Are Flunkers Social Outcasts? A Multilevel Study of Grade Retention Effects on Same-Grade Friendships

Peer Relationships in Adolescence

During recent decades, the attention paid to students’ peer relationships at school has grown substantially. A sizeable body of literature has emerged that is mostly rooted in the social learning hypothesis (Ladd, Kochenderfer-Ladd, Visconti, & Ettekal, 2012)—which gives social learning a central place in peer influence—or social capital theory (J. S. Coleman, 1988)—which considers relationships with peers as resources to be exchanged for “success.” In accordance with the expectations of both theoretical approaches, existing research points to the importance of peer relationships for school adjustment (see e.g., McDougall et al., 2001). Relationship formation is subject to proximity, which means that students form friendships with students they meet on a regular basis (Clark & Ayers, 1988). Due to the specific way in which education is organized, students most notably come into contact with classmates at school (Neckerman, 1996). Therefore, peer relationships within the classroom are considered the most consequential for social and scholastic learning (Ladd et al., 2012).

In previous research, outcomes of three specific classmate relationship patterns have been investigated (for a review of this research, see Ladd et al., 2012). First, researchers have studied the outcomes of peer acceptance or rejection. These studies have found, for instance, that rejection by classmates fosters negative school attitudes and lower levels of academic engagement (Ladd, 2003), a higher likelihood of depression and aggressive behavior (Ialongo, Vaden-Kieman, & Kellam, 1998), and ultimately, lower academic achievement (Ladd, 2003). The second relationship pattern is peer victimization, which involves negative behavior of a subgroup of peers directed against a classmate (Olweus, 1993). Studies consistently find that being victimized results in a multitude of psychosomatic symptoms, including depression, anxiety, feelings of tension, and sleep problems (Demanet & Van Houtte, 2012a). According to Buhs and Ladd (2001), victimization explains how peer rejection leads to academic withdrawal: Rejected students will be subject to more victimization, which in turn will diminish their engagement and achievement.

The third relationship pattern involves friendships, which are defined as voluntary relationships that embody an affective tie between pairs of students (Bukowski et al., 1993). It is important to note that whereas victimization and rejection or acceptance are concerned with the general classroom group, friendships are particular to dyads of students. With regard to adjustment during adolescence, researchers rate dyadic friendships as more important than general acceptance by the group (Bukowski et al., 1993). Moreover, the beneficial effects of acceptance only apply to students with friendships (Bukowski et al., 1993). Last, students who have friends have a lower likelihood of being victimized (Mouttapa, Valente, Gallaher, Rohrbach, & Unger, 2004). Accordingly, with regard to classroom peer relationships, friendships appear more important for school adjustment than the other two types.

Also, the number of friendships is related to several beneficial outcomes, for instance, to higher grades (Berndt & Keefe, 1995), involvement in classroom and school-related activities (Kingery & Erdley, 2007), and better performance on academic tasks (Newcomb & Bagwell, 1995). Students with more friends report higher school engagement (Ladd, 1990), higher levels of school belonging (Wentzel, 2009), and are less subject to deviant peer influence (Geven, Weesie, & van Tubergen, 2013). Given its implications for academic grades and school attitudes, the number of classroom friendships is an important indicator of peer group integration.

Grade Retention and Peer Relationships

It is inherent to the practice of grade retention that retained students enter a new class group. It is likely that being retained has an impact on the position in the social group of the new class (Pierson & Connell, 1992). The impact, however, may differ with the age of the student, as researchers have argued that retention is especially stigmatizing at later ages, such as in adolescence (Demanet & Van Houtte, 2013; Wu et al., 2010). At younger ages, retained students may be popular among their new classmates, as they will be comparatively older, more mature, and more confident about the subject matter (Pierson & Connell, 1992). When retention occurs later in the educational path, the retained student may be labeled a slow learner or a low achiever, which might lead to withdrawal from the class’s social group (Hong & Yu, 2008). Moreover, scholars point out that the label of retainee impedes a person’s ability to form new friendships (Hong & Yu, 2008; Jimerson et al., 2006; Stearns et al., 2007). Accordingly, most researchers assume that retained students are confronted with difficulties in rebuilding their social network and therefore will have fewer classroom friends compared with promoted students.

Most of these expectations and assumptions, however, remain uninvestigated to date, as empirical research on the consequences of grade retention for friendships is relatively scarce and has produced mixed results (Bonvin et al., 2008; Wu et al., 2010). Some existing studies confirm the general expectation that grade retention is detrimental for social relationships with classmates. For example, Pianta, Tietbohl, and Bennett (1997), focusing on kindergarten retention, show that retained students are less well liked by classmates than equally low-achieving but promoted counterparts. In a longitudinal study, Jimerson and colleagues (1997) confirmed these results and added that the negative social repercussions of primary school retention may affect retained students even up to the age of 16. Both of these studies are limited to teacher reports of peer acceptance. A more recent study utilizing peer-reported data was able to distinguish between the short-term and longer-term social consequences of grade retention in first grade (Wu et al., 2010). While in the short term grade retention increased liking by classmates, it was associated with growing negative effects over time. These findings are consistent with the idea that retained students are stigmatized more as they grow older (see also Wu et al., 2010, p. 148). However, not all research is equivocal on the social consequences of grade retention. Bonvin and colleagues (2008) carried out a longitudinal investigation into the social consequences of grade retention in primary education in Switzerland. They also found a beneficial short-term effect on social acceptance but did not find the detrimental long-term effect described by Wu and colleagues (2010). Other studies have found no association between retention and social outcomes. Hong and Yu (2008), for example, found that grade retention in kindergarten is not related to popularity or ease of making friendships in elementary education. Gottfredson, Fink, and Graham (1994) found that retention in middle school is unrelated to the retained students’ perceived social acceptance among peers.

This overview shows that existing research on the social outcomes of retention is inconclusive. This might be attributed to the fact that although studies mostly investigate social outcomes under the heading of social or peer acceptance (Bonvin et al., 2008; Jimerson et al., 1997; Wu et al., 2010), they operationalize this concept in many different ways. For instance, some scholars focus on psychosocial dimensions of interpersonal relationships, such as perceived competence or interest in peer relationships (Hong & Yu, 2008; Jimerson et al., 1997), while others rely on liking by peers (Pianta et al., 1997; Wu et al., 2010). Still others construct a measurement of the frequency of positive or negative interpersonal interactions in class (Bonvin et al., 2008) or investigate retained students’ perceptions of peer relatedness (Pierson & Connell, 1992). Most previous research focuses on peer acceptance outcomes, while friendship outcomes are largely neglected. However, as discussed in the previous section, having friendships in adolescence is more consequential for school adjustment than being accepted or rejected by peers. It appears that only focusing on acceptance does not provide a complete picture of the social outcomes of retention: There should also be a focus on the actual number of friendships with classmates. Another concern deals with the timing of retention. As the label of retainee is more stigmatizing at later ages, one would expect grade retention to be especially detrimental for social relationships in adolescence. Unfortunately, however, only one study on the social outcomes of grade retention has been situated in secondary education (Gottfredson et al., 1994). Given the lack of investigation into this developmental period, it is important to perform more research on social outcomes in adolescence.

Multilevel Questions in Retention Research

Since the 1960s, school characteristics have been shown to determine children’s cognitive and non-cognitive outcomes, over and above the influence of individual characteristics (see e.g., Teddlie & Reynolds, 2000). One important shortcoming of research into grade retention is that very few studies have accounted for this multilevel nature of the school context (for notable exceptions, see Demanet & Van Houtte, 2013; Hong & Raudenbush, 2005, 2006). Nevertheless, differences between schools’ retention policies produce differences in their retention composition—that is, the proportion of students in a school who have a history of retention (Hong & Raudenbush, 2005; Shepard & Smith, 1988). It is important to investigate whether this compositional characteristic influences student outcomes over and above the influence of individual grade retention.

Two specific kinds of composition effects can be investigated. First, the compositional characteristic may exert a direct effect on outcomes. Hong and Raudenbush (2005, 2006) hypothesized that retention policy might not only affect retained children but also the promoted ones by shaping the retention composition at a school. This consideration led them to assess the impact of the schools’ retention composition on the achievement of retained and promoted students. They discovered, however, no effect from the retention composition. According to the authors, this might be due to the study’s crude distinction between schools with a low and high rate of retention (Hong & Raudenbush, 2006). Another study investigated the effect of retention composition on school misconduct (Demanet & Van Houtte, 2013). This study utilized a continuous measurement for retention composition, which allowed for a more fine-grained analysis, and found that higher percentages of retained students at a school were associated with a higher likelihood of misbehavior. This effect was established over and above the individual effect of grade retention.

We have reasons to believe that retention composition also affects the number of classroom friendships directly. Specifically, the frequent application of this practice may eventually diminish the number of classroom friendships for all students in a school. For instance, if more former classmates of promoted students are retained, the promoted students will also eventually end up with fewer classroom friends, leaving them subject to the associated negative outcomes (see previous). No study to date, however, has investigated whether the retention composition of the school is related to the number of classroom friendships of its students.

The second type of composition effects are moderation effects. Hong and Raudenbush (2006) investigated whether the impact of individual retention on academic achievement differs between schools with low and high rates of retention. They found, however, no moderation effect from the retention composition. This might be due to the study’s dichotomous distinction between a low and high rate of retention (Hong & Raudenbush, 2006), as Demanet and Van Houtte (2013) found that retained students in schools with a low overall rate of retention are more likely to break rules than their equivalents in schools with a high rate.

Applied to the topic of social relationships, it is possible that the effect of being retained on the number of friendships differs between low and high retaining schools. Such a moderation effect may be expected based on two viewpoints: stigma literature (Coleman, 1986) and macrostructural theory (Blau, 1974). First, we have discussed previously that the stigma associated with being retained may impede the chances of developing classroom friendships. Scholars hold, however, that the severity of stigmatization is dependent on the context (see e.g., Coleman, 1986). In the case of grade retention, it has been suggested that if more fellow students have been retained, it is less stigmatizing to be in that category than when only a few peers are in the same situation (Demanet & Van Houtte, 2013). Hence, we may expect the label of retainee to have less of an impeding effect on the development of new classroom friendships in high-retention schools.

Second, a potential moderating influence of retention composition is consistent with macrostructural theory (Blau, 1974, 1977). This classical sociological theory links a context’s heterogeneity or differentiation with the integration between individuals from different groups. The theory starts from the principle that individuals prefer to associate with similar others—or in other words, with others belonging to the in-group. For instance, people tend to prefer same-gender or same-ethnic associations (Demanet, Agirdag, & Van Houtte, 2012; Jugert, Noack, & Rutland, 2013). Of course, individuals occupy an infinite number of characteristics, and the characteristic that is chosen as an indicator of similarity depends on the salience of the characteristic in a certain context (Blau, 1974). For example, in some contexts, gender is an important marker for in-group association, while in other contexts, ethnicity is more predominantly chosen (see e.g., Demanet et al., 2012; McPherson, Smith-Lovin, & Cook, 2001). Out-group associations occur, but these depend on status differentials: A person will be less likely to associate with others who are not equal in status (Blau, 1974).

A second principle of macrostructural theory is that structural conditions determine at least partly whether this preference for in-group association may be realized (Blau, 1974). In colloquial terms, it is easier to befriend in-group members when more in-group members are around. For instance, research shows that students engage more readily in interethnic friendships when meeting fewer same-ethnic peers at school (Van Houtte & Stevens, 2009).

When we apply this theory to the topic of grade retention, this reasoning implies that retained students will prefer to associate with other retainees. Moreover, if retained students are indeed stigmatized in adolescence as “incompetent” (see also Hong & Yu, 2008), we may expect a status difference between retainees and their promoted counterparts, which makes associations between retained and promoted students less likely to occur. The application of the second principle of the theory suggests a moderation effect of the retention composition. It suggests, namely, that it is easier for a retained student to befriend people in the same position in a school with a higher overall proportion of retained students than in a school with a lower proportion. In other words, the consequences of retention for students’ number of same-grade friendships may be expected to be less severe in schools where more students are retained.

The Flemish Educational Context

Flanders is the Dutch-speaking, northern part of Belgium, and since 1988, the Flemish government has had the jurisdiction to implement and govern its own educational system. The current study is limited to students and schools in this region. First, it should be noted that every school in Flanders is state subsidized—public and private schools alike. Public schools are non-sectarian, and with only a few exceptions, private schools are Catholic. The majority (around 70%) of students attend Catholic schools. Children usually attend nursery school (i.e., preschool) from the age of 2½, and education is compulsory from the age of 6 until the age of 18. At 12 years of age, after six grades of primary education, children transfer to secondary education. There are also six grades of secondary education (i.e., middle school and high school) comprising three main units, each subdivided into two grades. There are four main tracks in secondary education: academic education preparing for higher education, technical education, vocational education, and artistic education (the last of which is marginal in terms of the number of students). Between-school tracking is most common, though within-school tracking exists (Van Houtte, Demanet, & Stevens, 2012). The different tracks are commonly classified hierarchically, placing vocational tracks at the lower end of the scale.

At the end of each secondary grade, students receive a certificate indicating whether they can continue their current school path (certificate A) or not (certificate B or C). A certificate B allows progress into the next grade, but the student needs to join another track (usually lower), whereas a certificate C indicates grade retention. These certificates are based on the grade point average (GPA) obtained, but as no standardized tests exist in Flemish education (Stevens, 2007), it is the responsibility of the teachers to design, administer, and grade the examinations, and accordingly, teachers have considerable power to determine the educational progress of their pupils. In practice, therefore, the certificate granted is not only determined by a student’s academic achievement but also reflects teachers’ appraisal of other aspects of general conduct, such as behavior in class (Stevens, 2007).

Due to the hierarchical nature of the system, a “cascade effect” has emerged, in which students start in higher tracks, but if they fail, they move to lower—and less valued—tracks. This cascade effect is determined by B certificates, but it is also possible for students to change tracks voluntarily. While in theory it is possible to switch to higher tracks, in practice, the flow of students predominantly moves from the academic track, through the technical track, to the vocational track. As the subject matter taught in the various tracks and curricula builds on knowledge acquired in previous grades of the same curriculum or track, it is hard for students to join a higher track where the subject matter is more abstract and theoretical. Recent data show that the downward flow is quite large, as more than 70% of fifth-grade students in the vocational track had started secondary education in a higher track (Van Praag, 2014). Moreover, students in vocational tracks are more likely to have been retained in the past (for an extended discussion, see Van Houtte et al., 2012).

In Flemish secondary schools, in general, students belong to specific class groups—a group of students that is seen as a unit and has a teacher appointed to them responsible for administrative tasks. However, students have ample opportunity to meet students from other class groups, as they are frequently put together. For instance, it is commonplace for students with the same number of hours of math to be put together for math lessons even if they do not belong to the same administrative class group. Class group composition also shifts for other courses, such as physical education classes. Accordingly, in Flanders, there are few structural constraints preventing students coming into contact with pupils from the same grade who belong to administratively different classes. As classes are administrative units that have little significance in daily school life, for the current study, it is logical to investigate all friendships of students with same-grade others rather than to limit the study to investigating classroom friendships as studies in other national contexts have done (e.g., Buhs et al., 2006; Ladd et al., 2008).

The Current Study

In summary, we identify three gaps in scientific literature. First, previous studies have investigated popularity measures, while friendship measures are more consequential for students’ adjustment and therefore are more important to focus on. Second, retention researchers tend to study children at younger ages, and therefore they neglected to study social outcomes in adolescence—a developmental period in which youngsters are most exposed to the negative stigma that adheres to being retained. Third, previous studies on the social outcomes of retention have not yet studied school effects. Notably, schools differ in their retention policy and therefore also in their retention composition—the percentage of retained students at school. This retention composition may affect social outcomes directly, but it might also moderate the relationship between individual retention and social outcomes.

The literature review leads us to propose the following three hypotheses. These hypotheses are depicted graphically in the conceptual model (see Figure 1).

OPEN IN VIEWER

Figure 1.

Conceptual model.

Methods

Data

The data are taken from the FlEA (Flemish Educational Assessment), gathered in the 2004–2005 school year in 85 Flemish secondary schools. We collected these data as part of a larger project investigating the well-being and sexual experiences of students in the Flemish secondary school system. We used multistage sampling. First, we selected proportional-to-size postcodes, size being defined by the number of schools within each postcode using information provided by the Educational Department. From the 240 postcodes obtained, we selected 48 at random. This resulted in a desired overrepresentation of larger municipalities. The aim was to survey students in the third and fifth grades of the Flemish secondary school system (these correspond to Grades 9 and 11 in the U.S. or U.K. school system). Consequently, in the chosen postcodes, we selected all regular secondary schools that provide a third and fifth grade. The eventual participation rate was 31% of the selected schools, which is low due to schools in Flanders being swamped with research requests. Schools choose the research they take part in on a first come, first served basis. We ran analyses comparing our sample with the total Flemish school population, based on information obtained from the Flemish Educational Department. These show that the participating schools did not differ from those that opted out in terms of school sector, size, or curriculum (see Table 1 for school sample characteristics). Hence, there is no systematic bias, and the 85 schools in the sample are representative of the Flemish situation (Van Houtte, 2005).

OPEN IN VIEWER

Table 1

School Sample Characteristics

Variable		%	Mean	SD	N
School sector					85
	Private	49.40
	Public	50.60
Number of students			461.55	285.27	83
Number of teachers			79.76	33.74	82
Curriculum					85
	Multilateral	27.06
	Categorical: Academic	25.88
	Categorical: Technical/vocational	43.53
	Categorical: Artistic	3.53

Note. Multilateral schools: schools where next to the academic track at least one other track is offered (e.g., academic track and vocational track) (see Van Houtte, Demanet, & Stevens, 2012).

In the participating schools, we asked all third- and fifth-grade students present at the time of the visit to fill out a questionnaire. Students filled out the questionnaire in class, supervised by members of the research team and a teacher. A few students were absent for various reasons. In total, 11,872 students provided valid surveys, which amounts to a response rate of 87%. Of the valid surveys, 6,081 (response rate of 90%) were from students in the third grade, and 5,791 (response rate of 86%) came from students in the fifth grade. However, two schools in the sample did not provide information on their school size. As multilevel analysis does not allow missing data at the second level, we had to remove these two schools from the analyses. Consequently, the final analyses are based on 11,759 students across 83 schools. The questionnaires were not anonymous because we needed to link the students’ responses with other data provided by the schools. We later removed all the names, so the final analyses are performed on anonymous data.

Measurements

Outcome

The aim of the current study is to assess the number of same-grade friendships. As discussed in the Flemish context section, class groups in Flemish secondary schools are administrative units that have little significance in the daily lives of students. Therefore, to determine peer relationships that are consequential for school adjustment, it is more logical to investigate same-grade friendships than to restrict research to students in the same classroom. We assess this variable using information provided by the students, which is in line with most previous studies on the impact of peer relationships in education (see e.g., Clark & Ayers, 1988; Coie, Dodge, & Kupersmidt, 1990). The rationale for this is that previous studies have found teacher and student reports of peer relationships to differ. Because students can be seen as “insiders” and teachers as “outsiders” to peer relationships, it is expected that students will provide a better insight into peer relationships than teachers will (see Hartup, 1996, p. 2).

This variable is assessed by a nomination procedure. This has proved a successful method for gathering information on peer ties and interactions (for a discussion, see Coie et al., 1990). Each student was given a list of all the other students in the same grade in their school. Next to the names, we listed identification numbers, and the respondents were asked to provide the respondent identification number for their best friends. In asking students to nominate “best friends,” we follow previous studies (e.g., Bukowski et al., 1993). Scholars noted that if one would simply ask to nominate “friends,” the list for some students would be endless, as students can always name “friends” when asked to do so (Hartup, 1996). Therefore, researchers commonly restrict this choice to the “best friends.” It was not possible to nominate students from outside of school or from other grades in the same school. It was possible to nominate no friends, and the maximum was 12 friends.

No respondents indicated having no best friends, so the minimum is 1 friend (345 students), and the maximum of 12 was indicated by 1,717 (14.5%) students. Elsewhere (Demanet et al., 2012), we have noted that friendship nomination is related to student characteristics. For instance, girls, older students, and students in the vocational track nominate fewer friends. School characteristics also matter, most notably, students in lower socioeconomic status (SES) schools have a tendency to indicate fewer friends. The nomination procedure provided us with data of the social network of each grade. From these network data, we computed each student’s indegree, referring to the number of students by whom the respondent in question was indicated as a best friend. This can be considered a dimension of a student’s network centrality. There exist three types of network centrality, namely, degree, closeness, and betweenness (see Brass & Burkhardt, 1993). By using the indegree, we capture the first type—also labeled indegree centrality—which is a measure of “the number of direct ties to other actors” (Brass & Burkhardt, 1993). As such, it is a very simple but common way of measuring an actor’s centrality in a network (Rowley, 1997). This measurement ranges from 0 (for 249 students) to 15 (for 173 students). On average, the number of friendship nominations in the data set is 6.03 (SD = 3.27; see Table 2).

OPEN IN VIEWER

Table 2

Descriptive Statistics for Variables: Frequencies (%), Means, Standard Deviations, and N

Variables		%	M	SD	N
School level
Retention composition			22.47	15.00	83
School sector
	Public	49.40			83
Socioeconomic composition			4.86	1.20	83
Ethnic composition			15.50	21.05	83
School size			461.55	285.27	83
Student level
Number of friendships			6.03	3.27	11,761
Primary school retention
	Retained	14.80			11,635
Secondary school retention
	Retained	19.70			11,439
Gender
	Girls	51.30			11,732
Socioeconomic status			5.22	2.09	11,050
Ethnicity
	Ethnic minority	11.70			11,759
Age			16.44	1.30	11,694
	Academic	47.00
	Arts	2.80
	Technical	28.50
	Vocational	21.70
Prior achievement			69.43	9.23	10,612

Student-Level Independent Variables

The principal independent variables at the individual level are primary school retention and secondary school retention. We measured these by asking the respondents to report retrospectively on their history of grade retention. This results in two dichotomous variables (0 = never retained; 1 = retained at least once). Of our respondents, 14.8% indicated that they had been retained at least once in the course of primary education (see Table 2), and 19.7% had been retained at least once in secondary education. There was a small association between grade retention in primary education and grade retention in secondary education (Cramer’s V = 0.028; p < .01). Of the students retained in secondary education, 16.3% had also been retained in primary education. Out of all the respondents, 3.3% had been retained both in primary and secondary education. These numbers are quite high compared with other countries (see e.g., Goos et al., 2013). Previous research documents that the prevalence of grade retention in Flanders is high because schools and teachers are generally convinced about the benefits of grade retention and have considerable autonomy to decide whether students should be retained or not (Goos et al., 2013; see the Flemish educational context section).

As stated in the data analyses section (see the following), we include the sociodemographic variables of gender, SES, ethnicity, age, and attending a vocational track as individual-level control variables. The sample used for the analyses is quite equally divided by gender. Of the students in the sample, 51.3% were girls. This only slightly differs from the general population. In the school year 2004–2005, for example, 48.86% of all 15- and 17-year-olds in Flanders were female. The SES of students’ families is measured using the class scheme of Erikson, Goldthorpe, and Portocarero (1979) and is based on the occupational status of the students’ parents. For this, we take the highest ranked occupation of either the father or the mother (Erikson et al., 1979), or if they were unemployed, we use their previous occupation. This results in a SES scale ranging from 1 (unskilled manual labor) to 8 (professionals and major proprietors), with a mean of 5.22 (SD = 2.09, see Table 2). The principal criterion for determining students’ ethnicity is the birthplace of the maternal grandmother. Only 1% of the respondents did not answer this question. To determine the ethnicity of those students, we use the nationality of the students’ mother and father as most ethnic minority students are second- or third-generation citizens and hold Belgian nationality. As is common practice in European research, only West European birthplaces and nationalities are considered as being of native descent (e.g., Timmerman, Hermans, & Hoornaert, 2002). This results in a dichotomous variable (0 = ethnic majority student, 1 = ethnic minority student). The majority of respondents are ethnic majority students (88.3%). Of the ethnic minority students, most have Turkish or Moroccan backgrounds (both approximately 30%), the remainder have Southern European (10%), Eastern European (8%), North African (5%), or other (16%) backgrounds. The majority of students were 15 (35%) or 17 (32.6%) years old at the time of the survey, while the rest were slightly older than most in their grade, mainly due to retention (11.2% 16 years old, 14.2% 18 years old, 4.5% 19 years old, and 1.3% 20 years old). The mean age is 16.44 (SD = 1.3). Most respondents were attending the academic track (47%), with 28.5% attending the technical track, 21.7% the vocational track, and 2.8% the arts track. The last individual-level control variable is prior achievement. To grade their students, teachers use a percentage, so grades range from 0% to 100%. The GPA for the students in our sample ranges from 41% to 100%, with a mean of 69.43% (SD = 9.23; see Table 2). As already stated, grading is the responsibility of teachers in Flanders, and as there are no standardized tests, it is difficult to compare grades across schools. Therefore, it is not possible to interpret whether the mean of 69.43% is high or low. If a student’s grade is below 50%, however, a certificate C is often given. Above that threshold, teachers have a high level of autonomy to make decisions about the certificates they give. Accordingly, above 50%, there are no clear cutoff points for grades to determine whether students can continue their curriculum (certificate A), whether they need to join another track (certificate B), or whether they should be retained (certificate C).

School-Level Independent Variables

To construct the measurement for retention composition, we use the individual-level data to calculate the percentage of respondents in a school who were retained at least once during the course of secondary education. There is a substantial variation between schools in the percentage of retained students, ranging from 0% (two schools) to 58.33% (one school). On average, schools include 22.47% (SD = 15.00; see Table 2) of students who were retained at least once during the course of secondary education.

We include four school-level control variables. In this sample, 49.4% of the schools belong to the public sector, which is a slight overrepresentation of the overall Flemish situation. This is because we oversampled larger municipalities, where the majority of public schools in Flanders are located. To assess SES composition, we calculate the mean of students’ SES per school, which is the customary way to construct this measurement (see e.g., Demanet & Van Houtte, 2011, 2014). The schools have an average SES composition of 4.86 (SD = 1.20). We measure the ethnic composition by the proportion of ethnic minority students in the school. We asked the administrators to estimate this. However, 12 (14.12%) of the 85 administrators chose not to answer this question. In addition, we computed the proportion of ethnic minority students at school using the individual-level data (see aforementioned). The correlation of 0.88 (p <.001) between the two validates this aggregated measurement. The schools cover a wide range of ethnic composition, from 0% (6 schools) to 88.20% (1 school) of ethnic minority students. The mean ethnic composition is 15.50 (SD = 21.05). We asked the administrators to provide us with the school size—that is, the number of students at each school. However, we obtained information from only 83 of the 85 schools in our sample (see aforementioned). The mean school size in our sample is 461.55 (SD = 285.27).

Data Analyses

To test the hypotheses, we use multilevel modeling (HLM6) (Raudenbush & Bryk, 2002). We should make clear that we rely on cross-sectional data and that multilevel analysis remains a correlational technique. Hence, we note from the outset of this study that we cannot make any causal claims. As is common in multilevel analysis, we first estimate unconditional models, which enables us to determine the amount of variance in the number of same-grade friendships situated at the school level. We test four models. To ensure model stability, all independents except the dichotomous variables are grand mean centered.

In the first model, we assess the role of grade retention in the number of same-grade friendships. Because schooling is organized into two different systems in Flemish education—a primary school system and a secondary school system—and previous studies suggest that retention in these systems may have different effects on student outcomes (Demanet & Van Houtte, 2013), we distinguish between retention in primary and secondary education in the analyses. Moreover, in the first model, we assess the role of retention composition—that is, the percentage of retained students in a school. A compositional effect arises when the composite effect is found to be significant, over and above the individual effect. The multilevel equations for Model 1 are:

Level 1 Model

Number of friendships = β 0 + β 1 × ( Retention primary ) + β 2 × ( Retention secondary ) + r .

Level 2 Model

β 0 = γ 00 + γ 01 × ( Retention composition ) + u 0

β 1 = γ 10 + u 1

β 2 = γ 20 + u 2 .

In the second model, a cross-level interaction term between the proportion of retained students in school and retention in secondary education is added. This enables us to investigate whether the association between secondary grade retention and the number of same-grade friendships is dependent on the retention composition, as stated in the third hypothesis. The multilevel equations for the second model are thus:

Level 1 Model

Number of friendships = β 0 + β 1 × ( Retention primary ) + β 2 × ( Retention secondary ) + r .

Level 2 Model

β 0 = γ 00 + γ 01 × ( Retention composition ) +

β 1 = γ 10 + u 1

β 2 = γ 20 + γ 21 × ( Retention composition ) + u 2 .

In the third model, we add control variables to rule out spurious relationships and selection effects. At the school level, we control for school sector, SES composition, ethnic composition, and size, as these variables are related to students’ number of friendships (Demanet et al., 2012; Moody, 2001). At the individual level, we control for gender, SES, ethnicity, and age (Barone et al., 1995). It is especially important to account for the latter, as we have to be careful not to confound retention effects with age effects. Additionally, we control for whether students attended the vocational track (see the context section). We accordingly arrive at the multilevel equations for Model 3:

Level 1 Model

Number of friendships = β 0 + β 1 × ( Retention primary ) + β 2 × ( Retention secondary ) + β 3 × ( Gender ) + β 4 × ( SES ) + β 5 × ( Ethnicity ) + β 6 × ( Age ) + β 7 × ( Vocational track ) + r .

Level 2 Model

β 0 = γ 00 + γ 01 × ( Retention composition ) + γ 02 × ( School size ) + γ 03 × ( Ethnic composition ) + γ 04 × ( SES composition ) + γ 05 × ( School sector ) + u 0

β 1 = γ 10 + u 1

β 2 = γ 20 + γ 21 × ( Retention composition ) + u 2

β 3 = γ 30 + u 3

β 4 = γ 40 + u 4

β 5 = γ 50 + u 5

β 6 = γ 60 + u 6

β 7 = γ 70 + u 7 .

In the fourth model, we additionally control for students’ prior achievement, as students with lower prior achievement are more likely to have been retained in the past and achievement is related to friendships at school (Buhs et al., 2006; Jimerson & Ferguson, 2007; McCoy & Reynolds, 1999). This variable is only included in the last model because the prior achievement measurement should be considered carefully. As no standardized tests (e.g., state-administered tests) exist in Flemish education, it is hard to compare academic achievement across schools. Furthermore, as this is a self-reported item, it could contain biases due to memory problems and cover-up strategies. As a result, it has a large number of missing values (9.8%). The multilevel equations for the fourth model are:

Level 1 Model

Number of friendships = β 0 + β 1 × ( Retention primary ) + β 2 × ( Retention secondary ) + β 3 × ( Gender ) + β 4 × ( SES ) + β 5 × ( Ethnicity ) + β 6 × ( Age ) + β 7 × ( Vocationaltrack ) + β 8 × ( Prior achievement ) + r .

Level 2 Model

β 0 = γ 00 + γ 01 × ( Retention composition ) + γ 02 × ( School size ) + γ 03 × ( Ethnic composition ) + γ 04 × ( SES composition ) + γ 05 × ( School sector ) + u 0

β 1 = γ 10 + u 1

β 2 = γ 20 + γ 21 × ( Retention composition ) + u 2

β 3 = γ 30 + u 3

β 4 = γ 40 + u 4

β 5 = γ 50 + u 5

β 6 = γ 60 + u

β 7 = γ 70 + u 7

β 8 = γ 80 + u 8 .

Results

The unconditional “null” models show that 8.97% (σ² = 9.721; τ ₀ = 0.958; p < .001) of the variance in the number of friendships is situated at the school level. Hence, the unconditional model indicates that performing multilevel analyses is warranted.

Individual Retention and the Number of Same-Grade Friendships

The results of the multilevel analyses (shown in Table 3) indicate that grade retention is related to the number of same-grade friendships. First, not taking the control variables into account, retention in primary education (γ* = −0.027; p < .01; see Model 1) is associated with a lower number of same-grade friendships in secondary education, which aligns with the hypotheses. When we control for the sociodemographic variables in the third model, however, the association with primary school retention is no longer significant (γ* = 0.013; p > .05). Further analyses (not shown) indicate that the relationship with primary school retention is entirely mediated by vocational track enrollment. Accordingly, it appears that students who are retained at least once during the course of primary education are more likely to have fewer same-grade friends in secondary education because they tend to end up in the vocational track. Indeed, the results of these analyses show that students in the vocational track are less frequently nominated as a best friend (γ* = −0.067; p < .001). Furthermore, secondary school retention is also related to the number of same-grade friendships (γ* = −0.067; p < .001). When we take the control variables into account, this association reduces (γ* = −0.041; p < .01 in Model 3; γ* = −0.027; p < .1 in Model 4) but remains significant. We should point out that the association with secondary grade retention is relatively small. Without accounting for contextual associations (see the following) and when controlling for all covariates in the model, the difference between retained and promoted students on average is only 0.249 friends, which is not large. However, this means that even when controlling for sociodemographic characteristics, having been retained during secondary education is related to a lower number of friendships in the same grade, which supports Hypothesis 1.

OPEN IN VIEWER

Table 3

The Association Between Primary School Retention, Secondary School Retention, Percentage of Retained Students, and Number of Friendships: Results of Stepwise Multilevel Analyses

		Model 1	Model 2	Model 3	Model 4
Intercept	γ	5.869***	5.851***	5.988***	6.103***
	SE	0.102	0.100	0.157	0.167
School level
Retention composition	γ	−0.027***	−0.037***	−0.026***	−0.026***
	γ*	−0.125***	−0.171***	−0.120***	−0.119***
	SE	0.005	0.006	0.006	0.007
School sector	γ			0.003	−0.047
	γ*			0.0001	−0.007
	SE			0.162	0.170
Socioeconomic status composition	γ			−0.121	−0.127
	γ*			−0.044	−0.047
	SE			0.116	0.118
Ethnic composition	γ			−0.0001	0.003
	γ*			−0.001	0.021
	SE			0.006	0.006
School size	γ			0.001*	0.001*
	γ*			0.051*	0.050*
	SE			0.0001	0.0001
Individual level
Retention primary	γ	−0.249**	−0.221*	0.118	0.169 †
	γ*	−0.027**	−0.024*	0.013	0.019 †
	SE	0.088	0.087	0.092	0.095
Retention secondary	γ	−0.546***	−0.644***	−0.332**	−0.218 †
	γ*	−0.067***	−0.079***	−0.041**	−0.027 †
	SE	0.110	0.102	0.112	0.114
Gender	γ			−0.122	−0.152*
	γ*			−0.019	−0.023*
	SE			0.077	0.077
Socioeconomic status	γ			0.100***	0.091***
	γ*			0.064***	0.058***
	SE			0.021	0.021
Ethnicity	γ			−0.006	−0.043
	γ*			−0.001	−0.004
	SE			0.128	0.125
Age	γ			−0.224***	−0.221***
	γ*			−0.089***	−0.088***
	SE			0.055	0.055
Vocational track	γ			−0.347*	−0.533***
	γ*			−0.043*	−0.067***
	SE			0.146	0.149
Prior achievement	γ				0.015**
	γ*				0.042**
	SE				0.004
Cross-level interactions
Retention Secondary × Retention Composition	γ		0.029***	0.021**	0.024**
	SE		0.008	0.007	0.007
Variance components:
Intercept	U0	0.721***	0.685***	0.960***	0.900**
Retention primary	U1	0.063	0.052	0.160	0.116
Retention secondary	U2	0.418	0.280**	0.288	0.305
Gender	U3			0.148**	0.108*
Socioeconomic statu	U4			0.012	0.012 †
Ethnicity	U5			0.150	0.145
Age	U6			0.159***	0.147***
Vocational track	U7			0.592***	0.664**
Prior achievement	U8				0.0001
Intraclass correlation (ICC)		0.111	0.095	0.208	0.202

Note. The unstandardized (γ) and standardized (γ*) gamma coefficients are presented, with the standard errors (SE).

†

p≤ .10. *p≤ .05. **p≤ .01. ***p≤ .001.

The Moderating Role of Retention Composition

Having established that the retention composition is related to the number of same-grade friendships, we still need to investigate whether the composition moderates the association with individual retention in secondary education (Hypothesis 3). It appears that the relationship between secondary school retention and the number of friendships is moderated by the retention composition. Specifically, the cross-level interaction term is consistently significant across all models (γ = 0.029; p < .001 in Model 2; γ = 0.021; p < .01 in Model 3; γ = 0.024; p < .01 in Model 4). Accordingly, it is apparent that individual retention is less detrimental for the number of same-grade friendships in schools where more retained students are present.

This moderation requires closer attention. The size of the regression coefficient suggests that the association with individual secondary school retention becomes positive in schools with a high percentage of retained students. This would mean that in schools with a high retention rate, retained students have more same-grade friends than promoted students. More specifically, the analyses suggest that this association becomes positive when the measure for retention composition—which is centered around the grand mean—is higher than 9, or in other words, once schools reach a threshold of 32.47% (= 9 + the mean of 22.47) retained students.

One may question whether the association with individual secondary school retention continues to rise linearly with the percentage of retained students at school or whether there is a tipping point where the relation reaches a maximum. To investigate this, we perform additional analyses in which the continuous variable for retention composition is broken down into its quintiles. Accordingly, the retention composition variable becomes a categorical one, with the values indicating the quintile of retention composition a school belongs to. The five quintiles have the following boundaries: Qu₁: 0, 9.12; Qu₂: 9.12, 16.15; Qu₃: 16.15, 25.62; Qu₄: 25.62, 36.35; Qu₅: 36.35, 58.33. Taking Qu₁ as the reference category, we construct four dummy variables indicating which of the four other quintiles a school belongs to and add these dummy variables to the models. To investigate the linearity of the cross-level interaction, we add cross-level interaction terms of individual secondary school retention with the dummies for Qu₂, Qu₃, Qu₄, and Qu₅, respectively. We perform the same four steps as in the previous analyses (for the multilevel equations for these additional analyses, see Appendix in the online journal).

The results of these analyses show that the association of secondary school retention with the number of same-grade friendships is the same in schools in the first and second quintile of retention composition (γ = −1.039; p < .001; see Table 4, Model 4). For schools in the third quintile, the association is much smaller (γ = −0.150 [= −1.039 + 0.889]; p < .01) but still significantly negative. For schools in the fourth quintile, the relationship is significantly positive (γ = 0.427 [= −1.039 + 1.466]; p < .001). Although the relationship is still significantly positive for schools in the fifth quintile (γ = 0.111 [= −1.039 + 1.150]; p < .001), the coefficient shows that the difference between retained and promoted students in schools in the highest quintile of retention composition is virtually zero.

OPEN IN VIEWER

Table 4

The Association Between Primary School Retention, Secondary School Retention, Percentage of Retained Students (Categorical), and Number of Friendships: Results of Stepwise Multilevel Analyses

			Model 1	Model 2	Model 3	Model 4
Intercept		γ	6.667***	6.836***	6.661***	6.791***
		SE	0.244	0.255	0.287	0.286
School level
Retention composition (reference category Qu1)	Qu2	γ	−0.900**	−0.924**	−0.561*	−0.640*
		γ*	−0.113**	−0.116**	−0.070*	−0.080*
		SE	0.277	0.297	0.276	0.266
	Qu3	γ	−0.839*	−1.005**	−0.652*	−0.653*
		γ*	−0.103*	−0.123**	−0.080*	−0.080*
		SE	0.323	0.346	0.317	0.291
	Qu4	γ	−0.929**	−1.353***	−0.858*	−0.831*
		γ*	−0.114**	−0.166***	−0.105*	−0.102*
		SE	0.348	0.349	0.342	0.338
	Qu5	γ	−1.328***	−1.661***	−1.231***	−1.215***
		γ*	−0.166***	−0.208***	−0.154***	−0.152***
		SE	0.282	0.305	0.325	0.322
School sector		γ			−0.035	−0.097
		γ*			−0.005	−0.015
		SE			0.156	0.159
Socioeconomic status composition		γ			−0.128	−0.141
		γ*			−0.047	−0.052
		SE			0.118	0.121
Ethnic composition		γ			−0.001	0.002
		γ*			−0.003	0.016
		SE			0.006	0.006
School size		γ			0.001*	0.001*
		γ*			0.059*	0.061*
		SE			0.0001	0.0001
Individual level
Retention primary		γ	−0.249**	−0.229*	0.115	0.164 †
		γ*	−0.027**	−0.025*	0.013	0.018 †
		SE	0.088	0.088	0.092	0.096
Retention secondary		γ	−0.543***	−1.629***	−1.137***	−1.039***
		γ*	−0.066***	−0.199***	−0.139***	−0.127***
		SE	0.110	0.299	0.272	0.276
Gender		γ			−0.126	−0.157*
		γ*			−0.019	−0.024*
		SE			0.077	0.077
Socioeconomic status		γ			0.102***	0.092***
		γ*			0.065***	0.059***
		SE			0.021	0.021
Ethnicity		γ			−0.004	−0.044
		γ*			−0.0004	−0.004
		SE			0.129	0.126
Age		γ			−0.226***	−0.222***
		γ*			−0.090***	−0.088***
		SE			0.055	0.055
Vocational track		γ			−0.326*	−0.516***
		γ*			−0.041*	−0.065***
		SE			0.144	0.147
Prior achievement		γ				0.015**
		γ*				0.041**
		SE				0.004
Cross-level interactions
Retention Secondary × Retention Composition	Qu2	γ		0.584	0.467	0.404
		SE		0.396	0.362	0.347
	Qu3	γ		1.053**	0.830**	0.889**
		SE		0.340	0.305	0.312
	Qu4	γ		1.675***	1.422***	1.466***
		SE		0.326	0.307	0.313
	Qu5	γ		1.406***	1.110***	1.150***
		SE		0.350	0.303	0.315
Variance components:
Intercept		U0	0.710***	0.682***	1.058***	1.005***
Retention primary		U1	0.047	0.044	0.152	0.111
Retention secondary		U2	0.406***	0.170 †	0.246	0.264
Gender		U3			0.153**	0.114*
Socioeconmic status		U4			0.012	0.012 †
Ethnicity		U5			0.138	0.143
Age		U6			0.160***	0.146***
Vocational track		U7			0.587***	0.654**
Prior achievement		U8				0.0001
Intraclass Correlation (ICC)			0.107	0.085	0.211	0.206

Note. The unstandardized (γ) and standardized (γ*) gamma coefficients are presented, with the standard errors (SE). Qu = quintile.

†

p≤ .10. *p≤ .05. **p≤ .01. ***p≤ .001.

In short, this more fine-grained analysis provides more insight into the nature of the cross-level interaction effect (see Figure 2). It shows that the difference in the number of same-grade friendships between retained and promoted students is greatest in schools with a low retention rate (respectively, D = 1.039 for schools in the first quintile; D = 0.635 for schools in the second quintile). For schools in the third quintile, the difference is virtually zero (D = 0.150). The retained students even surpass the promoted students in schools in the fourth quintile of the retention composition distribution (D = −0.427). In the schools with the highest retention rate, the difference between retained and promoted students is again negligible (D = −0.111). These analyses show that retention is quite strongly related to the number of same-grade friendships in the schools with the lowest retention rates: In these schools, retained students have approximately one friend less than their promoted counterparts. The analyses point to the importance of the context, as the difference between retained and promoted students is substantially less in schools with more retained students.

OPEN IN VIEWER

Figure 2.

Number of friendships: cross-level interaction effect between retention composition and individual secondary school retention (results of multilevel analyses).

Discussion

We aimed to investigate the implications of grade retention for the number of same-grade friendships in adolescence. This study is unique in three respects. First, we focus on adolescence, a neglected developmental period in retention research. Second, we investigate the number of same-grade friendships as an outcome of grade retention, an outcome that has, to the best of our knowledge, not previously been linked empirically to grade retention. Third, we adopt a multilevel approach as we investigate whether the percentage of retained students at school affects the number of same-grade friendships directly and whether the retention composition determines the impact of individual retention on the number of same-grade friendship relationships.

First, the analyses show that students retained in primary education have fewer friendships in secondary education, but this appears to be due to attending the vocational track. It is likely that this finding is due to the specific nature of the Flemish secondary school system, which groups students in a rigid track system. This track system in Flanders has resulted in a cascade effect: the tendency to start in more academically oriented tracks and change to more vocationally oriented tracks during the course of secondary education. Van Praag (2014) found that this has led to a very heterogeneous student body in the lower tracks in terms of the educational trajectories followed. It may be that students therefore have difficulties in establishing friendship relationships with one another—at any rate, we may expect them to have more difficulties than students in the more homogeneous higher tracks. The fact that primary school retention has no direct association with friendships in secondary school, moreover, seems to suggest that these students are not labeled as retainees anymore or that this label is not stigmatizing to the extent that it influences their friendships. These findings are consistent with previous research. For instance, Hong and Yu (2008) suggest that kindergarten retention does not have negative effects on outcomes in primary education because the retention stigma has been eroded by then. Accordingly, it seems that the stigma of grade retention may “wash away” over the years (Demanet & Van Houtte, 2013; Hong & Yu, 2008, p. 418).

Retention in secondary education is, as expected (Hypothesis 1), associated with the number of same-grade friendships. Specifically, not taking into account the role of the retention composition, students retained in secondary education have fewer friendships. The analyses support the assumption that retention has implications for the friendships of retained students with others in their grade (Pierson & Connell, 1992; Stearns et al., 2007). When not accounting for the context, the difference between retained and promoted students is quite small. An important finding in the current study, however, is that the nature of the association depends on the retention composition of the school (see Figure 2). Accordingly, the analyses support the third hypothesis. More specifically, the negative association between grade retention and the number of friendships only shows in schools with a low rate of grade retention. In schools with many retained students, these pupils may even have more friends than promoted students, although we should note that this difference is small and actually becomes negligible in the highest retaining schools in our sample. Accordingly, we conclude that grade retention relates to students’ same-grade friendships but that the exact implications are dependent on the proportion of retained students in the school.

These findings may be explained by the basic tenets of macrostructural theory (Blau, 1974). As discussed previously, this theory is guided by two principles. First, people prefer to associate with in-group members, and second, the structural properties of a context determine the extent to which this preference can be realized. According to the first principle, retained students might be expected to prefer friendships with others in the same position. The second principle, therefore, might explain why the impact of grade retention on the number of friendships is dependent on the retention composition. In schools with a low retention rate, it may be difficult for retained students to realize their preference to establish friendships with others in the same position as there are fewer of them in such schools. In schools with a high retention rate, retained students are much more readily found, so the in-group preference can be realized. Another possibility is that the distinction between retained and promoted students is more salient for retained students as a marker for in-group association than it is for promoted students. If so, retained students would especially choose friendships based on the retained/promoted student distinction while promoted students would continue to select based on other characteristics, such as gender or ethnicity. In that case, the retention composition would particularly affect the chances of retained students associating with in-group members, which might explain the finding that retained students have slightly more friends than promoted students in schools with a high retention rate. In the highest retaining schools, then, there may be so many retained students that the label loses its salience as marker for in-group association, which makes the differences between retained and promoted students obsolete. We must state, however, that this reasoning remains tentative and that as the current study is a unique effort in trying to reconcile individual and school-level influences of retention, further research on this topic is needed to test some of these post hoc propositions. In addition to having a moderating influence, the results point out that retention composition is directly related to the number of same-grade friendships. More specifically, we find that all students in higher retaining schools have fewer same-grade friendships, which supports our second hypothesis. These results on students’ social outcomes align with research that has investigated the impact of retention composition on school misconduct (Demanet & Van Houtte, 2013).

These findings have implications for educational policy for the Flemish system but also for educational systems in other national contexts. If the proportion of retained students in school can be seen as an indicator of a school’s retention policy, as other authors have argued (see Hong & Raudenbush, 2006; Shepard & Smith, 1988), we can state that our findings do not endorse the widespread use of grade retention in secondary schools. Indeed, if the practice of grade retention can foster social isolation, it might bring a host of unwanted side effects, such as decreased enjoyment of school, lower academic performance, and increased school avoidance (Hartup, 1996; Ladd, 1990). We should bear in mind that the Flemish system is a rather unique one, most notably because so many students are retained in comparison to other systems. As our findings suggest that the stigma of retention for adolescent students is dependent on the context, we expect retention to be even more stigmatizing in an educational system where it is seldom applied, as, for example, in the U.S. system. Another peculiar feature of the Flemish system is the role of the teacher, who has large autonomy to decide about students’ fate and bases his or her appraisal of students also on other aspects besides their academic ability (Stevens, 2007). In systems with standardized testing, this role is diminished, and one can expect the decision to retain students to be based more exclusively on students’ ability. Whether accurately based on achievement or not, however, being retained always is a stigmatizing experience for adolescents, a point that is also underscored by U.S. studies (Wu et al., 2010). So, based on these findings in the Flemish system, we expect the negative effect of retention on social integration in the same-grade peer group in adolescence to be present in other national contexts as well, possibly even stronger in systems with a low incidence of retention. Moreover, the finding that primary school retention is inconsequential for friendships in secondary education may be specific to the particular tracked educational system in Flanders, as we found the effect to disappear when we controlled for track attended. As Wu and colleagues (2010) in the United States found retention at early ages to impact social acceptance more negatively as students grow older, it is possible that a direct effect of primary retention is found in other national contexts.

Although we find that the negative association between grade retention and friendships is moderated by retention composition—and one could conclude that a more frequent application of retention at school counteracts the negative social implications of retention—we should note that the negative social repercussions are only absent when a very high proportion of students in a school—at least 30%—have been retained. Taking into account studies in various contexts—including the U.S. context—that show the ineffectiveness of grade retention as a practice for remedying poor academic results (e.g., Holmes, 1989; Jimerson, 2001) and studies that point to the negative non-cognitive side effects of grade retention (Goos et al., 2013; Jimerson et al., 1997; Pagani et al., 2001), we argue that the practice of grade retention should be abandoned. In secondary education, it seems appropriate to search for other strategies to remedy poor educational performance. Other programs have been suggested, including summer schools, increased positive parental involvement, remedial activities during and after school hours, and individualized educational programs (for an extended discussion, see Jimerson, 2001; McCoy & Reynolds, 1999).

It is important to acknowledge the limitations of the current study and propose some directions for future research in this area. First, we should reiterate that this study does not have a longitudinal design. Therefore, we cannot capture causality and remain unsure whether grade retention invokes social outcomes or the other way around. For instance, in line with a social capital framework—that sees peer relationships as precursors to academic success (Coleman, 1988)—it has been suggested that retained students are already overrepresented among the socially isolated students prior to being retained (Hong & Yu, 2008). If this is the case, the causal direction would be from social isolation to grade retention, which is contrary to our expectations throughout this study. However, most previous studies into the social outcomes of retention have utilized a longitudinal design (see e.g., Bonvin et al., 2008; Wu et al., 2010). These studies show that the dominant causal direction is from grade retention to peer relationships. For example, as these studies have demonstrated differing short- and long-term outcomes of grade retention, it is likely that the social outcomes are affected by whether or not the student is retained and are not only due to some preexisting social situation. We do call for future longitudinal studies to incorporate the number of same-grade friendships as a dependent variable to investigate these claims. Furthermore, longitudinal designs that span the beginning of primary education through to adolescence usually have to deal with attrition, resulting in small sample sizes that limit the generalizability of the results (Jimerson & Ferguson, 2007). Indeed, most previous retention studies have been limited by small sample sizes and unrepresentative data (Ehmke, Dreschel, & Carstensen, 2010). In this study, we have accounted for this well-known critique of retention research by using an extensive data set that is representative of the Flemish situation. Moreover, we account for another well-established critique of retention research by demonstrating that the multilevel nature of the school context should be taken into account when determining grade retention effects (Demanet & Van Houtte, 2013; Hong & Raudenbush, 2005). However, we do propose that future longitudinal research should attempt to replicate these multilevel findings.

A second limitation of the current study concerns its outcome. As we only focus on the number of same-grade friendships, we do not take into account the friendships retained students might have with their former classmates. While other researchers have suggested that grade retention largely breaks these cross-grade friendships (Pierson & Connell, 1992; Stearns et al., 2007), it is possible that retained students still maintain their former friendships outside regular lessons, for example, during breaks or after school hours. Hence, while we found retained students to have less same-grade friendships, they do not necessarily have a smaller overall friendship network. Previous research, however, has demonstrated that same-grade friendships have beneficial effects on school adjustment (see e.g., Hartup, 1996; Ladd, 1990; Ladd et al., 2012), while cross-grade friendships mostly involve out-of-school activities and therefore may decrease students’ attachment to school (Im, Hughes, Kwok, Puckett, & Cerda, 2013; Mahoney & Stattin, 2000). Moreover, from a structural point of view, same-grade friendships are more likely to last than cross-grade friendships simply because they are less subject to opportunity constraints (see Shrum, Cheek, & Hunter, 1988). Students from different grades—even if they attend the same school—can only meet outside lesson hours, so contact time is rather limited. Because of these structural constraints, students tend to be limited in the number of their cross-grade friendships (see e.g., Moody, 2001). It is possible however that retained students occupy a unique position and are more likely to hold on to their cross-grade friendships, which might explain their lower number of same-grade friendships. Lacking data on cross-grade friendships, we could not determine to what extent this is the case. It might therefore be important for future research to compare the number of cross-grade and same-grade friendships of retained students. Furthermore, through our exclusive focus on friendship quantity, we have not provided any insight into the qualitative aspects of the friendships of retained students, which deal, for example, with the level of received support or the relationship cohesion (see e.g., Claes, 1992; Demanet & Van Houtte, 2012b). A complete study of friendships takes into account both their quantity and their quality (Bukowski et al., 1993). Unfortunately, it was not possible to investigate friendship quality in the current study. It could be important to compare the quantity and quality of friendship relationships in this regard. For instance, it might be true that retained students eventually end up with fewer friendships, but these need not necessarily be less cohesive. At the other extreme, it is possible that friendships between retained students and others in the same grade are less cohesive because of the age difference and the fact that the friendships are relatively new. With all these points taken into consideration, we suggest that future research should distinguish between the quantity and quality of retained students’ friendships.

A last limitation lies in the operationalization of grade retention. Previous studies showed that the timing of retention is important; that is, whether retention has positive, negative, or no social consequences depends on how recently it occurred (Bonvin et al., 2008; Wu et al., 2010). As we had no information about what specific grade students repeated in the past, we could not carry out a fine-grained analysis regarding the timing of retention. To try to account for this issue, we incorporated whether retention had occurred in primary or secondary education, but we admit that this is a crude indicator of the timing of retention.

Conclusion

This study is unique in investigating the association between grade retention and the number of same-grade friendships in adolescence and in researching multilevel questions with regard to the social outcomes of grade retention. First, we find that although grade retention in secondary education is related to the number of same-grade friendships, retention in primary education is unrelated to the number of same-grade friendships in secondary education. Furthermore, the retention composition of the school is found to be directly related to same-grade friendship quantity as all students have fewer same-grade friendships in schools with a higher proportion of retained students. Third, retention composition should be taken into account when investigating the nature of the association between retention and the number of friendships at the individual level. We find that the association between retention in secondary school and the number of same-grade friendships is dependent on retention composition, being negative in schools with low rates of retention but becoming less negative and eventually positive in schools with higher proportions of retained students.