This study aimed to examine the utility and validity of the CBCL in assessing behavioural/emotional problems in a rural Kenyan preschool sample. After translation and slight adaptation of the CBCL, overall internal consistency properties were excellent, the test–retest correlation coefficients were good, and the inter-informant agreements with mothers were acceptable for other close caretakers, as well as for fathers. Additionally, most factor loadings and fit statistics for the seven-syndrome CBCL structure were acceptable, establishing the use of these behavioural/emotional constructs in this population.
CBCL scores and cut-off ranges
The mean CBCL scores (27) in this sample is comparable to 33 from an American sample [3], but lower than those in a Taiwanese (42) [20] and Chinese sample (45); although the latter included adopted children who may have more psychopathology than in the general population [21]. Parents may have underreported the extent of behaviour/emotional problems considering the stigma associated with mental health illnesses [22], particularly as this was the first psychopathology survey of preschool children in this area. Behavioural/emotional scores were similar between sexes and between externalising and internalising scales, consistent with some previous studies [3, 21], but not others [20].
The cut-off CBCL scores for use in epidemiological and intervention studies based on the 90th percentile as recommended by Achenbach and Rescorla [3] is comparable to those of 50–65 reported in other countries [3, 20]. This cut-off score likely represents those at risk of severe behavioural/emotional problems rather than a clinical diagnosis of mental health problems since it is derived from a random rather than a normative sample. The high behavioural/emotional scores in our study are consistent with a high prevalence of neuropsychiatric conditions in this area [14]; the prevalence of behavioural/emotional problems may be higher than the 8–15 % reported in most studies from high income countries [1].
Associations for discriminant validity
Behavioural/emotional scores were associated with pregnancy complications and adverse perinatal events, supporting the discriminant validity of the CBCL in differentiating at-risk children from those not at risk [3]. No significant associations were observed with seizures and socioeconomic information, but this may be explained by the smaller number screening for seizures, for example. Nonetheless, all these factors investigated should be accounted in associations with behavioural/emotional scores since they can be potential confounders. The CBCL may therefore be used by clinicians to identify children at risk of behavioural/emotional problems, following medical conditions or early life exposures, who would benefit from behavioural/emotional interventions.
Test retest and inter-informer reliability
The good test–retest reliability scores asserts the stability of the CBCL in assessing behaviour over time, although psychopathology can change in developing children [23]. Our test–retest reliability was better than that reported from a Luganda version of the CBCL (0.76 vs. 0.67), but the Uganda study used the school-aged CBCL [13]. Inter-informant agreement was acceptable for both fathers and caretakers, although the former was lower than the latter; which is similar to UK studies using the Strengths and Difficulties Questionnaire [24]. Indeed in anecdotal reports from the field team a number of fathers noted that they were not very familiar with their children’s behavioural/emotional patterns. On the contrary, caretakers such as grandmothers, stepmothers and/or aunts showed good inter-informant agreement with the mothers; as they spend more time caring for these children.
Internal consistency
All empirically-based seven-syndromes, as defined by ASEBA [3], were associated with acceptable to excellent reliability coefficient alphas, underscoring the value of the CBCL in assessing behavioural patterns in this Kenyan rural population. A Luganda version of the school-aged CBCL had good reliability coefficient alpha (0.83) [13], which is slightly lower than in our preschool CBCL (0.95). Total problem coefficient alpha of 0.95 is highly similar with those documented in the USA (0.95) [3], China (0.93) [21], and Taiwan (0.95) [20]. The coefficient alpha for “withdrawn” and “attention problems” were slightly lower than in other studies [3, 20, 21], perhaps because in this population emotional behaviours are considered less serious than disruptive behaviours. This finding may suggest that some items describing withdrawn and aggressive behaviours are understood differently in Kenya than in the USA.
Seven-syndrome structure and fit indices
Our Confirmatory Factor Analysis, implemented with structural equation modelling, supported the seven-syndrome CBCL structure, whose fit indices were acceptable. In particular, the standardised factor loadings are comparable to the ranges provided by Achenbach and Rescorla who first validated the CBCL in the USA [3]. The slightly smaller loadings in a few items in our study (withdrawn and attention problems) are in part explained by performing polychoric (for 3-point response scales) rather than tetrachoric (for 2-point response scales) item correlations; the former is deemed appropriate for the CBCL but may be associated with lower factor loadings [18]. The few items with very low standardised coefficients may have been misunderstood by parents and should be investigated further in future studies before they omitted from future assessments using CBCL to examine behavioural/emotional problems in Kenyan populations. All RMSEA and most CFI and TLI indices suggested an acceptable to good fit for the seven-syndrome CBCL structure in our population. In particular, our overall RMSEA of 0.035 is better than the 0.06 from the USA [3], 0.053 from China [6], 0.055 in Taiwan [20] and up to 0.059 from 23 other societies [6], probably because we allowed item error terms to correlate [19]. These findings support configural invariance of the CBCL and its application across diverse societies, including rural Kenya. Since the internal structure of the CBCL in this population is satisfactory, future studies can evaluate other properties, in particular the predictive validity as these children grow older [11].
Strengths and limitations
The strength of this study is the careful translation of the CBCL into the local languages and use of trained and experienced field assistants to administer the tool. Training of fieldworkers by one psychologist and comparison of their scoring for concordance before collection of the CBCL data helped avoid introduction of inter-rater bias. The sample size was acceptable to run confirmatory factor analysis and to determine overall internal consistency. The sample size may however have been small for some sub-analysis. Withdrawn and attention problems scales were associated with low internal consistency. Test–retest reliability and interinformant agreement were not performed for subscales of the CBCL, since these scales had low scores which were skewed, and these factors would overinflate the correlation coefficients. The derived cut-off score doesn’t represent a clinical diagnosis of a mental health problem since it is based on a random rather than a normative sample.