You are on page 1of 7

This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON EDUCATION 1

Comparison of the Effectiveness of Online


Homework With Handwritten Homework in
Electrical and Computer Engineering Classes
H. Joel Trussell , Life Fellow, IEEE, and Marcia L. Gumpertz

Abstract—Contribution: This article compares the predictive and Computer Engineering (ECE) at NC State University.
performance of the scores on WeBWorK homework (online) Effectiveness will be measured by the performance of the
with those of standard handwritten homework. The compar- students on the final exams. The classes analyzed range
ison is done across six undergraduate electrical engineering
classes where each of the nine instructors have used both from sophomore to senior-level undergraduate classes. The
homework modalities. classes selected used traditional homework in the past and
Background: Online homework systems have been used for changed to the online system more recently. The same instruc-
many years, but analysis of their effectiveness is mixed. Previous tors taught the classes using the same textbooks and syllabi
work has been limited to a small number of classes in a wide for both modalities.
variety of disciplines. This article has a larger number of classes
and instructors than previous studies. The classes cover many There are several aspects that may be considered when com-
basic topic areas in electrical and computer engineering, so is paring homework modalities, including effectiveness, student
directly applicable to the audience of these transactions. preference, faculty preference, efficiency, and cost. It is noted
Research Question: What is the effect of online homework com- that other aspects of education may also be considered and the
pared to traditional handwritten homework on the performance definition of effectiveness may vary from that used here.
of the students on the final exams in selected ECE classes?
Methodology: Mixed-effects analysis of variance models are The results of any study of online homework may depend
used to determine the predictive ability of performance on home- on the subject areas and the software. The WeBWorK
work of the two modalities on the performance on the final exams. system [1] was designed by mathematicians for undergrad-
The data are limited to classes where the instructors have taught uate math classes. It has been used in classes in disci-
the class using both modalities. These models incorporate the plines that have a large math component, e.g., physics,
effect of modalities for each instructor and the effect of the
modalities across all classes. chemistry, statistics, computer science, and various engineer-
Findings: The result is that there is no significant statis- ing disciplines [2]–[7]. There are other online systems that
tical difference in the two modalities to predict final exam are used in these disciplines but a comparison of those is
scores. This indicates that the advantages of using the automated beyond the scope of this article. It is also true that a soft-
online system can be obtained with no detrimental effect on the ware system may be appropriate for some topics within
students’ learning.
a discipline and not for others. For example, WeBWorK
Index Terms—Effectiveness, handwritten homework, is quite useful for classes that introduce and analyze cir-
online homework, statistical analysis, traditional homework, cuits but is not appropriate for the design of circuits. The
WeBWorK.
classes in which it is used in the NCSU ECE Department
include: 1) math for ECE (ECE220); 2) electromagnetics
(ECE303); 3) electromechanical energy conversion (ECE305);
4) control systems (ECE308); 5) electrical engineering for
I. I NTRODUCTION non-ECE engineering majors (ECE331); 6) digital signal
HE USE of online systems for student homework and
T testing is becoming more common with the improve-
ments in software and its availability. There are still many
processing (ECE421); 7) random processes (ECE514); 8)
power system analysis (ECE451); and 9) electric motor drives
(ECE453).
questions about the effectiveness and advantages of its use The properties of the WeBWorK system that are required
compared to the traditional handwritten submissions. This arti- for effective usage are as follows.
cle will examine the effectiveness of using the WeBWorK 1) Allow input and check of math formulas with multiple
online system for homework in several classes in Electrical independent variables in a syntax-free format.
2) Allow parameter randomization of problem variables,
Manuscript received September 3, 2019; revised December 15, 2019; so the students obtain unique problems; the random-
accepted January 18, 2020. (H. Joel Trussell and Marcia L. Gumpertz
contributed equally to this work.) (Corresponding author: H. Joel Trussell.) ization of the parameters for the problems means that
H. Joel Trussell is with the ECE Department, NC State University, Raleigh, the problems can be used in successive offerings of the
NC 27695 USA (e-mail: hjt@ncsu.edu). class without the worry of the answers being available
Marcia L. Gumpertz is with the Statistics Department, NC State University,
Raleigh, NC 27695 USA (e-mail: gumpertz@ncsu.edu). from previous semesters. (Note that this also reduces the
Digital Object Identifier 10.1109/TE.2020.2971198 likelihood of cheating.)
0018-9359 
c 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: The Ohio State University. Downloaded on February 27,2020 at 20:05:25 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON EDUCATION

3) Allow dynamic graphs, e.g., signal input and output the student’s answer. There is a button to “Email the
can be drawn for the random parameters of a specific Instructor.” This allows the students to ask their ques-
problem. tions and the email to the instructor includes a link to
4) Allow interactive graphics, e.g., students can create the problems with the students’ answers. Very often,
graphs and vector diagrams. the instructor can identify the error from the answer.
These properties allow the creation of online problems that Occasionally, the instructor may require more detailed
reflect realistic engineering problems. In fact, the problems information about the student’s approach, which would
created in WeBWorK at NCSU are closely related to prob- be sent the old fashioned way of emailing the scanned
lems in the texts that are used for the classes. This avoids work. From the NCSU experience, the vast majority
a common problem of matching online library problems to the of questions are answered with the information in the
text used in the class. In the cases where the instructors have “Email the Instructor” notice.
their own problems, custom problems were created for the 4) Problems allow hints based on incorrect student answers.
instructors. 5) WeBWorK collects information that can be used to eval-
The ECE department at NCSU supported the adoption uate problems. Various statistics are computed to help
of WeBWorK by funding teaching assistants (TAs) to work evaluate problems and students, e.g., the average number
with one of the authors to develop the problems for vari- of attempts required for a given percentage of students
ous classes. This extended the work done for his own course solving a problem, a percentile score of correct answers
to others that were mathematically based. The relief of the for each problem. The detailed data can be downloaded
programming burden made the adoption process much more and processed by offline software, such as Excel, for
palatable for instructors who did not have the time to invest more detailed studies. For example, a detailed study of
in learning the WeBWork system. The instructors had only to the number of attempts for each problem could be used
indicate the problems that they wished to have coded from to study the correlation between attempts and overall
past semesters of their courses. It was possible to adapt the performance in the course.
original problems to WeBWorK format in the vast majority
of cases. Instructors could augment the online problems with
handwritten problems if desired, but this was done rarely.
The development of the problems for a semester was done II. P REVIOUS W ORK ON O NLINE H OMEWORK
for each class using the time of the author plus the time E FFECTIVENESS
of a half-time (20 h per week) TA. For the original class The work of [8] gives a good survey of benefits and per-
of the author, his time was probably about 15 h per week. ceived problems of using online homework. While this article
This included the time required to learn the details of the emphasizes its use in accounting classes, the authors mention
WeBWorK system. the results from other disciplines. This is good since there are
WeBWorK was chosen over other systems because it was not a lot of papers that address the effectiveness of online
easy to program the problems of interest for ECE classes and homework. The authors say, “The literature on whether OHS
it was free to faculty and students. This is mentioned in [5], [online homework systems] improves student performance is
which discusses the application to signal processing. The mixed.” They cite some papers that show limited effectiveness
WebAssign system was also available at NCSU. It seemed to and others that have found little or no improvement. Rather
require more effort in programming and had a significant cost than repeat [8]’s references, this article adds more recent
passed along to the students. In talking with the representa- work. Bowman et al. [7] and Buttner and Black [9] found
tives from WebAssign, they indicate that their system provides that online homework helped and Chow [10], Dodson [11],
much the same capabilities mentioned in the list above. Babaali and Gonzalez [12], Woolley [13], and Callahan [14]
Other aspects of WeBWorK that are similar to other systems found no significant differences between online and traditional
and very useful include the following. homework.
1) Problems are easily grouped into problem sets, which There were two papers that indicated limited negative
basically define individual homework assignments. It effects. The work of [15] found that students using online
is easy to rearrange problems into updated sets as homework performed better on solving quantitative prob-
schedules change. lems, but performed more poorly on multiple-choice questions.
2) Problems allow multiple attempts for the students to get A related finding in [16] finds that on tests that are primar-
credit. In WeBWorK, the problems are submitted indi- ily multiple choice, students who are allowed two attempts
vidually, not as a group in the whole assignment. The to solve problems perform more poorly on tests than those
deadline for the homework set applies to all problems in that are allowed only one attempt. This negative effect was
that set. The number of attempts per problem is deter- on multiple-choice problems, which are used rarely in ECE
mined by the instructor and depends on the difficulty and problems developed at NCSU. Those problems predominantly
complexity of the problem. One instructor has recently have answers that are numerical, formulas, or systems of equa-
allowed unlimited attempts for most problems. These tions. Because the problems call for writing numbers and
data have not been analyzed yet. formulas, instructors give students several attempts to get the
3) WeBWorK allows the student an easy way to ask ques- correct answer for each problem. The relation of the number
tions and gives the instructor or TAs information about of attempts and performance on testing is beyond the scope

Authorized licensed use limited to: The Ohio State University. Downloaded on February 27,2020 at 20:05:25 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

TRUSSELL AND GUMPERTZ: COMPARISON OF EFFECTIVENESS OF ONLINE HOMEWORK WITH HANDWRITTEN HOMEWORK IN ECE CLASSES 3

of this article. However, preliminary work at NCSU has indi- WeBWorK, “online homework (WeBWork) with instant feed-
cated no significant statistical correlation between the number back was helpful for learning.” This statement was rated higher
of attempts a student makes and their performance on tests. than the general question, even though all the assignments
The lack of proof that online homework improves stu- were WeBWorK, (3.5/4.1 and 4.2/4.3). The NCSU experi-
dent performance is often noted as a reason that many ence also agrees with other authors that students who have
instructors do not use it in their classes. Most of the refer- taken classes with online homework will sometimes request
ences that found no statistically significant effect on student WeBWorK in classes that do not offer it.
performance, still indicated that the advantages of the online The bottom line is that even if online homework does not
homework make it preferable to traditional handwritten home- significantly improve student performance, it has many other
work. benefits that make it desirable.
While there is no substantial evidence that online home-
work improves performance, there are only limited results
III. D ESIGN OF ECE S TUDY
in the literature that indicate that it is harmful. Indeed, the
results discussed here indicate there is no statistically sig- Online homework was introduced in the ECE department
nificant difference between traditional handwritten homework at NCSU in 2013 in ECE220, analytical methods in Electrical
and online homework on the students’ performances on final and Computer Engineering. The instructor for that class had
exams. Humphrey and Beard [8] and Chow [10] noted many taught the class several times before using traditional home-
of the benefits of the online homework systems. These are work. He developed online problems that were keyed to the
elaborated here. WeBWorK provides the following benefits. problems in the text for the class. All successive instructors
1) Instant Feedback to the Students: Usually, the student have opted to use the online system.
can correct her work if told there is an error. The online system was offered to other ECE classes that
2) Greater Access to Help: WeBWorK allows the student appeared to be appropriate candidates. The authors worked
to e-mail the instructor and teaching assistants a ques- with the instructors who taught the classes to create the
tion about a problem. The e-mail contains a link that problems that resembled those they were currently using.
allows the recipient to see the exact problem and its Some instructors used problems directly related to those in
solution. The student does not have to wait until the the text they were using; others had their own versions of
next scheduled office hours to get help. Note: The problems for their class. It was relatively easy to accommo-
number of e-mailed questions in the classes is usually date both.
small, e.g., the average is less than 5 per homework Since the instructors that used the online system had taught
assignment. the same class previously, the results from the previous
3) Grading is instantaneous and thus requires greatly semesters were used to compare the effectiveness of the online
reduced instructor or TA grading time. This time system. There were no cases where instructors tried the online
can be used for more productive development of the system and reverted to traditional homework. This is likely
class. because they were instrumental in creating the online problems
4) Analysis of the Class Results is Easy: WeBWork pro- for their class.
duces a large database of statistics that can help evaluate A table of the classes for those that use both online and tra-
the students and the homework problems. For example, ditional homework and their related data is given in Table I.
it is easy to see which problems are the most difficult. Three examples of WeBWork problems for ECE220 and
Other data, such as recorded incorrect answers when ECE303 are shown in Figs. 1–3. An online demonstration of
multiple attempts are permitted, can be used to indi- the WeBWork problems created for ECE classes is available
cate guessing; timing of the submissions can be used to at [18].
indicate study habits.
An additional advantage of the WeBWorK software is that IV. DATA A NALYSIS AND R ESULTS
the problems can be reused by the students after grading. There There are two aspects of the analysis that are of interest.
is a button that allows the student to “get another version of The first is the simple relationship between the students’
this problem.” The student can see the correct answers to such performance on the final exam and the modality of the home-
problems, since they are no longer in graded status. Since stu- work used in the class. The second is the effect of homework
dents often ask for additional examples of problems on which style on the relationship between performance on the home-
to practice, this is a benefit to them. work and performance on the final exam, while explicitly
Trussell and Baron [5], Humphrey and Beard [8], considering the natural variation of the populations of the
Woolley [13], and Zerr [17] noted that generally students like classes from semester to semester. SAS Proc Mixed was used
the online homework systems. The evaluations at the end of the to fit both models of Sections IV-A and IV-B.
semester for the classes of interest indicate similar results. The
general evaluation question on the standard NCSU evaluation
form asks for a rating from 1 to 5 of the statement, “the course A. Relation of Final Exam Scores and Modality
assignments were valuable aids to learning.” The average for First, the effect of the WeBWorK modality on the final exam
classes with Webwork was 4.2, the same as handwritten home- score outcome is discussed. To test whether the mean final
work. In a couple of classes that added the specific question on exam scores were different between sections using WeBWork

Authorized licensed use limited to: The Ohio State University. Downloaded on February 27,2020 at 20:05:25 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON EDUCATION

TABLE I
ECE C LASS I NFORMATION , H—H ANDWRITTEN , W—W E BW OR K. B ECAUSE THE N UMBER OF P ROBLEMS VARIES S LIGHTLY F ROM S EMESTER TO
S EMESTER , THE N UMBER OF S ETS AND P ROBLEMS R EFLECT THE M OST R ECENT O FFERING . T HE T OTAL N UMBER OF S TUDENTS
R EFLECTS T HOSE IN A LL S EMESTERS ECE C LASSES

Fig. 1. Example problem ECE220, differential equation, and blank answers.

Fig. 3. Example problem ECE303, electromagnetics, and preview and


answers shown.

Fig. 2. Example problem ECE303, electromagnetics, and answers inserted.

allows for variation among classes, (mc)ij represents a ran-


and sections using paper homework, the following mixed, dom interaction between the modality effect and the class,
effects analysis of the variance model is used to fit the data: sk(ij) allows for random variation among semesters of the same
yijkl = μ + Mi + cj + (mc)ij + sk(ij) + ijkl . (1) class, and ijkl allows for variation among individuals within
the same section of the class.
In this model, (1), yijkl is the final exam score for the lth stu- The estimated means and standard errors for the final exam
dent in the kth section of the jth class using the ith modality scores are:
(handwritten homework or WeBWorK). Note that for the sta- 1) mean of final exams for online mode = 76.35 and std
tistical analysis, a “class” refers to a particular instructor error = 2.69;
teaching a particular course. The μ term represents the mean 2) mean of final exams for handwritten mode = 75.33 and
final exam score, Mi represents the effect of the modality, cj std error = 2.68.

Authorized licensed use limited to: The Ohio State University. Downloaded on February 27,2020 at 20:05:25 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

TRUSSELL AND GUMPERTZ: COMPARISON OF EFFECTIVENESS OF ONLINE HOMEWORK WITH HANDWRITTEN HOMEWORK IN ECE CLASSES 5

Fig. 4. Box plots of distribution of homework averages in each section of each course, sorted in order of increasing median homework score for each course.
WeBWorK denoted (+), traditional homework (o).

There is no significant difference between the two modali- For most sections, the homework averages tended to be highly
ties (p-value = 0.3964). skewed toward the upper end of the 100-point scale (Fig. 4).
As a point of interest, there is not a significant differ- In Fig. 4, the boxes enclose 75% of the homework scores in
ence in mean homework scores between the two modalities each section. In most cases, the boxes lie above 80% and in
(p-value = 0.34). The estimated means and standard errors some, the entire box is between 90% and 100%. Unusually
for the homework scores are: low homework averages for individual students are denoted
1) mean of the homework averages for online by asterisks. The median, denoted by a bar in the figures, is
mode = 86.64 and std error = 2.34; higher than the mean, denoted by a plus or circle, in every
2) mean of the homework averages for handwritten section but one. There is no obvious consistent difference in the
mode = 85.31 and std error = 2.34. distribution of homework scores for sections using WeBWorK
(+) compared to sections using traditional homework (o).
B. Relation of Final Exam Scores to Homework Performance The results of a simple linear regression of final exam score
Next, the relationship between homework and final exam on homework average for each section of each class are shown
scores is discussed. Does one modality exhibit a stronger rela- in Fig. 5. There is wide variability in the slopes and intercepts
tionship with the final exam score than the other? All classes among sections, even within the same course and modality.
but one had graded homework and a final exam. The exception No consistent differences in slope or intercept between sec-
was ECE308, which substituted a project for the final exam tions with WeBWork or paper homework are apparent. In
score. For ECE308, the average of the two in-class exams was most sections, there was a significant correlation between aver-
used in place of the final for the analysis. age homework score and final exam grade, but in seven of
The classes varied in several respects from each other. Even the 40 sections, there was not. Of the seven, three instruc-
sections of the same course, e.g., ECE331, which is taught tors had nonsignificant correlation in both handwritten and
by several instructors, have different numbers of homework online homework. While this might be caused by the instruc-
assignments and different numbers of exams. The numbers tors’ testing styles, it does not indicate any difference in the
of homework sets and problems listed in Table I reflect the performance caused by the modalities.
latest offerings. For this analysis, this article uses the aver- The fitted regression lines are displayed by the instructor
age homework grade and the final exam grade, both of which in Fig. 6. Again, there is wide variability among sections
were normalized to a 100-point scale. The final class grades taught by the same instructor, though instructors 1 and 5 look
could not be used since they differed by weightings of different more consistent than the others. Instructors 7 and 9 each
amounts for different assignments in the classes. had one semester, their first semester teaching the course,

Authorized licensed use limited to: The Ohio State University. Downloaded on February 27,2020 at 20:05:25 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON EDUCATION

Fig. 5. Regression for each section of each course. The fitted lines are displayed separately for each course.

Fig. 6. Regression of final exam score on homework average for each section. Regression lines are displayed separately for each instructor.

where the relationship between homework scores and final slope in Figs. 5 and 6. For one instructor (instructor 6),
exam scores was different from later semesters. The extremely the slopes for WeBWorK look steeper than for traditional
small range of homework scores results in the only negative homework, but for the others, there is no obvious consistent

Authorized licensed use limited to: The Ohio State University. Downloaded on February 27,2020 at 20:05:25 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

TRUSSELL AND GUMPERTZ: COMPARISON OF EFFECTIVENESS OF ONLINE HOMEWORK WITH HANDWRITTEN HOMEWORK IN ECE CLASSES 7

difference between WeBWork and paper homework. Formal and the student evaluations of the online modality are posi-
tests comparing WeBWorK and homework with respect to the tive, readers are encouraged to use online homework where it
relationship between homework scores and final exam scores is applicable.
are provided using data that produced Fig. 6.
A random coefficient regression model was used to test the R EFERENCES
relationship between homework score and final exam score. [1] WeBWorK. Accessed: Feb. 12, 2020. [Online]. Available:
http://webwork.maa.org
The model regressed final exam score on homework score, [2] J. Baldwin et al., “Examples of WeBWorK programming assign-
with separate intercepts and slopes for each modality. The ments,” in Proc. 37th SIGCSE Tech. Symp. Comput. Sci. Educ.,
2006. [Online]. Available: http://ollygotel.com/downloads/examples-of-
model also allowed the intercepts and slopes to vary randomly webwork-programming-assignments.pdf
among classes and sections within a class. The model is [3] C. W. Liew, J. A. Shapiro, and D. E. Smith, “Assessing
student performance in solving complex physics problems,” in
yijkl = μ + Mi + β1 xijkl + Mβ1i xijkl Proc. Conf. Intell. Tutoring Syst., Mar. 2004. [Online]. Available:
http://www.physics.rutgers.edu/∼shapiro/tutor/its04.pdf
+ mk(ij) + bk(ij) xijkl + ijkl . (2) [4] A. G. d’Entremont and J. Abello, “Creating problem taxonomies for
WeBWorK in mechanical engineering,” in Proc. ASEE Annu. Conf.
In (2), yijkl is the final exam score and xijkl is the home- Expo. Mech. Eng. Divis. Tech., Salt Lake City, UT, USA, Jun. 2018,
work score for an individual student. The parameters Mi , β1 , pp. 24–27.
[5] H. J. Trussell and D. Z. Baron, “Creating analytic online homework for
and Mβ1i represent the modality effect, the slope for home- digital signal processing,” IEEE Signal Process. Mag., vol. 32, no. 5,
work scores, and the modality by homework score interaction, pp. 112–118, Sep. 2015.
respectively. The random coefficients mk(ij) and bk(ij) allow [6] D. Cubranic, B. Dunham, and D. Kim, “On-line homework in prob-
ability and statistics: WeBWorK incorporating R,” in Proc. 9th Int.
the coefficients to vary randomly among classes and sections Conf. Teach. Stat., Flagstaff, AZ, USA, 2014. [Online]. Available:
of a class, and ijkl accounts for variation among individual http://icots.info/icots/9/proceedings/pdfs/ICOTS9_C178_DUNHAM.pdf
[7] C. R. Bowman, G. Ozcan, and D. B. King, “Predicting student success
students in the same section of a class. via online homework usage,” J. Learn. Design, vol. 7, no. 2, pp. 47–61,
The results of interest in combining all sections and classes 2014.
[8] R. L. Humphrey and D. F. Beard, “Faculty perceptions of online home-
are: work software in accounting education,” J. Account. Educ., vol. 32, no. 3,
1) the estimated slope for the online mode = 0.3983 and std pp. 238–258, 2014.
error = 0.026, the estimated slope for the handwritten [9] E. H. Buttner and A. N. Black, “Assessment of the effectiveness of an
online learning system in improving student test performance,” J. Educ.
mode = 0.3544 and std error = 0.027, and p-value for Bus., vol. 89, no. 5, pp. 248–256, 2014.
the test of equal slopes = 0.2478; [10] A. F. Chow, “Online homework impact in undergraduate mathematics
and business statistics courses,” Educ. Stud., vol. 41, no. 3, pp. 244–248,
2) the estimated intercept for the online mode = 42.86 and 2015.
std error = 2.66, the estimated intercept for the hand- [11] J. R. Dodson, “The impact of online homework on class productivity,”
Sci. Educ. Int., vol. 25, no. 4, pp. 354–371, 2014.
written mode = 44.78 and std error = 2.83, and p-value [12] P. Babaali and L. Gonzalez, “A quantitative analysis of the relationship
for the test of equal intercepts = 0.6251. between an online homework system and student achievement in pre-
From this, one can see that there is no statistically signifi- calculus,” Int. J. Math. Educ. Sci. Technol., vol. 46, no. 5, pp. 687–699,
2015.
cant difference between the two homework modalities in the [13] D. J. Woolley, “Which helps accounting students learn more: Traditional
relationship between homework scores and final exam scores. homework, online homework, or clickers?” Acad. Educ. Leadership J.,
vol. 19, no. 3, pp. 337–343, 2015.
Note that because the exam scores and homework aver- [14] J. T. Callahan, “Assessing online homework in first-semester calculus,”
ages are capped at 100%, some of the usual assumptions of Problems Resources Issues Math. Undergraduate Stud., vol. 26, no. 6,
pp. 545–556, 2016.
the mixed-effects model (2) are not satisfied. This can be [15] D. J. Fatemi, L. Marquis, and S. Wasan, “Student performance in
addressed by regressing the logit or log odds of the exam intermediate accounting: A comparison of the effectiveness of online and
scores on the logit of the homework averages rather than manual homework assignments,” Account. Educ. J., vol. 24, pp. 1–19,
Jan. 2014. [Online]. Available: http://aejournal.com/ojs/index.php/aej/
regressing the original exam scores on the homework aver- article/view/259/138
ages. If this is done, neither the test for equality of slopes nor [16] R. K. Orchard, “Multiple attempts for online assessments in an opera-
tions management course: An exploration,” J. Educ. Bus., vol. 91, no. 8,
the test for equality of intercepts shows a significant difference pp. 427–433, 2016.
due to modality (p-value = 0.76 for the test for equal slopes [17] R. Zerr, “A quantitative and qualitative analysis of the effectiveness
of online homework in first-semester calculus,” J. Comput. Math. Sci.
and p-value = 0.59 for the test for equal intercepts). Teach., vol. 26, no. 1, pp. 55–73, 2007.
[18] NCSU_ECE_Demo. Accessed: Feb. 12, 2020. [Online]. Available:
V. C ONCLUSION https://demo.webwork.rochester.edu/webwork2/NCSU_ECE_Demo/
Based on the results from 40 sections of six different ECE H. Joel Trussell (Life Fellow, IEEE) received the Ph.D. degree in electri-
courses taught by nine different instructors from 2011 to 2018, cal engineering and computer science from the University of New Mexico,
there was no significant difference between the performance of Albuquerque, NM, USA, in 1976.
He is currently a Professor of electrical and computer engineering with
students that took classes using the online homework modal- NC State University, Raleigh, NC, USA. He has a long research history in
ity and those that took classes using traditional handwritten signal and image processing and a shorter history in education.
homework from the same instructor. It was noted that online Marcia L. Gumpertz received the Ph.D. degree in statistics from North
homework has several advantages for students, such as imme- Carolina State University (NC State), Raleigh, NC, USA, in 1989.
diate feedback, timely assistance on questions, and availability She is currently a Professor of statistics with the Department of Statistics,
North Carolina State University. She has coauthored the book Planning,
of practice problems. The online homework has the obvious Construction and Statistical Analysis of Comparative Experiments. She is
benefits for instructors of time savings on grading, uniform a PI of the AGEP-North Carolina Alliance: An Institutional Transformation
grading of problems, reduction of the likelihood of cheating, Model to Increase Minority STEM Doctoral Student and Faculty Success.
Prof. Gumpertz is a member of the Statistical Consulting Core providing
and recording data for easy analysis of problem difficulty and statistical consulting services to the NC State community. She is a fellow of
effectiveness. Since there are no disadvantages for the students, the American Statistical Association.

Authorized licensed use limited to: The Ohio State University. Downloaded on February 27,2020 at 20:05:25 UTC from IEEE Xplore. Restrictions apply.

You might also like