Journal of Adolescence 32 (2009) 651e670

www.elsevier.com/locate/jado

Student engagement and its relationship with early high

school dropout
Isabelle Archambault*, Michel Janosz, Jean-Sébastien Fallu, Linda S. Pagani
École de Psychoe´ducation, Universite´ de Montre´al, Canada

Abstract

Although the concept of school engagement figures prominently in most school dropout theories, there
has been little empirical research conducted on its nature and course and, more importantly, the association
with dropout. Information on the natural development of school engagement would greatly benefit those
interested in preventing student alienation during adolescence. Using a longitudinal sample of 11,827
French-Canadian high school students, we tested behavioral, affective, cognitive indices of engagement
both separately and as a global construct. We then assessed their contribution as prospective predictors
of school dropout using factor analysis and structural equation modeling. Global engagement reliably pre-
dicted school dropout. Among its three specific dimensions, only behavioral engagement made a significant
contribution in the prediction equation. Our findings confirm the robustness of the overall multidimen-
sional construct of school engagement, which reflects both cognitive and psychosocial characteristics,
and underscore the importance attributed to basic participation and compliance issues in reliably esti-
mating risk of not completing basic schooling during adolescence.
Ó 2008 The Association for Professionals in Services for Adolescents. Published by Elsevier Ltd. All rights
reserved.

Keywords: School engagement; Achievement motivation; Involvement; Commitment; Participation; School dropout

* Corresponding author. École de Psychoéducation, Université de Montréal, C.P. 6128, Succ. Centre-ville, Montréal,
Québec, Canada H3C 3J7. Tel.: þ1 514 279 7529; fax: þ1 514 343 6951.
E-mail addresses: [email protected] (I. Archambault), [email protected] (M. Janosz),
[email protected] (J.-S. Fallu), [email protected] (L.S. Pagani).

0140-1971/$30.00 Ó 2008 The Association for Professionals in Services for Adolescents. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.adolescence.2008.06.007
652 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670

Limited achievement and academic attainment represent two important consequences of

growing up poor (Entwisle, Alexander, & Olson, 2005; Pagani, Boulerice, Vitaro, & Tremblay,
1999). More than ever, younger generations need a basic education in order to successfully partic-
ipate in the demanding labour force that awaits them (Heckman, 2006). Those not earning a high
school diploma face a life-course of underemployment and its correlates, which, for many perpet-
uates their economically disadvantaged origins (Rumberger & Lamb, 2003). Aiming to circum-
vent student dropout and ultimately reduce the intergenerational cycle of poverty, researchers
and clinicians have increasingly become interested in studying the nature and course of its putative
mechanisms.
The construct of engagement is central to most theories of school dropout (Finn, 1989; Tinto,
1975). The present study aims to develop and examine the predictive value and multidimensional
nature of student engagement. We test the contribution of its specific components in predicting
student dropout. Our hope is to generate knowledge that will help orient the development of
educational practices which promote success in high school (Christenson & Thurlow, 2004).

Engagement in theories of school dropout

The construct of student engagement originates in part from Social Control Theory (Hirshi,
1969) which places a great deal of emphasis on individual feelings of attachment and belonging-
ness to social institutions. Youthful antisocial behavior is viewed as a breakdown of the bonds
between the individual and society. Likewise, disengagement could result from a weakened rela-
tionship between the individual and educational institutions. The bonds in Social Control Theory
are characterized by commitment, beliefs, attachment, and engagement. These theoretical
elements have greatly influenced conceptualizations of student engagement in recent theories of
dropout.
In Tinto’s (1975) mediation model, school dropout represents an ongoing and unfolding
process. From the time students enter school, they interact within its academic and social system.
Individual and family background characteristics contribute to their commitment toward this
institution and its academic goals. Individual commitments to specific academic goals directly
influence involvement in school-related tasks and activities. In turn, commitment to school influ-
ences the time invested toward this institution. Taken together, goals and institutional commit-
ments set the course of student engagement from school entry onward. These two
characteristics and their evolution are believed to influence a youngster’s academic and social
experience at school and, in unfavourable conditions, can eventually play a role in the decision
to leave the system altogether.
Finn’s (1989) participation-identification model of school withdrawal also considers the
construct of student engagement. In this model, engagement is defined by identification and
participation at school. Identification refers to a sense of belongingness and the perceived worth
of schooling. Participation comprises four distinct components that range from minimum to
maximum engagement. These are responsiveness to requirements, participation in class-related
initiatives and extracurricular activities, and decision-making. We expect students to identify
more with school as they increase their participation. Conversely, low or absent participation,
predicts gradual disengagement and eventual school withdrawal.
 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670 653

In their dropout prevention model, Wehlage, Rutter, Smith, Lesko, and Fernandez (1989)
introduce concepts of educational engagement and school membership as intermediate steps
that contribute to individual and social development in school. Educational engagement favors
student effort and promotes personal academic success. Membership is successfully reached
when students generate social bonds with peers or adult authorities in the school context.
From this perspective, students who fail to achieve these two goals present higher risks of drop-
ping out.
Finally, in Rumberger and Larson’s (1998) model, engagement is distinguished by two compo-
nents, social and academic, which contribute to academic adjustment. Social engagement is
defined by behaviors such as class attendance, rule compliance, and active participation in
school-related activities and venues. Academic engagement includes student attitudes toward
school and the ability to meet performance expectations. For this model, both types of engage-
ment are essential for understanding the process that underlies school dropout.
Two major similarities emerge between the aforementioned conceptualizations. First, they
conceive engagement as a process that evolves over the course of the school experience. Second,
they underscore the importance of behavioral and motivational aspects of engagement. Behav-
ioral and motivational dimensions of engagement seem more obvious in three of the models.
As an exception, these aspects are less clearly established in Tinto’s model, given that goal and
institutional commitment are both considered to be motivational processes. Institutional commit-
ment, as it is defined, could also represent what the other models view as behavioral engagement.
Through concepts of participation-identification, educational engagement, membership, and
social and academic engagement, we can conclude that both behaviors and attitudes are equally
important in the hypothetical process that leads to school dropout.

Toward a framework of school engagement

Even if similarities exist in the theoretical understanding of student engagement, conceptuali-

zations remain fragmented and untested. As highlighted in literature reviews (Fredricks, Blumen-
feld, & Paris, 2004; Libbey, 2004), authors refer to it in various ways as school bonding and
connectedness (Eggert, Thompson, Herting, Nicholas, & Dickers, 1994; Jenkins, 1997), attach-
ment (Gottfredson, Fink, & Graham, 1994), belongingness (OECD, 2003), involvement (Caspi,
Entner, Moffitt, & Silva, 1998; Finn, 1989), and commitment (Janosz, LeBlanc, Boulerice,
& Tremblay, 1997).
Some authors have recently proposed a more integrated definition of student using multiple
dimensions student engagement (Appleton, Christenson, Kim, & Reschly, 2006; Fredricks
et al., 2004; Jimerson, Campos, & Greif, 2003). A primary advantage of such multidimensionality
is that the concept addresses central and related facets of human development (i.e., behaviors,
affect, and cognition). Another advantage clearly concerns prescriptions for prevention and inter-
vention strategies (Christenson & Thurlow, 2004). In this paper, we adopt the comprehensive defi-
nition advanced by Fredricks et al. (2004). According to their perspective, student engagement
encompasses behavioral, affective, and cognitive dimensions.
Behavioral engagement refers to student conduct that is beneficial to psychosocial adjustment
and achievement at school. This dimension can be divided into three main axes: positive
654 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670

behaviors, involvement in school-related tasks, and participation in extracurricular activities (Fre-

dricks et al., 2004). Behaviors defining all three axes range on a continuum and may be positive
and negative. For the positive behavior axis, school attendance vs. skipping class and compliance
with rules vs. oppositional behavior and represent good examples along the engagement/disen-
gagement continuum (Costenbader & Markson, 1998). When it comes to involvement in
school-related tasks, behaviors could be addressed as to whether and to what extent students
do their homework and participate in classroom-related work and discussions (Posner & Vandell,
1999; Tymms & Fitz-Gibbon, 1992). Finally, studies that focus on participation in extracurricular
activities generally address their frequency of participation (Mahoney & Cairns, 1997).
The affective dimension of engagement refers to feelings, interests, perceptions, and attitudes
toward school. Researchers have operationalized this variable using perceptions of belongingness
(Goodenow, 1993), the perceived benefits and value of education (Eccles, Wigfield, Harold,
& Blumenfeld, 1993), and specific importance of school in helping students reach specific goals
(Bouffard & Couture, 2003; Watt, 2004). Most of these concepts can be treated more generally
or specifically (i.e., according to different subject matter), depending upon the research question.
Cognitive engagement addresses two variables that might affect achievement and psychosocial
adjustment. These are student psychological investment in learning and the use of self-regulation
strategies by students. Cognitive investment in learning covers perceptions of competency, willing-
ness to engage in learning activities and engage in effortful learning, and establishing task-oriented
goals (i.e., performance, mastery, and performance-avoidance goals; DeBacker & Nelson, 2000).
Self-regulation strategies focus on specific learning tools such as memorization, task planning, and
self-monitoring (Ablard & Lipschultz, 1998).

Important considerations in the study of school engagement

As underscored by Appleton et al. (2006), despite the recent efforts consolidate the concept of
student engagement, the theoretical and practical overlap between this multidimensional
construct and the motivational literature remains obvious. Nevertheless, when conceptualized
as having multiple dimensions beyond goal-directed issues, school engagement derives its rele-
vance in its own right, especially given the complexity of school dropout. For example, Janosz,
LeBlanc, Boulerice, and Tremblay (2000) prospectively tested a typology of school dropout
with 1582 Montreal high school students from an urban disadvantaged setting. They were fol-
lowed from early adolescence onward. A third (n ¼ 507) did not complete high school by age
22. Surprisingly, they found that 40% of high school dropouts had previously reported high levels
of school motivation. These were referred to as quiet dropouts. Surprisingly, individuals following
this life-course also showed similar, and sometimes even better, behavioral and psychological
profiles than the average graduate. Another 40% (maladjusted dropouts) experienced severe levels
of school and psychosocial difficulties. Two other interesting life-course patterns also emerged
with the remainder of the dropouts: The disengaged dropout (10%), a strongly unmotivated
student with few socio-emotional difficulties and average grades; and the low-achiever dropout
(10%), who is typically unmotivated with a school experience of failure yet not showing any exter-
nalizing problem behaviors.
 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670 655

The multiconceptual approach we favor in this study integrates several motivational concepts
(e.g., value and interest attributed to school and its academic goals) while also considering
behavior as an important indicator of student engagement/disengagement and dropping out.
Since behavior is more easily measurable than motivation, it remains very useful for identifying
at-risk students. Furthermore, given that students are expected to disengage first psychologically
and then behaviorally (Eccles, 2004), the multidimensional construct of engagement could repre-
sent an effective way to tap different but related levels of student risk. We favor a multidimensional
approach that covers both motivational and psychosocial characteristics. As such, the evolution
of the school experience of the individual can be addressed as a whole. We can also potentially
address whether there are distinct affective, cognitive, and behavioral needs that influence whether
students stay in school or not. The natural evolution and course of student engagement and its
relationship with school dropout has yet to be viewed from a process standpoint. A better devel-
opmental understanding of its complexity will not only help advance theory but also help clini-
cians, teachers, and parents keep students from prematurely withdrawing from basic schooling.

Objective

Theoretical and empirical findings indicate that dropout represents a complex and gradual
process of diminishing school engagement (Ensminger, Lamkin, & Jacobson, 1996; Finn,
1989). This requires an integrative approach to capture the nature and course of its complexity
(Elder, 1995). School transitions are often challenging (Eccles, Barber, Stone, & Hunt, 2003)
and dropout often occurs before 10th grade. As such, the developmental period leading up to
this life-course turning point might be helpful in the study of engagement.
Using the conceptualization offered by Fredricks et al. (2004), this study aims to examine behav-
ioral, affective, cognitive indices of engagement both separately and as a global construct. Next, we
test the contribution of global engagement and its components as prospective predictors of school
dropout. We expect that student low engagement will predict school dropout. Because students
disengage first psychologically and then behaviorally (Eccles, 2004), we expect that the behavioral
component will more reliably predict dropping out because of its proximity to the outcome.

Method

Participants and procedure

Consent forms were sent to parents and obtained for 77.4% of candidates for this study. Partic-
ipants were 11,827 seventh to ninth graders (corresponding to 12 to 16-year-olds, 44.6% boys) from
69 high schools in the French-Canadian province of Quebec, Canada. The average participant age
was 13.1 years (SD ¼ 0.97). Assessments took place in the spring of 2003 when teachers distributed
questionnaires to students in their classes with the assistance of trained graduate students.
Our sample is mostly homogeneous in terms of ethnicity but remains representative of the prov-
ince. Most participants were born in Quebec (89.2%), with the remainder born in another Cana-
dian province (4.9%) or in another country (5.9%). About two-thirds of participants (69.3%)
656 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670

lived with two parents, 20.3% lived with their mother or their father (4.5%), and the remainder
lived in shared custody (4.6%) or in various other family arrangements (1.2%).

Measures

School engagement
As reported in Table 1, 18 self-report items were used to measure overall student engagement
from the beginning of the school year. These items were hypothesized to reflect six first-order
latent concepts that represent three dimensions of engagement that would further converge into
one global engagement construct. Behavioral engagement, which assesses school attendance
and compliance with rules (e.g., ‘‘Have you disrupted the class on purpose?’’), was assessed using
four negatively recoded items on a four-point Likert scale (never to quite often). Affective and
cognitive engagement were both measured using a seven-point Likert scale (which ranged from
strongly agree to strongly disagree). The seven-item affective dimension assessed student enjoy-
ment and their level of interest in school-related challenges and tasks (e.g., ‘‘Do you like
school?’’). The seven-item cognitive dimension assessed student willingness to learn language
arts (French) and mathematics (e.g., ‘‘How much effort are you ready to spend in mathematics?’’).
All original items comprising the behavioral engagement dimension were transformed into

Table 1
Items measuring the first-order latent concepts related to student engagement.
Items
Behavioral engagement
School attendance 1. Missed school without a valid reason.
2. Skipped a class while you were at school.
Discipline 3. Disrupted the class on purpose.
4. Been rude to your teacher.
Affective engagement
Liking school 5. I like school.
6. I have fun at school.
7. What we learn in class is interesting.
8. I enjoy what we do at school.
Interest in school work 9. I am happy when the work is quite challenging.
10. Often, I do not want to stop working at the end of a class.
11. I am very happy when I learn something new that makes sense.
Cognitive engagement
Willingness to learn French languages arts 12. How much time are you ready to spend in French?
13. How much effort are you ready to put into French?
14. How much energy are you willing to put into French?
Willingness to learn mathematics 15. How much effort are you willing to spend in mathematics?
16. How much time are you ready to spend in mathematics?
17. I would like to do/learn more than what we are actually
doing/learning in mathematics class?
18. I can easily spend a lot of time on mathematics problem.
 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670 657

z-scores before conducting the analysis to minimize any potential variance and distribution prob-
lems caused by the limited item-response constraints (Gorsuch, 1983).

School dropout
Information on school registration status was obtained through official records on an annual
basis. Students who never obtained a high school diploma and who were no longer enrolled at
school in the province of Quebec by the end of September 2005 (i.e., 2 years after data collection)
were identified as school dropouts. Out of the 11,827 students, 404 students (3.4%) were identified
as dropouts.

Maternal education
Students were asked to indicate their mother’s highest level of schooling (1 ¼ primary school,
2 ¼ seventh grade, 3 ¼ eighth grade, ., 8 ¼ university).

Course retention
Information on class retention in secondary school was reported by students (e.g., ‘‘Have you
ever failed a secondary school course, but not had to repeat your entire year?’’).

Data analytic strategy

We first generated descriptive statistics and correlations for all studied variables. Next, Explan-
atory Factor Analyses (EFAs) with Promax rotation were conducted in Mplus using the 18
engagement items (Muthén & Muthén, 2005). Based on theory, Scree plots, and eigenvalues,
we identified several candidate psychometric models. These models were then subjected to first-
order Confirmatory Factor Analysis (CFA). We subsequently compared the fit for each model.
Factor loadings were fixed arbitrarily at one unit for each first variable. The other factor loadings
and the variance of all latent constructs were left to vary. Given the non-normal distribution of
most of the scales in our data, maximum likelihood mean-adjusted estimators for non-normal
distributions were used (Muthén & Muthén, 2005).
Once the first-order model was established, we then tested second and third-order models that
were hypothesized to represent the specific dimensions and the global construct of engagement.
Multiple group invariance of the final model was also tested for boys and girls. Using five hierar-
chically nested steps, we determined: (1) a baseline model for each group, (2) configural invari-
ance, (3) invariance of first-order factor loadings, (4) invariance of first and second-order
factor loadings, and (5) invariance of first and second-order factor loadings, measured variables,
and first-order intercepts (for technical details, see Byrne & Stewart, 2006; Chen, Sousa, & West,
2005). For model identification, we constrained the mean of the second and third-order factors
from the baseline model to zero.
Finally, we conducted two separate Structural Equation Models (SEMs). The first model aims
to verify the prospective link between student global engagement and school dropout over 2 year
period. To this end, we used the third-order construct of engagement obtained from our final
CFA. The second SEM prospectively tested the predictive links between the behavioral, affective,
and cognitive second-order components of student engagement and dropout.
As recommended by Hu and Bentler (1999), we compared different parameters in order to eval-
uate the EFA, CFA, and SEM model fit: Comparative Fit Index (CFI); TuckereLewis Index
658 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670

(TLI); Root Mean Square Error of Approximation (RMSEA); Standardized Root Square Mean
Residual (SRMR); and the c2/df Ratio. We chose to rely on the indices that are less sensitive to
sample size (TLI, RMSEA; Sharma, Mukherjee, Kumar, & Dillon, 2005). Further, because
SRMR is not available for SEM categorical outcome testing in Mplus, this model specification
index was only used to evaluate the measurement model. Although values of 0.06 or less are consid-
ered an adequate fit for SRMR and RMSEA (MacCallum, Browne, & Sugawara, 1996), values of
0.05 or less represent a more conservative choice. A value of 0.95 and above is considered an excel-
lent fit for CFI and TLI. Also, a 3.0 value or less represents the best ratio for c2/df (Bentler
& Bonett, 1980). To compare CFA models with a different number of factors, we also used Akaike
Information Criterion (AIC; Akaike, 1987), Bayesian Information Criterion (BIC: Schwartz,
1978), and Sample-Size Adjusted Bayesian Information Criterion (ABIC; Sclove, 1987). A better
fit is indicated when absolute values of these indices become smaller. However, it is preferable to
rely on the BIC and ABIC rather than the AIC because they usually select more parsimonious
models. Finally, adequacy of factor loadings and Cronbach’s alpha coefficients were examined
for both models. Although factor loadings exceeding 0.40 are considered acceptable (Hair, Ander-
son, Tatham, & Black, 1998), we chose to adopt a more conservative standard of 0.50.

Results

The multidimensionality of student engagement

Table 2 reports correlations between the 18 items included in the psychometric model, as well as
their means and standard deviations. Inter-item correlations were all positive and significant
(P < 0.001). According to the criteria of Cohen (1988), they were ranging from small (0.06) to
large in size (0.85), with most being of moderate size (mean of 0.29).
The EFA of the 18-item version identified the most plausible models based on the Scree plot. As
illustrated in Fig. 1, these models were composed of three, four, and six factors. We also chose to
select the five-factor model because of its location prior to the inflexion of the curve on the Scree
plot. The three and four-factor models were the only ones with all eigenvalues greater than 1 (for
the five and six-factors models, the final eigenvalues were 0.972 and 0.963, respectively). We relied
on this test to select our best fitting models because the results of the Scree test are more reliable
than eigenvalues with larger sample sizes (Gorsuch, 1983). Item factor loadings in all models were
quite high (greater than 0.55); however, three factor loadings did not reach 0.40 (items 11, 17, and
18). These items were thus removed.
With the remaining 15-items, we conducted separate CFAs to force them into three, four, five,
and six-factor models.1 Table 3 reports the fit indices of these four models. None of the c2/df
ratios reached the value of 3; however, considering that c2 is very sensitive to sample size, we
decided not to rely on this fit to select our best model. The BIC, ABIC, and AIC consistently

1
Distribution of items for the three (factor 1: 1, 2, 3, 4; factor 2: 5, 6, 7, 8, 9, 10; factor 3: 12, 13, 14, 15, 16), four
(factor 1: 1, 2, 3, 4; factor 2: 5, 6, 7, 8, 9, 10; factor 3: 12, 13, 14; factor 4: 15, 16), five (factor 1: 1, 2, 3, 4; factor 2:
5, 6, 7, 8; factor 3: 9, 10; factor 4: 12, 13, 14; factor 5: 15, 16), and six-factor models (factor 1: 1, 2; factor 2: 3, 4; factor
3: 5, 6, 7, 8; factor 4: 9, 10; factor 5: 12, 13, 14; factor 6: 15, 16).
Table 2
Inter-correlations and descriptive statistics of the different items measuring student engagement.
Variables 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1. Missed school without a valid
reason.
2. Skipped a class while you were at 0.40
school.
3. Disrupted the class on purpose. 0.21 0.20

I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670

4. Been rude to your teacher. 0.29 0.25 0.48
5. I like school. 0.18 0.13 0.20 0.21
6. I have fun at school. 0.15 0.09 0.14 0.17 0.72
7. What we learn in class is 0.17 0.12 0.20 0.23 0.61 0.56
interesting.
8. I enjoy what we do at school. 0.19 0.15 0.22 0.25 0.62 0.59 0.58
9. I am happy when the work is quite 0.10 0.08 0.09 0.12 0.36 0.31 0.38 0.29
challenging.
10. Often, I do not want to stop 0.11 0.07 0.13 0.13 0.37 0.32 0.38 0.30 0.41
working at the end of a class.
11. I am very happy when I learn 0.12 0.08 0.14 0.16 0.44 0.38 0.45 0.38 0.40 0.37
something new that makes sense.
12. How much time are you ready to 0.15 0.10 0.19 0.15 0.36 0.31 0.42 0.32 0.27 0.26 0.31
spend in French?
13. How much effort are you ready to 0.15 0.10 0.19 0.16 0.37 0.33 0.42 0.33 0.27 0.26 0.33 0.79
put into French?
14. How much energy are you willing 0.16 0.10 0.19 0.16 0.38 0.34 0.45 0.33 0.29 0.27 0.35 0.78 0.85
to put into French?
15. How much effort are you willing 0.14 0.10 0.16 0.14 0.33 0.29 0.33 0.31 0.27 0.21 0.33 0.43 0.44 0.43
to spend in mathematics?
16. How much time are you ready to 0.15 0.10 0.16 0.15 0.36 0.31 0.36 0.32 0.31 0.26 0.37 0.47 0.48 0.47 0.72
spend in mathematics?
17. I would like to do/learn more 0.13 0.09 0.12 0.15 0.38 0.34 0.40 0.33 0.42 0.32 0.43 0.28 0.30 0.31 0.46 0.51
than what we are actually doing/
learning in mathematics class?
18. I can easily spend a lot of time on 0.09 0.06 0.12 0.09 0.27 0.22 0.24 0.22 0.20 0.18 0.25 0.23 0.24 0.23 0.34 0.43 0.30
mathematics problem.
Mean 2.62 2.74 2.34 2.43 3.88 4.24 3.8 4.01 3.63 2.75 5.05 4.93 5.23 5.00 5.81 5.44 4.51 4.17
S.D. 0.64 0.55 0.81 0.75 1.82 1.67 1.62 1.42 1.83 1.85 1.63 1.55 1.54 1.62 1.33 1.47 1.71 1.64
All correlations are significant at P < 0.001.

659
660 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670

Eigenvalue
5

Eigenvalue 4

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Factor Number
Fig. 1. EFA Scree plot of the 18 items.

decreased as the number of factors increased; whereas, only the five and six-factor models reached
the ideal fit level of 0.95 for CFI and TLI and 0.5 for RMSEA and SRMR. Consequently, we next
compared factor loadings of these two models. They were all higher in the six-factor structure
(Table 4). More specifically, factor loadings of items 1 and 2 were higher in this model (respec-
tively, 0.57 and 0.66) than in the five-factor model (respectively, 0.38 and 0.43). Furthermore,
the item distribution in the six-factor model showed adequate internal consistency (Table 5).
For all these reasons and considering that this model corresponds to our initial theoretical struc-
ture (Table 1), we selected it as the best first-order model.
We subsequently tested a second-order model measuring the behavioral, affective, and cogni-
tive dimensions of engagement. This model was based on the six concepts previously identified.
As expected, this model resulted in a good fit solution index (CFI ¼ 0.98; TLI ¼ 0.97;
RMSEA ¼ 0.05; SRMR ¼ 0.02; c2/df ¼ 13.17) which confirmed the presence of behavioral, affec-
tive, and cognitive dimensions of engagement. Factor loadings for this second level structure were
salient (reported in second column of Table 4), as they were all above 0.70 and ranged as high as

Table 3
Fit indices for the CFA models.
Model CFI TLI RMSEA SRMR BIC ABIC AIC c2/df
1. Three-factor 0.870 0.845 0.089 0.066 429207 429045 428835 38.34
2. Four-factor 0.946 0.933 0.058 0.042 423038 422866 422644 87.78
3. Five-factor 0.967 0.957 0.047 0.032 421375 421191 420951 24.88
4. Six-factor 0.979 0.963 0.038 0.025 420406 420205 419945 16.84
CFI, Comparative Fit Index; RMSEA, Root Mean Square Error of Approximation; SRMR, Standardized Root
Square Mean Residual; BIC, Bayesian Information Criterion; ABIC, Sample-Size Adjusted Bayesian Information
Criterion; AIC, Akaike Information Criterion.
 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670 661

Table 4
Factor loadings of the first and second-order CFA models.
First-order CFA Second-order
15-item model CFA
Behavioral engagement
School attendance e during this school year, how many times have you: 0.86
1. Missed school without a valid reason. 0.57
2. Skipped a class while you were at school. 0.66
Discipline e during this school year, how many times have you: 0.85
3. Disrupted the class on purpose. 0.63
4. Been rude to your teacher. 0.77

Affective engagement
Liking school 0.71
5. I like school. 0.82
6. I have fun at school. 0.79
7. What we learn in class is interesting. 0.71
8. I enjoy what we do at school. 0.72
Interest in school work 0.79
9. I am happy when the work is quite challenging. 0.59
10. Often, I do not want to stop working at the end of a class. 0.56
11. I am very happy when I learn something new that makes sense. e
Cognitive engagement
Willingness to learn French languages arts 0.81
12. How much time are you ready to spend in French? 0.80
13. How much effort are you ready to put into French? 0.88
14. How much energy are you willing to put into French? 0.88
Willingness to learn mathematics 0.83
15. How much effort are you willing to spend in mathematics? 0.80
16. How much time are you ready to spend in mathematics? 0.86
17. I would like to do/learn more than what we are actually e
doing/learning in mathematics class?
18. I can easily spend a lot of time on mathematics problem. e
N ¼ 11,827.

0.86. Internal consistency estimates also support the adequacy of this second level structure. The
coefficients, reported in Table 5, are satisfactory for all three dimensions: behavioral (a ¼ 0.65),
affective (a ¼ 0.83), and cognitive (a ¼ 0.88).
We combined our first and second-order models together and further tested the global
construct of engagement. Overall, this last model indicated an excellent fit for all indices
(CFI ¼ 0.97; TLI ¼ 0.98; RMSEA ¼ 0.04; SRMR ¼ 0.03; c2/df ¼ 11.81). This is illustrated in
Fig. 2. Factor loadings assessing relationships between measured variables, first and second
sets of the latent concept, and the global measure of engagement were all very adequate. With
coefficients varying from 0.62 to 0.84, all factor loadings appear especially high. As expected,
the behavioral (0.62), affective (0.92), and cognitive (0.84) dimensions all seem to converge into
one major construct. Given its positively skewed distribution, the proportion of variance
662 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670

Table 5
Number of items and internal consistency of the first and second-order models.
Items Internal
consistency
Behavioral engagement 4 0.65
School attendance e during this school year, how many times have you: 2 0.60
Missed school without a valid reason.
Skipped a class while you were at school.
Discipline e during this school year, how many times have you: 2 0.66
Disrupted the class on purpose.
Been rude to your teacher.
Affective engagement 6 0.83
Liking school 4 0.86
I like school.
I have fun at school.
What we learn in class is interesting.
I enjoy what we do at school.
Interest in school work 2 0.65
I am happy when the work is quite challenging.
Often, I do not want to stop working at the end of a class.
Cognitive engagement 5 0.88
Willingness to learn French languages arts 3 0.93
How much time are you ready to spend in French?
How much effort are you ready to put into French?
How much energy are you willing to put into French?
Willingness to learn mathematics 2 0.84
How much effort are you willing to spend in mathematics?
How much time are you ready to spend in mathematics?
N ¼ 11,827.

explained by the behavioral dimension (R2 ¼ 0.32) is significantly smaller than that explained by
the affective (R2 ¼ 0.84) and cognitive (R2 ¼ 0.70) dimensions.

Invariance across gender

In terms of the baseline models, results revealed an excellent data fit for both boys (c2
(96) ¼ 8239.481; CFI ¼ 0.982; SRMR ¼ 0.024; RMSEA ¼ 0.036, with 90% CI ¼ 0.034 to 0.038)
and girls (c2 (96) ¼ 1119.371; CFI ¼ 0.981; SRMR ¼ 0.025; RMSEA ¼ 0.037, with 90%
CI ¼ 0.035 to 0.039). These parameters were all statistically significant. Table 6 reports the results
of factorial invariance testing. Only the final step indicated a small but significant variation of the
chi-square as a change of the CFI (0.019).2 This change indicates that the measured variables and

2
According to Cheung and Rensvold (2002), invariance is reached when changes on the CFI are no larger than 0.01
and when the c2 differences between the models’ fit is not significant. However, considering that the c2 is very sensitive
to sample size, we relied on the CFI change.
 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670 663

First Order Model Second Order Model Third Order Model

.56 Item 1 .66

School
.68 Item 2 attendance .70
.57
Behavioral
.60 Item 3 .63
Dimension
Discipline .81

.41 Item 4 .77

.32 Item 5
.82 .62

.78 Liking
.40 Item 6
school .84
.71
.50 Item 7
Affective .92 School
.72
Dimension Engagement
.49 Item 8

.59 .76
.65 Item 9 Interest in
school work
.70 Item 10 .56
.84

.35 Item 11 .80

Willingness
.88 .81
.23 Item 12 to learn
French Cognitive
.23 .88 Dimension
Item 13

.35 .80 Willingness .74

Item 14
to learn
mathematics
.27 Item 15 .86

Fig. 2. Final measurement structure of school behavioral, affective, and cognitive engagement.

first-order intercept were not completely invariant for boys and girls. However, considering that
all models presented excellent fit indices and that the first three steps provided evidence of invari-
ance, we can conclude that this structure operates equivalently for both genders.

School engagement and dropout

In the second stage of our investigation, we prospectively tested two separate SEM models; one
to verify the predictive links between student global engagement and dropout and the other to test
the predictive links between the three second-order engagement dimensions and the same
outcome. Theoretically significant covariates (i.e., age, course retention, maternal education)
were included in both models. As recommended by MacCallum et al. (1996), the first step
involved testing a saturated model which includes all possible paths from the covariates and
the school engagement latent factor (global or specific) to the measured outcome of school
dropout. Next, all non-significant paths were removed and the model was retested. Fig. 3
664 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670

Table 6
Tests for invariance of the school engagement structure by gender: goodness-of-fit statistics.
Model c2 df CFI RMSEA SRMR Model D*CFI Dc2 Ddf
comparison
Model 1: configural 1999.180 162 0.981 0.040 0.026 e e
invariance
Model 2: invariance of 2041.207 171 0.981 0.039 0.027 2 vs. 1 0.000 42.027* 9
first-order factor
loadings
Model 3: invariance of 2176.661 177 0.979 0.040 0.033 3 vs. 2 0.002 135.454* 2
first and second-order
factor loadings
Model 4: invariance of 2437.334 192 0.960 0.053 0.052 4 vs. 3 0.021 260.673* 15
first and second-order
loadings, measured
variables and first-
order intercept
CFI, robust CFI; RMSEA, robust root mean squared error of approximation; SRMR, standardized root mean squared
residual.
*P < 0.001.

illustrates final results of the first model predicting school dropout from the global construct of
engagement. Once again, because of the sample size, the c2/df ¼ 19.89 was far from the ratio
of 3. Furthermore, the CFI was close but did not reach the 0.95 ideal fit (CFI ¼ 0.928). However,
considering that overall, the other fit indices were adequate (TLI ¼ 0.963; RMSEA ¼ 0.040), con-
firming a good fit. As expected, school engagement predicted school dropout (b ¼ 0.15,
P < 0.001) beyond the contribution of the covariates. This model explained 12% of the variance
in school dropout.
The second SEM structure also presented a good data fit (CFI ¼ 0.934; TLI ¼ 0.972;
2
c /df ¼ 17.25; RMSEA ¼ 0.038). As expected and illustrated in Fig. 4, a decrease in behavioral
Student Maternal
Age Education

-.13*
-.13*
.15*

.06*

School -.15* School

Engagement Dropout R2= .12
.90

.09*
-.08*

Retention in
Secondary
School

Fig. 3. Structural equation model of global school engagement to predict school dropout. Large circles represent latent
factors and small circles with numbers reflect residual variances. *P < 0.001.
 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670 665

Student Maternal
Age Education

-.07*
-.17*
-.14*
Behavioral .16* -.14*
Engagement
.48*
.88
-.15*
.92 -.16*
Affective School
-.11 R2 = .12
Engagement Dropout
.41*
.91 .06
.65*

Cognitive .09*
Engagement
-.07*
-.04*

Retention in
Secondary
School

Fig. 4. Structural equation model of behavioral, affective, and cognitive engagement to predict school dropout. Large
circles represent latent factors and small circles with numbers reflect residual variances. *P < 0.001.

engagement significantly predicted school dropout (b ¼ 0.15, P < 0.001). Affective and cognitive
engagement were not directly associated with dropout, and the three dimensions of engagement
were highly covariant, especially the affective and cognitive dimensions (r ¼ 0.65, P < 0.001).
This model also explains 12% of the variance in school dropout.

Discussion

The primary purpose of this study was to develop a reliable measure of student engagement.
We then tested how this construct and its components related to eventual school dropout. Overall,
our results support the use of a multidimensional framework of engagement in predicting student
dropout. They also underscore the necessity to better understand how the individual factors
comprising the construct of student engagement relate to each other both concurrently and
over the long-term.
As expected, our first set of analyses modeled a multidimensional structure, built on six
concepts that are closely related to the different facets of engagement: school attendance and disci-
pline; liking school; interest in academic work; and willingness to learn language arts and math-
ematics. These concepts all represent important indicators of academic success (Alexander,
Entwisle, & Kabbani, 2001; Janosz et al., 1997). In subsequent steps, these concepts converged
into behavioral, affective, and cognitive components, and finally, into one global construct of
engagement. Although invariance of this structure was only partially confirmed for boys and girls,
666 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670

these variations were to be expected given that boys tend to be comparatively less engaged in
many behavioral and emotional aspects of schooling (Wigfield & Eccles, 2002).
In keeping with previous research and theory (Ensminger et al., 1996; Finn, 1989), our results
also indicated that, beyond the contribution of important individual and family risk factors, global
student disengagement was associated with eventual dropout over the short term. When broken
down into components, only the behavioral dimension predicted dropout. That is, student compli-
ance and attendance forecasted dropout better than student willingness and effort to learn the basic
curriculum and how much pleasure was associated with school-related issues. This finding is not
surprising, as impoliteness, truancy, and absenteeism are all behaviors that express some degree
of alienation from school (Battin-Pearson et al., 2000; Janosz et al., 1997; Newcomb, Abbott, &
Catalano, 2002; Rumberger, 1995). Students are often sanctioned by school staff in response to
these behaviors, and unfortunately, this contributes to further negative perceptions about investing
in school. Eventually, school disengagement becomes expressed by dropout (Finn, 1989).
We found it surprising that affective and cognitive engagement did not matter in the prediction
of eventual school dropout. Past research has suggested that all three dimensions are important in
forecasting school dropout (Janosz et al., 1997). It is plausible that the specific influence of affec-
tive and cognitive engagement on student propensity to dropout prematurely might be mediated
by behavioral engagement. Entwisle et al. (2005) observed that student disengagement can start as
early as the first grade transition or at least a number of years before the actual interruption of
schooling (Alexander et al., 2001; Entwisle et al., 2005; Rumberger, 1987; Vitaro, Brendgen, Lar-
ose, & Tremblay, 2005). Further longitudinal research is needed to examine any indirect influen-
tial role played by behavioral engagement in the hypothesized relationship between psychological
engagement and dropout.
The concept of heterotypic continuity, whereby the same underlying latent factor manifests
itself differently across development (Cicchetti & Rogosch, 2002), suggests that student disengage-
ment likely has distinct dynamics as students progress across their school experience. In its initial
stages, student disengagement might first be expressed by a psychological state; while later, as this
negative emotional and cognitive state evolves, the nature of disengagement becomes more
observable, and thus easier to reliably assess (Eccles, 2004). Decreases in school interest, motiva-
tion, and willingness to learn eventually lead to school alienation and misbehaviors. In other
words, student effect and cognitions regarding school and learning-related variables influence
behaviors, eventually culminating in the decision to dropout. Within the realm of this interpreta-
tion, our results are consistent with the idea that behavioral manifestations of student disengage-
ment process are more proximal to dropout and are likely a consequence of affective and cognitive
disengagement. Unfortunately, the concurrent nature of our engagement measures precludes
a prospective approach to the differential influence of student affect, cognition, and behavior in
the prediction of school dropout. A longer-term approach in data collection is thus needed to
establish how these variables unfold and foretell dropout at a later age.
It is likely that the behavioral component accounted for the association between the inclusive
engagement variable and dropout. This component is also easier to operationalize compared to
the affective and cognitive components. Nevertheless, considering engagement as a composite
variable might enable researchers to account for putative influential mechanisms that exist
between its components. Even if our findings highlight the unique direct contribution of behav-
ioral engagement in the prediction of student dropout, there remains a need to better understand
 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670 667

the longitudinal mechanisms underlying this outcome. As demonstrated previously (Alexander

et al., 2001; Janosz et al., 1997), student effect and cognition act as important indicators of student
alienation and this contribution, even if indirect, should not be neglected in our efforts to under-
stand the problem. Yet, in order to do so, it might be more beneficial to treat engagement as
a multidimensional experience. Attention to the developmentally specific roles of each dimension
and their potential interactions in the school dropout prediction equation merits some attention.
Some limitations of this study bear mentioning. First, some of the children in our sample were
not yet of legal age for dropout (age 16) when the dependent variable was measured. As such, the
prevalence of this outcome in our sample was small and represents only the early withdrawers.
The fact that our model explained a modest effect size with such a particular sample remains inter-
esting. It leads us to believe that with a longer follow-up, the prediction might be stronger. One
further limitation of this study relates to the measure of student engagement itself. Considering
that various skills, behaviors, and attitudes are shaped at different life stages (Cunha, Heckman,
Lochner, & Masterov, 2006), the dimensions of engagement might be dependent upon student
age. As recounted by Cicchetti and Rogosch (2002), heterotypic continuity and change across
development is expected. As we consider developmental constraints, it is likely that the way
behavioral engagement is defined in this study is not representative of engagement during elemen-
tary school. Finally, there are compromises made with secondary analysis of existing data.
Although our three dimensions of engagement are supported by theory, their operationalization
is constrained by the data. For example, our cognitive dimension only reflected student psycho-
logical investment in learning, when, in principle, its conceptualization is also defined in terms of
metacognition and self-regulation strategies (Fredricks et al., 2004). Consequently, it is possible
that these concepts could be more associated with dropout than psychological investment in
learning. Independently of these limitations, a definition of engagement based on few items
remains advantageous as a cost-effective risk indicator of school dropout.
In summary, this study demonstrated that, beyond its theoretical contribution, the multidimen-
sional construct of student engagement is indeed associated with school dropout. Student engage-
ment can be defined by three specific dimensions: behavioral, affective, and cognitive. Globally,
these dimensions somehow stand for the complexity of student experience. However, only the
behavioral dimension seems to contribute to the estimation of school dropout. In order to under-
stand the more indirect mechanisms underlying this outcome, future research should address the
specific interactions between the dimensions of engagement over a longer follow-up period. Such
efforts will contribute to meaningful knowledge that will enable tailored and differentiated preven-
tion strategies that reduce dropout. The identification of students presenting greater behavioral,
affective, and cognitive risks will help researchers to develop or improve realistic, cost-effective,
and context-based prevention and intervention strategies one step further toward fostering
successful development.

Acknowledgments

This research was supported by a doctoral grant from the Programme Persévérance et Réussite Sco-
laire du Fond Québécois pour la Recherche sur la Société et la Culture. This research was approved by
the ethics board of the Université de Montréal and informed consents were obtained for all subjects.
668 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670

References

Ablard, K. E., & Lipschultz, R. E. (1998). Self-regulated learning in high-achieving students: Relations to advanced
reasoning, achievement goals, and gender. Journal of Educational Psychology, 90(1), 94e101.
Akaike, H. (1987). Factor analysis and AIC. Psychometrika, 52, 317e332.
Alexander, K. L., Entwisle, D. R., & Kabbani, N. S. (2001). The dropout process in life course perspective: early risk
factors at home and school. Teachers College Record, 103(5), 760e822.
Appleton, J. J., Christenson, S. L., Kim, D., & Reschly, A. L. (2006). Measuring cognitive and psychological engage-
ment: validation of the Student Engagement Instrument. Journal of School Psychology, 44(5), 427e445.
Battin-Pearson, S., Newcomb, M. D., Abbott, R. D., Hill, K. G., Catalano, R. F., & Hawkins, J. (2000). Predictors of
early high school dropout: a test of five theories. Journal of Educational Psychology, 92(3), 568e582.
Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness-of-fit in the analysis of covariance structures.
Psychological Bulletin, 88, 588e600.
Bouffard, T., & Couture, N. (2003). Motivational profile and academic achievement among students enrolled in
different schooling tracks. Educational Studies, 29(1), 19e38.
Byrne, B. M., & Stewart, S. M. (2006). The MACS approach to testing for multigroup invariance of a second-order
structure: a walk through the process. Structural Equation Modeling, 13(2), 287e321.
Caspi, A., Entner, W., Moffitt, T. E., & Silva, P. (1998). Early failure in the labor market: childhood and adolescent
predictors of unemployment in the transition to adulthood. American Sociological Review, 63, 424e451.
Chen, F. F., Sousa, K. H., & West, S. G. (2005). Testing measurement invariance of second-order factor models. Struc-
tural Equation Modeling, 12(3), 471e492.
Cheung, G. W., & Rensvold, R. B. (2002). Evaluating goodness-of-fit indexes for testing measurement invariance.
Structural Equation Modeling, 9(2), 233e255.
Christenson, S. L., & Thurlow, M. L. (2004). School dropouts: prevention considerations, interventions, and chal-
lenges. Current Directions in Psychological Science, 13(1), 36e39.
Cicchetti, D., & Rogosch, F. A. (2002). A developmental psychopathology perspective on adolescence. Journal of
Consulting and Clinical Psychology, 70(1), 6e20.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum.
Costenbader, V., & Markson, S. (1998). School suspension: a study with secondary school students. Journal of School
Psychology, 36(1), 59e82.
Cunha, F., Heckman, J. J., Lochner, L. J., & Masterov, D. V. (2006). Interpreting the evidence on life cycle skill forma-
tion. In E. A. Hanushek, & F. Welch (Eds.), Handbook of the economics of education (pp. 697e812). Amsterdam:
Forthcoming.
DeBacker, T. K., & Nelson, R. (2000). Motivation to learn science: differences related to gender, class type, and ability.
Journal of Educational Research, 93(4), 245e254.
Eccles, J., Wigfield, A., Harold, R. D., & Blumenfeld, P. (1993). Age and gender differences in children’s self- and task
perceptions during elementary school. Child Development, 64(3), 830e847.
Eccles, J. S. (2004). Schools, academic motivation, and stage-environment fit. In R. M. Lerner, & L. Steinberg (Eds.),
Handbook of adolescent psychology (2nd ed.). (pp. 125e153) Hoboken, NJ: John Wiley & Sons.
Eccles, J. S., Barber, B. L., Stone, M., & Hunt, J. (2003). Extracurricular activities and adolescent development. Journal
of Social Issues, 59(4), 865e889.
Eggert, L. L., Thompson, E. A., Herting, J. R., Nicholas, L. J., & Dickers, B. G. (1994). Preventing adolescent drug
abuse and high school dropout through an intensive school-based social network development program. American
Journal of Health Promotion, 8(3), 202e215.
Elder, G. H. (1995). The life course paradigm: social change and individual development. In P. Moen, G. H. Elder,
& K. Lüscher (Eds.), Examining lives in context: Perspectives on the ecology of human development (pp. 101e139).
Washington, DC: APA Press.
Ensminger, M. E., Lamkin, R. P., & Jacobson, N. (1996). School leaving: a longitudinal perspective including neigh-
bourhood effects. Child Development, 67, 2400e2416.
Entwisle, D. R., Alexander, K. L., & Olson, L. S. (2005). First grade and educational attainment by age 22: a new story.
American Journal of Sociology, 110, 1458e1502.
 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670 669

Finn, J. D. (1989). Withdrawing from school. Review of Educational Research, 59(2), 117e142.
Fredricks, J. A., Blumenfeld, P. C., & Paris, A. H. (2004). School engagement: potential of the concept, state of the
evidence. Review of Educational Research, 74(1), 59e109.
Goodenow, C. (1993). Classroom belonging among early adolescent students: relationships to motivation and achieve-
ment. Journal of Early Adolescence, 13(1), 21e43.
Gorsuch, R. L. (1983). Factor analysis. Hillsdale, NJ: Lawrence Erlbaum.
Gottfredson, D. C., Fink, C., & Graham, N. (1994). Grade retention and problem behavior. American Educational
Research Journal, 31(4), 761e784.
Hair, J. F., Anderson, R. E., Tatham, R. L., & Black, W. C. (1998). Multivariate data analysis (5th ed.). Upper Saddle
River, NJ: Prentice Hall.
Hirshi, T. (1969). Causes of delinquency. Berkley, CA: University of California Press.
Heckman, J. J. (2006). Skill formation and the economics of investing in disadvantaged children. Science, 312, 1900e1902.
Hu, L., & Bentler, P. M. (1999). Cut-off criteria for fit indices in covariance structure analysis: conventional criteria
versus new alternatives. Structural Equation Modeling, 6, 1e55.
Janosz, M., LeBlanc, M., Boulerice, B., & Tremblay, R. E. (1997). Disentangling the weight of school dropout predic-
tors: a test on two longitudinal samples. Journal of Youth and Adolescence, 26(6), 733e762.
Janosz, M., LeBlanc, M., Boulerice, B., & Tremblay, R. E. (2000). Predicting types of school dropouts: A typological
approach with two longitudinal samples. Journal of Educational Psychology, 92, 171e190.
Jenkins, P. H. (1997). School delinquency and the school social bond. Journal of Research in Crime and Delinquency,
34(3), 337e367.
Jimerson, S. R., Campos, E., & Greif, J. L. (2003). Towards an understanding of definitions and measures of school
engagement and related terms. The California School Psychologist, 8, 7e28.
Libbey, H. P. (2004). Measuring student relationship to school: attachment, bonding, connectedness, and engagement.
Journal of School Health, 74(7), 274e283.
Mahoney, J. L., & Cairns, R. B. (1997). Do extracurricular activities protect against early school dropout? Develop-
mental Psychology, 33(2), 241e253.
MacCallum, R. C., Browne, M. W., & Sugawara, H. M. (1996). Power analysis and determination of sample size for
covariance structure modeling. Psychological Methods, 1, 130e149.
Muthén, L. K., & Muthén, B. O. (2005). Mplus user’s guide. Los Angeles, CA: Muthén & Muthén.
Newcomb, M. D., Abbott, R. D., & Catalano, R. F. (2002). Mediational and deviance theories of late high school
failure: process roles of structural strains, academic competence, and general versus specific problem behavior. Jour-
nal of Counseling Psychology, 49(2), 172e186.
OECD. (2003). Education at a glance: OECD indicators 2003. Paris: OECD. http://books.google.ca/book-
s?vid¼ISBN9264102337&id¼Ndsqu2gRWGwC&dq¼Educationþatþaþglance:þOECDþindicatorsþ2003.
Accessed 12.11.05.
Pagani, L., Boulerice, B., Vitaro, F., & Tremblay, R. E. (1999). Effects of poverty on academic failure and delinquency
in boys: a change process model approach. Journal of Child Psychology and Psychiatry, 40(8), 1209e1219.
Posner, J. K., & Vandell, D. L. (1999). After-school activities and the development of low-income urban children:
a longitudinal study. Developmental Psychology, 35(3), 868e879.
Rumberger, R. W. (1987). High school dropouts: a review of issues and evidence. Review of Educational Research, 57,
101e121.
Rumberger, R. W. (1995). Dropping out of middle school: a multilevel analysis of students and schools. American
Educational Research Journal, 32(3), 583e625.
Rumberger, R. W., & Lamb, S. P. (2003). The early employment and further education experiences of high school drop-
outs: a comparative study of the United States and Australia. Economics of Education Review, 22, 353e366.
Rumberger, R. W., & Larson, K. A. (1998). Student mobility and the increased risk of high school dropout. American
Journal of Education, 107, 1e35.
Schwartz, G. (1978). Estimating the dimension of a model. Annals of Statistics, 6, 461e464.
Sclove, L. S. (1987). Application of model-selection criteria to some problems in multivariate analysis. Psychometrika,
52, 333e343.
Sharma, S., Mukherjee, S., Kumar, A., & Dillon, W. R. (2005). A simulation study to investigate the use of cutoff
values for assessing model fit in covariance structure models. Journal of Business Research, 58, 935e943.
670 I. Archambault et al. / Journal of Adolescence 32 (2009) 651e670

Tinto, V. (1975). Dropout from higher education: a theoretical synthesis of recent research. Review of Educational
Research, 45(1), 89e125.
Tymms, P. B., & Fitz-Gibbon, C. T. (1992). The relationship of homework to A-level results. Educational Research,
34(1), 3e10.
Vitaro, F., Brendgen, M., Larose, S., & Tremblay, R. E. (2005). Kindergarten disruptive behaviors, protective factors,
and educational achievement by early adulthood. Journal of Educational Psychology, 97(4), 617e629.
Watt, H. M. G. (2004). Development of adolescents’ self-perceptions, values, and task perceptions according to gender
and domains in the 7th-trought 11th-grade Australian students. Child Development, 75(5), 1556e1574.
Wehlage, G. G., Rutter, R. A., Smith, G. A., Lesko, N., & Fernandez, R. R. (1989). Reducing the risk: Schools as
communities of support. London: The Falmer Press.
Wigfield, A., & Eccles, J. S. (2002). The development of competence beliefs, expectancies for success, and achievement
values from childhood through adolescence. In A.Wigfield., & J. S. Eccles (Eds.), Development of achievement moti-
vation (pp. 173e195). San Diego, CA: Academic Press.