Learning from bad peers? Influences of peer deviant behaviour on adolescent academic performance

ABSTRACT

In addition to family and school factors, adolescent academic performance can also be influenced by the surrounding peer groups. Based on the data from China Education Panel Studies (CEPS) between 2013 and 2014, this paper examined the association between peer deviant behaviour and adolescent academic performance. It is found that peer deviant behaviour negatively predicted adolescent academic performance, and the negative association between peer deviant behaviour and academic performance in migrant and school-boarding adolescents was obviously weaker than that in native and home-resident adolescents. This negative association was slightly reduced but still significant after controlling for potential selection bias. More peer deviant behaviours reduced adolescent academic performance by increasing their own deviant behaviours and reducing their own educational expectation. Therefore, it is necessary to guide peer behaviours and interactions and regulate adolescent deviant behaviours from a global perspective to avoid negative peer effects.

KEYWORDS:

• Peer deviant behaviour
• academic performance
• peer effects
• Chinese adolescents

Introduction

The healthy growth and development of school-aged adolescents is crucial to the future development of the country and society, and how to improve the quality of education and the academic performance of adolescents is one of the most important issues. China has a long-standing tradition of valuing education, as expressed in idioms such as hoping for children to become dragons and making a name on the gold list, showcasing the importance that families and society place on the education of children. China also has the biggest compulsory educational system worldwide, and it has performed a far-reaching educational reform called the ‘double reduction’ policy since 2021, so the academic performance of adolescents in compulsory education has attracted much attention. As early as half a century ago, the Coleman Report has pointed out that family cultivation, school education and peer factors all have profound impacts on adolescent academic development (Coleman, 1966). There are numerous studies on how family and school factors influence adolescent academic performance, and most of them have observed that family socioeconomic status, family social capital, school teaching quality, teacher-student relationship and other factors have fundamental influences on adolescent academic performance (Tian & Jing, 2018; Wang et al., 2017; Zhao, 2013).

In the past few decades, China has undergone a profound social transition, and citizen’s social values and behavioural norms have experienced significant ruptures and reconstructions. During the period of rapid social change, children and adolescents are more prone to deviant and delinquent behaviour (Li et al., 2010). Particularly, the urbanization and population migration have led to a mass of migrant and left-behind children. These children mostly grow up in an environment where their parents are absent, which may contribute to more instances of deviant behaviour such as excessive internet use, drinking and brawl (Wu et al., 2018; Yang & Jiang, 2023), which indirectly contribute to their deviant peer interactions. The allusion of three house moves of Mencius’s mother reveals the important influence of surrounding peers on adolescent behaviour and academic development. Some recently published studies have revealed the peer effects on adolescent academic performance. For example, numerous studies document that the improvement of classmates’ average grades significantly promotes adolescent academic performance, but this promotion effect is nonlinear, and the negative effects brought by a poor learning environment far outweigh the positive effects brought by a good one (Kang, 2007; Min et al., 2019; Wang et al., 2022; Wu & Zhang, 2020; Yang & Huang, 2020; Yuan et al., 2018). Some studies also report that a larger proportion of migrant adolescents in the class is not conducive to the improvement of class’s learning environment, which negatively influences the academic performance of native adolescents (Hu, 2018). However, other evidence shows that a certain number of migrant adolescents in a class can improve the academic performance of native adolescents to some extent (Wu & Qi, 2019). Furthermore, a good academic performance of roommates can positively influence individual academic performance through such channels as value cultivation and learning behaviour demonstration, but such influence is different among adolescents of different sexes and from different social classes (Cheng, 2017; Ma & Huang, 2021).

It can be seen that prior studies have tried to examine the peer effects from the perspective of class or peer academic characteristics (Wu & Zhang, 2020; Yang, 2021; Yang & Huang, 2020), and the observed evidence of peer effects on adolescent academic performance is inconsistent, which need more investigations. This class-level research design based on overall class characteristics neglects the adolescent-centred peer relationships and more specific and diverse behavioural and attitudinal characteristics within peer groups, which can also hardly explore whether and how specific peer behaviours affect adolescent academic performance from a more microscopic perspective. Very few studies have focused on the peer effects of learning behaviours within adolescent daily interaction circles (Cheng, 2017), but the investigation into negative or deviant behaviours of peers is limited. Only two preliminary studies document that establishing social relationships with peers who exhibit negative behaviours may have a negative impact on children’s academic performance (Han, 2022; Gao & Xue, 2020), but they do not further explore potential mediating pathways.

Under the background of the drastic transition of Chinese society and the increase of deviant behaviour among adolescents, exploration of the association between deviant peer behaviour and academic performance holds significant theoretical and practical importance. On the one hand, it can expand the connotation of peer effects in academic performance research and uncover its multiple manifestations. On the other hand, peer relationships play a crucial role in the socialization process of adolescents, and its influence on adolescent development is no less important than that of family and school factors (Gao & Xue, 2020). Revealing the process by which peer relationships affect adolescent academic performance can help contribute to targeted interventions aimed at improving adolescent academic development. Thus, based on the adolescent-centred peer relationships, this study attempted to examine the relationship between peer deviant behaviour and adolescent academic performance, as well as the heterogeneity of this relationship, in China. Furthermore, this study also aimed to uncover the mechanisms through which peer deviant behaviour influences adolescent academic performance from both behavioural and attitudinal perspectives.

Literature review and research hypotheses

Studies in the area of educational sociology have commonly found that there are significant peer effects on adolescent academic achievement; that is to say, there is an obvious positive relationship between adolescent own academic performance and their peer group’s academic performance (Liang & He, 2017; Yang, 2021; Yang & Huang, 2020; Zhang et al., 2021). Peer effects usually play a role through the norm of social conformity (Wu & Zhang, 2020), which means that adolescents tend to behave similarly compared to their peers and they also share similar values. Peer group seems to be a catalyst or bridge that transmits social norms, educational values and learning skills. Living and learning with outstanding peers can motivate and promote adolescent academic performance, while being with peers that own poor academic performance or behavioural norms usually lead to some negative consequences (Putnam, 2015; Zhang et al., 2021).

An obvious challenge is that the formation of peer relationships among adolescents is highly selective. On the one hand, the peer groups around school-aged adolescents are often school-aged adolescents in the same school, grade, class or community. At present, there is a considerable socioeconomic differentiation in dwelling and adolescent schooling in China (Wu & Huang, 2016). Thus, the peer groups of school-aged adolescents are highly similar in family socioeconomic status, education and cultural environment. On the other hand, adolescents will also actively choose adolescents with similar behaviours and values to build peer relationships (Jiang & Wang, 2020), which also leads to the phenomenon of birds of a feather flock together. This selection bias can lead to estimation bias in peer effects and exaggerate peer effects on adolescent academic performance.

Some empirical studies have also paid attention to and tried to address the above selection bias using multiple strategies. For example, based on the stable inverse probability weighting and county-fixed effect models, Wu and Zhang found that the quality of close friends network and average academic performance within class were positive predictors of adolescent academic performance (Wu & Zhang, 2020). Based on the propensity score matching method, Yang and Huang observed that average academic performance within class and peer academic performance both positively predicted adolescent academic performance, and peer effects had a stronger influence on mathematics achievement than Chinese and English achievement (Yang & Huang, 2020). Two recent studies found that peer good behaviours could significantly improve adolescent academic performance, while peer deviant behaviour had a negative effect on adolescent academic performance (Gao & Xue, 2020; Zhang et al., 2021). However, these two recent studies do not overcome the selection bias mentioned above, so further tests are needed. Accordingly, the following hypothesis was proposed.

H1:

Peer deviant behaviour can negatively predict adolescent academic performance.

Prior evidence suggests that the peer effects on adolescent academic performance cannot be constant across different groups. Cheng, for example, found that the academic performance of peers within the same social class had a stronger long-term impact on adolescent academic performance, while peers across different social classes had a stronger short-term impact (Cheng, 2017). Considering the systematic differences in habits, cognitive abilities and social networks of peers across different groups, there may also be obvious disparities in the association between peer deviant behaviour and adolescent academic performance. There has been evidence that boys have more behavioural problems than girls (Yin et al., 2014), so the group aggregation of peer deviant behaviour is also more common in boys; and boys also tend to have more rebellious personality traits, making their academic performance more susceptible to the negative effects of peer deviant behaviour.

The large-scale urbanization and population migration in China over the past several decades have led to a significant influx of rural labourers into cities for work, as well as an increasing number of rural children attending schools in other regions. The issue of migrant children receiving education in different locations and its impact on their academic performance and overall well-being has received widespread attention (Meng & Yang, 2019; Xu, 2020; Zhu & Wang, 2019). Studies have shown that compared to urban native children, migrant children tend to have lower family socioeconomic status and social capital (Yang & Duan, 2008). However, compared to children who stay in rural areas, migrant children have access to better educational resources and therefore have improved academic performance (Li & Qiu, 2016; Zhu & Wang, 2019). Furthermore, due to the selection of migration, parents of migrant children tend to prioritize their children’s education more than parents of native children (Zhu & Wang, 2019), which may indirectly reduce the negative effect of peer deviant behaviour on migrant children’s academic performance.

Boarding arrangement is closely related to social stratification and population migration. The rapid urbanization process in China has led to an increase in school-boarding students, including those who accompany their parents to cities for education, those who are forced to board due to the merging of rural primary and middle schools, and those who choose to board due to the long distance between home and school. Home-resident students usually have more time to spend with their parents and other families after school, and they have more family economic resources and social capital, helping reduce the negative impact of peer effects on academic performance. By contrast, the main reasons for students to board at school are that their homes are far from the school and they are either migrant or left-behind children. These characteristics determine that school-boarding students are more likely to be in the lower location of the social stratification spectrum (Chen & Miao, 2021; Wang & Mao, 2015). In addition, school-boarding students usually spend more time in daily contact with their peers and roomies after school, which may lead to stronger peer effects of peer deviant behaviour on academic performance. Accordingly, the following hypotheses were proposed.

H2:

There are group heterogeneities in the negative association between peer deviant behaviour and adolescent academic performance.

H2a:

Boys are more affected by peer deviant behaviour;

H2b:

Migrant adolescents are less affected by peer deviant behaviour;

H2c:

School-boarding adolescents are more affected by peer deviant behaviour.

Peer effects are essentially a kind of endogenous interaction, and adolescent behaviour will change with the behaviour of the group they are in (Lu & Zhang, 2007). The exemplary role of good peers can restrain adolescent deviant behaviours to some extent, while making friends with bad peers may increase adolescent deviant behaviours (Su & Xing, 2019). According to the norm of social compliance and contagion, this phenomenon reflects members within specific peer groups tend to follow consistent norms with each other. Just as there are relatively consistent learning behaviours within specific peer groups (Cheng, 2017; Ma & Huang, 2021), in the case of more deviant behaviours within the peer group, adolescents are also more inclined to behave illegally, so as to make their own behaviours conform to the norms within the peer group (Newman, 2017). Thus, peer deviant behaviour may increase adolescent deviant behaviour and negatively influence adolescent academic performance. Accordingly, the following hypothesis was proposed.

H3:

Adolescent own deviant behaviour mediates the influence of peer deviant behaviour on their academic performance.

Adolescent own educational expectation is closely related to those of their peers (Haller & Butterworth, 1960). Under the rule of social compliance and contagion, adolescent educational expectation will increase with the increase in average educational expectation within the class, and the enterprising behaviour of friends is also positively related to adolescent educational expectation (Cao & Wu, 2019), which reflects the fact that adolescents usually follow the values within the peer group, either actively or passively (Cheng, 2017; Newman, 2017; Wu & Zhang, 2020). Peers with more deviant behaviours tend to have lower educational expectations and poorer academic performance. Under the rule of social compliance and contagion, adolescents tend to reduce their educational expectation when they are in such a peer group, which may further lead to poor academic performance. Accordingly, the following hypothesis was proposed.

H4:

Adolescent own educational expectation mediates the influence of peer deviant behaviour on their academic performance.

Data and methods

Data source

The data used in this study were obtained from the China Education Panel Survey (CEPS) carried out in the autumn of 2013 and spring of 2014. The survey was designed and conducted by the National Survey Research Center at Renmin University of China, which was conducted among students in the 7^th and 9^th grades. At the first stage, CEPS2013–2014 selected 28 county-level units (counties/cities/districts) as the survey sites based on the average schooling years and the proportion of migrant population, including 15 counties/cities/districts randomly selected from 2870 counties/cities/districts, three counties/districts selected from Shanghai city, and 10 counties/cities/districts randomly selected from 120 counties/cities/districts with a large proportion of migrants, which was proven to be nationally representative (Shen, 2020). At the second stage, the probability proportionate to size (PPS) sampling method was used, and a total of 112 schools and 438 classes were randomly selected from the 28 county-level units mentioned above. Finally, all students in the selected classes were interviewed in the survey. A total of 19,487 students were interviewed in the survey, and their parents and teachers were also interviewed. In this study, samples with values missing on peer deviant behaviour and adolescent academic performance were deleted, and a total of 18,420 valid samples were used in the following analysis.

Variables

Explained variable: academic performance

The mid-term examination results of Chinese, mathematics and English of junior high school students in 2013 were collected in the survey, and they were standardized at the school level, with a mean of 70 points and a variance of 10 points, which made them more comparable. The present study used the weighted average score of Chinese, mathematics and English ranging from 0 to 100 points as the outcome.

Explanatory variable: peer deviant behaviour

The peer network of adolescents was investigated in CEPS, and they were asked to enumerate whether there were the following deviant behaviours among their good friends: (1) truancy, (2) being punished for violating school disciplines, (3) brawl, (4) smoking or drinking, (5) going to the Internet cafés or video arcades, (6) puppy love, and (7) quit school. The Cronbach’s α was 0.856, and the options for each question included ‘0=none’, ‘1=one or two’ and ‘2=many’. These adolescent behaviours have been widely used as deviant behaviours in prior studies especially in those from China (Li, 2022; Yang & Jiang, 2023). The above items were then added to yield the peer deviant behaviour variable ranging from 0 to 14, and higher score meant more peer deviant behaviours. Subsequently, considering that most of the peers had no deviant behaviour, the continuous peer deviant behaviour variable presented a highly right-skewed distribution, so this study further recoded it as a binary variable classifying those with zero value as having no peer deviant behaviour (n = 12246) and those with positive values as having peer deviant behaviour (n = 6174).

Mediating variables: adolescent deviant behaviour and educational expectation

CEPS investigated adolescent being late for school, truancy behaviours and being criticized by teachers (both their parents and themselves); the Cronbach’s α was 0.666, and the options for each question included 0 to 3 indicating from ‘none’ to ‘often’. Thus, the exploratory factor analysis (EFA) was performed to yield a continuous variable (KMO = 0.653, cumulative variance = 52.5%), and higher score meant more deviant behaviours. Adolescent educational expectation was measured using a single question, and what level of education the interviewees hope to achieve was asked in the questionnaire. It was a continuous variable ranging from 0 to 6 that indicated the expected educational level from junior high school to doctoral level.

Control variables

Sex, age, grade, being the only child or not, being a migrant child or not, living in school or not, family economic status, parental educational years, family social capital, and the fixed effect at the school level were used as control variables. Similar to Tian and Jing’s study (Tian & Jing, 2018), family social capital was measured from three aspects, including parental supervision on adolescent learning, and intergenerational daily interactions and activities. Parental supervision on adolescent learning was obtained by adding two specific variables, while intergenerational daily interactions and activities were obtained using EFA (for the former: KMO = 0.772, cumulative variance = 47.1%; for the latter: KMO = 0.716, cumulative variance = 54.0%). More details can be seen in Appendix 1.

Statistical methods

Academic performance was a normally distributed continuous variable, so the general linear regression model based on ordinary least squares (OLS) estimate was used as the basic model to examine the association between peer deviant behaviour and academic performance (Information on whether four basic assumptions of OLS model were supported can be seen in Appendixes 2, 3 and 4). Subsequently, considering the endogenous selection bias of peer effects (Wu & Zhang, 2020; Yang & Huang, 2020), this study further used the propensity score matching (PSM) approach to overcome the selection bias of peer relationships, so as to estimate the impact of peer deviant behaviour on adolescent academic achievement more accurately. In this study, the inverse probability weighted (IPW) method based on the Stata syntax ‘teffects’ was used to estimate the propensity score (Hong, 2015; Stata, 2017). The above statistical methods were used to test H1.

Considering the fact that statistical significance does not indicate economic or practical significance, the Cohen’s d coefficient was calculated to evaluate the effect size of binary peer deviant behaviour on academic performance,¹ and adjusted Cohen’s d coefficient was calculated to evaluate the effect size of continuous peer deviant behaviour on academic performance² (Cohen, 1988; Milner et al., 2023). To test H2a to H2c, the Stata syntax ‘bdiff’ based on permutation test (Efron & Tibshirani, 1993; Lian et al., 2010) was used (seed number = 19950111, time of repeated sampling = 1000). To test H3 and H4, the direction dependence (DD) method was used in this study because continuous peer deviant behaviour variable presented a highly right-skewed distribution (Wiedermann & Eye, 2015). Subsequently, the Karlson-Holm-Breen (KHB) mediation model (Karlson & Anders, 2011; Karlson et al., 2012) was further used for a robustness check (retest H3 and H4).

Results

Association between peer deviant behaviour and adolescent academic performance

Models 1 to 3 in Table 1 report the results of OLS models. It can be seen that peer deviant behaviour was negatively related to adolescent academic performance, with adolescent own characteristics, school characteristics, family socioeconomic status and family social capital controlled for. Model 2 suggests that each unit increase in peer deviant behaviour reduced adolescent academic performance by 0.464 points on average (p < 0.001). Furthermore, the adjusted Cohen’s d was 0.324, which indicated a medium effect size.

Table 1. Association between peer deviant behaviour and adolescent academic performance.

	Model 1		Model 2		Model 3		Model 4
	β	R_SE	β	R_SE	β	R_SE	β	R_SE
Peer deviant behaviour_continuous	−0.516***	0.037	−0.464***	0.042
Peer deviant behaviour_binary (ref: no)					−1.187***	0.149	−0.968***	0.162
Sex (ref: girls)	−3.794***	0.126	−3.427***	0.131	−3.545***	0.131	Yes
Age	−1.483***	0.096	−1.222***	0.101	−1.257***	0.101	Yes
Grade (ref: 7^th)	2.914***	0.232	1.998***	0.244	2.054***	0.245	Yes
Only child (ref: no)	0.822***	0.157	0.397*	0.165	0.378*	0.165	Yes
Migrant child (ref: no)	0.710***	0.187	0.743***	0.195	0.736***	0.195	Yes
School-boarding (ref: no)	0.707**	0.236	0.483*	0.247	0.501*	0.248	Yes
Family SES (ref: low)
middle			0.261	0.177	0.267	0.178	Yes
high			−0.179	0.324	−0.220	0.326	Yes
Parental educational years			0.408***	0.028	0.412***	0.028	Yes
Parental supervision			−0.541***	0.036	−0.545***	0.036	Yes
Intergenerational daily interactions			1.010***	0.074	1.028***	0.074	Yes
Intergenerational daily activities			−0.172*	0.085	−0.193*	0.085	Yes
School fixed effect	Yes		Yes		Yes		Yes
Constant	91.256***	1.495	84.113***	1.634	84.673***	1.634	70.328***	0.118
R^2	0.093		0.122		0.117		—
n	17956		15660		15660		15660

***p < 0.001, **p < 0.01, *p < 0.05. R_SE meant robust standard error. Control variables were included in Model 4, but the IPW method did not report their estimates.

Considering the highly right-skewed distribution of peer deviant behaviour, Model 3 used binary peer deviant behaviour as the treatment. It is found that compared to adolescents with no peer deviant behaviour, those with peer deviant behaviour had a worse academic performance, and they got a lower score in the mid-term examination (β=-1.187, p < 0.001). Furthermore, the Cohen’s d was 0.255, which indicated a medium to small effect size.

Considering the selection bias in the process of peer relationship establishment, Model 4 further used the IPW method to estimate the influence of peer deviant behaviour on adolescent academic performance. It is found that compared with Model 3, the negative association between peer deviant behaviour and adolescent academic performance was slightly weakened, but it still strongly showed statistical significance. It is reported that, having peers with deviant behaviours would reduce adolescent academic performance by 0.968 points (p < 0.001).

The above results indicate that the potential selection bias will make us overestimate the negative association between peer deviant behaviour and adolescent academic performance. Even so, the selection bias seems not to be very critical, and peer deviant behaviour indeed has a substantial association with adolescent academic performance, with both statistical and economic or practical significances. Accordingly, H1 was supported.

Group heterogeneities in the association between peer deviant behaviour and adolescent academic performance

Models in Table 2 test whether the association between peer deviant behaviour and adolescent academic performance varies across groups. The results in Model 1 (both Panels A and B) show that the negative associations between peer deviant behaviour and academic performance were similar between boys and girls (p for difference > 0.05), so H2a was not supported. Boys usually have more peer deviant behaviours and poorer academic performance than girls, so this negative association may decline under the rule of diminishing marginal effect. On this condition, although boys may be more strongly influenced by peer deviant behaviours, the sex disparity in this association may be nonsignificant. However, the phenomenon of boys’ poor academic performance and peer deviant behaviours still deserves more attention.

Table 2. Group heterogeneities in the association between peer deviant behaviour and adolescent academic performance, OLS model.

	Model 1		Model 2		Model 3
Panel A	boys	girls	migrants	natives	school-boarding	home-residence
Peer deviant behaviour	−0.441***	−0.479***	−0.263**	−0.518***	−0.393***	−0.501***
	(0.051)	(0.071)	(0.096)	(0.047)	(0.067)	(0.054)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
School fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
p-value for difference test	0.340		0.006		0.056
R^2	0.071	0.088	0.149	0.132	0.126	0.134
n	7759	7901	2674	12986	4998	10662
Panel B	Model 1		Model 3		Model 4
Peer deviant behaviour (ref: no)	−1.310***	−0.931***	−0.658	−1.311***	−0.711**	−1.405***
	(0.213)	(0.206)	(0.354)	(0.165)	(0.252)	(0.185)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
School fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
p-value for difference test	0.079		0.006		0.004
R^2	0.065	0.083	0.147	0.125	0.120	0.130
n	7759	7901	2674	12986	4998	10662

***p < 0.001, **p < 0.01, *p < 0.05. Robust standard error in parentheses.

Model 2 in both Panels A and B shows that, compared with native adolescents, the academic performance of migrant adolescents was less related to peer deviant behaviour. Compared with no peer deviant behaviour, having peer deviant behaviours reduced the academic performance of migrant adolescents by 0.654 points, but it reduced the academic performance of native adolescents by 1.311 points (p for difference = 0.006). Therefore, H2b was supported.

Model 3 in both Panels A and B shows that peer deviant behaviour had a stronger negative association with the academic performance of home-resident students than that of school-boarding students. Compared with no peer deviant behaviour, having peer deviant behaviours reduced the academic performance of school-boarding students by 0.711 points, but it reduced the academic performance of home-resident students by 1.405 points (p for difference = 0.004). Therefore, H2c was not supported.

4.3. Mediations of adolescent own deviant behaviour and educational expectation

The results in Panel A, Table 3 show that adolescent academic performance was more likely to be the outcome and peer deviant behaviour is more likely to be the predictor, when controlling for adolescent own deviant behaviour and other controls. Because the residuals were more skewed when using peer deviant behaviour as the outcome and academic performance as the predictor. Based on this criterion, the results in Panel B, Table 3 also show that adolescent academic performance was more likely to be the outcome and peer deviant behaviour is more likely to be the predictor, when controlling for adolescent own educational expectation and other controls. It should be noted that, under large sample conditions, even a small skewness can yield statistically significant results in a skewness test. Therefore, in this study, the focus is solely on the magnitude of skewness itself as a criterion for judgement, rather than the significance level of the skewness test.

Table 3. Mediations of adolescent own deviant behaviour and educational expectation, based on DD method.

Panel A	Outcome	Predictors	β	R_SE	Skewness of residuals
Model 1	M1	X	0.107***	0.007	2.942***
Model 2	Y	X	−0.464***	0.042	−0.700***
Model 3	Y	M1	−1.555***	0.081	−0.680***
Model 4	Y	X	−0.301***	0.042	−0.678***
		M1	−1.440***	0.082
Model 5	X	M1	0.384***	0.027	3.300***
Model 6	X	Y	−0.023***	0.002	3.424***
Model 7	M1	Y	−0.021***	0.001	3.001***
Model 8	X	Y	−0.015***	0.002	3.287***
		M1	0.362***	0.027
Panel B	Outcome	Predictors	β	R_SE	Skewness
Model 1	M2	X	−0.064***	0.008	−0.301***
Model 2	Y	X	−0.464***	0.042	−0.700***
Model 3	Y	M2	2.323***	0.048	−0.687***
Model 4	Y	X	−0.301***	0.039	−0.682***
		M2	2.292***	0.047
Model 5	X	M2	−0.102***	0.013	3.527***
Model 6	X	Y	−0.023***	0.002	3.424***
Model 7	M2	Y	0.069***	0.001	−0.071***
Model 8	X	Y	−0.017***	0.002	3.505***
		M2	−0.063***	0.013

***p < 0.001, **p < 0.01, *p < 0.05. X meant peer deviant behaviour, Y meant academic performance, M1 meant adolescent own deviant behaviour, M2 meant adolescent own educational expectation. R_SE meant robust standard error. Skewness test for normality of residuals was performed. All other controls and school fixed effects were controlled for in all models.

Based on the results in Models 1 to 4 of Panel A, it is reported that adolescent own deviant behaviour may be an important channel from peer deviant behaviour to adolescent academic performance, and the mediation effect was −0.154 ( = 0.107$\times$-1.440). Based on the results in Models 1 to 4 of Panel B, it is reported that adolescent own educational expectation may also be an important channel from peer deviant behaviour to adolescent academic performance, and the mediation effect was −0.147 (=-0.064 × 2.292).

However, Table 3 did not test whether the mediation effect was statistically significant, so a further check using KHB method was performed. Table 4 shows that adolescent own deviant behaviour and educational expectation indeed mediated the association between peer deviant behaviour and adolescent academic performance. The pooled mediation of adolescent own deviant behaviour and educational expectation accounted for 58.1% (p < 0.001) of the total effect when using continuous peer deviant behaviour. Further, the binary peer deviant behaviour was used as the explanatory variable, and the results were highly consistent (mediation effect = 60.2%, p < 0.001). Thus, H3 and H4 were both supported.

Table 4. Mediations of adolescent own deviant behaviour and educational expectation, based on KHB method.

Panel A		β	SE	Proportion of mediation
Peer deviant behaviour (continuous)	Reduced model	−0.438***	0.034
	Full model	−0.183***	0.035
	Difference	−0.254***	0.017	0.581
	Decomposition of mediation
	M1	−0.115***	0.008	0.263
	M2	−0.139***	0.015	0.318
Model information: R^2 = 0.270，n = 15030
Panel B		β	SE	Proportion of mediation
Peer deviant behaviour (binary)	Reduced model	−1.126***	0.132
	Full model	−0.448***	0.133
	Difference	−0.678***	0.063	0.602
	Decomposition of mediation
	M1	−0.302***	0.026	0.269
	M2	−0.375***	0.056	0.333
Model information: R^2 = 0.270，n = 15030

***p < 0.001, **p < 0.01, *p < 0.05. SE meant standard error, M1 meant adolescent own deviant behaviour, M2 meant adolescent own educational expectation.

Robustness check

Considering the evident missing data issue in the previous models, which may possess non-randomness and affect the accuracy of model estimates, this study further employed the multiple imputation method to conduct robustness analysis. The results in Table 5 indicate that the core findings from the aforementioned analysis basically remained unchanged after the multiple imputation. That is to say, peer deviant behaviour still negatively predicted adolescent academic performance, with heterogeneity observed across groups such as migrants and natives, and school-boarding and home-resident students. Additionally, adolescent own deviant behaviour and educational expectation continued to serve as important mediations.

Table 5. Robustness check after multiple imputation, n = 18420.

Panel A	β	R_SE	Controls	School fixed effect
Total sample	−0.489***	0.036	Yes	Yes
Boys	−0.464***	0.044	Yes	Yes
Girls	−0.505***	0.065	Yes	Yes
Migrants	−0.339***	0.082	Yes	Yes
Natives	−0.526***	0.041	Yes	Yes
School-boarding	−0.444***	0.059	Yes	Yes
Home-residence	−0.500***	0.046	Yes	Yes
Panel B		β	SE	Proportion of mediation
Peer deviant behaviour	Reduced model	−0.489***	0.030
	Full model	−0.228***	0.030
	Difference	−0.261***	0.015	0.534
	Decomposition of mediation
	M1	−0.116***	0.008	0.238
	M2	−0.145***	0.013	0.296

***p < 0.001, **p < 0.01, *p < 0.05. M1 meant adolescent own deviant behaviour, M2 meant adolescent own educational expectation. SE meant standard error. R_SE meant robust standard error.

Conclusions and discussion

Adolescent education and healthy growth have always been the top priority of social development. Besides family and school, peer group is also an important factor affecting their education and healthy growth. Numerous studies have revealed the peer effects in academic performance among adolescents (Putnam, 2015; Yang & Huang, 2020). This study further discovers that deviant peer behaviour is strongly and negatively associated with adolescent academic performance. Even after controlling for the selection bias of peer relationships, this negative association remains significant. It indicates that peer behaviour indeed play a crucial role in influencing adolescent academic performance beyond the family and school environment, essentially representing a specific form of peer effects. In addition, bad or deviant behaviour of peers may eventually lead to the decline of adolescent academic performance through a series of pathways. This study finds that peer deviant behaviour mainly reduces adolescent academic performance by increasing their own deviant behaviours and reducing their educational expectation. These findings align with previous research and further confirm the social compliance and contagion effects within peer groups (Cao & Wu, 2019; Su & Xing, 2019; Wu & Zhang, 2020). Thus, this study not only provides concrete evidence of peer effects and its negative association with adolescent academic performance but also uncovers two specific mechanisms through which peer effects influence academic performance.

There were group heterogeneities in the influence of peer deviant behaviour on adolescent academic performance, and the negative impact is significantly lower in migrant and school-boarding adolescents than in their counterparts. Some studies report that a large proportion of migrant students in school or class may not be conducive to creating a good learning environment and therefore negatively influences the academic performance of native students (Hu, 2018). The CEPS data show that migrant students had poorer academic performance and more peer deviant behaviours, so it may negatively affect the academic performance of native students to some extent, resulting in a stronger negative influence of peer deviant behaviour on native students’ academic performance. In China, education is closely related to real estate, giving rise to the phenomenon of school district housing. Families with better socioeconomic conditions are more able to afford houses near high-quality schools, resulting in fewer needs for boarding at school. On the other hand, families with poorer socioeconomic conditions and rural families mostly reside in areas further away from schools, requiring their children to board at school in order to facilitate attending school (Chen & Miao, 2021; Liu & Zhou, 2019). This essentially creates a kind of social stratification. In this context, although boarding students may have more interactions with peers while non-boarding students have less, the former tend to have poorer socioeconomic status and academic performance than the latter. This may also lead to a diminishing marginal effect, where the negative influence of unfavourable peers on non-boarding students is stronger than that on boarding students.

Implications and limitations

The findings of this study have strong policy and practical implications. It is evident that adolescent academic performance is indeed affected by peer effects. A positive peer relationship is crucial for the academic performance of adolescents in middle school, suggesting that we should pay attention to both the quantity (how many friends they make with) and the quality (how the peers behave) of adolescent peer relationships. It is necessary to monitor adolescent peer relationships from multiple levels, including the family, school, and society, and encourage them to avoid negative peer relationships. Furthermore, although the negative association between peer deviant behaviour and academic performance is smaller among migrant and school-boarding students, these two specific groups are considered high-risk and academically vulnerable. Hence, it is essential to pay special attention to these disadvantaged groups and improve the overall quality of peer relationships and academic performance for all adolescents. On the other hand, under the influence of social compliance and contagion, peer deviant behaviour will lead to negative consequences through both behaviour imitation and value infection, such as adolescent own deviant behaviour and educational expectation mentioned in this study. Therefore, we need to strengthen ideological guidance for adolescents and assist them in developing correct educational and moral values. Parents and teachers should closely observe adolescent behaviour and promptly address any negative behaviour.

There are still some limitations to this study. First, the results obtained from cross-sectional data in this study can hardly provide strong causal evidence. Although this study considered the selection bias in the construction of peer relationships and tried to overcome it using school fixed effect and IPW method, it did not examine the long-term effect of peer deviant behaviour on adolescent academic performance. Second, the mediation channels discussed in this study were proximal factors of adolescent academic performance, and the process of peer effects on adolescent academic performance may have more remote pathway mechanisms worth exploring, such as health behaviour and psychological factors. Thus, it is encouraged that future research can delve deeper into the aforementioned issues based on longitudinal design and data.

Statement regarding informed consent

The CEPS data are publicly available, the survey was approved, and informed consents were obtained from both adolescents and their parents.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Junfeng Jiang

Junfeng Jiang is an associate research fellow at School of Sociology at Central China Normal University. His research interests include medical sociology and educational sociology.

Notes

1. $Cohen{\hspace{0.17em}}_{\hspace{0.17em}}^{\prime }sd=\left(x_1-x_2\right)/s$, Cohen (1988) recommended effect size benchmarks of small, medium, and large for effect size of d = 0.3, d = 0.5, and d = 0.8, respectively. In this study, $x_1$ meant the average of adolescent academic performance in those without peer deviant behaviour, $x_2$ meant the average of adolescent academic performance in those with peer deviant behaviour, and $s$ meant the combined standard deviation of the above two groups.

2. $AdjustedCohen{\hspace{0.17em}}_{\hspace{0.17em}}^{\prime }sd=2r/\sqrt{1-r^2}$. For this adjusted d, Cohen (1988) recommended effect size benchmarks of small, medium, and large for effect size of d = 0.1, d = 0.3, and d = 0.5, respectively. In this study, r meant the Pearson’s correlation coefficient between adolescent academic performance and peer deviant behaviour.

Appendices

Appendix 1 Basic information of the sample

Table

Variables	n (%)	mean (SD)	range
Sex			0–1
girls	9045 (49.1)
boys	9375 (50.9)
Age		13.88 ± 1.33	11 ~ 18
Grade			0–1
7^th	9741 (52.9)
9^th	8679 (47.1)
Family economic status			0–2
low	3822 (20.8)
middle	13446 (73.2)
high	1107 (6.0)
Paternal educational year		10.86 ± 3.07	0 ~ 19
Only child			0–1
no	10358 (56.2)
yes	8062 (43.8)
Migrant child			0–1
no	15138 (82.6)
yes	3179 (17.4)
School-boarding			0–1
no	12487 (67.8)
yes	5933 (32.2)
Parental supervision		2.43 ± 2.06	0 ~ 6
Intergenerational daily interactions		0.01 ± 1.00	−2.01 ~ 1.92
Intergenerational daily activities		0.01 ± 1.00	−1.63 ~ 2.41
Adolescent deviant behaviour		−0.02 ± 0.98	−0.65 ~ 6.18
Adolescent educational expectation		3.89 ± 1.51	0 ~ 6
Peer deviant behaviour (continuous)		0.90 ± 1.94	0 ~ 14
Peer deviant behaviour (binary)			0–1
no	12246 (66.5)
yes	6174 (33.5)
Academic performance		70.16 ± 8.58	0 ~ 100

SD denoted standard deviation.

 

Appendix 2 Distribution of the outcome and residuals

Appendix 2 shows that both academic performance variable and the residuals (based on Model 2 in Table 1) approximately followed a normal distribution. Thus, the normality assumption is considered to be supported.

 

Appendix 3 Scatter plot of deviant peer behaviour and academic performance

Appendix 3 shows that the scatter plot between peer deviant behaviour and academic performance showed an approximate linear relationship, no obvious curve relationship was exhibited. In addition, I also tried to add the squared peer deviant behaviour based on the Model 2 in Table 1 to test whether a quadratic relationship existed. The model result shows that the coefficient of squared peer deviant behaviour was 0.021 (SE=0.0.013, p=0.11), suggesting no quadratic relationship. In this case, a higher order curve relationship is probably impossible. Thus, the linearity assumption is considered to be supported.

 

Appendix 4 Test of homoscedasticity and residual independence assumptions

Table

	Model 1 OLS		Model 2 OLS+R_SE		Model 3 Robust OLS
	β	SE	β	R_SE	β	SE
Peer deviant behaviour	−0.464***	0.036	−0.464***	0.042	−0.462***	0.036
Controls	Yes		Yes		Yes
School fixed effect	Yes		Yes		Yes
Heteroscedasticity test	385.05***
Ramsey Reset test	4.97***		4.97***
n	15560		15560		15560

***p < 0.001, **p < 0.01, *p < 0.05. SE meant standard error. R_SE meant robust standard error. Heteroscedasticity test was based on the Breusch-Pagan test. Ramsey Reset test was used to test whether there was omitted variables in the model, and a significant estimate meant the residual dependence.

Model 1 in Appendix 4 shows that the OLS model exhibited significant heteroscedasticity (Chi2 in heteroscedasticity test=385.05, p<0.001). Thus, this study adopted the “OLS+robust SE” strategy in the results section. This strategy allows for the presence of heteroscedasticity and provides more reliable estimations. Furthermore, the robust OLS model is also used in Model 3, which assign higher weights to points near the fitted line to overcome potential heteroscedasticity problems caused by potential outliers. The results indicate that the estimate in robust OLS model is nearly identical to those in Models 1 and 2. That is to say, although the homoscedasticity assumption is not met, it does not significantly affect the estimation of this research.

Furthermore, the omitted variable test for Models 1 and 2 suggests the presence of important omitted variables in both models (F=4.97, p<0.001). This implies that the distribution of residuals cannot be considered completely independent, violating the assumption of residuals independence. Therefore, in Model 4 of Table 1, this study reports the results of PSM, which helps address the issue of non-independent residuals (also called endogeneity problem). The PSM estimation shows a weaker effect (-0.968) compared to the estimate in Model 3 of Table 1 (-1.187, a reduction of about 20%). This suggests that this potential endogeneity problem leads to an overestimation of the negative association between peer deviant behaviour and academic performance. However, overall, the PSM estimation still indicates a significant negative relationship. Thus, it is concluded that the issue of non-independent residuals does not significantly affect the fundamental conclusions and judgements in this study.