Abstract
This study employs Macro-level Strain Theory (MST) as a framework to provide a better understanding of the way in which the structural and social context of Stockholm schools covaries with self-reported violent and general offending. The findings contribute to the literature in this area by directing a special focus at the interplay between the theory’s macro-level components and some individual-level mechanisms that may be assumed to condition the effect of strain on offending. Using multi-level data on 4789 students nested in 82 schools (violent offending) and 4643 students nested in 83 schools (general offending) in the City of Stockholm, the study notes significant contextual effects of anger, meaninglessness and life dissatisfaction on offending. School-level deprivation appears to have a confounding effect on the relationship between school-contextual negative affect and offending. Further, school-contextual anger influences some individuals more than others. Implications of these findings are discussed.
Introduction
This study analyses the variance in crime between schools from a contextual strain perspective applied in Stockholm. Since the millennium, Western criminology has been characterized by a focus on individual-level explanatory models, a trend that in a number of countries, Sweden included, has occurred in parallel with increases both in levels of inequality and in social differences between groups of individuals (Nilsson et al., 2016). According to Hagan and Rymond-Richmond (2013), recent years have witnessed something of a re-emergence of the status ascribed to the contextual, urban sociological perspective, among other things thanks to Sampson’s (2012) comprehensive analyses of Chicago’s residential neighbourhoods. Contemporary analyses employing contextual approaches are thus often dominated by neighbourhood-focused analyses based on either social disorganization theory (Shaw and McKay, 1942; Welsh et al., 2000) or more recent versions of this theory, such as the theory of collective efficacy (Sampson, 2012). This approach primarily examines variations in crime between social units in terms of informal social control and social trust (Brezina et al., 2001; Felson et al., 1994). Another theoretical perspective that also allows for the combination of micro- and macro-level mechanisms, but which has been the focus of less empirical research, is Agnew’s (1999) contextual version of General Strain Theory (GST), more commonly known in the literature as Macro-level Strain Theory (MST).
During recent decades, the variation between Swedish schools in students’ educational achievements has increased (Yang Hansen and Gustafsson, 2016), and this variation is now considerably greater than in comparable European countries (OECD, 2016). The degree to which crime is concentrated in certain places is assumed to increase with increased levels of segregation (Desmond and Kubrin, 2009; Sampson, 2012), which makes it particularly interesting to study the metropolitan school context in Sweden. Based on the social disorganization perspective, an increase in the geographical concentration of crime may be understood as the result of reduced informal control and reduced social trust (Sampson, 2012). The contextual strain perspective introduces a cultural dimension, which emphasizes extending our understanding of how the social context may produce a culture of motivation for criminal acts (Agnew, 1999). According to MST, variations in crime between schools may be partially explained by the mean level of negative affect within the student population, inter alia because there is a high probability that angry and frustrated students will tend to be concentrated in socially disadvantaged environments (Agnew, 1999). Adopting the MST perspective thus provides an opportunity to develop a deeper understanding of how structural and social variables at the group level exert an influence on variations in crime. This is important, in part because it increases our theoretical understanding of the interaction between the context and the individual, but also because it may have implications for policy. For example, it draws attention to the social density of strained individuals and to the fact that this may vary from one school context to another. Thus on the basis of an MST framework, the overarching objective of the current study is to examine the impact of school-level strain (negative affect and deprivation) on offending.
School context vs neighbourhood context in Stockholm
MST was primarily developed to account for differences in crime between residential neighbourhoods. European studies have produced varying findings regarding neighbourhood effects on crime, however, and some have questioned the significance of the neighbourhood for differences in crime (Gerell, 2014; Op de Beeck et al., 2012; Pauwels, 2010). Results showing that neighbourhood effects are weak may be due to the fact that the units of analysis employed at the neighbourhood level are often of a relatively heterogeneous nature, thus equalizing neighbourhood characteristics within each of the neighbourhoods examined. Agnew (1999) has therefore argued that MST might be most appropriate for analyses of data based on smaller social units, such as schools.
In an earlier study, Lindström (1993) found that the school context was of limited significance for offending among Stockholm youth. Since his analyses were conducted, however, Sweden has witnessed major reforms to the school system during the 1990s. Responsibility for schools passed to the municipalities, and a reform was passed that increased the number of independent schools and also introduced freedom of choice for students/families to choose school via a universal voucher system. Recent studies have shown that school results, as well as levels of mental ill-health, drug use and offending, vary between different schools in Stockholm (Eklund and Fritzell, 2014; Modin et al., 2015; Sandahl, 2016), even given controls for school composition with regard to individual-level characteristics. A central point of departure for the current study is that school constitutes an important context in which adolescents’ life conditions are consolidated and reproduced. Another is that Sweden, which is often presented as a leader in the areas of welfare and equality, but which has been characterized by increasing welfare inequalities over recent years (Nilsson et al., 2016), constitutes a highly relevant study object in this regard.
MST – A macro-level extension of GST
Unlike the classical and institutional versions of strain theory (see, inter alia, Cloward and Ohlin, 1960; Cohen, 1965; Merton, 1938; Messner and Rosenfield, 2001), Agnew’s (1985) initial conceptualization of strain focused less on structural injustices between social settings and more on the individual-level negative emotions produced by these injustices. In his General Strain Theory (GST), Agnew (1992) stated that strain – conceptualized as feelings of anger, frustration, fear or depression – is the cause of offending (Agnew, 1992). As a causal theory at the micro level GST has been criticized for failing to explain why most forms of strain seem to lead to responses other than offending (see, for example, Wikström, 2010). This is not an issue in this particular study because it focuses on the interplay between contextual and individual mechanisms and differences in offending between schools, rather than on individual-level causal explanations. GST has also been criticized by structural theorists for overemphasizing the socio-psychological consequences of strain, thus missing the macro-level implications of the theory (Bernard, 1984; Young, 2010). Having recognized that GST was mainly being used with a focus on the individual level of analysis, Agnew (1999) addressed this issue by formulating MST, an elaborated macro-level version of GST.
MST broadens the focus to include not only individual differences in crime, but also differences between neighbourhoods and possible causes of these differences at both the social and structural levels. According to Agnew (1999), neighbourhood characteristics such as poverty and unequal opportunities not only serve to increase individuals’ perceptions of strain but also give rise to some form of aggregate, group-level sense of anger and frustration. This aggregate-level strain is assumed to produce a direct effect on the propensity to engage in crime, even after controlling for individual strain. However, subcultural mechanisms, such as values conducive to crime, are also assumed to condition the contextual effect of strain on crime (Agnew, 1999). Thus, aggregate levels of negative affect should not replace structural mechanisms in the explanation of variations in crime between schools. Instead, economic deprivation is assumed to lie behind not only variations in crime but also possible variations in the mean level of perceived strain. This is reminiscent of social disorganization theory in the sense that the point of departure is that social injustices are predictors of other mechanisms at the neighbourhood level (formal and informal social control), which are in turn predictors of crime. According to Agnew (1999), the strain perspective can complement this approach by expanding the focus to the ways in which the context may also contribute to the motivation to offend. Applied to the school context in the current study, this means that structural factors (for example, in the form of segregated/deprived schools) result in certain schools producing more students who are experiencing strain, and thus a higher level of contextual strain.
Previous empirical studies on MST
Several studies have examined the relationship between different aspects of the school context and offending, contributing essential theoretical and practical knowledge by illustrating the significance of school composition, culture and organization for school disorder (see, for example, Welsh et al., 2000). A number of studies focusing on GST have also found support for the existence of a relationship between individual-level strain and violence (for example, Mazerolle and Maahs, 2006; Piquero and Sealock, 2006). However, only a small number of studies have combined school-level and individual-level data to test hypotheses based on MST (Brezina et al., 2001; Hoffmann and Ireland, 2004; Op de Beeck et al., 2012). Some studies have used the theory to examine variations in crime between neighbourhoods (Wareham et al., 2005; Warner and Fowler, 2003). The current study focuses specifically on the school context, thus the following review is restricted to studies that have applied MST to school-level data.
Brezina et al. (2001) found support for MST by showing that school-level anger is predictive of violence and conflicts among students, even given controls for anger at the individual level. However, the aggregate measure of student anger failed to produce a significant effect on disruptive behaviour, which included conflicts with teachers, indicating that school-level anger was associated with a behaviour-specific effect. As noted by Brezina et al. (2001), the study was limited to exploring only one of the various school-level aspects of negative emotions suggested by Agnew (1999) to be predictive of offending (anger). The study also lacks a measure that might capture possible sources of anger and other negative emotions at the school level, such as school organizational factors or school deprivation.
Op de Beeck et al. (2012) tested MST on a sample of schools in Belgium and found partial support for the theory in that school-contextual strain (measured in terms of aggregate levels of future prospects, negative affect and life dissatisfaction) significantly affected violent offending but not general offending. As in Brezina et al.’s study, Op de Beeck et al. did not include any other measures at the school level in addition to the strain-related negative emotions. Nor did they include any of the potential individual-level mechanisms that Agnew suggests might moderate the relationship between school-level strain and offending. Hoffmann and Ireland (2004), who studied the potential moderating effect of school opportunity structures on the relationship between individual-level strain and general offending, were not able to confirm this hypothesis. However, they did find that the relationship between strain and delinquency varied randomly across schools, thus suggesting that other school characteristics, not measured in their study, might affect strain and delinquency and should be investigated in future research.
MST is a very broad framework, integrating a number of theoretical perspectives and several different levels of analysis. The variation in findings may in part be due to studies having employed the theoretical framework of MST differently. It may also be due to differences in outcome measures and to the many different possibilities provided by Agnew for the conceptualization of strain. The current study contributes to the research in this field by testing a combination of different aspects of MST that have previously been completely or partially excluded from empirical studies. In order to avoid losing sight of the macro implications, this study takes into account the view that the contextual motivation to commit offences described in MST is, in turn, rooted in structural conditions. The influence of school characteristics on self-reported offending will therefore be examined in part by means of school mean levels of perceived negative affect, which is in line with the approaches adopted in previous studies, and in part by means of more objective indicators that are intended to measure the schools’ level of deprivation. 1 Previous studies (Agnew, 1999; Brezina et al., 2001; Op de Beeck et al., 2012) have also called for research that may shed light on individual-level mechanisms with potential to condition the effects of contextual strain on offending (often referred to as subcultural mechanisms). The current study therefore also includes an examination of interactions between strain at the school level and values conducive to crime at the individual level. According to Agnew (1999), the origin of such values can partially be explained in terms of strain theory. Finally, the study is the first to apply the MST perspective in a Northern European (Swedish) metropolitan context.
Study outline and hypotheses
The objective of this study is to examine the impact of school-level strain (negative affect and deprivation) on individual-level violent and general offending (see Figure 1). More specifically, the study will test the following hypotheses:
Study outline.
Methods
Data
The data analysed in this study are primarily drawn from the Stockholm School Survey (SSS) of 2014. The SSS is a large-scale school questionnaire survey conducted every two years by the City of Stockholm. The survey includes all Year 9 classes in compulsory schooling (15 year olds) and all second-year upper-secondary school students (17 year olds), at schools managed by the City of Stockholm. A large proportion of independent schools also participate on a voluntary basis. The present study concerns the Year 9 students, since relevant register data are not available for the upper-secondary school level. The data collection was administered, and the data coded, by the company Markör marknad och kommunikation (www.markor.se). The questionnaires were distributed to the schools in March and collected during April and May. The students completed the questionnaires anonymously during lesson time and returned them to their teachers in a sealed envelope. Markör has estimated the level of missing data due either to students being absent from school or to questionnaires being completed incorrectly to 22 percent. In order to adapt the material to the multi-level structure employed in the analyses, questionnaires from schools with fewer than 10 students were removed from the data set. Respondents with missing data on one or more of the outcome or school-level variables have also been excluded. The final study samples were based on 4789 students distributed over a total of 82 schools (violent offending) and 4643 students distributed over 83 schools (general offending). Following the principle that comparisons of estimates across different models should as far as possible be based on the same observations, students with missing data on any of the relevant questionnaire items have been excluded from the analyses. See Table 1 for descriptive statistics.
Descriptives.
The structural characteristics of the schools are measured by means of a register-based index calculated by Statistics Sweden for the schools administration at the City of Stockholm, which is used to distribute resources among schools on the basis of the extent of the problems the schools are assumed to have.
Measures
Because the SSS questionnaire was not originally developed to capture theoretical concepts relating to MST mechanisms, the measures employed in the current paper have been chosen with the aim of coming as close as possible to the mechanisms defined by Agnew (1999) as (aggregated) negative affect. With the exception of anger, Agnew is not very specific in defining which particular negative emotions are comprised by this concept. Therefore this study follows recommendations by Brezina et al. (2001), which suggest that future tests of MST should include a broad range of negative emotions.
Outcomes
In Agnew’s (1999) original formulation of MST no crime-specific outcome variable was specified. Thus, since previous contextual studies in general (see, for example, Oberwittler, 2004) and previous testing of MST in particular (Brezina et al., 2001; Op de Beeck et al., 2012) have shown that contextual effects vary depending on the type of crime used as the outcome, the present study makes a distinction between violent and general offending. Furthermore, self-reported minor offences (such as fare-dodging, shoplifting and graffiti) were excluded from the analysis in order to avoid including behaviours that have relatively high prevalence rates in the particular age group examined in this study.
The study employs logistic regression with dichotomous outcome variables, because this is assumed to provide more reliable results and better opportunities to interpret the findings (Feng et al., 2014), compared with using a skewed continuous outcome, which would have required transformation. The reliability of the variables based on indexes is reported in terms of the value of Cronbach’s alpha in connection with the description of each operationalization. Cronbach’s alpha provides information on the internal consistency of these composite measures.
Violent offending was measured using the only question available in the questionnaire that aims to capture physical violence: ‘How many times have you done the following things during the past 12 months?’ – ‘Intentionally hit someone in such a way that you think or know that he/she required medical care’. Those who reported having done so at least once were assigned the value 1; others were assigned the value 0.
General offending was captured by the questions: ‘How many times have you done the following things during the past 12 months?’ – ‘stolen a car’, ‘stolen a moped/motorbike’, ‘broken in somewhere’ or ‘forced someone to give you something valuable’. Those who reported having done so at least once were assigned the value 1; others were assigned the value 0 (α = 0.834).
School-level independent variables (level 2)
The school mean level of deprivation is measured using a resource distribution index employed by the Stockholm City schools administration, which is comprised of the following background factors: whether the student is immigrant, and, if so, the student’s year of arrival in Sweden; the educational background of the student’s guardians; whether the student’s family is in receipt of income support; whether the student lives with both, one or no parents/guardians; residential neighbourhood and school environment. 2 The index is based on a model produced using statistics from the most recent academic year on the proportion of students who are not expected to achieve their educational targets. The background factors are weighted such that the index value for Stockholm as a whole is 100, and each school is then assigned an individual index score in relation to the average for the city. During the year of the study (2014), the individual schools had index scores ranging between 14 and 490, with high scores indicating high levels of socio-structural disadvantage (referred to in this study as a high degree of school deprivation). In order to facilitate interpretation of the results, the index has been divided by 10 and thus varies between 1.4 and 49.0, with the city average score being 9.93.
School-level strain is captured by using aggregate measures of a number of student-reported factors that, according to MST or previous research on MST, may constitute examples of the negative emotions produced by strain. These are: perceptions of anger; that school is meaningless; life dissatisfaction; and a threatening school environment. Agnew (1999) explicitly argues that communities with high levels of anger, criminal victimization and fear produce the most serious types of strain. Thus, the measures of anger and perceiving the school environment as threatening may be the dimensions of negative affect that are most consistent with the terminology of MST. The present study also incorporates school meaninglessness as an attempt to capture frustration in the form of an atmosphere of hopelessness. Notably, the dimension of life dissatisfaction (although measured somewhat differently) was found to be the only school-level strain indicator that significantly affected general offending in Op de Beeck et al.’s study (2012).
Anger is measured as the school’s mean score on an index based on the questions: ‘I can’t cope with being provoked – if I am I may hit someone’, ‘If I get angry with someone, I’m not afraid to injure him/her’, ‘I’ll set about someone who makes me angry, even if he/she hasn’t hit me first’, ‘I do the opposite of what people tell me to do, just to make them angry’. The response alternatives are: ‘strongly disagree’, ‘somewhat disagree’, ‘somewhat agree’, ‘strongly agree’. Index scores range between 4 and 16 (α = 0.646).
School meaninglessness is measured in terms of the school’s mean score on an index based on the following items: ‘Schoolwork feels meaningless’, ‘Schoolwork makes me confused’. ‘My teachers give interesting lessons’ (reverse coded). The response alternatives are: ‘strongly disagree’, ‘somewhat disagree’, ‘somewhat agree’, ‘strongly agree’. Index scores range between 3 and 12 (α = 0.59).
Life dissatisfaction is measured as the school’s mean score on an index based on the questions: ‘Do you feel sad and depressed without knowing why?’, ‘Do you ever feel frightened without knowing why?’, ‘How often do you think that it’s really great to be alive?’ (reverse coded). The response alternatives are: ‘seldom’, ‘occasionally’, ‘sometimes’, ‘fairly often’ and ‘often’. Index scores range between 3 and 15 (α = 0.83).
Threatening environment is measured using the items: ‘I worry about being exposed to crime at school’, with ‘strongly disagree’ and ‘somewhat disagree’ being assigned the value 0, and ‘somewhat agree’ and ‘strongly agree’ assigned the value 1; and ‘How often have you been bullied or harassed at school during this school year?’, with the response alternatives ‘I have not been bullied’ and ‘once or twice’ being assigned the value 0, and the response alternatives ‘about once a week’ and ‘several times a week’ being assigned the value 1. The school-level measure is based on the mean proportion of students with a score of 1 on these two measures (they are either worried about being exposed to crime or have been harassed).
Individual-level independent variables (level 1)
Individual-level strain employs the measures of anger, meaninglessness, threatening school environment and life dissatisfaction described above, but at the individual level.
Sex is measured by the question: ‘Are you a boy or a girl?’ Being male is employed as the reference category in the analysis.
Parental education is measured using the question: ‘What are your parents’ highest levels of education?’, with separate response alternatives for the mother and father: compulsory education (no more than nine years of education), upper-secondary, university or equivalent, and a ‘don’t know’ alternative. The variable has four categories, with two parents having a university education being used as the reference category. The other three alternatives are: one parent with a university education, no parent with a university education and ‘don’t know’ / information missing. 3
Conditioning mechanisms
Agnew (1999) hypothesizes that subcultural mechanisms, such as values conducive to crime, may condition the effect of contextual strain on offending. In line with Felson et al.’s (1994) understanding of subcultural values as attitudes that approve of or tolerate norm-breaking behaviour, the current study measures this mechanism by means of an index comprising a number of items focused on the students’ attitudes towards rules: ‘I like doing things that are dangerous and exciting even if they are forbidden’, ‘I ignore rules that prevent me from doing what I want to do’, ‘I do silly things, even if they are a bit dangerous’, ‘I think it’s okay to take something without asking, if you don’t get caught’. The four response alternatives are: ‘strongly agree’, ‘somewhat disagree’, ‘somewhat agree’ and ‘strongly agree’. Index scores range between 4 and 16, with high scores representing values conducive to crime (α = 0.774).
Analysis
Although it is reasonable to assume that the greatest degree of variance in offending will be found at the individual level, this study assumes that there are also systematic differences depending on which school the respondents attend. The analyses have been conducted in STATA version 16.0.
The covariance between the two binary outcomes of crime and the indicators of school-level strain are examined empirically by means of multi-level logistic regression analysis (using a two-level random intercept model). The question of whether there are significant differences in self-reported offending between schools is first examined in an empty model. The empty model contains no independent variables but allows for the variance in self-reported violent and general offending to be separated into two components, one for each level (individual and school). Thus the empty model shows whether or not there are any contextual differences between schools, which is presented in terms of the ICC measure of variance provided in the STATA output. 4 Subsequently, the analysis examines the study’s four hypotheses on the basis of what the multi-level literature refers to as ‘the structure of macro-micro propositions’ (Snijders and Bosker, 2012). In stage one (Models 1:1, 2:1, 3:1 and 4:1 in Tables 2 and 3), the analysis examines whether there are significant correlations between the various indicators of school-contextual strain and individual offending given controls for perceived strain at the individual level (H1). In stage two (Models 1:2, 2:2, 3:2 and 4:2), the focus is directed at whether these correlations remain following the introduction of controls for individual background factors in the form of sex and the parents’ level of education (H2). The analysis then moves on to include the more objective measure of the schools’ socio-structural characteristics (Models 1:3, 2:3, 3:3 and 4:3) in order to examine whether the schools’ level of deprivation may explain the correlation between school-level subjective strain and violent and general offending respectively (H3). The final model includes a measure of attitudes towards rules at the individual level as a means of examining the possible effects of subcultural mechanisms, which MST argues may condition the correlation between school-contextual strain and offending. Therefore, the current study specifically examines whether the relationship between strain at the school level and individual offending is dependent on the students’ own attitudes towards delinquency. This assumption is examined separately by means of so-called cross-level interactions (Snijders and Bosker, 2012). For reasons of clarity, significant interactions are presented graphically in Figure 2.
Schools’ mean level of strain, level of deprivation and individual conditioning mechanisms in relation to violent offending. Logistic random intercept models.
Note:
Tests for interactions were carried out separately, which means that the main effects presented in the models are not adjusted for the interaction terms. No significant interactions were found with violent offending as the outcome.
*p < .05; **p < .01; ***p < .001.
Schools’ mean level of strain, level of deprivation and individual conditioning mechanisms in relation to general offending. Logistic random intercept models.
Note:
Tests for interactions were carried out separately, which means that the main effects presented in the models are not adjusted for the interaction terms. One significant interaction was found with general offending as the outcome, specifically the interaction between school-level anger and values conducive to crime (p = .036). See Figure 2 for an illustration of the significant interaction.
*p < .05; **p < .01; ***p < .001.
Illustration of the estimated association between a school’s mean level of anger and general offending depending on students’ attitudes towards crime (cross-level interaction).
For each model, information is provided on the unexplained variance in the outcome measure at the school level, which enables the MOR (median odds ratio) to be calculated. When working with a dichotomous outcome variable, Merlo et al. (2006) argue that the MOR is preferable to the ICC (intra-class correlation), which is otherwise more common in multi-level analysis. In the current study, the MOR may be described as the median change in the odds of a student reporting one of the relevant offences if he or she were to move from one school to another. A MOR value equal to 1 would mean that there is no variation in self-reported offending between schools (Merlo et al., 2006)
Results
Between-school variance
Table 2 (violent offending) and Table 3 (general offending) present the results of the analysis in a series of models that reflect the study’s hypotheses. The empty model indicates that the variance in crime between schools is small but statistically significant with regard to both violent (ICC = 0.08) and general offending (ICC = 0.07). These estimates mean that 8 percent and 7 percent of the variance in violent and general offending, respectively, may be ascribed to the school level. These proportions may appear small, but they are in line with the findings of similar studies. According to Felson et al. (1994), it is unusual for more than between 5 and 10 percent of the total variance in ‘any’ dependent variable to be located at level 2 (see also Welsh et al., 2000). Studies employing similar outcomes have shown that the variance in crime between schools usually varies between 2 and 10 percent, with the between-school variance found in European studies often being lower compared with US studies (Brezina et al., 2001; Op de Beeck, 2012). The MOR estimates in the empty models may be interpreted as showing that the odds of reporting violent offending would increase by a factor of 2.01 at the median if a student were to move from one school to another that has a higher risk of offending. The corresponding estimate for general offending is 1.95.
Test of school-contextual strain (H1 and H2)
The relationships between self-reported offending and school-level strain are presented in the first and second models for each of the strain indicators in Tables 2 and 3. School-contextual strain reflects the level of negative emotions (anger, meaninglessness, life dissatisfaction and threatening environment) that according to MST are assumed to be produced by strain.
When the school-level measures of strain are included in the model, the MOR value decreases in all cases by comparison with the empty models, indicating that these strain indicators are capturing something in the school culture that may contribute to explaining the between-school variance in offending. For both violent and general offending, school-level anger (OR = 1.34 and 1.31, respectively) appears to be significantly related to the outcome variables, given controls for anger at the individual level (H1). These estimates are somewhat strengthened by the inclusion of individual-level controls (H2). The odds of having reported at least one violent offence increase by 43 percent for each unit increase in the school-level index of anger, while the corresponding increase for general offending is 36 percent. The effect of school-level life dissatisfaction on violent- and general offending also appears to remain when controls are included both for the effect of the individuals’ own perceptions of dissatisfaction (OR = 1.57 and 1.69 respectively) and for sex and parental education (OR = 1.61 and 1.69). However, the measures of meaninglessness and of a threatening school climate are significantly related to offending only at the individual level in the two first models.
Further, the analysis shows life dissatsfaction to constitute the school-level strain indicator that most clearly confirms H1 and H2 in relation to both violent and general offending, which corresponds to the findings of Op de Beeck et al. (2012).5 Anger at the school level also appears to increase the risk for offending following the inclusion of controls for individual level anger and background factors (models 1:2 in tables 2 and 3).. This provides support for one of the central hypotheses of MST, further confirming findings produced by earlier studies in this area (Agnew, 1999; Brezina et al., 2001). The findings of the current study thus provide only partial confirmation for H1 and H2 since the other hypothesized indicators of school-contextual strain are not significantly correlated with the outcomes examined once controls are included for the individual-level measures.
The explanatory value of school-level deprivation (H3)
The current study hypothesizes that the effects of the school-contextual indicators of strain may be explained by school deprivation (H3). Models 2:3, 3:3 and 4:3 in Tables 2 and 3 show that the indicator of school deprivation is significantly related to both outcome measures, even when considering the other variables in the analysis. In cases where violent and general offending were predicted by school-level life dissatisfaction, the inclusion of the school measure of deprivation appears to lead to school-contextual strain no longer exhibiting a significant correlation with the outcome variables (Models 3:3 in Table 2 and Table 3). Regarding school level anger (models 1:3 in Table 2 and 3). Here, the two school-level indicators of anger and deprivation instead seem to cancel each other out. This is also indicated by the MOR value in this case, since it is largely unaffected by the introduction of deprivation, unlike the other models. This indicates that school structural characteristics partially explain the school-level effects of subjective strain, or that these school-level variables substantially overlap. An overlap of this kind would suggest that school-contextual anger and dissatisfaction are only indirectly related to offending. This would provide support for H3, while simultaneously introducing a greater degree of uncertainty in relation to H1 and H2. The relationship between school-level meaninglessness and general offending (model 2:3 in table 3) constitutes an exception in this regard. This association is somewhat strengthened and reaches statistical significance, when the measure of school deprivation is included in the model, indicating that the school’s level of meaninglessness is not explained by the level of deprivation as formulated in H3. Rather the effect of school meaninglessness seems to be dependent on the school’s level of deprivation in either a mediating or a moderating way.
The effects of school-level deprivation appear to be relatively weak, but it is important to remember that the index ranges between 1.4 and 49.0. In Model 3:3 (Table 2), for example, this indicates that the odds of reporting a violent offence increase by 3 percent for each unit increase in the school deprivation index. With the exception of the analyses focusing on anger (models 1:3 and 1:4), these effects remain significant in all cases. In addition to showing that school-level deprivation appears to explain the effect of subjectively perceived strain at the school level, this also indicates that there are substantial differences in self-reported offending between schools that have the lowest and highest scores on the deprivation index. Even when the odds ratio for the socio-structural index is at its lowest (OR = 1.01), this means that the odds of having offended in one of the most deprived schools (which have the highest scores on the deprivation index) are 61 percent higher than in the least deprived schools with the lowest index scores. 6 Attending a ‘deprived’ school thus substantially increases the likelihood of engaging in both violent and general offending, even when individual differences and different aspects of school-contextual strain are held constant. This provides support for the macro-theoretical additions to strain theory that MST provides in relation to Agnew’s original theory (GST), and for the criticism that emphasizes the importance of also including more objective structural mechanisms in the analysis.
Test of conditioning mechanisms at the individual level (H4)
Based on Agnew’s (1999) arguments that there are a number of subcultural mechanisms that increase the likelihood of responding to strain by engaging in offending, this study has examined the potential conditioning effect of students’ attitudes towards crime.
The analysis of whether the relationship between school-contextual strain and offending is conditioned by students’ attitudes was examined through cross-level interactions between each indicator of school-contextual strain and values conducive to crime at the individual level. Only one interaction was found to be significant in the full model (p < .05), namely that between school-level anger and values conducive to crime in relation to general offending. The fact that the interaction term is statistically significant means that the effect of the one variable is dependent on, or affected by, the value of the other. The joint effects of these two variables on general offending are presented in Figure 2, in which both of the interacting variables have been dichotomized at the midpoint in order to facilitate interpretation of the results. 7
The separate interaction analysis indicates that values conducive to crime appear to moderate the effect of school-level anger on general offending, thus confirming H4 in this specific instance. Figure 2 illustrates that the relationship between school-contextual anger and general offending is considerably stronger among students with pro-delinquent attitudes. Compared with the reference category (not having pro-delinquent attitudes and attending a school with a low mean level of anger), the odds of having committed a general offence are almost fourteen times greater among students with pro-delinquent attitudes who also attend a school with a high mean level of anger. Among students with pro-delinquent attitudes who attend a school with a low mean level of anger, the odds ratio is 9.3, while the excess risk among students who attend a school with a high mean level of anger but who do not have pro-delinquent attitudes is relatively small (OR = 1.5). This indicates that school-contextual strain primarily affects those students who are already at heightened risk of committing offences.
Summary and discussion
This study has employed Macro-level Strain Theory as a framework for improving the understanding of how the structural and social context of Stockholm schools covaries with violent and general offending. The study contributes to the existing literature in this area by focusing specifically on the interplay between macro-components of the theory (aggregate levels of negative emotions and deprivation) and the individual (subcultural) mechanisms that are assumed to potentially condition the relationship between contextual factors and offending.
Firstly, the between-school variance in crime does not differ greatly between the two outcome variables employed in the study. Op de Beeck et al. (2012) found convincing contextual effects in relation to violent but not general offending. This study found that 8 percent and 7 percent of the variance in self-reported violent and general offending, respectively, was located at the school level.
Secondly, compared with the more subjective indicators of strain that were tested in this study, the deprivation index appears to constitute a more prominent school-level factor in relation to both outcome variables. Attending a deprived school is associated with higher levels of self-reported offending among students, irrespective of the students’ individual-level characteristics. At the same time, the aggregate levels of the negative emotions examined in this study were found to have either a modest direct effect on self-reported offending or no effect. Experiencing the school environment as threatening, appear to be significantly related to offending only at the individual level. However, the school-level measures of anger and life dissatisfaction was found to be significantly related to both violent and general offending, even when controls for individual-level characteristics were included.
The significant effect of these school-contextual factors disappears, however, when the measure of deprivation is introduced into the model, indicating that the effect of school-level anger and life dissatisfaction on offending is in whole or in part explained by the schools’ socio-structural characteristics. This provides support for H3 in the current study, and it also introduces a greater degree of uncertainty in relation to H1 and H2 at the theoretical level. School-level deprivation appears to be directly correlated with both the school-contextual indicators of strain and the two outcomes, thus producing a confounding effect on the relationship between school-contextual strain and offending.
Finally, the interaction analysis provided support for H4 in one instance, because the association between school-level anger and general offending was conditioned by the students’ own attitudes towards offending. Thus school-contextual negative affect seems to predict offending particularly among students with values conducive to crime. According to Agnew (1999), a conditioning effect of this kind may explain why people do not always react to strain by engaging in offending. Thus one possible interpretation is that a school climate characterized by anger may be particularly important for students who are already at heightened risk of offending, whereas students who are not at such high risk do not appear to be affected by the context to the same extent. In this sense, the current study may in part be seen as providing support for the arguments presented by Brezina et al. (2001) and Op de Beeck et al. (2012) that the weak effects on offending noted in relation to school-contextual strain may be due to the failure to include conditioning mechanisms in the analysis. Therefore, further research on MST should focus on these conditioning mechanisms and place more emphasis on analysing the interplay between contextual and individual-level factors as a means of improving the understanding of school-level variation in offending.
Although Agnew’s (1999) specification of MST provides a theoretical framework that is easily applied to data from school surveys, it is primarily those aspects of the theory that originate from other theoretical perspectives (classical strain and subcultural theories) that find support in the analyses presented in this study. For example, Cloward and Ohlin (1960) argued that the presence of illegitimate opportunity structures is necessary for strain to be directed into delinquent adaptations. The measures of aggregated negative affect used in the current study indicate that these contextual aspects are of limited significance for offending following the inclusion of the more objective measure of school deprivation. Therefore, future tests of MST should not exclude such an important aspect of the theory. On the basis of the analyses presented above, another strength of MST may be found in the opportunities it provides for the integration of macro- and micro-mechanisms as a means of improving the theoretical understanding of the interplay between structural, social and individual factors. The classical strain theories (Cloward and Ohlin, 1960; Cohen, 1965; Merton, 1938) are often criticized for ignoring the links between offending and frustration at the individual level and for being difficult to test empirically (Bernard, 1984). At the same time, more modern, often micro-oriented, versions of strain are instead commonly criticized for focusing only on the socio-psychological consequences of strain, and thus of ignoring structural factors (Young, 2010). Hagan and Rymond-Richmond (2013) emphasize the contribution to criminology made by Sampson’s (2012) concept of collective efficacy. They argue that it is an integrated focus on the individual and the social context in combination, using modern statistical methods (such as multi-level analysis), that enables us to challenge the individualistic bias that has characterized the field in recent years, at both theoretical and policy levels.
The MST perspective may be viewed as complementing such approaches by focusing on how societies, in addition to controlling crime, may also produce motivations for crime (Agnew, 1999). Applied to the school context, and based on the results produced by the current study, this would mean that social policy interventions should not only focus on schools as agents of social control but also acknowledge that both the social and the structural context at school may serve to motivate offending. According to MST, deprived schools are more likely to attract and retain individuals with high levels of strain, thus producing a compositional effect, which will in turn most likely result in higher levels of contextual strain (Agnew, 1999). This implication is particularly relevant for countries with school systems similar to Sweden’s, with the freedom to choose one’s school. Research has shown that it is primarily socially advantaged families who utilize the opportunity to choose which school to attend (Bunar, 2010), which tends to further increase both school segregation and the variation between schools in the social density of angry and dissatisfied students. Furthermore, structural and context-based interventions reduce the risk of labelling effects and control-related harms among individual students. Since the 1990s, the Swedish school system has been characterized by a substantial focus on the individual, with the emphasis being directed at the control mechanisms that schools are assumed to be able to mobilize for the purposes of crime prevention (Wahlgren, 2014). Thus there is a need for a more context-based view of the significance of the school system for crime and other social problems.
Limitations and directions for further research
Although the data set employed in this study is large and ideal for multi-level analysis, it is also subject to certain limitations. One of these is that the data are secondary, which means that the variables used in the current study, like previous MST studies, were originally constructed to capture other theoretical concepts than the ones derived from MST. Therefore some of the measures could be criticized for not capturing the strain indicators sufficiently. Another limitation is that the data are cross-sectional, which prevents causal interpretations. The variables based on the students’ questionnaire responses may also be affected by both over- and under-reporting. One strength of the study in this regard, however, is that the deprivation index at least is based on register data. Another limitation associated with the data is the lack of information on the students who were absent from school at the time of the data collection, or who have not answered certain of the questions for various reasons. As in all studies that employ self-report data to study social problems, it is reasonable to assume that students characterized by such problems are over-represented among those with missing data. At the same time, an analysis commissioned by the City of Stockholm to examine the significance of missing data at the school level for the results of the survey found that the schools that chose not to participate primarily reported that small student numbers constituted the principal reason for their non-participation (Eklund and Fritzell, 2014).
Agnew’s (1999) MST is a comprehensive, integrated theoretical perspective. As has been noted earlier, previous tests of MST have been forced to exclude certain important aspects of the theory. This study has been able to investigate some of these, for example the school-level measure of deprivation and the individual-level conditional mechanisms. Owing to data limitations, however, some central aspects of MST have been ignored even in the current analysis. Given the focus on the theory’s macro-level implications, the most important mechanisms in this regard are perhaps perceived injustice and relative deprivation. 8 The concept of relative deprivation has found support in a study of Icelandic schoolchildren, which found that the effect of economic deprivation on offending is contingent on the standard of living in the students’ reference groups (Bernburg et al., 2009). Based on longitudinal individual-level data, Alm and Estrada (2018) have found that subjective perceptions of deprivation in adolescence are related to offending later in life among less privileged individuals. Since the current study includes a measure only of absolute deprivation, future school-contextual analyses in this field should also include a measure of subjective deprivation, for example via students’ perceptions of their opportunities relative to those of others. Another finding in the current study was that the contextual effect of meaninglessness became statistically significant only when control for school-level deprivation was included in the analyses. Although an in-depth analyses of this possible suppressor effect of deprivation on the relationship between school context and offending was out of the scope for the current study, it may be a venue for future research that could take the theory’s macro-level implications a step further.
Footnotes
Acknowledgements
I would like to thank Susanne Alm, Olof Bäckman and two anonymous reviewers for their constructive comments on earlier versions of this manuscript. I also would like to thank Dave Shannon for reviewing the language of the article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by the Social Affairs Administration in the City of Stockholm.