Simultaneous multidimensional impacts of active learning revealed in a first implementation in the MENA region

Significance

The United Nations Sustainable Development Goals underscore the principle of “leaving no one behind” through “a holistic approach to achieve sustainable development for all.” Quality education, the fourth goal, targets learning outcomes and skills. In the Middle East and North Africa (MENA) region, while access to education is growing, the quality of education and education-driven economic growth are lagging. MENA has the world’s largest youth population and the highest youth unemployment rate. The impact of unrealized education promise cannot be excluded from drivers of youth’s frustration and instabilities. Active learning can potentially reform education, particularly in science, and materialize the education promise. We show that a grassroots implementation outside developed countries using open resources is feasible, resulting in multidimensional gains.

Abstract

Enhancing science education in developing countries has been a focal point of many studies and efforts, but reform has mainly been driven by top-down approaches that often face impediments. A shift to active learning pedagogies can potentially address these challenges, but it has thus far been predominantly implemented and understood in developed countries. Thanks to the growing accessibility of open education resources and ubiquitous technologies, education reform can now be carried out from the bottom up. Here, we present the results of a two-year implementation of active learning in five core physics and astronomy courses comprising 2,145 students from the Middle East and North Africa (MENA) region. Simultaneous improvements are observed in both students’ performance and their perception of the quality of learning; means improved by 9% (0.5 SD) and 25% (1.5 SD), respectively. Moreover, the performance gap between students in the bottom quartile and those in the top quartiles was narrowed by 17%. The failure rate was reduced to a third of that in traditional classes; this is 36% better than the results in developed countries, indicating a greater need for active pedagogies by MENA students. Our findings reveal a multidimensional positive influence of active learning, the viability of its grassroots implementation with open resources, and its sustainability and reproducibility. We suggest that wider implementation can boost education-driven economic growth by 1% in per capita gross domestic product [GDP], substantially cut costs of repeating courses, and produce a more competent STEM workforce—all of which are urgently needed to stimulate development and growth.

Homegrown scientific competitiveness is essential to a sustained development and economic growth, especially in our time where global challenges demand unconventional scientific innovation (1). But a healthy educational system must be in place to (a) entice new generations to science and equip them with the qualities they need to drive innovation and development in order to compete in today’s labor market and (b) provide science literacy to all, including nonscience students, so the values of science—including inquiry, critical thinking, and the practice of the scientific method—are widely adopted by the population, thereby creating more rational societies (2–5). Such an educational system is indispensable, with up to half of the traditional jobs expected to become obsolete due to automation and the new ones demanding more interdisciplinary skills (1). Evidence shows that leveraged education-driven economic growth is tied to the actual cognitive and technical skills acquired from quality learning and not merely to the attainment of certain years or levels of education (2, 3). The need is more urgent in vulnerable countries in Africa and Asia (6–11). Despite a series of reforms, particularly in the Middle East and North Africa (MENA; over 364 million in population, with youth (aged under 30) constitute more than half (55%), compared with 36% across OECD countries), the number of professional scientists and researchers is well below the global average, and the region’s average students have the lowest results in international standardized tests, the largest performance gap (between top 75% and lowest 25% performing students), and the highest impact of mindset on performance (12–14). Outdated and disengaging science education, which teaches science as a body of facts and assesses mainly on rote memorization, does not only turn youth away from science, technology, engineering, and mathematics (STEM) disciplines (15–17) but also degrades the overall learning and skill outcomes of the graduates, reducing the contribution of education to economic growth.

Reforming STEM education to engage students and enhance their learning and skills has its own challenges (18, 19), particularly in physics as a core and foundation subject in virtually all STEM disciplines. Teaching through research-validated pedagogies, such as by active learning, has proven to yield positive outcomes and is being adopted in many developed countries (20–23). Besides the anticipated learning gains, this approach restores a consistency between the way science is practiced and the way it is taught, with the role of evidence, continuous improvement, and scrutiny reflected in science teaching as it is in research. Active learning works because it enhances teaching effectiveness and student engagement by assigning active roles to students in the learning process during its different phases (24). In contrast to traditional teaching, which relies primarily on lecturing with the students as passive recipients, active pedagogies create an atmosphere that supports personalized learning and reflection through in- and out-of-class activities that helps students develop a deeper understanding by connecting new ideas and concepts to existing knowledge and experiences (25). While it does not necessarily require expensive resources, active learning can substantially benefit from, and makes good use of, ubiquitous and affordable technology (26, 27). Types of active learning include collaborative learning or peer instruction, inquiry-based projects, technology and interactive visualization, and conceptually oriented tasks (28). Pedagogies often use several of these types separately or in combination.

The actual implementation of science education reform faces challenges, including the availability of educational material and technology solutions that truly engage students and fit the learning goals, resistance to change (from students, faculty, or the administration), and the need for training of faculty and students in the new approach. In developing countries, a shortage of funds and resources amplifies these challenges even more. Recommendations for successful reform include building on existing resources and adopting them to local circumstances; taking into account faculty and students’ existing knowledge, beliefs, and attitudes; and giving activities and pedagogies a local context and practical relevance (10, 29).

Nevertheless, the present understanding of the impact of active learning comes almost entirely from its implementation in developed countries, as primarily measured from the enhancement of performance in examinations and concept inventories as well as from the reduction of failure rates. Most individual studies on physics courses have reported an average increase in the mean of 0.60 SD in examinations and similar assessment components (30–34), with a larger 0.94 SD impact on concept inventories (35–38). In meta-analyses that compared active learning with traditional teaching controls and included other STEM fields, Ruiz-Primo et al. measured an improvement by 0.47 SD in the mean of assessment components from 166 STEM studies and an improvement by 0.59 SD from 71 physics studies (28). Freeman et al. (22) measured an improvement in the mean by 0.47 SD, which corresponds to half a letter grade from 158 STEM studies, with that of physics improving by 0.72 SD from 31 physics studies, with a corresponding letter grade enhancement from B to a B+. Expanding the impact to include the reduction of failure rate, Freeman et al. (22) measured a 35.5% drop in failure rate from 67 STEM studies.

Here, we present the impact assessment of transforming five undergraduate physics courses to active learning using nearly identical approaches and curricula to those in the references above on a sample of students from the MENA region at Sultan Qaboos University (SQU) in Oman. The courses are taught in English to nonnative speakers, a practice that is now common in the MENA universities, particularly those with internationally accredited programs, and are part of the Bachelor of Science (B.Sc.) degree program in physics (accredited by Accreditation Agency for Study Programmes in Engineering, Informatics, Natural Sciences and Mathematics [Germany]). We explore the impact of the active pedagogies on performance (course means and failure rates), as in the references above, e.g., refs. (22, 28), and also assess the impact on other aspects, including perception, attitude, and the performance gap (between top and lowest performing students). We compare the results to students in the refs. above and discuss their implications and prospects of a wider implementation.

1. Motivation and Plan

As typical for MENA students, the students in this study have lower average science scores on international tests due to their inadequate science education during K to 12 (12, 13). At the university level, this leads to greater rates of failure and low enrollment and retention in STEM majors. For example, the average failure rate in all foundation physics courses at SQU is about 25% with traditional teaching, the highest at the SQU College of Science. To tackle these issues, we developed a bottom-up strategy to implement active learning in a way that is agile, relevant, and appealing to the students, while being cost effective. The objective is to improve students’ learning and conceptual understanding of physics and engage them in a manner that strengthens their critical thinking, collaborative, and technical skills. We applied active learning to five courses during the 2017 to 2018 and 2018 to 2019 academic years: PHYS I, PHYS II, PHYS III, ASTRO I, and QM (SI Appendix, Table SI.1). The courses are offered by the department of physics at SQU as part of its B.Sc. program and service courses. The levels span from introductory physics (PHYS I and II) to intermediate foundation physics (PHYS III) and advanced quantum physics (QM) (all are core courses in the Physics B.Sc. program), and they also include an elective astronomy course (ASTRO I). All five courses are calculus based and taken by both physics majors and students from other majors at the College of Science, plus students from the Colleges of Education and Agriculture. Each course offering was a full, semester-long course, following the United States model, with the standard course letter grades of A to F.

2. Methods

2.1. The Pedagogy.

Active learning in PHYS I, II, and III included reading quizzes, peer instruction, and context-rich–based tutorials, where students were encouraged to work collaboratively. The students were distributed in groups of three that were rotated every 3 wk. In the large introductory PHYS I course, low- and high-performing students were mixed to stimulate engagement among all students, whereas in the PHYS II and III courses, the group distribution was random. In the ASTRO I and QM courses, interactive visualizations, peer instruction, as well as inquiry-based projects were utilized. All five courses included prelecture textbook readings, videos, and conceptual questions, with peer instruction exercises incorporated during the class. We used open educational resources in tutorials and peer-instruction exercises that had been developed at the University of Minnesota and the University of Colorado (39, 40) but edited them to generate context-rich activities with local context and practical relevance (SI Appendix, section SI.2). These active approaches were utilized throughout the semester, including in classes, tutorials, and assignments, to maximize the impact on the students at an early point and remedy accumulated weaknesses from preuniversity. The free app Google Classroom was used as a communication and sharing platform for all course activities and material while mainly relying on students accessing it with their smartphones. This not only reduced costs vis a vis clickers but also better engaged the students, both inside and outside the classroom, with a technology that they are extremely adept at. This sustainable bottom-up learning and teaching environment is currently running, having expanded to include all physics foundation courses at SQU. The midterm and final examinations were kept at the same level as those in the traditional offerings that served as controls. Although students worked in groups for peer instruction and tutorial activities, they were assessed individually in all assessment components.

2.2. The Experiment.

The five courses described above represent all the courses that were taught using active learning in the Department of Physics during two academic years between 2017 and 2019. Eight classes were taught actively (“A classes/offerings”), whereas 11 classes were traditionally taught (“T classes/offerings”) and served as controls (Table 1). In the T classes, the instructor delivered the content using the T lecturing mode on the board or through a presentation based on the textbook material, with the students being passive learners except for brief periods of questions; in the T class tutorials, the instructors solved problems. Virtually all the instructors of the A offerings (six out of seven) had previously taught the same or equivalent courses in the T mode, not necessarily back to back with the A offering (SI Appendix, section SI.5). The transformation to active learning in the multisection courses PHYS I, II, and III was coordinated over three consecutive semesters among the faculty teaching them and independent of that in ASTRO I and QM. The students in this study all had the same academic preparation in both the T and A offerings, as measured by their high school grades, whose averages are 88.4% and 88.8% for the T and A students, respectively.

Table 1.

Performance impact

Courses	Offering comparison: active (A) vs. traditional (T)	Active learning (A)			Traditional teaching (T)			% gain in mean	t test			Effect size in Hedge’s g
		nA	MA	SA	nT	MT	ST		t	SE (t)	P	g	SE (g)	95% CI (−)	95% CI (+)
PHYS I	3A vs. 3T	725	63.17	12.77	928	58.29	15.06	8.38	6.99	0.69	≪10−4	0.35	0.05	0.25	0.44
PHYS II	2A vs. 2T	99	70.35	9.44	77	62.28	12.06	12.97	4.98	1.67	<10−4	0.75	0.16	0.45	1.06
PHYS III	1A vs. 2T	56	65.22	8.87	84	61.30	14.80	6.39	1.78	2.00	0.077	0.31	0.16	0.00	0.61
ASTRO I	1A vs. 1T	55	77.84	8.62	51	74.17	9.01	4.95	2.14	1.72	0.034	0.41	0.19	0.03	0.80
QM	1A vs. 3T	12	62.03	7.38	58	54.91	14.73	12.96	1.62	2.88	0.109	0.51	0.21	0.11	0.91
All courses	8A vs. 11T	947	67.72	9.42	1198	62.19	13.13	8.90	10.93	0.49	≪10−4	0.48	0.04	0.39	0.56

The mean scores in the active (A) and traditional (T) offerings of the five courses, showing the total mean course points (M) (unweighted), class (sample) size (n), sample SD (S), percentage gain in the mean, the t test, and the effect size in Hedge’s g. The unweighted means were used to remove the bias due to class size among the different courses and offerings. The Nos. in the offerings comparison, e.g., 3A versus 3T, refers to the No. of course offerings used in the A and T modes.

To study the impact, the official final course results were obtained from the university admission and registration office. The students’ course evaluation results were also obtained (with the courses and instructors’ names removed and all A and T offerings of the five courses being averaged respectively to maintain instructor anonymity). The final course scores (total course points out of 100) measure the terminal and true impact on performance at the end of the course, rather than emphasizing any specific active learning component within the course.

Impact was measured based on the reported mean of total course points in the A offering relative to the most recent previous T offerings with an equivalent assessment. Similarly, the failure rate was obtained from the reported ratio of students with D and F grades. The A and T offerings of the same course were ensured to have equivalent examinations and assessment components through the departmental standard practice (since its establishment in 1988) that (a) each final examination is formally reviewed by cochecker(s) and (b) the overall results of all courses are reviewed, discussed, and approved by department examination board meetings, during which the courses’ averages, grade distributions, failure rates, ratios of letter grades, numerical cutoffs, and percentages of syllabus covered are reported and compared with previous offerings. This practice is designed to minimize human error and ensure consistency in the assessment of courses. The cocheckers for the single-section courses (PHYS II and III, QM, and ASTRO I) were other independent instructors who had taught these courses before (i.e., in the T offerings in this case). The cocheckers for PHYS I were the same faculty members who taught it during the data collection period, several of whom had taught it multiple times in the T mode before the T to A transformation (the general practice in the department is that multisection courses are cochecked by their instructors, particularly those who taught the courses multiple times). The A and T offerings of all these courses, as all other courses in the college, go through additional checks in a college board meeting to look for any inconsistencies among the various offerings. It should be noted that the A offering received the same level of scrutiny at all check levels as those taught in the T mode. The collection of the data and analysis reported in this study took place after the conclusion of the implementation of active learning and the reporting of the course results, including the review and approval mentioned above.

Our analysis strategy focused on measuring the inclusive impact aspects, including (a) all performance indicators (mean, failure rate [odds and risk ratio], grade ratios, and performance gap), (b) students’ attitudes and perceptions of the quality of teaching and learning experience from the official anonymous course evaluations conducted independently by the university toward the end of the semester, and (c) the interplay and correlations among these variables.

We grouped the A and T offerings of each course and obtained the mean of the total course points (M), class (sample) size (n), sample SD (S), and the percentage of students with the letter grades A+B, C, and D+F (Tables 1 and 2). The difference in the mean (t test) and mean effect size (Hedges’ g) were calculated for each course as shown in Table 1. The impact on failure rate was assessed from the ratio of D+F grades in the A and T offerings as shown in Table 2. This approach is consistent with previous studies [e.g., refs. (22, 28)] and allows a comparison between the students’ groups. Withdrawal rates were not included in our analysis. The numeric values of the course Ms that define the impact on performance and those of the D and F cutoff scores are given in Table 1 and SI Appendix, Table SI.2, respectively.

Table 2.

Grade ratios, failure rate, and odds and risk ratios of failure

	A+B		C		D+F failure rate (FR)		% reduction in FR	FR ratio (T/A)	ORTA (odds ratio of failure in T relative to A offerings)					RRAT (risk ratio of failure in A relative to T offerings)
	T (%)	A (%)	T (%)	A (%)	T (%)	A (%)			ORTA	Ln ORTA	SE Ln ORTA	Z	P	RRAT	Ln RRAT	SE Ln RRAT	Z	P
PHYS I	34.94	48.26	39.55	35.19	25.11	16.71	33.45	1.50	1.67	0.51	0.35	1.38	0.167	0.67	−0.41	0.28	1.55	0.1200
PHYS II	45.48	67.39	32.42	30.69	22.08	2.02	90.85	10.92	13.73	2.62	0.75	3.48	0.0005	0.09	−2.39	0.72	3.31	0.0009
PHYS III	42.95	46.43	36.70	48.21	20.34	5.36	73.65	3.80	4.51	1.51	0.51	2.98	0.0029	0.26	−1.33	0.46	2.89	0.0038
ASTRO I	70.59	78.18	21.57	18.18	7.84	3.64	53.57	2.16	2.25	0.81	0.65	1.17	0.243	0.46	−0.77	0.62	1.48	0.1385
QM	38.98	66.67	28.83	25.00	32.07	8.33	74.03	3.85	5.19	1.65	0.42	3.96	0.0001	0.26	−1.35	0.36	3.76	0.0002
All courses	46.59	61.38	31.81	31.45	21.49	7.21	66.44	2.98	3.52	1.26	0.46	2.73	0.0064	0.34	−1.09	0.41	2.79	0.0053

Grade ratios in the T and A offerings of the five courses, showing the upper grades (A+B), middle grade (C), and the lower grades (D+F) (that represent the failure rate) as well as the percentage reduction in the failure rate. The odds and risk ratios of failure and their statistical significance are also shown.

To probe how active learning was perceived by the students, the official course evaluations administered independently by the university were used. Each such evaluation consisted of 15 questions covering the various aspects of learning and course management (Table 3 and SI Appendix, Table SI.3B).

Table 3.

Perception impact

Course evaluation question	A			T			% gain in mean	t test			Effect size in Hedge’s g
	nA	MA	SA	nT	MT	ST		t	SE (t)	P	g	SE (g)	95% CI (−)	95% CI (+)
All questions (Q1–Q15)	6893	81.08	6.90	5760	73.02	13.22	11.05	44.02	0.19	≪10−4	0.79	0.019	0.749	0.823
Learning experience questions (Q6–Q11, Q14)	3271	83.00	6.68	2733	70.71	13.90	17.38	44.75	0.29	≪10−4	1.16	0.027	1.106	1.213
Perception & attitude questions (Q7, Q8, Q9)	1428	82.93	6.88	1185	66.24	14.25	25.21	39.12	0.45	≪10−4	1.54	0.042	1.455	1.618
Response rate (Q1–Q15)	6893	459.53	35.69	5760	384	25.20	19.67	134.94	0.54	≪10−4	2.41	0.017	2.375	2.443

Students’ evaluations survey scores averaged among the A and T offerings of the five courses (to maintain instructor anonymity). The total average score of the 15 questions is given as well as those of the questions probing students’ learning experiences and perceptions and attitudes. Course-evaluation questions are listed in SI Appendix, section SI.3. Shown are the total mean score (M), sample size (n) (which represents the No. of responses and can vary from question to another because students can skip questions), sample SD (S), percentage gain in the mean, the t test, and the effect size in Hedge’s g.

An institutional review board approval was not required for this study. The study was reviewed and deemed exempt based on the nature of the data and methods presented herewith.

3. Results

3.1. Performance Impact.

The A offerings for the five courses showed an increase in the mean of the total course points that ranged from 5 to 13% compared with the T offerings (Table 1). This corresponds to an improvement by 0.31 ± 0.16 to 0.75 ± 0.16 SD, which also represents the Hedges’ g effect size in units of SD.

The failure rate was greatly reduced under active learning in each of the five courses, and the reduction rate ranged from 33 to 91%, corresponding to an odds ratio of failure OR_TA of 1.67 to 13.73 (Table 2), i.e., students on average are 1.67 to 13.73 times more likely to receive D or F under T teaching compared with active learning. The corresponding risk ratio RR_TA is in the range of 1.50 to 10.93, i.e., when active learning is absent, low-performing students are 1.5 to 11 times more likely to receive D or F. A more intuitive way to assess the reduction in failure is through the risk ratio of failure in A relative to T offerings (RR_AT), whose range is 0.1 to 0.7, indicating that active learning has lowered a 1.0 probability of failure for the lowest performing students to 0.1 to 0.7. Note that this is despite the fact that the cutoff scores for grades D and F in the A offerings in all required core courses (PHYS I, II, and II and QM) were higher than those of the T offerings (SI Appendix, Table SI.2).

On the other hand, the impact on the upper grades for each of the five courses show an improvement in the A+B grades whose percentage increase ranges from 8.09 to 73.60%, whereas the grade C remained nearly unchanged (Table 2). Fig. 1A graphically compares the improvement in performance in the five courses, as measured by the effect size in SD (Hedge's g effect size), the percentage reduction of failure rate (D+F grades), and the percentage increase in the upper grades (A+B). A comparison between the distributions of the final scores of the entire T and A offerings of the five courses is shown in Fig. 1B. The overall active learning distribution is enhanced by a 9% increase in the mean compared with the T offerings, and the region with the lowest-performing students is populated only by students from the T offerings.

Fig. 1.

The impact of active learning among the individual five courses studied (A) and using joint five-course results among the active learning and traditional teaching offerings (B). A shows the interplay among the performance indicators in each of the five courses in terms of the enhancement in the final course mean in units of SD (also represents the Hedges’ g effect size), the improvement in the upper grades (A+B), and the reduction of failure rate (D+F grades). B shows the kernel density distribution with rug marks of the final course scores (total points of all components out of 100) of the entire traditional teaching (red) and active learning (blue) offerings of the five courses. The overall active learning distribution is enhanced by a 9% increase in the mean, and the lowest-performing students with scores less than ∼25 points are present only in the traditional teaching offerings, although both samples had similar population of at-risk students with low GPA (Fig. 2).

3.2. Performance Gap.

To bring the students’ overall academic standings into the picture, we utilized the cumulative grade point average (GPA) at the end of the semester*, and as shown in Fig. 2A, we plotted the GPA and the final course scores for each student, comparing the entire A (blue dots) and T (red dots) offerings. It is striking that the student group shown in the orange box with the lowest GPA and lowest total course scores belongs entirely to the T offerings: in the A offerings, students with the same lowest GPA range are pulled up with higher scores; that is, the very low-end tail in the final course scores is eliminated. The comparison between the distributions of the final scores of the T and A offerings of the five courses, shown in Fig. 1B, indeed shows that the lowest performing students with scores less than ∼25 points have been pushed into higher grades in the A offerings and the overall mean of the active learning distribution is enhanced by 9%.

Fig. 2.

Distribution of students’ final course scores (total points of all components out of 100) and the cumulative GPA (at the end of the semester) in the entire traditional teaching (red) and active learning (blue) offerings of the five courses. (A) A scatter plot with one-dimensional kernel density and rug marks on the right and top axes. (B) Two-dimensional kernel density contours. In A, the lowest-performing students in the orange box with scores less than ∼25 points (also indicated on the right x axis) come exclusively from the traditional teaching offerings, although at-risk students with the lowest GPA (less than ∼2; top y axis) are equally present in both the traditional and active offerings. Vertical and horizontal dotted lines are the mean values, and the solid lines are the best-fit linear regression and their 95% confidence regions. The linear regression has a steeper slope of 20.42 ± 0.64 for traditional teaching offering (t = 32.04, Pearson’s R = 0.68, P ≪ 10⁻⁴) compared with a slope of 16.14 ± 0.50 for the active learning offerings (t = 32.45, Pearson’s R = 0.73, P ≪ 10⁻⁴). In B, the two-dimensional kernel density contours are plotted on identical scales for accurate comparison. The red and blue dotted lines are the linear regression shown in A. The comparison reveals (i) less spread of the overall distribution of the active offerings by 10% (along the linear regression lines across the GPA scale) and (ii) a 25% narrower spread in the scores at the average GPA in the active offering (by 36.62 points) compared with the traditional offerings (48.90 points), and (iii) the orange box of the lowest-performing students that is populated by some students in the traditional offerings is completely empty in the active offerings, with the students of the same lowest GPA range lumped up into higher scores.

The contrast between the A and T offerings shown in Fig. 2A through the final course scores and GPA relation is further illustrated through the two-dimensional kernel density contour plots shown in Fig. 2B, where the student population in the A offerings show narrower spread, absence of the lowest-performing students, and less steep overall slope and the tail of the A offerings distribution is pushed up toward higher scores compared with that of the T offerings.

Even disregarding the low-end tail, we see in Fig. 2 that the score distribution for A courses has a narrower spread compared with that of the T courses, with more students lumped into higher scoring ranges, which suggests a reduction in the performance gap between the top- and lowest-performing students. We calculated the performance gap from the difference between the average final scores of the top 75% and lowest 25% students (a convention widely used in the references above when assessing performance gap [also called achievement gap]). Indeed, we find that a performance gap of 38.64 ± 0.61 points in the T offerings was narrowed to 32.24 ± 0.52 points in the A offerings, as shown in Fig. 3, at a 16.56% reduction.

Fig. 3.

Comparison between the performance gap (defined as the difference between the average final course scores of the top 75% students and the lowest 25% students) in the traditional teaching and active offerings of the five courses. A performance gap of 38.64 points (out of 100) in the traditional teaching offerings is reduced to 32.24 points in the active offerings, signifying a 16.56% reduction.

The narrower distribution of the A offerings relative to the T offerings shown in Fig. 2B (by 10% across the GPA scale and by 25% at the average GPA of each offering) supports the observed reduction in the performance gap above but also signifies a broader reduction in performance across the GPA scale for the students of the A offerings relative to those T teaching offerings.

3.3. Perception Impact.

The impact of active learning on students’ perception and attitude (as determined from the course evaluations) showed gains that are even larger than the gains in performance (Table 3). The overall mean evaluation score of the five courses grew in the A offerings by 11% (0.79 SD) relative to the T teaching offerings, with a mean effect size g_Perception = 0.79 ± 0.02 (P ≪ 10⁻⁴) (a five-course mean effect size was used to maintain the anonymity of the instructors). The component that probes the learning experience (Q6 to Q11 and Q14 in SI Appendix, Table SI.3A) grew by 17.38% (1.16 SD), with a mean effect size g_{Learning Experience} = 1.16 ± 0.03 (P ≪ 10⁻⁴).

A positive impact on the students’ attitudes and perceptions of their learning is inferred from the evaluation questions Q7, Q8, and Q9^†, whose combined mean grew by 25.21% (1.54 SD), corresponding to a mean effect size of g_Attitude = 1.54 ± 0.04 (P ≪ 10⁻⁴). The individual means of these questions increased by 26.60%, 23.09%, and 25.71% at respective mean effect sizes of g_Q7 = 1.70 ± 0.07 (P ≪ 10⁻⁴), g_Q8 = 1.16 ± 0.07 (P ≪ 10⁻⁴), and g_Q9 = 1.82 ± 0.07 (P ≪ 10⁻⁴). The students’ response rate to the evaluations also grew by 19.67% (2.41 SD), with its mean effect size g_{Response Rate} = 2.41 ± 0.02 (P ≪ 10⁻⁴). Note that the T teaching offerings had 26% more students than those in the A offerings (SI Appendix, Table SI.3B).

4. Discussion

The implementation of active learning in the courses presented here revealed positive impacts on the students for all the categories probed, including performance (course mean), failure rate, performance gap, and perception and attitude, all of which occurred simultaneously.

The measured gains can be understood and attributed to a number of factors. With regard to strategy, (1) we utilized active learning in all course components, including lectures, tutorials, and assignments, throughout the semester; (2) we tapped into a familiar technology, students’ smartphones and free apps for in-class activities and course management, which required no additional resources; (3) course materials were accessible from any device and across operating systems, increasing accessibility; (4) peer instruction motivated students to exercise open inquiry and immediately face concepts just learned in lectures, giving them the chance to practice critical thinking and team-based learning; (5) tutorials and exercises played a vital role in engaging all students, in particular the low performers, who started playing a proactive role in their own learning by being mixed with high performers (and hence trying to compete with them); and (6) giving physics real-life context through context-rich problems and project-based learning customized to the local environment, thus tying physics to students’ cultural experiences.

Observing the highest reduction in failure rate (and lowest risk ratio of failure) where there is the largest effect size in course mean, as shown in Fig. 1 and Table 1, confirms that active learning impacted the low-performing students efficiently, as also revealed by Fig. 2, where the lowest-performing students exclusively come from the T offerings. It is significant that this occurred with higher passing grade requirements in the A offerings compared with the T ones.

The observed improvements in performance by 5 to 13% (g = 0.31 to 0.75) signify a considerable change. It has been argued that, in such real-world education experiments, “a fifth of a standard deviation”, i.e., g of 0.20, “is a large effect” (41) and that the best empirical benchmarks are obtained from the highest-quality results (42–44). Indeed, our result compares reasonably with the most recent meta-analyses, which reported g ∼0.60 (22, 28).^‡ The average improvement in the five courses (g = 0.49) corresponds to a GPA enhancement by 0.37, which would shift the average letter grade from C to C+. This is compared with a GPA enhancement by 0.34 and a letter grade shift from B to B+ reported in ref. (22).^§

Our improvement for the odds ratios of failure (OR_TA = 1.67 to 13.73 and its average of 3.52; that is, students being 3.52 times as likely to fail with T teaching) is considerably larger than that associated with the similar performance improvement in the ref. (22) meta-analysis, which reported an odds ratio of failure OR_TA = 2.07 for physics. Thus, when our nonnative speakers were exposed to essentially the same active learning approaches used with native speakers in developed countries, the result was a greater reduction in failure. That this occurs despite a comparable effect size in performance is attributed to a disproportionately large impact on the low-performing students, who benefited more than the high-performing students. A small increase in scores can allow students to cross the failure threshold, increasing the reduction of failure rate considerably with only a small impact on the total effect size, as also revealed by comparing the lowest-performing students in the T and A offerings in Fig. 2.

The reported gains on students’ performances were matched by even greater gains on their perception and attitude toward the overall quality of the teaching received and their learning experience, compared with T teaching. The improvements in the overall students’ evaluations (11.05%, 0.79 SD), their learning experiences (17.38%, 1.16 SD), and their perceptions and attitudes toward learning (25.21%, 1.54 SD) are key indicators that will await comparison with students from other regions. Our lowest-performing students (Fig. 2) tend to be the ones with lower self-esteem and interest in the course. Active learning has enhanced their participation and thus their perception, leading to a more positive attitude. Such enhancement in students’ perceptions is integral to reforming science education. It leads to better retention and enrollment and offers vital feedback to reform efforts. Recent surveys^¶ showed that 8 in 10 (83%) of the MENA region youth (over 200 million in population) are concerned about the quality of education in their country and nearly half think that the education they receive does not prepare them for the jobs of the future. Concerns about access to quality education and finding jobs ranked consistently as the top concerns of the MENA youth for over a decade.

The reported performance gap (39 points under traditional teaching and 32 points under active learning, a 17% reduction) remains considerably larger than the gap in preuniversity standardized tests among regional students, which ranges between 17 and 23 points (12) (all are out of 100 points and represent the difference between the top 75% and the lowest 25% performers). This signifies the challenges faced at the university level in STEM disciplines in less industrialized countries when courses are conducted in English at a standardized (accredited) level, compared with the preuniversity level, where teaching is in the native language using local curricula. This difference is most relevant to the first-year-level college physics courses. In the upper-level courses, students will have had more exposure and adaptation to English. We note that reading in our first-year-level courses (A offerings) was encouraged through prelecture “reading quizzes”.

It is indicative that the active learning approaches used with the students in this study have effectively created a support system that gives students an early warning on their difficulties and weaknesses and prompts them to continuously troubleshoot their issues and improve before taking the examinations (e.g., through peer instruction during lectures). This offered extra layers of diagnosis and treatment and a social experience that boosts the students’ morale. Obviously, the low performers benefit from this tremendously. In T teaching, on the other hand, even a fine lecture can give students a false sense of security, where they think they understand the material due to the lecture clarity, when, in fact, they can still have problems with the topics and their application (45).

Furthermore, to a student body of nonnative speakers, the active learning methodology adopted here lowered the effects of the language barrier during class instruction and with the textbook, which can hinder academic progress (46). They also helped students with vulnerabilities, including socioeconomic and attitude factors (attitude, motivation, cultural, and/or emotional). Such students, especially those with low GPA, tend to be reluctant to seek help through the available support channels or be identified (47). These issues may particularly affect students from developing countries. Indeed, students in the MENA region were found to be more vulnerable to attitude factors than their peers elsewhere, and those with positive motivation were found to outperform students with a poorly calibrated attitude by 14%, higher than any other region (13). Positive attitude and supportive instructors (which active learning can induce as part of its activities) have been shown to induce higher academic achievements from low-performing students and to narrow the performance gap (48, 49). We argue that the engagement activities in small groups and the practice of inquiry and reflection utilized here created personalized atmospheres and learning pathways that catered to the students’ individual needs, leading to improvements in attitude and calibrated motivation. Evidence of a positive impact on our students’ attitude toward their learning is inferred from the evaluation questions Q7, Q8, and Q9^ǁ, whose five-course average increased by 25% with their means increased by 1.54 SD, reflecting a significant positive outlook and perception of learning. Additional evidence comes from students’ written comments in the same course evaluations (SI Appendix, section SI.6).

Thus, in conclusion, the implementation factors, the nature of the student body, and the impact on attitude can plausibly explain the observed gains. Students in developing countries, and particularly those in the MENA region, are in great need of and can clearly achieve higher overall results with active learning. The narrowing of the performance gap lends further support to this proposition. All in all, we infer that active learning has created extra layers of support and offered efficient learning opportunities that enhanced academic performance and lowered the impact of student vulnerabilities, leading to simultaneous positive impact on performance, failure rate, perception, and the performance gap, with the gain in perception being more than twice that of performance. Looking ahead, a tweaking of active learning approaches and their better alignment with course learning goals may further increase student success. This is particularly important during the current post-COVID period where universities are facing cohorts of students with greater needs for support.

Our work shows that active learning is flexible enough to implement without undue cost and can, if widely implemented, lead to a substantial reduction of lost tuition due to repeating courses. The reduction of failure achieved in the five courses is tantamount to saving $158K (SI Appendix, section SI.4). At this rate, we can estimate a saving of $610K per semester for our foundation physics, chemistry, biology, and mathematics courses (15 courses taken every semester by an average 3,653 students from various STEM majors at SQU). This represents $724K per 100 students per degree program (of about 130 credit hours), a substantial saving that can help enhance other aspects of learning. Policy makers planning for such transformations should take into account the available resources and cultural contexts to achieve the desired outcomes.

Quality education has impact far beyond improving performance in courses and cutting costs; it has been shown to boost economic growth through improving cognitive skills and innovation vis-à-vis existing and new technologies, with a positive correlation between the growth in annual GDP per capita and international preuniversity test scores, a correlation that is particularly strong for science (50–52). If we assume that a comparable rate applies at the university level, our mean result on the performance improvement of nearly half an SD would correspond to 1% of the annual growth in GDP per capita; that is, active learning would boost the education contribution to economic growth by adding an additional one percentage point to the GDP annually^††.

5. Conclusion

The present study reveals a successful first implementation of active learning on our sample of students from the MENA region, revealing significant gains in performance, perception, reduction of the failure rate, and narrowing the performance gap (the gap between the top 75% and the lowest 25% students). The results pinpoint the individual constituents of the impact indicators and probe their underlying dynamics. Findings include (a) observing simultaneous gains in performance and perception (enhanced final mean course score, reduced failure rate, enhanced students’ learning experiences, and having a more positive perception of the quality of teaching), (b) narrowing of the performance gap, (c) revealing the interplay among the active learning impact variables^‡‡, and (d) showing a comparable performance improvement by MENA students as that by students from developed countries under the same set of active learning pedagogies and curricula but with MENA students having a considerably larger reduction of failure rate (by 36%), which shows that there is an even greater need for these improved pedagogies for students in developing countries.

Our results provide evidence that policy recommendations that make use of available international resources and adapt them to cultural contexts, where stakeholders are involved and take ownership of the reform from the start, can indeed yield positive results (9–11, 53, 54). The fact that we have completely offset the traditional need for dedicated resources makes our model applicable to universities in the MENA region and other developing countries.

Mounting evidence now shows that learning is more effective through active pedagogies, as cited in many of our references. Our study shows that these well-researched pedagogies can be adapted to different cultural contexts and are feasibly implementable at the grassroots level, producing positive impacts outside developed countries with nearly the same approaches. Because of open education resources and the spread of new technologies, such as smartphones and education apps, switching teaching to active modes is not resource intensive. Indeed, it is feasible with already available resources, as we have shown. This will help realize the economic potential of education, currently lagging in the MENA nations and other developing countries, and materialize the United Nations Sustainable Development Goals (SDG), particularly those on quality education and economic growth, while also acting as a catalyst for the other SDG.

Through investing in new evidence-based pedagogies and pushing the frontiers and boundaries of learning, universities around the world can initiate a timely upgrade to the present centuries-old teaching model, making them not just a place to provide a tertiary education but also advocates for advancing the practice of learning at all levels.

Data, Materials, and Software Availability

Raw data (students course results, GPA, and course survey, PHYS 2101 [PHYS I], PHYS 2102 [PHYS II], PHYS 3103 [PHYS III], PHYS 4101 [QM], PHYS 2901 [ASTRO I], Deanship of Admission and Registration, Sultan Qaboos University, Muscat, Oman, 2020) are available upon request from the corresponding author.