British Educational Research Journal

Emotional disengagement from schoolwork

This study is concerned with a particular aspect of emotional (dis)engagement: students’ emotional experiences of doing schoolwork. Schoolwork‐related emotions and their representative attitudes are distinct from other types of emotional engagement in school, as discussed earlier. Schoolwork emotions can be conceptualised as being nested alongside feelings about peers and teachers within classrooms. In turn, those classroom‐oriented feelings are nested alongside emotional engagement with other aspects of school experience such as extracurricular activities and student councils, in the larger school context (Skinner & Pitzer, 2012).

Although we take students’ emotional attitudes to represent their emotional experiences, there is a degree of bidirectional influence between the two factors. Hidi and Renninger's (2006) Four‐Phase Model of Interest Development outlines how trait emotions (felt in the moment, such as situational boredom), and state emotions (tendency for an emotion, such as dispositional boredom) inform each other developmentally through time. They propose that state emotions are triggered by an activity, then are maintained to some extent throughout it. Over time, individuals develop emerging trait emotions based on their emotional states. Then, those traits become well‐established. Over time, trait emotions can drive further state emotions (Linnenbrink‐Garcia et al., 2010) which again become internalised within a person's self‐schema. Accordingly, it is critical that researchers use carefully worded items in order to capture the subtleties of these different phenomena, as we demonstrate in the following example statements:

• I had an enjoyable learning experience in my lesson (triggered state emotion)
• My lesson was enjoyable (maintained state emotion)
• I tend to enjoy my lessons (emergent emotional trait)
• Lessons are enjoyable (well‐established emotional trait)

However, as our final example shows, although lessons are the attitudinal object it is actually the individual's tendency that is being expressed. Therefore, researchers should be wary of evaluating the school environment with emotional attitudinal statements as these can reflect trait emotions (Symonds, 2014).

In this study, we focus on two primary indicators of emotional disengagement from schoolwork: boredom and interest. Although students have many different schoolwork emotions, including anxiety, frustration, pride, joy and disappointment (Pekrun & Linnenbrink‐Garcia, 2012), boredom and interest are fundamental to the concept of engagement as deep and sustained involvement in an activity (Csikszentmihalyi, 1990; Schneider et al., 2016). As affective states, they mirror each other in many ways; with boredom often being accompanied by feelings of emotional pain and displeasure, and interest typically coinciding with enjoyment and satisfaction (Pekrun et al., 2002). When people are intensely interested in what they are doing, they can experience cognitive absorption (Salmela‐Aro & Upadyaya, 2013) and flow (Csikszentmihalyi, 1990). In contrast, boredom can signal emotional disaffection (Skinner & Pitzer, 2012).

Boredom and interest, therefore, can be perceived as opposite ends of a certain type of measurable engagement continuum. This continuum should have good internal validity, as boredom and interest associate with opposite motivational variables (e.g., lower versus higher activity value) and outcomes (e.g., lower versus higher mastery motivation), as a range of studies demonstrate (Pekrun et al., 2010; Linnenbrink‐Garcia et al. 2013; O'Keefe & Linnenbrink‐Garcia, 2014). Likewise, in line with Heckhausen et al.'s (2010) A Motivational Theory of Life‐Span Development, boredom and interest predict opposite types of engaged behaviour, with boredom driving people to disengage from an activity, and interest motivating people to engage with it.

Emotional disengagement, motivational theory and students’ developmental outcomes

Emotional attitudes are part of a larger self‐system of motivation and engagement (Skinner & Pitzer, 2012). Therein, emotional attitudes have mediated the connections between behavioural engagement and motivational resources such as goals and self‐concept (Green et al., 2012). Using Heckhausen et al.'s (2010) perspective, this mediation can be protective, with students using emotional attitudes to dismiss or reinforce feedback between behaviour and motivation. For example, disliking mathematics can support students’ decisions to attribute their failures to pedagogy and curriculum, rather than to personal ability, which would impact their mathematics’ self‐concepts. Likewise, being emotionally disengaged from an activity may justify students’ decisions to value it less, with the end goal of avoiding engaged behaviour. Accordingly, lower emotional engagement has predicted lower effort in learning in Finland (Pietarinen et al., 2014), and greater off‐task behaviour, truancy and absenteeism in the US (Li & Lerner, 2013).

In turn, disengaging behaviourally from schoolwork can have feed‐forward effects on emotional disengagement (Skinner et al., 2008) as students limit their opportunities to obtain environmental rewards and supports available for that activity. For example, trying less hard may result in more negative feedback from teachers, thus creating negative state emotions that develop into negative trait emotions over time. These types of feedback loops can sustain students’ emotional disengagement from schoolwork, and increase it when students’ negative thoughts and actions, and a lack of environmental support, amplify each other over time (Skinner et al., 2008).

This dynamic system of disengagement from schoolwork and schooling has been associated with a range of emotional, personal and social factors. In the US, spending less time at school has related to elevated substance use which is often a key feature of adolescent peer socialisation (Henry et al., 2012). Also in the US, emotionally disengaged students have reported poorer mental health and more negative growth in mental health across schooling, than engaged students (Wang & Peck, 2013), perhaps due to their alienation from important resources for mental health in the school context (Salmela‐Aro & Upadyaya, 2013). In Finland, emotional disengagement from schoolwork has been associated with lowered educational aspirations (Tuominen‐Soini & Salmela‐Aro, 2014) perhaps as feeling little emotional connection to schoolwork does not inspire further study goals. Emotional disengagement has also associated with lower achievement outcomes in one US sample (Li & Lerner, 2011) although this link was not present in another sample, when modelled with cognitive and behavioural engagement (Wang & Eccles, 2012).

Disengagement trajectories and their characteristics

Several person‐oriented studies have identified different patterns of emotional (dis)engagement growth, although none have focused solely on emotional attitudes towards schoolwork as we have defined them. In latent‐profile models, students have remained in disengaged profiles, or moved in and out of those, across time in England (Ross, 2009) and Finland (Tuominen‐Soini & Salmela‐Aro, 2014). In growth‐mixture models, students have been classified into multiple disengagement trajectories, where they disengaged at different rates from different baselines in the US (Li & Lerner, 2011), and had stable disengagement, disengaged then rebounded, or disengaged continually, in Canada (Janosz et al., 2008). Together, these studies indicate the presence of multiple disengagement trajectories in secondary schooling.

Those trajectories have been marked by individual differences in ethnicity, socio‐economic status (SES), gender and achievement. In the US, more African‐American students were classified into trajectories of emotional, cognitive and behavioural disengagement from schooling that began from a lower level of engagement (Li & Lerner, 2011), whereas in England, more students of South‐Asian, Black‐African and Black‐Caribbean descent fitted into profiles of emotional disengagement with schooling (Ross, 2009), in line with the high aspirations for educational attainment typically transmitted to them by their first generation immigrant parents (Strand, 2007).

The findings for SES in these studies have been reasonably consistent, with lower SES predicting greater and increasing emotional disengagement in the US (Li & Lerner, 2011) and the UK (Ross, 2009) where around 2.6 million children lived in low‐income households during our study period (HM Treasury, 2004). Studies from both countries have documented that children of parents with lower levels of formal education and fewer economic resources, tend to have lower achievement and educational aspirations (Schoon 2010, 2014; Ermisch et al., 2012), although there is less research on SES and emotional attitudes to schoolwork as we have defined them here. In another study of the LSYPE data, Chowdry et al. (2011) found a weak positive association between SES and students’ enjoyment and valuing of schooling (2‐items) and stronger relationships between SES and students’ educational aspirations, however they did not test the associations between SES and students’ interest and boredom in lessons.

In the US and England, being male has predicted greater disengagement in secondary schooling (Ross, 2009; Li & Lerner, 2011), whereas in Finland, more males have fitted into a profile of high cynicism towards school (Tuominen‐Soini & Salmela‐Aro, 2014). These gender differences in emotional engagement with schooling have been attributed to many factors including male students’ desire to obtain popularity through publicly dismissing schoolwork and their lack of enjoyment of subjects without a practical component (Symonds et al., 2014).

Regarding achievement, students in Finland who were cynical and exhausted also had lower achievement, however there was one profile of engaged‐exhausted students who were higher achievers (Tuominen‐Soini & Salmela‐Aro, 2014). Similarly, in Canada, trajectories of engaging and disengaging have both been predicted by lower initial achievement (Janosz et al., 2008); demonstrating that lower achievement does not uniformly predict disengagement.

The present study

Drawing on the evidence reviewed earlier, we made the following seven assumptions:

1. There will be multiple trajectories of schoolwork emotional disengagement and engagement in the LSYPE.
2. The disengagement trajectories will be characterised by lower SES, being white and male, with those individual differences being more prevalent in trajectories of greater disengagement.
3. After accounting for those individual differences, the disengagement trajectories will not necessarily associate with prior achievement, as for example, higher achievers can become burned out and lower achievers can retain emotional engagement with schoolwork.
4. However, the disengagement trajectories will associate with lower achievement by the end of school, as students’ achievement wanes in response to the broader dynamic system of emotional and behavioural disengagement.
5. The disengagement trajectories will also associate with greater behavioural disengagement, poorer psychological wellbeing and more risk behaviour at school, as school avoidance tactics are played out and environmental supports at school decrease in a continuous feedback loop.
6. After comprehensive secondary school, the disengagement trajectories will associate with reduced participation in upper secondary, further and higher education, and more participation in employment, as students strive to avoid further experiences of activity boredom and disinterest generated in a highly structured educational environment.
7. After school completion, the disengagement trajectories will no longer associate with poorer psychological wellbeing and greater risk behaviour, as students are released from the negative person–environment interactions they experienced at school, and enter a more self‐directed and positive phase of development.

To test these assumptions, we conducted a person‐oriented analysis of the development of emotional disengagement from schoolwork in the LSYPE. We compared students in the main disengagement trajectories to their counterparts who remained stably engaged or disengaged across compulsory secondary schooling, to ascertain whether there were between‐group differences in gender, ethnicity, SES, achievement, psychological wellbeing, risk behavior, and career pathways. The analyses were completed for four different time periods: (1) early adolescence at the start of the study, (2) mid adolescence at the end of compulsory secondary school, (3) late adolescence after the transition out of compulsory secondary schooling, and (4) young adulthood.

Participants and procedures

Data from Waves 1 to 7 of the nationally representative LSYPE,¹ were used in the current research. In the LSYPE, participants were surveyed annually from age 13–14 years (in 2004) to 20–21 years (in 2011). Schools were used as the primary sampling unit with 646 maintained and 28 independent schools involved. The study aimed to achieve target numbers of at least 1,000 students in each of five major minority ethnic groups (Indian, Pakistani, Bangladeshi, Black African, Black Caribbean and Mixed) to facilitate research on ethnic minorities, and to oversample deprived schools also for analytic purposes. Accordingly, state‐funded schools were stratified by deprivation status based on free school meals data. From within each deprivation stratum, schools were selected according to the number of Year 9 ethnic minority students enrolled. Schools in the top quintile of the deprivation distribution were oversampled by 50%. Students attending schools in the maintained sector were sampled using data from the Pupil Level Annual School Census (PLASC), according to their gender, ethnicity and probability of being in a sampled school. In the independent sector, schools and students were sampled based on measures of academic performance and gender.

Weights used in the study were calculated by the LSYPE administration, who used logistic regression models to calibrate the sample to population totals for ethnicity, gender and region drawn from the National Pupil Database at Wave 1. At each subsequent wave, a non‐response weight was calculated using either regression models or cell weighting, and was based on design weights from previous waves where applicable, and on a range of data drawn from that wave. More detailed information on weighting can be found in the LSYPE user guide online.²

Study participants were in comprehensive or independent schools between 14 and 16 years of age, before transferring to education, employment or training or following alternative pathways at age 17–18. Of the 16,122 participants involved across the seven waves, only those with emotional disengagement data from 2 or more time points during the 3 years of compulsory secondary school were included in this study (total N = 13,734). This selection procedure guaranteed that no cases would be arbitrarily fitted using Growth Mixture Modelling (GMM). As described in our analysis and results sections, we used this technique to uncover multiple trajectories of disengagement that could overlap at different time points as they descended, ascended or remained stable. Therefore, cases needed at least 2 time points of data in order to classify them accurately into a specific growth pattern.

Compared to those in the analytic sample, participants with missing data were equally balanced in gender, but had slightly lower SES (t = 5.978, df = 2,721, p = .000, d = .14), childhood achievement (t = 9.658, df = 2,608, p = .000, d = .23) and were more likely to be white (t = 7.043, df = 3,262, p = .000, d = .16). Calculation of Cohen's d revealed that those effects were negligible or small. The present study used data from Waves 1, 2 and 3 to represent compulsory secondary school, Wave 4 to represent the school transition period, and Waves 6 and 7 to represent young adulthood.

Measures

Disengagement trajectories in compulsory secondary school

Growth mixture modelling (GMM) in Mplus version 7.3³ was used to determine the most common (dis)engagement trajectories in secondary school. GMM assumes that distinct trajectories exist within a dataset, and seeks to classify participants into trajectories that have good internal homogeneity, for the number of trajectories that the researcher specifies (Muthén & Muthén, 2000). We ran unconditional models using full‐information maximum likelihood estimation, which uses all available data (Little & Rubin, 2014) and assumes that data are missing at random or completely at random (Schafer & Graham, 2002). Multiple starting values were used to avoid localised solutions.

Different models, containing an increasing number of trajectories, were evaluated for their best fit to the data, and compared to each other. We evaluated the classification accuracy of each model overall (entropy of .7 or above) and for each class (average posterior probabilities of group membership at .7 or above). Successive solutions were deemed to fit better than the former when they returned a lower statistic for the Akaike information criterion (AIC), Bayesian information criterion (BIC) and sample size adjusted BIC; and significant values for the Vuong‐Lo‐Mendell‐Rubin (VLMR) and Lo‐Mendell‐Rubin (LMR) likelihood ratio tests (Nagin & Odgers, 2010).

Of the 10 models evaluated, the model containing 8 trajectories emerged as the best fit (Table 1), with the highest entropy (0.89), a notably lowered BIC (44,164) and significant VLMR and LMR p values (both at 0.000). We named each trajectory according to the steepness of its slope (stable, mild, rapidly, Figure 1), and as to whether its intercept was above or below the middle score in the disengagement scale (disengaged versus engaged). The largest trajectory had relatively stable, moderate levels of engagement (stable engaged, N = 5,840, 42.5%; Table 2). There, its slope reached significance probably only because of the large number of cases. The next three largest trajectories had similar, moderate levels of engagement at baseline but then disengaged gradually (disengaging mildly, N = 2,558, 18.6%) or rapidly (disengaging rapidly, N = 1,306, 9.5%), or had a rapid increase in engagement (engaging rapidly 1, N = 1,492, 10.9%). The two next largest trajectories were disengaged at baseline, and either remained stable (stable disengaged, N = 868, 6.3%) or became more disengaged (disengaging disengaged, N = 898, 6.5%). Finally, two smaller trajectories increased in engagement from an engaged baseline (engaging rapidly 2, N = 485, 3.5%) or disengaged baseline (engaging rapidly 3, N = 287, 2.1%). These findings fitted our first assumption, that there would be different trajectories of (dis)engagement. Across trajectories, the broader pattern was of being stable engaged or engaging (59%), or stable disengaged or disengaging (41%).

Disengaging/disengaged students in early adolescence

In order to study the effects of the different disengagement trajectories, we allocated each with a comparison trajectory (see Table 7). Specifically, we compared the disengaging mildly, disengaging rapidly and disengaging disengaged trajectories to their stable counterparts with the closest disengagement intercept, to identify the effects of differently positioned disengagement slopes. Then, we paired the stable disengaged trajectory with the stable engaged trajectory to identify the effects of stable disengagement.

Using stepwise multiple linear regression, we tested the effects of individual differences and age 11 achievement on trajectory membership. For the first 3 comparison groups, we controlled for Wave 1 disengagement in the first step to control for slight differences in intercept and reveal the effect of slope, then entered the individual differences and achievement in the second step. We removed this control when comparing the two stable groups, in order to focus on level of disengagement. The regressions tested assumption b that the trajectories would be predicted by being male, having lower SES and being white, and assumption c that there would be no difference in age 11 achievement once controlling for individual differences. Hereon we use Cohen's (1992) terminology for reporting the effect sizes of mean differences (small = .20, medium = .50, large = .80), which are indicated by the Beta weights in the multiple regression models.

Disengaging/disengaged students in mid adolescence

As observed in the regressions, the disengaging trajectories had slightly different Wave 1 disengagement than their stable counterparts. This raised the possibility that any between‐group differences observed in development might be accountable to both this different initial level of disengagement, and the subsequent disengagement process. Accordingly, we removed this confounding effect by using propensity score matching (PSM) to create similar groups of students who then followed different disengagement trajectories. We did this by balancing each pair of trajectories on Wave 1 disengagement, background factors, and all other longitudinal variables with Wave 1 data (gender, SES, white ethnicity, age 11 achievement, effort, truancy, university aspirations, parental relationship, bullying, smoking, drinking and drug use). Only the stable disengaged versus stable engaged trajectories were not balanced on Wave 1 disengagement, to free the effect of disengagement level.

The PSM R plugin for SPSS developed by Thoemmes (2012) was used for the matching. First, we replaced the Wave 1 missing values with average scores computed from five multiple imputed datasets created in SPSS version 23.0. Then, we created new data files for each pair of trajectories. Within each file we created a trajectory variable of (1 = disengagement, 0 = comparison). Then, we regressed the matching variables on the trajectory variable to create propensity scores that indicated which comparison case was closest to each disengagement case (nearest neighbour matching). Strict protocols were employed to produce a precise match (one to one matching strategy, calliper of 0.2, and discarding cases outside the common area of support). The final sub‐samples for each pair of trajectories are displayed in Table 7.

The success of each PSM was evaluated as follows. We observed a reduction in between‐trajectory variance indicated by the relative multivariate imbalance statistic. Then we imported the matched trajectory variable into the main dataset, and examined the between‐group differences in the matching variables using the original, non‐imputed data. Levene's tests and independent samples t‐tests confirmed that the matched trajectories were equal in their distribution and mean values for the matching variables, with the following exceptions: cannabis use was higher for the disengaging rapidly trajectory (M = .11, SD = .31) versus the stable engaged trajectory (M = .07, SD = .25; B = .507, SE = .142, Wald = 12.65, df = 1, p = .000, d = .82), and smoking was negligibly higher for the disengaging mildly trajectory (M = 1.29, SD = 1.03) versus the stable engaged trajectory (M = 1.23, SD = .94; B = .061, SD = .029, t = 4.258, p = .039, d = .06).

Using linear and logistic regressions, we then examined the trajectory pairs for differences in the Wave 3 variables, in order to test assumption d, that disengaging students would have lower later achievement to other trajectories, and assumption e, that they would develop greater behavioural disengagement, poorer psychological wellbeing and more risk behaviours at the end of compulsory secondary school.

Disengaging/disengaged students in late adolescence

Next, we used linear and logistic regressions to test for differences between the disengaged students and their counterparts, after finishing compulsory comprehensive school (Wave 4). There we assumed (f) that disengaging students would participate less in upper secondary and further education and more in employment; but (g) that they would have similar psychological wellbeing and risk behaviours.

Disengaging/disengaged students in young adulthood

Our final set of regressions tested the same assumptions (f and g) in young adulthood. However, anxiety and subjective ill‐being were not measured in Waves 6 and 7, therefore we used life‐satisfaction as a complementary measurement of psychological wellbeing.

Multiple disengagement trajectories

We identified eight different disengagement and engagement trajectories in the LSYPE data, extending the latent transition analysis of the LSYPE by Ross (2009) by demonstrating slopes of (dis)engagement in school. The first two disengagement trajectories were emotionally engaged with schoolwork at age 14–15 years but then disengaged at different rates. This experience of initially being engaged but then losing interest in schoolwork is widespread in the school transition literature. Immediately following transfer to secondary school, students often report a ‘honeymoon’ period (Hargreaves, 1984) where they view their experiences through rose tinted glasses. However, this initial excitement can quickly wear off, as students realise that not all aspects of school environment are a good fit with their personal and developmental needs, as outlined by stage‐environment fit theory (Eccles et al., 1993; Symonds & Galton, 2014).

Second, we identified a group of students who remained at a stable level of disengagement during the study period. As disengagement was first measured at age 14 years, rather than at 12 years immediately after transfer to secondary school, it is impossible to tell whether these students had been engaged beforehand. However, the consistency of their disengagement drew our attention to exploring which mechanisms might have kept them at that stable level, by comparing them to the disengaging disengaged trajectory.

A final group of students were also disengaged at the start of the study, but then had rapidly increasing disengagement across secondary school. Possibly, whatever personal and social mechanisms were influencing their emotional disengagement from schoolwork became stronger, as their disengagement and these factors amplified each other through time (Skinner et al., 2008). Interestingly, this trajectory of disengaging from a disengaged baseline has not been observed in other large scale studies (Janosz et al., 2008; Li & Lerner, 2011), therefore does not necessarily relate to the greater heterogeneity of larger samples.

Individual characteristics of disengagement

Our next assumption (b) was that the trajectories would be marked by known risk factors for disengagement from schooling and schoolwork, including being male, having greater social disadvantage and being white. However, in contrast to studies from the US (Li & Lerner, 2011) and Canada (Janosz et al., 2008), and to Ross’ (2009) analysis of the LSYPE data, we observed no gender differences between trajectories, and no SES difference in three of the four disengagement trajectory pairs, after controlling for baseline disengagement. Further investigation using SES as an ordinal variable revealed that SES did not correlate with our measure of emotional disengagement at any wave. Also, the full range of SES categories were fairly normally distributed within each trajectory. Therefore, the lack of SES differences in our study was not accountable to our multiple regression method where gender and ethnicity were entered before SES in the models.

Our finding of a lack of relationship between individual background factors and emotional attitudes towards school is comparable to Gorard and Huat See's (2010) cross‐sectional study of around 3,000 students in 45 secondary schools across England. There, background factors including gender, SES and eligibility for free school meals accounted for only 4% of the variation in students’ perceptions of how interesting their lessons were, and 10% of the variation in how much they enjoyed school. As these authors commented, this finding is in stark contrast to the positive associations generally found between SES and other aspects of engagement such as educational aspirations (Chowdry et al., 2011) and the amount of time not being in education, employment or training (NEET) (Schoon, 2014).

So why do our findings contrast with the prior longitudinal studies of emotional engagement? This may be explained by our narrower measurement of boredom and interest. Ross (2009) combined items on students’ attitudes towards school, school work and lessons (a 10‐item scale) with a measurement of their truancy and school leaving aspirations. In England, students’ expectations for attending higher education are associated with their social class (Chowdry et al., 2011; Parker et al., 2016) therefore possibly explaining why SES differences were observed in Ross’ (2009) study and not in ours. Also, Janosz et al. (2008) used a combined measure of emotional, cognitive and behavioural engagement, and Li and Lerner (2011) measured emotional engagement as students’ emotional attitudes towards teachers and peers, not schoolwork. In all of these studies, social factors such as SES and gender may have influenced items within those measure that did not regard boredom and interest, calling attention to the importance of clear and specific construct definition.

In our study, being white was the only consistent predictor of trajectory membership with effect sizes that ranged from small to medium across the four disengagement trajectories. This could represent the protective process of being an ethnic‐minority student in England, as they are often second‐generation immigrants, whose attitudes towards schooling are positively influenced by their parents’ high expectations for achievement, even if their achievement is often lower than their white peers (Strand, 2007).

Disengagement and achievement: A dynamic developmental system

Third we predicted that the disengagement trajectories would have similar achievement to their comparison trajectories at baseline (assumption c) but that their achievement would wane across the school years (assumption d). Accordingly, in the LSYPE, students disengaged in various ways, regardless of their childhood achievement. However, by the end of compulsory school, students in all disengaging trajectories had worse achievement on average than their counterparts. Our suggestion is that this decline in achievement was part of a broader dynamic system of emotional and behavioural disengagement, for, in line with assumption e, by Wave 3, disengaged students also tried less hard at school and attended less often, and had lower aspirations for academic success, which presumably would have negatively impacted their achievement. This might also explain why emotional engagement did not predict achievement in Wang and Eccles (2012), after they added behavioural engagement to their model, as behavioural engagement may have fully mediated the connection between emotional attitudes and achievement.

Longitudinal associations with education and employment

Next we assumed (f) that emotional disengagement from schoolwork would associate with less participation in full time education and more uptake of employment, after compulsory schooling. At age 17–18 years, our assumption was met for all disengagement trajectories.

The differences in educational participation might relate to the disengagement trajectories having lower end‐of‐school achievement, as students in England need to get reasonably high grades in order to continue to A‐Level courses. Also, they might reflect an avoidance strategy (Heckhausen et al., 2010), where students sought to protect themselves against having similar negative emotional experiences in the same institution or in a sixth form or Further Education college. In all those institutions, there are some environmental similarities including timetables and examinations, which can act as acute stressors in late adolescence (McCoy et al., 2014). The lower rates of participation in full‐time education at age 17–18 might also have affected the subsequent lower rates of participation in university and further education in young adulthood, observed for several of the disengagement trajectories. Together these results suggest that emotional disengagement from schoolwork affected students’ school and career pathways by altering their chances of, and dispositions towards progression. Likewise, in Finland, having a cynical attitude towards school has predicted dropping out of education (Bask & Salmela‐Aro, 2013).

Disengaged students were also more likely to enter employment in late adolescence as expected, but not in young adulthood; perhaps as more comparison students finished their educational courses and entered the workforce. Another interesting result surfaced in young adulthood, where the disengaging mildly, disengaging rapidly and disengaging disengaged students were more likely to be unemployed than their comparison students. Possibly their lower educational aspirations and achievement might have been a risk factor for unemployment, as has been found already in the LSYPE, and in other British cohorts (Duckworth & Schoon, 2012; Schoon 2014).

Longitudinal associations with wellbeing

Finally, we assumed (g) that disengaged students would have similar psychological wellbeing and risk behaviours to their counterparts in young adulthood, despite having poorer psychological wellbeing and riskier behaviour at the end of compulsory school, indicating some type of personal recovery as they engaged in new activities that were presumably less boring and more interesting to them. However, in the year after compulsory school, students in the disengagement trajectories still had higher anxiety and subjective ill‐being. Possibly these were recent spillover effects from their negative emotional experiences of doing schoolwork (Salmela‐Aro & Upadyaya, 2013) that were only temporary, for two years later in young adulthood, only the disengaging rapidly students were notably less satisfied with their lives. Interestingly, even the stable disengaged students had very similar life satisfaction to those who were stable engaged, indicating that despite their prior differences in wellbeing, and their different uptake of education and employment, they evaluated their lives as being just as good in young adulthood. Possibly, their greater uptake of vocational pathways acted as a protective process, by which their mental health improved (Symonds et al., 2016).

A similar pattern appeared for substance use, where the higher levels of smoking, drinking and drug use observed at the end of compulsory schooling were not present in young adulthood. Our suggestion is that those enhanced levels at compulsory school were part of the broader dynamic disengagement system described earlier, where students avoided school and spent more time engaged in risk behaviours outside school (Henry et al., 2012). However, after leaving school, these students might have resumed a more normative level of substance use. Their engaged counterparts who were more likely to attend university might also have overtaken them in substance use, for in the US, young adults attending university have reported higher levels of substance use than their non‐university peers (Carter et al., 2010). Regardless, the findings present a picture of improved health behaviours after leaving school for emotionally disengaged students.