The spillover effect and external welfare of extracurricular tutoring: evidence from academic tutoring for middle-school students in China

ABSTRACTThis paper investigates the spillover effects of middle school students taking up academic tutoring on students' cognitive and noncognitive skills in the same class in China. The identification relies on variations across classes within schools, which are formed roughly randomly. Correcting for measurement error, we find a significantly positive spillover effect of peers' academic tutoring on students' English test scores, and no spillover effect exists on students' ranks at the school level. For noncognitive performance, we find a positive spillover effect of academic tutoring on improving confidence. Our results show no positive external welfare exists for academic tutoring for students' grades or ranks, consistent with the newly released 'Double Reduction' policy that the Chinese government prohibits extra academic tutoring.KEYWORDS: spillover effectacademic tutoringexternal welfaremiddle-school studentsJEL CLASSIFICATION: I21I25I28J13 Disclosure statementNo potential conflict of interest was reported by the author(s).Notes1 It is called contextual effects defined by Manski (Citation1993).2 The Compulsory Education Law of the People's Republic of China was amended and adopted at the 22nd Meeting of the Standing Committee of the Tenth National People's Congress on June 29, 2006.3 It is still an open question on whether classrooms or schools are the more appropriate unit of peer interactions. We use the classroom level for the identification strategy. However, it is also natural to expect that a large fraction of total peer interactions should arise at the school level.4 The CEPS applied a stratified, multistage sampling design with probability proportional to size (PPS). It used the average education level of the population and the proportion of the floating population as stratified variables. First, 28 counties were randomly selected from a total of 2,870 counties. Second, in each selected county, four schools were randomly selected. Third, two classes from 7th grade and two classes from 9th grade were randomly selected. All students, their parents and teachers, as well as administrators of the selected schools were surveyed.5 We use classrooms assigned randomly. Alternative rules include: assign based on students' residency or past scores.6 Hu (Citation2015) was the first to use the random assignment identification strategy. Recent research that exploits the same dataset and identification strategy includes Gong et al. (Citation2021); Eble and Hu (Citation2019); Xu et al. (Citation2020); Eble and Hu (Citation2020); Zhao and Zhao (Citation2021).7 East region: Jiangsu, Guangdong; Middle region: Henan, Shanxi; West region: Xinjiang.8 6,241 students across 150 classrooms were interviewed in the 2013 – 2014 academic year but the CEPS lost track of 370 students in the 2014 – 2015 academic year.9 The Hukou system is the unique household registration system in China. Public schools are dominant in China and non-local residents are restricted from enrolling in public elementary and middle schools. Students who are not living in their Hukou registered place could be considered immigrant students.10 Ammermueller and Pischke (Citation2009)argued that the test performed well in a Monte Carlo experiment, with good power, although the rejection rate was somewhat high under the null. They found an empirical rejection rate of 0.13 for a 5% nominal size.11 For estimated coefficients of family's socio-economic status, the magnitudes are pretty small. For example, being the only child in the family is only associated with a 0.4% point increase in the fraction of peers who take Chinese tutoring.12 The variation in Table A6 is in classroom means and most student-level variation is within classes..13 Using the standard deviations reported in Table 1, we calculate 0.235×8.766=2.06 and 2.06/29.073=0.071.14 Middle school students in China usually start their new academic year in September.15 We consider the expenditure for academic tutoring and interest-oriented tutoring such as painting, calligraphy, music, dancing, chess, or sports together as we are not able to separate them from the answers.16 Table A7 shows heterogeneous effects of tutoring expenditure on both cognitive and noncognitive performance and the estimated coefficient of the interaction term between tutoring expenditure and peers' cores tutoring are insignificant.