Int'l Conf. Frontiers in Education: CS and CE | FECS'15 | 215 Flipping a Data Structures and Discrete Mathematics Class Anja Remshagen Computer Science Department, University of West Georgia, Carrollton, Georgia, US Abstract - This paper investigates the impact of the flipped classroom in a data structures and discrete mathematics course. The retention rates and students’ grades when the flipped classroom model was applied in the course are compared with the retention rates and grades in the traditional lecture model. A two-week study was performed to compare the students’ performance in the two different pedagogical models. Results are mixed and don’t show a clear performance improvement in favor of the flipped classroom. However, grades and retention rates were significantly higher in the flipped classroom, indicating that the flipped classroom might help raising the confidence level of students. Keywords: computer science education, flipped classroom, traditional lecture, data structures, discrete mathematics 1 Introduction Data Structures and Discrete Mathematics I (CS3151) is a required course to complete the Bachelor of Science in Computer Science at the University of West Georgia. The University of West Georgia is a regional, public institution. Most undergraduate students are traditional students that have started college right after finishing high school in the West Georgia region. Many of these students are first-generation college students. The racial demographics of the students is similar to the racial demographics of Georgia. For example, in the academic year 2013/2014, 33.5% of the students were Black/African Americans and 56.3% were Caucasion/White. In the course CS3151, Data Structures and Discrete Mathematics I, computer science majors learn about elementary data structures and algorithms and about fundamental concepts in discrete mathematics. They implement and apply the data structures and related algorithms in small-scale Java programs. The course is typically taken by computer science majors after they have completed the introductory computer science sequence CS1 and CS2, which introduces students to the object-oriented programming language Java. As one of the first courses of theoretical and abstract nature, many students do not feel prepared for the rigor of CS3151. This is reflected by a high withdrawal rate. For example, in Fall 2012, out of 44 students 11 students withdrew from the class. In Spring 2013, out of 36 students 6 students withdrew. Some of the withdrawing students change their major; others attempt the class a second time. To improve the retention rate in CS3151, the flipped classroom had been applied in Spring 2014. The flipped classroom is a pedagogical model where students learn new content on their own time outside the classroom and apply the material in problems and hands-on-activities during class sessions [5], [6]. The flipped classroom has been explored in several computer science disciplines, including introductory computer science courses [1], [7], computer architecture [4] and software engineering [3], to name only a few. Most studies demonstrate more engagement and enthusiasm among the students. Some studies show that there was no performance improvement [1], [4], or they had not yet sufficient data to study the students’ performance in a flipped computer science course [3], [7]. This paper compares the students’ performance in the flipped classroom with the performance in the traditional lecture model during a twoweek study. It also explores retention rates and changes in grades among the recent course offerings. 2 The Flipped Classroom in CS3151 Before the introduction of the flipped classroom in CS3151, the class followed the traditional lecture model. Two issues that seem to impede students’ success came up repeatedly when the traditional lecture model had been applied: (1) Students asked for more practice problems. (2) Some of the weaker students needed significant help to get started on the programming exercise. In Spring 2014, the flipped classroom was implemented to help students with the two issues addressed above and, in turn, to improve retention rates. As the current generation of students is used to consume visual media, video recordings for the course material were produced in the form of narrated slides, instead of assigning readings from a textbook. The students were required to watch the videos before class. In class, students worked through exercises individually or in teams of two. After working through one or two exercises, typically a student presented his or her solution. Sometimes the instructor worked out the solution with the students on the whiteboard. As in preceding semesters, every week homework was assigned in Spring 2014. Most of the homework exercises 216 Int'l Conf. Frontiers in Education: CS and CE | FECS'15 | were very similar to the exercises worked out in class and contained a mix of exercises at different levels of Bloom’s taxonomy. In addition to written exercises, in many classroom meetings the students worked on a programming exercise that required the implementation or the application of a data structure and its associated algorithms. During the class meetings, students could program in teams of two. The programming exercises typically could not be finished in class due to time-constraints. Thus the completion of the programming exercises was assigned as part of the homework. Students had to complete all homework individually. At the end of the time reserved in class for the hands-on exercise, ideas of how to complete the programming task were discussed. This gave the students with weak programming skills a good start on programming exercises. 3 Table 1. Schedule of interventions. Wednesday (Mar 24) Pretest taken by Group L and F Assignment of videos only accessible to Group F Monday (Mar 31) Lecture for Group L Assignment of homework to Group L Wednesday (Apr 2) Flipped classroom meeting for Group F Assignment of homework to Group F Monday (Apr 7) Posttest taken by Group L and F Homework due for Group F and Group L Study and Data Collection In order to determine whether students had a better understanding of the material, the concepts, and of the algorithms by using the flipped classroom, the class was split into two groups for one week. Group L was exposed to the traditional lecture model during that week, and Group F to the flipped classroom. Both groups were taught exactly the same material. Group L met on a Monday for a traditional lecture that introduced the abstract data type priority queue, the heap data structure, its associated operations, and heap sort. The same material that was covered in the lecture was recorded through narrated slides. The recordings were assigned to Group F to watch before the class meeting on Wednesday. The recordings were not available to Group L. Group F worked through exercises in their class meeting. A pre- and posttest assessed the progress of each group. Table 1 displays the schedule for this study. Different problems were tackled by the students in examples during the lecture, in exercises for the flipped classroom, in a homework assignment, and in the pre- and posttest. We can split all these problems into four categories outlined in Table 2. The first category (KNOW) required students to understand concepts and recall related facts. These problems are located at the first and second level of Bloom’s Taxonomy. The second category (PRP) required students to interpret the representation of heaps and heap properties. These problems are at the second and third level of Bloom’s Taxonomy. The third category (ALG) demanded students to perform and trace the algorithms that are executed by the operations on priority queues and heaps. These problems are at the third level of Bloom’s Taxonomy. The last category Table 2. Problem categories. Category KNOW PRP ALG OTHER Description General knowledge about heaps and priority queues Properties of heaps and their representation Operations on heaps and priority queues Includes problems at the Bloom’s synthesis level Examples in lecture for Group L x x Exercises in class for Group F x x x x x Homework for Group L and F Pre/Posttest for Group L and F x x x Int'l Conf. Frontiers in Education: CS and CE | FECS'15 | 217 (OTHER) included problems at higher Bloom levels. These problems were not suited for a quiz-like test that can be completed in a few minutes. Thus only the first three categories were assessed in the pre-and posttest. Table 2 displays the occasions at which the students had been exposed to problems of the corresponding categories. KNOW Table 3. Problem categories. ALG PRP Students in Group L were occasionally asked problemsolving questions during the lecture, but participation was voluntary. The recorded slides for Group F did not contain all the examples that were worked out on the whiteboard as part of the lecture. Instead, Group F worked through related exercises on Wednesday. In addition to problems of category ALG and PRP, Group F worked through a problem of category OTHER. To assess the difference in performance, a pretest and posttest were administered in the class meeting in the week before, respectively after, the groups met separately. The preand posttest contained eight questions as shown in Table 3. The questions are split into the three categories KNOW, PRP, and ALG. Q1 Determine whether a heap is a suitable data structure to implement a waiting line in discrete event-driven simulation. Q2 Identify the concept that underlies heap sort. Q3 Identify the time complexity of heap sort. Q4 Determine the arrays representation of a heap. Q5 Determine whether a given tree is a heap. Q6 Trace a given sequence of operations on a priority queue. Q7 Trace the insert operation on a given heap. Q8 Trace the remove operation on a given heap. problems of category OTHER. Hence not only the students in Group F, but also the student in Group L, that followed the lecture model, had worked on exercises themselves as well. Programming exercises were not assigned in that week. The goal of the assessment is to determine whether students learn better through the use of the flipped classroom. In the lecture model learning takes place mostly outside of the classroom. In CS3151, many students delay studying the class material until they have to complete a homework assignment. Hence to fairly compare the learning in both models, all students had to complete the same homework assignment before the posttest was given. The assignment consisted of two problems of the category ALG and of two 4 Results Table 4 shows the results of the pre- and posttest. The maximum points for each question were 1.25. The mean and the standard deviation is displayed for each test question for Group L and F. The column labeled DIFF displays the difference among the means of the pre- and posttest. Table 4. Results of the pre- and posttest. Group L ALG PRP KNOW Pretest Posttest Group F Diff ES MEAN SD MEAN SD Q1 0.63 0.66 0.25 0.53 -0.38 Q2 0.88 0.60 1.00 0.53 Q3 0.50 0.65 0.63 Q4 0.00 0.00 Q5 0.41 Q6 Pretest Posttest Diff ES MEAN SD MEAN SD -0.44 0.42 0.65 0.42 0.65 0.00 0.00 0.12 0.16 0.42 0.65 0.83 0.65 0.41 0.46 0.66 0.13 0.14 0.00 0.00 0.42 0.65 0.42 0.65 1.13 0.40 1.13 2.85 0.21 0.51 0.63 0.68 0.42 0.49 0.49 1.13 0.40 0.72 1.14 0.42 0.65 1.20 0.127 0.78 1.19 0.50 0.65 1.13 0.40 0.63 0.83 0.83 0.65 1.04 0.51 0.21 0.25 Q7 0.25 0.53 1.13 0.40 0.88 1.33 0.21 0.51 1.04 0.51 0.83 1.15 Q8 0.25 0.53 0.63 0.66 0.38 0.44 0.42 0.65 1.04 0.51 0.62 0.76 218 According to Table 4, the lecture group showed gains in some areas while Group L showed gains in others. We discuss the results for each category. Both groups did not perform very well on the questions in the first category (KNOW). This is not very surprising, as the questions in this category had not been practiced in the homework or in one of the class meetings in any form. Group F shows a stronger improvement than Group L. It seems that the students were better able to retain the knowledge through the posted videos than through the lecture. Possibly, the students in Group F profited from the advantage that they could review the videos repeatedly if they could not follow the presented material the first time. Recall that the lecture group had seen examples and the flipped classroom group had worked through exercises related to the questions in the second category (PRP). But there had been no homework exercises on the corresponding topic. Group L improved significantly in the first question of category PRP, while Group F improved only slightly. Both Groups showed a similar improvement in the second question, where the improvement was slightly stronger for Group F. In the last category (ALG), Group L improved more than Group F in the first two questions, and Group F improved more in the last question. Recall that both groups had seen examples and exercises, respectively, of the corresponding topic. Both groups had seen questions in the homework assignment that were very similar to the questions in the pre-and posttest. These questions practice operations on priority queues and heaps. The complexity of the operations in Q6, Q7, and Q8 increases with Q6 covering the least complex operations and Q8 covering the most complex one. It seems that the hands-on exercises in the flipped classroom helped the students learn in particular the more complex the problems better. Another reason for the better performance on complex tasks might be that students in the flipped classroom had gone through the correct solution to similar exercises in the class. For simpler exercises most students are able to determine the correct solution on their own. But for complex ones, students may need to see more solutions of related problems. We also compare the final grades of students in Spring 2014 with the grades in preceding semesters when the traditional lecture model had been applied. Before Spring 2014, CS3151 had been offered in Spring 2013, Fall 2012 and Fall 2011 and had been taught by the same instructor as in Spring 2014. There has been significant improvement concerning the withdrawal and passing rates. In Spring 2014, 81.5% of the students passed the class with a C or better. The passing rates in the preceding semesters were 66.7% (Spring 2013), 63.6% (Fall 2012), and 53.7% (Fall 2011), respectively. However, it is hard to compare final grades as the grade calculation has differed in different semesters. For example, the grades from Fall 2011 to Spring 2013 included a Int'l Conf. Frontiers in Education: CS and CE | FECS'15 | quiz grade that was determined by the use of iClickers. Also the grading rubric for homework had been changed in Spring 2014, typically resulting in higher grades. Thus, it is hard to determine if the students’ performance has indeed improved, and if so, in how far the improvement is based on the flipped classroom model. Table 5 lists for each semester and for each final letter grade the number of students achieving that letter grade and the corresponding percentage of students with that letter grade in the class. For example, in Spring 2014 four students received an A in the class, which is 14.8% of the students. This is the highest percentage of students that have earned the letter grade A compared with the other class offerings. No student withdrew from the class in Spring 2014 while about 17% to 25% withdrew from the class in preceding semesters. Table 5. Grade distribution per semester. Class size A B C D F W Fall 2011 Fall 2012 Spring 2013 Spring 2014 54 44 36 27 3 (5.6%) 14 (25.9%) 12 (22.2%) 10 (18.5%) 4 (7.4%) 1 (2.3%) 15 (34.1%) 12 (27.3%) 4 (9.1%) 1 (2.3%) 2 (5.6%) 5 (13.9%) 17 (47.2%) 2 (5.6%) 4 (11.1%) 4 (14.8%) 6 (22.2%) 12 (44.4%) 2 (7.4%) 3 (11.1%) 11 (20.3%) 11 (25.0%) 6 (16.7%) 0 (0%) Considering that no student withdrew from CS3151 in Spring 2014, the flipped classroom seems to strengthen the students’ confidence level, which has also been confirmed by [2], for example. One might argue that the higher grades might be the cause for the low withdrawal rate. Probably the grades of a student play an important role in the decision to withdraw from a course. But this is likely not the only cause since in previous semesters many student withdrew from the course even though they still had passing grades. As pointed out before, the letter grade does not offer a true comparison as far as mastering the course content is concerned. A better indicator whether students perform better might be the test grades. In all semesters very similar tests were given. Table 6 lists the total average test grade. In each semester, except for Fall 2011, three tests were given. In Fall 2011, four tests were given. The maximum test grade for each Int'l Conf. Frontiers in Education: CS and CE | FECS'15 | test is 100. The test grades do not demonstrate a clear performance improvement. Note the total average includes only the test grades of students that had not withdrawn from the course. Thus the calculated average may favor semesters at which more students withdrew. Table 6. Total average tests grades per semester. Semester 5 Average total test grade Spring 2014 72.07 Spring 2013 67.81 Fall 2012 73.95 Fall 2011 71.18 Conclusions Based on the two-week study and on test grades, it does not seem that students are performing better overall on exercises in the flipped classroom compared to the traditional lecture model. The performance was compared after the corresponding material had been covered and practiced in both classroom models. There is some indication that the students perform better in the traditional classroom on less complex problems for which they have seen examples in class. Students may perform better on complex problems in the flipped classroom if those problems have been practiced in class. The final grades have been significantly better in the flipped classroom. However, a comparison among different semesters is difficult due to the different student population and the difference in grade calculation. No student withdrew from the flipped classroom while about 17% to 25% of the students had withdrawn from previous course offerings that applied the lecture model. This may be contributed to the higher grades students received before the withdrawal deadline. But the flipped classroom may also raise the 219 students’ confidence level as not all withdrawals are made by students with failing grades. 6 References [1] Campbell, J., Horton, D., Craig, M., Gries, P., Evaluating an Inverted CS1, Proceedings of the 45th ACM Technical Symposium on Computer Science Education, 307312, 2014. [2] Enfield, J., Looking at the Impact of the Flipped Classroom Model of Instruction on Undergraduate Multimedia Students at CSUN, TechTrends, 57 (3), 14-27, 2013. [3] Gannod, G. C., Burge, J. E., Helmick M. T., Using the inverted classroom to teach software engineering, Proceedings of the 30th International Conference on Software Engineering, 770-786, 2008. [4] Gehringer, E. F. & Peddycord III, B. W., The InvertedLecture Model: A Case Study in Computer Architecture, Proceedings of the 44th ACM Technical Symposium on Computer Science Education, 489-494, 2013. [5] Hughes, H., Introduction to Flipping the College Classroom, In World Conference on Educational Multimedia, Hypermedia and Telecommunications, 2012 (1), 2434-2438, 2012 [6] Lage, M. J., Platt, G. J., Treglia, M., Inverting the Classroom: A Gateway to Creating an Inclusive Learning Environment, Journal of Economic Education, 31 (1), 30-43, 2000. [7] Lockwood, K., Esselstein, R., The Inverted Classroom and the CS Curriculum, Proceedings of the 44th ACM Technical Symposium on Computer Science Education, 489494, 2013.