Stevenson and Baker (1992) used the term shadow to denote the strong connection between educational allocation rules and non-formal schooling, implying that such rules tend to be hidden in countries such as Taiwan (Lin, 1983) and Hon Kong (Sweeting, 1983). In fact, in some educational systems (e.g., in Japan and China), these activities make up a large, open enterprise. Private supplementary curriculum tutoring beyond the hours of formal schooling is regarded as shadow education (Bray, 1999, 2009; Buchmann et al., 2010). Shadow education, which is also called private tutoring and additional instruction interchangeably, occurs because of fierce competition for entry into college, education inequity, and the high college aspirations of students. Kim and Jung (2019) disclosed five shadow education practices in South Korea as follows: home-visit private tutoring, private tutoring in institutes, subscribed learning programs, internet-based private tutoring, and after-school programs. Zhang and Bray (2020) denoted three modes of shadow education, consisting of providers, forms, and seasons. For example, providers include individuals and institutions; forms include online, dual-tutor, and offline.

The direct purpose of shadow education is to improve academic performance but studies have shown that its impact on performance is not consistent. Kim (2015) found that private tutoring expenditure appeared to have significant effects on standardized test achievement but not have statistically significant effects on school performance achievement. Guill and Bos (2014) showed that neither hierarchical regression analysis nor propensity scores matching approach revealed any positive effect of private tutoring on student achievement when several cognitive, motivational, family, and school context covariates were controlled for.

PISA investigated extracurricular activities and additional instruction in 2018, which is part of shadow education, in each educational cycle. Useful results were generated. Extracurricular activities mainly cover activities in-school, whereas additional instruction refers to extra time outside of mainstream school hours; such time may be provided at private institutions, at home, or elsewhere. PISA research has been developed for more than 20 years with subjects and time as two essential variables used to evaluate shadow education. Examples of questions posed are as follows: in the year 2000, “During the last 3 years, have you attended any of these special courses outside of your school to improve your results?” and, in 2003, “On average, how many hours do you spend each week on the following?” In 2015, the term “additional instruction” and question items such as “In this school year, approximately how many hours per week do you attend additional instruction in the following domains in addition to mandatory school lessons?” were included in the framework. Questions and instructions that compare in-school and outside-school factors were included in PISA 2015 such as “Compare your lessons at school and your additional science instruction?” and “Where are the following teacher characteristics more likely to occur?” In PISA 2018, motivation information was sought through items such as “Why do you attend additional instruction in this school year.”

Wellbeing is a complex, multidimensional construct that cannot be properly measured by a sole indicator in a single domain (Borgonovi and Judit, 2016). The concept of wellbeing concerns optimal psychological functioning and experience (Ryan and Deci, 2001). Researchers, although finding that the formulation of a universal concept of wellbeing was challenging, attempted nevertheless to build up various conceptual models. For example, Scanlon (1998) based his contractualism on a broader account of wellbeing; Sumner (1998) pointed out that virtue is the most important constituent of wellbeing. Student wellbeing, defined as students' overall development and quality of life, is increasingly accounted for in education policy (OECD, 2017). Especially for young pupils in all-level schools, educational outcomes have a strong relation with wellbeing. Konu and Rimpelä (2002) attempted to construct a school wellbeing model by identifying four indicators to measure student wellbeing, that is, school conditions (having), social relationships (loving), means for self-fulfillment (being), and health status. Ryan and Deci (2001) proposed that hedonic wellbeing depicts spontaneous feelings of happiness and that eudaimonic wellbeing concerns deeper self-realization. The varying views of wellbeing have shaped two distinct, yet overlapping, perspectives for empirical inquiry, namely, subjective and psychological wellbeing (Ryan and Deci, 2001). However, the above definitions of student wellbeing involve spontaneous feelings, self-realization, etc. Thus, it is clear that student wellbeing is a complex, multidimensional construct that cannot be properly measured by a sole indicator in a single domain (Borgonovi and Pál, 2016).

OECD (2017) proposed the how's life framework to measure wellbeing using 11 dimensions under two broad headings. The “Material Condition” covers a dimension grounded in market transactions: income and wealth, jobs and earnings, and housing. “Quality of Life” encompasses factors that are essential to welfare: health status, work-life balance, education and skills, social connections, civic engagement and governance, environmental quality, personal security, and subjective wellbeing. The Framework for the Analysis of Student Well-Being (FASWB) proposed by PISA is one of their widely utilized tools, which measure student wellbeing. In 2015, PISA became the first large-scale testing program to extract student wellbeing by consisting of independent indicators, and it developed the FASWB in 2018. The questionnaire covers the dimensions of life as a whole, self-related wellbeing (health, education and skills, and psychological functioning), school-related wellbeing (social connections and schoolwork), and wellbeing outside of school (social connections, material conditions, and leisure time), all of them also representing objective and subjective perceptions, affect, and satisfaction perspectives. OECD (2019) categorized student wellbeing into five domains, namely, cognitive, psychological, physical, social, and material wellbeing. The nomination of student wellbeing as a key research field was due to various factors such as lengthy schooling, high suicide rate (prompted by fierce competition), dislike of schooling, diminished school engagement, and school anxiety (Natsuaki et al., 2009; McGill et al., 2012). In 2021, PISA extended its survey to assess teachers' occupational wellbeing, thus building up a student-teacher binary research base for further studies.

As mentioned earlier, shadow education consists of various factors, such as in-school vs. out-school, public vs. private, online vs. offline, instruction by teacher vs. by tutor, and during semester vs. during vacation. This study adopts a specific conceptional definition from PISA 2018, additional course, which is presented in our student questionnaire and which refers to time spent on extra courses as well as which kinds of subject students attend after school. Student wellbeing in our study includes affective, life meaning, and belonging to school. We set out to explore the factors influencing student wellbeing, under the framework of PISA 2018, in order to shed light on how shadow education influence student wellbeing and to provide an empirical basis for further research. In terms of the factors affecting student wellbeing, “student” and “school” would be extracted as dimensions and would form the key issues in our research. The following questions underpinned our study:

A total of 6,037 students completed the PISA 2018 survey from 152 Hong Kong schools (7.2% from government; 78.1% from aided or capital; 2.3% from private or international; 12.4% from a direct subsidy scheme). Students' age ranged from 15.25 to 16.25 years, with an average of 15.73 years. The sample contained 2,955 (48.9%) female respondents and 3,082 (51.1%) male respondents.

The variables utilized were taken from the PISA 2018 questionnaire and consisted of student, school, education career, and student wellbeing domains.

This study performs data processing utilizing SPSS.25 preliminary sorting and transformation and employs HLM6.08 data analysis on the significant factors of wellbeing in Hong Kong for two horizontal linear model analyses. To find out the effect of shadow education on student wellbeing, two-level HLM is used to examine the relationships between student and school-level variables.

The cross-level coefficient ρ = τ∞/(τ∞+δ2) = 147.083/(147.083 + 163.547) = 0.473 shows that 47.3% of the total variance of the dependent variable affective student wellbeing derives from differences between schools, whereas 52.7% is due to differences within schools. When the dependent variable is belonging to school, the cross-level correlation coefficient is ρ = τ∞/(τ∞ + δ2) = 0.585, indicating that 58.5% of the total variance comes from differences between schools, and 41.5% comes from differences within schools. When the dependent variable is life meaning, the cross-level correlation coefficient is ρ = τ∞/(τ∞ + δ2) = 87.367/(105.347 + 87.367) = 0.453, indicating that 45.3% of the total variance comes from differences between schools and 54.7% is derived from differences within the school. According to Cohen's (1988) definition, ρ > 0.138 indicates a high degree of correlation. Therefore, the differences between schools cannot be ignored; it suggests that student wellbeing can be explained partly by factors within the school and partly by factors between schools.

In our analysis of student gender, ESCS, shadow education attendance, and outside-school learning time, the random effects covariance model analysis (refer to Table 3) shows that there were no significant differences for the dependent variable effect. For the variables, gender (p = 0.255 > 0.05), ESCS (p = 0.276 > 0.05), and shadow education attending time (p = 0.442 > 0.05), there were also no significant differences but for shadow education attending (p = 0.002 < 0.05), there was a significant positive predictive effect for affective wellbeing. This result indicates that: (1) there are no significant differences in affective wellbeing for gender and ESCS. The correlation between shadow education attendance and affective wellbeing reached significance. The dependent variable belonging to school did not have a significant effect on gender (p = 0.294 > 0.05), ESCS (p = 0.539 > 0.05), or shadow education time (p = 0.147 > 0.05). Results for the variable shadow education attendance were significant (p = 0.035 < 0.05). (2) There were no significant differences in belonging to a school for gender and family economic and cultural status; there was no correlation between outside-school learning time and belonging to school. There were no significant differences for life meaning in ESCS (p = 0.065 > 0.05) and shadow education time (p = 0.198 > 0.05), but there were significant differences for gender (p = 0.031 < 0.05) and shadow education attendance (p = 0.000 < 0.01). (3) There were no significant differences in life meaning for students with different family economic and cultural statuses; shadow education time was negatively correlated with effective wellbeing, although this finding was not significant; students attending shadow education have a higher level of psychological resilience than those who do not attend.

When school dimension, school type, supportive of shadow education, schoolwork pressure, and competition climate were gradually added and a non-random intercept model analysis was performed (refer to Table 3), the results showed that when the dependent variable was affective wellbeing, school type (p = 0.579 > 0.01), supportive of shadow education (p = 0.953 > 0.01), and schoolwork pressure to attendance of shadow education (p = 0.184 > 0.01) were significant. The competition climate (p = 0.012 < 0.05) variable had a significant impact on affective wellbeing, suggesting that competition climate in the school is conducive to affective wellbeing. When the dependent variable is belonging to school, the school type (p = 0.502 > 0.01), supportive of shadow education (p = 0.938 > 0.01), and schoolwork pressure leading to attendance of shadow education (p = 0.097 > 0.01) had no effect on belonging to school. Competition climate (p = 0.009 < 0.01) had a significant impact on belonging to school. When the dependent variable is life meaning, the types of school (p = 0.496 > 0.01), supportive of shadow education (p = 0.734 > 0.01), and schoolwork pressure leading to attendance of shadow education (p = 0.077 > 0.01) had no significant impact on life meaning. In terms of the variable of competition climate (p = 0.011 < 0.05), the competition climate has a significant impact on life meaning, indicating that the competition climate can stimulate students' life meaning.

The student is a key component in school, so the characteristics of shadow education in the school dimension can affect student wellbeing to a certain extent.

In the cross-level interaction of the integrated model (refer to Table 4), when the dependent variable was wellbeing, school supportive of shadow education was shown to play a moderating role in attendance of shadow education in affective wellbeing. In that case, supportive of shadow education had a negatively weakening effect on attendance of shadow education and affective student wellbeing. For each additional unit of supportive shadow education, the impact of attendance shadow education on affective student wellbeing decreased by one. These findings suggest that if a school strongly supports students' shadow education, their affective wellbeing will increase.

When the dependent variable was belonging to school, attendance of shadow education caused by competition climate was shown to play a moderating role between the length of attendance at shadow education and belonging to school. The relationship has a negative weakening effect. In schools with high attendance at shadow education caused by competition, students with longer extracurricular hours have a stronger sense of belonging to school.

When the dependent variable is life meaning, attendance at shadow education caused by competition has a negative effect on the relationship between the length of extracurricular study and life meaning. It is concluded that in schools with high attendance at shadow education caused by the competition climate, students with longer extracurricular hours have a strong sense of life meaning; and in schools with high attendance at shadow education caused by support for shadow education, students with longer extracurricular hours have a stronger sense of life meaning.

The results suggest that attendance at shadow education has a significant impact on affective student wellbeing, belonging to school, and life meaning. The length of extracurricular study has a negative association with a sense of affective student wellbeing, belonging to school, and life meaning. Students who attend shadow education have a higher level of affective wellbeing, belonging to school, and meaning in life than students who do not. Female students have a higher level of life meaning than male students. The result testifies that shadow education is not detrimental; rather, attending shadow education improves student wellbeing in Hong Kong. Influenced by a mixed model of Confucian and western cultures, Hong Kong students demonstrate some features of Confucian culture. Participants from the Confucian cultural sphere have a highly positive parenting style, affecting the relationship between education and social achievement. The positive impact of shadow education on students' wellbeing has been greatly influenced by this educational boom. As mentioned earlier, the incentive for attendance at shadow education derives mainly from the students themselves, which means that the students who attend shadow education are mostly self-motivated. Kevin (2020), who conducted narrative research on learning under shadow education in Hong Kong, pointed out that shadow education plays a pivotal role in the advancement and future careers, helping students in Hong Kong successfully complete their education. Kevin (2020) also found that most students accept shadow education, which may be viewed as a necessity, and that those students may not feel disadvantaged. The results of this study are also consistent with Cayubit et al. (2014)'s findings, that is, that the inferiority and inadequacy related to learning will be transformed into a sense of efficacy, self-confidence, and high self-esteem after attendance at shadow education. Thus, the provision of private education has a rational basis and is an inevitable trend, consistent with the results of studies in Western countries (Bray, 2009; Burch, 2009; Davies and Guppy, 2010; Silova, 2010). Zhang and Bray (2020) highlighted how learning and teaching are personalized processes in which the private sector can provide practical answers to the problems in public schools. As a result, attendance at shadow education will indirectly improve happiness, with the impact not limited to those often reported in the literature.

Among the variables of school shadow education, our results suggested that school competition climate has a significant impact on affective student wellbeing, belonging to school, and life meaning, whereas the three variables of school type of shadow education caused by schoolwork pressure and shadow education support have no significant impact on affective student wellbeing, school sense of belonging, and life meaning. Bray (2013) also found that competition climate has a significant impact on student wellbeing, and students in schools with a strong competition climate are more likely to fear failure, so they tend to relieve their emotional pressure by attending shadow education.

We found that shadow education attendance due to schoolwork pressure and school support did not have a significant impact on students' wellbeing; Bray (2013), however, found that students who attend shadow education due to schoolwork pressure can significantly improve their academic performance. Another key finding in this study was that most of the Hong Kong students who attend shadow education do so because of their own preferences instead of recommendations from teachers. This finding is in accordance with Zhan et al.'s (2013) research which showed that Hong Kong students believe that shadow education helps them cope with examinations more efficiently compared with mainstream education in schools. It can be inferred that shadow education prompted by school factors has a significant effect on testing scores but not on student wellbeing: students who are in a fierce competition climate and who put more time into shadow education will have lower levels of wellbeing.