Pre-Recorded Lectures, Live Online Lectures, and Student Academic Achievement

Abstract

In the midst of the COVID-19 pandemic, universities throughout the world are embracing online learning, often depending on synchronous and asynchronous digital communications. In this paper, we compare the impacts of live online (synchronous) and pre-recorded (asynchronous) lectures on student achievement using a randomized experiment. We discovered that pre-recorded lectures reduce lower-ability students’ academic achievement but have no effect on higher-ability students’ academic achievement. In particular, being taught via pre-recorded lectures as opposed to live online lectures decreased the likelihood of answering exam questions correctly by 1.6 percentage points for students in the bottom 50th percentile of the ability distribution (measured by GPA at the beginning of the semester). Furthermore, being taught via pre-recorded lectures in the starting weeks of the semester compared to the later weeks tended to be more harmful to students’ academic achievement. These findings have important implications for the effective design of education policies.

1. Introduction

The COVID-19 pandemic has imposed negative consequences on the Sustainable Development Goals (SGDs) approved by the United Nations in 2015. Among the goals, SGD-4 was created to tackle global education quality. This goal promotes the idea that learners should be equipped with the skills and knowledge necessary to foster global sustainable development for the benefit of the environment, population, and future generations. However, the pandemic has forced many education institutions across the globe to organize their classes online to continue teaching and learning while curtailing mass infection. In Vietnam, many universities have been pushed to make a rapid transition from traditional face-to-face instruction to online classes since the beginning of the pandemic. Such immediate transition coupled with the inexperience of both instructors and students in Vietnam threatens the quality of its educational system and performance. Therefore, there is an urgent need for research on different designs in online education to ensure the quality of that education, and thus progress towards SGD-4 in the country.

Although there remain numerous problems, meta-analyses reveal that online learning can provide viable alternatives to face-to-face instruction [1]. In Vietnam, the most common strategies for delivering online lectures are synchronous online education (e.g., live online lectures via Google Meet, Zoom, Microsoft Teams, etc.) and asynchronous online education (e.g., pre-recorded lectures, email exchanges, forum discussion, etc.). Nevertheless, scientific studies exploring the effects of these two approaches on students’ academic achievement are still limited, and the results are inconclusive thus far. For example, Kubey, Lavin, and Barrows [2] discovered that students receiving synchronous online education tend to have lower academic achievement than those receiving asynchronous online education. The superior amount of content obtained via asynchronous approaches, according to Perera and Richardson [3], is the main factor contributing to better performance on the final examination. Somenarain et al. [4], on the other hand, demonstrate that synchronous online education is more successful in improving students’ conceptual comprehension. In addition, students evaluated synchronous virtual classrooms more highly because of the convenience and quality of live discussions being delivered [5]. Several studies, such as the works of Love and Fry [6] and Wells et al. [7], uncover that students see synchronous online education as only a kind of tutor section, thus averse to engaging in synchronous online classes and worsening academic achievement. There are also studies documenting that both synchronous and asynchronous online education are equally helpful in improving students’ conceptual knowledge and academic achievement [8].

In this paper, we contribute to the literature by comparing the effectiveness of live online (synchronous) and pre-recorded (asynchronous) lectures on students’ academic achievement using a randomized experiment in a Vietnamese university. We conducted a randomized experiment in which first-semester second-year university students attending live online lectures were randomly selected to be taught via the pre-recorded lectures for their mandatory courses in economics. Moreover, being taught via the pre-recorded lectures was also randomized within, and not only across, students. In other words, a given student was taught via the pre-recorded lectures in only some random weeks during the semester. In addition, we connected all of the exam questions to the weeks of the semester in which the content needed to answer them was taught via either online live or pre-recorded lectures. As a result, we can take advantage of the differences in question scoring both between students and within students (i.e., weeks with online live lectures vs. weeks with pre-recorded lectures for a given student). Our findings not only provide additional evidence to the debate on the relative effectiveness of synchronous and asynchronous online education, but also offer meaningful implications to policymakers, especially in the midst of the COVID-19 pandemic.

Our empirical analysis revealed that pre-recorded lectures reduce lower-ability students’ academic achievement but have no effect on higher-ability students’ academic achievement. In particular, being taught via pre-recorded lectures as opposed to live online lectures decreased the likelihood of answering exam questions correctly by 1.6 percentage points for students in the bottom 50th percentile of the ability distribution (measured by GPA at the beginning of the semester). Furthermore, being taught via pre-recorded lectures in the starting weeks of the semester compared to the later weeks tended to be more harmful to students’ academic achievement. Specifically, being taught via the pre-recorded lectures as opposed to live online lectures in the early, middle, and late semester decreased the likelihood of answering exam questions correctly by 1.1, 0.9, and 0.8 percentage points, respectively.

A possible explanation for our findings is that in crowded classes (such as those in our experiment, which ranged from 55 to 83 students per instructor), interactions with the instructor could be limited, and the benefits of being taught via the live online lectures might only be marginally higher than the benefits of being taught via the pre-recorded lectures, especially for the students who can possibly grasp much of the material on their own. The remaining students, on the other hand, might find it more difficult to grasp the content without asking instructors for more explanations during live online lectures. In other words, self-study from pre-recorded lectures might be less effective when compared to attending live online lectures for these students. In addition, providing pre-recorded lectures in the starting weeks of the semester could be more harmful because the first few weeks may set the tone for the rest of the semester in many aspects. For example, getting used to the course knowledge early might be helpful in digesting the later course contents that are increasingly difficult and require previously studied knowledge. Furthermore, students often make early decisions about whether or not they will enjoy the course, its material, the instructor, and their peers during the first few weeks.

The heterogeneity in the impacts of providing pre-recorded lectures, as opposed to the live online ones, on students of high and low abilities are a discovery that may readily guide the design of distance learning education policies. For instance, universities concerned about their students’ academic achievement might gain from our research that only offering pre-recorded lectures could deepen the academic inequality between students with higher and lower abilities. Under limited resources (such as bad internet connection and crowded classes), a better solution for distance learning could be to recommend pre-recorded lectures for the students with high GPAs and live online lectures for those with low GPAs. Furthermore, when considering the combined option of live online and pre-recorded lectures, it would be better for students’ academic achievement to schedule live online lectures early in the semester.

2. Literature Review

Synchronous online education is real-time communication between instructors and students when they are geographically distant but online at the same time to receive instantaneous responses from each other [9]. This type of instructional approach is often delivered through real-time communication technologies, such as instant messaging and online video/audio conferences [10]. Asynchronous online education refers to delayed communication over a period of time between instructors and students. This type of instructional approach usually relies on text-based or recording technologies such as online forums and e-mails [10].

Both synchronous and asynchronous online education have their own advantages and disadvantages. Synchronous online education allows students to discover social indicators such as body language and facial expression during the conversation, which can lead to the establishment of social bonds and improve students’ participation in interactions [11]. Miscommunications can also be reduced by instantaneous questions and answers during synchronous online education. In addition, synchronous online education can save time for both instructors and students since the discussions happen simultaneously, i.e., repeated questions and answers are significantly reduced.

Asynchronous online education, on the other hand, might make the learning atmosphere more comfortable by allowing for greater freedom in communicating time and mode. Students will have more time to enjoy the lectures and formulate their comments. As a result, exchanges between instructors and students in asynchronous online education are often more relevant and meaningful [12]. This advantage of asynchronous online education is particularly beneficial to introvert, hesitant, or language-challenged students [13]. Nevertheless, asynchronous online education has obstacles such as required commitment from the student and inefficient communication caused since exchanges are delayed [14].

Thus far, the majority of empirical research directly comparing synchronous and asynchronous online education has only focused on students’ perceptions. For example, Chundur and Prakash [15], Rockinson-Szapkiw and Wendt [16], and Peterson et al. [11] find that students prefer synchronous online education, while Griffiths and Graham [17] and Buxton [18] discover the opposite. Some studies further document that students are indifferent between attending synchronous and asynchronous online education [10,19]. The primary cause of why students prefer one approach to the other in these studies is mostly related to the advantages of each approach, such as the interaction of synchronous online education and the convenience of asynchronous online education.

Nevertheless, scientific studies exploring the effects of these two approaches on students’ academic achievement are still limited, and the results are inconclusive thus far. For example, Kubey, Lavin, and Barrows [2] discovered that students receiving synchronous online education tend to have lower academic achievement than those receiving asynchronous online education. The superior amount of content obtained via asynchronous approaches, according to Perera and Richardson [3], is the main factor contributing to better performance on the final examination. Somenarain et al. [4], on the other hand, demonstrate that synchronous online education is more successful in improving students’ conceptual comprehension. Several studies, such as the works of Love and Fry [6] and Wells et al. [7], uncover that students see synchronous online education as only a kind of tutor section, thus averse to engaging in synchronous online classes and worsening academic achievement. There are also studies documenting that both synchronous and asynchronous online education are equally helpful in improving students’ conceptual knowledge and academic achievement [8].

In this paper, we contribute to this growing literature by comparing the effectiveness of live online (synchronous) and pre-recorded (asynchronous) lectures on students’ academic achievement in a Vietnamese university. Our findings not only provide additional evidence to the debate on the relative effectiveness of synchronous and asynchronous online education, but also offer meaningful implications to policymakers, especially in the midst of the COVID-19 pandemic.

3. Methodology

During the Fall semester of 2021 at Ho Chi Minh City Open University, we delivered the live online lectures in two mandatory courses of economics (Principles of Macroeconomics and Principles of Microeconomics) for the bachelor’s program via Google Meet (the default platform of the university). Depending on their major, students are required to take the subject in either Vietnamese or English.

Each subject-by-language was delivered by a different instructor. All classes of the same subject, regardless of the delivered language, were completely harmonized. In other words, the course contents, lecture notes, exercises, and homework are exactly the same in all classes of the same subject (only translated into Vietnamese or English depending on the language requirement). Furthermore, the students having the same subject were given the identical exam (again, only translated into the two languages). Students were registered in a class at the beginning of the semester and were not allowed to change due to the university policy.

All the live online lectures were delivered and recorded via Google Meet. These recorded lectures were then used as the pre-recorded lectures. The pre-recorded lectures were accessible 24 h after the live online lecture. The delay was unavoidable because we needed to process the recorded live online lectures and upload them to the school platform. Students accessed the pre-recorded lectures via the school platform using their usual university accounts.

At Ho Chi Minh City Open University, a semester consists of 10 weeks/sessions. Each session lasts 270 minutes (i.e., 4.5 h). In the first week, the instructors of these classes introduced the experiment to the students. Then, they were given one week to register for participation in the experiment. Overall, 154 out of 552 students decided to participate in the experiment. Here it is important to note that there could be bias from inviting voluntary participation. To put it differently, if the students who did not choose to participate in the experiment had different views from the volunteers and such differences were correlated with examination performance, the estimations in the later sections could be biased. In

Table 1. Classes and students involved in the experiment.

Class	Language	Class Size	Participation
Principles of Macroeconomics	English	55	16
Principles of Macroeconomics	English	55	12
Principles of Macroeconomics	Vietnamese	83	24
Principles of Macroeconomics	Vietnamese	83	31
Principles of Microeconomics	English	55	17
Principles of Microeconomics	English	55	15
Principles of Microeconomics	Vietnamese	83	23
Principles of Microeconomics	Vietnamese	83	16

, we summarize the participating classes, the language of instruction, the class size (i.e., the number of students), and the number of students participating in the experiment (Participation).

Based on the participation lists of each class, we proceeded to randomly divide the students into three groups. The first group, Never, consisted of 31 students (20% of the experiment-participating students) that had not been taught via pre-recorded lectures at any point during the semester. The second group, Always, consisted of 31 students (i.e., another 20%) that had been taught entirely via pre-recorded lectures in all weeks throughout the semester. The third group, Sometimes, consisted of the remaining 92 students (i.e., the remaining 60%) who had been taught via pre-recorded lectures in only some random weeks. Being taught via the pre-recorded lectures was randomized not only across, but also within, students. We did so by randomly assigning 50% of the Sometimes students to be taught via the pre-recorded lectures for each individual week. This additional layer of randomization allowed us to take advantage of the differences in academic achievement both between students and within students (i.e., weeks with online live lectures vs. weeks with pre-recorded lectures for a given student). At the end of the first week, the experiment-participating students were provided the exact schedule for the semester (i.e., which weeks were being taught via which modes). The experiment started in the second week until the end of the semester.

We also collected information of the experiment-participating students on individual characteristics such as gender, age, GPA, residential areas, and their performance in the final exams of the experiment-semester. All of the classes included in the experiment featured multiple-choice exams with a total of 40 questions in 60 minutes. More importantly, we designed the exam questions and answers in a matrix format that precisely mapped to individual weeks. This matrix detailed the week and the lecture mode in which the material required to answer a given exam question was given.

Our final sample consisted of 154 students. In

Table 2. Summary statistics.

	All	Never	Sometimes	Always
	(N = 154)	(N = 31)	(N = 92)	(N = 31)
Female	0.601	0.613	0.603	0.585
	(0.491)	(0.489)	(0.491)	(0.495)
Age	20.118	19.825	20.169	20.259
	(2.228)	(2.063)	(2.197)	(2.493)
GPA	3.347	3.397	3.316	3.394
	(1.865)	(1.933)	(1.846)	(1.885)
Mother, High School	0.373	0.388	0.376	0.349
	(0.479)	(0.484)	(0.478)	(0.471)
Father, High School	0.444	0.441	0.443	0.451
	(0.494)	(0.494)	(0.493)	(0.497)
Exam Grade	0.001	−0.002	0.006	−0.009
	(0.939)	(0.870)	(0.958)	(0.923)
Share of Correct Answers	0.644	0.656	0.641	0.644
	(0.203)	(0.198)	(0.203)	(0.209)

, we provide the summary statistics of the experiment-participating students. Approximately 60.1% of them were female. Their average age was 20.118, and their GPA at the beginning of the semester was 3.347 (out of 4). In addition, 37.3% and 44.4% of them had a mother and a father who had completed high school, respectively. The final exam grades in all classes were in a 10-point scale and were standardized within each class to have a mean of zero and a variance of one. Note that the numbers of experiment-participating students differed across classes; thus, the overall mean is 0.001, and not exactly zero.

Next, using the questions–answers matrix mentioned previously, we computed our main outcome, which was the share of correct answers to the exam questions related to the material covered at the class-week level (Share of Correct Answers). If only one question was related to the material of the week, the Share of Correct Answers took the value of one if that answer was correct, and zero otherwise. If two or more questions were related, the Share of Correct Answers equaled the total number of correct answers divided by the total number of related questions. On average, the experiment-participating students correctly answered 64.4% of the questions. We also provided statistics by the assignment groups. Consistent with randomization, differences in the statistics across the assignment groups were small in magnitude and statistically insignificant.

4. Results

In this section, we discuss the estimating strategy and provide the estimated impacts of being taught via pre-recorded lectures as opposed to live online lectures on students’ performance on the exams. To do so, we estimated the following regression equation:

$$Y_{icw}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{Treated}_{icw}+{\mathsf{\delta }}_i+{\mathsf{\theta }}_c+{\mathsf{\lambda }}_w+{X’}_{icw\mathsf{\Omega }}+{\in }_{icw}$$

(1)

where the subscripts i, c, and w correspond to student, class, and week, respectively. The main outcome variable, $Y_{icw}$, is the Share of Correct Answers, as discussed above. The main explanatory ${Treated}_{icw}$ is a zero-one indicator taking a value of one if the referred student of a given class was randomly assigned to be taught via the pre-recorded lectures in the referred week. The covariate $X_{icw}$ denotes a set of individual characteristics including age, squared-age, sex, GPA, maternal education, paternal education, and the fixed effects of the birth district, current district, as well as enrolment cohorts. We also denote by ${\mathsf{\delta }}_i$, ${\mathsf{\theta }}_c$, and ${\mathsf{\lambda }}_w$ individual, class, and week fixed effects, respectively. Finally, ${\in }_{icw}$ is the error term.

The estimating results are reported in

Table 3. Main results.

	Y = Share of Correct Answers
	(1)	(2)	(3)	(4)
Treated	−0.018 ***	−0.012 ***	−0.012 ***	−0.010 ***
	(0.005)	(0.004)	(0.004)	(0.003)
Observations	4968	4968	4968	4968
Student-Fixed Effects				X
Class- and Week-Fixed Effects			X	X
Characteristics		X	X	X

Note: *** p < 0.010. Characteristics include age, squared-age, sex, GPA, maternal education, paternal education, and the fixed effects of the birth district, current district, as well as enrolment cohorts.

. Column 1 displays the estimate from the most parsimonious specification in which we controlled only for the main explanatory variable, Treated. In Column 2, we added the set of individual characteristics to the most parsimonious specification. In Column 3, we further controlled for the class and week-fixed effects. Finally, Column 4 represents the most extensive specification in which we accounted for student-fixed effects in addition to the previous controls.

According to Column 1, we find that being taught via the pre-recorded lectures, as opposed to the live online lectures, decreased the likelihood of answering exam questions correctly by 1.8 percentage points. However, this estimate from the most parsimonious specification only reflects the correlation between pre-recorded lectures and students’ academic achievement, as important factors are not accounted for. Therefore, from Column 2 to Column 3, we gradually added the individual characteristics (age, squared-age, sex, GPA, maternal education, paternal education, and the fixed effects of the birth district, current district, as well as enrolment cohorts) and class as well as week-fixed effects, and the estimate becomes a little smaller in magnitude but remains statistically significant.

Finally, according to our most extensive specification in Column 4 in which the student-fixed effects model was utilized, we find that being taught via the pre-recorded lectures as opposed to the live online lectures decreased the likelihood of answering exam questions correctly by 1.0 percentage points. The estimate is statistically significant at the 1% level. In addition, since the total number of questions given in the final exam was 40, the treatment effect was equivalent to 0.4 additional wrong answers for the students.

So far, we detected unfavorable effects of being taught via pre-recorded lectures as opposed to the live online lectures on student performance on the exam. We then proceeded to explore the heterogeneous effects along the lines of student ability and the timing of the delivery of the pre-recorded lectures. First, we divided the students into two subsamples: (i) the Lower Ability subsample, including the students with GPAs below the 50th percentile, and (ii) the Higher Ability subsample, including those with GPAs above the 50th percentile. We then estimated the regression equation (Equation (1)) separately for each subsample.

The estimating results derive from the most extensive specification (similar to Column 4 in Table 3) and they are reported in

Table 4. Heterogeneity in ability.

	Y = Share of Correct Answers
	Low Ability	High Ability
	(1)	(2)
Treated	−0.016 *** (0.004)	−0.003 (0.003)
Observations	2502	2466
Student-Fixed Effects	X	X
Class and Week-Fixed Effects	X	X
Characteristics	X	X

Note: *** p < 0.010. Characteristics include age, squared-age, sex, GPA, maternal education, paternal education, and the fixed effects of the birth district, current district, as well as enrolment cohorts.

. Interestingly, we discovered that pre-recorded lectures reduced lower-ability students’ academic achievement but had no effect on higher-ability students’ academic achievement. In particular, being taught via the pre-recorded lectures as opposed to the live online lectures decreased the likelihood of answering exam questions correctly by 1.6 percentage points for students in the bottom 50th percentile of the ability distribution (measured by GPA at the beginning of the semester). The estimate was small and statistically insignificant for the subsample of higher-ability students. Since the total number of questions was 40, the treatment effect was equivalent to 0.6 additional wrong answers for the lower-ability students.

A possible explanation for the findings is that in crowded classes (such as those in our experiment, which ranged from 55 to 83 students per instructor), interactions with the instructor could be limited, and the benefits of being taught via live online lectures might only be marginally higher than the benefits of being taught via pre-recorded lectures, especially for students who can possibly grasp much of the material on their own. The remaining students, on the other hand, might find it more difficult to grasp the content without asking instructors for more explanations during live online lectures. In other words, self-study from pre-recorded lectures might be less effective when compared to attending live online lectures for these students.

Next, we further explored the relative importance of the delivery timing of the pre-recorded lectures. Specifically, we wanted to know which period for the delivery of pre-recorded lectures had the largest impacts on student achievement in the exam. To do so, we replaced the single indicator Treated in Equation (1) with three indicators, namely, Treated First 3 Weeks, Treated Second 3 Weeks, and Treated Third 3 Weeks, which indicated whether the student was taught by the pre-recorded lectures in the first, second, or third 3 weeks of the experiment period, respectively. The estimating results derived from the most extensive specification (similar to Column 4 in Table 3) and they are reported in

Table 5. Heterogeneity in treatment time.

	Y = Share of Correct Answers
	(1)
Treated First 3 Weeks	−0.011 ***
	(0.003)
Treated Second 3 Weeks	−0.009 **
	(0.004)
Treated Third 3 Weeks	−0.008 **
	(0.003)
Observations	4968
Student-Fixed Effects	X
Class and Week-Fixed Effects	X
Characteristics	X

Note: ** p < 0.050, *** p < 0.010. Characteristics include age, squared-age, sex, GPA, maternal education, paternal education, and the fixed effects of the birth district, current district, as well as enrolment cohorts.

.

Overall, we find that providing the pre-recorded lectures in any of the three 3-week periods compared to the live online lectures was detrimental to student achievement. However, we find that being taught via the pre-recorded lectures affected students’ academic achievement the most in the first 3 weeks by decreasing the likelihood of answering exam questions correctly by 1.1 percentage points (i.e., more than 0.4 additional wrong answers). The impact was slightly lower in the second 3 weeks, with a reduction in the likelihood of answering exam questions correctly by 0.9 percentage points (i.e., almost 0.4 additional wrong answers). Lastly, delivering the pre-recorded lectures in the last 3 weeks of the semester had the smallest effect, with a decrease of 0.8 percentage points in the likelihood of answering exam questions correctly (i.e., 0.3 additional wrong answers).

A potential explanation for this pattern is that the first few weeks may set the tone for the rest of the semester in many aspects. For example, getting used to the course knowledge early might be helpful in digesting the later course contents that are increasingly difficult and require previously studied knowledge. Furthermore, students often make early decisions about whether or not they will enjoy the course, its material, the instructor, and their peers during the first few weeks.

5. Conclusions

Many universities across the globe have been pushed to make a rapid transition from traditional face-to-face instruction to online classes due to the outbreak of COVID-19. The most common strategies for delivering online lectures are synchronous online education (e.g., live online lectures via Google Meet, Zoom, Microsoft Teams, etc.) and asynchronous online education (e.g., pre-recorded lectures, email exchanges, forum discussion, etc.). Nevertheless, scientific studies exploring the effects of these two approaches on students’ academic achievement are still limited, and the results are inconclusive thus far. In this paper, we contribute to the literature by comparing the effectiveness of live online (synchronous) and pre-recorded (asynchronous) lectures on students’ academic achievement using a randomized experiment in a Vietnamese university.

We conducted a randomized experiment in which first-semester second-year university students attending live online lectures were randomly selected to be taught via pre-recorded lectures. Being taught via pre-recorded lectures was randomized not only across, but also within, students in our experimental design. As a result, we were able to take advantage of the differences in question scoring both between students and within students. Our empirical analysis reveals that pre-recorded lectures reduce lower-ability students’ academic achievement but have no effect on higher-ability students’ academic achievement. In particular, being taught via the pre-recorded lectures as opposed to live online lectures decreased the likelihood of answering exam questions correctly by 1.6 percentage points for students in the bottom 50th percentile of the ability distribution (measured by GPA at the beginning of the semester). Furthermore, being taught via the pre-recorded lectures in the starting weeks of the semester compared to the later weeks tended to be more harmful to students’ academic achievement.

The heterogeneity in the impacts of providing pre-recorded as opposed to the live online lectures on students of high and low abilities is a discovery that may readily guide the design of distance learning education policies. For instance, universities concerned about their students’ academic achievement might gain from our research that only offering pre-recorded lectures could deepen the academic inequality between students with higher and lower abilities. Under limited resources (such as bad internet connection and crowded classes), a better solution for distance learning could be to recommend pre-recorded lectures for the students with high GPAs and live online lectures for those with low GPAs. Furthermore, when considering the combined option of live online and pre-recorded lectures, it would be better for students’ academic achievement to schedule live online lectures early in the semester.

Our work also contributes to a growing body of knowledge about the relationship between alternative instructional modes and learning processes. For instance, some studies document that students enrolled in Massive Open Online Courses (MOOCs) have higher dropout rates than students enrolled in traditional courses [20]. Others have shown that flipped classes (involving a teaching method in which students watch pre-recorded videos of lectures and use classroom time to discuss problem sets with instructors) disproportionally benefit students of higher ability [21]. Within online education, several studies examining the role of social presence in online classrooms have found that presenting the instructor’s face increases students’ learning outcomes and enjoyment [22]. However, others suggest that the potential benefit of eliciting social interactions by presenting the instructor’s face is offset by the extra visual processing caused by the face [23].

The heterogeneous effects of different modes of online teaching may extend beyond students’ academic achievement, since the decline in education could affect a variety of aspects such as earnings, health, and social capital, among others [24,25,26,27,28,29,30,31]. Affected individuals may transmit these socioeconomic disadvantages to the future generation, exacerbating social inequality and impeding economic development [32,33,34]. Therefore, our findings not only provide additional evidence to the debate on the relative effectiveness of synchronous and asynchronous online education, but also offer meaningful implications to policymakers, especially in the midst of the COVID-19 pandemic.