences in final grade between the two years, where attendance was required in one year and not in the other. High school grade point average had the highest correlation with the dependant variable, but only accounted for 8% of the variance. The SAT-M accounted for an additional 2% of the variance and was not significant.

Betts and Morell (1999)

OLS Regression analysis of GPA to measure relative effectiveness of high school resources (model GPA as function of characteristics of the students’ schools)

Students who enrolled at U. of California between 1991 and 1993, who had previously attended California public high schools, N=5,000

Cumulative university Grade Point Average on a scale of 0 to 4

High school profile data, including average student and family demographic characteristics associated with their high school

High school environmental effects were measured as well as student characteristics, vs. direct effects of their behaviour (e.g. class attendance)

Most family background variables were highly significant, including: males and ethnic minorities tend to have significantly lower GPA than females or whites; parental income below $50,000 neg. while above $200,000 tapered off. GPA was lowest in engineering and sciences, and highest in arts and humanities

24

Table B1: Summary of Literature Documenting Association between Attendance and Grades (continued) Study/ Details




Brown et al. (1999)

Descriptive study examined the relationship between class attendance in a nursing course and its effect on grades achieved

Students in nursing courses at Humber College. N=342 nursing students of which 80 were first year and 262 were second year

Course grades. All students within a year wrote the same test, not with-standing having different teachers

Class attendance.

4 teachers collected data from nine class sections involving three different nursing courses. A correlation coefficient was calculated for each of the 9 class sections. The impact of the different courses, instructors, and year, were not analyzed. Classes missed were recorded vs. hours missed

4 of the 9 class sections found a significant negative correlation between absenteeism and grades, but 5 groups found no significant relationships. Range of absenteeism was 0-5 for students with 80+% (mean of 1.4 classes missed); students with grades 50% to 60% missed an average of 4.4 classes while students with grades 40% to 50% missed and average of 5.7 classes)

Buckles and McMahon (1971)

Experimental design with random allocation of students to one of two groups: one with, and the other without required class attendance

Two introductory microeconomics course sections, taught by two different instructors at Vanderbilt University

Test score on the final exam, based on the content of a programmed course text.

TUCE pre-test, number of hours studied, course load, high school rank, SAT score, GPA, educational status, dummy variables for gender, section, major, and lecture attendance

Final regression equations included only 6 independent variables as those variables that were found to be unrelated to the dependant variable were removed

Lectures which do no more than explain the material covered in assigned reading do not significantly improve student academic performance

25





Didia and Hasnat (1998)

Modelled a standard production function approach using questionnaire and institutional data

210 SUNY students enrolled in 7 sections of an intro financial management course offered in Fall 1994 and Spring 1995, taught by 4 faculty

Final course letter grade

Prerequisite grades, age, class standing, hours of study, gender, instructor dummy, and transfer status

Used OLS regression as well as ordered-probit estimation (because of discrete dependant variable), and got similar results with each technique

Found strong positive relation between cumulative GPA and course grade. Also true for prerequisite courses. Weak relation between age and course grade Gender played no role. Surprisingly, found a marginally significant negative relation between hours studied on grades

Durden and Ellis (1995)

OLS regression study of survey questionnaire data

Multiple instructor data from several introductory economics courses

Student grades were normalized to a 10-point grading scale to minimize grading effects across different instructors

Self-reported data on absences entered as continuous data and as a dichotomous variable

Used dummy variables to investigate possible impact of: MSAT, VSAT, race, calculus course taken, prior economics exposure, gender, high school preparatory program, extra-curricular activities, credit hours taken per semester, hours worked, and if from N. Carolina

(1) Attendance effect is non-linear and mattered only after a student missed a threshold of four classes, with the size of the negative impact increasing with each additional absence. (2) As number of absences increased beyond 4 the negative impact on grades increases. (3) GPA and college-entrance exam scores (MSAT, VSAT) are among most important determinants of student academic performance; also calculus course had significant positive effect.

Durden and Ellis (2003)

Empirical study using an OLS Regression model on the experience of two different instructors

Principles of Economics classes N = 252

Overall class average grade

Class attendance, GPA and SAT scores

See Table B2 Results suggest, rather strongly that motivation is an independent factor with regard to average scores earned. Some weak evidence that classroom attendance may serve as a proxy for the effects of internal motivation

26





Emerson and Taylor (2004)

Regression of a production function approach, modelling student learning as fn(aptitude, educational background, educational environment, and instructor effects)

9 sections of a university microeconomics course (900 students) of which 2 sections (59 students) relied on classroom experimental learning vs. traditional lectures

Used 33 questions on economics portion of the TUCE on the first and last day of classes

Used measures of aptitude (e.g. GPA and SAT scores), major, student’s prior high school economics course, gender, ethnicity, also used a dummy variable for the experimental learning group

Both dependent variables are potentially subject to censoring problems. Also a potential positive selection bias exists. Pre-course TUCE score is considered as a proxy for pre-course aptitude, so used 2SLS vs. OLS. Instructor level effect difficult to measure

Students in the experimental sections (59) did significantly better on TUCE economic scores than did the 241 students attending traditional lectures. The results also indicate that certain student characteristics, including gender, major, and grade point average, can be used to predict a student’s likely success when choosing between courses that rely on experiments vs. traditional pedagogy

Hanushek et al. (2003)

Econometric study of peer group data. Tried to overcome problems of omitted variables and simultaneous equations biases through use of a fixed effects framework and lagged measures of peer achievement

Makes use of a unique matched panel data set on students and schools to identify the impacts of specific peer group characteristics on academic achievements.

Individual grades not specifically examined. Regressed average per achievement in a grade

Separated peer influence into endogenous (behavioural) effects and exogenous or predetermined (contextual) effects, e.g. Family (race, socio-econ) and school variables

Because members of peer groups tend to have similar experiences over time through systematic neighbourhood and school choice, many omitted historical factors will be common to the peer group. Many poorly measured, or omitted, contemporaneous inputs will also tend to be common to the group

Most important finding is that peer average achievement has a highly significant effect on learning across the test score distribution.

27





Jaggia and Kelly-Hawke (1999)

Econometric production function study using an ordered logit model

Grades 4, 8, and 12 students taking the Massachusetts Educational Assessment Program (MEAP)

1992 MEAP test scores, placing students in one of five outcome categories, analogous to letter grades

1) School input measures: teacher-pupil ratio, per pupil and administrative expenditures 2) socioeconomic measures: percent of single mothers, professionals/ managers, rental units, and crime rate 3)a dummy variable if an urban area

Even though the underlying dependant variable is continuous, only the discrete responses are observed.

Results strongly suggest that higher levels of spending do not have any consistent or systematic relation with student performance. Smaller class sizes (teacher-pupil ratios) lead to better student performance only in the early stages of education

Johnson et al. (2002)

OLS regression study to measure relation between student performance and effort

70 students in an intro financial management course at a Mid Western business school, over two semesters

Final course grade

College GPA, gender, ACT score

Instead of self-reported data on student effort, measured effort by number of attempts and time spent on computer quizzes (maximum of 9)

Positive relation between performance and both measures of effort – number of attempts and log time. No relation for gender. Noted a positive relation natural aptitude and ability (ACT score, GPA) and performance.

Jones and Field (2001)

ANCOVA-based study to evaluate impact of supplemental instruction on final grade performance.

1,359 students in 9 classes of Principles of Accounting

Final course grades

Attendance in supplemental instruction offerings, gender, student’s major and instructor.

Controlled for self-selection by students. Two covariates (students’ SAT and prior GPA) were included to capture students’ aptitude and prior academic performance.

Participation in both voluntary and mandatory supplemental sessions was found to positively associate with final grades. Notably, level of supplemental instruction attendance was a factor

28





Krieg and Uyar (1997)

286 students enrolled in an Intro to Business and Economics Statistics course in 6 separate sections, over 3 semesters, but all taught by the same instructor at the University of the Mid West

Education production function with student performance measured by exam scores and course grade

Function of: a) dummy variables for gender, time of class, semester, dorm residence college transfer, funding, and year and b) continuous % variables for class missed, homework done, program GPA, math %, hours worked

Using teacher prepared test (vs. std. tests) enable measurement of students’ ability to master specific content of the course as defined by the instructor. Single course and instructor controls for data consistency and external determinants for success (see Romer, 1993)

Student performance was significantly better in Spring than Fall semesters. Program GPA, math scores, parental funding were also positive significant factors. Exam schedule, gender, class time, and whether transfer student or not were insignificant factors. Living in a dorm, hours in workplace and percent of Friday classes missed lowered scores.

Krueger (1999)

Econometric analysis of a large scale randomized experiment on class size

Kindergarten students in the Tennessee Student/ Teacher Achievement Ratio (STAR) project N=11,600

Standardized test scores on reading, word recognition and math (SAT and BSF tests)

Class sizes and teacher backgrounds

Project STAR was a 4 yr. longitudinal study in which kindergarten students and their teachers were randomly assigned to 1 of 3 class size groups (small = 13-17, regular = 22-25, and regular with an aid)

Average performance on standardized tests increases by 4 percentile points the first year students attend small classes; class size has a larger effect for minority students; teacher characteristics (master’s vs. bachelor’s, length of teaching experience) has little effect

29





Marburger (2001)

Empirical study using regression specifically to measure the cause and effect relationship between absence from class and academic performance as measured by exam scores. Also used a survey questionnaire

Author’s microeconomics university students N=56 students (main sample)

Exam scores on three separate exams where absences were controlled for when the topic was covered in class to calculate the “pure” attendance effect

Absences linked to when specific exam question topics were covered

Used dummy variables instead of continuous data used dummy variables to regress “pure” attendance effect. Estimated the mean performance on each test if 100% attendance had prevailed Did a survey to elicit student attitudes toward absenteeism and study habits

the sample was divided into two groups: low absenteeism (<6 absences) and high absences (>=6 absences) The mean exam score was significant. Affected by absenteeism. The precise impact of absenteeism is likely institution and instructor specific. Less effective instructors may experience more absenteeism but with less of an impact (inferred from Romer, 1993)

Moore et al. (2003)

Empirical study to determine the relationship between class attendance and course grades in conditions where the importance of class attendance was not stressed and when it was

1,400+ students from various intro science classes. No grade was assigned for attendance.

Course grades from different instructors

Did linear regression on historical data. Class attendance data collected by various instructors

Variability between instructors not accounted for

Y(grades)= 33.1 + 0.55*(attendance) with a correlation coefficient r=0.78 In a large Introductory Biology class, where class attendance was stressed throughout the course, the regression equation was: Y(grades)=40.2+0.52*(attendance) with a correlation coefficient r=0.75

30





Newman et al. (1981)

Correlation study and linear regression calculated. Chi Square statistic determined for 5 grade categories and absences

144 third year dental students for a class where attendance was required

Final letter grades based on 3 multiple choice tests

Absenteeism recorded by a secretary based on an assigned seating chart 10-20 minutes after start of one hour class.

Since attendance was required and students knew that attendance was being recorded, low absences were recorded (2.72 ± 1.72 STD). Course had 39 one-hour classes.

Correlation coefficient of r=-0.437 was determined for numerical grades vs. absences. Observed absences were lower than the expected absences for students in the grade categories. Y = -1.4(x) + 80.4 A correlation between absences and seating order was also found suggesting an environmental factor possibility

Nichols (2003) Empirical study of high school assessment data from one high school in Indiana. Did correlational matrices on the relationship among yearly absences for failing students

Study based on data collected for graduates for 3 years: 2000 (n= 2,000), 2001 (n=2,056), and 2002 (n=2,364)

State proficiency data in English/ language arts and mathematics. Testing data mandated by the state to graduate with a high school diploma

Gender ethnic, economic status, and history of school attendance. Also earlier grade point averages for earlier grades was collected which identified students at-risk

Noted that standardized tests do not measure educational quality and that standardized achievement test, rather than standardized aptitude tests, be used to evaluate school effectiveness.

For each graduation class and for each test of proficiency, lower income students (those qualifying for lunch supplements) had a greater failure rate (sometimes 2x than higher income). School absences become more frequent for students who struggle academically. Ethnic majority females tend to have higher grade point averages than lower income and majority males. Correlational analysis found consistent strong negative correlation among language, math and reading scores and yearly average absences, indicating that standardized test scores vary inversely with average yearly absences. Relation between absences and reading is less clear, while for math and English is t was clear. Low attendance began early and worsened for low-income students

31





Petress (1996) Policy analysis essay

Economic college student attendance policy arguments

Reasons for and against instituting an undergraduate attendance policy

Pro attendance policy arguments seem more constructive to quality education than do anti attendance policy claims

Roby (2004) Empirical study correlating average attendance data by school

Examined schools in Ohio for grades 4, 6, 9, and 12 N=3,171

Results from the Ohio Proficiency tests found Ohio Dept of Education web site

Average attendance data from school records

Problem of non-homogeneity of schools and confounding of other background variables

Showed moderate positive relationship between student achievement and student attendance. Grades 9 and 12 showed the most significant correlation (r=0.78, r2=0.60 and r=0.55, r2=0.29 respectively)

Romer (1993) Two-part statistical research study on incidence of absences and the effect of absences on final grades

1) Absenteeism rates studied at 3 different university macroeconomics courses. 2) Experience of a single class used to regress attendance and grades

2) Overall score on 3 exams, and converted to letter grades

Single instructor’s attendance records for 6 classes and if did problem sets

2) Regressed both attendance data and number of students who had completed all 9 problem sets, plus dropped lowest problem set score, to address differences in student motivation

Estimated that approximately 1/3 of students did not attend classes and that course size and level of math involved appears to have an important effect. Absences had 3x more significant negative impact on course grades than did proxy on problem set completion. “Significant” absences dropped final grades by at least one letter

Salemi (2003) Discusses aspects of an ideal teacher training program to improve the quality of teaching economics

Theoretical discussion of teaching

Mostly notable for the lack of discussion of classroom management, specifically how to deal with absenteeism, but also cheating, test administration, maintaining a conducive learning environment (noise, lateness, student to student interaction

32





Shimoff and Catania (2001)

Quasi-experimental (not clear how groups were assigned)

114 students complete the course of 132 in an intro psychology course were assigned (?) to 2 groups of 57 each: one group had to sign in other not

Grades as measured by 10 best quizzes, but actually only used top 7 quiz marks because students who did well could afford to miss the last quizzes

Class attendance as measured by sign in and teacher

Grades were recorded as letter grades. The grades differences were most pronounced for students with grades in the range B and C

Sign-in no sign in Mean attend. 85.6% 78.5% SD=9.66 SD=11.35 difference was significant t(17)=2.82, p=0.01, r2=0.24 Absenteeism reduced from 14.4% to 21.5% or by 33.3% Sign-in no sign-in correct answer 111.3 105.12 of 144 questions results statistically significant t(112)=2.21, p=0.03, r2=0.04 Recording students’ attendance increased mean % correct from 73% to 77%

Siegfrried and Walstad (1998)

Research review Economic undergraduate students at various universities

Various student outcomes, usually final grades

Various student characteristics, notably class format, student effort, attendance, motivation

Generally finds that student effort, study time and attendance have an important influence on student outcomes, specifically higher grades.

Sosin et al. (2004)

Test and control groups, single equation regression

Multi university students in 30 micro- and macro-economic courses N = 3,494

Performance as measured by TUCE scores

Technology, Web-CT, extra class contact, prior math, GPA, gender, part-time work, class size, credit enrolment

Much of data was self-reported

The effect of the aggregate technology, prior math, and GPA variables were positive and significant in both macro- and micro-economic courses

33





Thompson and Plummer (1979)

Statistical study comparing successful with unsuccessful remedial university students

Remedial freshman English university students N=187, with 133 (71%) males and 54 (29%) females.

Analysed in terms of ability (ACT scores), attendance (measured as a percent of class attendance), and declared majors

Final grades as letter grades: A, B, C, D, and F. No students achieved an A.

Compared to the entire class, males were over-represented in this remedial class

Failing students, although having the highest ability scores, had very low attendance rates compared to successful students, but there was little difference between B, C, and D grade students. In fact C students had the highest and B the lowest attendance rates with D students being in the middle (58.1%, 49.9%, 51.4% respectively)

Van Blerkom (1992)

Quantitative correlation study

Two part study of students enrolled in 17 sections of an undergraduate psychology course over 3 years: Part A involved 959 students at Pennsylvania State University. Part B involved 354 students at the University of Pittsburgh

Grades Attendance as measured by sign-in attendance sheets and self-reporting on a questionnaire

In Part A, a sign-in attendance sheet was used to collect data and give 10% grade credit for class attendance, while a questionnaire was administered for the Part B cohort

The correlation between class attendance and course grades was significant for all 17 sections, ranging from .29 to .73 (median .55). Questionnaire data showed that students estimated that the had missed between 0 and 75 classes during the previous academic year (mean 11.0, STD 11.9, and median 8.0) ANOVA revealed there was a significant class standing effect of this variable, F(4,227) = 2.59, p<0.04, although the effect on gender was not significant, F<1. Sophomores reported missing the most classes

34

Table B2: Additional Information from Selected Studies

Study/ Details

Year Additional Information

Durden and Ellis

2003 Key issue dealt with: “Do variables such as class attendance, GPA, and SAT scores act as proxy measures for motivation”

Moore et al. 2003 Noted several authors (Launius, 1997; Moore, 2003) who found that students expect to receive credit for attending classes, and their rates of attendance drop if they do not receive the credit. Also noted several studies (Berenson et al, 1992;; Hammen & Kelland, 1994) where grades were not related to attendance. In a comparison of 280 students where attendance was stressed for one group and not for another found that average attendance increased from 59 to 70 and the attendance and grade distributions improved remarkably (authors could have calculated R2)

Petress 1996 Five arguments against having an attendance policy for undergraduate college economic students were put forward and refuted: 1) students are adults and can decide what is in their best interest – notably students most likely to object to attendance policies are likely to be in trouble due to lack of attendance, 2) students are customers and it is the seller (ie. teacher) who is obliged to make the product (class) appealing enough to attract students – this quasi-economic model falsely suggests that education is a linearly sold tangible product rather than a multi-person developmental process that is diminished for all students if some students do not participate in the process, 3) the difficulty of differentiating “excused” from non-acceptable absences puts an unreasonable responsibility on faculty – lack of attendance does not need to be categorized, but rather simply noted and only the degree of acceptable absences needs to be determined, 4) strict policies are an infringement on faculty and student academic freedom – where attendance is not legally required, there is some moral dimension if the education is subsidized by a third party for that party to get its money’s worth when paying for classes to be taught, and 5) it is the responsibility of the absent student to catch up on work missed – this is an infringement on the other students and teacher in that they must tolerate the time, resource, and missed peer involvement, accommodation necessary for the wasted catch-up activities

Romer

1993 Attendance could be a proxy for motivation, which is negatively associated with low performance so student motivation to do well should be controlled for. Otherwise, the effect of attendance and other possible motivation-related factors may be overstated and the predictive value of empirical results may be compromised.

Siegfried and Walstad

1998 Notably found that when measurements were not disaggregated, that significant results were not observed. It was the quality and just the quantity of effort in studying that may have had the greatest influence on grade outcomes. Some suggested class format strategies for reducing absenteeism included frequent quizzes, course exams at regular intervals, and/or numerous homework problem sets to encourage student study and participation

Van Blerkom 1992 In the last part of the study examined attendance at several different points throughout the semester in a larger course section and then correlated these to exam grades. The overall average attendance was 87.8% but showed a steady decline during the semester (during the first 2 weeks 93.1% vs. last 2 weeks 82.0%. This decline represented a significant trend, r=-0.82, p<0.001. Also a tendency for attendance to be lowest on Fridays, but was not significant, F (2, 39) = 2.31, p<0.12. Means for Monday = 88.7%, Wednesday = 89.0%, Friday = 85.7%) Six most frequently given reasons for missing class were: a) need to complete and assignment or extra credit project, or to study for another class; b) the class was boring; c) sever illness such as flu; d) minor illness such as headache, cold, or sore throat; e) to tired to go to class because of active social life; and f) oversleeping.

35

References Arce, D.G., J.P. Formby and B. Zheng (1996) “Student Performance and Induced Class Attendance in Introductory Economics: Evidence from a Controlled Experiment.” Working Paper No. 263, Department of Economics, University of Alabama [Abstract only] Berensen, Sarah B., Genda Carter and Karen S. Norwood (1992) “The At-Risk Student in College Developmental Algebra.” School Science and Mathematics, 92(2), 55-58 Betts, Julian R. and Darlene Morell (1999) “The Determinants of Undergraduate Grade Point Average: The Relative Importance of Family Background, High School Resources, and Peer Group Effects.” Journal of Human Resources, 34(2): 268-293 Brown, Beth, Cynthia Graham, Sheila Money and Mary Rakoczy (1999) “Absenteeism and Grades in a Nursing Curriculum.” Michigan Community College Journal: Research and Practice, 5(2): 81-84 Buckles, Stephen G. and Marshall E. McMahon (1971) “Further Evidence on the Value of Lecture in Elementary Economics.” Journal of Economic Education, 2(2): 138-141 Chow, Henry P. (2003) “Exploring the Predictors of Educational Experience and Academic Performance among University Students in Regina.” Alberta Journal of Educational Research, 49(1): [ ] Davidson, Russell and James G. MacKinnon (1993) Estimation and Inference in Econometrics (New York: Oxford University Press) Didia, Dal and Baban Hasnat (1998) “The Determinants of Performance in the University Introductory Finance Course.” Financial Practice and Education, 8(1): 102-107 Druger, Marvin (2003) “Being There.” Journal of College Science Teaching, 32(5): 350-351 Durden, Garey C. and Larry V. Ellis (1995) “The Effects of Attendance on Student Learning in Principles of Economics.” American Economic Review (Papers and Proceedings), 85(2): 343-346 Durden, Garey C. and Larry V. Ellis (2003) “Is Class Attendance a Proxy Variable for Student Motivation in Economic Classes? An Empirical Analysis.” International Social Science Review, 78(1/2): 42-46

36

Emerson, Tisha L.N. and Beck A. Taylor (2004) “Comparing Student Achievement across Experimental and Lecture-Oriented Sections of a Principles of Microeconomics Course.” Sourthern Economic Journal, 70(3): 672-693 Gilbert, Sid (1995) “Quality Education: Does Class Size Matter?” Association of Colleges and Universities of Canada Research File, 1(1), 1-8 Graham, Amy E. and Thomas A. Husted (1993) “Understanding State Variations in SAT Scores.” Economics of Education Review, 12(3): 197-202 Hammen, C.S. and J.L. Kelland (1994) “Attendance and Grades in a Human Physiology Course.” The American Journal of Physiology, 267(6): 105-108 Hanushek, Eric A., John F. Kain, Jacob M. Markman and Steven G. Rivkin (2003) “Does Peer Ability Affect Student Achievement?” Journal of Applied Econometrics, 18: 527-544 Hyde, R.M. and D.J. Flournoy (1986) “A Case Against Mandatory Lecture Attendance.” Bulletin of the Psychonomic Society, 24: 63-64 [Abstract only] Jaggia, Sanjiv and Allison Kelly-Hawke (1999) “An Analysis of the Factors that Influence Student Performance: A Fresh Approach to an Old Debate.” Contemporary Economic Policy, 17(2): 189-198 Johnson, David (2005) “Signposts of Success: Interpreting Ontario’s Elementary School Test Scores.” Policy Study No. 40 (Toronto: C.D. Howe Institute) Johnson, Dean L., Patrick Joyce and Sen Swapan (2002) “An Analysis of Student Effort and Performance in the Finance Principles Course.” Journal of Applied Finance, 12(2): 67-72 Jones, Jefferson and Kent Fields (2001) “The Role of Supplemental Instruction in the First Accounting Course.” Issues in Acconting Education, 16(4): 531-548 Krieg, Randall G. and Bulent Uyar (1997) “Correleates of Student Performance in Business and Economics Statistics.” Journal of Economics and Finance, 21(3): 65-74 Krueger, Alan B. (1999) “Experimental Estimates of Education Production Functions.” Quarterly Journal of Economics, 114(2): 497-532 Loeb, Susanna and Katherine Strunk (2003) “The Contribution of Administrative and Experimental Data to Education Policy Research.” National Tax Journal, 56(2): 415-438 Marburger, Daniel R. (2001) “Absenteeism and Undergraduate Exam Performance.” Journal of Economic Education, 32(2): 99-110

37

Moore, Randy, Murray Jensen, Jay Hatch, Irene Duranczyk, Susan Staats and Laura Koch (2003) “Showing Up: The Importance of Class Attendance for Academic Success in Introductory Science Courses.” The American Biology Teacher, 65(5): 325-329 Newman, Sheldon, Norman J. Schuman, W. Thomas Fields and Loys Nunez (1981) “Dental Students’ Grades and their Relationship to Classroom Attendance.” Journal of Dental Education, 45(6): 360-361 Nichols, Joe D. (2003) “Prediction Indicators for Students Failing the State of Indiana High School Graduation Exam.” Preventing School Failure, 47(3): 112-120 Nofsinger, John and Glenn Petry (1999) “Student Study Behaviour and Performance in Principles of Finance.” Journal of Financial Education, 25: 33-41 Park, Kang H. and Peter M. Kerr (1990) “Determinants of Academic Performance: A Multinomial Logit Analysis.” Journal of Economic Education 21(2): 101-111 Petress, Kenneth C. (1996) “The Dilemma of University Undergraduate Student Attendance Policies.” College Student Journal, 30(3): 387-389 [Abstract only] Pindyck, Robert S. and Daniel L. Rubinfeld (1998) Econometric Models and Economic Forecasts, 4th ed. (New York: Irwin/McGraw-Hill) Raptis, Helen and Thomas Flemming (2003) “Reframing Education: How to Create Effective Schools.” C.D. Howe Institute Commentary 188 (Toronto: C.D. Howe Institute) Roby, Douglas E. (2004) “Research on School Attendance and Student Achievement.” Educational Research Quarterly, 28(1): 3-14 Romer, David (1993) “Do Students Go to Class? Should they?” Journal of Economic Perspectives, 7(3): 167-174 Salemi, Michael K. (2003) “A Model Teacher-Education Program for Economics.” American Economic Review (Papers and Proceedings), 93(2): 455-459 Shanahan, Martin, Christopher Findlay, Judy Cowie, David Round, Ron McIver and Steven Barrett (1997) “Beyond the ‘Input-Output’ Approach to Assessing Determinants of Student Performance in University Economics: Implications from Student Learning Centred Research.” Australian Economic Papers (Special Issue): 17-37 Shimoff, Eliot and A. Charles Catania (2001), “Effects of Recording Attendance on Grades in Introductory Psychology.” Teaching of Psychology, 28(3): 192-195 Siegfried, John J. and William B. Walstad (1998) “Research on Teaching College Economics.” In Teaching Undergraduate Economics: A Handbook for Instructors, William B. Walstad and Philip Saunders, eds.: 141-166 (Boston: Irwin McGraw-Hill)

38

Sosin, K., B.J. Blecha, R. Agarwal, R.L. Bartlett and J.I. Daniel (2004) “Efficiency in the Use of Technology in Economic Education: Some Preliminary Results.” American Economic Review (Papers and Proceedings), 94(2): 253-258 Thompson, Mark E. and Bonnie C. Plummer (1979) “Remedial College Freshemen English Students: Descriptions and Characteristics.” Reading Horizons, 19(3): 248-255 Van Blerkom, Malcolm L. (1992) “Class attendance in undergraduate courses.” The Journal of Psychology, 126(5): 487-494 Wooldridge, Jeffrey M. (2002) Econometric Analysis of Cross Section and Panel Data (Cambridge, MA: The MIT Press)

Does Attendance Matter? Evidence from an Ontario ITALeconomics.ca/2005/papers/0483.pdf · Does Attendance Matter? Evidence from an Ontario ITAL Pierre-Pascal Gendron and Paul Pieper

Documents