The wider social environment and changes in self-reported quality of life in the transition from late childhood to early adolescence: a cohort study

Research design

Maastricht is a small city located in the extreme south of the Netherlands (122 000 inhabitants) with a predominantly white population. The present study aimed to follow a cohort of 1007 adolescents living in the 36 Maastricht residential neighbourhoods and attending the same grade [3]. At baseline and follow-up, adolescents and their parents were asked to fill in a quality of life questionnaire that also included individual and family confounding variables. All adolescents meeting inclusion criteria, rather than baseline responders only, were asked to fill-in the follow-up questionnaire. The present paper reports on the follow-up measurement two to three years after baseline (2002/2003).

The study was been approved by the Medical Ethics Committee of the Maastricht University Hospital (AZM). Both at baseline and at follow up, parents were asked to sign an informed consent form conform legal requirements in The Netherlands.

Neighbourhood variables

The measure of NSD was based on various neighbourhood socioeconomic characteristics obtained from the Maastricht Statistics Department and Statistics Netherlands (CBS) [3]. To summarise these data an exploratory factor analysis (principal factors without rotation) was carried out. Two identified factors explained 70.0% of the total variance [3]. Percentage single parent families, ethnicity, non-voters, unemployment, unemployment more than 1 year, social security, social security more than 3 years, mean income, mean income for persons employed 52 weeks a year, percentages high and low incomes, and percentage economically inactive loaded on the first factor NSD. Single persons and various mobility variables loaded on 'residential instability'. Regression factor scores were calculated for NSD, yielding a continuous variable with mean 0 and unity standard deviation. Higher scores indicated more socioeconomic disadvantage. This variable had a normal distribution.

In order to assess social capital, approximately 200 adult inhabitants from each of the 36 Maastricht neighbourhoods, aged 20 to 65 years, were randomly selected, using the municipal database. Forty-eight percent of the 7236 selected adults responded (hereafter: community survey). These inhabitants received a questionnaire on social capital, which they were asked to send back. Social capital was measured using the two collective efficacy scales: informal social control (ISC) and social cohesion and trust (SC&T), developed by Sampson and colleagues [15]. The two sum scores were obtained from individual answers in the community survey and aggregated to the neighbourhood level [3]. The ISC scale measures the willingness to intervene in hypothetical neighbourhood-threatening situations, for example in the case of children misbehaving. The SC&T scale measures bonds and trust among neighbourhood residents. These data were collected separately and the neighbourhood-level variables can be matched with all Maastricht data collections, such as the family cohort studied in the present paper.

The validity of ISC and SC&T was demonstrated in a selection of Chicago neighbourhoods representing the whole city [14, 15]. Collective efficacy was also associated with mortality in Hungary [21] and with child mental health in Chicago [22]. The construct and concurrent validity was also examined at the Maastricht study site: both informal social control and social cohesion and trust were highly associated with NSD [3] (construct validity) and with measures of social contacts and cosiness as described in [23] at the neighbourhood level (concurrent validity, unpublished results: Spearman's rho between 0.60 and 0.91, p < 0.001, n = 36). The concurrent validity of social cohesion and trust was also tested at the individual level (unpublished results: Spearman's rho 0.56 and 0.53, respectively, p < 0.001, n = 3047).

The three neighbourhood variables were standardized to mean zero and unity standard deviation. Higher scores indicated more NSD and lower levels of social capital [3].

Adolescents' health-related quality of life

Both baseline and follow-up questionnaires of adolescents consisted mainly of the Child Health Questionnaire (CHQ) child form (87 items) [24]. The CHQ-subscales general health, mental health, and self-esteem were selected for the analyses [3]. Furthermore, items on satisfaction (including amongst others: satisfaction with relationships, with friends and teacher, school performance, leisure activities) were added to the questionnaire and a satisfaction scale was constructed combining self-esteem and satisfaction items [3]. Because this scale was constructed in the same way as the original CHQ scales [24], general health, mental health, self-esteem, and satisfaction scales could all range from 1 to 100 and were all positively scored, higher scores indicating better outcomes.

Individual and family variables

Family socioeconomic status was assessed using occupational status and educational status. Occupational status was measured using the current or last profession of the parents, and scored according to the International SocioEconomic Index of occupational status ISEI-92 [25]. Parent questionnaires also assessed the highest level of completed education. Family occupational and educational status were based on the parent with the highest score. In order to ensure control for family level processes, a variable measuring the quality of child-parent interaction at baseline was included in the models as the family-level equivalent of neighbourhood social capital. This variable, parental perceived difficulty (in child raising), was measured using the NOSIK (Nijmegen Parental Stress Index Short Version), a Dutch 25 item questionnaire (items such as "I have much more problems raising my child than expected", and "I notice that I am less able to take care of my child than expected") [26]. Sum scores of the 11 items of the parent domain were used in the present analyses.

Statistical analysis

All analyses were performed using STATA (version 7/SE) [27]. Hierarchically structured data were subjected to multilevel regression analysis [28] in order to investigate neighbourhood effects while controlling for individual effects. Multilevel or hierarchical linear regression techniques are a variant of the more often used unilevel linear regression analyses and are ideally suited for analysis of clustered data, in this case consisting of multiple persons clustered within a single neighbourhood. The βs are the regression outcomes of the predictors in the multilevel model and can be interpreted identically to the estimates in the unilevel analyses.

Regression models analysing quality of life (general health, mental health, self-esteem, or satisfaction) included baseline quality of life, parental occupational status, parental educational status, parental welfare recipient status, single parent family status, child's gender, grade retention (age), and parental perceived difficulty (NOSIK). Occupational status, parental perceived difficulty, and parental educational status were entered as dummy variables in the equation, high occupational status, high educational status, and low perceived difficulty being the reference categories. The above-mentioned individual and family variables were all individual-level variables in the analysis because only one child per family was included in the cohort. Neighbourhood-level variables were NSD, ISC, and SC&T (included separately).

Two a priori interaction terms were added to the models: NSD * individual socioeconomic status and NSD * SC&T. When results suggested interaction, two methods were used to further clarify the dynamics of the interaction. (1) If one of the interacting variables was educational status, analyses were performed stratified by combined categories (university or higher vocational, higher secondary or intermediate vocational, lower secondary or elementary). (2) If SC&T was the interacting variable, the model including the interaction term was used to calculate effects of NSD for very low social cohesion and trust neighbourhoods (i.e. SC&T variable-2SD), low social cohesion and trust (i.e. SC&T variable-1SD), average social cohesion and trust (SC&T variable), high social cohesion and trust (SC&T variable+1SD), and very high social cohesion and trust (SC&T variable+2SD), respectively.

Descriptives and correlations

Associations between neighbourhood factors and changes in general and mental health

Neither crude analyses nor analyses controlling for confounders (see statistical analyses) showed large or statistically significant associations between neighbourhood factors and changes in general or mental health (data not presented).

Associations between neighbourhood variables and changes in self-esteem and satisfaction

Individual-level variables and outcomes

Baseline values of general health, mental health, self-esteem, and satisfaction were all highly associated with their respective follow-up variables. All quality of life outcomes were lower in girls and in families with more perceived difficulty. Adolescents from single parent families reported lower levels of general health, self-esteem, and satisfaction, after controlling for baseline values. Lower parental educational status was negatively associated with changes in general health, albeit non-linearly – and it was positively associated with changes in satisfaction (data not presented).

Methodological issues

Social capital is an umbrella term including many different constructs [37]. Only two of these constructs were included in the present study (informal social control; social cohesion and trust). These two scales were selected because these were the best validated measures at the time and the scales were used in a large cohort study in Chicago, with a very similar design, so that we could compare effects [38]. It is possible that analysing other constructs would yield different results. These previous analyses also showed that in Maastricht, ethnicity is not associated with quality of life outcomes [38]. Therefore, ethnicity was not included as a confounder in the present analyses. Because the Maastricht population is predominantly white, it may not be possible to extrapolate findings to larger and ethnically more diverse cities. However, for European small cities and towns like Maastricht, that typically have a low proportion of non-Western immigrants [39], the results may provide a public health perspective. In addition, a study in a relatively homogeneous population avoids difficulties of studying ethnically heterogeneous populations, such as language problems and cultural differences. More research in ethnic minority groups and ethically diverse populations can give more insight in the external validity of the present findings. Results regarding self-esteem and satisfaction may be stronger in ethnically more diverse populations because minority children living in a dissonant environment were reported to have lower levels of self-esteem than minority children from segregated but protected environments [30] (see above).

The strength of the present study is its longitudinal design that enables the prospective investigation of changes in quality of life in the transition to early adolescence. Furthermore, a principle objective of our methodology was to examine effects of neighbourhood variables that were obtained independently of the responding adolescents. Because perceptions of social capital are always biased by individual mental health status, it is difficult to disentangle cause and effect. The purpose of studying more distal mechanisms constituting objective social capital was realized by measuring social capital scale items in a group of informants that was different than the cohort investigated [40]. However, although both ISC and SC&T were measured independently of the study sample, answers of all informants are coloured by their individual characteristics [41]. On the other hand, individual socioeconomic and demographic composition provide the basis for social interactions in a neighbourhood and, therefore, controlling for individual characteristics leads to over adjustment [41].

The present paper has some limitations. First, in a longitudinal study there is always loss to follow-up. Although 79% of the baseline responders also responded at follow-up, which is relatively high, parental educational status differed between those who dropped out after baseline and those who responded to the follow-up questionnaire (t-test, p = 0.01). However, it is unlikely that this impacted on the results, as results (table 3 and 4) were stratified by parental educational status because of interaction between individual and neighbourhood socioeconomic status. This selective non-response could have led to a decreased power in the stratum of adolescents with low parental education, but table 3 shows greatest effects in this group.

Second, the response rate in the social capital community survey in adults was only 48% [3]. However, the community sample respondents and the general population between 20 and 65 years of age do have similar distributions in age, gender and ethnicity. Furthermore, all respondents were considered to be "key" informants about their own neighbourhood, with the implicit assumption that responders gave the same information about the neighbourhood as the non-responders would have given. The validity of the sample might have been judged differently if the principle objective was to obtain information on the person, not his or her neighbourhood. Thus, this information is more or less independent of the response rate. In order to verify this assumption, we examined post hoc associations between ISC and SC&T collected in the family cohort (parents), and those collected in the community survey (reproducibility). Neighbourhood scores on ISC and SC&T based on these questionnaires were highly correlated.

Inclusion of educational status and occupational status guarantees satisfactory control for individual level socioeconomic status in the Netherlands [42]. However, the possibility remains that residual confounding may have lead to spurious results at the neighbourhood level, because of omitted variable biases [1]. Families moving into poor or not moving out of poor neighbourhoods may differ from their peers although equally poor or affluent (e.g. in motivation, literacy etc). Smoking and obesity are factors that are associated with neighbourhood of residence and can hypothetically influence health outcomes [43–45]. Because at baseline none of the adolescents smoked, post hoc we repeated the analyses including obesity only as an extra confounder. Results were very similar, but effects in self-esteem in the middle stratum of parental education were somewhat stronger and statistically significant. In addition, although physical activity may be associated with health and quality of life outcomes, this measure was not included in the present study. Physical activity may be more easily obtainable or attractive in advantaged neighbourhoods, because of the neighbourhood environment or the presence of better equipped facilities. Given the fact that adolescents' quality of life in advantaged neighbourhoods as a result of more and better sports facilities, controlling for physical activity would result in smaller effects. Therefore, it is highly unlikely that physical activity is the reason that we did not find an effect of neighbourhood variables on quality of life. On the other hand, the problem of unhealthy reductions in physical activity tends to increase only in late adolescence [46], while our study ended in young adolescence. In addition, it is not likely that it impacts our main finding: the interaction between individual socioeconomic status and neighbourhood socioeconomic disadvantage.

Furthermore, none of the models showed statistically significant variance at the neighbourhood level (σ_μ ²), and intra class correlations (ρ) were low. Theoretically, variance at each level warrants including that level in the analyses [28]. However, neighbourhood researchers tend to analyse neighbourhood effects, even when intra class correlations and neighbourhood variation are low, and it is generally held that this is warranted [47]. In addition, in line with low neighbourhood-level variance, results showed no main effects of any of the neighbourhood variables. This does not rule out hypothesized interaction effects: neighbourhood-level variables were associated with outcomes in subgroups of adolescents.

The main outcomes of our study were quality of life and mental health and, therefore, the CHQ was included in the research instruments. CHQ-subscales are all continuous variables. Some prefer the use of dichotomous health outcomes. However, dichotomization results in loss of information and was not necessary here. This could reduce the comparability of the results to studies that did use dichotomous outcomes, but in neighbourhood research both dichotomous and continuous outcomes have been studied. Although the CHQ is a comprehensive instrument, the number of items per psychological domain is relatively low compared to psychological questionnaires like the CBCL. Therefore, the questionnaire is suited for research in the general population. However, in order to enable studies to address multiple research questions after a single data collection, one may prefer to further reduce the number of items. For general health, there is a widely-used and validated one-item alternative: "How do you perceive your health?" (answers on a 5-item Likert type scale: 1 excellent, 2 very good, 3 good, 4 fair, or 5 poor) [38]. This question (in Dutch) as well as one question on psychological problems (yes/no) are included in a new Maastricht data collection. Both concurrent (e.g. with the strengths and difficulties questionnaire, SDQ) and predictive validity will be studied in the future, using, amongst others, matching procedures with the psychiatric case register that records psychiatric service consumption. More research is needed to find and validate one-item alternatives for the self-esteem and satisfaction questions.

Associations between NSD, informal social control, and social cohesion and trust were so strong that collinearity problems would likely have arisen had these three variables been entered jointly in one regression model. Therefore, all neighbourhood variables were entered in the models separately, except when analysing interaction effects between two neighbourhood variables.

Finally, a previous study in another Dutch city on changes in behavioural problems between the age of 11 and 13 years, showed a statistically significant association between NSD and only one of the six behaviour outcomes [4]. Because all changes in behaviour were in the expected direction, the authors proposed that a longer follow-up period could reveal statistically significant changes. Future data collections with longer follow up periods may reveal more associations between neighbourhood factors and changes in general and mental health, and associations with self-esteem and satisfaction (in subgroups) could be replicated.