Life stage of boarding at school and middle school student victimization in rural China

Abstract

Boarding at school is a prevalent arrangement for rural students in China. Previous studies about boarding at school and students’ victimization experiences seldom considered the life stages when the children are first placed in boarding facilities. Using a unique data set collected from 26 junior high schools in rural China, we discovered that boarding at school is associated with increased victimization. Further analysis shows that such vulnerability mainly applies to boys. Furthermore, earlier exposure to boarding, such as being placed in a boarding facility at the primary school stage, has a stronger positive association with victimization than exposure to boarding at a later stage, such as at secondary school, indicating the importance of starting life stage of boarding when examining the boarding outcome. Our attempt at the mediation analysis of attachment to schools and families shows that boarding at school reduces students’ school attachment, making it a potential factor for their susceptibility to victimization. This study contributes to understanding the consequences of boarding at school on student well-being under the national school merging initiatives in China in the past decades.

Introduction

Due to a huge rural-urban education gap in China, improving the quality of rural schools has been a long-time national concern (Hannum and Adams 2009). Moreover, with the relaxation of hukou restrictions and the increase in rural-to-urban migration, many rural migrants bring their children. Rural-to-urban migration has resulted in declining school-age children in rural areas, despite some parents choosing to leave their children behind in their rural homes (Han 2018). The widespread adoption of boarding schools is one of several strategies to promote rural education and tackle reducing school-age children in rural areas (Tong, Gan, and Wen 2024). The provision of boarding facilities by schools brings together children from dispersed villages into a central location, which can leverage economies of scale. Since the late 1990s, boarding schools have been extended to the level of primary education as part of the School Merger Policy (Liu et al. 2010), which aims to improve the quality of primary school education by shutting down small schools and concentrating students in large and better-equipped schools. Between 1997 and 2006, over two hundred thousand rural primary schools were closed down (Xiong 2009; Liu et al. 2010), and the number of junior high schools was reduced by 10 percent between 2000 and 2006 because of the school merging. This school merging has increased the boarding student population. Although the pace of merging has largely slowed and even halted in the last decade, boarding school students have been prevalent in rural China. From 2006 to 2015, the percentage of boarders increased from 19.4% to approximately 28% (Wu and Hou 2017). In 2018, China’s central government launched a nationwide program to improve the facilities, teacher resources, and management of boarding schools in rural areas (The State Council of the People’s Republic of China 2018).

Boarding school, also known as residential education, is a setting that integrates both home and school life (Lee and Barth 2009). Nevertheless, boarding schools have historically been associated with incidents of bullying and victimization, partly due to the perception that they serve as a refuge for children from underprivileged backgrounds (Chui and Chan 2015; Colmant et al. 2004; Lee and Barth 2009; Pfeiffer and Pinquart 2014). Despite this, there have been limited studies investigating the victimization experiences of children in situations where boarding schools are not intended to provide a means of escape from dysfunctional families. China provides such a case. Boarding schools in rural China possess distinctive characteristics that are essential when investigating the relationship between boarding and student victimization, especially in contrast to their urban or Western counterparts. A primary reason for students residing at boarding schools is the geographical remoteness of their homes from educational facilities. Consequently, many rural schools host a heterogeneous population of boarders and day students, often dictated by the varying proximities of students’ homes to the school. These institutions typically operate under a quasi-military regimen, which presents a unique context for the boarding experience. Additionally, the widespread implementation of school consolidation initiatives has resulted in an increasing number of children beginning their boarding experience as early as in their primary educational phase. Furthermore, the boarding quality in these rural areas has been documented as substandard, with deficiencies in infrastructure and services (Luo et al. 2009), and notably lacking in meeting young boarders’ nutritional and emotional needs (Liu and Villa 2020). This shortfall in care provision has the potential to yield long-term detrimental effects on the well-being of the children in these environments.

However, prior research on students’ experiences of boarding and victimization has rarely considered the life stage at which students were initially enrolled in boarding schools. Separation from parents at an earlier age has been associated with an increased risk of problematic behavior (Chen and Jiang 2019; Howard et al. 2011; Mok et al. 2018) and symptoms of anxiety and depression (Liu, Li, and Ge 2009), which may heighten the likelihood of experiencing bullying victimization. In addition to the immediate impact, early exposure to such risk factors may have an accumulative and enduring effect on children, perpetuating their vulnerability to future victimization. Conversely, when children begin boarding school comparatively later, their increased maturity and preparedness may help reduce their vulnerability to victimization. Building on prior literature and situating our study in rural China, the primary objective of this study is to examine whether boarding at school leads to higher levels of victimization among middle school children. If so, does the life stage at which children are first placed in boarding school matter for subsequent victimization, and what are the potential mechanisms? In addition, boys and girls may face different risks of victimization since they are often involved in different routine activities/lifestyles and have different paces on maturity in child development. It is essential to ascertain which gender is more vulnerable in this context to facilitate the implementation of targeted interventions designed to mitigate victimization.

Using a unique data set collected from 26 junior high schools in three provinces of rural China, our study advances the literature by using a life-stage-specific approach to investigate how the timing of children’s initial exposure to boarding may be associated with their subsequent victimization experiences. Through this approach, we provide theoretical insights and policy implications for enhancing the well-being of children in China and other developing countries facing similar social challenges.

Theoretical background

The theoretical link between boarding at school and victimization

Previous research conducted in several countries has indicated that ­students who attend boarding schools are more prone to experiencing victimization than those who attend day schools (e.g., Hu, Xue, and Han 2021; Huang, Wu, and Song 2017; Pfeiffer and Pinquart 2014; Wu and Hou 2017). The primary line of explanation is rooted in lifestyle/routine activity theory (LS/RAT) in victimology (Cohen and Felson 1979). This theory emphasizes the contextual factors (i.e., time and space) in terms of the intersection of three elements temporally and spatially: the presence of motivated offenders, suitable victim targets, and an absence of capable guardians (Branic 2015). It posits that the likelihood of victimization will grow when those three elements converge; conversely, the vulnerability to victimization will decrease without any one of them (Cohen and Felson 1979). Certain non-deviant, risky, and commonplace activities and lifestyles, such as socializing with peers and sneaking out of the house at night, often place adolescents in situations where they are more likely to become targets of victimization. This is because these routines and lifestyles entail a reduction of social guardianship from parents or other adults and increased proximity to motivated delinquents (Antunes and Ahlin 2017; Cheung and Zhong 2022).

This theoretical framework is relevant to understanding the experiences of adolescents in boarding schools in rural China, as it involves a shift in their routine activities and lifestyles. The reason is twofold. First, boarding at school increases exposure to the peer environment. Due to the absence of parental monitoring, boarding students typically spend considerably more time with their peers than with adults. As a result, they may become more closely associated with motivated delinquent peers, encounter provocative behavior, or even become involved in interpersonal conflicts (Pfeiffer and Pinquart 2014; Schreck and Fisher 2004). Under such circumstances, boarding students may be more susceptible to becoming the target of bullying or delinquency. Based on self-reported bullying and victimization in Germany, Pfeiffer and Pinquart (2014) show that adolescents at boarding schools report higher levels of bullying victimization in comparison to their counterparts in day schools. Hu, Xue, and Han (2021) study reveals a similar finding in China. Overall, past studies suggested that boarding at school is positively associated with juvenile victimization experiences.

Second, guardianship should be conceptualized as encompassing both the physical presence and the emotional availability to children who are enrolled in boarding schools. The perception of inadequate guardianship can precipitate feelings of loneliness and insecurity among these students. In this context, examining the impact of boarding on familial and school attachment provides a novel perspective for understanding social guardianship within boarding environments. It is commonly posited that strong attachment to either family or school functions as a protective factor for student well-being, correlating with a reduced likelihood of victimization. However, competing arguments regarding the impact of boarding on such attachment exist. At first glance, boarding at school could strengthen a child’s sense of belonging due to the extended time spent in the school environment. Conversely, there is a concern that teachers in rural China may not be sufficiently prepared to offer the necessary support. Without well-defined protective standards, this could lead to inadequate prevention of victimization and a shortfall in assisting boarders with the coping mechanisms needed to deal with such experiences (Fredrick et al. 2021; Yin et al. 2017). Thus, the quality of the boarding facility and the care provided by teachers may significantly influence students’ sense of belonging to the school. In rural China, boarding facilities frequently fall short in terms of adequacy (Liu and Villa 2020). This may have a negative influence on student’s attitudes toward school. Indeed, existing research in China shows that rural children, especially boys, have a worse attitude toward campuses when they are in boarding schools (Wang and Mao 2018).

For whether boarding at school can enhance or erode family bonding, one line of research argues that separation from parents by boarding at school during adolescence is constructive to adolescents’ emotional attachment to their parents (Liu et al. 2017). Adolescence is a stage in which self-autonomy is developed, and it can be stressful for parents and children if there is a conflict between children’s pursuit of self-autonomy and parental control. As a result, boarding at school can offer children time and room to relax, and such emotional separation is beneficial for children to develop healthier relationships with families. Another source of parent-child stress comes from the academic supervision of parents. Due to the fierce competition in China’s education system, Chinese culture greatly emphasizes children’s educational achievement (Xing, Leng, and Ho 2021). If students board at school, parents can be fully substituted for the supervision of homework, preventing any potential parent-child conflict due to homework.

On the contrary, another research perspective proposes that early parental separation and boarding at school from a young age may lead to psychological detachment, a condition frequently linked to a spectrum of emotional challenges and stress in managing peer interactions. (Chen, Chen, and Zhao 2020; Wu, Song, and Huang 2016). All these factors could heighten the susceptibility to being victimized at school. The reason is that the parent-child bond has not been well established in early childhood. Such separation may make children feel distant and alienated from their parents, so they would be reluctant to communicate with parents when they have issues, such as victimization, at boarding schools. On this account, it is crucial to consider the life stage when the students are first placed in boarding schools.

The relevance of life stage when starting boarding

In previous studies, boarding status has typically been evaluated as a static binary variable, indicating whether a student is currently boarding at school or not. The timing of a student’s initial enrollment in a boarding school has received little attention in the literature. The significance of this research gap is underscored by the importance of the timing of family separation and the age at which children are separated from their parents, as emphasized by child development theories, including attachment theory (Bretherton 1992). According to Bowlby’s (1977) and Little’s (1965) work, adults who experienced parental separation or loss during their early childhoods are more likely to exhibit behavior problems such as aggression, and they have a higher tendency to commit crimes. Following the attachment theory, Troy and Sroufe (1987) also found a higher likelihood of victimization among preschool students with an “avoidant” attachment style, which is allegedly induced by a problematic parent-child relationship or parental deprivation. Drawing on the ­accumulated disadvantages associated with early parental deprivation, we anticipate that earlier exposure to boarding may be linked to more severe levels of victimization.

One of the few exceptions in the literature that has examined the relationship between the life stage of boarding and student outcomes is Wu and Hou (2017) study, which utilized data from selected provinces in China. The study revealed that boarding at school before the third grade has a considerably more negative impact on child development than boarding after the third grade. Nonetheless, this study’s focus on primary school students precludes an exploration of whether there is a cumulative effect of earlier exposure to boarding on the victimization experiences of adolescents following their transition to secondary school. As separation from parents at an earlier age is associated with an increased likelihood of problematic behaviors (Chen and Jiang 2019; Howard et al. 2011; Mok et al. 2018) and symptoms of anxiety and depression (Liu, Li, and Ge 2009), these effects may accumulate and persist throughout adolescence, potentially exerting an enduring impact. Therefore, it is plausible that family life deprivation resulting from earlier exposure to boarding may have lasting implications for children, including an elevated risk of victimization when compared with later exposure to boarding.

Gender heterogeneity

A significant body of research has demonstrated that victimization, particularly school-based victimization, is gendered (Deryol, Wilcox, and Dolu 2021), albeit the role of boarding is barely considered in this line of literature. Two primary explanations can be posited for the observed differences between boys and girls. Firstly, divergent growth spurts and maturation patterns between genders play a significant role. The child development literature consistently indicates that boys typically exhibit a slower trajectory in developmental milestones. This is particularly evident in the comparative analyses of similar age cohorts, where boys often lag behind girls in adhering to instructions and exercising self-control (Hamama and Hamama-Raz 2021). Second, scholars of lifestyle/routine activity theory (LS/RAT) provide a perspective highlighting gender-specific risk profiles for victimization, closely related to the distinct routine activities associated with each gender (Wilcox, Tillyer, and Fisher 2009). For instance, prior studies found that boys tend to confront overt victimization (physical attack and verbal threat), whereas girls tend to receive indirect (or relational) aggression (Biggs et al. 2010; Card et al. 2008; Crick and Grotpeter 1995; Khong et al. 2020; Wang et al. 2010). Despite a dearth of evidence on gendered routine activities, previous studies on extracurricular routine activities may shed some light on this issue. For example, research has shown that participation in clubs and interscholastic sports, which involve competition between teams from different schools, reduces the likelihood of victimization for females compared to males; conversely, participation in intramural sports (i.e., competition between teams or students within the same school) boosts the odds of victimization for females versus males (Lasky et al. 2021; Peterson et al. 2018; Popp and Peguero 2011). The reason is that girls who participate in interscholastic sports may receive better supervision from coaches and other school personnel than those who participate in intramural sports.

Additionally, social norms and expectations often dictate how boys and girls should behave, which can influence the type of victimization they encounter (Popp and Peguero 2011). For instance, boys are often socialized to be more aggressive, and this might lead them to experience or engage in more physical forms of victimization. Girls, on the other hand, might be socialized to be more relational and could experience more relational aggression, such as gossiping, exclusion, and social manipulation. Indeed, research has suggested that the gender differences in the impact of routine activities on victimization depend on the type of victimization experienced, with gender being more important in explaining the risk of violent victimization than property victimization (Peterson et al. 2018; Popp and Peguero 2011). This is because the gender of victim targets may be more closely related to the psychological or social status rewards conferred by violent delinquency than the economic rewards conferred by property delinquency (Popp and Peguero 2011). Taken together, it is reasonable to expect that gender may condition the influence of boarding at school on student victimization and the types of victimization they encounter.

Hypotheses

We set up our hypotheses according to the flow of our theoretical arguments. We started with the current boarding status and victimization among middle school students. To examine the life stage of starting boarding, we considered more nuanced measures on the timing when children were first placed in boarding school and their effects on student victimization. Next, we proposed attachment to school and family as the social guardianship mechanism to protect against student victimization beyond the increased exposure to the peer environment in boarding schools. In other words, we sought to elucidate whether such attachment mediates the relationship between boarding and victimization. Notably, we also explored competing hypotheses regarding the nature of attachment to school and family in the context of boarding at school. Finally, we anticipated the gender heterogeneity in the association between boarding experiences and victimization. Our hypotheses are as follows.

• Hypothesis 1: The status of currently boarding at school is associated with more victimization among middle school students.

• Hypothesis 2: Exposure to boarding school at an earlier life stage is associated with more victimization among students compared to exposure at a later life stage.

• Hypothesis 3a: Boarding at school either increases or reduces students’ attachment to school, which is associated with less victimization among students.

• Hypothesis 3b: Boarding at school either increases or reduces students’ attachment to family, which is associated with less victimization among students.

• Hypothesis 4: There are gendered differences in the relationship between boarding at school and victimization among students. Given that boys’ readiness for boarding at school tends to be later than that of girls, the positive association between boarding and student victimization is stronger among boys than girls. This gender difference also applies to the relationship between boarding at school and various types of victimization.

Method

Data

We drew on data collected from 26 rural middle schools in three provinces of China, including Anhui, Hunan, and Shaanxi, located in the eastern, central, and northwestern regions, respectively. In each province, three counties were sampled by their level of economic development (low, medium, and high). Within each county, three rural towns with lower, middle, and higher ranks were identified for selecting their middle schools according to their economic development levels. All schools we selected are public schools. Within each school, we randomly selected one class from each grade (Grades 7–9). Nine middle schools (3 counties * 3 schools) from each province were selected except for one county in Shaanxi province, where only two schools were chosen in this particular county due to budget limitations. This missing school resulted in a loss of three classes in the sample. In Anhui, two classes declined to participate in our interview due to exams and other reasons. As a result, the total number of schools is 26 (9 + 9 + 8), and the total number of classes is 76 (24 schools*3 classes + 2 schools*2 classes). The authors’ university approved the research ethics.

A structured and anonymous questionnaire survey was conducted in the classroom in each selected class from the selected schools. In each class, two to three research assistants were assigned to assist the students in completing the questionnaires. All three provinces fall within the middle level of economic development in China and are significant sources of rural-to-urban migration. As a result, about half of the students in the sampled schools were from families with at least one migrant parent (as shown in Table 1). Approximately three-quarters of the sampled students were boarding at school at the time of the survey, partly due to their parents’ out-migration and/or the non-walkable distance between their home and the school. A total of 2,777 students were interviewed. About 20% of students had some missing values in the variables we used in the study. We adopted multiple imputations, with 20 imputations, to address the missing values. However, not all variables were successfully imputed due to the non-converging issue when imputing the nominal variables. The sample size after the imputation varies for each variable, varying from 2,672 to 2,777 (as shown in Table 1). This is sufficient compared to the sample size of variables before the imputation, ranging from 2,174 to 2,777. We also put the number of sample sizes before the imputation in brackets in Table 1.

Table 1. Measurements and descriptive statistics.

Variable Name	N^a	Mean (%)		S. D.	Measurements
Victimization (overall)	2,728 [2,727]	1.777		2.226	Count measure (range: 0–9)
Violent Victimization	2,740 [2,739]		1.153	1.494	Count measure (range: 0–5)
Property Victimization	2,755 [2,755]		0.307	0.585	Count measure (range: 0–2)
Sexual Victimization	2,755 [2,755]		0.338	0.567	Count measure (range: 0–2)
Current Boarding Status	2,764 [2,764]		0.727	0.446	Binary variable (0 = no; 1 = yes)
Starting Stage of Boarding	2,777 [2,724]		1.151	0.780	Nominal variable (0 = never boarded 23%; 1 = boarding experience started in primary school 35%; 2 = boarding experience started in junior high school 42%)
Female	2,763 [2,763]		0.513	0.500	Binary variable (0 = male; 1 = female)
Parental Migration Status	2,777 [2,174]		1.048	1.226	Nominal variable (0 = both stay at home 49%; 1 = only father migrates 22%; 2 = only mother migrates 5%; 3 = both migrate 24%)
Co-residence with Grandparent(s)	2,765 [2,765]		0.544	0.498	Binary variable (0 = no; 1 = yes)
Sibship Size (including the respondent himself/herself)	2,777 [2,704]		2.246	0.800	Ordinal variable (1 = only child; 2 = 1 sibling; 3 = 2 siblings; 4 = 3 or more siblings)
Grade	2,777 [2,777]		7.960	0.810	Nominal variable (1 = Grade 40.4%; 7; 2 = Grade 8 32.7%; 3 = Grade 9 26.9%)
Local Hukou Status	2,737 [2,737]		0.947	0.225	Binary variable (0 = no; 1 = yes)
Relative Family SES	2,777 [2,704]		2.014	0.713	Ordinal variable (1 = lower; 2 = similar; 3 = higher)
Parental Highest Educational Attainment (years)	2,777 [2,593]		9.503	2.272	Interval-ratio variable (range: 1.963–19)
Commuting Time from Home to School (hours)	2,777 [2,754]		0.490	0.495	Interval-ratio variable (range: 0–9)
Attachment to Family	2,676 [2,676]		3.409	0.906	Interval-ratio variable (range: 1–5; the average of respective items; the range of each item: 1 = never to 5 = always)
Attachment to School	2,693 [2,693]		4.107	0.680	Interval-ratio variable (range: 1–5; the average of respective items; the range of each item: 1 = fully disagree to 5 = fully agree)
School-level Parental Highest Educational Attainment (years)	2,777 [2,777]		9.520	0.671	Interval-ratio variable (range: 8.103–10.616)
Merged School	2,777 [2,777]		0.424	0.494	Binary variable (0 = no; 1 = yes)

^aThe number outside the bracket is the sample size of variables after the imputation, whereas the number inside the bracket is the sample size of variables before the imputation.

There are several unique advantages to using this dataset to address our research questions. First, this survey was designed to understand rural adolescent students’ boarding experiences, physical and psychological well-being, and subjective attitudes. In particular, the survey contains an education history section that captures detailed information about students’ boarding experiences since the primary school. Second, the survey included a section dedicated to understanding the victimization experiences of rural Chinese adolescents. The setting of schools in the present study may differ vastly from previous studies on boarding schools in Western countries, as boarding students are less selected based on their family background than students attending day schools in the same region. In addition, students often have a choice not to board at schools.

Variables

Dependent variable

The dependent variable is the self-reported victimization experienced by students in the past twelve months. We collected information on three types of victimization: violent victimization, property victimization, and sexual victimization. The five items assessing violent victimization captured the frequency of being threatened with violence, being physically assaulted, being threatened or forced to do something, and being attacked because of the respondent’s appearance and family background. The two items on property victimization inquired about the frequency of having property stolen and being robbed. Lastly, the two items addressing sexual victimization gauged the frequency of being sexually harassed (i.e., the respondent was afraid or made to feel bad when someone else said sexually indecent things to him/her) and being sexually assaulted (an adult of the opposite sex intentionally touched the respondent’s body, such as thighs, buttocks, and chest, or kissed the respondent by using force). All items were measured on a Likert scale, with responses to each of the nine items ranging from “never” (scored 1) to “often” (scored 4). The higher the value, the more frequently the respondent was victimized. The Cronbach’s alpha is 0.829, showing high internal consistency. Since the frequency of victimization is highly skewed, we first recorded each item as a binary one, with “never” as 0 and the other three answers as 1. Then, we combined the nine items to generate a count variable indicating the overall victimization, ranging from 0 to 9. Three types of victimization (i.e., violent victimization–consisting of five items; property victimization–consisting of two items; and sexual victimization–consisting of five items) were coded in the same way.

Key independent variables

The first set of key independent variables consists of the student’s current boarding status in middle school and the life stage at which the student is initially placed in boarding schools (hereafter, the starting stage of boarding). Current boarding status is a binary variable indicating whether the respondent was boarding or not in their junior high school when the interview was conducted. The starting stage of boarding incorporates the variation in life stages by considering students’ boarding history since primary school. It is a nominal-level variable with three categories, indicating the timing of the student’s first enrollment in boarding school. The three categories are “never boarded,” “boarding experience started in primary school period,” and “boarding experience started in junior high school period.” Gender is also another key independent variable.

The respondent’s attachment levels to school and family are the two mediating variables. Attachment to school was measured by nine items, such as “feeling happy with school life,” “having a sense of belonging,” “having close relationships with teachers,” and others (Cronbach’s alpha = 0.809). Attachment to family was measured by ten items, such as whether family members “seek help from each other,” “are willing to do things with each other,” “have close relationships,” and others (Cronbach’s alpha = 0.893). In line with lifestyle/routine activity theory, these two variables often indicate social guardianship, which can allay adolescents’ vulnerability to victimization (Cheung and Zhong 2022; Han, Zhang, and Zhang 2017; Posick and Rocque 2015).

Control variables

To minimize the estimation bias induced by the confounding factors associated with boarding experiences and victimization, we controlled for a number of individual and school-level variables in the model. For the individual-level variables, we controlled for the respondent’s local hukou status (whether the respondent resides in his or her household registration place, which is tied to a person’s social welfare entitlement). We used local hukou status instead of rural/urban hukou because most students are rural hukou, and there are few beneficial differences between rural and urban hukou in rural areas of China. Since the data did not contain any objective measures of the respondent’s family socioeconomic status (SES), we included the relative family SES perceived by the respondent (the subjective rating of the respondent’s family SES compared with that of his or her neighbors in the village), and the highest level of educational attainment (in years) of either parent, as proxies for the respondent’s family SES status. Parental migration status was measured by four categories. The reference category “both parents stay at home” indicates that the respondent’s parents currently reside with him or her. The category “only father migrates” indicates that the father has out-migrated, but the mother stays at home, whereas “only mother migrates” refers to the opposite situation. The category “both parents migrate” refers to the situation in which the respondent’s mother and father have out-migrated. Other individual-level control variables included grade (Grades 7–9), sibship size (including the respondent him/herself), whether the respondent co-resides with his/her grandparent(s), and time spent on commuting from home to school (in hours).

Regarding the school-level variables, we aggregated the highest educational attainment (in years) of parents in each school to assess the school SES. We considered whether the school is merged after the consolidation process noted in the paper’s front part.

Table 1 displays the measurements and the descriptive statistics of all the variables. The mean count of victimization is 1.777, and close to three-quarters of respondents are currently boarding at school. For the starting stage of boarding, about 35% of students had boarded when they were in primary school. More details about other variables can be found in Table 1.

Analytical approach

Two steps of data analysis are involved to achieve the research objectives. For the first step, we adopted the two-level negative binomial model to estimate how the current boarding status and the starting stage of boarding are associated with victimization and whether the association varies by the student’s gender. There are three reasons for adopting this estimation method. First, since the dependent variable is a count measure and over-dispersed, the negative binomial model is more appropriate for the analysis than other methods, including Poisson regression. Second, considering the hierarchical structure of the data with the individual-level information nested within schools, we specified individual students at the first level and the school at the second level (Raudenbush and Bryk 2002). Third, as our focus is not on examining the cross-level interactions between individual boarding at school and the school-level characteristics, we employed the random intercept modeling by only allowing the intercept to vary by school but not the coefficients of current boarding status and the starting stage of boarding. The following three equations can express the first step of analysis. Equation (1)(1) $$V_{\mathit{\text{ij}}}={\beta }_{j0}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+\sum _{k=1}^{10}{\beta }_{\mathit{\text{kj}}}{\mathit{\text{INDVARS}}}_{\mathit{\text{kij}}}+\sum _{m=1}^2{\beta }_{\mathit{\text{mj}}}{\mathit{\text{SCHVARS}}}_{\mathit{\text{mij}}}+{\epsilon }_{\mathit{\text{ij}}}$$(1) illustrates the individual-level model, indicating the effect of boarding at school (i.e., current boarding status and the starting stage of boarding) on victimization. Equation (2)(2) $$\begin{array}{l}{V}_{\mathit{\text{ij}}}={\beta }_{j0}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+{\beta }_{j3}{B}_{\mathit{\text{ij}}}\times G_{\mathit{\text{ij}}}+{\displaystyle \sum }_{k=1}^{10}{\beta }_{\mathit{\text{kj}}}\mathit{\text{INDVAR}}{S}_{\mathit{\text{kij}}}\\ +{\displaystyle \sum }_{m=1}^2{\beta }_{\mathit{\text{mj}}}\mathit{\text{SCHVAR}}{S}_{\mathit{\text{mij}}}+{\epsilon }_{\mathit{\text{ij}}}\end{array}$$(2) incorporates the interaction term between two variables of boarding at school and gender into the model. Equation (3)(3) $${\beta }_{j0}={\gamma }_{00}+{\mu }_{j0}$$(3) illustrates the school-level models. (1) $$V_{\mathit{\text{ij}}}={\beta }_{j0}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+\sum _{k=1}^{10}{\beta }_{\mathit{\text{kj}}}{\mathit{\text{INDVARS}}}_{\mathit{\text{kij}}}+\sum _{m=1}^2{\beta }_{\mathit{\text{mj}}}{\mathit{\text{SCHVARS}}}_{\mathit{\text{mij}}}+{\epsilon }_{\mathit{\text{ij}}}$$(1) (2) $$\begin{array}{l}{V}_{\mathit{\text{ij}}}={\beta }_{j0}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+{\beta }_{j3}{B}_{\mathit{\text{ij}}}\times G_{\mathit{\text{ij}}}+{\displaystyle \sum }_{k=1}^{10}{\beta }_{\mathit{\text{kj}}}\mathit{\text{INDVAR}}{S}_{\mathit{\text{kij}}}\\ +{\displaystyle \sum }_{m=1}^2{\beta }_{\mathit{\text{mj}}}\mathit{\text{SCHVAR}}{S}_{\mathit{\text{mij}}}+{\epsilon }_{\mathit{\text{ij}}}\end{array}$$(2) (3) $${\beta }_{j0}={\gamma }_{00}+{\mu }_{j0}$$(3)

In Equations (1)(1) $$V_{\mathit{\text{ij}}}={\beta }_{j0}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+\sum _{k=1}^{10}{\beta }_{\mathit{\text{kj}}}{\mathit{\text{INDVARS}}}_{\mathit{\text{kij}}}+\sum _{m=1}^2{\beta }_{\mathit{\text{mj}}}{\mathit{\text{SCHVARS}}}_{\mathit{\text{mij}}}+{\epsilon }_{\mathit{\text{ij}}}$$(1) and (2)(2) $$\begin{array}{l}{V}_{\mathit{\text{ij}}}={\beta }_{j0}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+{\beta }_{j3}{B}_{\mathit{\text{ij}}}\times G_{\mathit{\text{ij}}}+{\displaystyle \sum }_{k=1}^{10}{\beta }_{\mathit{\text{kj}}}\mathit{\text{INDVAR}}{S}_{\mathit{\text{kij}}}\\ +{\displaystyle \sum }_{m=1}^2{\beta }_{\mathit{\text{mj}}}\mathit{\text{SCHVAR}}{S}_{\mathit{\text{mij}}}+{\epsilon }_{\mathit{\text{ij}}}\end{array}$$(2) , $V_{\mathit{\text{ij}}}$ refers to the dependent variable-victimization experiences of the ith student in the jth school. ${\beta }_{j0}$ indicates the intercept in the individual-level model, which is affected by the school-level effect of the jth school as shown in Equation (3)(3) $${\beta }_{j0}={\gamma }_{00}+{\mu }_{j0}$$(3) . $B_{\mathit{\text{ij}}}$, and $G_{\mathit{\text{ij}}}$ refer to two variables on boarding at school and gender of the ith student in the jth school respectively, and ${\beta }_{j1}$ and ${\beta }_{j2}$ are the corresponding coefficients. ${\mathit{\text{INDVARS}}}_{\mathit{\text{kij}}}$ is the generic notation of ten individual-level control variables of the ith student in the jth school and ${\beta }_{\mathit{\text{kj}}}$ is the notation of coefficients. ${\mathit{\text{SCHVARS}}}_{\mathit{\text{mij}}}$ is the generic notation of two school-level control variables of the ith student in the jth school merged into the individual level, and ${\beta }_{\mathit{\text{mj}}}$ is the notation of coefficients. ${\epsilon }_{\mathit{\text{ij}}}$ is the individual-level error term and is uncorrelated with covariates.

Equation (3)(3) $${\beta }_{j0}={\gamma }_{00}+{\mu }_{j0}$$(3) illustrates the function of how school-level factors affect the intercept of Equation (1)(1) $$V_{\mathit{\text{ij}}}={\beta }_{j0}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+\sum _{k=1}^{10}{\beta }_{\mathit{\text{kj}}}{\mathit{\text{INDVARS}}}_{\mathit{\text{kij}}}+\sum _{m=1}^2{\beta }_{\mathit{\text{mj}}}{\mathit{\text{SCHVARS}}}_{\mathit{\text{mij}}}+{\epsilon }_{\mathit{\text{ij}}}$$(1) . ${\gamma }_{00}$ and ${\mu }_{j0}$ are the intercept and the random error term of the school-level model, respectively. We only let the intercept vary by school, not the coefficient of two variables measuring boarding at school, as it is not our interest to examine the cross-level interaction.

We substituted the Equation (3)(3) $${\beta }_{j0}={\gamma }_{00}+{\mu }_{j0}$$(3) into Equations (1)(1) $$V_{\mathit{\text{ij}}}={\beta }_{j0}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+\sum _{k=1}^{10}{\beta }_{\mathit{\text{kj}}}{\mathit{\text{INDVARS}}}_{\mathit{\text{kij}}}+\sum _{m=1}^2{\beta }_{\mathit{\text{mj}}}{\mathit{\text{SCHVARS}}}_{\mathit{\text{mij}}}+{\epsilon }_{\mathit{\text{ij}}}$$(1) and (2)(2) $$\begin{array}{l}{V}_{\mathit{\text{ij}}}={\beta }_{j0}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+{\beta }_{j3}{B}_{\mathit{\text{ij}}}\times G_{\mathit{\text{ij}}}+{\displaystyle \sum }_{k=1}^{10}{\beta }_{\mathit{\text{kj}}}\mathit{\text{INDVAR}}{S}_{\mathit{\text{kij}}}\\ +{\displaystyle \sum }_{m=1}^2{\beta }_{\mathit{\text{mj}}}\mathit{\text{SCHVAR}}{S}_{\mathit{\text{mij}}}+{\epsilon }_{\mathit{\text{ij}}}\end{array}$$(2) to get Equations (4)(4) $$V_{\mathit{\text{ij}}}={\gamma }_{00}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+{\displaystyle \sum }_{k=1}^{10}{\beta }_{\mathit{\text{kj}}}{\mathit{\text{INDVARS}}}_{\mathit{\text{kij}}}+{\displaystyle \sum }_{m=1}^2{\beta }_{\mathit{\text{mj}}}{\mathit{\text{SCHVARS}}}_{\mathit{\text{mij}}}+{\mu }_{j0}+{\epsilon }_{\mathit{\text{ij}}}$$(4) and (5)(5) $$\begin{array}{l}{V}_{\mathit{\text{ij}}}={\gamma }_{00}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+{\beta }_{j3}{B}_{\mathit{\text{ij}}}\times G_{\mathit{\text{ij}}}+{\displaystyle \sum }_{k=1}^{10}{\beta }_{\mathit{\text{kj}}}\mathit{\text{INDVAR}}{S}_{\mathit{\text{kij}}}\\ +{\displaystyle \sum }_{m=1}^2{\beta }_{\mathit{\text{mj}}}\mathit{\text{SCHVAR}}{S}_{\mathit{\text{mij}}}+{\mu }_{j0}+{\epsilon }_{\mathit{\text{ij}}}\end{array}$$(5) for our final analysis: (4) $$V_{\mathit{\text{ij}}}={\gamma }_{00}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+{\displaystyle \sum }_{k=1}^{10}{\beta }_{\mathit{\text{kj}}}{\mathit{\text{INDVARS}}}_{\mathit{\text{kij}}}+{\displaystyle \sum }_{m=1}^2{\beta }_{\mathit{\text{mj}}}{\mathit{\text{SCHVARS}}}_{\mathit{\text{mij}}}+{\mu }_{j0}+{\epsilon }_{\mathit{\text{ij}}}$$(4) (5) $$\begin{array}{l}{V}_{\mathit{\text{ij}}}={\gamma }_{00}+{\beta }_{j1}{B}_{\mathit{\text{ij}}}+{\beta }_{j2}{G}_{\mathit{\text{ij}}}+{\beta }_{j3}{B}_{\mathit{\text{ij}}}\times G_{\mathit{\text{ij}}}+{\displaystyle \sum }_{k=1}^{10}{\beta }_{\mathit{\text{kj}}}\mathit{\text{INDVAR}}{S}_{\mathit{\text{kij}}}\\ +{\displaystyle \sum }_{m=1}^2{\beta }_{\mathit{\text{mj}}}\mathit{\text{SCHVAR}}{S}_{\mathit{\text{mij}}}+{\mu }_{j0}+{\epsilon }_{\mathit{\text{ij}}}\end{array}$$(5)

In this step, we first investigated whether the two boarding school variables, including the student’s current boarding status and the starting stage of boarding, is positively or negatively associated with victimization. Secondly, in order to examine the gender difference, we conducted the analysis by gender and used the interaction term to determine whether the gender difference in coefficients is statistically significant. Third, we analyzed the association between students’ boarding at school and three types of victimization by using the same models.

The second step is to determine whether the attachment levels to school and family mediate the association between students’ boarding at school (measured by the current boarding status and the starting stage of boarding) and victimization. We specified the negative binomial specification to deal with the dependent variable, which is a count measure, and incorporated the school-level effect into the model.

Results

The association between boarding at school and victimization experiences

Table 2 presents the count of victimization based on student’s current boarding status and the starting stage of boarding respectively, without controlling for any variables or accounting for the school effect. A statistical significance test follows. The table shows that students who are currently boarding score higher on the victimization scale than those who are not boarding. Moreover, those who started boarding during the primary school period experience the highest level of victimization, surpassing both those who have never boarded at school as well as those who started their boarding in secondary school. Interestingly, students who started boarding in junior high school have a lower level of victimization than those who never boarded at school. These findings warrant a further analysis of the effects of boarding by including the control variables with the proper regression analysis.

Table 2. Group difference of mean of victimization by the two measures of boarding.

	Mean of Victimization (count measure)	N	Pair	Group Difference^a
Current Boarding Status
Not boarding	1.700	741	Not boarding vs. Currently boarding	0.107
Currently boarding	1.807	1979
Starting Stage of Boarding
Never boarded	1.677	657	Never boarded vs. In primary school	0.444***
In primary school	2.088	994	In primary school vs. In junior high school	0.563***
In junior high school	1.552	1077	In junior high school vs. Never boarded	0.120

* p < .1^{; **}p < .05^{; ***}p < .01.

^a Using the pairwise comparisons of means to estimate the group difference.

Table 3 presents the results of the multilevel negative binomial models, which estimate the association between the current boarding status and victimization, taking into account the control variables and the school effect. The modeling started with a full sample, followed by separate analyses for female and male students, and the interaction between gender and current boarding status. In Model 1, after considering all the individual and school-level control variables, we found a positive and statistically significant association between the current boarding status and victimization experiences. That is, the difference in the logs of expected counts of victimization is expected to be 0.155 units higher for students currently boarding at school compared to those not currently boarding at school. The result suggests that boarding harms students in terms of the victimization. The result supports Hypothesis 1. The subsequent gender-specific analyses in Models 2 and 3 show that the association between the current boarding status and victimization is statistically significant only among boys. Holding the other variables constant in the model, the difference in the logs of expected counts is expected to be 0.219 units higher for boys currently boarding at school compared to boys not currently boarding at school. The inclusion of the gender and boarding interaction in Model 4 also demonstrates a similar finding, meaning that being male elevates the odds of victimization associated with boarding. The corresponding interaction term is statistically significant. This provides initial evidence for the gender difference hypothesis.

Table 3. The effect of boarding status on victimization estimated by multi-level negative binomial model.

	(1)	(2)	(3)	(4)
	Full model	Female	Male	Interaction
Current Boarding Status (ref.: Not boarding)
Currently boarding	0.155**	0.034	0.219**	0.248***
	(0.073)	(0.106)	(0.095)	(0.088)
Female (ref.: Male)	−0.314***			−0.162*
	(0.048)			(0.094)
Current Boarding Status (ref.: Not boarding) $\times$ Female (ref.: Male)
Boarding $\times$ Female				−0.208*
				(0.109)
Parental Migration Status (ref.: Both stay at home)
Only father migrates	0.019	0.005	0.036	0.024
	(0.064)	(0.091)	(0.089)	(0.064)
Only mother migrates	0.018	0.003	0.021	0.010
	(0.108)	(0.160)	(0.146)	(0.108)
Both parents migrate	0.029	−0.099	0.126	0.026
	(0.069)	(0.096)	(0.098)	(0.069)
Co-residence with Grandparent(s)	−0.004	0.038	−0.066	−0.005
	(0.052)	(0.074)	(0.074)	(0.052)
Sibship Size	0.032	0.040	0.031	0.032
	(0.032)	(0.046)	(0.045)	(0.032)
Grade (ref.: Grade 7)
Grade 8	−0.068	−0.088	−0.051	−0.070
	(0.057)	(0.082)	(0.081)	(0.057)
Grade 9	−0.276***	−0.284***	−0.273***	−0.278***
	(0.061)	(0.086)	(0.087)	(0.061)
Local Hukou Status	0.083	0.132	0.021	0.085
	(0.110)	(0.157)	(0.155)	(0.110)
Relative Family SES	−0.129***	−0.181***	−0.085*	−0.131***
	(0.035)	(0.051)	(0.047)	(0.035)
Parental Highest Educational Attainment	−0.025**	−0.027	−0.024	−0.026**
	(0.011)	(0.018)	(0.015)	(0.011)
Commuting Time from Home to School	0.060	0.076	0.093	0.059
	(0.054)	(0.086)	(0.067)	(0.054)
School-level Parental Highest Educational Attainment	0.018	−0.022	0.019	0.019
	(0.099)	(0.121)	(0.097)	(0.099)
Merged School	0.135	0.137	0.162	0.132
	(0.134)	(0.163)	(0.129)	(0.134)
Constant	0.777	1.007	0.647	0.715
	(0.941)	(1.166)	(0.940)	(0.942)
N	2,672	1,382	1,290	2,672

Note: standard error is in parentheses. *p < .1; **p < .05; ***p < .01.

In Table 4, we further investigated whether the life stage at which the student is first placed in a boarding facility is associated with victimization experiences. In Panel A, we treated the never boarded students as the reference group. Model 1 in Panel A indicates that boarding at school started in primary stage increases students’ victimization incidence comparing to those who never boarding at school, but this is not statistically significant when they started to board at school in junior high stage. The result aligns with Hypothesis 2, suggesting that having boarding experience at an earlier life stage makes children more vulnerable to victimization compared to boarding started at relatively mature ages. The gender-specific analysis shows that both starting stages of boarding increase boys’ chance of being victimized comparing to never boarded boys, suggesting boys’ vulnerability when boarding regardless of stages. Model 4 of Panel A introduces the interaction term between gender and the starting stage of boarding. It shows a significant interaction effect between gender and starting boarding from junior high schools, meaning that being girls can reduce the incidence of victimization if they started boarding from junior high schools.

Table 4. The correlation between the starting stage of boarding and victimization estimated by multi-level negative binomial model.

	(1)	(2)	(3)	(4)
	Full model	Female	Male	Interaction
Panel A:
Starting Stage of Boarding (ref.: Never boarded)
In primary school	0.210**	0.114	0.290***	0.295***
	(0.084)	(0.122)	(0.109)	(0.102)
In junior high school	0.119	−0.040	0.206*	0.245**
	(0.083)	(0.117)	(0.110)	(0.103)
Female (ref.: Male)	−0.309***			−0.137
	(0.049)			(0.101)
Starting Stage of Boarding (ref.: Never boarded) $\times$ Female (ref.: Male)
In primary school $\times$ Female				−0.189
				(0.129)
In junior high school $\times$ Female				−0.264**
				(0.128)
Panel B:
Starting Stage of Boarding (ref.: In primary school^a)
Never boarded	−0.210**	−0.114	−0.290***	−0.295***
	(0.084)	(0.122)	(0.109)	(0.102)
In junior high school	−0.091	−0.154	−0.084	−0.051
	(0.068)	(0.097)	(0.091)	(0.086)
Female (ref.: Male)	−0.309***			−0.326***
	(0.049)			(0.079)
Starting Stage of Boarding (ref.: In primary school) $\times$ Female (ref.: Male)
Never boarded $\times$ Female				0.189
				(0.129)
In junior high school $\times$ Female				−0.075
				(0.111)
Control Variables^b	Yes	Yes	Yes	Yes
Constant	0.937	1.068	0.843	0.962
	(0.909)	(1.108)	(0.906)	(0.911)
N	2,680	1,386	1,294	2,680

Note: Standard error is in parentheses^{. *}p < .1; **p < .05; ***p < .01.

^a When we examined whether the starting stage of boarding matters, we intended to highlight the differential effect of boarding on victimization across two life stages (i.e., starting in primary schools vs. starting in junior high schools). Hence, we set “in primary school” rather than “never boarded” as the reference group.

^b Control variables are controlled in the model. They are Parental Migration Status, Co-residence with Grandparent(s), Sibship Size, Grade, Local Hukou Status, Relative Family SES, Parental Highest Educational Attainment, Commuting Time from Home to School, School-level Parental Highest Educational Attainment, and School Merged Status.

To illustrate the starting stage effect between primary- and middle-school boarders, we further treated students having boarding experience during primary school as the reference group (shown in Panel B). Panel B of Model 1 shows that victimization is higher for students who “started boarding in primary school” than those who have never boarded and “started boarding in junior high school”, although the latter lacks of statistical significance. The difference in the logs of expected counts is expected to be 0.210 units lower for never-boarding students compared to those who started boarding in primary school while holding the other variables constant in the model. In Panel B of Models 2 and 3 of Table 4, the gender-specific analysis shows that girls are less likely to be victimized if they start boarding in junior high school, although the coefficient is only close to show statistical significance. The difference in the logs of expected counts is expected to be 0.154 units lower for girls who started to board in junior high school compared to those who started boarding in primary school while holding the other variables constant in the model. Boys are significantly less likely to be victimized if they have never boarded. The difference in the logs of expected counts is expected to be 0.290 units lower for boys who never boarded at school than those who started boarding in primary school while holding the other variables constant in the model. There is no significant variation in victimization observed between boys who started boarding in junior high school and those who started in primary school. These results suggest that boys have a higher tendency to be victimized regardless of when they start boarding at school, whether in primary or junior high, while starting boarding at a later stage can reduce victimization for girls. Therefore, the results further highlight that the life stage at which students start boarding matters, and the extent to which it matters varies by the student’s gender. Overall, girls are more susceptible to the timing of boarding at school than boys. Such gender differences might be attributed to the differential development of social intelligence and maturity among boys and girls (Björkqvist 1994). This evidence provides support that the gender variation in boarding and victimization is a function of the life stages when children are initially placed in boarding facilities.

Mediating effects of school and family attachment

In Tables 5 and 6, we estimate the mediating effect of attachment to school and family in the association between boarding at school and victimization. As depicted in Table 5, the current boarding status is not associated with family attachment across the full and gender-specific samples, although family attachment itself plays an important role in abating the victimization experience. In contrast, currently boarding significantly reduces attachment to school, while attachment to school can reduce the victimization experience. This mediating effect is observed in the full and the gender-specific samples. Table 6 reports the mediating effects of attachment to family and school on the association between the starting stage of boarding and victimization. Similar to the findings shown in Table 5, the family bond does not mediate the association between the starting stage of boarding and victimization, although the family bond itself significantly decreases victimization. In contrast, beginning boarding in both primary and junior high schools undermine school attachment. Further analysis uses primary school as the reference group (not shown) suggests that those who started boarding in primary school have weaker school attachment compared to those who started boarding in junior high school and those who never boarded at school. Furthermore, attachment to school reduces victimization in the full sample and across genders. In short, attachment to school mediates the association between the starting stage of boarding and victimization. The results altogether provide evidence for Hypothesis 3a, but not 3b.

Table 5. Multi-level model estimating the mediating effect (Current boarding status as the key independent variable).

	(1)	(2)	(3)
	Full Model	Female	Male
Dependent Variable: Victimization
Current Boarding Status (ref.: Not boarding)
Currently boarding	0.182**	0.093	0.224**
	(0.074)	(0.108)	(0.094)
Attachment to Family	−0.187***	−0.215***	−0.176***
	(0.031)	(0.045)	(0.042)
Attachment to School	−0.369***	−0.392***	−0.358***
	(0.040)	(0.060)	(0.054)
Female (ref.: Male)	−0.260***
	(0.052)
Control Variables^a	Yes	Yes	Yes
School-level Effect^b	Yes	Yes	Yes
Constant	2.347***	2.808***	2.070***
	(0.728)	(0.946)	(0.780)
Mediating Effect
Dependent Variable: Attachment to Family
Current Boarding Status (ref.: Not boarding)
Currently boarding	0.046	0.026	0.066
	(0.044)	(0.061)	(0.061)
School-level Effect	Yes	Yes	Yes
Constant	3.372***	3.427***	3.320***
	(0.062)	(0.071)	(0.070)
Dependent Variable: Attachment to School
Current Boarding Status (ref.: Not boarding)
Currently boarding	−0.085***	−0.079*	−0.099**
	(0.033)	(0.043)	(0.048)
School-level Effect	Yes	Yes	Yes
Constant	4.168***	4.226***	4.114***
	(0.046)	(0.050)	(0.057)
N	2,754	1,410	1,334

Note: Standard error is in parentheses^{. *}p < .1^{; **}p < .05^{; ***}p < .01.

^a Control variables are controlled in the model. They are Parental Migration Status, Co-residence with Grandparent(s), Sibship Size, Grade, Local Hukou Status, Relative Family SES, Parental Highest Educational Attainment, Commuting Time from Home to School, School-level Parental Highest Educational Attainment, and School Merged Status.

^b School-level effect is controlled in the model.

Table 6. Multi-level model estimating the estimating the mediating effect (starting stage of boarding as the key independent variable).

	(1)	(2)	(3)
	Full Model	Female	Male
Dependent Variable: Victimization
Starting Stage of Boarding (ref.: Never boarded)
In primary school	0.261***	0.216*	0.282**
	(0.086)	(0.123)	(0.110)
In junior high school	0.165*	0.041	0.227**
	(0.084)	(0.118)	(0.108)
Attachment to Family	−0.190***	−0.218***	−0.178***
	(0.031)	(0.045)	(0.042)
Attachment to School	−0.363***	−0.393***	−0.350***
	(0.040)	(0.060)	(0.054)
Female (ref.: Male)	−0.256***
	(0.052)
Control Variables^a	Yes	Yes	Yes
School-level Effect^b	Yes	Yes	Yes
Constant	2.251***	2.681***	1.984**
	(0.718)	(0.917)	(0.784)
Mediating Effect
Dependent Variable: Attachment to Family
Starting Stage of Boarding (ref.: Never boarded)
In primary school	0.039	−0.014	0.085
	(0.051)	(0.070)	(0.072)
In junior high school	0.074	0.049	0.104
	(0.050)	(0.067)	(0.072)
School-level Effect	Yes	Yes	Yes
Constant	3.362***	3.427***	3.297***
	(0.062)	(0.072)	(0.072)
Dependent Variable: Attachment to School
Starting Stage of Boarding (ref.: Never boarded)
In primary school	−0.115***	−0.060	−0.175***
	(0.038)	(0.051)	(0.055)
In junior high school	−0.094**	−0.112**	−0.077
	(0.037)	(0.048)	(0.055)
School-level Effect	Yes	Yes	Yes
Constant	4.186***	4.235***	4.139***
	(0.046)	(0.052)	(0.056)
N	2,763	1,414	1,339

Note: Standard error is in parentheses^{. *}p < .1; **p < .05; ***p < .01.

^a Control variables are controlled in the model. They are Parental Migration Status, Co-residence with Grandparent(s), Sibship Size, Grade, Local Hukou Status, Relative Family SES, Parental Highest Educational Attainment, Commuting Time from Home to School, School-level Parental Highest Educational Attainment, and School Merged Status.

^b School-level effect is controlled in the model.

Boarding at school and three types of victimization

Tables 7 and 8 present the results about the association between two variables measuring student’s boarding at school and three types of victimization (violent, property and sexual), with the latter considering the life stage of boarding. In Table 7, all three interaction models based on the three forms of victimization suggest that boarding is positively associated with victimization when they are boys. However, the gender-specific models only shows that boys who are currently boarding at school are more likely to experience property victimization. Overall, the three sets of models point to the boys’ tendency to victimization among all three types of victimization. Table 8 concerns the starting stage of boarding as the key independent variable. For the models analyzing the full sample, results show the importance of the life stage of boarding on violent and property victimizations. However, for sexual victimization, it seems that both stages of exposure are harmful among the full sample. The gender-specific analysis shows that only for boys, such detrimental influence of boarding in terms of three types of victimization is larger when they start boarding at primary school. This analyses further attest to Hypothesis 4 on the gender differences in the impact of boarding on the types of victimization.

Table 7. The correlation between current boarding status and types of victimization estimated by multi-level negative binomial model.

	Violent				Property				Sexual
	Full	Female	Male	Interaction	Full	Female	Male	Interaction	Full	Female	Male	Interaction
Current Boarding Status (ref.: Not boarding)
Currently boarding	0.110	−0.006	0.163	0.192**	0.182	−0.014	0.339**	0.289**	0.247**	0.208	0.203	0.293**
	(0.079)	(0.116)	(0.102)	(0.095)	(0.119)	(0.181)	(0.151)	(0.140)	(0.117)	(0.180)	(0.144)	(0.138)
Female (ref.: Male)	−0.268***			−0.132	−0.471***			−0.258	−0.373***			−0.285*
	(0.053)			(0.102)	(0.082)			(0.162)	(0.079)			(0.160)
Current Boarding Status (ref.: Not boarding) $\times$ Female (ref.: Male)
Boarding $\times$ Female				−0.188				−0.278				−0.117
				(0.119)				(0.186)				(0.182)
Control Variables^a	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Constant	0.042	0.387	−0.085	−0.010	0.187	0.868	−0.475	0.109	−1.736	−1.753	−1.544	−1.759
	(1.039)	(1.284)	(0.987)	(1.041)	(1.182)	(1.619)	(1.335)	(1.183)	(1.228)	(1.725)	(1.274)	(1.224)
N	2,684	1,384	1,300	2,684	2,698	1,391	1,307	2,698	2,696	1,390	1,306	2,696

Note: Standard error is in parentheses. *p < .1; **p < 0.05; ***p < 0.01.

^a Control variables are controlled in the model. They are Parental Migration Status, Co-residence with Grandparent(s), Sibship Size, Grade, Local Hukou Status, Relative Family SES, Parental Highest Educational Attainment, Commuting Time from Home to School, School-level Parental Highest Educational Attainment, and School Merged Status.

Table 8. The correlation between the starting stage of boarding and types of victimization estimated by multi-level negative binomial model.

	Violent				Property			Sexual
	Full	Female	Male	Interaction	Full	Female	Male	Interaction	Full	Female	Male	Interaction
Starting Stage of Boarding (ref.: Never boarded)
In primary school	0.163*	0.101	0.227*	0.229**	0.327**	0.072	0.528***	0.483***	0.284**	0.176	0.311*	0.343**
	(0.091)	(0.133)	(0.117)	(0.110)	(0.136)	(0.205)	(0.174)	(0.162)	(0.134)	(0.204)	(0.167)	(0.160)
In junior high school	0.067	−0.082	0.123	0.176	0.216	−0.063	0.425**	0.353**	0.229*	0.115	0.246	0.332**
	(0.090)	(0.129)	(0.118)	(0.111)	(0.115)	(0.200)	(0.178)	(0.167)	(0.134)	(0.199)	(0.169)	(0.161)
Female (ref.: Male)	−0.264***			−0.120	−0.463***			−0.164	−0.370***			−0.218
	(0.053)			(0.110)	(0.082)			(0.177)	(0.079)			(0.172)
Starting Stage of Boarding (ref.: Never boarded) $\times$ Female (ref.: Male)
In primary school $\times$ Female				−0.151				−0.402*				−0.149
				(0.140)				(0.218)				(0.212)
In junior high school $\times$ Female				−0.232*				−0.341				−0.240
				(0.139)				(0.220)				(0.213)
Control Variables^a	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Constant	−0.011	0.250	−0.172	−0.056	−0.064	0.813	−0.832	−0.205	−1.736	−1.611	−1.685	−1.768
	(1.011)	(1.243)	(0.959)	(1.014)	(1.155)	(1.598)	(1.299)	(1.163)	(1.222)	(1.689)	(1.293)	(1.217)
N	2,692	1,388	1,304	2,692	2,706	1,395	1,311	2,706	2,704	1,394	1,310	2,704

Note: Standard error is in parentheses. *p < .1; **p < .05; ***p < .01.

^a Control variables are controlled in the model. They are Parental Migration Status, Co-residence with Grandparent(s), Sibship Size, Grade, Local Hukou Status, Relative Family SES, Parental Highest Educational Attainment, Commuting Time from Home to School, School-level Parental Highest Educational Attainment, and School Merged Status.

Discussion and conclusion

This study examined the relationship between boarding at school and victimization experiences among rural middle school students in China. We took a novel approach to examine whether having boarding experiences in earlier versus later life stages relates to different victimization outcomes among junior secondary school students. We situated our study in the context of rural China where the school merger program has been implemented for several decades due to the decline in student population in both primary and middle schools in rural China. The results of this study inform the debate on whether boarding leads to an increase in victimization, particularly in a context where students enrolling in boarding schools are less likely to be selected based on their family dysfunction status. Importantly, they provide a broader understanding of the association between boarding and student well-being. Three major conclusions emerge from our results.

First, while our findings accord with most of the existing literature in highlighting the overall detrimental effect of boarding, they corroborate the importance of dissecting the heterogeneous effects of gender and life stage of boarding if we seek to better understand boarding students’ victimization experiences in secondary schools. Specifically, boarding at school is found to correlate with a higher level of overall victimization, with boys being more susceptible to victimization when boarding at school. When the timing of boarding is further considered, exposure to boarding school at an earlier stage in life increases a student’s vulnerability to victimization compared to being exposed at a later stage. Among boys, boarding at school during any life stages tends to experience more victimization than non-boarders. In contrast, for girls, starting to board during middle school tends to mitigate their propensity to experience victimization compared to starting to board in primary school.

These findings overall tend to favor our conjecture built on lifestyle/routine activity theory (Cohen and Felson 1979) that boarding experiences could represent a shift in routine activities and lifestyles, which in turn precipitate victimization in different genders. However, this study does not delve into specific routine activities, such as types of peer activities and extracurricular activities during boarding, that make boarding boys far more susceptible to victimization than their female peers. This merits investigation in future research when applying lifestyle/routine activity theory to understand boarding-related victimization. On the other hand, more focused arrangements should be made to prevent boys from being victimized when they are boarding at schools. Both primary and middle schools should implement interventions aimed at eliminating victimization. Moreover, when parents have a choice, they should consider that boarding at school could be more harmful to boys at both primary and middle school stages. More preventive measures should be taken beforehand when parents must send their boys to boarding schools. Parents and education policies should strive to minimize children’s enrollment in boarding schools during the early stages of life, such as the primary school age years.

Second, our results address the competing arguments as to whether boarding at school enhances or erodes social guardianship in terms of family and school attachment, which can serve as social guardianship against victimization according to lifestyle/routine activity theory. We show that boarding at school does not matter to students’ attachment to family, but it erodes their attachment to school. Also, only attachment to school mediates the positive association between boarding and victimization. When the life stage of boarding is considered, boys are more likely than girls to be affected by such mediation: beginning to board in primary school reduces boys’ attachment to school, which can prevent their victimization. Theoretically, these findings add to lifestyle/routine activity theory by illustrating the salience of school guardianship within boarding school environments. Practically, these findings suggest that education policy makers need to manage a paradoxical situation in which more time spent on school due to boarding does not really raise the sense of belonging in school. This paradox may be attributed to the boarding environment and the array of activities provided within it. Education policy makers need to develop strategies to improve the boarding experience, aiming to promote students’ happiness and boost their sense of belonging at school. These improvements are crucial in mitigating the negative aspects of boarding, including the risk of victimization. To effectively address this issue, policy initiatives should extend beyond simply upgrading the physical boarding facilities. It is equally crucial to provide a wider and more engaging range of extracurricular programs. These programs should expand the scope of the educational strategies to encompass both environmental enhancements and life-quality improvements. For instance, they can be designed to minimize contact with potentially negative peer groups and to enhance students’ school engagement.

Third, different forms of victimization should be taken into consideration when studying the gender heterogeneity in the impact of boarding. Boys are more vulnerable to all three types of victimization when they started to board in primary schools than never boarding, but they are also more likely to suffer from property victimization when boarding started from junior high schools. These results echo our proposition that when both boys and girls board at school, spend the same amount of time with peers, and receive less parental guardianship compared to non-boarding students, boys’ higher propensity to get involved in direct aggression, such as verbal and/or physical attack, might make them more likely to be victimized than girls. As we have argued at the outset of the paper, gender norms might matter in this regard. More research is needed to measure those norms and investigate how those norms may intersect with the impact of boarding on various forms of victimization.

Our study has a few shortcomings that warrant attention. Our sample focuses on three provinces, so it is limited in terms of generalizability to other places in China, as well as the rest of the world. Second, our study does not directly investigate the reasons behind the association between boarding at school and increased victimization, nor does it explore why boarding experiences in earlier life stage, i.e., in primary school, has a long-lasting effect on students’ subsequent experiences of victimization. While we can only theoretically argue that it is possibly due to earlier deprivation of family life, we are unable to identify the mechanisms in a more direct way. Additionally, while we have argued that rural boarding schools are less selective, whether in terms of family dysfunction or socioeconomic status, there is still potential selectivity among rural students who are attending boarding schools. Therefore, caution should be taken when drawing causal inference in this study. Despite these limitations, our study makes an important contribution by enhancing our understanding of the relationship between the life stage at which students are initially placed in a boarding school setting and their experiences of victimization in the context of large-scale school merging and parental out-migration. In this sense, it also provides valuable insights for future research directions in examining the well-being of boarding students.

Disclosure statement

No potential conflict of interests was reported by the authors.

Additional information

Funding

This study was supported by the General Research Fund of Research Grants Council of Hong Kong (GRF 14642816), and the Direct Grant of Research Committee, The Chinese University of Hong Kong.

Notes on contributors

Yuying Tong

Yuying Tong is a Professor of Sociology at the Chinese University of Hong Kong. Her research areas crosscut social demography, migration and immigration, family and life course, gender, and population health/well-being.

Yutong Hu

Yutong Hu is a PhD student in sociology at the Chinese University of Hong Kong. Her main research interests include the sociology of education, children and youth studies, and gender.

Nicole Wai-ting Cheung

Nicole Wai-ting Cheung is an associate professor at the Department of Sociology of the Chinese University of Hong Kong. Her research interests crosscut sociology of deviance, sociology of youth, and addiction.