It has been suggested that children with positive attributes, such as being affectionate, responsible or generous, may be protected from developing psychopathology. Reference Bromley, Johnson and Cohen1–Reference Gillham, Adams-Deutsch, Werner, Reivich, Coulter-Heindl and Linkins3 If this is true, facilitating the emergence of positive attributes – in addition to, or instead of attempts to reduce psychiatric symptoms – may be a particularly attractive way to promote well-being in youth. Reference Park and Peterson4,Reference Park and Peterson5 However, the evidence-base for such an assertion is still weak in several respects. First, there is little research about the positive attributes of children and adolescents from the general population and no longitudinal studies starting in childhood. Reference Bromley, Johnson and Cohen1,Reference Gillham, Adams-Deutsch, Werner, Reivich, Coulter-Heindl and Linkins3 Second, there is little research into whether positive attributes are merely the flipside of the absence of psychiatric symptoms, with reflectiveness for instance, being a positive attribute that is more or less synonymous with the absence of a psychiatric symptom, namely impulsiveness; yet it is crucial to demonstrate that positive attributes have predictive value across time even when adjusting for psychopathology. Third, most previous research failed to control for factors – including social or family characteristics – that are plausible confounders of the relationship between positive attributes and psychiatric outcomes.
Our aim with this study was to address these shortcomings and test the importance of positive attributes by using data from The 2004 British Child and Adolescent Mental Health Survey (B-CAMHS04), Reference Green6,Reference Parry-Langdon7 a longitudinal multi-informant study of young people in the general population. To date, few studies have employed parent-reported measures of positive attributes. Reference Park and Peterson5,Reference Williams and McGee8 Therefore, we first established how well parents’ ratings of children’s positive attributes can be measured in the general population employing the Youth Strengths Inventory (YSI) from the Development and Well-Being Assessment (DAWBA), Reference Ford, Goodman and Meltzer9,Reference Goodman, Ford, Richards, Gatward and Meltzer10 by analysing its psychometric properties and the interrater correlations. Second, using path analysis we examined whether children’s positive attributes are a construct sufficiently distinct from parental accounts of children’s mental illness. Third, we tested the hypothesis that the report of higher levels of positive attributes at baseline reduces the risk of psychopathology over a 3-year period. We measured outcome using dimensional measures of psychopathology, psychiatric diagnoses as well as psychosocial adjustment indices. Finally, we used propensity scores Reference Austin11–Reference Rosenbaum and Rubin13 to match young people on baseline psychopathology and plausible confounders – to estimate the potential treatment effects that positive attributes might have on the risk for future psychopathology. Propensity score matching is a process that attempts to emulate a randomised controlled trial (RCT). Thus, to reduce selection bias we matched groups with high and low positive attributes on observed covariates, such as psychopathology, family variables and socioeconomic status.
Method
Participants
The B-CAMHS04 involved a sample of 5–16 year olds (n = 7977) representative of the general British population; it has previously been described in detail. Reference Green6 The study used ‘child benefit’ (a state benefit payable at that time in Great Britain for each child in a family) to develop a sampling frame of 5- to 16-year-olds in different postal sectors in England, Wales and Scotland. After excluding families with no recorded postal code, it was estimated that this represented 90% of all British children. Out of the 12 294 contacted, there were n = 1085 who opted out and n = 713 who were non-eligible or had moved without trace, leaving 10 496 who were approached in person. Of those, n = 7977 participated (65% of those selected; 76% of those approached). If the child was aged 5–10, a face-to-face interview was conducted with the parent and if the child was aged 11–16, the parent was interviewed first followed by the young person. In 2007, i.e. 36 months after the baseline survey, Reference Parry-Langdon7 families were approached once more unless they had previously opted out or the child was known to have died. Of the original n = 7977 participants, n = 5325 (67%) participated in the detailed follow-up. Reference Parry-Langdon7 All study procedures received multicentre research ethics committee approval and informed consent was obtained from parents and assent from children participants.
Assessment
Development and Well-Being Assessment
The DAWBA Reference Ford, Goodman and Meltzer9,Reference Goodman, Ford, Richards, Gatward and Meltzer10 is a structured interview administered by lay interviewers who also record verbatim accounts of problems. The questions are closely related to DSM-IV and ICD-10 14,15 diagnostic criteria and focus on current problems. The kappa (κ) statistic for chance-corrected agreement between two raters was 0.86 for any disorder (s.e. = 0.04), 0.57 for internalising disorders (s.e. = 0.11) and 0.98 for externalising disorders (s.e. = 0.02). Reference Ford, Goodman and Meltzer9 Values of κ <0 indicate no agreement, 0–0.20 slight agreement, 0.21–0.40 fair agreement, 0.41–0.60 moderate agreement, 0.61–0.80 substantial agreement and 0.81–1 almost perfect agreement. Reference Landis and Koch16 Children were assigned a diagnosis only if their symptoms were causing significant distress or social impairment The DAWBA was completed at baseline and at 36 months (further information on the DAWBA is available from http://www.dawba.info). This paper focuses on the overall presence of disorder (i.e. any DSM-IV disorder), externalising disorders (the combination of conduct, oppositional defiant and attention-deficit hyperactivity disorders) and internalising disorders (the combination of depressive and anxiety disorders).
Youth Strengths Inventory
The YSI forms part of the DAWBA (Section N) and asks about positive attributes of the child, with similar but not identical parent-report and youth-report versions. The first part of the YSI asks how applicable various descriptions are to the child (for example generous, affectionate, caring), while the second part asks about things the child does that please the parents (for parent report) or that the child is proud of (for self-report). Examples include being good with friends, helpful at home and polite. The parent version has 12 items in each part, whereas the self-report version has 8 items in the first part and 11 items in the second part. Each item is scored on a three-point Likert scale (i.e., no, 0; a little,1; a lot, 2), with scores ranging from 0 to 48 for the parent version and 0 to 37 for the self-report version. Examining the psychometric properties, we found that all YSI scales showed high internal consistencies, similar to other existing strengths measures. Reference Park and Peterson4,Reference Peterson and Seligman17,Reference Epstein, Ryser and Pearson18 For parent report, Cronbach’s alphas were 0.84 at baseline (mean 39.2, s.d.= 6.0, median 40, IQR = 36–44) and 0.86 at 36 months (mean 38.9, s.d. = 6.4, median 40, IQR = 36–44). For child report, they were 0.77 at baseline (mean 27.1, s.d.= 5.1, median 27, IQR = 24–31) and 0.73 at follow-up (mean 28.3, s.d. = 3.7, median 28, IQR = 26–31). A preliminary factor analysis yielded to an unrotated single-factor structure for all informants and time points. For the parent-reported YSI scores, this single factor explained 79% of the variance at both time points. For the youth-reported YSI scores, the variance explained by this single factor was 83% in 2004 and 81% in 2007. It should be noted that the item ‘polite’ was dropped from the factor analysis of youth-reported positive attributes in 2007 because of zero variance (i.e. all children answered ‘A lot’ in this item). Parent-rated YSI scores were used in the analyses unless otherwise specified.
Strengths and Difficulties Questionnaire
The Strengths and Difficulties Questionnaire (SDQ) asks about 25 attributes, some positive and others negative; respondents use a three-point Likert scale to indicate how far each attribute applies to the target child. Reference Stringaris and Goodman19 The 25 items are divided between five scales of five items each, generating scores for emotional symptoms, conduct problems, hyperactivity–inattention, peer problems and prosocial behaviour; all but the last are summed to generate a total difficulties score ranging 0–40. The SDQ also has an impact supplement Reference Stringaris and Goodman19 that asks whether the respondent thinks that the child or youth has a problem, and if so, enquires further about overall distress, social impairment, burden and chronicity. This instrument has robust psychometric properties. Reference Goodman20,Reference Goodman21 In our sample, Cronbach’s alphas for parent reports were 0.84 at baseline (mean 5.9, s.d. = 4.8) and 0.86 at 36 months (mean 5.7, s.d. = 4.6). It is important to note that in this study, the SDQ was used to generate a difficulties score, but not to generate a strengths score: the strengths score used in the analyses reported in this paper was based on the YSI. The SDQ was used in addition to the DAWBA because there is increasing recognition of the great importance of assessing psychopathology as a dimension in addition to the diagnostic approach. Reference Plomin, Haworth and Davis22
Psychosocial adjustment
Measures of psychosocial adjustment such as contact with psychiatric services, self-harm, truancy and contact with police were gathered in the baseline survey (2004). A participant was coded as having ever experienced one of these outcomes as rated by youth, teacher or parent report. In the follow-up survey (2007), informants were asked again about the same outcomes – we analysed new onsets of these outcomes.
Family factors
Sociodemographic details were collected at the parental interview covering the following information (a) ethnic group, (b) housing tenure (rented accommodation v. owner-occupiers), (c) gross household income, (d) maternal highest educational qualification, (e) family type (reconstituted with step-parent and/or step/half-siblings v. other, lone-parent v. other), (f) parental anxiety and depression, assessed with the 12-item version of the General Health Questionnaire, Reference Goldberg, Gater, Sartorius, Ustun, Piccinelli and Gureje23 (g) family discord, assessed with the general functioning scale of the McMaster Family Assessment Device, which includes 12 items, scored on a 1–4 scale with a maximum score of 48, Reference Miller, Epstein, Bishop and Keitner24 and (h) stressful life events during the child’s lifetime, including parental separation, court appearance, bereavement and serious illness or accident. Reference Goodyer, Wright and Altham25
Child factors
Parents also provided information about (a) age, (b) gender, (c) their child’s general health, using a five-point Likert scale, including the report of any physical disorders affecting their child, (d) neurodevelopmental disorders like cerebral palsy, difficulties with coordination, epilepsy and muscle disease or weakness and (e) ‘generalised learning disability’ (intellectual disability, referred to as learning disability in UK health services). Parents and teachers were asked to estimate each child’s mental age, and teachers reported whether a child had a written statement of special educational needs related to cognitive and intellectual needs (including specific, moderate, severe and profound intellectual difficulties, but not distinguishing between them). For the purpose of these analyses, a child was considered to have a ‘generalised learning disability’ when one or both informants estimated that mental age was 60% or less of the chronological age (such as a mental age of 6 or less at a chronological age of 10).
Statistical analyses
Interrater associations and longitudinal stability
We examined whether there was a correspondence between parent-rated positive attributes and their children’s self-perceptions. Concurrent and longitudinal associations of positive attributes scores were tested using Pearson’s correlations within and across informants.
Distinction between positive attributes and psychiatric symptoms across time
A path analysis model was estimated to test whether total SDQ score (symptoms) and YSI score (positive attributes) exhibited distinguishable predictions across time.
Positive attributes predicting future psychiatric symptoms
The prediction of psychiatric symptoms by positive attributes was estimated in regression models with the total SDQ symptom score as the outcome (i.e. at 36 month) and baseline YSI score (positive attributes) as the predictor. In subsequent steps, regression models were adjusted for total SDQ symptom score at baseline, as well as relevant family factors (i.e. ethnicity, single-parent family, reconstituted family, maternal highest educational qualification, gross household income, housing tenure, family functioning, General Health Questionnaire and life events) and child factors (i.e. age, gender, general health, neurodevelopmental disorder, generalised learning disability and any psychiatric disorder at baseline) as covariates. Reference Bromley, Johnson and Cohen1,Reference Williams and McGee8,Reference Ford, Collishaw, Meltzer and Goodman26–Reference Walrath, Mandell, Holden and Santiago28 This way we were able to test the specific association between positive attributes and psychiatric outcomes without the overestimation bias because of common related factors.
Positive attributes predicting future psychiatric disorders
The prediction of psychiatric diagnoses at 36-month follow-up (dependent variable) by baseline positive attributes was estimated in logistic regression models where baseline YSI score was used as an independent variable. We predicted three domains of psychiatric disorders; these were ‘any psychiatric disorder’, ‘emotional-internalising disorder’, and ‘disruptive behaviour-externalising disorder’. In adjusted models, diagnoses at baseline were used as covariates. In subsequent steps, the same child and family factors employed as covariates in the prediction of psychiatric symptoms were also added to these models. In addition, to examine whether parental ratings of positive attributes may lead to psychiatric disorders or be a consequence of them, we employed path analytical models. Doing so, we were able to look at the longitudinal association taking into account the correlation at baseline and follow-up.
Association with psychosocial adjustment
We employed logistic regression models in which the independent variable was positive attributes at baseline, and the dependent variables were the new emergence of the following factors in the 36-month follow-up: contact with psychiatric services, self-harm, truancy and contact with police. The models were examined unadjusted and adjusted for baseline psychiatric symptoms.
Difference in symptoms at follow-up between matched groups differing in level of positive attributes
Propensity score matching is an attempt to reduce bias in causal inference in observational studies. Through this method, a ‘treated’ group (for example high positive attributes) is matched on plausible confounders to a ‘control’ group (for example low positive attributes). If a good balance is achieved (i.e. approximately equal distribution of baseline covariates in both groups) and assuming that this model includes all relevant confounders, then this observational study should emulate what a RCT does by randomisation. Hence, a difference in the outcome (for example symptoms at follow-up) can be attributed to being ‘treated’ or not – unless there were strong confounders that were neither measured directly nor correlated with matching variables that were measured.
Propensity score analysis has several advantages over multiple linear regression approaches. First, when a good covariate balance is achieved, propensity score analysis does not rely on the correct specification of the functional form of the relationship (for example linearity or log linearity) between the outcome and the covariates, which is not the case with linear regression models when covariate distributions are very different between groups. Reference Zanutto29,Reference Dehejia and Wahba30 Second, propensity score methods make it easier to determine whether the model has been correctly specified than with regression approaches. Reference Austin11 Finally, propensity score analyses are considered objective in the sense that the model is specified without relying on the outcome. (It is also possible to combine propensity score analyses with regression adjustment to reduce slight imbalances in the covariates and increase precision.) Reference Rubin31–Reference Rubin and Thomas33
In this study we used a propensity score matching approach to test whether children only differing in levels of positive attributes at baseline showed different levels of psychiatric symptoms at follow-up. Reference Austin11 Our aim was to compare children at both extremes of the distribution of levels of positive attributes. A binary variable of baseline positive attributes was computed selecting the extremes of the distribution of this variable (high positive attributes v. low positive attributes) using percentiles 20 and 80 as cut-off points. Doing so, we kept approximately the same proportion of people in each group. We matched both groups on the same covariates that were adjusted for in the previous regression models, plus psychiatric symptoms and disorder at baseline. In this case, a logit regression model was used to estimate the propensity score. The resulting propensity score was the predicted probability of belonging to the high positive attributes group for each child. Propensity score matching was performed employing a one-to-one nearest-neighbour method within a caliper (or distance) of 0.5. Reference Austin11,Reference Heckman, Ichimura and Todd34 With one-to-one nearest-neighbour matching, only one ‘control’ participant is selected for each ‘treated’ participant, namely the one whose propensity score is closest to the ‘treated’ participant. Reference Stuart12 Caliper is defined as the difference in propensity scores between selected matches. By setting a caliper of 0.5 we pre-specify the largest allowable absolute difference in propensity scores for matched participants, thus ensuring closer balance. We only analysed observations that were inside the common support area. That is, there might be substantial overlap of the propensity score distributions in the two groups, but potentially density differences. Therefore, we discard individuals with propensity score values outside the range of the other group, thereby also ensuring a better balance. Reference Stuart12 Balance of the covariates was assessed before and after matching using a measure of standardised bias. Reference Rosenbaum and Rubin13 Standardised differences of means <0.20 are acceptable, Reference Stuart12,Reference Rosenbaum and Rubin13 and differences <0.10 are considered negligible Reference Normand, Landrum, Guadagnoli, Ayanian, Ryan and Cleary35 (i.e. no mean differences in a covariate between groups). Regression models were employed to test the difference in means of SDQ total score at follow-up between matched groups as well as the change of SDQ total score over time in each group. Cohen’s d were calculated as a measure of effect size.
The Stata 11 software package for Windows was employed to test all study hypotheses, except for the path analyses, which were run with MPlus version 7. Reference Muthen and Muthen36
Results
Interrater correlation and longitudinal stability
As shown in Table 1, there was a moderately strong cross-sectional correlation between parent- and child-rated positive attributes as measured with the YSI. The stability of positive attributes scores within informants ranged from moderate in children to high in
Association of positive attributes within and across informants at baseline and follow-up
| Positive attributes score, r (95% CI) n | |||
|---|---|---|---|
| Baseline | 36 months | ||
| Positive attributes score | Parent | Child | Parent |
| Baseline | |||
| Parent | - | - | - |
| Child | 0.30 (0.24-0.36) 838 | - | - |
| 36 months | |||
| Parent | 0.64 (0.62-0.65) 4921 | 0.24 (0.18-0.31) 838 | - |
| Child | 0.29 (0.23-0.35) 838 | 0.46 (0.41-0.51) 923 | 0.30 (0.24-0.36) 838 |
r, Pearson correlations; n, number of observations.
P<0.001 in all cells. All findings in bold are significant (P<0.05).
parents. Finally, positive attributes scores at baseline were significantly better (as evidenced in non-overlapping CIs) in predicting positive attributes scores at 36 months within informants than across informants.
Distinction between positive attributes and symptoms across time
The path analysis model in Fig. 1 shows that the within-domain prediction is much stronger than the across-domain prediction: positive attributes are a better predictor of positive attributes, whereas symptoms are a better predictor of symptoms (as evidenced in non-overlapping CIs).
Positive attributes predicting future psychiatric symptoms
As seen in Table 2, higher levels of positive attributes measured at baseline were significantly predictive of fewer psychiatric symptoms
Path analysis of the relationship between positive attributes and Strengths and Difficulties Questionnaire (SDQ) total difficulties (symptoms) score across time.
Significant paths (P<0.05) and correlations with standard errors and 95% confidence intervals are presented as straight and curved lines respectively. R 2, proportion of variance explained.
Association between positive attributes score at baseline and Strengths and Difficulties Questionnaire (SDQ) total difficulties (symptoms) score at follow-up in adjusted and unadjusted modelsFootnote a
| Outcome: SDQ score at 36-month follow-up predicted by: | β (95% CI) R 2 |
|---|---|
| Positive attributes score only | –0.48 (–0.51 to –0.46) 0.22 |
| Positive attributes score adjusted for baseline SDQ difficulties score | –0.10 (–0.12 to –0.07) 0.47 |
| Positive attributes score adjusted for baseline SDQ difficulties score and family factors | –0.10 (–0.12 to –0.07) 0.47 |
| Positive attributes score adjusted for baseline SDQ difficulties score and child factors | –0.09 (–0.12 to –0.07) 0.48 |
β, standardised regression coefficient; R2, proportion of variance.
a. Family factors are ethnicity, single parent family, reconstituted family, maternal highest education, gross household income, housing tenure, family functioning, General Health Questionnaire and life events. Child factors are age, gender, general health, neurodevelopmental disorder, generalised learning disability and any psychiatric disorder at baseline. All findings in bold are significant (P50.05).
at follow-up. This was even true in models adjusted for SDQ total symptom score at baseline, family factors and child factors.
Positive attributes predicting future psychiatric disorders
Table 3 shows that a higher level of parent-rated positive attributes at baseline was a significant predictor of less psychiatric disorders at follow-up, no matter what domain of disorder was predicted. This was even true when adjusting for baseline disorder, family factors and child factors.
Figure 2 shows that positive attributes predicted less psychiatric disorders (beta coefficient range: –0.19 to –0.32) to a significantly higher extent than disorders predicted less positive attributes (beta coefficient range: –0.05 to –0.08).
Association with psychosocial adjustment
As can be seen in Table 4, parent-rated positive attributes were significant predictors of less subsequent psychiatric and police
Standardised positive attributes at baseline as a predictor of psychiatric disorders in adjusted and unadjusted modelsFootnote a
| OR (95% CI) | |
|---|---|
| Any disorder predicted by: | |
| Positive attributes only | 0.42 (0.39-0.46) |
| Positive attributes adjusted for any disorder at baseline | 0.57 (0.51-0.63) |
| Positive attributes adjusted for any disorder at baseline and family factors | 0.59 (0.53-0.66) |
| Positive attributes adjusted for any disorder at baseline and child factors | 0.60 (0.53-0.66) |
| Any emotional disorder predicted by: | |
| Positive attributes only | 0.62 (0.56-0.70) |
| Positive attributes adjusted for any emotional disorder at baseline | 0.69 (0.61-0.78) |
| Positive attributes adjusted for any emotional disorder at baseline and family factors | 0.74 (0.64-0.85) |
| Positive attributes adjusted for any emotional disorder at baseline and child factors | 0.73 (0.64-0.83) |
| Any externalising disorder predicted by: | |
| Positive attributes only | 0.40 (0.36-0.45) |
| Positive attributes adjusted for any externalising disorder at baseline | 0.54 (0.48-0.62) |
| Positive attributes adjusted for any externalising disorder at baseline and family factors | 0.61 (0.52-0.70) |
| Positive attributes adjusted for any externalising disorder at baseline and child factors | 0.55 (0.48-0.63) |
a. Family factors are ethnicity, single parent family, reconstituted family, maternal highest education, gross household income, housing tenure, family functioning, General Health Questionnaire and life events. Child factors are age, gender, general health, neurodevelopmental disorder and generalised learning disability. All findings in bold are significant (P<0.05). Note that odds ratios significantly less than 1 represent a protective effect.
Path analyses of the relation between positive attributes and any domain of disorder (a), any internalising disorder (b) and any externalising disorder (c) across time.
Significant paths (P<0.001) and correlations with standard errors and 95% confidence intervals are presented as straight and curved lines respectively. R 2, proportion of variance explained.
Standardised positive attributes at baseline as predictors of a new onset of psychiatric contact, self- harm, truancy and police contact at follow-up, with and without adjustment for baseline symptomsa
| Association of positive attributes at baseline with outcome at 36 months, OR (95% CI) | ||
|---|---|---|
| Unadjusted for baseline symptoms | Adjusted for baseline symptoms | |
| Predicting new psychiatric contact | 0.60 (0.53-0.69) | 0.83 (0.70-0.99) |
| Predicting new self harm | 0.70 (0.63-0.79) | 0.87 (0.75-1.01) |
| Predicting new truancy | 0.58 (0.49-0.70) | 0.81 (0.63-1.04) |
| Predicting new police contact | 0.64 (0.58-0.71) | 0.75 (0.66-0.86) |
Odds ratios with confidence intervals are presented from logistic regression models with each of the outcomes as dependent variables and positive attributes as an independent variable either unadjusted or adjusted for baseline total Strengths and Difficulties Questionnaire score. All findings in bold are significant (P<0.05). Note that odds ratios significantly less than 1 represent a protective effect.
contact in individuals who had not had such contact previously, even with adjustment for baseline psychiatric symptoms. By contrast, unadjusted prediction to new truancy or self-harm did not remain significant once adjusted for baseline psychiatric symptoms.
Difference in symptoms at follow-up between matched groups differing in level of positive attributes
As seen in Fig. 3, the reduction in standardised bias was acceptable after carrying out the propensity score matching; there was only one covariate with a standardised bias >0.10. The mean standardised bias in the covariates was 42.3% (s.d. = 39.5, range 9.2–162.3) before matching, and decreased to 3.7% (s.d. = 2.8, range 0.3–10.5) after matching. The outcome analyses (after matching) revealed that there were significant differences in the SDQ total score at follow-up between matched groups with high and low levels of
Quality of propensity score matching: standardised bias (%) across covariates before and after matching groups with high and low positive attributes.
positive attributes (effect size 0.25, 95% CI 0.11–0.38, P<0.001). Moreover, the change in SDQ total score over time was significant for the matched low positive attribute group (effect size 0.39; 95% CI 0.08–0.69, P<0.05) but not for the matched high positive attribute group (effect size 0.15, 95% CI –0.18 to 0.47, P = 0.372) (Fig. 4).
Discussion
Main findings
This study examined the extent to which parental perceptions of positive attributes in their own children predicted psychiatric outcomes. Analyses were carried out in a large epidemiological sample of young children and adolescents employing a longitudinal design. Overall, we found that higher levels of parental perceptions of positive attributes added to the prediction of fewer future
Standardised increase of psychopathology over time in matched groups: effect sizes of change in the Strengths and Difficulties Questionnaire total score (symptoms) at follow-up in matched groups with high and low levels of positive attributes.
Bars represent 95% confidence intervals. Overall effect size, 0.25 95% CI 0.11–0.38, P<0.001.
psychiatric symptoms and disorders when adjusting for relevant confounders. Although effect sizes were relatively small, especially for psychiatric symptoms, predictions were particularly strong for ‘all types of disorder combined’ and ‘externalising disorder’. The results for the self-reported positive attributes were in the line with parent-reported measures; that is, higher levels of self-reported positive attributes at baseline were significant predictors of less psychiatric symptoms at follow-up. In addition, higher levels of self-reported positive attributes were significant predictors of fewer externalising disorders. These results are available in the online Tables DS1–3.
To date, most of the literature on this topic has been based on clinical samples and has used self-reported positive attributes. To our knowledge there are very few works employing parent-report measures of positive attributes Reference Park and Peterson5,Reference Williams and McGee8 and even fewer examining the relationship with psychiatric outcomes. Reference Bromley, Johnson and Cohen1,Reference Radigan and Wang2 Using two informants, we found a moderately strong agreement between youth and parent-reported positive attributes at baseline and follow-up. We also found that levels of positive attributes were highly stable across time within parents, and moderately stable within youth and across informants.
In line with previous studies, Reference Radigan and Wang2,Reference Gillham, Adams-Deutsch, Werner, Reivich, Coulter-Heindl and Linkins3,Reference Lyons, Uziel-Miller, Reyes and Sokol37 higher levels of positive attributes at baseline were predictive of fewer psychiatric symptoms 3 years later. As expected, the predictions were still significant even after accounting for symptoms at baseline and relevant child and family factors. As we predicted, effect sizes were moderate in the unadjusted models (for example –0.48 for parent-reported positive attributes to symptoms 3 year later) and substantially attenuated in adjusted models (for example about –0.10 for parent-reported positive attributes to symptoms 3 years later). However, the propensity score matching approach demonstrated that children who only differ in levels of positive attributes at baseline showed different levels of psychiatric symptoms at follow-up. As expected, differences in psychiatric symptoms were lower but still significant. This approach, which is more similar to an RCT, confirms that positive attributes are independently related to psychiatric symptoms.
In addition, a path analytical model showed that positive attributes were best predicted by positive attributes and symptoms were best predicted by symptoms, suggesting different constructs. Our data are more in keeping with a vulnerability model than a model in which positive attributes and psychopathology are on the same continuum. Reference Tackett38
Supporting previous results in adolescent samples, Reference Bromley, Johnson and Cohen1 we also found that higher positive attributes scores at baseline added to the prediction of reduced risk for psychiatric disorders 3 years later. For example, our results showed that every decrease in parent-reported positive attributes score increased the probability of future psychiatric disorder (all sorts combined) by 76%, of emotional disorder by 45% and of externalising disorder by 85%. These probabilities were only attenuated by 10% after accounting for child and family factors. Of note, higher positive attributes scores were more likely to predict reduced risk for externalising disorders than for emotional disorders. This was an unexpected finding and should be confirmed in future studies. Previous research found a significant association with emotional disorders and any psychiatric disorder, although only the last remained significant after controlling for socioeconomic confounders. Reference Bromley, Johnson and Cohen1 In contrast to our results, the same study did not find an association with disruptive disorders.
The difference in the prediction of internalising and externalising disorders might be for different reasons. First, the prevalence of each domain of disorder differs across different ages. Second, externalising problems may cause more burden in families Reference Meltzer, Ford, Goodman and Vostanis39 and therefore parents may underreport their children’s positive attributes; however, in our sample, adjusting for parental burden did not change the pattern of the results.
In path analyses we showed that higher positive attributes at baseline predicted fewer psychiatric disorders 3 years later; this direction of effect was significantly stronger than that of psychiatric disorders predicting fewer positive attributes.
The weaker prediction to symptoms than to disorders may be explained by the methods employed for collecting the data. Psychiatric diagnoses were based on clinical ratings of interviews, including extensive open-ended comments – a substantially different method than the respondent-based YSI measure of positive attributes. By contrast, SDQ-based symptoms and YSI-based positive attributes shared a respondent-based approach, so that adjusting for baseline symptoms will substantially attenuate prediction as a result of common method variance. Reference Podsakoff, MacKenzie, Lee and Podsakoff40 Indeed, if adjustment is restricted to child and family factors, without adjustment for baseline symptoms, effect sizes remained around –0.40 for parent-reported positive attributes predicting symptoms 3 years later.
Finally, lower parent-reported positive attributes scores increased the probability of new future contact with psychiatric services by 20% and contact with police by 33% after adjusting for baseline symptoms. Interestingly, contact with police was the category most strongly associated with externalising disorder whereas contact with psychiatric services was the category most strongly associated with all types of disorder combined.
Implications
Our findings suggest benefits of measuring positive attributes alongside symptoms in day-to-day mental health assessments of children and young people. This adds to the prediction of psychiatric symptoms and psychiatric disorders, in line with previous results. Reference Bromley, Johnson and Cohen1,Reference Gillham, Adams-Deutsch, Werner, Reivich, Coulter-Heindl and Linkins3 This enables clinicians to recognise a particularly high-risk group independently of the level of symptoms at admission – something that may be especially useful as regards externalising disorders. It is plausible that interventions aiming to increase the recognition of and further develop positive attributes may alleviate or prevent externalising disorders.
Furthermore, previous research suggests that strength-based assessment approaches promote family involvement in treatments Reference Rudolph and Epstein41 and result in more caregiver satisfaction and treatment adherence. Reference Cox42 Assessing a child’s positive attributes can be a less stigmatising way of discussing treatment needs and may make it easier for clinicians to engage families.
Limitations
The findings of this study must be considered in the light of its limitations. First, as is often the case in longitudinal studies, there was considerable attrition. Although this would be most likely to bias prevalence estimates rather than the associations between disorders and other factors, Reference Wolke, Waylen, Samara, Steer, Goodman and Ford43 it has meant that those with high levels of psychopathology at baseline were less likely to participate in the study. Second, propensity score matching in non-experimental studies cannot adjust for unobserved covariates, so residual confounding is still possible. Reference Stuart12 A third limitation is that the YSI, unlike other longer strengths measures such as the Values in Action Inventory of Strengths (VIA-IS) Reference Park and Peterson4 or the Behavioral and Emotional Rating Scale (BERS) Reference Epstein and Sharma44 measures positive attributes as a single factor. Although simple and quick, this could be a disadvantage if different positive attributes show different associations with specific psychiatric outcomes. Future studies should compare the YSI with longer multiscale measures to establish what, if anything, is lost through using a brief unidimensional measure of positive attributes. Finally, it is conceptually unclear what is actually relevant in the assessment of positive attributes. That is, we cannot determine how far reported positive attributes reflect the presence of objective strengths in the child – as opposed to subjective rater biases. However, the same limitation applies to all parent-reported psychopathology, and is not specific to positive attributes. Future research to examine this issue might use a variety of informants (including teachers and peers) and approaches (including systematic observations schedules, intervention studies and qualitative studies)
To sum up, our findings suggest that positive attributes can be easily and reliably measured with the YSI from the DAWBA and can provide useful information regarding psychosocial and psychiatric outcomes in children and young people.
eLetters
No eLetters have been published for this article.