Online Homework and Student Performance: Evidence from a Natural Experiment Karen Menard, Bridget O’Shaughnessy and Abigail Payne Online Homework and Student Performance: Evidence from a Natural Experiment Karen Menard** Bridget O’Shaughnessy* Abigail Payne* With Olesya Vovk* and Katherine Swierzewski* May, 2014 Preliminary Draft Do not cite Abstract: Recent research has shown that online homework in economics courses can provide a small but statistically significant improvement in student achievement. During the fall 2010 semester, it was brought to the attention of instructors at McMaster that the provincial government's policies regarding ancillary fees at universities prohibit the mandatory use of online resources linked to students' grades. In Introductory Macroeconomics, the enforcement of the policy had a dramatic impact on how the course was delivered. Prior to fall 2010 online homework, delivered through a website called "Aplia", comprised twenty percent of students' grades. Access to the website came free with the purchase of a new text book. Students who did not wish to purchase a new hard copy book could purchase access to the Aplia website which included an electronic copy of the text. For the winter 2011 semester, the online homework component was dropped, and students were evaluated using only mid-term tests and a final exam. During the 2011-2012 academic year, students had the option of doing online homework for grades, or just having their grade made up of the tests and exam. The enforcement of the provincial policy regarding the use of online tools provides a natural experiment that can be used to study the effectiveness of such tools - over 5000 students from across campus are involved in the study. Preliminary results indicate that the online homework improves student learning outcomes, particularly for weaker students. *Department of Economics and Public Economics Data Analysis Laboratory (PEDAL), McMaster University ** Ontario Institute for Cancer Research Online Homework and Student Performance: Evidence from an Online Experiment 1. Introduction With class sizes ever increasing, and teaching resources constrained, instructors would like to use high quality online resources to assess students. The current study will address the following question: are online resources effective as teaching and learning tools? As explained in more detail below, we study this question by taking advantage of the natural experiment that occurred when the provincial policy around the use of online resources was strictly enforced in 2011. From 2008 to 2012, all sections of a large enrollment economics course during the regular school year (September-April) were taught by the same instructor, using the same text book and largely the same evaluation methods. Until the Fall 2010 term, students wrote two term tests and a final exam and did mandatory online homework, which was provided by a publishing company for a fee. The fee included an online version of the text book at a lower price than purchasing a new edition of the paper book. During the fall 2010 term, it was brought to the attention of the instructor that the provincial policy regarding ancillary fees prohibits instructors from mandating that students purchase access to such websites for graded work. As a result of this new information, the online homework portion of the grading scheme was dropped for the winter 2011 term. After consultation with administrators, it was determined that the use of such online learning tools was acceptable as long as they were optional to students. Beginning in fall 2011, students could purchase access to the online homework and use it for grades. Students who did not choose to purchase the online homework package had the weight transferred to their final exam. A breakdown of the grading scheme for each term in the study is given in Table 1. The two term tests were not weighted equally – the test with the higher score was given more weight in the calculation of a student’s final grade. The exact weights are given in parentheses. The online homework included a math test designed to review basic math concepts taught in high school, weekly homework assignments based on lectures and textbook readings, and, when the homework was mandatory, assignments based on online experiments. During the experiments, hundreds of students met online in a simulated market. Each student was given a role as either a buyer or seller, or a lender or borrower, and placed bids through their internet browser. When the computer found a match between two parties, the transaction was recorded. Online discussions followed several rounds of trading in the market. For each of the two experiments in 2010, students were required to complete two homework assignments, one prior to the experiment and one after the experiment was completed. When the online homework system became optional, the experiments were dropped from the course and the only online homework assignments were based on the lectures and textbook readings. In our analysis, we will examine student performance in the course overall as well as their performance on individual tests and the exam. Table 2 shows the timing of the homework assignments relative to the tests and final exam for the two terms with online homework. The remainder of the paper is structured as follows: Section 2 presents a literature review; Section 3 describes the methods used and data set development. Section 4 presents our results, and Section 5 concludes. 2. Literature Review There are two bodies of literature that directly inform this study. The first relates to student retention. Finnie and Qiu (2008) find that first year students are more likely to leave university than students who make it into their second year of studies. We know that students who have a connection to faculty tend to perform better than students who do not. Most first year students face large classes and we also know that it is difficult to get this connection to faculty in classes of 400 to 600 students (Nagda et al, 1998, Chickering and Gamson, 1987). Finnie and Qiu also report that the most common reason students leave university is “didn’t like it/not for me”. Chickering and Gamson emphasize time on task as a core principle in good teaching. With very large classes, offering traditional paper homework assignments, graded by the instructor or by teaching assistants, is impractical. Instructors are using online homework both to engage students and to provide students with more time on task, so gathering evidence of its efficacy is an important area of work. Bosshardt (2004) studies the decision of first year economics students to drop the course. Based on student characteristics, he divides the class into groups that are “good” students or “at-risk” students, those with a high probability of receiving a grade of D or F. Some of the at-risk students dropped the class while others did not. Bosshardt finds that of the at-risk students who dropped the course, only 16% remained at the university they were attending. One particularly interesting aspect of our project will be whether or not there are differential impacts with online homework – does the online homework best serve good students or atrisk students? Does the answer depend on whether or not the homework is required or optional? Another body of work that is relevant pertains to the effectiveness of online homework in economics. In a survey of 700 students conducted by Kennelly and Duffy (2007), students overwhelmingly favoured online homework compared with traditional paper homework, felt that using the publishers’ online homework tool helped improve their understanding of the course material, and wanted their instructors to use the website in the future. Nguyen and Trimarchi (2010) study two different publishers’ online homework tools, finding that online homework had a small but statistically significant impact on student grades. Use of such websites led to a 2% increase in grades whether or not the online homework was required. The experiments Nguyen and Trimarchi (2010) conducted involved several sections of the course taught by the same instructor, some sections having online homework and others not having online homework. In one experiment, no variables were included to control for ability or other student characteristics. In the other experiment information about student characteristics was obtained via a student survey. Our study uses administrative data and information from students’ university applications to determine student characteristics, which are likely to be more reliable than survey responses. Additionally, we are including three distinct cases in our study – one with mandatory online homework, one with optional online homework, and one with no online component at all. Lee, Courtney, and Balassi (2010) compare online homework with traditional paper homework graded by a teaching assistant. The core measure of student learning is the difference between pre- and post-test scores on the Test of Understanding in College Economics (TUCE), a commonly used measure in the economic education literature. They find that differing homework methods do not affect students’ improvement in TUCE scores, despite anecdotal evidence that students prefer online homework. One drawback of the study is the small sample size – they consider three sections of a microeconomics principles course with enrollments of between 46 and 77. When we are considering large classes with approximately 1,500 students in each of the three semesters of analysis, the choice is often not between paper-based homework and online homework, but between online homework and no homework at all. Clearly, this ties in with Chickering and Gamson’s emphasis on time on task. 3. Methodology and Dataset Development 3.1 Natural Experiment Framework To understand better the effectiveness of online homework on student performance in large enrollment classes, we are utilizing a natural experiment type of methodology. As explained above, prior to the winter term of 2011, the instructor of a large enrollment economics course (five sections over an academic year, 2500 students across these sections) the online homework was mandatory. For the winter term of 2011, no online homework was offered. Starting in the fall term of 2011, online homework was offered but was an optional component. Thus, we can study two natural experiments. Experiment #1 is the comparison of a control group that does not have any online homework (winter 2011) with a treatment group that has required online homework (winter 2010). Experiment #2 is the comparison of a treatment group of students that are offered optional online homework (winter 2012) with two control groups: those with no online homework (winter 2011) and those with required online homework (winter 2010). Note that we are only using students enrolled in the winter terms as we have observed slight differences in student background characteristics across the fall and winter terms. Thus the comparison of the effects no online homework, mandatory online homework, and voluntary online homework regimes are best done only on winter term students. In an ideal setup, we would observe the same student under both the control and the treatment conditions. When it comes to matters such as education and other social science based experiments, however, it is rare to accomplish this. Instead we must compare similar students that are exposed to different treatments. This requires a level of confidence that the students exposed to the different treatments are similar in all ways except for the instrument being tested (the use of online homework). Fortunately, in our set up, we have the students being exposed to the same instructor, the same lectures, the same text book, and the same test bank of multiple choice questions. The students may differ in terms of program of study (e.g. engineering, commerce, economics, social science) and they may differ in terms of the level of preparation. We have developed a rich data set that will help us to control for these differences in our multivariate analysis. 3.2 Data We have constructed the data set by linking the following measures from four main sources of data. The primary source of data was the course data collected by the instructor. The data are collected for all students who registered to take the course in three winter terms: winter 2010, 2011 and 2012. We have developed measures to reflect marks received by the student on two term tests, the final exam, as well as marks on each of the 14 homework assignments and math evaluation test for the terms where homework was offered. Additional measures were created to capture the weight of each test and each homework assignment used to calculate the final course grade. Both the term tests and the final exam consisted of multiple choice questions and short answer questions. To ensure consistency, we are using only the multiple choice sections from the term tests and final exam for all terms under study. The second data source was obtained from the university’s Office of the Registrar and includes information on other courses taken by the students. From that source, we created measures to capture overall student performance in the fall term preceding the course as well as the winter term of the course. We calculated the number of credits taken (full or partial load), average GPA in each term, and flagged any other economics courses taken by the student, as well as previous attempts at the course under study. Other measures include the level of enrolment (level 1, level 2, etc), and program of study where the student is registered during the term in question (e.g. engineering, commerce, social science, humanities). The Registrar’s data also provide us with information on students who withdrew from the university entirely but were listed in the course data as “failed the course”. The third data source was information on the applications submitted by the students for admission to the university through the Ontario University Application Centre. The data records include applications for students applying directly from an Ontario high school (known as “101” students) and delayed entry and/or non-Ontario high school students (known as “105” students). The 101 set of applications capture information on the students’ performance in level 4 (grade 12) courses in high school and their home postal code. From the 105 set of applicants the high school grade information was more limited. The location of their residence and home postal code (if from Canada), however, were available. Using the high school course information, we calculated the average of the best 6 grade 12 U-level courses, which is equivalent to the general admission requirements for the university. The fourth source of the data was the 2006 Canadian Census. Using the home postal code measures, we added the information on socio-economic characteristics of students’ neighbourhoods such as average household income, share of immigrants, age and education profiles. The census geography utilized was the dissemination area, a geography that covers roughly 500 households. Table 3 captures basic information on the students who took the economics course in 2010, 2011, or 2012 and compares them with entry level students at the university in these years as well as entry level students across all universities in Ontario. Compared to the university as a whole, the economics course contains a higher share of male students, higher shares of commerce and engineering students, and higher shares of immigrant students. There are lower shares of science/health and humanities students. Overall, however, the students share similar high school GPAs and other characteristics. In Table 4 we report summary statistics that compare students enrolled in the three terms the macroeconomics course was offered. Many of the demographic characteristics of students in the three terms are comparable. About the same proportion of students attended 6 or more years in a Canadian school system, speaks English as their first language, applied for OSAP, and come from low-, middle-, and high-income neighbourhoods. Performance in high school is also similar across students registered in all three terms. The average of the students’ best six grade 12 courses is virtually identical, at about 86%, and almost all 101 students had at least one grade 12 math course. Another strikingly similar characteristic is the proportion of students carrying a full time course load during the terms under study, 83-84% in each term. There are a few moderate differences across terms. In 2010 the proportion of males taking the economics course was lower while the proportion of Canadian citizens was higher. A smaller proportion of the class was comprised of business and engineering students in 2010 than in the other terms. This is likely due to scheduling issues, and it seems plausible that business and engineering students took the course in the fall 2009 term rather than the winter term in 2010. The largest differences can be seen in the students’ performance on the online math test (for the two semesters the online homework was available) and in the percentage of the class enrolled in level 1. The average in the online math test was 92% in 2010 and 82% in 2012. Incentives may matter – in 2010 the math test was worth 5% of the students’ final grade, while in 2012 it was only worth 3%. Level 1 students make up a higher proportion of the class in 2012 than either 2010 or 2011 (80%, 68%, and 72% respectively). Overall, the students appear quite similar in all three of the terms under study. 3.3 Empirical Model Our focus is on student performance on the multiple choice section of two term tests and the final exam. Student i’s test score, Ti, depends individual student information obtained from course records, the Registrar and the student’s university application, Xi, and information about the dissemination area the student resided in when they applied for university, Ni. (1) T = Ⱦ + Ⱦଵ Hi + X ȾX + N ȾN + ɂ Individual characteristics include a dummy variable for online homework, Hi, faculty, gender, level at the university (first year or higher), English as a second language, high school average grade (best six courses), OSAP, years in Canadian school system, age, citizenship, and a dummy variable for ‘105’ students. Neighbourhood characteristics include measures of income and educational attainment, visible minority percentage, percentage of one-parent families, youth unemployment, population, and proportion of population between the ages of 15 and 24. Robust standard errors are clustered by DA level. Standard errors are clustered when members of a group (in this case census dissemination area, approximately 500 households) are likely to share some characteristics that are unobservable. 4. Results 4.1 Mandatory Homework Compared with No Homework Figure 1 shows the average score on the multiple choice section of two term tests and the final exam for the year in which online homework was mandatory and year without online homework. Mean scores are slightly lower on the term tests, dramatically lower on the final exam. When online homework was mandatory, the mean on the final exam MC section was 67.8%, compared with 55.2% in the term with no homework. Distributions of the test and exam grades are given in Figure 2. Again, the difference in term test grade distributions is not large. For the final exam, the shapes of the distributions are strikingly similar, but the entire distribution is shifted to the left in the term with no online homework. Regression results are provided in Tables 5A and 5B. Our key variable of interest, a dummy variable equal to one for students with online homework, is statistically significant for both term tests and the final exam. A student’s grade on each term test is likely to be about two percentage points higher in the term with mandatory homework, while the final exam grade is over twelve percentage points higher. Faculty of enrolment has a statistically significant impact on test performance in most cases. Relative to Social Sciences students, Business and Science/Health Science students perform better on both tests and the final exam. The only significant difference for Engineering students is on the final exam, outperforming Social Sciences students by over three percentage points. Humanities students fare worse on all three assessments, but the difference on the first test is not statistically significant. Any grade difference of more than three percentage points is meaningful to the students in the sample. The university under study uses a twelve point grading system to calculate cumulative averages, where an “F” equals zero, “D-” equals one, “D” equals two, up to “A+” equalling twelve. For grades in the “D”, “C”, and “B” ranges, an increase in grade of three percentage points is enough to move students up to the next grade level. For example, a student with a grade of 67% has a C+, which translates to six out of twelve, while a student with 70% has a B-, which is seven out of twelve. Gender, high school grades, and being an upper year student all have statistically significant and meaningful impacts on test and exam performance. The largest of these coefficients is on “upper year”, with upper year students scoring over five percentage points better on the term tests and almost three percentage points better on the final exam. Females perform between one and two percentage points worse than males, and a one percentage point increase in high school average leads to about a one percentage point increase in test and exam performance. Census measures are all numerically small and insignificant, with the exception of income. Students in lower income neighbourhoods have slightly lower scores on both term tests and the final exam. However, only two coefficients are statistically significant: the final exam for the lowest tercile and the first term test for the middle tercile. 4.2 Mandatory Homework Compared with Optional Homework Figures 3 and 4 show the test/exam means and grade distributions for mandatory homework and optional homework. These figures look strikingly like Figures 1 and 2, where we compared mandatory homework with no homework. Performance on the first term test and final exam are lower in the year with optional homework, with the final exam mean dropping by more than eleven percentage points when homework is optional. The mean scores on the second term test are identical in the two years. Optional homework leads to a four percentage point decrease in Test #1 and almost twelve percentage point decrease in the final exam compared with mandatory homework. Performance on test 2 is not affected by the optional nature of the online homework. A student’s faculty of enrollment affects test/exam performance, with Business, Engineering, and Science/Health students outperforming Social Science students on all assessments by 1.5 to 5.6 percentage points. All are statistically significant except for Engineering students’ performance on the second term test. Being an upper year student, male, a native English speaker, and having a higher high school average all contribute to higher scores on the tests and final exam. Living in a low income neighbourhood negatively impacts test/exam performance. None of the other census variables are statistically significant or meaningful in terms of coefficient size. 5. Conclusions Does online homework improve student performance on term tests and exams? When the homework is mandatory, the answer is yes. Students with mandatory online homework did better on term tests by two to four percentage points than students with no homework or with optional homework. Final exam results were even more striking – mandatory online homework leads to an eleven to twelve percentage point increase in the final exam score, again compared with no homework or optional homework. These results were obtained using large samples of over 2500 students, controlling for many individual and neighbourhood characteristics. The next step in this research is to compare the performance of students who chose to do the homework with those who did not during the term when homework was optional. We also need to examine the relationship between homework and test performance for weaker students compared with stronger students. We will conduct propensity score matching analysis to further explore the relationship between online homework and test/exam performance. References Bosshardt, W. (2004) "Student drops and failure in principle courses", Journal of Economic Education, 35(2)111-28. Chickering, A., Gamson, Z. (1987) Seven principles for good practice in undergraduate education, American Association of Higher Education Bulletin, 39 (7) 3-7. Finnie, R., Qiu, H. (2008) The patterns of persistence in post-secondary education in Canada: evidence from the YITS-B dataset, accessed 19 March 2014 at http://higheredstrategy.com/mesa/pub/pdf/MESA_Finnie_Qiu_2008Aug12.pdf. Kennelly, B., Duffy, D. (2007) Using Aplia software to teach principles of economics, paper presented to Developments in Economic Education Conference in Cambridge, UK, 6-7 September. Lee, W., Courtney, R., and Balassi, S. (2010) Do online homework tools improve student results in principles of microeconomics courses? American Economic Review, 100(2) 283-86. Nagda, B. A., Gregerman, S. R., Jonides, J., von Hippel, W., Lerner, J. S. (1998) Undergraduate student-faculty research partnerships affect student retention, The Review of Higher Education 22 (1) 55-72. Retrieved from http://wwwpersonal.umich.edu/~jjonides/pdf/1998_6.pdf Nguyen,T. and Trimarchi, A. (2010) Active learning in introductory economics: do MyEconLab and Aplia make any difference?, International Journal for the Scholarship of Teaching and Learning, 4(1)10. Table 1 Winter, 2010 Term Tests (25%, 15%): Final Exam: Online Math Test: Online Homework: 40% 40% 5% 15% Winter, 2011 Term Tests (30%, 20%): Final Exam: 50% 50% Winter, 2012 Term Tests (25%, 20%): Final Exam: Online Math Test: Online Homework: 45% 40% 3% 12% Table 2: Course Schedule, Winter 2010 and Winter 2012 Winter, 2010 Week 2 Date Jan. 15 3 4 Jan. 18 Jan. 22 Jan. 25 5 6 Jan. 29 Feb. 6 Feb. 12 7 8 9 Feb. 15-19 Feb. 24 Mar. 5 10 Mar. 13 11 Mar. 19 12 Mar. 22 Mar. 26 Mar. 29 13 14 Apr. 5 Apr. 9 Apr. 14 Winter, 2012 Homework Math Test Ch. 2 Experiment #1 Prep Ch. 3 Experiment #1 Analysis Ch. 4 Test Test # 1 (1-4) Ch. 5 Ch. 6 (practice) Reading Week Ch. 7 Ch. 9 (practice) Ch. 10 Test #2 (5-10) Ch. 11 Ch. 12 (practice) Experiment #2 Prep Ch. 13 Experiment #2 Analysis Ch. 14 Ch. 15 Final Exam Week 2 Date Jan. 13 Homework Ch. 2 3 Jan. 17 4 Jan. 24 Ch. 3 Math Test Ch. 4 5 6 Jan. 31 Feb.7 7 8 9 Feb. 11 Feb. 14 Feb. 20-24 Mar. 2 10 Mar. 6 11 Mar. 10 Mar. 13 Ch. 12 12 Mar. 20 Ch. 13 13 Mar. 27 Ch. 14 14 Apr. 3 Ch. 15 Apr. 7 Test Ch. 5 Ch. 6 Ch. 7 Test # 1 (1-6) Ch. 8 Reading Week Ch. 9 Ch. 10 Ch. 11 Test # 2 (7-11) Final Exam Table 3: Sample comparison with the student population Sample students Total students in sample Delayed or non-Ontario students (% 105 students) Gender (% male) Immigrant status Canadian Citizen Permanent Resident Other Years in Canadian K-12 school system 6 years or more 3-5 years 2 years or less % with English reported as their primary language Indication on application that an application for financial aid was submitted (% applied) Students with ON postal code (3) % living in low income neighbourhood High School GPA (Best 6 University or Mixed Courses) Average Best 6 GPA (standard deviation) (Note: Differences of group means are statistically different from zero at 1% confidence level across all groups) 5th percentile of best 6 Median best 6 95th percentile of best 6 Students by Program of Registration, Year 1 (4) Commerce (includes Social Sciences) Engineering Science/Health Humanities Other NOTES: (1) 4,295 11.0% 63.4% All (1) students at university under study (2) 18,505 All(2) Ontario students (3) 255,633 46.7% 43.9% 77.8% 11.0% 11.2% 88.4% 7.3% 4.3% 91.1% 5.7% 3.1% 76.0% 13.4% 10.6% 65.6% 88.3% 7.2% 4.5% 75.8% 90.8% 5.1% 4.1% 76.9% 61.7% 67.6% 65.4% 4,019 15.3% 18,427 15.3% 254,933 20.3% 86.2% (5.1) 86.1% (5.5) 83.3% (6.5) 78.2% 85.7% 95.5% 76.8% 86.0% 95.5% 72.5% 83.5% 93.8% 26.7% 28.2% 24.5% 13.5% 7.2% 11.9% 19.1% 33.2% 30.8% 5.0% 14.1% 9.0% 28.4% 40.0% 8.5% (1) All students who applied directly from HS in 2010, 2011 and 2012 and were registered at the university under study. (2) All students who applied directly from HS in 2010, 2011 and 2012 and were registered at any University. (3) OUAC 101 students (direct applicants) may have a non-Ontario postal code if they attend an International Ontario approved High School. For schools located in Ontario, if student postal code was invalid, school postal code was used. (4) Program of registration is based on transcript data for column 1 and application data for columns 2 and 3 Table 4A: Demographic Characteristics of Sample Students: Comparison Across Terms Absence of Online Homework Presence of Online Homework Winter 2011 (1) 1432 64.73% Winter 2010 (2) 1417 59.14% Winter 2012 (3) 1,454 66.23% 76.40% 11.45% 12.15% 80.17% 10.87% 8.96% 76.82% 10.80% 12.38% Students by number of years in Canadian school system (at admission) 6 years or more 73.95% 77.77% 3-5 years 15.08% 12.35% <= 2 years 10.96% 9.88% 76.07% 12.93% 11.00% Total students per term Gender (% male) Immigrant Status Canadian Citizen Permanent Resident Other % with English reported as their primary language 64.53% 66.97% Students by OSAP application status Appied for OSAP (% applied) 60.68% 62.74% Students by neighbourhood of residence (if residing in Ontario at admission) Students with ON postal code 1319 1351 % living in low income neighbourhood (bottom tercile) 20.28% 20.95% % living in mid-income neighbourhood (middle tercile) 23.25% 23.59% % living in high income neighbourhood (top tercile) 56.46% 55.46% Median distance to the University, km 50.5 47.7 65.27% 61.62% 1,356 18.66% 21.50% 59.84% 46.9 Table 4B: High School Information of Sample Students: Comparison Across Terms Absence of Online Homework High School GPA (Best 6 University or Mixed Courses) Average Best 6 Standard Deviation of Best 6 Average Grade difference with Winter 2010 students Presence of Online Homework Winter 2011 Winter 2010 Winter 2012 (1) (2) (3) 86.1 (5.0) 0.1 (0.2) 86.0 (5.3) 85.7 85.3 86.4 (4.9) 0.4 (0.2) 0.2 (0.2) 85.8 1259 97.5% 88.9% 1263 97.8% 86.9% 2.48% 10.53% 57.78% 27.27% 1.94% 86.6 (8.4) 88 2.24% 5.19% 61.46% 30.80% 0.31% 87.6 (7.9) 89 Average Grade difference with Winter 2011 students Median Best 6 High School Math (average of all grade 12 University stream courses) # students with at least one Math course 1221 (% 101 students) 98.2% (% of all students) 85.3% % taking level 12 Math courses (101 students) 0 math courses 1.85% 1 math course 7.07% 2 math courses 62.22% 3 math courses 28.54% 0.32% 4 or 5 math courses 1 Average of Best Math grade 87.0 (standard deviation) (8.2) Median Best Math grade 89 1 Only 5 students in the entire sample took 5 grade 12 math courses, all of whom were in the economics course in 2010. Table 4C: University Information of Sample Students: Comparison Across Terms Absence of Online Homework Presence of Online Homework Winter 2011 Winter 2010 Winter 2012 (1) (2) (3) 27.65% 12.08% 27.72% 29.12% 2.72% 0.63% 23.78% 13.62% 24.84% 34.65% 2.40% 0.64% 28.27% 11.83% 31.64% 24.97% 2.89% 0.41% % students registered in Level 1 71.79% 68.38% Concurrent or Past Enrollment in First Year Microeconomics Course Number of students 1,215 1,159 Mean performance in microeconomics course (scale of 7.8 8.5 12) (standard deviation) (2.9) (2.6) 79.78% Students by Enrolled Faculty At Time of Course Business Social Sciences Engineering Science/Health Humanities (N/A) Performance in Course Online Math Test (standard deviation) 1,305 7.5 (3.0) 92.0 (14.5) 81.9 (31.3) 13.7 (3.2) 3 15 21 83.98% 13.9 (3.5) 3 15 21 83.22% Overall Term Performance What was the total # credits taken in the Term Average # of credits (standard deviation) Minimum credits Median credits Max credits Share of students with "full time loads" 13.8 (3.4) 3 15 22 83.80% Figure 1 80 Average Term Test and Exam Scores Based on Online Participation 78.8 76.7 74.2 40 Percentage on Test 50 60 70 71.8 67.8 55.2 Term Test One Term Test Two Winter 2010 Mandatory Online Homework Final Exam Winter 2011 No Online Homework Figure 2A Distribution of Midterm Grades by Term Fraction of Students .02 .03 .01 0 0 .01 Fraction of Students .02 .03 .04 Term Test Two .04 Term Test One 0 10 20 30 40 50 60 70 80 90 100 Distribution of Grades Winter 2010 0 10 20 30 40 50 60 70 80 90 100 Distribution of Grades Winter 2011 Figure 2B 0 Fraction of Students .01 .02 .03 .04 Distribution of Final Exam Grades by Term 0 10 20 30 40 50 60 Distribution of Grades Winter 2010 70 Winter 2011 80 90 100 Table 5A Winter 2010 (Mandatory Online Homework) and Winter 2011 (No Online Homework) Midterms and Final Exam Regressions VARIABLES Online Homework Mandatory Transcript Measures Business Engineering Science/ Health Humanities OUAC Measures Upper Year Student Distance from University Female Non-English First Language Average Best 6 High School Grade OSAP Applicant Years in Canadian School System Age at time of Course (year taking course - year of birth +1) Canadian Citizen OUAC 105 Students (1) (2) (3) Final Exam Grade, Percentage on Multiple Choice Test 1 Grade, Percentage on Multiple Choice Test 2 Grade, Percentage on Multiple Choice 2.106 (0.484)** 2.342 (0.475)** 12.613 (0.393)** 2.626 (0.891)** 0.934 (0.921) 4.011 (0.999)** -1.102 (1.793) 3.571 (0.947)** 1.622 (0.985) 4.158 (1.083)** -3.541 (1.729)* 1.670 (0.747)* 3.358 (0.758)** 5.643 (0.873)** -3.275 (1.340)* 5.810 (0.645)** 0.006 (0.003)* -1.659 (0.499)** -0.634 (0.628) 1.013 (0.056)** -0.186 (0.506) 0.120 (0.106) 5.694 (0.706)** 0.004 (0.003) -1.346 (0.529)* -0.679 (0.616) 1.048 (0.055)** -0.101 (0.519) 0.156 (0.084) 2.750 (0.593)** 0.006 (0.002)** -1.974 (0.424)** 0.232 (0.516) 1.130 (0.047)** -0.139 (0.409) 0.208 (0.076)** 0.211 (0.193) 0.132 (1.038) 0.724 (0.291)* 0.202 (0.201) -1.515 (0.970) 0.032 (0.303) 0.013 (0.186) -0.270 (0.839) -0.083 (0.261) Table 5B Winter 2010 Mandatory Online Homework and Winter 2011 No Online Homework Midterms and Final Exam Regressions Test 1 Grade, Percentage on Multiple Choice Test 2 Grade, Percentage on Multiple Choice Final Exam Grade, Percentage on Multiple Choice 0.012 (0.011) -0.004 (0.030) 0.012 (0.012) -0.002 (0.030) 0.010 (0.009) 0.026 (0.025) % Pop. With High School Diploma 0.067 (0.049) 0.058 (0.050) 0.070 (0.042) % Pop. With Trade Certificate 0.034 (0.077) -0.008 (0.083) 0.021 (0.064) -0.046 (0.049) -0.013 (0.054) -0.006 (0.043) 0.031 (0.083) 0.015 (0.033) 0.024 (0.085) 0.025 (0.034) 0.031 (0.071) 0.077 (0.027)** Bottom Third DA Income Tercile -1.415 (0.773) -1.602 (0.835) -1.479 (0.641)* Middle Third DA Income Tercile -1.370 (0.641)* -0.708 (0.644) -0.733 (0.516) % Unemployed age 15-24 -0.064 (0.072) 0.000 (0.078) 0.034 (0.064) -0.067 (0.113) 0.079 (0.059) -94.779 (28.670)** -0.089 (0.101) 0.024 (0.066) -32.011 (30.031) -0.008 (0.094) 0.086 (0.049) -45.009 (25.505) 2,577 0.304 2,540 0.289 2,602 0.516 VARIABLES Census Measures (DA Level) % Pop. Visible Minority % Pop. 1 Parent Family % Pop. With College Certificate or Diploma % Pop with University Certificate of Diploma % Pop with BA or Higher Total Population in Thousands % Population age15-24 Constant Observations R-squared Clustered standard errors in parentheses ** p<0.01, * p<0.05 Figure 3 80 Average Test Scores Based on Online Participation 78.8 74.2 74.2 Percentage on Test 60 70 74.4 67.8 40 50 56.4 Term Test 1 Term Test 2 Winter 2010 Mandatory Online Homework Final Exam Winter 2012 Optional Online Homework Figure 4A Distribution of Midterm Grades by Term Term Test One 0 0 .01 .01 Fraction of Students .02 .03 Fraction of Students .02 .03 .04 .04 Term Test Two 0 10 20 30 40 50 60 70 80 90 100 Distribution of Grades Winter 2010 0 10 20 30 40 50 60 70 80 90 100 Distribution of Grades Winter 2012 Figure 4B 0 Fraction of Students .01 .02 .03 .04 Distribution of Final Exam Grades by Term 0 10 20 30 40 50 60 Distribution of Grades Winter 2010 70 80 Winter 2012 90 100 Table 6A Winter 2010 Mandatory Online Homework and Winter 2012 Optional Online Homework Midterm Regressions VARIABLES Online Homework Optional Transcript Measures Business Engineering Science/ Health Humanities OUAC Measures Upper Year Student Distance from University Female Non-English First Language Average Best 6 High School Grade OSAP Applicant Years in Canadian School System Age at time of Course (year taking course - year of birth +1) Canadian Citizen OUAC 105 Students (1) (2) (3) Test 1 Grade Percentage on Multiple Choice Test 2 Grade Percentage on Multiple Choice Final Exam Grade, Percentage on Multiple Choice -4.079 (0.429)** 0.307 (0.473) -11.065 (0.394)** 3.190 (0.813)** 2.114 (0.809)** 5.626 (0.863)** 0.793 (1.467) 2.139 (0.943)* 1.578 (0.938) 4.662 (1.015)** -2.398 (1.909) 3.049 (0.758)** 2.517 (0.769)** 5.635 (0.861)** -2.400 (1.379) 4.108 (0.618)** 0.004 (0.003) -1.695 (0.468)** -1.109 (0.562)* 0.940 (0.047)** 0.207 (0.452) 0.040 (0.100) 3.140 (0.679)** 0.007 (0.003)** -0.756 (0.506) -2.041 (0.638)** 1.052 (0.053)** -0.633 (0.515) 0.029 (0.110) 4.044 (0.610)** 0.008 (0.002)** -1.951 (0.441)** -1.281 (0.528)* 1.146 (0.046)** -0.177 (0.412) 0.053 (0.094) -0.002 (0.215) 1.106 (0.963) 0.484 (0.306) 0.088 (0.231) -0.333 (1.151) -0.068 (0.352) 0.037 (0.209) 1.015 (0.934) -0.003 (0.297) Table 6B Winter 2010 Mandatory Online Homework and Winter 2012 Optional Online Homework Midterm Regressions Census Measures (DA Level) % Pop. Visible Minority 0.015 (0.010) 0.015 (0.029) 0.030 (0.011)** 0.018 (0.030) 0.012 (0.009) 0.042 (0.025) % Pop. With High School Diploma 0.029 (0.048) 0.091 (0.050) 0.044 (0.042) % Pop. With Trade Certificate 0.075 (0.070) 0.016 (0.075) -0.048 (0.061) -0.020 (0.046) -0.038 (0.051) -0.042 (0.040) -0.028 (0.075) 0.036 (0.031) -0.105 (0.083) 0.021 (0.034) -0.128 (0.070) 0.041 (0.027) Bottom Third DA Income Tercile -1.568 (0.802) -1.966 (0.845)* -1.462 (0.651)* Middle Third DA Income Tercile -0.643 (0.589) -0.065 (0.069) -0.706 (0.641) -0.075 (0.082) -0.084 (0.516) -0.078 (0.064) -0.115 (0.110) 0.109 (0.063) -59.196 (30.045)* -0.046 (0.121) 0.030 (0.067) -16.348 (34.456) -0.137 (0.099) 0.078 (0.049) -38.033 (28.905) 2,640 0.313 2,523 0.266 2,649 0.493 % Pop. 1 Parent Family % Pop. With College Certificate or Diploma % Pop with University Certificate of Diploma % Pop with BA or Higher % Unemployed age 15-24 Total Population in Thousands % Population age15-24 Constant Observations R-squared Clustered standard errors in parentheses ** p<0.01, * p<0.05 The Effectiveness of Tutorials in Large Classes: Do they matter? Is there a difference between traditional and collaborative learning tutorials? Karen Menard** Bridget O’Shaughnessy* A. Abigail Payne* With Olesya Kotlyachkov* and Bradley Minaker* June 2014 *Department of Economics and Public Economics Data Analysis Laboratory (PEDAL), McMaster University **Ontario Institute for Cancer Research I. Introduction Across much of Ontario and Canada, a typical first-year university student experience involves the enrollment in a large class. While there are many benefits for universities to use large classes from a financial and resource perspective, the impact on students, particularly weaker students, tends to be overlooked. Struggling students may not seek help and/or may disengage from their studies. Ultimately this may lead to bad decisions around the program of study, academic choices made in later years and, in many instances, may lead to dropping out. As illustrated in Dooley, Payne, and Robb (2011), high school grades are a strong indicator of success in university and first-year students are more likely to leave university than students who make it into their second year of studies (see also Finnie and Qiu, 2008). In this report we examine the use of tutorials in a large enrollment course and analyze the benefits of tutorials in such courses. It has been demonstrated that students who develop a connection with faculty tend to perform better than students who do not develop such a connection. Studies have shown that it is difficult to develop a student/faculty connection in classes of 400 to 600 students (see Nagda, 1998, Chickering and Gamson, 1987, and Finnie and Qiu, 2008). If students are engaged in their first year, this may have a positive impact on their university progression. However, given the reality of resource constraints on universities and the new practice of large classes, how can students be appropriately engaged in the context of a large class? Admittedly, poor performance in any one class is unlikely to cause a student to leave post-secondary education (PSE), and good performance in any one class is equally unlikely to entice a student to stay in PSE. It seems reasonable to believe, however, that helping a student improve performance in one class could mitigate other factors that might cause the student to leave university. If an approach were successful in improving at-risk student performance in one course, it could possibly be replicated in many first year courses and have an impact on persistence. If tutorials in large enrollment courses are successful at engaging students, does it matter if the tutorial is conducted as a traditional tutorial (i.e., with a teaching assistant working through a problem with the students) or as a collaborative learning tutorial (i.e., groups of students working through a problem together and the teaching assistant providing assistance to the students). Collaborative learning has been extensively studied in science and engineering programs. Felder (1995) and Felder, Felder, and Dietz (1998) found that collaborative learning and active learning improved student outcomes and student satisfaction in a sequence of large (90 to 123 students) chemical engineering courses. The instructional techniques used in the Felder studies, however, did not seem to help the weakest students in the class. A meta-analysis of 39 studies in science, mathematics, engineering, and technology courses showed a positive and statistically significant impact on student achievement, motivation, and attitudes when cooperative learning or collaborative learning methods were used (see, Springer, Stanne, and Donovan, 1999). Two studies have examined collaborative learning in the social sciences but for relatively smaller classes than what is being proposed. Yamarik (2007) studied class sizes of 25-35 students and focused on students in second/third year of university. He found collaborative learning classes to be more effective in leading toward student success than traditional classes. Huynh, Jacho-Chaves, and Self (2010 and 2011) studied classes of 200 students in first-year studies. They found benefits to collaborative learning but lacked an experimental design that allowed them to compare the benefits to other methods such as traditional tutorials. A key finding of Huynh, Jacho-Chaves and Self (2011) is that collaborative learning had a particularly strong positive impact on students falling in the bottom 40th percentile. In this report we analyze an intervention conducted during the 2012/2013 academic school year for a large class in economics. This course typically enrolls over 2,500 students each year across five sections. The students in this course represent all faculties across the campus as the course is a pre-requisite for many programs. The backgrounds of students and their level of preparation for the course vary. Previous attempts at low-cost interventions to improve student performance in this course were ineffective. For example, in 2009/2010, a random set of students who performed poorly on the first test were personally emailed by the instructor and provided information on academic resources. Final course performance by these students was no different than students who were not emailed but performed similarly on the first test. The innovation in our study is to examine the effectiveness of collaborative learning in larger classes (approximately 500-600 students per class) and to compare and contrast the effectiveness of collaborative learning versus traditional tutorials. In previous studies with smaller class sizes, most of the students participated in the tutorials. With very large classes, students may feel anonymous and attendance at tutorials may be lower than in a small class setting. As many departments face increasingly tight budgets, there is a trend towards eliminating tutorials, partly due to a perceived lack of student participation. Our study will begin to establish participation rates and identify which type of tutorial methods may be more beneficial in engaging students in large enrollment courses. Overall we find a high proportion of students participate in the first tutorial. We also find that close to 70% of the students attend at least 3 tutorials but that less than half the students attend all five tutorials. We find a measurable impact of tutorial attendance on course exams and on the final grade. Students who participated in all five tutorials performed better than those who only attended three tutorials. The traditional tutorials have a stronger (positive) effect on course performance. There is, however, a stronger positive correlation between the collaborative learning tutorials and performance on optional online homework assignments. The following section of the report outlines the course and experimental design. In section III we discuss the data, the selection of students studied and present summary statistics. Finally in sections IV and V we present the analysis and the discussion, respectively. II. Experiment Design Our experiment focused on instruction of introductory macroeconomics at an Ontario university. This course was taught by a single instructor and followed a similar structure for seven years. Each year, five sections of the course were offered: two in the fall (September to December) and three in the winter (January to April). The enrollment in each section ranged between 400 and 600 students, with a total enrollment of approximately 2,400 students each academic year. At the university under study, students taking introductory macroeconomics, however, are registered in all of the faculties as there are several programs across the various faculties for which this course is a requirement (e.g., engineering, commerce). Thus the students taking this course are diverse in their backgrounds, particularly in their academic preparation. This course, as is typical with economics courses, requires strong math and analytical skills. Prior to the year in which tutorials were introduced, students were evaluated based on their performance on two term tests and a final exam. The instructor offered optional online homework. A student who completes the online homework assignments can use her performance on the homework to reduce the weight allocated to her final exam. For a student who chooses not to complete the online homework assignments, the final grade was allocated as follows: 25% of the marks received on the better of two term tests (typically test #1) 20% of the marks received on the other term test 55% of the marks received on the final exam If a student completed the online homework assignment then the weight of the final exam was reduced to 40%. The 15% for the online homework was allocated as 5% of the marks received on a basic math test offered in the first two weeks of the course and 10% of the marks received on weekly homework assignments. The math component of the online homework is designed to review concepts learned in high school. To ensure tutorial attendance, the instructor offered an incentive. The tutorials were not for grades, but students were offered a “grade weight shift” as an incentive to attend. The grade weight shift allowed students to shift a small portion of the weight from the final exam to their higher term test grade. Students who performed better on the final exam than both tests would not be penalized because their shift would work in the opposite direction. Historically, the course average was higher on Test #1 than on Test #2, and both term test averages were typically higher than the average on the final exam. Usually only five percent of the class performed better on the final exam than both term tests. The percentage shift was as follows: Attend all five tutorials: Attend four tutorials: Attend three tutorials: Shift 5% of final exam weight to highest term test grade. This results in the better term test being worth 30%. Shift 4% of final exam weight to highest term test grade. This results in the better term test being worth 29%. Shift 3% of final exam weight to highest term test grade. This results in the better term test being worth 28%. Attend zero to two tutorials: No weight shift, grade is calculated according to scheme described above. Tutorials were provided for the students in Introductory Macroeconomics during the 2012/2013 academic year for the first time in two decades. Tutorials were held bi-weekly. There were five tutorials held each semester, beginning in Week 4 in the fall and Week 2 in the winter. The schedule for the tutorials and tests and the subjects covered (chapters of a text book) were as follows: Week 2 3 Fall, 2012 Date Tutorial 4 Sept. 17 Sept. 19 Sept. 24 5 6 Oct. 1 Oct. 9 7 8 Oct. 15 Oct. 22 9 10 Oct. 29 Nov. 5 11 12 Nov. 13 Nov. 20 13 14 Nov. 26 Dec. 3 Test Chapter 1-3 Chapter 4-6 Test 1 (1-6) Chapter 7-9 Chapter 10,11 3 Winter, 2013 Date Tutorial Jan. 17 Chapter Jan. 18 1-3 Jan. 23 4 Jan. 30 5 6 Feb. 6 Feb. 13 7 8 Feb. 18-22 Feb. 27 Week 2 9 Test 2 (7-11) 10 Mar. 6 Mar. 13 11 12 Mar. 27 13 14 Apr. 3 Apr. 10 Chapter 12,13 Final Exam Test Chapter 4-6 Test 1 (1-6) Chapter 7,8 Chapter 9-11 Test 2 (7-11) Chapter 12,13 Final Exam In the fall, most tutorials had an enrollment of approximately 70 students, with two exceptions. The tutorials that met from 8:00 pm to 8:50pm on Tuesdays and Wednesdays had 15 and 19 students, respectively. Tutorials were fundamentally different in the fall and winter semesters. In the fall, Teaching Assistants (TAs) stood in front of the tutorial section and delivered a “traditional” tutorial by solving a set of problems on a black board. Students were expected to bring questions and problems to the tutorial, having printed them from the course learning management system the week before and attempted them on their own. The TA showed the solutions to the students and then provided a corresponding multiple choice question and also went through the answer to the multiple choice question. In the winter semester, collaborative learning among students was the goal. Students were given the problem sheets at the beginning of the tutorial and told to work on the questions in small groups. Again, each chapter had two or three short answer problems with accompanying multiple choice questions. Students worked in groups sizes of 3-4 for approximately thirty minutes and then share their answers with other groups working nearby. TAs traveled throughout the room and assisted groups as needed. They were instructed to provide guidance, but not to directly tell the students the answers to the problems. III. Data Description, Sample Selection and Summary Statistics To study the effects of offering tutorials in large classes four data sources were utilized. The primary data are the records of the students from the course. The measures from this source capture information on (a) student tutorial participation, (b) performance on two course exams and one final exam, (c) participation in and performance on the online homework assignments, (d) adjusted final course grade (excludes performance on the online homework assignments), (e) an anonymous identification of the tutor assigned to each section. These measures were collected on students who participated in the course for both the 2011/2012 academic year (pretutorials, control) and the 2012/2013 academic year (tutorials, treatment). The second data source was obtained from registrar data held at the university. The core measures collected from the registrar were program of registration and enrollment and performance in all courses by the student. The third data source was information on the applications submitted by the students for admission to the university. The data on applications included applications for students applying directly from an Ontario high school (known as “101” students) and delayed entry and/or nonOntario high school students (known as “105” students). The 101 set of applications capture information on the students’ performance in level 4 (grade 12) courses in high school and their home postal code. From the 105 set of applicants the high school grade information was more limited. The location of their residence and home postal code (if from Canada), however, were available. Using the home postal code measures from the fourth data set, namely the socioeconomic characteristics of the neighbourhood where the student’s family resided, were added from the 2006 Canadian census. The census geography utilized was the dissemination area, a geography that covers roughly 500 households. The core sample studied in this report is those students enrolled in the course and for whom we observe both information from their university application and they received a final grade. We, therefore, include in our sample a total of 4,384 students. Some of these students, however, are observed repeating the course, leaving us with a total 4,342 unique students that are studied. The enrollment in the fall term is lower than the winter term given there was one fewer section of the course offered. Table 1: Sample comparison with the student population Sample students All(1) students at university under study All(2) Ontario students (1) (2) (3) 4,342 9,541 130,282 64.7% 47.0% 44.1% 78.3% 9.1% 12.6% 88.5% 7.0% 4.6% 90.9% 5.7% 3.4% 77.5% 12.7% 9.8% 88.6% 6.7% 4.7% 90.8% 4.8% 4.4% 65.9% 75.5% 76.9% 60.9% 67.9% 65.8% Students with ON postal code (3) 4,024 9,510 129,925 % living in low income neighbourhood High School GPA (Best 6 University or Mixed Courses) Average Best 6 GPA (standard deviation) 13.2% 14.5% 20.3% 86.5% (5.0) 86.6% (5.5) 83.7% (6.5) Total students in sample Delayed or non-Ontario students (% 105 students) Gender (% male) 11.8% Immigrant status Canadian Citizen Permanent Resident Other Years in Canadian K-12 school system 6 years or more 3-5 years 2 years or less % with English reported as their primary language Indication on application that an application for financial aid was submitted (% applied) (Note: Differences of group means between columns 1 and 2 are not statistically different from zero. The differences in means between the university under study and all Ontario students, however, is statistically different at the 1% confidence level) 5th percentile of best 6 Median best 6 95th percentile of best 6 Students by Program of Registration, Year 1 78.3% 86.2% 95.5% 77.2% 86.7% 95.7% 72.7% 83.8% 94.2% 28.8% 28.2% 11.8% 20.3% 14.1% 9.3% (4) Commerce (includes Social Sciences) Engineering Science/Health Humanities Other 22.6% 15.5% 5.0% 32.7% 30.2% 5.1% 28.9% 39.0% 8.7% NOTES: (1) All students who applied directly from HS in 2011 and 2012 and were registered at McMaster. There are a few students that are observed taking the course in more than one term. (2) All students who applied directly from HS in 2011 and 2012 and were registered at any University. (3) OUAC 101 students (direct applicants) may have a non-Ontario postal code if they attend an International Ontario approved High School. For schools located in Ontario, if student postal code was invalid, school postal code was used. (4) Program of registration is based on transcript data for column 1 and application data for columns 2 and 3 In Table 1 the characteristics of the students under study were compared with the characteristics of the 101 students that first enrolled in the university under study in 2011 or 2012 (column 2) and all Ontario direct-entry students observed registering at an Ontario university in 2011 or 2012. As a portion of the students in the course are registered students at a higher level (level 2 mostly), we compare the entering characteristics of these students based on the information provided in their application. As shown in Table 1, there are more males in the course than at the university and in the entire system. There are fewer students who are Canadian citizens, spent 6 or more years in the Ontario K-12 system, and who report English as their primary language in the course than at the university or in the province. A lower proportion of the students indicated that they applied for financial aid on their university application form and of those with an Ontario postal code, there are a lower proportion of the students whose family address is located in a low-income neighbourhood (bottom tercile of neighbourhoods). There are small but not statistically significant differences in entering high school averages with the students under study reporting slightly lower entering averages relative to all students at the university under study but with these students reporting higher averages than that reported for all registrants. Finally, the students in the study more heavily represent commerce and engineering than science and humanities. Thus, overall the students under study are more likely to be foreign born, male, from higher income families, and more interested in commerce and engineering. Table 2: Demographic Characteristics of Sample Students: Comparison Across Terms Fall Term (Traditional Tutorial) Control (2011) (1) 704 62.2% Treatment (2012) (3) 861 63.8% Total students per term Gender (% male) Immigrant Status Canadian Citizen 77.4% 78.8% Permanent Resident 7.7% 9.2% Other 14.9% 12.1% Students by number of years in Canadian school system (at admission) 6 years or more 76.3% 77.6% 3-5 years 15.2% 12.7% <= 2 years 8.5% 9.8% % with English reported as their 64.6% 65.4% primary language Students by OSAP application status Applied for OSAP (% applied) 56.0% 63.9% Students by neighbourhood of residence (if residing in Ontario at admission) Students with ON postal code 643 800 % living in low income neighbourhood 12.3% 14.8% (bottom tercile) % living in mid-income neighbourhood 32.8% 33.4% (middle tercile) % living in high income neighbourhood 54.9% 51.9% (top tercile) Median distance to the University, km High School GPA (Best 6 University or Mixed Courses) 47.6 47.0 Winter Term (Collaborative Learning) Control Treatment (2012) (2013) (2) (4) 1,454 1,365 66.2% 65.1% 76.8% 10.8% 12.4% 79.3% 8.1% 12.6% 76.1% 12.9% 11.0% 78.8% 11.7% 9.6% 65.3% 67.1% 61.6% 60.2% 1,356 1,259 14.6% 14.9% 31.1% 31.2% 54.4% 53.9% 46.9 48.7 Average Best 6 Standard Deviation of Best 6 Grade difference with Fall 2011 students 85.6 (4.6) 86.3 (4.7) 0.7 (0.3)*** 86.4 (4.9) Grade difference with Winter 2012 students 87.2 (5.3) 0.8 85.8 (0.2)*** 86.7 1263 97.8% 86.9% 1166 97.4% 85.4% 29 67 794 402 162 87.6 (7.9) 89 82.0 (31.3) 31 64 749 353 168 88.0 (8.4) 90 80.0 (33.2) 26.71% 14.98% 24.85% 28.69% 3.72% 1.05% 28.27% 11.83% 31.64% 24.97% 2.89% 0.41% 24.25% 15.31% 28.57% 27.69% 3.66% 0.51% Share students registered in Level 1 73.15% 68.64% Concurrent or Past Enrollment in First Year Microeconomics Course Number of students 228 327 Mean performance in microeconomics 7.9 8.0 course (scale of 12) (standard deviation) (3.1) (3.1) 79.78% 76.12% 1,305 1,159 7.5 8.4 (3.0) (3.1) Median Best 6 85.3 85.9 High School Math (average of all grade 12 University stream courses) # students with at least one Math course 598 729 (% 101 students) 97.1% 96.2% (% of all students) 84.9% 84.7% # level 12 Math courses taken 0 18 29 1 44 56 2 372 432 3 or 4 182 241 (no HS course information ) 88 103 Average of Best Math grade 86.1 86.9 (standard deviation) (8.7) (8.6) Median Best Math grade 87 89 Performance in Course Online Math Test 81.9 79.4 (standard deviation) (30.4) (33.0) Students by Enrolled Faculty At Time of Course Business Social Sciences Engineering Science/Health Humanities Not Declared or Not Applicable 38.64% 14.20% 23.15% 19.74% 3.84% 0.43% Overall Term Performance What was the total # credits taken in the Term Average # of credits (standard deviation) Minimum credits Median credits Max credits Share of students with "full time loads" 13.2 (3.4) 3 12 21 79.12% 13.6 (3.3) 3 15 21 82.58% 13.9 (3.5) 3 15 21 83.22% 13.8 (3.6) 3 15 21 81.10% Do our students vary across terms? Table 2 shows the characteristics of our study students grouped by term of enrollment in the course. Across the terms there are marginal differences in most of the measures. The core differences are the proportion of students who are male (more in the winter terms), and the proportion residing in low-income neighbourhoods (smaller proportion in fall 2012). In terms of preparation, the students in the winter term have slightly higher high school GPAs and higher averages for the best level 4 math course in high school. Presumably these differences are driven by the fact that a greater share of students are enrolled in the engineering faculty in the winter term and engineering students typically enter the university with better grades due to high admission standards and are also more math oriented, an important factor in the course under study. IV. Analysis a. Tutorial Attendance The first set of analyses focuses on participation in tutorials. The structure of the introduction of tutorials was to use a quasi-experimental design given the students choose to take the course and the requirement that we offer tutorials to all students in a given term. Offering tutorials and participating in tutorials, however, are not identical. Given a core justification for introducing tutorials is to engage students, especially those likely to face difficulties in the first year of university studies, a first analysis focuses on participation in the tutorials and developing a better understanding of who participates as well as the extent to which a student participates in the tutorials throughout the term. Figure 1: Tutorial Attendance Figure 1 depicts the attendance of tutorials. In both terms, a high share of students attended at least 1 tutorial with most students (~70%) attending at least 3 tutorials. In each term, we observe a decrease in attendance, approximately a 5 percentage point drop from attending 1to 2 tutorials and from attending 2 to 3 tutorials and an approximately 10 percentage point drop from attending 3 to 4 tutorials. The biggest drop off in tutorial participation is from 4 to 5 tutorials with there being a 17-19 percentage point drop in participation. A key component to tutorial attendance is participation in the first and/or second tutorial. If a student failed to attend the first or second tutorial, the student was not observed attending the remaining tutorials. In the fall term, there were 168 students that did not attend at least one of the first two tutorials. Less than 10 percent of these students (14) were observed attending a future tutorial. Similarly, in the winter term, there were 299 students that failed to attend at least one of the first two tutorials. Less than 4 percent of these students (9) were observed attending a subsequent tutorial. To what extent is tutorial attendance correlated with being more academically prepared? While there is no perfect measure of academic preparation, we can look at the relationship between performance in high school, as measured by the best level 4 (grade 12) math grade and tutorial attendance. As mentioned in the introduction, for the course under study, math skills are required for successful completion of the course. In Figure 2 we group tutorial participation by the number of tutorials a student attends and depict these students based on whether their best level 4 math grade is above or below the median math grade for the students in the course. Across the board, it appears that approximately half of the students fall above and half of the students fall below the median. So there is not prima facie evidence that only the better academically prepared (as measured by math grades) students are attending the tutorials. Figure 2: Tutorial Attendance and Grade 12 Math Performance Are there other observed characteristics that are correlated with tutorial attendance? In Table 3, we report the results from a linear probability regression that uses as a dependent variable a zero/one indicator variable to indicate whether the student attends at least one tutorial as well as the number of tutorials attended. In columns 1 to 4 we report the results for the fall term and in columns 5 to 8 we report the results for the winter term. In the odd columns we use as a measure of preparedness for the course the score on the best high school level 4 math courses and the square of its term. In the even columns we use as a measure of preparedness performance on the online math test offered in the first two weeks of the course and the square of its term. For students with either no high school math mark or no in course online math mark we assign a value equal to the mean of the respective mark and include indicator variables equal to 1 if the student does not have the respective mark. The omitted category is not having taken the test. Table 3: Likelihood of Attending a Tutorial Dependent Variable: Attend At Least 1 Tutorial Best High School Math Mark Best High School Math Mark Squared (/100) Below Median on Best HS Math Mark No HS Math No Transcript In Course Online MathMark In Course Online MathMark Squared (/ 100) Below Median on Online Math Mark No Online Math Mark Level 1 Student Full Time Student Student = Female Enrolled in Business Faculty Enrolled in Engineering Faculty Enrolled in Science or Health Science Fall 2012 (1) (2) (3) (4) Attended At Least 1 # of Tutorials Tutoria1 Attended -0.03** -0.14** (0.01) (0.06) 0.02** 0.09** (0.01) (0.04) -0.01 -0.07 (0.05) (0.25) 1.10* 6.35** (0.58) (2.89) 1.18** 6.42** (0.58) (2.93) 0.48 3.62 (0.77) (3.27) -6.16 -121.31 (59.68) (270.13) -0.03 -0.25 (0.04) (0.22) -0.21*** 0.93*** (0.05) (0.28) 0.16*** 0.15*** 0.72*** 0.67*** (0.03) (0.03) (0.17) (0.16) 0.02 -0.01 0.46*** 0.35** (0.03) (0.03) (0.16) (0.16) 0.01 0.01 0.41*** 0.41*** (0.03) (0.02) (0.14) (0.13) 0.03 0.01 0.44** 0.33* (0.04) (0.03) (0.21) (0.19) 0.18*** -0.17*** 0.68*** 0.66*** (0.06) (0.05) (0.24) (0.22) 0.03 0.01 0.18 0.05 (0.05) (0.04) (0.23) (0.21) Winter 2013 (5) (6) (7) (8) Attended At Least # of Tutorials 1 Tutoria1 Attended -0.01 -0.17* (0.02) (0.09) 0.01 0.11** (0.01) (0.05) -0.00 0.02 (0.03) (0.19) 0.38 6.57* (0.80) (3.95) 0.66 7.81** (0.79) (3.90) -0.01 -1.04 (0.57) (2.54) 8.48 154.68 (45.51) (210.70) -0.03 -0.13 (0.03) (0.17) 0.23*** -1.37*** (0.04) (0.17) 0.18*** 0.16*** 0.93*** 0.82*** (0.03) (0.03) (0.13) (0.13) 0.10*** 0.08*** 0.38*** 0.25* (0.03) (0.02) (0.14) (0.14) 0.06*** 0.06*** 0.37*** 0.36*** (0.02) (0.02) (0.09) (0.09) -0.01 0.01 0.12 0.13 (0.03) (0.03) (0.18) (0.15) -0.04 (0.04) 0.06 (0.04) -0.02 (0.03) 0.06* (0.03) -0.04 (0.17) 0.39** (0.19) 0.04 (0.16) 0.37** (0.15) Enrolled in Humanities Canadian Citizen Language Spoken at Home is not English Indicated Applied for Financial Aid Age <= 17 Age = 19 Age = 20+ Neighourhood Share of Unemployment for individuals 25+ Neighbourhood Income in Bottom Tercile Neighbourhood Income in Middle Tercile Total Population in Neighbourhood (/1000) Share of the Population Aged 15-24 Share of Families = 1 Parent Share of Population with High School Degree Share of Population with University or Higher Degree Share of Population = Visible Minority Family Residence in Neighbourhood not in Canada Constant 0.00 (0.07) -0.01 (0.04) -0.02 (0.03) 0.06** (0.03) 0.00 (0.14) 0.01 (0.03) 0.04 (0.05) 1.73 (1.46) -0.03 (0.06) -0.02 (0.03) 0.15 (0.67) -0.61 (0.78) -1.11 (1.77) 0.91 (0.93) 0.36 (0.26) -0.07 (0.08) 0.08 (0.06) 2.43 (1.87) 0.02 (0.06) -0.03 (0.04) -0.02 (0.03) 0.05* (0.03) 0.03 (0.13) 0.02 (0.03) 0.04 (0.04) 2.00 (1.35) -0.06 (0.06) -0.03 (0.03) -0.28 (0.65) -0.18 (0.72) -0.85 (1.72) 0.91 (0.83) 0.31 (0.22) -0.07 (0.07) 0.04 (0.03) 0.88 (1.76) 0.08 (0.35) -0.27 (0.19) -0.18 (0.15) 0.42*** (0.13) 0.22 (0.45) 0.05 (0.15) 0.09 (0.25) 11.19 (7.59) -0.37 (0.29) -0.27 (0.17) 1.44 (3.69) -4.40 (4.49) 5.56 (9.25) 5.77 (4.76) 1.32 (1.31) -0.07 (0.42) 0.40 (0.32) 0.55 (9.98) 0.14 (0.36) -0.41** (0.20) -0.12 (0.14) 0.37*** (0.13) 0.32 (0.47) 0.07 (0.14) -0.00 (0.24) 13.09* (6.94) -0.52* (0.28) -0.34** (0.16) -0.75 (3.56) -2.35 (4.23) 6.93 (9.03) 5.70 (4.28) 1.03 (1.15) -0.07 (0.38) -0.08 (0.16) -8.35 (9.53) -0.10 (0.08) -0.05 (0.03) 0.04 (0.03) 0.01 (0.02) -0.00 (0.06) 0.01 (0.03) -0.05 (0.05) -0.64 (1.27) -0.05 (0.04) -0.02 (0.03) -0.07 (0.45) -0.32 (0.55) 0.77 (1.37) 0.02 (0.73) -0.05 (0.23) 0.06 (0.07) 0.01 (0.05) 0.35 (1.64) -0.14* (0.08) -0.06* (0.03) 0.02 (0.03) 0.02 (0.02) -0.01 (0.06) 0.02 (0.03) -0.02 (0.04) -0.40 (1.24) -0.06 (0.04) -0.02 (0.03) -0.06 (0.46) -0.49 (0.52) 0.97 (1.41) 0.25 (0.70) -0.02 (0.21) 0.06 (0.06) 0.01 (0.03) -0.37 (1.44) -0.20 (0.34) -0.12 (0.17) 0.16 (0.16) 0.02 (0.11) 0.34 (0.31) 0.12 (0.14) -0.34 (0.22) -2.96 (6.18) -0.22 (0.21) 0.05 (0.15) 3.47 (2.51) -0.63 (2.62) 9.82 (6.13) 1.35 (3.51) -0.52 (1.07) 0.39 (0.34) 0.18 (0.24) -1.91 (7.28) -0.32 (0.32) -0.18 (0.16) 0.09 (0.14) 0.07 (0.10) 0.33 (0.31) 0.21 (0.13) -0.23 (0.18) -1.69 (5.89) -0.29 (0.21) 0.05 (0.14) 3.36 (2.54) -1.49 (2.50) 11.77* (6.18) 2.42 (3.31) -0.43 (0.99) 0.39 (0.32) 0.07 (0.13) -10.23 (6.34) Observations R-squared Robust standard errors clustered by neighbourhood in parentheses *** p<0.01, ** p<0.05, * p<0.1 852 0.23 852 0.27 852 0.23 852 0.27 1,358 0.34 1,358 0.38 1,358 0.29 1,358 0.34 Focusing first on tutorial attendance in the fall term, the results suggest a negative correlation between the best high school mark and tutorial attendance but a positive correlation between the square term of the best high school mark. Across most marks, however, this suggests only a slight decline in attendance based on math marks. For most of the measures, the coefficient is not significantly different from zero. There is a higher likelihood of tutorial attendance by level 1 students, female students, full time students, and students that indicated they were applying for financial aid. Compared to students in Social Science, students in business are more likely to attend tutorials, while those in engineering are less likely to attend. Moving next to the results for the winter term, we observe similar results but with different magnitudes. For example, a level 1 student in the fall is more likely to attend an additional 0.7 tutorials whereas a level 1 student in the winter is more likely to attend an additional .9 tutorials. In addition, the coefficient on the measure for applying for financial aid is not significantly different from zero in the winter term. A student’s faculty plays a different role in the winter term. Business and engineering students are no more or less likely than social sciences students to attend, while science students are more likely and humanities students less likely to attend. Overall, while there are some differences based on observed characteristics in the likelihood of attending a tutorial, the evidence does not overwhelmingly support the notion that less prepared students do not attend the tutorials. This, in part, is likely attributable to the relatively high participation rates in the tutorials. b. Test Performance Does offering tutorials matter? We turn next to observed performance on the two term tests. In Table 4 we report the average performance on the two term tests across the four terms. As we demonstrated above, there are differences in the composition of students across terms. We, therefore, compare the students in tutorials based on tutorial type (traditional v. collaborative) and enrollment term. Beginning first with the fall term (columns 1 and 2), average performance on term test 1 is similar across the two years. Average student performance on term test 2 is higher in the term with tutorials but average student performance on the final exam is lower. Stronger performance on test 2 is counter to what has been observed historically in this course (that performance is lower on test 2), raising some suggestion that possibly the tutorials had an effect on understanding the concepts and, thus, test 2 performance. The decline in the final exam is somewhat puzzling. Recall, however, that the weight attributed to a student’s final exam is shifted to performance on the best test if the student attended tutorials. Thus, this decline in performance on the final exam might reflect a strategic decision by the students if students seek minimum grades versus maximum knowledge. Without more information, however, we cannot explore such a hypothesis. Turning next to the winter terms, average performance on the term tests and final exams is lower in the term with the collaborative tutorials. This is perplexing. Recall, however, that there are lower shares of students enrolled in engineering and business programs in Winter 2013 versus Winter 2012. Also, there is a lower share of students that took 2 or more grade 12 math in the 2013 versus 2012 term. Table 4: Average Student Performance on Term Tests and Final Exam Fall Term (Treatment=Traditional Tutorial) Total students in sample Control (2011) Treatment (2012) 704 861 71.8 (14.1) 63.9 (16.5) 68.7 (15.8) 71.7 (15.2) 68.0 (20.8) 62.8 (14.2) Fall v. Fall Treatment - Control # of students repeating the course Average Test 1 (0-100) (standard deviation) Average Test 2 (0-100) (standard deviation) Average Final Exam (0-100) (standard deviation) -0.1 4.1 -5.9 Winter Term (Treatment=Collaborative Learning) Control (2012) Treatmen t (2013) 1454 1365 10 8 76.4 (13.6) 72.4 (16.5) 65.0 (15.0) 72.8 (18.1) 65.4 (19.5) 61.4 (14.8) Winter v. Winter Treatment Control -3.6 -7.0 -3.6 In Table 5, we report the regression results using performance on the terms tests as the dependent variable. In columns 1 and 2, performance on term test 1 is the dependent variable. Tutorials 1 and 2 included material covered on this test, so we explore the effects of attendance in either or both of these tutorials on test performance, after controlling for background characteristics of the students and the neighbourhoods in which their parents reside. The results in column 1 reflect overall tutorial participation during the 2012/13 school year. The results in column 2 allows for a differential effect of the collaborative learning tutorial and also explores whether there is a differential effect for students that are observed with high school marks falling below the median of the students enrolled in the course. Overall, there is a positive effect of tutorial participation on test performance. Participation in one or both tutorials increases performance on the test an average of 2 to 2.5 percentage points. There is, however, no discernable difference of the collaborative learning form of tutorial on performance. The coefficient on the interaction term for the tutorial participation and having a below median math mark is negative but imprecisely measured. There is also a strongly positive effect of tutorial participation on term test 2 (columns 3 and 4). For this test, the material covered in tutorials 3 and 4 were most relevant. Given we observe a difference in participation in these two tutorials, we include separate measures for participating in the two tutorials. Overall, participating in tutorial 3 increased performance on the test by 2.2 percentage points, overall, and 3.7 percentage points for the traditional tutorials. Participating in both tutorials, increased performance an average of between 7 and 9.2 percentage points. Although not always precisely measured, there is some evidence that the collaborative learning tutorials did not increase performance to the same degree that the traditional tutorials did. Table 5: Effects of Tutorials on Term Tests Dependent Variable Attended Tutorial #1 and/or Tutorial #2 * Collaborative Learning * Below Median High School Average (Best 6 Grade 12) Attended Tutorial #3 * Collaborative Learning * Below Median High School Average (Best 6 Grade 12) Attended Tutorial #4 * Collaborative Learning * Below Median High School Average (Best 6 Grade 12) Best High School Average (Best 6 Grade 12) Below Median High School Average (Best 6 Grade 12) No High School Average Level 1 Student Full Time Student Student = Female Enrolled in Business Faculty Enrolled in Engineering Faculty Enrolled in Science or Health Science Enrolled in Humanities Additional Controls Test 1 (1) 2.25*** (0.53) Test 2 (2) 2.43*** (0.54) 0.3 (0.36) -0.69 (0.45) (3) 2.13* (1.13) (4) 3.67** (1.77) -2.24 (1.99) -0.39 (1.85) 4.93*** 5.60*** (1.06) (1.62) -3.54* (1.94) 2.75 (1.86) 1.14*** 1.14*** 1.17*** 1.22*** (0.09) (0.09) (0.09) (0.08) -0.07 0.41 0.58 0.32 (0.74) (0.87) (0.67) (0.81) 2.88** 3.01*** 3.66** 3.52** (1.13) (1.13) (1.42) (1.44) -4.23*** -4.26*** -3.74*** -3.54*** (0.54) (0.55) (0.89) (0.88) 0.69 0.75 1.48** 1.39* (0.52) (0.52) (0.73) (0.75) -1.59*** -1.60*** -0.36 -0.36 (0.50) (0.50) (0.59) (0.58) 4.17*** 4.14*** 2.04* 1.83 (0.80) (0.80) (1.16) (1.19) 4.18*** 4.12*** 1.71 1.73 (0.85) (0.85) (1.13) (1.14) 7.06*** 7.00*** 5.83*** 5.73*** (1.05) (1.05) (1.69) (1.69) 0.53 0.52 -2.11 -2.13 (1.51) (1.51) (1.84) (1.84) Neighbourhood Neighbourhood Neighbourhood Neighbourhood SocioSocioSocioSocioeconomic economic economic economic measures Observations 4,359 R-squared 0.27 Robust standard errors clustered by neighbourhood reported in parentheses *** p<0.01, ** p<0.05, * p<0.1 measures measures measures 4,359 0.27 4,359 0.20 4,359 0.20 c. Final Exam and Overall Course Performance Do the tutorials affect the final exam and performance in the course? In table 6 we report the results using the final exam marks and the overall course grade as dependent variables. For all specifications, we include indictor variables for participating in at least 1 tutorial, the total number of tutorials attended and interaction terms between the number of tutorials attended and the collaborative tutorial and being below the median on high school marks. In column 1 we use as a dependent variable the final exam marks. In columns 2 and 3 we use as a dependent variable the overall grade in the course. In column 2 the overall grade includes marks attributed for undertaking the online homework and participating in tutorials. In column 3, the overall grade excludes marks assigned for these two items, thus only accounting for performance on the two term tests and the final exam. Overall, attending a single tutorial has no, or a relatively small negative, effect on final exam and overall course performance. There is an increasingly positive effect of attending several tutorials on these measures of performance. The effect of the collaborative learning tutorials, however, is smaller than the traditional tutorials. For example a student who attended all five traditional tutorials, improved her final exam mark by 5.7 percentage points and her overall grade (adjusted) by 6.5 percentage points. If the same student attended all five collaborative tutorials, her final exam mark improved by 2 percentage points and her overall grade increased by 3.4 percentage points. There is no discernable effect of the tutorials on the performance of students that entered the course at the lower end of the distribution based on high school marks. Any grade difference of more than three percentage points is meaningful to the students in the sample. The university under study uses a twelve point grading system to calculate cumulative averages, where an “F” equals zero, “D-” equals one, “D” equals two, up to “A+” equaling twelve. For grades in the “D”, “C”, and “B” ranges, an increase in grade of three percentage points is enough to move a student up to the next grade level. For example, a student with a grade of 67% has a C+, which translates to six out of twelve, while a student with 70% has a B-, which is seven out of twelve. Table 6: Effect of Tutorials on Final Exam and Overall Course Grade Dependent Variable Ever Attend A Tutorial Number of Tutorials Attended * Collaborative Learning * Below Median High School Average (Best 6 of Grade 12) Best High School Average (Best 6 Grade 12) Below Median High School Average (Best 6 Grade 12) No High School Average Level 1 Student Full Time Student Final Exam Overall Grade: Unadjusted Overall Grade: Adjusted (1) -1.64 (1.13) 1.47*** (0.26) -0.75*** (0.16) -0.27 (0.17) 1.24*** (0.06) 0.33 (0.57) 3.91*** (1.33) -4.94*** (0.62) -0.00 (2) -0.68 (0.99) 1.61*** (0.22) -0.48*** (0.13) -0.03 (0.14) 1.07*** (0.05) 0.22 (0.50) 3.30*** (0.99) -3.93*** (0.49) 0.60 (3) -1.59 (1.11) 1.62*** (0.25) -0.65*** (0.13) -0.12 (0.15) 1.20*** (0.06) 0.35 (0.52) 3.64*** (1.08) -4.50*** (0.54) 0.44 (0.52) -1.42*** (0.38) Enrolled in Business Faculty 3.17*** (0.79) Enrolled in Engineering Faculty 2.45*** (0.84) Enrolled in Science or Health Science 7.17*** (1.23) Enrolled in Humanities -2.60 (1.60) Neighbourhood SocioAdditional Controls economic measures Observations 4,359 R-squared 0.34 Robust standard errors clustered by neighbourhood reported in parentheses *** p<0.01, ** p<0.05, * p<0.1 Student = Female (0.43) -1.09*** (0.33) 3.70*** (0.62) 2.80*** (0.62) 7.15*** (0.94) -1.51 (1.32) Neighbourhood Socioeconomic measures 4,359 0.39 (0.45) -1.27*** (0.36) 3.06*** (0.77) 2.68*** (0.75) 6.83*** (1.16) -1.81 (1.41) Neighbourhood Socioeconomic measures 4,359 0.38 V. Discussion Participation rates in the optional tutorials were quite high, with almost seventy percent attending at least three out of the five tutorials offered. Students who did not attend one of the first two tutorials were unlikely to attend any of the remaining tutorial sessions. First year students, females, and students applying for financial aid are more likely to attend tutorials. To the extent that these are high risk groups for withdrawing from a course, or the university, these findings support the idea of offering tutorials. Countering that, we find that students from high income neighbourhoods are also more likely to attend tutorials. Term test performance is, on average, improved by tutorial participation. Attending a single tutorial does not improve final exam or overall course performance, but attending more than one tutorial has a cumulative effect on exam and overall performance. A student who attends all five tutorials will likely improve their course grade by two full points on a twelve-point grade scale. Traditional tutorials appear to help students more than collaborative learning tutorials, which was not the expected. It may be that the tutorials were too large for the collaborative learning method to be effective. Each teaching assistant had almost seventy students per tutorial section; a size that may be better suited for the more traditional tutorial style. It may also be possible that having students in the traditional tutorial attempt the problem set ahead of the tutorial positively influenced performance in the course. With tight budgets and large section sizes, the traditional tutorial may be an effective means of using limited resources toward student academic success. We are also disappointed that students at the lower end of the distribution were not helped more by adding tutorials to the course. The research so far is mixed on this issue with some studies finding that tutorials have a greater impact on weaker student performance, while others do not (Felder, Huynh, Jacko-Chaves, and Self, 2010b). Overall, we found that student participation in tutorials helped performance in the course, and believe that these gains in learning are worth the cost of offering tutorials. References Chemers, M. M., Hu, L., Garcia, B. F. “Academic self-efficacy and first year college student performance and adjustment.” Journal of Educational Psychology, 93(1), 55-64. Chickering, A., Gamson, Z. (1987) Seven principles for good practice in undergraduate education, American Association of Higher Education Bulletin, 39 (7) 3-7. Dooley, Martin, A. Abigail Payne and A. Leslie Robb. 2011. “Understanding the Determinants of Persistence and Academic Success in University: An Exploration of Data from Four Ontario Universities.” Toronto, ON: Higher Education Quality Council of Ontario. Felder, R.M. (1995) A longitudinal study of engineering performance and retention. IV. Instructional methods. Journal of Engineering Education, 84(4) 361-367. Felder, R.M., Felder, G.N., and Dietz, E. J. (1995) A longitudinal study of engineering performance and retention. V. Comparisons with traditionally taught students. Journal of Engineering Education, 87(4) 469-480. Finnie, Ross and Theresa Qiu. 2008. “Is The Glass (or Classroom) Half-Empty or Nearly Full? New Evidence on Persistence in Post-Secondary Education in Canada.” In Who Goes? Who Stays? What Matters? Accessing and Persisting in Post-Secondary Education in Canada eds. R. Finnie, R.E. Mueller, A. Sweetman and A. Usher. Montreal and Kingston: McGill-Queens University Press. Finnie, R., Qiu, H. (2008) The patterns of persistence in post-secondary education in Canada: evidence from the YITS-B dataset. A MESA Project Research Paper. Toronto, ON: Educational Policy Institute. Huynh, K. P., Dt. T. Jacho-Chavez, and J. K. Self. 2010a. The efficacy of Collaborative Learning Recitation Sessions on Student Outcomes. American Economic Review, 100(2), 287-91. Huynh, K. P., Dt. T. Jacho-Chavez, and J. K. Self. 2010b. The distributional efficacy of Collaborative Learning Recitation Sessions on Student Outcomes. JSM Proceedings, Alexandria, VA: American Statistical Association, 4406-4418. Nagda, B. A., Gregerman, S. R., Jonides, J., von Hippel, W., Lerner, J. S. (1998) Undergraduate student-faculty research partnerships affect student retention, The Review of Higher Education 22 (1) 55-72. Springer, L., M. E. Stanne, and S. S. Donavan. 1999. Undergraduates in Science, Mathematics, Engineering, and Technology: A Meta-analysis. Review of Educational Research, 69(1): 21-51. Yamarik, Steven. 2007. Does Cooperative Learning Improve Student Learning Outcomes? Journal of Economic Education, 38(3): 259-77.