School-based extracurricular activity involvement and high school dropout among at-risk students: Consistency matters

Abstract

Encouraging involvement in school-based extracurricular activities (ECA) may be important for preventing high school dropout. However, the potential of these activities remains underexploited, perhaps because studies linking ECA involvement and dropout are rare and based on decades-old data. Previous studies also ignore key parameters of student involvement. The present study expands and updates this limited literature by using recent data from a high-risk Canadian sample (N = 545) and by considering a range of involvement parameters. Results showed that consistent involvement in the past year was associated with lower odds of dropout (OR = 0.32; 95% CI = 0.17–0.61). However, adolescents who interrupted their involvement during this period (e.g., because of cancelations or exclusions) were as much at risk of dropout as those who were not involved at all. Findings notably imply that excluding students from ECA (e.g., because of No Pass/No Play policies) may heighten their dropout risk.

Leaving high school without a diploma—dropping out—has apparently serious individual and social consequences. Adolescents who drop out of school are more likely to depend on social assistance or be unemployed later during adulthood (Hankivsky, 2008; Rumberger, 2020). They also experience more mental health issues than their graduate peers and commit more crimes (Cahuc et al., 2019; Maynard et al., 2015). Although dropout is probably multi-determined, it is generally thought to occur in the wake of a disengagement from school life (Dupéré et al., 2015; Tinto, 1975). Schools could possibly prevent disengagement and dropout by encouraging sustained involvement in extracurricular activities (ECA). However, the few extant studies linking ECA involvement and dropout are based on decades-old data. They also provide an incomplete description of the kind of ECA involvement that could prevent dropout. Their findings thus need to be replicated, updated, and extended.

Sustaining school perseverance via extracurricular activities

In some low-income high schools, dropout unfortunately remains quite common (Atwell et al., 2019). While some students leaving school prematurely had been struggling for years with academic or behavior problems, others were following, up to the point of their departure, rather unproblematic educational trajectories (e.g., Bowers & Sprott, 2012). Irrespective of long-term circumstances, their departure often coincides with crises such as an episode of intense bullying, the incarceration of a parent or even a bad romantic break-up (Dupéré et al., 2018; Kirk & Sampson, 2013; Samuel & Burger, 2020). The extent to which students will consider staying in school despite long-term difficulties or recent crises could depend on a genuine and steady engagement in at least some aspects of school life (Finn, 1989; Tinto, 1975). By the same token, disruptions that rupture this engagement could rapidly lead to dropout. It is thus important to examine how schools can preserve, especially during the critical period of late adolescence, whatever form of positive engagement students have in school, notably via ECA.

According to the positive youth development perspective (Lerner, 2015, 2017), organized activities offer unique environments for fostering engagement in institutions like one’s local community center or school, as well as healthy psychosocial outcomes more broadly. Involvement in quality activities with developmentally appropriate goals and supportive leaders is seen as a potentially potent way to enhance adolescents’ social connections and sense of belonging and to foster their acquisition of important life skills such as self-regulation and leadership (Lerner et al., 2011; Vandell et al., 2015). Beyond quality, benefits from activity involvement are thought to depend on multiple involvement parameters.

Although community-based organized activities can certainly lead to positive outcomes (Hirsch et al., 2011), school-based ECA appear uniquely likely to sustain engagement in school life and school perseverance (Eccles & Gootman, 2002). Skills and relationships developed within school-based ECA can be readily leveraged to foster school perseverance, especially if these activities are led by school staff (Bennett, 2015; Turnnidge et al., 2014). Other involvement parameters include the nature (e.g., arts, sports) and number (one versus multiple) of ECA. In general, the more varied the involvement, the more students are expected to benefit from it, even though some activities are thought to particularly foster certain outcomes (e.g., creativity through artistic activities; Vandell et al., 2015; see also Farb & Matjasko, 2012). Finally, another parameter, dosage, encompasses intensity (e.g., hours per week), as well as length (e.g., years of involvement) and consistency (e.g., continuous versus interrupted involvement; Vandell et al., 2015). Although sustained, uninterrupted ECA involvement is considered preferable, it cannot be assumed that it will be observed for all students, and the importance of examining the causes and consequences of interruptions has been noted, particularly when it comes to involuntary interruptions (e.g., that result from exclusions; Gilman et al., 2004).

In short, theory suggests that ECA involvement is likely to support school engagement and perseverance to varying degrees, depending on multiple parameters, specifically the number of activities and the nature, intensity, length and consistency of involvement. The next section overviews studies testing these assumptions empirically. Studies correlating ECA involvement with educational outcomes other than dropout are reviewed first, followed by an overview of the limited literature examining this last outcome specifically.

Extracurricular activities and school adjustment

Most empirical studies linking ECA involvement and adolescents’ adjustment in school are correlational rather than experimental (Farb & Matjasko, 2012; Roth & Brooks-Gunn, 2016; Vandell et al., 2015). Because adolescents who enroll in ECA tend to be more advantaged than those who do not, studies typically attempt to disentangle the contribution of ECA involvement by including statistical controls (see Morgan and Winship (2015) for the strengths and limitations of this approach). Controlling for potentially confounding factors (e.g., previous academic achievement), studies suggest that ECA involvement is associated with better grades and attendance (Busseri et al., 2006; Durlak et al., 2010), more appropriate behavior (Guest & McRee, 2009), higher educational aspirations (Fredricks & Eccles, 2006) and a greater attachment to school (Barnett, 2007; Dotterer et al., 2007).

Consistent with theory, the apparent benefits of ECA involvement seem to depend on various parameters. Intense, diverse and sustained involvement is generally related to more positive educational outcomes than limited involvement (e.g., Busseri et al., 2006; Fredricks & Eccles, 2006). In terms of the nature of the activity, whereas involvement in nonsport ECA is generally associated with favorable outcomes, involvement in sport ECA is not always associated with positive educational outcomes and is sometimes even associated with negative ones. For instance, Fredricks and Eccles (2008) found that involvement in sport ECA in grade 8 is associated with a decreased perception of school importance by grade 11. Such findings do not necessarily reflect negative socialization processes, as sports ECA sometimes attract adolescents who value drinking, substance use and even deviant behaviors, pointing to potential selection effects (e.g., Denault & Poulin, 2016; for a review, see Vandell et al., 2015).

Extracurricular activities and dropout

Compared with the literature just reviewed on schooling adjustment in general, the ECA literature focusing specifically on dropout is much more limited in terms of number of studies, range of parameters considered, and data sources. In fact, it relies on only four US samples from the 1980s and early 2000s.

Three of these data sources consist of large, nationally representative surveys. Given that sports ECA are not always associated with positive educational outcomes, analyses of these surveys have distinguished between sports and other ECA. Adjusting for background characteristics, findings initially suggested that adolescents involved in sports or other ECA at one point in time were less likely than uninvolved peers to drop out a few years later, with this apparent protective role particularly marked for students involved in many types of ECA or who spent more hours per week in such activities, although the observed trends varied somewhat as a function of race or ethnicity (McNeal, 1995; Neely & Vaquera, 2017; Peguero et al., 2016; Ream & Rumberger, 2008; Rumberger, 1995). However, a less consistent picture emerged when the same data were reanalyzed in two studies using more stringent approaches based on instrumental variables (predicting ECA involvement but not adjustment outcomes). A first discordant study focusing on sports ECA and using height as an instrument found no link between participation and dropout (Eide & Ronan, 2001). Yet, the second study, using ECA offerings and eligibility requirements as instruments and focusing on sports and civic club participation found that both types of ECA were associated with a lower probability of dropout (Crispin, 2017). In this last study, the apparent protective role of ECA involvement was observed for both at-risk and non-at-risk students (differentiated mostly on sociodemographic characteristics).

The last series of studies are based on a smaller convenience sample (Mahoney, 2000, 2014; Mahoney & Cairns, 1997). In this regional study launched in the 1980s, adolescents involved in ECA were less likely to eventually drop out than similar uninvolved peers. Effect sizes tended to be larger than in survey-based studies, especially within high-risk subgroups of students from low socioeconomic backgrounds showing poor academic performance (e.g., grades, retention) and marginal social integration (e.g., few friends, unpopular). To illustrate, in the at-risk subgroups, adolescents who were uninvolved in ECA during high school had a dropout rate up to 26 times higher than involved high-risk peers. For all subgroups, ongoing ECA involvement during high school was more strongly associated with school perseverance than earlier involvement during middle school, suggesting that the potential protective role of ECA involvement could be transitory. Also, the link between ECA involvement during high school and dropout was significant or marginally significant for all types of activities, including sports, and tended to be slightly stronger for involvement in multiple activities rather than in a single one.

Limitations of extant studies on dropout

In sum, although findings from the few extant studies are not entirely coherent, they suggest that ECA involvement could protect against dropout. These studies, however, suffer from two limitations. First, they are based on a handful of somewhat dated data sets in which participants’ ECA involvement occurred two to three decades ago. Over the last decades, changes have made adolescent involvement potentially more relevant for school perseverance than in the past. ECA involvement has notably grown significantly in the wake of massive recent efforts to increase the availability of such activities (see Hirsch et al., 2011; Vandell et al., 2015). For adolescents currently in high school, noninvolvement may thus have become quite rare and entail more marginalization than before. Furthermore, increased opportunities for ECA involvement occurred alongside broader transformations of educational systems, with schools becoming larger and more diverse (Crosnoe, 2011). In increasingly large and impersonal high schools, ECA may offer smaller social niches in which it is easier for students to develop a sense of belonging.

Second, studies focusing on dropout suggest that involvement may be beneficial only under certain circumstances. However, only partial guidance can be offered to schools regarding the type of ECA involvement needed to prevent dropout, because extent studies have not examined two key aspects of dosage, duration and consistency of involvement over time (see Vandell et al., 2015). This gap is significant given that the longer adolescents are involved in ECA, the more likely they are to develop skills, a sense of competence and positive relationships with adults and same-aged co-participants (Denault & Guay, 2017; Farb & Matjasko, 2012; Fredricks & Eccles, 2006).

Yet, some benefits of even long-term ECA involvement might not always outlast the end of a given activity and might in some cases dissolve rapidly. This possibility has received very little research attention, even though ECA interruptions are quite frequent. For instance, one study found that about a third of adolescents who were involved in a given activity at one point in time had discontinued their involvement a year later (Persson et al., 2007). Although some voluntary interruptions (e.g., to increase leisure time) are probably inconsequential, others may cause enough distress or frustration to amount to a crisis. To this point, it has been shown that ECA dismissals following selection procedures or auditions can cause clinically significant distress (e.g., see Barber et al., 2005; Barnett, 2007; Blakelock et al., 2016). Also, applied behavioral research shows that revoking students’ privileges often backfires and stokes a desire to revolt (Landrum & Kauffman, 2013), a situation that is bound to happen under common policies making ECA involvement conditional on adequate grades and good conduct (like No Pass/No Play, see Burnett, 2000; Crispin, 2017). Because internalized distress and external manifestations of anger are associated with dropout (Dupere et al., 2018; Esch et al., 2014), the protection apparently entailed by ECA involvement with regards to dropout might not endure after this involvement is interrupted, especially when interruptions occur under inauspicious circumstances.

The present study

The goal of this study is to use recently collected data to reexamine the association between ECA involvement and dropout. We first consider the general hypothesis that involvement in ECA, as opposed to noninvolvement, is associated with a lower risk of dropout. We then investigate specific hypotheses regarding the involvement parameters associated with this lower risk. It is expected that the risk of dropout will be particularly low when ECA involvement is diversified (multiple activities), intense, durable and uninterrupted, although sports might have somewhat less of a promotive role than other activities. Our hypotheses are examined in bivariate analyses as well as in multiple logistic regression analyses which control for risk factors for both dropout and noninvolvement in ECA at the individual and family levels (e.g., academic risk, parental unemployment). The data that support the findings of this study are available from the corresponding author, [ET], upon request.

Method

Sample

Twelve public high schools in and around the city of Montreal (Quebec, Canada), participated during either the 2012–13, 2013–14, or 2014–15 academic year.¹ Ten of the twelve schools were situated in communities of concentrated disadvantage, and the two others were located in middle to lower middle-class neighborhoods according to provincial official thresholds (Ministère de l’Éducation du Loisir et du Sport [MELS], 2014). On average across the twelve participating schools, the dropout rate was 36%, which is double the average dropout rate in the province of Quebec, and above the threshold applied in the United States to identify low-graduation-rate high schools (Atwell et al., 2019).

In each participating school, students were first administered, early in the school year, a short screening questionnaire (~10 min) that measured their initial risk for dropout, as well as basic socio-demographics (see Measures). All students at least 14 years of age from the regular or special education sectors were invited to participate, and the vast majority (97%) provided written consent and filled in the questionnaire (N_screened = 6,773). Following this initial screening phase, a second interview phase took place later during the year. The goal of this second phase was to conduct individual interviews with a subset of 45 students per school: 15 who had just dropped out, 15 matched at-risk peers, and 15 with an average level of dropout risk. Interviewed adolescents were asked about their experiences in school over the last 12-month period (or during the 12 months prior to departure from school for interviewees who had dropped out), notably in terms of ECA involvement.

Recruitment for the interviews started as soon as the screening phase was over and continued for 12 months. First, to reach students who dropped out in the calendar year following screening, designated school staff informed the research team as soon as a student dropped out, and meetings were quickly arranged for those who consented to be interviewed (usually within a few weeks after a student’s departure). Second, following a matched case-control logic, very shortly after each completed interview with a recent dropout (again, usually within a few weeks), a second interview was conducted with a persevering student from the same school, sector (regular or special education) and sex who showed the closest dropout risk score based on the index administered during the screening phase (see Measures). When two or more students were eligible, the one that was most similar in terms of age and family characteristics (ethnicity, socioeconomic status, structure) was selected. Third, schoolmates with scores on the risk index that were close to their school’s average were invited to participate to form an additional, not-at-risk or “average” comparison group.

Overall, 70% of the targeted adolescents accepted to be interviewed. This rate was slightly lower among dropouts (65%), but still substantially higher than what is usually expected among this subpopulation (typically below 35%, see Dupéré et al., 2015). Among the two other categories, 70% of the matched at-risk but persevering students consented to be interviewed, and 77% of the “average” comparison group participated. In total, 545 adolescents (M_age = 16.3, SD = 0.9) were interviewed.

The sociodemographic makeup of the sample is presented in Table 1, separately for each of the three groups of participating adolescents (dropout, matched at-risk and not-at-risk). Overall, the sample comprises a similar proportion of adolescent boys and girls and is ethnically diverse, with 35% of participating adolescents originating from immigrant families with parents born outside Canada. As expected by design, the not-at-risk group presents a generally favorable sociodemographic, individual and contextual profile compared with the two other groups. Also by design, the dropout and the matched at-risk groups do not significantly differ on the main screening variables used for matching (dropout risk score, sex and sector [regular versus special education]). Their socioeconomic and demographic background is also similar for a number of secondary variables used for matching to the extent possible, including age, immigrant and visible minority status, parental education and maternal employment. However, the matching procedure was not perfect since paternal unemployment and parental separation/divorce were more common among dropouts than matched at-risk peers. Differences between these two groups also emerged for variables assessed after the screening phase, during the individual interviews, and thus not considered in the matching procedure (externalized symptoms and contextual risks, see Measures for details).

Table 1.

Participants’ characteristics.

	Dropout		Matched at-risk		Not-at-risk
	(n = 183)		(n = 183)		(n = 179)
	M/%	SD	M/%	SD	M/%	SD
Socio-demographics
Male	54.1		54.1		48.6
Age	16.5a	0.9	16.4b	1.0	16.0a,b	0.8
Immigrant status	32.8		35.0		36.3
Visible minority	19.1		24.0		26.8
Parental education1	2.5a	1.0	2.6	0.9	2.7a	1.0
Maternal employment	69.4		70.5		69.8
Paternal employment	69.4a		80.3a		78.2
Separated/divorced parents	69.9a,b		53.6a		50.8b
Individual risk factors
Special education	42.6a		45.9b		4.5a,b
Dropout risk index (global)	1.1a	2.1	1.3b	1.9	0.6a,b	0.5
Dropout risk index (items)
Retention2	2.3a	1.0	2.3b	1.0	1.4a,b	0.6
Appreciation of school3	2.2a	1.0	2.4	0.8	2.5a	0.6
Importance of grades4	3.0a	0.8	3.1	0.6	3.3a	0.6
Academic aspirations5	4.2a	1.2	4.4b	1.4	4.9a,b	1.1
Perceptions of grades6	2.7a	0.9	2.7b	0.8	3.0a,b	0.6
Language arts grades7	7.2a	2.5	7.2b	2.3	8.0a,b	1.5
Math grades7	6.9a	3.1	6.3b	2.8	8.0a,b	2.4
Externalized symptoms8	1.1a,b	1.8	0.6a	1.4	0.4b	1.1
Contextual risk factors
Severe life events9	0.6a,b	1.1	0.3 a	0.6	0.2b	0.5
Moderate life events10	0.8a	1.0	0.6	1.1	0.4a	0.9
Severe chronic stressors11	0.9a,b	1.1	0.5a	0.8	0.5b	0.9

Note: Means and percentages sharing subscripts in each row differ significantly at p < .05, based on t tests (for means) or chi-square tests (for percentages).

¹Maximum level of education attained by one parent (1 = primary to 4 = university).

²Number of retentions (1 = none to 4 = three times or more.)

³Attitude toward school (1 = I do not like school at all to 4 = I like school a lot).

⁴Importance of grades (1 = not important at all to have good grades to 4 = very important to have good grades).

⁵Aspirations (1 = no particular aspirations to 6 = university aspirations).

⁶Perceptions of grades (1 = among the worst students to 5 = among the best students).

⁷Language arts/math grades (1 = 0%–35% to 14 = 96%–100%).

⁸Externalized symptoms (0 = no symptoms to 8 = eight symptoms or more).

⁹Severe life events (0 = none to 8 = eight).

¹⁰Moderate life events (0 = none to 8= six).

¹¹Severe chronic stressors (0 = none to 5 = five).

Measures

High school dropout

Adolescents were considered to have dropped out when they met at least one of three conditions according to school records. First, they filed an official notice of schooling termination before obtaining a diploma. Second, they asked for a transfer to the adult sector (GED equivalent). These students are considered as non-graduates because GED graduates’ outcomes are more similar to dropouts’ than to high school graduates’ (Heckman et al., 2014). Third, those who had stopped attending school for at least a month without justification and without filing a notice of termination or asking for a transfer were also counted as dropouts.

Control variables

Questions on students’ socio-demographic backgrounds were inserted in the brief questionnaire booklet completed during the screening phase. Students were asked about their sex, age and minority (i.e., non-White) and immigrant (i.e., at least one parent born outside Canada) status, as well as their family structure (i.e., intact versus non-intact), parents’ employment status (i.e., employed versus not employed), and parents’ education (1 = primary to 4 = university).

Three variables were also used to capture students’ initial individual risk profile. First, a validated risk index administered as part of the screening questionnaire captured participants’ general propensity for dropout based on seven key risk factors such as grade retention, low school engagement and negative attitudes toward school (Archambault & Janosz, 2009; see Table 1 for details). In the current sample, this index showed good predictive validity (with an area under the ROC curve = .81) and predicted dropout more accurately than administrative data about failure, truancy and disciplinary suspensions (Gagnon et al., 2015). Second, students were also asked in the screening questionnaire about their enrollment in special education (because of learning or conduct/emotional problems), another key marker of risk. Third, during the interviews, the presence of externalizing symptoms (related to inattention/hyperactivity and conduct problems) in the past year was assessed with the Structured Clinical Interviews for DSM (SCID; Spitzer et al., 1997), and scores were rated on an 8-point scale (0 = no symptoms to 8 = eight symptoms and more; see Dupéré, Dion, Leventhal, Archambault, Crosnoe et Janosz(2018) for details).

To further capture contextual vulnerabilities, interviewees were also asked about their exposure to stressful life events or chronic stressors in the past 12 months. Stressors were assessed with the Life Events and Difficulties Schedule (Dupéré, Dion, Harkness, McCabe, Thouin & Parent, 2016; LEDS-Brown & Harris, 1978), a semi-structured, interview-based instrument often considered the gold standard for measuring psychosocial stressors (Grant et al., 2004; Harkness & Monroe, 2016). Stressors were then classified according to their intensity in three categories: moderate life events (e.g., school suspension; 0 = no event to 6 = six events), severe life events (e.g., sexual abuse; 0 = no event to 8 = eight events) and moderate to severe chronic stressors (e.g., physical illness; 0 = no chronic stressor to 5 = five chronic stressors).

Involvement in extracurricular activities

Interviewees were asked about whether they were involved in ECA in the past year. To be considered an ECA, the activity had to be organized by the school, supervised by an adult and take place on a regular basis within the school’s premises. About a third of the students (30%, n = 166) were involved in activities that met these criteria. This rate is lower than those observed in general population samples, which is not surprising, given the oversampling of low-SES students at high risk for dropout. In fact, the 30% rate of involvement is virtually identical to that observed in another low-SES sample of Quebec adolescents (Denault & Guay, 2017).

To gather detailed information about each of these ECA, follow-up questions were asked to adolescents who had been involved in at least one ECA in the past year. Involved adolescents were asked to specify the nature of each activity (e.g., basketball team, drama club, orchestra, etc.). Their answers were used to derive a number of activities index, distinguishing 1 = involvement in only one activity (e.g., hockey only) versus 2 = two activities or more (e.g., hockey and football or student council and running club).

Answers were recoded in two dummy categories indicating whether their main ECA (i.e., the one they practiced more hours per week or had been involved in for the longest) fell in the “sport” or “other” category. The “other” category, which comprises arts, clubs and civic engagement ECA, could not be further separated owing to low prevalence (e.g., only ten adolescents were involved in clubs ECA). For the same reason, distinctions could not be made between team sports and individual sports. For each ECA, adolescents were also asked about start and end dates. Intervals between these dates were categorized to reflect the maximum length of involvement (1 = up to a few months; 2 = one year or more; 3 = two years or more; 4 = three years or more).

Additionally, start and end dates were used to assess ECA involvement consistency. ECA involvement was considered as potentially interrupted when an end date was reported, indicating that the interviewee had stopped being involved in a given activity while still attending school. In such cases, the interviewee was questioned about the reason for the interruption. When the interruption was related to seasonal cycles, and the adolescent had the intention of re-enrolling for the next season, ECA involvement was considered uninterrupted. With these criteria, 23% of involved interviewees had discontinued their ECA involvement (n = 38/166). The reasons underlying the interruptions were classified as voluntary (n = 11) and involuntary (n = 27). Causes of involuntary withdrawals included ECA cancelations for administrative reasons, transfers to a new school where the activity was not available, injuries or exclusions based on criteria related to ability, grades or behavior. In sum, for the consistency variable, ECA involvement was dummy-coded with one dummy capturing uninterrupted involvement (up to the time of dropout or of the interview) and another one capturing involvement interrupted voluntarily or involuntarily.

Interviewees were not asked about the intensity of their involvement, that is, about the number of hours per week spent in each activity. To access this information, schools’ recreation coordinators in each of the 12 participating schools were consulted about the hours per week associated with each activity offered in their school (for a similar approach, see Mahoney, 2000; Mahoney & Cairns, 1997). Coordinators’ answers were matched with the information provided by the students about the nature/type of activities in which they were involved in their school. Through this procedure, it was possible to derive an intensity index, coded in four categories: (1 = two hours/week or less; 2 = between two and five hours/week; 3 = between five and 10hours/week; 4 = more than 10 hours/week). In statistical analyses, only one ECA per student was included, due to the small number of participants involved in more than one activity. In cases where adolescents were involved in more than one ECA, the one retained in the analyses had either the highest intensity of involvement (number of hours per week) or the greatest length (when two ECA had the same intensity).

Finally, students were not asked during the interviews about how they got involved in a given ECA. However, recreation coordinators in the twelve participating schools were interviewed about how ECAs were generally advertised and managed in their respective school (see McCabe et al., 2018). All schools used mass advertisement techniques (e.g., posters, public address systems, mailers), but some invested special efforts to recruit struggling students (e.g., by touring special education classes). Other schools were less inclusive, for instance, because ECAs targeted in priority “talented” or high-GPA students. Such between-school differences are considered in our statistical analyses by modeling the nested nature of the data (i.e., students within schools).

Analytical plan

Our hypotheses were tested using basic bivariate analyses as well as multiple logistic regressions. The first type of analysis provides easy to interpret descriptions of findings while the second type controls for potential confounders (see Table 1) and clustering of students within schools. Each type of analysis involved two steps. The first step aimed at determining whether ECA, regardless of its parameters (e.g., intensity), was associated with a lower risk of dropout. The second step aimed at assessing whether this association changed when parameters of ECA involvement were considered.

As in previous analysis based on this sample (Dupéré et al., 2018), the regressions controlling for potential confounders were initially restricted to dropouts and their matched at-risk peers, as non-at-risk students were too different from these two groups to represent an appropriate comparison group (see Table 1 and sample description). However, to improve confidence that our findings did not reflect sampling biases, regression was rerun with the complete sample and with an alternative sub-sample. A further robustness check was also conducted by addressing selection issues with a complementary analytical approach (inverse propensity scoring).

Results

Bivariate analyses

In the first step, a chi-square test based on the full sample (N = 545) showed that the percentage of adolescents involved in ECA was significantly lower among dropouts (18.6%) than among matched at-risk (28.4%) and average (44.7%) students (see the upper panel of Table 2). In the second step examining involvement parameters, analyses (chi-square tests and ANOVAs) were restricted to adolescents who had been involved in ECA (n = 166). Focusing on this group allowed us to determine whether adolescents who had dropped out despite being involved in ECA had been involved in a less optimal manner (e.g., less intensely) than their involved peers who remained in school. For this subsample, only consistency of involvement was associated with dropout (see Table 2, lower panel). Specifically, almost half of the involved dropouts discontinued their involvement over the year, a situation that was 2.5 less likely to occur among their involved peers who remained in school. It is worth noting that although voluntary interruptions were frequent across groups, dropouts were especially likely to have involuntarily interrupted their ECA involvement. A descriptive examination of interviewees’ answers revealed that among dropouts, involuntary interruptions occurred for a variety of reasons, including some that were not related to their grades or behavior, such as school changes or injuries (detailed results available upon request).

Table 2.

ECA involvement parameters among dropouts, matched at-risk and average, not-at-risk students.

			Matched
	Dropout		at-risk		Not-at-risk
	M/%	SD	M/%	SD	M/%	SD
	In the full interviewed sample
	n = 183		n = 183		n = 179
ECA involvement (%)	18.6a		28.4a		44.7a
	Among those participating
	n = 34		n = 52		n = 80
ECA involvement parameters
Nature of the main activity
Sports	64.7		57.7		58.8
Other	35.3		42.3		41.2
Number of activities
> 1 activity	20.6		13.5a		33.8 a
Dosage
Intensity (Nb of hrs/week)1	2.0	0.9	1.9	0.8	2.1	1.0
Length (time since start date)2,3	2.2	1.1	2.2	1.1	2.4	1.1
Consistency
Interrupted involvement	44.1a, b		13.5a		20.0b
Voluntary	8.8		9.6		3.8
Involuntary	35.3a		3.8a		16.2a

Note: Means and percentages sharing subscripts in each row differ significantly at p < .05, based on ANOVAs (for means) or chi-2 tests (for percentages).

¹Response scale from 1= ≤ 2 hrs to 4 = ≥ 10 hrs.

²Two cases had missing values for this variable.

³Response scale from 1 = ≤ a few months to 4 = ≥ 3yrs.

ECA = Extracurricular activities.

Multiple logistic regressions

In the first step, dropout was predicted from the presence versus absence of ECA involvement. As can be seen in Table 3 (Model 1), the risk of dropout was lower among adolescents involved in ECA than among their non-involved peers (over and above other potentially confounding factors).

Table 3.

Multiple logistic regressions linking ECA involvement and high school dropout among dropouts and matched at-risk students (n = 366), controlling for sociodemographics, individual, and contextual risk factors.

Dummy ECA variables, with absence of involvement as the reference category	B	SE	Wald	OR	95% CI
Model 1—ECA involvement
Involvement (dichotomous)	−0.71**	0.27	7.00	0.49	0.29–0.83
Model 2—Nature of ECA involvement
Involvement in sports ECA	−0.59†	0.30	3.78	0.56	0.31–1.01
Involvement in “other” ECA	−0.93*	0.43	4.55	0.40	0.17–0.93
Model 3—Length of ECA involvement
A few months	−0.81†	0.43	3.59	0.45	0.19–1.03
At least one year	−0.50	0.49	1.05	0.61	0.23–1.58
At least two years	−0.98	0.87	1.28	0.38	0.07–2.05
Three years or more	−0.76	0.48	2.52	0.47	0.19–1.19
Model 4—Intensity of ECA involvement
Two hours per week or less	−0.98***	0.26	13.75	0.38	0.23–0.63
Two to five hours per week	−0.80†	0.41	3.75	0.45	0.20–1.01
Five hours per week or more	−0.35	0.38	0.82	0.71	0.33–1.50
Model 5—Number of activities
One activity only	−0.80**	0.30	7.09	0.50	0.25–0.81
Two or more activities	−0.21	0.46	0.22	0.81	0.33–1.98
Model 6—Consistency of involvement
Uninterrupted	−1.22***	0.32	14.74	0.30	0.16–0.55
Interrupted	0.57	0.61	0.86	1.76	0.53–5.83

Note: ECA = Extracurricular activities. The nested structure of the data was considered via the “cluster” option of the SAS SURVEYREG procedure. All the covariates listed in Table 1 were incorporated as controls, including sociodemographics, individual (for the dropout risk index, global scores were incorporated, not individual items) and contextual risk factors (full results available upon request).

^†p < .10;

^*p < .05;

^**p < .01;

^***p < .001.

In the second step, further models were conducted to predict dropout while considering the five parameters of ECA involvement. To avoid multicollinearity problems, a separate regression was carried out for each involvement parameter. One set of dummy variables was created per involvement parameter, with an absence of involvement in ECA as the reference category. The results of these five regressions are presented in Table 3, Models 2 to 6. To facilitate interpretation, the five graphs of Figure 1 show the strength of association between ECA involvement and dropout for each set of dummy variables. In each graph, the solid line indicates the beta observed (in Model 1, see Table 3) for presence versus absence of ECA involvement. To illustrate, the upper left graph for “Nature” shows that as compared to absence of ECA involvement, involvement in “sport” and in “other” activities apparently plays a similar promotive role, of a magnitude on par with that observed for involvement versus noninvolvement. Findings are similar for length, intensity and number of activities: compared with an absence of involvement, the promotive role of ECA involvement does not seem to vary much as a function of its length or intensity, or as a function of the number of activities. Finally, although interrupted involvement did not protect against dropout, uninterrupted ECA involvement appeared to play a particularly strong promotive role. The distinction between involuntary and voluntary interruptions was not examined at this step as it made the regression models unstable.

Figure 1.

Strength of the association between ECA participation and non-dropout, as function of five key dimensions of participation (nature, length, intensity, number of activities, consistency). The y-axis presents betas from logistic regression models regressing high school dropout on different dimensions of ECA participation (represented as dummy variables), while incorporating the full set of control variables (n = 366, see Table 3 for details). To facilitate interpretation, the y-axis scale was reversed [higher elevations = more negative betas] so that higher elevations represent a stronger promotive effect. The horizontal lines represent the elevation of the beta associated with the dichotomous variable representing participation versus nonparticipation, obtained in a preliminary model incorporating no distinctions as a function of the five dimensions of participation (but incorporating all control variables, see Table 3—Model 1). According to theoretical expectations, the blue lines should become gradually more elevated as they move towards the right-hand side of each panel. Such a trend is clearly observed only for consistency.

Second-step regression findings suggest that consistency of involvement is particularly relevant. The strongest effect size was indeed associated with uninterrupted involvement, with odds of dropping out about 70% lower for uninterrupted involvement as compared to an absence of involvement. Also, model fit statistics (AIC, BIC) calculated for each model (Table 4) show that inclusion of involvement parameters improved model fit only when consistency was considered. In contrast, model fit worsened when each of the other four parameters were considered. Fit statistics thus suggest that, apart from consistency, taking into account involvement parameters is not informative enough to improve the prediction of dropout based on presence or absence of ECA involvement (controlling for confounding variables). Fit statistics suggest a similar pattern when continuous rather than dummy variables were used for length, number of activities and intensity.

Table 4.

Multiple logistic regression model fit statistics for each ECA involvement model tested (Presented in Table 3).

	AIC	BIC
Initial model not considering the dimensions of ECA involvement
Model 1—Dichotomous ECA involvement (vs no involvement)	479.19	541.63
Models testing each of the five dimensions of ECA involvement
Model 2—Nature	480.74	547.08
Model 3—Length	484.69	558.84
Model 4—Intensity	481.66	551.91
Model 5—Number of ECAs	480.42	546.77
Model 6—Consistency	471.12	537.47

Note: ECA = Extracurricular activities. AIC = Akaike Information Criteria; BIC = Bayesian Information criteria. The AIC and BIC values in bold highlight the models with a better fit than the first basic model comparing ECA involvement and noninvolvement using a dichotomous variable, and without considering any of the five involvement parameters.

Robustness checks

A final analytical step consisted of probing the robustness of the main results by rerunning the final model (model 6—Table 3) under different conditions. The first robustness check reproduced model 6 while applying inverse propensity for treatment weighting (IPTW; Austin & Stuart, 2015) techniques. These techniques contribute to further reduce the likelihood that observed associations between a “treatment” (i.e., consistent ECA involvement) and an outcome (i.e., dropout) solely reflect selection issues. When Model 6 was rerun with IPTW, the association between consistent involvement and dropout remained significant and essentially unchanged (full results available upon request).

A second series of robustness checks involved rerunning Model 6 on slightly different subsets of participants. One variation involved running it on the complete sample of interviewees instead of only the two matched groups. Another variation involved running Model 6 with only the adolescents from the ten schools located in highly disadvantaged communities, with a restricted sample including only the two matched groups, and then with the three groups. Again, all the iterations produced results almost identical to those of Model 6 in Table 3, with uninterrupted ECA involvement associated with a lower risk of dropout, but no significant association for interrupted involvement (full results available upon request).

Discussion

Extant studies based on data from the 1980s and early 2000s suggested that ECA involvement lowered the risk of dropout among high school students. However, this link was not observed consistently, perhaps because the timing of ECA involvement in relation to dropout was not established precisely. Timing seems critical since the present findings suggest that ECA protect from dropout while students are involved in the activities. This apparent protective effect of sports and other ECA ceased to be detectable when students’ involvement was interrupted. We discuss below how our findings replicate and, to some extent, clarify findings from previous studies.

Replication and clarification of previous findings

Although a correlation should not be mistaken for a causal relation, it seems clear, based on both present and previous findings, that consistent ECA involvement is generally associated with a lowered risk of dropping out. Mahoney et al. (1997) had also reported a strong link between ECA involvement and school perseverance, probably because their analyses were based on the kind on consistent involvement that gets mentioned in yearbooks. We observed a lowered risk of dropout regardless of the nature of the activity, that is for both sports and other ECA, even though in previous studies sports ECA were not always associated with positive educational outcomes (e.g., Fredricks & Eccles, 2008). Based on informal discussions with recreation coordinators, it is our impression that, at least in some schools, sports ECA were important for academically marginalized students (see also Burnett, 2000). Perhaps sports ECA can support the school perseverance of some students without having a positive effect on grades.

Our findings do not provide unequivocal support for the idea that an intense involvement in multiple ECA is associated with better educational outcomes, as found in previous studies (see Vandell et al., 2015). Our participants were at a lower risk of dropout even when they were involved in only one activity or were involved two hours or less per week. Furthermore, considering the intensity of involvement or number of activities did not substantially improve the prediction of dropout according to indices of model fit. One possible explanation for this finding is that most of our participants were in their late teens. Many of them probably had a part-time job, a demanding social life and a relatively clear idea of their talents and interests. Perhaps a relatively light involvement in a single well-chosen activity was all they had the time for and was enough to make them feel connected to the life of their school.

However, being involved in a single ECA is a tenuous link to school life that can be easily severed for a variety of reasons not necessarily under students’ control (e.g., a decision by the school to discontinue the cheerleader activity). Although we could not explore these findings in regression analyses due to small numbers in some cells, there were indications in the bivariate analyses that involuntary interruptions of ECA involvement were associated with a high risk of imminent dropout among vulnerable students (bivariate analyses were conducted only on students who dropped out and their matched at-risk peers). Adolescents, even in their late teens, do not have an extensive experience in managing personal crises or the cognitive maturity to do so effectively (Steinberg, 2014), and they can indeed react impulsively when confronted with an intense frustration, such as being excluded from the football team. It is possibly a mistake to think that long term considerations loom large in adolescents’ decision to stay in school or to quit. A growing number of studies show their decision-making process is closely influenced by their current circumstances, especially when they are under stress or when their reputation within the peer group is threatened (e.g., Gardner & Steinberg, 2005; Somerville et al., 2019).

If this interpretation is correct, the protective effect of ECA involvement cannot be regarded as necessarily cumulative over time (e.g., Bornstein et al., 2006): the positive consequences of months of involvement could apparently be negated by one interruption, at least when it comes to school perseverance in environments where high school dropout is not rare. On the positive side, a short-term ongoing involvement of a few months seemed enough to protect against dropout. Other authors have similarly observed that recent involvement tends to be more strongly linked to school perseverance than earlier involvement (Mahoney & Cairns, 1997). The idea that adolescence is a tumultuous time (e.g., Barbot & Hunter, 2012) raises the possibility of abrupt fluctuations in adolescents’ adjustments and the need for a constant, uninterrupted support on the part of the school environment (see also Afia et al., 2019). This idea also raises the interesting possibility that encouraging ECA involvement (e.g., through school-wide interventions) could generate positive effects on school perseverance over the short term.

Practical implications

Our findings suggest that low-income high schools could prevent dropout by encouraging sustained ECA involvement. Specifically, it seems more important for schools to offer a variety of ECA in order to reach as many students as possible rather than to encourage involvement in a specific type of ECA (e.g., sports versus other). When it comes to preventing dropout, a diverse, intense involvement in multiple activities does not seem required. A relatively light involvement in one ECA could be enough if this involvement is not interrupted. Schools should thus prevent, to the extent possible, involuntary interruptions.

The present study has implications regarding No Pass/No Play policies in effect in many US states. These policies stipulate that students underperforming academically are to be excluded from ECA. These policies and similar ones implemented outside of the US, including in Canada, are meant to motivate students to do well in school, as a means of retaining the privilege of being allowed to be involved in ECA. Regardless of their effectiveness in this regard, the possibility that No Pass/No Play policies have significant downsides has long worried practitioners, especially regarding students at risk of dropping out (Burnett, 2000). Giving credence to these concerns, the present findings point toward potential counter-productive effects (see also Crispin, 2017). Thus, No Pass/No Play policies and other school-based practices that set stringent criteria for ECA involvement should be evaluated carefully to determine whether their benefits truly outweigh their potential costs (see also McCabe et al., 2018).

Limitations

The present study is not without limitations. Five are worth mentioning. First, small cell sizes limited our ability to fully explore the role of some involvement parameters, for instance involuntary interruptions. Second, even if interviewers were careful to elicit detailed and candid information from participants, ECA involvement was measured exclusively through self-report. Also, to minimize recall problems, interviewers collected information only on current or recent ECA involvement that had still been ongoing at least for some time in the past year. This strategy allowed us to capture recent patterns of ECA involvement and interruptions thought to be most important for school perseverance (Mahoney & Cairns, 1997) but provided no information on older interruptions dating back to over a year. Third, we deliberately over-sampled low-income students who had dropped out of school or were at risk of doing so. As a result, our findings may not generalize to less disadvantaged samples (e.g., middle-class adolescents who drop out of school). Fourth, we did not examine the quality of adolescents’ experiences in ECA (e.g., Denault & Poulin, 2016), a key parameter of involvement that could interact with interruptions (e.g., exclusions from high-quality ECA could carry an especially elevated risk of dropout). Finally, like most studies of ECA, our design was correlational rather than experimental, precluding firm conclusions on causal relations.

Future studies

To determine whether involvement in ECA is indeed causing a reduction in dropout risk, experimental designs should be considered when feasible (for a similar argument, see Roth & Brooks-Gunn, 2016). Schools would probably be highly reluctant to randomly restrict ECA involvement for some students, and asking schools to do this would arguably be unethical in a research context. Fortunately, other more acceptable and feasible experimental options can be tried. For instance, intervention schools could identify and proactively encourage students at-risk of dropout to be involved in at least one ECA of their choice throughout the school year (see McCabe et al., 2018). Alternatively, schools in which No Pass/No Play policies are in effect could be recruited, and a randomly selected sample of these schools could change their policies by offering tutoring sessions to underachieving students rather than interrupting their ECA involvement.