Adverse life events, area socio-economic disadvantage, and adolescent psychopathology: The role of closeness to grandparents in moderating the effect of contextual stress

Abstract

The study, using data from 801 11–16-year-olds clustered in 68 schools across England and Wales, tested whether closeness to grandparents moderates the association between contextual stress and adolescent psychopathology and prosocial behavior, measured with the strengths and difficulties questionnaire (SDQ). Contextual stress was measured at both school area level (assessed with the index of multiple deprivation) and child level (assessed, as life stress, with the number of proximal and distal adverse life events experienced). At baseline, area stress (multiple deprivation) was unrelated to psychopathology (SDQ), and although both proximal (during the last 12 months) and distal (before the last 12 months) life stress was associated with broad and specific child psychopathology, the association with proximal life stress was stronger. Closeness to the most significant grandparent moderated both the effect of proximal life stress on hyperactivity and broad psychopathology, and the effect of the interaction between distal and proximal life stress on broad and externalizing psychopathology. These findings suggest that the role of grandparents deserves further attention in future investigations of the development of resilience in youth.

Keywords::

• Adolescents
• area stress
• intergenerational relations
• life stress
• psychopathology
• SDQ

Introduction

It is now well established that life stressors do not occur in isolation, and that it is the combination of various stressors rather than any single stressor that portends negative child outcomes (Rutter 1979; Sameroff et al. 1993; Tiet et al. 1998; Ackerman et al. 1999, 2004; Burchinal et al. 2000; Evans 2003; Atzaba-Poria et al. 2004; Flouri and Kallis 2007). However, studies have yet to deal with several issues with respect to the modeling of cumulative stress effects on child psychopathology. First, research has not yet established convincingly the effect of the timing of cumulative stress (Burchinal et al. 2008). Second, with few exceptions (Simmons et al. 1987; Gerard and Buehler 2004; Flaherty et al. 2006; Morales and Guerra 2006; Flouri and Kallis 2007), studies do not usually examine the functional form of cumulative stress effect, with consequences for both theory development and intervention design. To illustrate, there is evidence for a linear effect whereby increments in stressors have a steady, additive impact on mental health problems in children (Deater-Deckard et al. 1998). As few researchers actually report whether their investigations included appropriate tests for nonlinear patterns of cumulative stress, this ignores the possibility of a nonlinear relationship that might, for instance, manifest itself as an acceleration (Simmons et al. 1987) or a leveling off (Morales and Guerra 2006) of problems at a critical level of stress. Third, despite the renewed interest in the role of neighborhoods in children's development (Leventhal and Brooks-Gunn 2000), which has rekindled the debate since Durkheim (1951), as to whether contextual stress should be operationalized at the family/individual or the area/neighborhood level, relatively few studies (McCulloch and Joshi 2001; Flouri et al. 2010) actually compare the effects of these two levels of contextual stress. Fourth and finally, it still needs to be established if there is an interaction between levels of stress, as it is possible that the effect of one level of stress on psychopathology is conditional upon the value of another.

What is more, only few of the studies modeling cumulative stress effects on child psychopathology have examined factors that protect children exposed to multiple stressors. The dearth of such research is unfortunate as various protective factors have been identified as moderating the impact of specific stressors. These factors are usually grouped under two domains, namely individual attributes and connections to competent and caring adults in the family and the community (Masten 2001; Masten et al. 2004). A connection to an involved grandparent is likely to be one such protective factor (King and Elder 1997) but only recently have studies started to explore the role of grandparents in the adjustment of youth experiencing significant adversity, such as parental separation (Bridges et al. 2007) or poverty (Pittman 2007). This is unfortunate as expecting a close connection to an involved grandparent to buffer the effect of contextual stress on child behavior would be in line with several theories. For example, it is in line with attachment theory (Bowlby 1969) since the argument is that a warm caregiver–child relationship may buffer the effect of stress on child psychopathology. It also draws on applications of social capital theories to population health (Kushner and Sterk 2005), which themselves draw on classic sociological theories (Durkheim 1951) for validation of the protective features of extended networks, social cohesion, and social integration. However, no study has yet examined, using a cumulative stress approach and modeling contextual stress at both individual and area level, the role of a close connection to a significant grandparent as a moderator of the stress/child adjustment association.

This study was designed to address this issue. In doing so, it also extended in several ways prior work on the role of cumulative stress in child psychopathology. First, it assessed cumulative stress appropriately. It used a well-validated measure of multiple stressors at individual level, Tiet et al.'s (1998) adverse life events scale, to measure life stress, and it measured area stress on the basis of area multiple deprivation scores. Measuring contextual stress appropriately is important because the variability in cumulative stressor measurement is often such that it makes comparisons of studies almost meaningless (Grant et al. 2006). Second, it measured both proximal (during the last 12 months) and distal (before the last 12 months) life stress to test for the effect of the timing of adverse life events. Third, it compared area stress with life stress effects. Fourth, it searched for an appropriate functional form of the effect of both area stress and life stress on psychopathology. Finally, it tested for the presence of interaction effects between area stress, proximal life stress, and distal life stress on psychopathology.

Methods

Participants and procedure

Secondary school age children in England and Wales, i.e. boys and girls aged 11–16 years (school years 7–11 years), were recruited by a survey company (GfK NOP) from schools drawn from the School Government Publishing Company list using probability proportionate to size sampling (i.e. the number of pupils per school was examined to ensure that larger schools had a greater chance of being selected). One year per school was randomly selected and within that year one class was also randomly selected. All of the pupils in the selected class were surveyed. Of the 103 schools, 70 schools in England and Wales returned questionnaires, resulting in a sample of 1566 secondary school-aged children. The study sample size (obtained listwise omitting missing data in all the response variables and covariates1) was 801 children, clustered in 68 schools. The number of children per school ranged from 1 to 24.

Permission to carry out the study was obtained from the University of Oxford Ethics Committee. Advance letters detailing the purpose of the study were sent to the sampled schools. Included in the letter was a return slip for the school to complete to indicate its willingness to take part in the study, and, if willing, to provide a contact name of the person at the school who would be in charge of arranging the survey process. Parents were told they could withdraw their children if they did not want them to take part. The adolescents provided informed consent to take part in the survey. They were made aware that they were free to withdraw from the study at any time for any reason. Confidentiality was ensured to all participants.

Measures

Area stress was measured with the index of multiple deprivation (IMD), a weighted area-level aggregation of specific dimensions of deprivation. While the methodologies used to construct the different UK countries' indices of multiple deprivation are conceptually similar, separate indices of deprivation have been developed in acknowledgment of the differences among UK countries. For example, the dimensions of deprivation used to construct the English IMD 2004 were (1) income deprivation, (2) employment deprivation, (3) health deprivation and disability, (4) education, skills, and training deprivation, (5) barriers to housing and services, (6) living environment deprivation, and (7) crime (Noble et al. 2006). The dimensions to construct the Welsh IMD were income deprivation, employment deprivation, health deprivation and disability, education, skills and training deprivation, barriers to housing and services, housing deprivation, and physical environment deprivation. In this study, for each child, the school's postcode was used to identify the school's area, and the IMD score associated with that particular area was used to assess the area stress. As the English IMD and the Welsh IMD are not comparable, in this study IMD ranks were used, which, for the purposes of this study, were standardized.

Life stress was assessed with Tiet et al.'s (1998) adverse life events scale, which in this study measured both proximal life stress (number of adverse life events experienced in the last year) and distal life stress (number of adverse life events experienced before the last year). This scale is composed of 25 possible events for which children had little or no control over (e.g. “someone in the family died”, “negative change in parents' financial situation”), and is a modification of the life events checklist (LEC; Coddington 1972a, 1972b; Brand and Johnson 1982), which has acceptable validity and test–retest reliability (Brand and Johnson 1982). The LEC is a measure of exposure to potentially traumatic events developed at the National Center for Posttraumatic Stress Disorder (PTSD) to facilitate the diagnosis of PTSD.

Psychopathology and prosocial behavior were assessed with the self-report version of the strengths and difficulties questionnaire (SDQ), a 25-item 3-point (ranging from 0 to 2) scale measuring four difficulties (hyperactivity, emotional symptoms, conduct problems, and peer problems), as well as prosocial behavior (Goodman 1994, 1997). Each subscale has five items. A total difficulties (broad psychopathology) score is calculated by summing the scores for externalizing (i.e. hyperactivity and conduct problems) and internalizing (i.e. emotional symptoms, and peer problems) difficulties (www.sdqinfo.com). Cut-off scores for the borderline/abnormal range (the SDQ cut-off score identifies 20% of the population) are 16 for total difficulties, 6 for emotional symptoms, 4 for conduct problems, 6 for hyperactivity, 4 for peer problems, whereas the borderline/abnormal range for prosocial behavior is 0–5 (www.sdqinfo.com). The SDQ has been extensively evaluated and applied in UK and abroad (Hawes and Dadds 2004; Woerner et al. 2004). Internal consistency (Goodman 2001), test–retest reliability (Goodman 1999), and concurrent and discriminant validity (Goodman et al. 1998; Goodman and Scott 1999) are excellent.

Closeness to grandparents was measured with Elder and King's (2000) scale of grandparent–grandchildren relationship quality. Children indicated on 4-point scales (ranging from 1 to 4) the extent to which they could depend on their grandparent, the extent to which they felt appreciated, loved or cared for by their grandparent, the extent to which the grandparent helped them in significant ways, the extent to which they perceived happiness in their relationship with their grandparent, and the extent to which they were close compared to other grandchildren to the grandparent. Children completed the scale for each living grandparent. The grandparent who received the highest scale score was identified as the most significant grandparent. Additional criteria, as suggested and used by Elder and King (2000), were used to identify the most significant grandparent if more than one significant grandparent emerged. These criteria were frequency of contact (the grandparent that they saw or talked to most was rated as the most significant), gender of grandparent (the same-sex grandparent was rated as the most significant), and lineage (maternal grandmother was chosen first, followed by maternal grandfather and then paternal grandparents).

Contextual and structural factors were the well-known child psychopathology correlates of poverty (measured by free school meals eligibility), of special educational needs, and of academic achievement (measured with the children's results in standard attainment tests (SATs)) taken at the end of primary school. SATs are statutory tests in the core subjects of the National Curriculum. At the end of primary school (Key Stage 2), children have to take national tests in English, Mathematics, and Science. There are certain levels which children should attain. According to the UK's Department for Children, Schools, and Families, the expected level for children in England and Wales at the end of Key Stage 2 is 4. As the study sample was in secondary schools in England and Wales, Key Stage 2 SATs results could be obtained from all pupils.

Results

Bias analysis and descriptive statistics

Comparing the study sample with the full sample, we found significant differences on the observable characteristics. As Table I shows, the study sample was generally less disadvantaged (the mean level attained in all three subjects was higher than average, and only a small percentage had been eligible for free school meals at least at some point during their school years) and more well adjusted than the whole sample. In all, of the study sample, 18.9% were in the borderline/abnormal range for total difficulties, 16.9% for prosocial behavior, 22% for hyperactivity, 14.1% for conduct problems, but only 8.9% for emotional symptoms and only 5.2% for peer problems. As Table II shows, closeness to grandparent was not related to contextual stress or to broad or specific psychopathology, although it was positively associated with prosocial behavior.

Table I.  Descriptive statistics and bias analysis results.

	Full sample		Study sample
Variables	Mean or %	SD	Mean or %	SD
Total difficulties	12.72	5.66	11.65*	5.23
Emotional symptoms	3.18	2.32	3.03*	2.26
Conduct problems	2.85	1.98	2.48*	1.84
Hyperactivity	4.72	2.42	4.45*	2.45
Peer problems	1.96	1.74	1.68*	1.54
Prosocial behavior	7.01	1.94	7.27*	1.86
Number of adverse life events in the last year	4.12	3.34	3.72*	3.03
Number of adverse life events before the last year	5.00	4.09	4.95*	3.97
Standardized IMD^† rank	0.00	1.00	0.01*	1.03
Age	13.37	1.40	13.51*	1.39
No free school meals	80.1%		84.8%*
No special educational needs	88.3%		93.6%*
SATs^‡ in English, Key Stage 2	4.49	0.63	4.54*	0.59
SATs in Mathematics, Key Stage 2	4.50	0.68	4.53*	0.64
SATs in Science, Key Stage 2	4.61	0.62	4.64*	0.60
Wales^¶	6.3%		7.0%
Closeness to grandparent	3.26	0.62	3.28*	0.62
Boy	51.0%		47.7%
N	1023–1566		801

The smallest full sample is for SATs in Science at Key Stage 2 while the largest full sample is for Wales. *Indicates the study sample is significantly different at 5% from the full sample.

^† Index of multiple deprivation

^‡ Standard assessment test

^¶ Child resident in Wales.

Table II.  Descriptive statistics and zero-order correlations.

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18
1. Total difficulties
2. Emotional symptoms	0.65**
3. Conduct problems	0.70**	0.17**
4. Hyperactivity	0.70**	0.13**	0.49**
5. Peer problems	0.48**	0.33**	0.15**	− 0.02
6. Prosocial behavior	− 0.17**	0.16**	− 0.34**	− 0.23**	− 0.03
7. Number of adverse life events in the last year	0.37**	0.20**	0.29**	0.27**	0.18**	0.00
8. Number of adverse life events before the last year	0.19**	0.16**	0.10**	0.14**	0.08*	0.02	0.45**
9. Standardized IMD^† rank	− 0.06	− 0.07	− 0.02	− 0.01	− 0.06	0.03	− 0.04	− 0.03
10. Age	− 0.03	− 0.01	− 0.06	− 0.01	− 0.01	− 0.06	− 0.06	0.16**	− 0.21**
11. No free school meals	− 0.12**	− 0.04	− 0.13	− 0.07	− 0.07	− 0.03	− 0.15**	− 0.15**	0.11**	− 0.04
12. No special educational needs	− 0.17**	− 0.07*	− 0.12**	− 0.10**	− 0.17**	0.01	− 0.19**	− 0.08*	− 0.05	0.13**	0.06
13. SATs^‡ in English, Key Stage 2	− 0.10**	− 0.01	− 0.11**	− 0.09**	− 0.03	0.06	− 0.11**	0.03	0.10**	0.06	0.07*	0.23**
14. SATs in Mathematics, Key Stage 2	− 0.08*	− 0.05	− 0.11**	− 0.07*	0.03	− 0.02	− 0.11**	0.12**	0.06	0.09**	0.06	0.16**	0.48**
15. SATs in Science, Key Stage 2	− 0.09*	− 0.04	− 0.11**	− 0.09*	0.04	− 0.01	− 0.09**	0.12**	0.06	0.03	0.09*	0.11**	0.51**	0.54**
16. Wales^¶	0.05	0.05	0.02	− 0.01	0.11**	− 0.05	− 0.04	− 0.07*	− 0.43**	0.07*	0.03	0.03	0.02	0.00	0.00
17. Closeness to grandparent	0.02	− 0.02	0.02	0.02	0.02	0.13**	− 0.03	− 0.06	0.02	− 0.09*	− 0.05	0.02	− 0.04	− 0.08*	− 0.04	0.01
18. Boy	− 0.04	− 0.27	0.12**	− 0.00	0.14**	− 0.31**	− 0.05	− 0.00	0.12**	− 0.05	0.01	− 0.09*	− 0.03	0.17**	0.06	− 0.03	− 0.10**
Skewness	0.34	0.61	0.78	0.09	1.21	− 0.68	1.54	1.06	− 0.07	− 0.04	− 1.94	− 0.358	− 0.37	− 0.81	− 0.80	3.38	− 1.18	0.09
Kurtosis	0.01	− 0.22	0.50	− 0.73	1.96	0.41	4.45	1.31	− 1.14	− 0.91	1.76	10.85	− 0.32	1.02	1.58	9.45	1.40	− 2.00
Min–max	0–30	0–10	0–10	0–10	0–10	0–10	0–24	0–25	− 1.63–1.53	10.00–16.00	1.00–2.00	1.00–2.00	3.00–6.00	1.00–6.00	1.00–6.00	1.00–2.00	0.40–4.00	0.00–1.00

*p < 0.05; **p < 0.01; p < 0.001

^† Index of multiple deprivation

^‡ Standard assessment test

^¶ Child resident in Wales.

Data analytic strategy

Our data have a nested structure as children are clustered within schools. We used MLwiN, a software that can account for the hierarchical structure of the data. To define the multilevel model for broad psychopathology (total difficulties), we used the notation in Goldstein (2003). In particular, the univariate response model was defined as,where Y denotes the vector, with elements y_ij, of responses (total difficulties) for child i clustered in school j, X is the design matrix with elements x_ij, B is the vector of regression parameters to be estimated, u is a vector, with elements u_j, of school random effects, and ϵ is a vector, with elements ϵ_ij, of child level residuals. To test the effect of contextual stress and its interaction with closeness to grandparent on the four difficulties and on prosocial behavior of the SDQ (specific psychopathology), a multivariate response multilevel model that allowed the error terms of the five different models to be correlated was fitted in MLwiN. This was because each child provided responses for each of the five SDQ scales, so responses were likely to be correlated. Ignoring this correlation by modeling each of the five SDQ scales separately may, therefore, lead to erroneously estimated SEs for the regression coefficients, and so to erroneous statistical inferences. Modeling specific psychopathologies in a multivariate (simultaneous) way also offers a flexible modeling framework, as it can accommodate different covariates for the different SDQ scales as well as allow for the correlation between unobserved factors affecting scores on the different SDQ scales. The multivariate response multilevel model was defined as,

The * superscript is used to denote multivariate vectors and matrices. In particular, Y^* denotes the vector, with elements y_ijl, of responses for children i clustered in schools j and SDQ scale l, X^* is the design matrix with elements x_ijl, B^* is the vector of the regression parameters—of the multivariate model—to be estimated, u^* is a vector, with elements u_jl, of school random effects for each SDQ scale l and ϵ^* is a vector, with elements ϵ_ijl, of child-level residuals in school j and SDQ scale l. In model (2), with denoting the variance covariance matrix between the level 1 error terms, and with denoting the variance covariance matrix between the level 2 error terms.

Multilevel models

We first estimated an empty two-level model for total difficulties (see Table III for the model summary for broad psychopathology), which showed that the average total difficulties score as reflected in the intercept was 11.732 (SE = 0.241). The child-level variance was 25.876 (SE = 1.346), and the variance due to differences in schools was 1.513 (SE = 0.650). This suggests a between-school variation in total difficulties. We first investigated the association between contextual stress and total difficulties by running three baseline models adjusting only for country. Area stress was not related to total difficulties, and although both distal and proximal life stress were associated with total difficulties, the association between total difficulties and proximal life stress (b = 0.631, SE = 0.057) was stronger. Next, the full model (Model 2) was introduced. This added to the empty model the following variables: country, proximal life stress, distal life stress, area stress, age, gender, free school meals, special educational needs, SATs in English, Mathematics, and Science, and closeness to grandparent. The effect of proximal life stress became slightly smaller (b = 0.547, SE = 0.066) but remained statistically significant at the 1‰ level. Of all the other variables entered, only special educational needs (b = 2.002, SE = 0.738) had a statistically significant effect on total difficulties. The difference in the − 2 log likelihood of the full random intercepts model described above and the full fixed-effects model was not significant (the one-sided test gave a p-value of p = 0.151). This suggests that there was no significant between-school variation in total difficulties after accounting for individual differences in the variables entered in the model.

Table III.  Model summary (broad psychopathology).

Model	Specification
Model 1	Area random effect
Model 1.1	Model 1+country+proximal life stress
Model 1.2	Model 1+country+distal life stress
Model 1.3	Model 1+country+area stress
Model 1.4	Model 1.3+proximal life stress+distal life stress
Model 2	Model 1.4+age+gender+free school meals+special educational needs+SATs in English+SATs in Mathematics+SATs in Science+closeness to grandparent
Model 2.1	Model 2+quadratic effect for area stress
Model 2.2	Model 2+quadratic effect for distal life stress
Model 3	Model 2+quadratic effect for proximal life stress
Model 3.1	Model 3+interaction between proximal life stress and area stress+interaction between proximal life stress squared and area stress
Model 4	Model 3+interaction between proximal life stress and distal life stress+interaction between proximal life stress squared and distal life stress
Model 4.1	Model 3+interaction between distal life stress and area stress
Model 5	Model 4+interaction between closeness to grandparent and proximal life stress+interaction between closeness to grandparent and proximal life stress squared+interaction between closeness to grandparent and the interaction between proximal and distal life stress+interaction between closeness to grandparent and the interaction between proximal life stress squared and distal life stress

SATs, standard attainment tests.

Next, we established the appropriate functional form of stress effect on broad psychopathology by introducing quadratic terms (one at a time) for area stress, distal life stress, and proximal life stress in the full model. Only proximal life stress squared was significant (b = − 0.018, SE = 0.009), which suggests that proximal life stress exhibited a quadratic-like effect that increased and then decreased across the range of observed values. Next, we tested if the effect of proximal life stress and, since it was significant, proximal life stress squared on total difficulties was conditional upon the value of distal life stress and area stress. Specifically, we calculated proximal life stress × area stress and proximal life stress squared × area stress interaction terms, and entered these in the model that added to Model 2 the quadratic term for proximal life stress (Model 3), but the interaction terms were neither jointly nor individually significant. Next, we added to Model 3 the proximal life stress × distal life stress and the proximal life stress squared × distal life stress interaction terms (Model 4), and found that these interactions were both jointly (χ²(2) = 8.218, p = 0.016) and individually (b = − 0.049, SE = 0.024, and b = 0.003, SE = 0.001, respectively) significant. This suggests that the effect of proximal life stress and proximal life stress squared on broad psychopathology depended on the level of distal life stress. Finally, we added to Model 3 the interaction between area stress and distal life stress, which was, however, non-significant.

To explore if closeness to grandparent moderated the effect of proximal life stress on broad psychopathology we calculated and entered in Model 4 the following interaction terms: closeness to grandparent × proximal life stress, closeness to grandparent × proximal life stress squared, as well as (given the significant interactions above) closeness to grandparent × the interaction between proximal and distal life stress, and closeness to grandparent × the interaction between proximal life stress squared and distal life stress. This model (Model 5) showed that all the interactions were both jointly (χ²(4) = 10.811, p = 0.029) and individually (b = 0.911, SE = 0.311; b = − 0.080, SE = 0.024; b = − 0.057, SE = 0.026; and b = 0.006, SE = 0.002, respectively) significant. This suggests that closeness to the most significant grandparent moderated the effect of how distal life stress impacted on the effect of proximal life stress on psychopathology, and buffered the effect of proximal life stress on broad psychopathology. For example, as Figures 1 and 2 show, the association between proximal life stress and broad psychopathology was weaker in children reporting very high levels (75th percentile; N = 121) of closeness than in children reporting very low levels (15th percentile; N = 137) of closeness. The zero-order correlation between broad psychopathology and proximal life stress for high and low levels of closeness to the most significant grandparent also shows that proximal like stress was more strongly related to broad psychopathology when closeness was low (r = 0.47, p < 0.001; N = 137) than when closeness was high (r = 0.24, p < 0.01; N = 121).

Figure 1  The association between proximal adverse life events and total difficulties in children reporting high levels of closeness to grandparents (N = 121). r = 0.24, p < 0.01; compared with stronger correlation in Figure 2.

Figure 2  The association between proximal adverse life events and total difficulties in children reporting low levels of closeness to grandparents (N = 137). r = 0.47, p < 0.001; compared with weaker correlation in Figure 2.

Multivariate response multilevel models

We then replicated these analyses on the four difficulties (hyperactivity, emotional symptoms, conduct problems, and peer problems) and prosocial behavior, by fitting in MLwiN a multivariate response regression model that allowed the error terms of the five models to be correlated and that included random school effects. This resulted in a two-level multivariate response model which is treated in MLwiN as a three-level model (with the responses treated as an additional lower level).

First, we ran an empty multivariate response multilevel model. The variance partition coefficients obtained showed that 2% of the total variation in emotional symptoms, 5.3% of the total variation in conduct problems, 3.5% of the total variation in hyperactivity, 3% of the total variation in peer problems, and 3.5% of the total variation in prosocial behavior was explained by children clustering within schools. This indicates that using a multilevel approach to model our data was appropriate. We ran three baseline models (Models 6–8) adjusting only for country. Model 6 examined the effect of proximal life stress, which was found to be significant in predicting emotional symptoms (b = 0.156, SE = 0.026), conduct problems (b = 0.175, SE = 0.021), hyperactivity (b = 0.209, SE = 0.028), and peer problems (b = 0.094, SE = 0.018), but not prosocial behavior (b = 0.004, SE = 0.022). Model 7, which examined the effect of distal life stress, also showed significant, albeit weaker, effects on emotional symptoms (b = 0.090, SE = 0.020), conduct problems (b = 0.043, SE = 0.016), hyperactivity (b = 0.085, SE = 0.022), peer problems (b = 0.032, SE = 0.014), but not prosocial behavior (b = 0.009, SE = 0.017). Model 8 examined the effect of area stress which was found to be non-significant.

Second, the full model (Model 9) was introduced. This added to Model 6, the following variables: distal life stress, area stress, age, gender, free school meals, special educational needs, SATs in English, Mathematics and Science, and closeness to grandparent (Table IV). The effect of proximal life stress remained statistically significant in predicting the four difficulties, but the effect of distal life stress remained statistically significant in predicting only emotional symptoms. Closeness to grandparent was positively related to prosocial behavior even after adjusting for other factors.

Table IV.  Specific psychopathology (coefficients and SEs).

	Emotional symptoms		Conduct problems		Hyperactivity		Peer problems		Prosocial behavior
Variables	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff.	SE
Number of adverse life events in the last year	0.097	0.029	0.169	0.024	0.188	0.032	0.091	0.020	− 0.024	0.024
Number of adverse life events before the last year	0.059	0.022	− 0.015	0.018	0.021	0.025	− 0.008	0.016	0.023	0.018
Standardized IMD^* rank	− 0.034	0.087	− 0.025	0.086	0.016	0.104	− 0.034	0.063	0.052	0.087
Age	− 0.062	0.057	− 0.036	0.054	0.019	0.067	0.020	0.041	− 0.100	0.055
No free school meals	− 0.017	0.215	− 0.473	0.176	− 0.160	0.240	− 0.186	0.150	− 0.109	0.178
No special educational needs	− 0.570	0.325	− 0.234	0.265	− 0.387	0.362	− 0.876	0.226	− 0.220	0.268
SATs^† in English, Key Stage 2	0.056	0.156	− 0.044	0.127	− 0.130	0.174	− 0.057	0.109	0.214	0.129
SATs in Mathematics, Key Stage 2	0.086	0.149	− 0.106	0.121	0.017	0.165	0.068	0.103	0.027	0.122
SATs in science, Key Stage 2	− 0.136	0.159	− 0.085	0.128	− 0.203	0.176	0.177	0.110	− 0.120	0.130
Wales^‡	0.512	0.328	0.171	0.326	− 0.037	0.394	0.687	0.239	− 0.215	0.331
Closeness to grandparent	− 0.156	0.121	0.084	0.099	0.102	0.135	0.111	0.084	0.314	0.100
Boy	− 1.229	0.156	0.556	0.129	0.085	0.175	0.403	0.109	− 1.207	0.131

Statistically significant (p < 0.05) effects are in italics.

^* Index of multiple deprivation

^† Standard assessment test

^‡ Child resident in Wales.

Third, we established the appropriate functional form of a contextual stress effect on specific psychopathology by introducing quadratic terms for contextual stress in the full model. Although the quadratic terms for distal life stress and area stress were not significant in predicting prosocial behavior or any specific psychopathology, the quadratic term for proximal life stress was significant in predicting conduct problems (b = − 0.006, SE = 0.003).

We then tested if the effect of proximal life stress and (since it was significant) proximal life stress squared on specific psychopathology was conditional upon the value of distal life stress and area stress. We calculated proximal life stress × distal life stress and proximal life stress squared × distal life stress interaction terms and entered these in the model that added to Model 9 the quadratic term for proximal life stress (Model 10). This model (Model 11) showed that the proximal life stress × distal life stress interaction term was significant in predicting peer problems (b = − 0.015, SE = 0.007), suggesting that the effect of proximal life stress on peer problems depended on the level of distal life stress. Next, we added in Model 11, the proximal life stress × area stress and proximal life stress squared × area stress interaction terms (Model 12), but these were not significant. Finally, we added to Model 12 the interaction between area stress and distal life stress (Model 13), which was also non-significant.

To explore if closeness to grandparent moderated the effect of proximal life stress on specific psychopathology, we calculated and entered in Model 11 the following interaction terms: closeness to grandparent × proximal life stress, closeness to grandparent × proximal life stress squared, and (given the significant interactions above) closeness to grandparent × the interaction between proximal and distal life stress, and closeness to grandparent × the interaction between proximal life stress squared and distal life stress. This model (Model 14) showed that all the interactions were significant in predicting hyperactivity (b = 0.477, SE = 0.153; b = − 0.032, SE = 0.012; b = − 0.036, SE = 0.013, and b = 0.003, SE = 0.001, respectively). This suggests that closeness to grandparent moderated both the effect of proximal life stress on hyperactivity, and the effect of how distal life stress impacted the effect of proximal life stress on hyperactivity. In addition, closeness to grandparent moderated the effect of proximal life stress squared on emotional symptoms (b = − 0.025, SE = 0.011), and the effect of the interaction between proximal and distal life stress on conduct problems (b = − 0.018, SE = 0.009).

Discussion

This study was carried out to test if closeness to the most significant grandparent moderates the association between contextual stress (proximal life stress, distal life stress, and area stress) with psychopathology and prosocial behavior in a community sample of 11–16-year-old children clustered in schools in England and Wales. Indeed, it showed that closeness to the most significant grandparent moderated the association of proximal life stress with hyperactivity and broad psychopathology. Importantly, closeness to the most significant grandparent also moderated how distal life stress interacted with proximal life stress to predict both broad and externalizing psychopathology.

In testing the moderator effect of closeness to grandparents on the contextual stress/child psychopathology association, the study also extended in several ways prior work on cumulative stress and adjustment in youth. First, assessing with well-validated measures proximal and distal life stress as well as contextual stress at school area level, it showed that although distal life stress did predict broad and specific psychopathology, proximal life stress was a better predictor of broad and specific psychopathology. In line with previous research (Flouri and Kallis 2007), distal life stress became non-significant once proximal life stress was accounted for in predicting broad, and most of specific psychopathology. Area stress was not predictive of either broad or specific psychopathology in this sample, corroborating previous research (Ford et al. 2004), which suggests that factors that impede children's adjustment should be located at the individual/family rather than the area, and neither area stress nor life stress was related to prosocial behavior. Second, by testing for the appropriate form of the effect of life stress on psychopathology, it showed that the association of proximal life stress with broad psychopathology and conduct problems was nonlinear. In modeling the appropriate functional form of the relationship between life stress and child outcomes, this study, therefore, joined the few other studies (Simmons et al. 1987; Gerard and Buehler 2004; Flaherty et al. 2006; Morales and Guerra 2006; Flouri and Kallis 2007) that have tested for “threshold” models of cumulative stress.

The strengths of this study should be seen in light of its limitations. First, selection bias is likely as the study participants had experienced relatively few adverse life events and had done well academically, and a much smaller than expected proportion scored above cut-off for peer problems and emotional symptoms. This may explain the quadratic effect found. Morales and Guerra (2006) suggested in explaining the quadratic effect of cumulative stress found in their disadvantaged sample that perhaps this was evidence of a level of severity in adjustment difficulties beyond which children would be dismissed or failing in school. In our privileged sample, the quadratic effect can be suggested to indicate evidence of a level of severity in adjustment difficulties beyond which children would be targeted for help. Second, not having grandparents was an exclusion criterion for this study, so our findings may not be generalized to children who do not have grandparents. Third, this, as the well-known adverse childhood experiences study (Anda et al. 1999; Dube et al. 2001; Chapman et al. 2004; Whitfield et al. 2005), is a retrospective longitudinal study of adverse life events. Although retrospective designs may suggest possible risk factors for outcomes, the test of the validity of these hypothetical relationships lies in prospective designs (Widom et al. 2004) and experiments (Costello et al. 2004). Reporting of recent adverse events may be highly related to the current state of behavioral or emotional difficulties due only to issues of recency and coloring of recall by current state. Fourth, our models necessarily treated life events as if they were independent events when in fact they may not be. Fifth, our methods have a problem with single source confounding: the SDQ and the adverse life events have the same reporter, the individual youth. This could falsely raise correlations. Sixth, the threat to reliability and validity of using retrospective reporting of life events becomes an increasing problem as the reporting interval lengthens. This suggests that the measure of distal adverse life events used in this study may be problematic. Seventh, some of our correlations, although significant, were quite low. Eighth and finally, as neither closeness to parents nor closeness between parents and grandparents were adjusted for, it is possible that closeness to grandparent might be simply a reflection of the closeness of the whole family unit.

Despite these limitations, this study showed that closeness to the most significant grandparent was positively related to adolescent prosocial behavior but also buffered the effect of proximal life stress on adolescent hyperactivity and broad psychopathology, and moderated the effect of how distal life stress impacted the effect of proximal life stress on adolescent broad and externalizing psychopathology. Taken together, these findings suggest that the role of grandparents deserves further attention in future investigations of children's positive development in general, but also of resilience in the face of adversity.

Acknowledgements

This program of research was supported by an Economic and Social Research Council grant to the first two authors.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.

Notes

¹ If multiple variables are analysed, as is the case here, then listwise deletion removes cases (subjects) if there is a missing value on any of the variables.