Learning From Home: Widening Rural-Urban Educational Inequality and High School Students’ Self-Control in China During the COVID-19 Pandemic and School Closure

Abstract

Worldwide school closures and remote learning have been implemented during the COVID-19 pandemic. These measures’ impact on young populations’ academic achievements is unclear. This study (N = 1,736, ages 14–20 years, 53% female, and Chinese) analyzed academic examination scores for students at a high school in Eastern China between January and July 2020. Results showed that overall, students’ academic achievements appeared to be negatively affected amid a school closure. More importantly, students’ self-control was introduced as a moderating factor that partially accounted for this difference in the context of remote learning at home. These findings extended our understanding of school closures’ unequal impact on young populations. Education and social policies should respond to these challenges during times of crisis.

Introduction

The closing of schools worldwide due to the COVID-19 pandemic were unprecedented. As of April 27, 2020, about 1.6 billion learners had been affected by school closures world-wide (United Nations, 2020). School measures were taken to ensure continuity of teaching and learning around the world. Nonetheless, the impact of these measures remains unclear. In a survey of 3,275 Chinese parents’ evaluation of online learning during the COVID-19 pandemic, Dong et al. (2020) found that parents generally held negative beliefs about the impact of online learning on their children’s education. Kuhfeld and colleagues estimated that most students in the United States returning from COVID-19 school closures were expected to experience a significant decline in their academic achievements (Kuhfeld et al., 2020). These exploratory studies prompted us to question the extent of the impact on children’s academic achievement and how young populations may be unequally affected.

We sought to examine the impact of COVID-19 related school closures and remote learning on young people’s development in the context of China. Using data collected from a high school during its closure and subsequent reopening at the initial stage of the pandemic, we aimed: 1) to describe the changes that took place in students’ academic achievement over time; 2) to explore the extent to which rural-urban inequality in education changed during this period of time; and 3) to extend a psycho-social explanation on the unequal effects of school closure and remote learning on rural and urban students’ education. We found that academic achievements of the Chinese high school students who participated in this study had been negatively affected by the school closure. Moreover, the school closure differentially affected urban and rural students’ academic achievements, leaving the latter at a disadvantage. Instead of using the usual socio-economic explanation, we found that students’ personal self-control partially accounted for the widening rural-urban inequality in education in this context. As such, we will elaborate on these arguments through a critical literature review.

Rural-Urban Disparity and Academic Achievement

Rural-urban disparities posed significant challenges to students’ academic achievements in China prior to COVID-19 (Chan, 2019). These disparities were known to be associated with the hukou system, which divides Chinese citizens into agricultural (rural) and non-agricultural (urban) statuses (Cheng & Selden, 1994). Despite the hukou reform in recent years, empirical studies have shown that the system still has a great impact on education. For example, Hannum (1999) identified a dual-tiered educational system that had been formed along urban-rural lines in the 1980s. In addition, Wu (2011, p. 48) used data from a national representative survey in 2005 to show that rural populations had long-standing and consistently shorter years of schooling. In a longitudinal analysis of post-reform periods of time, Yang et al. (2014) found that hukou status was the strongest predictor of children’s educational outcomes. Chinese scholars have argued that the hukou system should be viewed as one of the most far-reaching and enduring factors for understanding educational inequality in China’s context (Chan, 2019).

The pandemic may be exacerbating the enduring effects of rural-urban disparity on education. Recent educational reports by the United Nations have warned that COVID-19 may hinder the accumulation of human capital while simultaneously increasing educational inequality (United Nations, 2020). By using a projection of 5 million students in Grade 3 through Grade 7 across two school years (2017–2019), researchers estimated that all students returning to school after COVID-19-related school closures would suffer lower academic achievements (Kuhfeld et al., 2020). Furthermore, evidence from H1N1-related school closures in 2009 indicated that the educational performance of children from poor and vulnerable backgrounds was more likely to be affected (Cauchemez et al., 2009). However, scientific evidence on these effects remains scarce. To our knowledge, rarely has previous research examined the losses of learning during the COVID-19 pandemic, especially in the Chinese context. This paper aims to fill this knowledge gap by studying the impact of hukou status (a proxy for rural-urban disparity) on Chinese high school students’ academic achievement during the closures and reopening of schools.

Socio-Economic and Psycho-Social Explanations

Our second objective is of a theoretical nature, that is to explore explanations on the underlying mechanisms by which the rural-urban disparity affects students’ academic achievement, which shall be considered in the context of remote learning, or more specifically “learning from home” during the school closures. After the COVID-19 outbreak in late January 2020, the Chinese government suspended the opening of schools for the spring semester. At this time, national initiatives were launched to ensure that students’ education would continue amid school closures (Ministry of Education, China, 2020), including the wide adoption of online teaching and learning. China’s Ministry of Education established a National Cloud Platform in mid-February 2020 to deliver online education. By mid-May 2020, the platform had registered about 1.7 billion visits (Ministry of Education, China, 2020). However, switching to a learning-from-home method precipitated public concerns and heated debates about the implications of the new education modes on students’ learning experiences and outcomes.

A literature review suggests that both environmental and personal factors may affect academic achievement for students who learn from home. On the one hand, many empirical studies have demonstrated that socio-economic factors, such as family income, parental education, and occupation, are closely associated with students’ educational outcomes (Chung, 2015; Sirin, 2005; White, 1982). Compared with adolescents growing up in rich families, studies showed that adolescents from families with economic disadvantages tend to grow up with lower paternal knowledge, expectation, monitoring, discipline, and demandingness (Shek, 2007; Conger et al., 1999). Normally, school environment can serve a buffering role for the economically disadvantaged children. For example, Tan and Bodovski’s (2020) research indicated that boarding at school compensates for family disadvantages, especially for students with little parental support and involvement. However, during the pandemic, without the school environment, family socio-economic status (SES) may directly affect students’ access to and utilization of information resources at home, which can affect students’ academic achievements in turn (Cheshmehzangi et al., 2022).

Apart from environmental factors, student’s personal traits, especially self-control, may also be vital to their educational success, especially during times of school closure (Dong et al., 2020). Self-control is defined as “the self-initiated regulation of thoughts, feelings, and actions when enduringly valued goals conflict with momentarily more gratifying goals” (Duckworth et al., 2019, p. 374). Self-control has two features: it is self-motivated, and it requires a choice to be made between a valuable long-term goal and a less valuable short-term pleasure (Duckworth et al., 2019). Self-control may play critical roles in predicting academic achievement, because it is one of the most important traits to consider in relation to student’ independent and self-directed learning (Duckworth et al., 2012). For example, by designing the Brief Self-Control Scale (BSCS), Tangney et al. (2004) showed that self-control is closely related to academic achievement and psychosocial wellbeing. With a longitudinal study using the BSCS measure, Duckworth and Seligman (2005) demonstrated that self-control is more important than IQ in predicting academic performance of Grade 8 students. Duckworth et al. (2012) further explicated that high self-control helps students study and complete homework, and in conjunction with IQ, affects their academic achievement. Troll et al. (2020) found that smartphone use boosted academic achievement for students with higher self-control. Therefore, during school closures and online education, self-control can be crucial for understanding a student’s decision to study for the future rather than, for example, to play mobile phone games for immediate pleasure.

Based on the literature review above, we argue for a new theoretical framework and try to explore whether SES and psycho-social traits together can explain the effects of rural-urban disparity on students’ academic achievement. Our study will contribute to the literature in the following two ways: firstly, it is, to our knowledge, the first study to explore the mechanism underlying the widening rural-urban inequality in education in the context of the COVID-19 pandemic. The model integrating both SES and psycho-social traits, as a pioneering discovery, can enrich our understanding on youth development during school closure and online education. Secondly, given the difficulties in conducting fieldwork during the outbreak of the pandemic in 2020, our data are quite unique and provide us a rare opportunity to observe how rural-urban inequality in education changed during the crisis.

Methods

Data and Participants

For analyzing changes in academic achievements during the COVID-19 pandemic, a sample of high school students is desirable because these students take regularly scheduled examinations. By examining academic examination scores over time, we may know whether and how students’ academic achievements changed during the pandemic. For this purpose, we chose a county-level key high school (Zhongdian Gaozhong) in Eastern China and collected all attending students’ academic records during the pandemic. Like other key high schools, the institution under study receives funding from both the state and local levels of government. As a result, this school is not only equipped with more advanced facilities, but also attracts the most competitive teachers and students in the region.

Despite the non-representative sampling strategy, our sample bears similarities to other key high schools since every county has a key high school in China (Liu et al., 2020). With a gross domestic product (GDP) per capita of US$10,084 in 2020, the county, where the school is located, was at a moderate level of economic development approximating the national average. The school was closed due to the pandemic from late January 2020 to late April 2020 under an order from the State Department of Education applicable to all high schools nationwide. In late April, Level 3 (or Grade 12) students returned to the school, and so did Level 1 (Grade 10) and Level 2 students (Grade 11) in early May 2020.

We collected a three-point time series of students’ academic scores. The first time point was at the final examination of the Fall 2019 semester in January 2020, before the outbreak of the pandemic. The second time point was in May 2020 during the end-of-closure returning examinations for all students. The last time point was in July 2020, when the finals of the Spring 2020 semester were held. The tests were organized and administered by governmental educational bodies. The July test for the Level 3 students was the 2020 national college entrance exam. These scores were collected directly from the school.

To collect students’ information and to measure key variables for the present study, we surveyed all students shortly after the Level 3 students had completed their college entrance exams. We received 1,982 responses, covering all students in the school. Following a rigorous screening to find duplicate and invalid questionnaires, a total of 1,736 questionnaires (88%) were included in the current analysis. We conducted a missing data analysis and found that the overall distribution of the sample did not significantly change after removing the missing data. The study process complied with ethical standards for research involving human subjects and cleared ethics review processes at the lead author’s university. The survey questionnaire was reviewed and approved by the high school administration before being used. Prior to completing the survey, every student was informed by their teachers that participation was voluntary.

To prepare the data for analysis, we merged the survey data with the score data by matching the students’ IDs. We constructed a final balanced dataset for all students including students’ socio-demographic information, their personal and family resources related to learning from home, self-control, and the academic score data from the three time points. To protect the school’s and students’ anonymity, the dataset was completely anonymized.

Measures

The students’ academic achievements were measured by their academic scores in the subjects of Chinese, mathematics, English, and an integrated curriculum. The integrated curriculum represented the students’ selection of three subjects from the options of physics, chemistry, biology, history, political studies, or geography. We established three waves of students’ total academic scores by aggregating Chinese, mathematics, English, and the integrated curriculum scores. Standardized scores were used for the hierarchical linear modeling analysis.

We conducted interviews with teachers at the school to gain knowledge about the difficulty levels of the three waves of exams and learned that the May exams could be relatively easier, while the January and July exams were of standard difficulties. The lower difficulty level for the May tests tends to yield higher scores and thus would not invalidate our analysis. If the students’ scores were still lower in May in comparison to January and July, we would have stronger evidence to suggest that the school closure might have had a negative impact on the students.

Rural-urban disparity was captured by students’ reported hukou status. Within our sample, 1,262 students had an agricultural or rural hukou status, and 474 students had a non-agricultural or urban hukou status. We treated the rural students as the reference group.

Self-control was assessed with the BSCS. Developed by Tangney et al. (2004), the BSCS includes 13 items that are rated on a 5-point scale, ranging from 1 (not at all like me) to 5 (very much like me). The BSCS has high reliability with a Cronbach’s α of 0.83. We calculated the total BSCS score by aggregating all 13 items. Students whose total BSCS scores were higher than the average were categorized in the group of high self-control. Students whose total scores were equal to or lower than the average were classified in the group of low self-control. We used low self-control as the reference.

The control variables included gender, age, family structure, siblings, parental support, highest education and occupation levels of parents, family income, residential type, study room availability, access to a laptop or desktop computer, and quality of online class. For this study, we used two categories for gender, including male and female, with male acting as the reference. Age was measured by subtracting the participants’ birth years from 2020. Family structure was measured by whether students lived with one parent, both parents, or with other relatives. We used both parents as the reference. Siblings were measured by the number of brothers and sisters. Parental support was measured based on a question asking how often the student communicated with their parents during the school closure and was scored on a scale of 1 to 5, from very rarely to very often. Parents’ educational level was grouped into primary, junior middle, senior middle, university, and graduate levels. Parental occupation was divided into upper class, professional class, petty bourgeoisie, peasant class and unemployed according to the Erikson–Goldthorpe–Portocarero classification of occupation (Erikson & Goldthorpe, 2010). We controlled both the highest levels of parents’ education and highest level of parents’ occupation. Family yearly income was grouped into six categories in Chinese Yuan: below 10,000, 10,000 to 50,000, 50,000 to 100,000, 100,000 to 150,000, 150,000 to 200,000, and more than 200,000. Residential type was categorized into real estate in a city, real estate in a town, rural housing, rental apartment, factory dormitory, and others. Study room availability was assessed by asking whether a student had a dedicated study room at home. Students who had a study room acted as the reference group. Access to a laptop or desktop computer was measured by asking whether a student had such a device. Students who had a laptop or desktop computer acted as the reference group. We also controlled for the quality of online class by asking students whether the school should repeat classes to help them digest contents.

Analytical Strategy

We utilized Hierarchical Linear Models (HLM) and Change Score Models (CSM) to analyze the data. HLM enabled us to simultaneously explore within-person and between-person questions about changes. The within-student change of scores was regarded as the Level-1 submodel in the HLM. As shown in equation (1), $Y_{ij}$, which represents the value of total academic score for student i at time j, is a linear function of the months in which a student took an exam ($Mont{h}_{ij}$). This model assumes that any deviation from linearity observed within the sample data is a result of random measurement error (${\epsilon }_{ij}$). Between student variables were included in the Level-2 submodel. Equations (2) and (3) treat the intercept π0i and the slope π1i in equation (1) as the Level-2 outcomes that may be associated with the study’s predictors, hukou, SES, and self-control. For the present study, hukou status was included as the main predictor, and the other two as control variables. Each formula has its own residual term, ${\zeta }_{0i}\mathrm{and}{\zeta }_{1i}$, which permit one student to differ from others. The formulas for the HLM are as follows:

$$Y_{ij}={\pi }_{0i}+{\pi }_{1i}Mont{h}_{ij}+{\epsilon }_{ij}$$ (1)

$${\pi }_{0i}={\gamma }_{00}+{\gamma }_{01}Huko{u}_i+{\gamma }_{02}SE{S}_i+{\gamma }_{03}SEL{F}_i+{\zeta }_{0i}$$ (2)

$${\pi }_{1i}={\gamma }_{10}+{\gamma }_{11}Huko{u}_i+{\gamma }_{12}SE{S}_i+{\gamma }_{03}SEL{F}_i+{\zeta }_{1i}$$ (3)

HLM has certain limitations. It is highly complicated to interpret results when an interaction in the Level-2 submodel is included. To illustrate, if the interaction of hukou status and SES were added to equation (3), there would be multiple interactions in equation (1). Considering these complexities, we employed a relatively simple model to analyze the panel data. As shown in equation (4), a CSM assesses predictors of change in students’ academic scores between two points in time. This simple model enabled us to investigate whether score changes were related to the fixed characteristics of students. Here, $Y_{i2}and{Y}_{i1}$ are the total scores of student i at time points 2 and 1. In addition, $H_{i1}$ is the value of the hukou predictor, $S_{i2}$ is the value of the self-control predictor, $H_{i1}*S_{i2}$ is the value of interaction between hukou status and self-control, and $C_{i4}$ is the value of various controls. $e_i$ is an error term. Equation (4) is an unconditional change score model, as it assumes that the change is independent of the score at the first time given the predictor variable. This assumption may not hold in practice. Therefore, we included the total academic score at the earlier point in time and provided the conditional change model, which is shown in equation (5). To illustrate, we included the January score as a predictor for the period of school closure and the May score as a predictor for the period of school reopening. The formulas for this aspect of the modelling are as follows:

$$Y_{i2}-Y_{i1}={\beta }_0+{\beta }_1{H}_{i1}+{\beta }_2{S}_{i2}+{\beta }_3{H}_{i1}*S_{i2}+{\beta }_4{C}_{i4}+e_i$$ (4)

$$Y_{i2}-Y_{i1}={\beta }_0+{\beta }_1{H}_{i1}+{\beta }_2{S}_{i2}+{\beta }_3{H}_{i1}*S_{i2}+{\beta }_4{C}_{i4}+{\beta }_5{Y}_{i1}+e_i$$ (5)

Results

Descriptive Statistics

The total academic scores of the students at the observed high school changed during the COVID-19 pandemic. As shown in Table 1, the average total score for all high school students was about 514 in January 2020. However, the students’ average scores decreased dramatically during the school closure, falling about 44 points from January to an average 470 in May. After the reopening, when all the students had returned to school, the average score recovered to 501, falling short of the pre-closure levels recorded in January.

Table 1.

Descriptive Statistics (n = 1,736).

Variables	All children	Rural	Urban	Mean diff	p Value
Average total academic score in January	513.481 (145.341)	515.458 (146.934)	508.167 (140.990)	7.291	.357
Average total academic score in May	469.893 (126.306)	468.829 (126.667)	472.756 (125.423)	−3.927	.567
Average total academic score in July	500.521 (129.098)	499.576 (129.540)	503.047 (128.009)	−3.470	.620
Gender (female %)	0.533 (0.499)	0.542 (0.498)	0.508 (0.500)	0.034	.212
Mean age	17.187 (1.013)	17.204 (1.040)	17.143 (0.938)	0.060	.270
Mean grade	1.950 (0.832)	1.908 (0.835)	2.063 (0.815)	−0.155	.001***
Self-control (high self-control %)	0.423 (0.494)	0.421 (0.494)	0.430 (0.496)	−0.0100	.718
Family structure (both parents %)	0.423 (0.494)	0.465 (0.499)	0.312 (0.464)	0.153	.000***
Mean of siblings	2.249 (0.783)	2.364 (0.761)	1.943 (0.760)	0.421	.000***
Mean of parental support	3.507 (0.970)	3.498 (0.952)	3.530 (1.018)	−0.0310	.552
Mean of parents’ highest education level	3.753 (1.509)	3.375 (1.243)	4.762 (1.683)	−1.387	.000***
Mean of parents’ highest occupation level	3.866 (0.966)	4.044 (0.756)	3.390 (1.258)	0.654	.000***
Mean of family income	2.711 (1.122)	2.616 (1.085)	2.964 (1.180)	−0.348	.000***
Mean of house type	2.821 (1.250)	2.997 (1.194)	2.354 (1.278)	0.642	.000***
Study room (own a room %)	0.530 (0.499)	0.572 (0.495)	0.418 (0.494)	0.154	.000***
Digital device (own a laptop or desktop computer %)	0.442 (0.497)	0.494 (0.500)	0.304 (0.460)	0.191	.000***
Mean of quality of online class	2.262 (0.969)	2.253 (0.959)	2.285 (0.995)	−0.0320	.540
Sample	1,736	1,262	474

⁺p < .10. *p < .05. **p < .01. ***p < .01 (two-tailed)

The changes in total academic scores during the COVID-19 pandemic were different for rural and urban students. Table 1 shows that in January, the average total score of students with a rural hukou status was 516, which was relatively higher than the average total score of urban students at 508. During the school closure, while rural students’ average total score decreased to 469, urban students’ average total score decreased to 473. After the school reopening, the urban students maintained a minor advantage until the end of semester. t-Tests for each of the three waves showed no significant differences between the rural and urban students’ total scores. This result indicated that the hukou factor might be primarily related to the changes in academic scores over time. We then investigated this question using HLM.

Hierarchical Linear Modeling

Table 2 shows the analysis results using the HLM method. Growth models A1 and A2 represented the academic score change of each student during the COVID-19 pandemic. We paid closer attention to Model A2 because its slope was the most important parameter in the Level-1 submodel. Model A2 showed that students’ scores were negatively associated with time (${\pi }_{0i}=0.100;{\pi }_{1i}=-0.024$). The slope π1i revealed that as student i moved one-point time, the total score dropped by $0.024$. This result indicated that academic achievements of the high school students declined during the COVID-19 pandemic over time.

Table 2.

Hierarchical Linear Models (n = 1,736).

Variables	Level-1 model		Level-2 model
	A1	A2	B1	B2	B3
Intercept	−0.004	0.100***	0.091***	0.120***	−0.202
Month		−0.024*** (−0.002)	−0.024*** (−0.002)	−0.028*** (−0.002)	−0.027 (−0.016)
Hukou			0.032 (−0.05)	−0.071 (−0.055)	−0.173** (−0.066)
Month × Hukou				0.015** (−0.005)	0.018*** (−0.005)
Self-control					0.004 (−0.054)
Month × Self-Control					0.017*** (−0.004)
Income					0.072** (−0.024)
Month × Income					−0.003+ (−0.002)
Parents’ education					0.050* (−0.022)
Month × Education					−0.001 (−0.002)
Parents’ occupation					−0.009 (−0.033)
Month × Occupation					0.001 (−0.003)
Variance (month)		0.003 (0.000)	0.003 (0.000)	0.000 (0.000)	0.003 (0.000)
Variance (intercept)	0.867 (−0.031)	1.106 (−0.043)	1.107 (−0.043)	0.871 (−0.031)	1.091 (−0.042)
Variance (residual)	0.132 (−0.003)	0.099 (−0.003)	0.099 (−0.003)	0.123 (−0.005)	0.099 (−0.003)

Note. Sample size = 1,736. Standard deviation in parentheses.

⁺p < .10. *p < .05. **p < .01. ***p < .01 (two-tailed).

Models B1 and B2 showed the structural parts of the Level-2 submodel for between-individual differences in academic score change. Although both the intercept and slope of month in Model B1 remained significant (${\pi }_{0i}=0.091;{\pi }_{1i}=-0.024$), the coefficient for hukou status on the total score was not significant (${\gamma }_{11}=0.032$). This result confirmed the findings from the t-tests in Table 1. However, as shown in Model B2, the interaction between month and hukou status was positive and significant (${\gamma }_{11}=0.015$). This means that while rural students had higher initial scores than urban students before the pandemic, their scores decreased more than those of their urban peers did when learning from home during the pandemic.

Model B3 extended the analyses by including the variables of family income, self-control, parental education, and parental occupation. According to our theoretical model, we predicted that the widening rural-urban inequality in education during the pandemic could be attributed to both SES and psycho-social traits. However, the results seemed to reject our prediction about SES. According to Model B3, the variables of family income, parental education, and parental occupation showed no significant association with students’ academic score changes. In the meantime, self-control was significantly associated with the changes in total academic score. It indicated that the rural-urban inequality in academic score changes was not attributable to students’ SES, but self-control. The following section attempts to explore these findings in further details.

Change Score Models

Table 3 shows the analysis results of the CSMs. As described in the methods section, we divided the 7-month time span into three periods: January to July (the entire period), January to May (the school closure period), and May to July (the school reopening period). Model C1 and C2 showed the coefficients for the predictors between January and July. According to Model C1, the coefficient for self-control on the change score was significant and positive, indicating that students with high self-control had a 4.697-point higher increase in total academic score than students with low self-control from January to July.

Table 3.

Change Score Models (n = 1,736).

Variables	Change score from January to July			Change score from January to May			Change score from May to July
	Model C1	Model C2	Model C3	Model D1	Model D2	Model D3	Model E1	Model E2	Model E3
Hukou	3.857 (3.25)	−2.548 (4.083)	−2.544 (4.077)	3.857 (3.250)	−2.209 (3.695)	−2.210 (3.683)	0.985 (2.820)	0.740 (3.553)	0.743 (3.555)
Self-control	4.697* (2.672)	0.643 (3.095)	0.055 (3.099)	4.697* (2.672)	−2.102 (2.809)	−1.390 (2.808)	3.640 (2.317)	3.485 (2.691)	3.475 (2.695)
Hukou × Self-Control		15.087*** (5.841)	15.177*** (5.832)		14.009*** (5.294)	13.911*** (5.278)		0.576 (5.075)	0.564 (5.079)
Gender	−2.794 (2.664)	−3.072 (2.662)	−2.551 (2.666)	−2.794 (2.664)	0.580 (2.413)	−0.048 (2.412)	−3.193 (2.313)	−3.204 (2.316)	−3.192 (2.322)
Age	−1.166 (2.253)	−0.991 (2.25)	−0.568 (2.252)	−1.166 (2.253)	−2.958 (2.036)	−3.463* (2.036)	1.033 (1.948)	1.040 (1.950)	1.052 (1.957)
Grade	61.386*** (2.746)	61.034*** (2.745)	64.894*** (3.133)	61.386*** (2.746)	51.519*** (2.481)	46.877*** (2.829)	10.794*** (2.380)	10.781*** (2.384)	10.832 (2.481)
Family structure	−1.733 (2.686)	−1.938 (2.683)	−1.800 (2.679)	−1.733 (2.686)	0.232 (2.433)	0.078 (2.426)	−2.494 (2.330)	−2.503 (2.332)	−2.501 (2.333)
Siblings	1.671 (1.796)	1.7 (1.793)	1.721 (1.790)	1.671 (1.796)	1.458 (1.625)	1.434 (1.620)	0.454 (1.560)	0.455 (1.561)	0.454 (1.561)
Parents’ education	−1.451 (1.12)	−1.506 (1.118)	−1.709 (1.119)	−1.451 (1.120)	−0.354 (1.015)	−0.106 (1.015)	−1.258 (0.968)	−1.260 (0.969)	−1.264 (0.971)
Parents’ occupation	−1.332 (1.6)	−1.168 (1.598)	−1.311 (1.597)	−1.332 (1.600)	−0.256 (1.451)	−0.066 (1.448)	−0.719 (1.384)	−0.712 (1.386)	−0.715 (1.387)
Family income	−1.489 (1.24)	−1.444 (1.238)	−1.610** (1.237)	−1.489 (1.240)	−1.454 (1.124)	−1.260 (1.122)	−0.456 (1.078)	−0.455 (1.078)	−0.458 (1.080)
Parental support	3.113** (1.39)	3.032** (1.388)	3.051 (1.386)	3.113** (1.390)	1.807 (1.258)	1.777 (1.254)	0.793 (1.203)	0.790 (1.204)	0.788 (1.204)
House type	0.191 (1.118)	0.080 (1.117)	−0.029 (1.116)	0.191 (1.118)	−0.297 (1.015)	−0.168 (1.012)	0.204 (0.966)	0.199 (0.967)	0.196 (0.968)
Study room	1.459 (2.791)	1.383 (2.786)	1.418 (2.782)	1.459 (2.791)	−3.603 (2.523)	−3.666 (2.516)	4.230* (2.423)	4.227 (2.424)	4.230 (2.425)
Digital device	−4.627* (2.797)	−4.56* (2.793)	−4.151 (2.793)	−4.627* (2.797)	−5.407** (2.535)	−5.875** (2.531)	0.926 (2.427)	0.929 (2.428)	0.944 (2.437)
Quality of online class	2.945** (1.346)	3.024** (1.344)	3.001** (1.342)	2.945** (1.346)	1.861 (1.217)	1.891 (1.214)	0.699 (1.167)	0.703 (1.168)	0.701 (1.168)
Total academic score in January			0.033** (0.013)			−0.040*** (0.012)
Total academic score in May									0.001 (0.011)
Constant	−117.699*** (36.77)	−118.315***	−148.487*** (38.523)	−117.699*** (36.77)	−95.208*** (33.233)	−59.033* (34.816)	−6.975 (31.814)	−6.996 (31.824)	−7.629 (32.957)
R 2	.484	.486	.488	.484	.427	.431	.054	.054	.054

Note. Standard deviation in parentheses.

⁺p < .10. *p < .05. **p < .01. ***p < .01 (two-tailed).

Model C2 tested how self-control moderated the effect of hukou status on the score change. The coefficient for the interaction between self-control and hukou status was significant and positive. Among high self-control students, those with an urban hukou status had a 12.539-point higher gain than rural students, while, among low self-control students, those from rural areas scored 2.548 points higher than their urban peers. Model C3 was a conditional change model that included the total score in January as a control variable. The result was similar to Model C2. Figure 1 illustrates Model C3.

Figure 1.

Interaction effect of hukou status and self-control on academic score change between January and July.

We also analyzed the change score during the school closure from January to May. In Model D1, the effect of self-control was still positive and significant ($\beta =4.697$). Model D2 illustrated the interaction between hukou status and self-control. The significant coefficient meant that for high self-control students, those with an urban hukou status had a 11.800-point higher gain than rural students; for low self-control students, those from rural areas were scored 2.209 points higher than their urban peers. While Model D3 included the January score as a control variable, the result was nearly the same.

Model E1 and E2 described the change score from May to July, when the school was reopened to students. In consistency with our prediction, Model E1 showed that the effect of hukou status was not significant. Model E2 indicated that the interaction between hukou status and self-control was not significant either. After the control variable of the May score was added to Model E3, the result was nearly the same. It suggested that when the students returned to school, the effect of self-control could begin to weaken or even disappear.

Discussion

The students in this study were spared from any COVID-19 infection due to stringent measures taken to contain the spread of the virus. Nonetheless, the school closure might negatively impact their cognitive development. In addition, these consequences might be unequally distributed among the student populations. We explored three critical research questions with the aim of uncovering the relationship between school closures and educational inequality. Based on a field study of a high school in Eastern China, we generated three important findings: first, that academic achievements of the high school students declined during the closure and reopening of the school due to the COVID-19 pandemic; second, that changes in academic achievements were not equally distributed among rural and urban students; and third, that psycho-social traits, rather than SES, explained the rural-urban inequality in education during the crisis. These findings will be interpreted and discussed below.

The results revealed a decreasing trend in students’ academic test scores between January and July, especially in the period of school closure from January to May. In addition, urban students tended to be more adaptable to the pandemic’s impact on learning, especially for those with high self-control. These findings resonated with extant research in other countries. For example, Aucejo et al.’s (2020) survey of 1500 American students showed that COVID-19 negatively affected educational experiences and expectations, and that the low-income groups disproportionately bear more economic and health repercussions of the COVID-19 pandemic. Engzell et al. (2021), using about 350,000 samples in Netherland, found that students made little, or no progress when they learned from home. Moreover, they showed that the learning losses were even larger for those students from less-educated homes or weaker infrastructure. Therefore, these findings suggest that the pandemic brought about both learning losses and educational inequality.

Our main question was to explore whether the widening rural-urban inequality in education during the crisis could be attributed to students’ SES and psycho-social traits. Surprisingly, we found that socio-economic factors, such as parental education, occupation and income, could not adequately explain the widening rural-urban inequality in education, whereas the students’ self-control partially accounted for the unequal changes in academic achievements between the student populations. We speculate that self-control may function like a “conversion factor” (Sen, 1992). Students with good self-control characteristics may be more likely to convert home resources into learning advantages, boosting their academic achievements in turn. By contrast, among students with relatively low self-control, the effects of school closures and online education on rural and urban children’s academic achievements may not be significantly different. Furthermore, these findings prompt us to revise our theoretical framework and embrace a more integrated model in future research. We assumed that SES and psycho-social traits were independent factors in the present study. It is possible that students’ personal traits can interact with family characteristics (Duckworth et al., 2019). In other words, self-control might play a mediating role between SES and learning losses during the crisis. We hope that revised models can be examined in future studies.

While this study provided new evidence, it bears some limitations. One is that the data were collected from a single high school in Eastern China. Generalization of the findings warrants caution. Moreover, since all students were exposed to the same mode of school closure, the data did not allow us to analyze the causal effect of the COVID-19-related school closure on academic achievement. A control group or additional data points before and after the school closure would be required to make an inference about causality.

Despite these limitations, the study has theoretical and policy implications. First, losses in learning during the COVID-19-related school closures suggest that education and social policies face considerable challenges. As online education was used as an important part of the response policy package in China (Lu et al., 2020), measures should be taken to guarantee the quality of online education. For example, Clark et al. (2021) showed that students earned higher scores when they received recorded online lessons from higher-quality teachers. Policymakers should also prioritize reopening schools once acceptable public health conditions have been maintained. Another implication speaks to the nature of policy measures that target the rural-urban educational inequality. Between the two options of providing monetary and material resources and extending psycho-social support to rural students, based on the present study’s results, the latter may prove more beneficial in the context of China. It is therefore a pertinent task for educators and social service providers to promote capabilities and resilience in young people and their families so that they continue to grow during times of crisis.

Author Biographies

Gaoming Ma is an assistant professor at the Center of Social Welfare and Governance, Department of Social Welfare and Risk Management, School of Public Affairs, Zhejiang University. His research interest is child development and social policy.

Jiayu Zhang is a doctoral candidate at the Department of Social Welfare and Risk Management, School of Public Affairs, Zhejiang University. Her research interest is intergenerational care and child health.

Liu Hong is an associate professor at School of Social Development and Public Policy, Fudan University. His research interests involve social policy and youth development.