Does parent–child agreement vary based on presenting problems? Results from a UK clinical sample

Externalising problems

Several studies have found that agreement between parents and children was higher when reporting externalising symptoms rather than internalising ones and this has mostly been interpreted to be due to the externalising behaviours being more readily observable by the parent than the internalising difficulties [3, 7, 19]. However, disagreements still remain and children often report less behavioural problems than their parents, which might indicate that the underlying reason for the discrepancies is the child’s limited self-awareness [20, 21]. It has been suggested that externalising disorders are often characterized by the failure to reflect on the self and evaluate one’s own behaviour based on feedback from others [22], resulting in positive biases and impaired self-perceptions [23, 24]. This could have a protective and adaptive function, as an attempt to cope with the difficulties of the disorder [25].

Internalising problems

Self-reports are considered particularly important for investigating internalising problems, because these concern internal subjective experiences that might not be observed by others [26, 27]. Indeed, parent–child agreement when reporting emotional difficulties is often lower than for externalising, with children reporting more problems than their parents [19, 28]. One common characteristic of internalising is the distortion of cognition [29–31]. An alternative controversial school of thought introduced the concept of ‘depressive realism’, which can be defined as the propensity of depressed individuals to have more accurate perceptions of reality, while non-depressed people are more likely to exhibit positive biases when evaluating themselves [32–34]. This is consistent with studies such as that of Oland and Shaw [22], which highlighted the key role of self-reflection in the development of internalising disorders and the lack of this in externalising problems.

Comorbidity of disorders

Hoza, Murray-Close, Arnold, Hinshaw and Hechtman [35] used a longitudinal design over a 6-year period (assessed at 4 time points) to investigate the link between externalising and internalising problems and limited self-awareness. The findings indicated that children with ADHD 8–13 years old presented with more positively biased self-perceptions about their behaviour relative to reports from teachers across the 6 years, compared to the control group of their healthy peers. Their aggression levels at Times 1 and 2 also significantly predicted positive biases in the perception of their own behaviour at later time points, and at the same time positive biases of behaviour at Time 3 predicted later aggression. One explanation provided by the authors for these findings was the self-protection hypothesis, which suggests that positive biases serve as protection to cope with one’s own deficits [25]. Another important finding by Hoza and colleagues [35] indicated that depressive symptomatology was associated with a reduction of these inflated self-perceptions over time. Therefore, since externalising difficulties were associated with an increase in positive biases and internalising with their reduction, it would be interesting to investigate these biases in the context of comorbidity of difficulties.

Other child characteristics

With regards to age, Achenbach and colleagues [3] demonstrated that agreement between parents and children was higher for younger children (mean r = .51) than for adolescents (mean r = .41). The authors suggested that this may be because younger children spend more time with their parents than adolescents do, thus their behaviour is more observable. Similar findings were demonstrated by other studies using samples from the general population demonstrated that agreement between parents and children was higher for younger children (mean r = .51) than for adolescents (mean r = .41). The authors suggested that this may be because younger children spend more time with their parents than adolescents do, thus their behaviour is more observable. Similar findings were demonstrated by other studies using samples from the general population demonstrated that agreement between parents and children was higher for younger children (mean r = .51) than for adolescents (mean r = .41). The authors suggested that this may be because younger children spend more time with their parents than adolescents do, thus their behaviour is more observable. Similar findings were demonstrated by other studies using samples from the general population [36]. However, these results were not replicated when investigating clinical samples [37, 38]. Additionally, the effect of gender on parent–child-agreement has also been examined and like with age the results are inconsistent [37, 38].

Clinician-reported presenting problems

Clinicians completed a form rating twelve presenting problems for each child at the assessment stage. Ratings are based on the clinical judgement of individual clinicians and do not need to imply a diagnosis. The twelve presenting problems included in the form were: hyperkinetic, emotional, conduct, eating, psychosis, deliberate self-harm, autism spectrum disorder, learning disability, developmental, habit, substance misuse and other problems. The clinician was asked to provide ‘yes’ or ‘no’ answers, as to whether each of these problems was present for a child. The most common presenting difficulties reported in this sample were emotional (57%) and conduct problems (15%).

Strengths and Difficulties Questionnaire (SDQ)

The SDQ is a short questionnaire of 25 items used to assess the positive and negative behaviours of children and indicate the extent of their difficulties [6, 39]. The SDQ contains five subscales with 5 items each, representing different behavioural, social and emotional domains. These include conduct problems, emotional problems, hyperactivity, peer problems and prosocial behaviour. Scores ranging from 0 (no difficulties) to 10 (severe difficulties) are generated for each individual scale to indicate the extent of the child’s difficulties for each domain. In terms of outcome information, the main variables used for this study were the scores from the conduct and emotional scales of the SDQ, for both parents and children. Data were collected using the self-report version, which was developed for young people between the ages of 11–17 [6] and the parent-rated version aimed to be completed by parents/carers of children aged 4–17 [39–41].

Findings concerning the validity of the parent version of the SDQ indicated that it operated equally well as other well-established measures, such as the Rutter questionnaires or the Child Behaviour Checklist [39, 42]. It also demonstrated adequate criterion validity in relation to clinical diagnosis, as a correlation of .47 was found between the total difficulties score and diagnostic interview features [43]. Moreover, Goodman and colleagues [6] found satisfactory internal consistency for the self-report version of the SDQ in an adolescent population (emotional scale a = .75; conduct scale a = .72) and also confirmed that the self-report version could be used effectively to distinguish between children in a clinical sample from those in a community sample (concurrent validity = .82).

Exclusion criteria

The initial dataset provided by CORC contained 263,927 clinical records. However, large amounts of essential data (e.g. presenting problems) were missing, necessitating sample selection based on the following three main exclusion criteria: (1) records with no information about clinician-reported presenting problems for the child, as these formed the basis for dividing the sample into groups; (2) records with insufficient information to compute SDQ Emotional Problems or SDQ Behavioural Problems score for children or parents, as the main premise of this project was to investigate the reporting behaviours of children and parents; (3) records of young people under the age of 11 or over 17, in accordance to SDQ guidelines about the age suitability of the self-report version [6]. Figure 1 demonstrates a flowchart of the selection process. Sample selection was closely monitored, by comparing the descriptive statistics and distributions of the main variables of interest (such as the SDQ conduct and emotional scale scores) before and after the sample selection, and analyses indicated that the selection process did not change the data significantly or introduced bias in the distribution of key variables.

Data analysis

Before conducting the analysis, it was important to acknowledge the possible influence of missing values on the results. Based on the sample inclusion criteria, all the children had self-report data but not all had parent data. Effectively, this meant that records with missing parent scores would not be included in the correlation analysis. In order to identify whether these missing values would create a bias in the sample, the proportion of parents and children who both completed the SDQ was investigated for each age. It was found that the older the children were, the smaller the percentage was of those who had both child and parent reports. The distributions of the child scores on both SDQ scales for those with only child reports were found to be similar to the distributions of those who had both child and parent reports, thus it was decided to conduct the analysis on the latter group.

In order to test whether parent–child agreement on the SDQ conduct and emotional scales varied by the child’s presenting problem, the data was analysed using Pearson’s r correlations.¹ Following that, a test of multiple independent correlations [44] was conducted for each scale, in order to identify whether the aforementioned group coefficients significantly differed from each other, thus representing a real difference in the population. Then, Fisher’s Z transformations [45] were used to investigate whether agreement differed between groups when reporting on the problems that defined the child’s presentation; for example, by comparing the INT group agreement on the emotional scale with the EXT group agreement on the conduct scale. Lastly, pairwise tests of correlated correlations [46] were run to test the difference between the coefficients of the conduct scale and those of the emotional scale for each problem group. Pearson’s r correlations were finally conducted to explore agreement for different ages and gender.

All in all, this study employs ten statistical hypothesis tests. For each test, we report the uncorrected p value. This is recommended for research situations such as ours, where tests are used to investigate specific hypotheses developed prior to seeing the data [47, 48]. For the exploratory analyses relating to the second research aim, we do not employ significance tests, but report confidence intervals to indicate the uncertainty around the observed correlations.

Correlations for the conduct scale

The test of multiple independent correlations indicated that within the conduct scale, at least some of the correlations significantly differed between the groups (C(α) = 64.4, df = 6, p < .0001). More specifically, the correlation coefficients indicated that parents and children in problem groups that excluded externalising problems (INT, INT & OTHER, OTHER) seemed to agree more on conduct scores than groups that included externalising (EXT, COM, EXT and OTHER, COM and OTHER). Note in particular that the COM and COM and OTHER groups had smaller correlation coefficients than groups that did not include externalising difficulties, despite having comorbid internalising problems. These findings partly support our hypothesis, as they demonstrate that internalising groups have better parent–child agreement than the externalising ones whilst reporting conduct problems. However, it seems that it is the absence of externalising, rather than the presence of internalising difficulties that relates to higher parent–child agreement.

Correlations for the emotional scale

The second set of correlations was comparing child and parent scores on the emotional dimension of the SDQ. For the hypothesis to be supported it was again expected that correlations would be higher for groups with internalising problems and particularly the COM group, than the group with only externalising difficulties. As can be seen in Table 3, parent and child scores were positively correlated for all problem groups, with the COM and OTHER problem group having the highest correlation, while the COM group had the lowest correlation. Overall, it appeared that groups including internalising problems tended to have slightly lower parent–child agreement (INT, COM, INT and OTHER), with the exception of the COM and OTHER. These results do not support the initial hypothesis; rather, if anything the opposite tended to occur, that groups with externalising difficulties showed greater agreement on scores of emotional difficulties. However, the test of multiple independent correlations found no significant difference between any of the group correlations for the emotional scale (C(α) = 8.6, df = 6, p = .198). Therefore, there was no evidence that the presenting difficulties of the child affect the level of parent–child agreement on the emotional scale.

Comparing agreement between scales