Chapter 5 The British Child and Adolescent Mental Health Surveys 1999 and 2004
5.3 Mental health measures used in B-CAMHS
5.3.2 The Strengths and Difficulties Questionnaire (SDQ)
Description of measure
The Strengths and Difficulties Questionnaire (SDQ) is a 25-item questionnaire with supplementary questions on distress and impairment. It can be administered to parents and teachers of children aged 4 to 17, and to children aged 11 or over.
The SDQ was designed to include five subscales relating to emotional problems, behavioural problems, hyperactivity, peer problems and prosocial behaviour [194, 448].
Each subscale is comprised of five items with ‘Not true’ (scored 0), ‘Somewhat true’
126 (scored 1) and ‘Certainly true’ (scored 2) as response options, giving a range of 0-10. Five of the 20 items on the emotional, peer, behavioural and hyperactivity subscales are
‘strenghts’ and are reverse scored (the five prosocial items are also all positively worded).
Responses to the 20 items from the emotional problems, peer problems, behavioural problems and hyperactivity subscales are added to give a total difficulties score (TDS), with a range of 0-40. An annotated copy of the parent version of the SDQ is included in Section 14.2.1 Appendix 2, together with a summary of the constituent SDQ items of the five subscales and TDS. All scales (and particularly those from teachers) are usually positively skewed to some extent, as illustrated in Figure 5.5 using the B-CAMHS data.
Figure 5.5: Distribution of the parent, teacher and child TDS in B-CAMHS
In addition to these 25 symptoms, the SDQ contains supplementary questions on impact.
The parent and child SDQs contain one item on distress to the child, and four items on interference in home life, friendships, classroom learning and leisure activities. The response options are ‘Not at all’/’Only a little’ (both scored 0), ‘Quite a Lot’ (scored 1) and
‘A great deal’ (scored 2), giving an impact score of 0-10. The teacher SDQ contains one item on distress to the child and two on interference with peer relationships and classroom learning, giving an impact score of 0-6. There is also a single four-point item about burden to others for all informants.
Comparison with other child psychiatric brief questionnaires
There exist several other brief questionnaires which provide dimensional measures of child mental health. Among the longest-established are the Rutter scales for parents and teachers [68], which have proved reliable, valid and useful in a wide range of settings [193]. Also widely used are the Achenbach System of Empirically Based Assessment for parents, teachers and children [ASEBA: 187, 188-191], which I introduced in Chapter 2, p.48.
127 The development of the SDQ started with an extended version of the Rutter questionnaire which included ‘strengths’ items [449]. The SDQ was then refined by using factor analyses to identify its five key subscales; by including items which explicitly relate to core diagnostic symptoms (e.g. concentration, restlessness and impulsiveness for hyperactivity);
and by omitting behaviours like thumb-sucking which are no longer of interest to mental health professionals [194]. The result is that the SDQ has fewer items on emotional and behavioural problems than the Rutter; covers hyperactivity in a way which corresponds more closely to current diagnostic criteria; and has greater coverage of peer problems and prosocial behaviour. Unlike the Rutter, there also exists a child-report version of the SDQ.
The SDQ differs from the Rutter and the ASEBA in other important ways. The SDQ includes a substantial number of strengths items, as opposed to the exclusive focus of the Rutter and the ASEBA upon difficulties. The parent, teacher and child SDQs have identical items, facilitating comparisons across informants. By contrast, there are important differences in the content of the parent and teacher Rutter, and the parent, teacher and child ASEBA questionnaires. Finally, at 25 items the SDQ is slightly shorter than the Rutter (31 items in the parent version, 26 in the teacher version) and substantially shorter than the ASEBA (120 in the parent and teacher versions, 105 items on the child version). This may explain some evidence that the SDQ is more popular than the ASEBA with parents [450-451].
The reliability and validity of the SDQ total difficulties score (TDS) and subscales in Britain
Total difficulty score
There is substantial evidence that the SDQ’s total difficulty score (TDS) provides a reliable and valid measure of child mental health problems in Britain, with psychometric properties comparable or superior to other brief questionnaires. This evidence is summarised below and described in full in Table 14.8 of Section 14.2.2 Appendix 2.
Evidence of reliability of the TDS includes reasonable test-retest reliabilities (0.65-0.85) [452-453] and high Cronbach’s alphas (≥0.80 in B-CAMHS99) [448]. Agreement between parents, teachers and children was only moderate in B-CAMHS99 (0.33-0.48) [448] but
128 nonetheless substantially better than the mean agreement reported in a meta-analysis of previous measures (0.20-0.27) [226]. Moreover, as discussed in Chapter 2 (p. 60), these comparatively low rates of agreement do not necessarily simply reflect poor measurement – they may also result from substantive differences in how a child behaves in different settings.
The TDS also shows good evidence of construct validity. Despite being briefer, the TDS is highly correlated (0.78-0.92) with the parent and teacher Rutter and parent ASEBA and functions at least as well at detecting high risk groups [194, 451]. ‘High’ scores show good sensitivity (≈80-90%) and specificity (≈80%) relative to clinical diagnosis [454-455].
Moreover, in the combined B-CAMHS surveys the TDS seems to represent a truly dimensional measure of mental health problems, with each one-point increase in parent, teacher and child TDS being associated with an increased prevalence of DAWBA diagnosis [456].13 Further evidence to this effect comes from the fact that the same risk factors that predict change in TDS across the entire range, also predict it in children one standard deviation above and one standard deviation below the mean [457]. These risk factors are, moreover, the same as for the DAWBA and are in line with the previous literature [458].
SDQ subscales
There has been less evaluation of the reliability and validity of the five hypothesised SDQ subscales and the evidence is more mixed (see Table 14.9, Appendix 2). Cross-informant correlations are again not high but better than average for measures of child mental health, ranging from 0.25-0.48 for parent-teacher correlation, 0.30-0.44 for parent-child correlation and 0.21-0.32 to teacher-child correlation. The Cronbach alphas for some subscales are somewhat low (0.60-0.70), which may partly reflect the small number of items per scale [448]. The subscales showed evidence of good convergent and discriminant validity relative to DAWBA diagnoses in B-CAMHS99 [459]. The correlation with scores on the PACS investigator-based, semi-structured interview [460] are also least as high for the SDQ as for the ASEBA [451]. The subscales show the expected pattern of cross-scale correlation, with the behavioural and hyperactivity subscales being more highly correlated with each other than with the emotional subscale [448, 451]. Principal component analyses of B-CAMHS99 also broadly confirm the expected factor structure; in five-factor principal
13 On this paper I was the lead author; see Appendix 3.
129 component analyses all 25 items loaded onto their expected scale in all three informants, with 72/75 of these loadings being >0.4 [448].
Chief limitations of existing psychometric evaluations of the SDQ
Yet while these principal component analyses are reassuring, the existence of a proposed factor structure means that confirmatory factor analyses (CFA) would be a more appropriate technique. This is because CFA provide a hypothesis driven and model-based framework which can formally test the relative fit of different models (see Appendix 1, Section 13.1.5 for a fuller discussion).
This gap in the evidence is particularly important because the international literature does not fully support the proposed five-factor structure. As in Britain, most international studies are limited to exploratory principal component analyses or exploratory factor analyses (EFA). Generally these support the proposed five-factor solution (for a recent review, see [42]). At least one EFA from Finland [461], however, and two CFA from the US [462] and Belgium [463] suggest a three-factor solution. These three factors comprise an ‘internalising’ scale of the emotional and peer items, an ‘externalising’ scale of the behavioural and hyperactivity items and the standard ‘prosocial’ subscale. Table 14.7 (Section 14.2.1 Appendix 2) summarises this alternative three-factor structure alongside the hypothesised five-factor structure.
Finally, a thorough CFA on 914 Australian parent, teacher and child SDQs found that the five-factor solution did not provide adequate fit to the data and that many items loaded onto multiple factors [464]. The authors note that many psychological scales fail to meet rigorous psychometric criteria, and that model fit was near-acceptable by some indices.
Nevertheless, this study raises further grounds for caution regarding the SDQ factor structure, particularly since a CFA on 4167 Norwegian child SDQs also found a questionable model fit [465].
I therefore evaluate the SDQ’s internal structure further in Section 8.1 Chapter 8, conducting CFA and other analyses in order to inform my subsequent comparison of Indians and Whites.
130 Evaluation of the SDQ outside Britain
The satisfactory psychometric properties and clinical utility of the SDQ has been demonstrated in many high-, middle- and low-income settings including the US, Italy, Scandinavia, Brazil and the Yemen (for reviews, see [3 (Appendix D), 191, 200]). In India, the SDQ has not been validated but has been used with apparent success in Goa [426]. The SDQ has also been evaluated in other parts of the Indian subcontinent. Specifically, studies in Bangladesh [466-467] and Pakistan [468] have demonstrated that the SDQ differentiates clinic and community samples, strongly predicts clinical diagnosis, and can discriminate between different types of disorder within the clinic sample. Particularly noteworthy is the simultaneous evaluation of the multi-informant SDQ algorithm for probability of disorder in 101 children from a London clinic and 89 from a clinic in Dhaka, Bangladesh [454]. In both settings, the category of ‘probable’ diagnosis correctly identified most children (81-91%) who were independently assigned a clinical diagnosis, and the observed association was as strong in Dhaka as in London.