Impact statement
Access to reliable and culturally appropriate screening tools for mental health disorders in children is a key challenge, particularly in low- and middle-income countries.
The Strengths and Difficulties Questionnaire (SDQ) is a 25-item instrument structured into five scales assessing children’s internalizing behaviors (emotional symptoms and peer relationship difficulties), externalizing behaviors (conduct problems and hyperactivity/inattention) and prosocial behaviors. The availability of freely accessible versions in numerous languages and cultural contexts has contributed to the widespread use of the SDQ as a behavioral screening tool for children aged 2–17 years.
In sub-Saharan African populations, the SDQ has been primarily applied to high-risk groups (orphans and HIV/AIDS populations), mainly from Southern and Eastern Africa. However, only few studies have rigorously evaluated the psychometric properties of the SDQ in community samples.
We examined the reliability and validity of SDQ scores in a sample of 566 school-aged children from semi-rural communities in Benin (West Africa). Parents responded orally to SDQ items presented by trained local interviewers in their primary language (Fon). We examined the internal consistency and factorial structure of the SDQ items, and the association of scores with known socioeconomic and health risk factors.
The distribution of SDQ scores was skewed toward higher values compared with the UK reference population. The Total Difficulties and Externalizing Behavior scores showed acceptable reliability and external validity. In contrast, the Internalizing score showed weaker internal consistency, partly reflecting the difficulty of translating constructs such as peer-relationship difficulties across cultures.
Our findings suggest that the SDQ P should be used and interpreted with caution when assessing young children from community samples in sub-Saharan West Africa. Rigorous cross-cultural adaptation (including qualitative interviews, back-translations and field testing) is warranted. In clinical or screening contexts, region-specific normative data should be established, and application of Western-derived cutoff scores avoided.
Introduction
Latest estimates indicate that between 10% and 20% of children and adolescents worldwide experience a clinically relevant emotional or behavioral difficulty, and roughly one in seven 10- to 19-year-olds has a diagnosed mental health disorder (Kieling et al., Reference Kieling, Baker-Henningham, Belfer, Conti, Ertem, Omigbodun, Rohde, Srinath, Ulkuer and Rahman2011; UNICEF, 2021; WHO, 2025). These concerns are particularly present in sub-Saharan Africa, where the proportion of young people is the highest worldwide, and cumulative risks linked to poverty, exposure to infectious diseases, collective trauma and limited mental health resources substantially increase children’s vulnerability (Jorns-Presentati et al., Reference Jorns-Presentati, Napp, Dessauvagie, Stein, Jonker, Breet, Charles, Swart, Lahti, Suliman, Jansen, Van, Seedat and Groen2021). In such contexts, identifying children with emotional and behavioral difficulties from childhood onwards is crucial to prevent the emergence of mental health difficulties in adolescence and adulthood (Sharp et al., Reference Sharp, Venta, Marais, Skinner, Lenka and Serekoane2014; Jorns-Presentati et al., Reference Jorns-Presentati, Napp, Dessauvagie, Stein, Jonker, Breet, Charles, Swart, Lahti, Suliman, Jansen, Van, Seedat and Groen2021). However, most of the available screening tools (multi-item questionnaires) have been developed in Western, high-income countries, and their reliability and validity in other cultural contexts is a growing concern (Diener et al., Reference Diener, Oishi and Lucas2003; Jukes et al., Reference Jukes, Gabrieli, Mgonda, Nsolezi, Jeremiah, Tibenda and Bub2018; Brown et al., Reference Brown, Weiss, Suntheimer, Appiah, Aurino and Wolf2025). Indeed, children’s behavior is shaped directly by their sociodemographic environment and related cultural models that impact socialization goals and perceptions by the parents and families (Keller et al., Reference Keller, Lamm, Abels, Yovsi, Borke, Jensen, Holub, Lo, Tomiyama, Carmiol, Baruil, Salazar, Arguedas, Ruiz, Solamo, Carvajal, Fai, Papaeliou, Papadopoulos, Apostolou, Kopakaki, Nikolakaki, Vrettopoulou and We2006; Jukes et al., Reference Jukes, Gabrieli, Mgonda, Nsolezi, Jeremiah, Tibenda and Bub2018).
The 25-item Strengths and Difficulties Questionnaire (SDQ) (Goodman, Reference Goodman1997, Reference Goodman2001; Goodman et al., Reference Goodman, Ford, Simmons, Gatward and Meltzer2000) is the most widely used behavioral screening tool measuring emotional and behavioral difficulties and competencies for 2- to 17-year-olds (Goodman, Reference Goodman1997, Reference Goodman2001; Goodman et al., Reference Goodman, Ford, Simmons, Gatward and Meltzer2000; Hoosen et al., Reference Hoosen, Davids, de Vries and Shung-King2018). Initially developed in the United Kingdom, the SDQ consists of five scales, each composed of five items (Supplementary Table S1): Hyperactivity/Inattention, Emotional Symptoms, Conduct Problems, Peer Relationship Problems and Prosocial Behavior. Each item is rated as 0 (“not true”), 1 (“somewhat true”) or 2 (“certainly true”). The score for each scale is obtained by summing the scores of its five items (after prior score inversion of the positively formulated items), resulting in a scale score ranging from 0 to 10. A Total Difficulties score is derived as the sum of the scores of the five scales, ranging from 0 to 40, with higher scores indicating more problems. Additionally, the Emotional Symptoms and Peer Relationship Problems scales can be summed into an Internalizing behavior score, and the Conduct Problems and Hyperactivity/Inattention scales into an Externalizing behavior score. Cutoffs for the total difficulties and internal and external behavior difficulty scores have been proposed by the developers in the UK population to identify cases of children with behavior difficulties (Goodman, Reference Goodman1997). The SDQ’s brevity, free accessibility and availability in multiple informant formats (parent, teacher and self-report) have led to its worldwide growing use (Warnick et al., Reference Warnick, Bracken and Kasl2008). Validated cross-cultural versions are currently available in over 90 languages (https://www.sdqinfo.org/). In 2018, a scoping review identified 54 studies using the SDQ in African populations (12 countries) (Hoosen et al., Reference Hoosen, Davids, de Vries and Shung-King2018). Most of these studies have primarily focused on high-risk groups such as orphans or children affected by HIV/AIDS, predominantly in Southern and Eastern Africa (South Africa being overrepresented). This scoping review highlighted inconsistencies in the SDQ’s use in Africa, and limited research on methodological issues related to translation, cultural equivalence and administration to low-literacy populations (Hoosen et al., Reference Hoosen, Davids, de Vries and Shung-King2018; Nkuba et al., Reference Nkuba, Hermenau, Goessmann and Hecker2018). At that time, only one study had conducted a full psychometric validation (internal consistency and factorial structure) of the teacher version of the SDQ (Kashala et al., Reference Kashala, Elgen, Sommerfelt and Tylleskar2005). After Hoosen’s review, a dozen studies have examined the psychometric properties of the SDQ in SSA (South Africa, Nigeria, Congo, Ghana, Uganda and Zimbabwe), five of them focusing on young children assessed by their parents (SDQ-P version) (Sharp et al., Reference Sharp, Venta, Marais, Skinner, Lenka and Serekoane2014; Nakigudde et al., Reference Nakigudde, Bauta, Wolf and Huang2016; Mellins et al., Reference Mellins, Xu, Nestadt, Knox, Kauchali, Arpadi, Kvalsvig, Shrout and Davidson2018; Lai et al., Reference Lai, Small, Parchment and Lau2023; Brown et al., Reference Brown, Weiss, Suntheimer, Appiah, Aurino and Wolf2025). The others involved adolescents using the self-report version (SDQ-S) (Skinner et al., Reference Skinner, Sharp, Marais, Serekoane and Lenka2014; Adeniyi and Omigbodun, Reference Adeniyi and Omigbodun2017; De Vries et al., Reference De Vries, Davids, Mathews and Aarø2018; Aarø et al., Reference Aarø, Davids, Mathews, Wubs, Smith and de Vries2022; Lovero et al., Reference Lovero, Adam, Bila, Canda, Fernandes, Rodrigues, Sander, Mellins, Duarte, dos Santos and Wainberg2022; Katus et al., Reference Katus, Olalekan and Oyekola202 Reference Boivin, Zoumenou, Sikorskii, Fievet, Alao, Davidson, Cot, Massougbodji and Bodeau-Livinec6). Most of them reported a low internal consistency for certain dimensions (Peer Relationship, Hyperactivity/Inattention and Conduct) and overall poor fit to the original factorial structures. Poor fit and unclear factorial structure reflect construct validity issues potentially caused by cultural differences. Some of the adapted items may incorrectly reflect the intended underlying trait, and measurement errors are consequently expected. Before using the SDQ in clinical and epidemiological studies in the SSA context, a rigorous examination of its measurement properties in the targeted population is thus necessary. To date, no study has evaluated the validity and reliability of the parent-report version of the SDQ (SDQ-P) in school-aged children from a West African general population. Benin is a French-speaking SSA country, characterized by relative political stability and a low incidence of HIV infections compared to other SSA countries. This study involved 566 school-age children from a semi-rural general population in southern Benin, with the objective of testing whether the SDQ-P was a relevant and reliable tool to assess children’s socio-emotional behaviors and informing its future use in similar contexts.
Population and methods
Population
Our study population consisted of children born to pregnant women initially included from 2009 to 2010 in a trial comparing preventive malaria treatment (MiPPAD – Malaria in Pregnancy Preventive Alternative Drugs) in Benin. Mothers were enrolled during their second trimester of pregnancy (<28 weeks) and followed up until delivery (N = 1,027 live-born singletons). To be included, mothers had to be HIV negative at recruitment and have no history of severe renal, hepatic, psychiatric or neurological disease. These broad inclusion criteria suggest that the study population is representative of the general population. Two follow-ups were organized 1 and 6 years after birth to assess the development of the children. At age 6 (June 2016 to October 2018), 580 children were examined by psychologists and nurses specifically trained to administer neurodevelopmental tests, in two centers – Sékou and Attogon. Both are located in southern Benin, about 40 km from the economic capital, Cotonou, and 10 km apart from each other. More details about the study can be found in previous publications (Boivin et al., Reference Boivin, Zoumenou, Sikorskii, Fievet, Alao, Davidson, Cot, Massougbodji and Bodeau-Livinec2021; Garrison et al., Reference Garrison, Boivin, Khoshnood, Courtin, Alao, Mireku, Ibikounle, Massougbodji, Cot and Bodeau-Livinec2021). Among the 580 participating children at 6-year follow-up, 7 children were excluded for medical conditions (severe neurological disorders identified from their local medical records) and 566 had completed the SDQ questionnaire (Supplementary Figure S1). All caregivers of the children (mostly mothers, but also fathers, other family members or legal guardians) provided informed consent. Illiterate mothers gave their consent by fingerprint.
Assessment of children’s behavior
French is the official language in Benin, but the population from the Allada–Sekou area is mainly influenced by the Aja-Fon cultural and linguistic sphere, and many residents speak Fon and related languages. As no officially licensed version of the SDQ in Fon was available, a common oral translation of the French version of the SDQ-P was developed collectively under the supervision of a native speaker mental health professional (RZ), without formal translation/backtranslation process. Because of the high prevalence of illiteracy in the population, the SDQ-P was administered to caregivers by an interviewer using the local language (Fon). A pilot version of the questionnaire was tested by the field workers with 15 children of similar ages and their caregivers. All interviewers attended the pilot interviews in order to fix a common way to administer the questionnaire (wording, instructions and help without influencing the answers).
Covariates
At the 6-year follow-up, caregivers provided information about the socioeconomic environment of the child, including familial structure (single mother, married in a monogamous or polygamous situation), maternal level of education (schooled or unschooled at the primary school level at least), literacy and primary language spoken. A composite index was used to summarize the family’s socioeconomic status (SES) based on material possessions (radio, television, bike, motorbike and car), possession of cows and access to electricity (Koura et al., Reference Koura, Boivin, Davidson, Ouédraogo, Zoumenou, Alao, Garcia, Massougbodji, Cot and Bodeau-Livinec2013). The Edinburgh Postpartum Depression Scale was used to identify mothers with depression symptoms at 6 years of the child (EPDS score > 12) (Cox et al., Reference Cox, Holden and Sagovsky1987). At the 1-year follow-up, the level of cognitive stimulation and emotional support available to the child during infancy was evaluated using the Caldwell HOME Scale (Home Observation Measurement of the Environment) (Bradley and Caldwell, Reference Bradley and Caldwell1976). The BMI z-score at the age of SDQ assessment was also calculated (WHO references) (De Onis et al., Reference De Onis, Onyango, Borghi, Siyam, Nishida and Siekmann2007) as a proxy for the nutritional status of the child, and severe episodes of malaria during infancy with hospitalization were asked.
Statistical analyses
Distribution of the SDQ items and scores
The distributions of the answers to SDQ items were represented in bar plots to identify the most prevalent/absent behaviors in the population. Scores of the five scales, Internalizing and Externalizing scales and Total Difficulties were compiled using the algorithm provided by the developers (https://www.sdqinfo.org/).
Internal consistency and factorial structure of the SDQ questionnaire
The Cronbach alpha coefficient (Cronbach, Reference Cronbach1951) was calculated for the five scales, the Internalizing and Externalizing scales and the Total Difficulties scores. We stratified the internal consistency analysis on several characteristics of the mother–child dyads to identify specific contexts associated with lower or higher reliability of the SDQ. The inter-item correlations matrix was examined and submitted to a Confirmatory Factorial Analysis (CFA) to check the structural validity of the SDQ. We tested the three dominant factorial structures proposed in the literature: the original five-factor structure (conduct problems; hyperactivity/inattention; emotional symptoms; peer relationship problems; prosocial behavior), the three-factor structure (internalizing behavior, combining emotional symptoms and peer relationship problems); externalizing behavior (combining hyperactivity/inattention and conduct problems; and prosocial behavior); and the second order model, where the internalizing and externalizing factors are placed on a second level (Goodman et al., Reference Goodman, Ford, Simmons, Gatward and Meltzer2000; Goodman et al., Reference Goodman, Lamping and Ploubidis2010). Considering the ordinal nature of the SDQ items, model parameters were estimated with the Weighted Least Squares Mean and Variance adjusted (WLSMV) method. Items were standardized and factor variances fixed to be 1, so that factor loadings can be interpreted as correlations between items and factors. Model fit was evaluated using several fit indices: the Comparative Fit Index (CFI), Goodness-of-Fit statistic (GFI), Root Mean Square Error of Approximation (RMSEA) and Standardized Root Mean Square Residual (SRMR) (Goretzko et al. Reference Goretzko, Siemund and Sterner2024). Model fit was considered acceptable with RMSEA <0.07, CFI > 0.90, GFI > 0.95 and SRMR <0.05 (Hu and Bentler, Reference Hu and Bentler1999; Hooper et al., Reference Hooper, Coughlan and Mullen200 Reference Bradley and Caldwell7). Standardized factor loadings ≥0.30 were considered as minimally meaningful, and ≥ 0.50 as significant, in accordance with established practices (Tabachnick and Fidell, Reference Tabachnick and Fidell2001; Hair et al., Reference Hair, Black, Babin, Anderson and Tatham2010). Expecting, from previous studies in the literature, a potential lack of fit, we additionally ran a five-factor exploratory factor analysis (EFA) with oblimin rotation to relax the original structural hypotheses and allow cross-loadings between latent factors. The objective was not to propose an alternative form for the SDQ in Benin, but to document and better interpret the structural inconsistencies we observed.
Predictors and external validity of the SDQ scores
Associations of the SDQ scores with characteristics of the children and their familial socioeconomic context were studied to identify predictors and get some evidence about their external validity. The effects of sex and age of the child were first tested together in a linear regression model, then each other maternal and child characteristic was additionally included one by one in a multivariate linear regression model (sex, age and the covariate of interest). Missing values were not imputed; no multiple testing adjustment was applied, as our approach was exploratory and covariates are likely to be correlated, so that the regressions were not independent (Benjamini and Hochberg, Reference Benjamini and Hochberg1995). The residuals of the regression were graphically checked for normality.
Moreover, the first impact question proposed in the SDQ tool, in addition to the 25 individual items (“Overall, do you think that your child has difficulties in one or more of the following areas: emotions, concentration, behavior or being able to get on with other people?”) was dichotomized (Yes/No) and compared to the SDQ scores using a logistic regression model. Odds Ratios (ORs) and Areas Under the Operating Characteristic Curve (AUC) identified which SDQ scales were mostly considered problematic by the families, and Chi-square tests identified the most predictive items of the impact question.
Results
Population
The study included equal numbers of boys and girls (50%) at a median age of 6.1 years (IQR: 6–6.3, min: 5.5, max: 8.4), mainly from the Sekou region (74.6%) (Table 1). Thirty-four percent of them were underweight and 22.2% had experienced a severe malaria episode during infancy. The proportion of children currently attending school was 67.1%. Their mothers were mostly married (48% in monogamous and 46% in polygamous families), 64% spoke Fon and 21.6% Aïzo at home, 64% did not attend school (21.6% attended partially primary school) and 11.6% had experienced postnatal depression (EPDS > 12). The median SES score of the families was 5 (IQR = 4–9), and the median HOME score was 27 (IQR = 26–28).
Characteristics of the study sample (N = 566 children)
Table 1. Long description
The table is titled Characteristics of the study sample which includes.566 children. It consists of two columns: Characteristics and N with percentage or median with the first and third quartilesQ
Child Demographics:
* Site of inclusion: Attogon 144 25.4 percent; Sekou 422 74.6 percent.
* Child sex: Male 283 50 percent; Female 283 50 percent.
* Child age at SDQ completion: Median 6.1 Q 1 to Q 3 6.0 to 6.3.
* Child nutritional status: Underweight 183 34.0 percent; Normal 355 66.0 percent; Missing 28.
* Severe malaria episode during infancy: Yes 125 22.2 percent; No 439 77.8 percent; Missing 1.
* Child attending school: Yes 379 67.1 percent; No 186 32.9 percent; Missing 1.
Maternal and Household Factors:
* Maternal marital status: Single/Divorced/Widowed 33 5.9 percent; Married monogamous 271 48.4 percent; Married polygamous 256 45.7 percent; Missing 6.
* Number of children in household: 1 to 2 110 19.4 percent; 3 to 4 300 53.0 percent; 5 and more 156 27.6 percent.
* Maternal education: No schooling 359 64 percent; Partial primary 121 21.6 percent; Primary complete 21 3.7 percent; Secondary and more 60 10.7 percent; Missing 5.
* Maternal language: Fon 260 46.1 percent; Aïzo 290 51.4 percent; Other 14 2.5 percent; Missing 2.
* Socioeconomic status S E S score: Median 5 Q 1 to Q 3 4 to 9; Missing 3.
* H O M E Inventory: Median 27 2.2; Missing 32.
* Maternal E P D S: Median 5 Q 1 to Q 3 0 to 8; Missing 11.
* Informant: Mother 430 76 percent; Father 82 14.5 percent; Other family 54 9.5 percent.
S D Q Scores (Median, P 10, Q 1 to Q 3, P 90):
* Total difficulties: 16 7, 12 to 19, 21.
* Internalizing: 7 3, 4 to 9, 10.
* Emotional symptoms: 4 1, 2 to 5, 6.
* Peer relationship problems: 3 1, 2 to 4, 5.
* Externalizing: 9 4, 7 to 11, 13.
* Hyperactivity/Inattention: 5 2, 4 to 7, 9.
* Conduct problems: 3 1, 2 to 5, 6.
* Prosocial behavior: 6 5, 5 to 8, 9.
a For the SDQ scores, the median (10th percentile, 1st quartile - 3rd quartile, 90th percentile) are given.
Abbreviations: P10, 10th percentile; P90, 90th percentile; Q1, first quartile; Q3, third quartile.
Distributions of the scores
The SDQ questionnaire was mainly answered by the mothers (76%). Fathers represented 14% of the informants, the others being family members (grandmother, brother, sister, aunt and uncle) or legal guardians of the children. The distribution of responses to items is presented in Supplementary Figure S2. The median Total Difficulties score in the Beninese cohort was 16 (IQR = 12–19) (Table 1). The median scale scores were 4 (2–5) for Emotional symptoms, 3 (2–4) for Peer relationship problems, 3 (2–5) for Conduct problems, 5 (4–7) for Hyperactivity/Inattention and 6 (5–8) for Prosocial behavior. The distributions of the Total Difficulties, Internalizing and Externalizing scores in our sample are compared to the UK cutoffs in Supplementary Figure S3.
Internal consistency
In the inter-item correlation matrix, no within subscale correlations exceeded 0.50, except between the “has good friend” and “is generally liked” items from the prosocial subscale (r = 0.57). Many cross-correlations were simultaneously observed (e.g., the item “Often fights with other children or bullies them” from the Conduct subscale correlated over 0.30 with three items of the Hyperactivity subscale) (Supplementary Figure S4).
The internal consistency was moderate for the Total Difficulties score (α = 0.68), Externalizing (α = 0.65) and Hyperactivity/Inattention (α = 0.65), but lower for Internalizing (α = 0.52), Prosocial behavior (α = 0.55), Emotional symptoms (α = 0.51) and particularly low for Conduct problems (α = 0.34) and Peer relationship problems (α = 0.34) (Table 2). The Total Difficulties score appeared significantly more consistent in the Attogon center (α = 0.81 vs. 0.64 in Sekou). We found no differences in the consistency of the Total Difficulties score and subscales according to sex or nutritional status, nor to school attendance (except higher consistency of the prosocial score in children attending school). The socioeconomic status (SES) of the family was associated with higher internal consistency for all scales except Conduct problems (the lowest SES group having lower alpha values). Maternal education did not impact the reliability of the Total Difficulties score, but higher consistency was observed for Prosocial behavior and Emotional symptoms, while lower consistency of Conduct and Hyperactivity/Inattention scores was observed when mothers had a formal education. Maternal depression (EPDS≥12) was associated with lower reliability of the Total Difficulties score, Internalizing, Peer relationships and Conduct problems. Maternal and other informant reporting was associated with higher consistency of the Total Difficulties and Externalizing scales, compared to father reporting.
Internal consistency (Cronbach’s alpha coefficient) of SDQ scores according to child and mother characteristics
Table 2. Long description
The table presents internal consistency scores for the S D Q across eight scales: Total, Internalizing, Emotional, Peer, Externalizing, Hyperactivity, Conduct, and Prosocial.
* Item Counts: Total (20), Internalizing (10), Emotional (5), Peer (5), Externalizing (10), Hyperactivity (5), Conduct (5), Prosocial (5).
* Overall Sample (N = 566): Total alpha is 0.68. Subscale alphas range from 0.34 (Peer and Conduct) to 0.65 (Externalizing and Hyperactivity).
* Center: Attogon (N = 144) shows a higher Total alpha of 0.81 compared to Sekou (N = 422) at 0.64.
* Sex: Boys and girls show similar Total alphas (0.68 and 0.69 respectively).
* School Attendance: Both groups (Yes N = 380, No N = 186) show a Total alpha of 0.69 and 0.68.
* Nutritional Status: Underweight (N = 183) alpha is 0.66; Normal weight (N = 355) alpha is 0.70.
* Severe Malaria: Those with a history (N = 125) have a Total alpha of 0.74; those without (N = 441) have 0.66.
* S E S Quartiles: S E S 1 (lowest) has the lowest Total alpha at 0.61, while S E S 2 and 3 are both 0.70.
* Maternal Education: No education (N = 364) alpha is 0.69; Yes education (N = 202) alpha is 0.68.
* Maternal Depression: No (N = 500) alpha is 0.68; Yes (N = 66) alpha is 0.57, with a notably low Peer score of 0.08.
* Informant: Mothers (N = 430) alpha is 0.61; Fathers (N = 82) alpha is 0.65; Others (N = 54) alpha is 0.70. Fathers reported a negative Conduct coefficient of minus 0.03.
Abbreviations: EPDS, Edinburgh Postpartum Depression Scale; SES, socioeconomic score (quartiles).
Factorial structure
Using CFA, the five-factor and three-factor models did not achieve acceptable fit indices relative to the recommended thresholds, with GFI values of 0.846 and 0.799, respectively, and CFI and TLI indices not exceeding 0.72 (Supplementary Table S5). The RMSEA values ranged from 0.117 to 0.134, while the SRMR values ranged from 0.132 to 0.152. The five-factor model performed better than the three-factor model. Moreover, the two-stage model failed to converge, probably because the algorithm could not estimate stable higher-order loadings, as the covariance structure among first-order factors was too weak or inconsistent.
In the five-factor model, the Hyperactivity/Inattention and Emotional symptoms factors had consistent acceptable loadings (>0.30) for all the expected items (Table 3). The highest loadings were observed within the Hyperactivity/Inattention factor, with three items exceeding 0.50 (0.70 for “Easily distracted, concentration wanders,” 0.63 for “Constantly fidgeting or squirming” and 0.55 for “Restless, overactive, cannot stay still for long”). Loadings within the Emotional scale were lower and homogeneous (0.42–0.48). The Conduct scale showed only two items with acceptable loadings (“Generally obedient, usually does what adults request”: 0.48, “Often lies or cheats”: 0.39). The Peer relation scale presented two items with high loadings (“Has at least one good friend”: 0.68 and “Generally liked by other children”: 0.84), but the item “Picked on or bullied by other children” had an unexpected negative loading (−0.20). The Prosocial scale presented all expected loadings above 0.30 (0.76 for “Kind to younger children”), except the “Helpful if someone is hurt, upset or feeling ill” item (0.25).
Confirmatory Factor Analysis of the SDQ questionnaire (25 items): Standardized factor loadings (N = 550 complete questionnaires, among 566)
Table 3. Long description
The table presents Confirmatory Factor Analysis results for 550 questionnaires. It is organized by broad categories (Externalizing, Internalizing, and Prosocial) which are further divided into sub-factors.
Externalizing category:
* Hyperactivity sub-factor (Items 2, 10, 15, 21, 25): Loadings range from 0.37 to 0.78 in the Five-factor model and 0.28 to 0.82 in the Three-factor Externalizing column. Item 15 (pdistrac) has the highest loading at 0.78 and 0.82 respectively.
* Conduct sub-factor (Items 5, 7, 12, 18, 22): Loadings range from 0.27 to 0.55 in the Five-factor model. In the Three-factor Externalizing column, loadings range from minus 0.14 to 0.64.
Internalizing category:
* Emotion sub-factor (Items 3, 8, 13, 16, 24): Loadings range from 0.46 to 0.54 in the Five-factor model and 0.05 to 0.31 in the Three-factor Internalizing column.
* Peer sub-factor (Items 6, 11, 14, 19, 23): Loadings range from minus 0.21 to 0.94 in the Five-factor model. In the Three-factor Internalizing column, items 11 and 14 show strong negative loadings of minus 0.79 and minus 0.86.
Prosocial category:
* Prosocial sub-factor (Items 1, 4, 9, 17, 20): Loadings range from 0.21 to 0.86 in the Five-factor model and 0.15 to 0.93 in the Three-factor Prosocial column. Item 17 (pkind) shows the highest loading at 0.86 and 0.93.
Note: Bold values indicate loadings over 0.50. Italics indicate negative values. Estimates use W L S M V.
Note: WLSMV estimates. Factor loadings with absolute value over 0.50 are in bold, and negative factor loadings are in italics.
The three-factor structure showed consistent loadings within the Externalizing behavior factor, except for three items from the Conduct scale that correlated poorly (“Steals from home, school or elsewhere,” “Generally obedient” and “Often lies or cheats”). Conversely, the Internalizing behavior factor presented many highly inconsistent loadings, including high negative ones (“Has at least one good friend” [−0.69] and “Generally liked by other children” [−0.83]) (Table 3).
The EFA results (Supplementary Table S6) suggested several groupings of items from different originally expected scales. On Factor 2, the existing Hyperactivity/attention items grouped with “Often fights with other children or bullies them” (from Conduct scale) and “Picked on or bullied by other children” (from Peer relationship scale). Factor 1 combined items from the Peer relationship and Prosocial behavior (“Has at least one good friend,” “Generally liked by other children,” “Shares readily with other children” and “Kind to younger children”). Factor 4 combined items from the Emotion subscale, but with moderate loadings. Factors 3 and 5 presented more erratic loadings and were difficult to interpret.
Predictors of SDQ and external validity
Girls presented significantly lower Externalizing symptom scores (OR = −0.63 [−1.20; −0.07]) (Figure 1 and Supplementary Table S2). Older children showed significantly higher Total Difficulties and Internalizing scores. The maternal mental health emerged as the strongest predictor of SDQ scores: children whose mother was depressed (EPDS ≥12) had markedly elevated scores across Total Difficulties (OR = 3.61 [2.31; 4.91]), Internalizing (OR = 1.56 [0.82; 2.29]) and Externalizing (OR = 2.06 [1.18; 2.93]) domains. Children not attending school presented higher Total Difficulties (OR = 1.25 [0.34; 2.16]) and Externalizing scores (OR = 1.02 [0.41; 1.62]). A socioeconomic gradient was observed: children from higher SES backgrounds and raised in stimulating caregiving environments (as measured by HOME score) had lower externalizing symptom scores. Conversely, children whose mothers had no formal education exhibited higher Total Difficulties and Externalizing scores. Children living in households with three or more children tended to have lower Total Difficulties, Externalizing, and Internalizing scores, though not significantly. Children whose caregivers spoke Fon had lower Externalizing scores. Underweight was not associated with differences in SDQ scores, while children who had experienced at least one severe malaria episode during infancy exhibited significantly higher Externalizing scores. The informant of the interview (mother, father or another person) was not significantly associated with the scores, but fathers tended to give lower scores than mothers.
Associations between SDQ scores and characteristics of the mother–child dyads.
Abbreviations: EPDS, Edinburgh Postpartum Depression Scale; HOME, Home Observation Measurement of the Environment; Q1, first quartile; Q2, median; Q3, third quartile; Q4, fourth quartile; SES, socioeconomic score (quartiles).
Interpretation: Blue squares (Total Difficulties), red diamonds (Externalizing behavior) and green triangles (Internalizing behavior) show differences in SDQ scores predicted by regression models (SDQ as dependent variable; each characteristic as independent variable), adjusted for sex and age, with their 95% confidence intervals. Confidence intervals not including 0 indicate significant differences. For binary characteristics (site, sex, underweight, malaria, child at school, number of children at home [≥3], maternal education, language spoken, maternal depression and perceived impact), the figure shows differences in scores for the indicated category relative to the reference. For categorical variables (marital status, SES quartile and informant), the reference categories were “married monogamous,” median quartiles (“Q2 and Q3”) and “mother,” respectively. For continuous characteristics (child age and HOME score), differences in scores correspond to changes in the SDQ score for a one-unit increase in the characteristic.
Figure 1. Long description
A forest plot with a horizontal x-axis labeled Differences in S D Q scores with 95% C I, ranging from negative 2 to 5. A vertical dashed line marks the zero baseline. The y-axis lists 18 characteristics. Data is plotted using blue squares for Total Difficulties, green triangles for Internalizing, and red diamonds for Externalizing behaviors.
Key findings from top to bottom:
* Site SEKOU vs ATTOGON: Total and Externalizing scores are significantly higher in SEKOU, while Internalizing score does not differ.
* Child sex: Total and Externalizing scores are significantly lower in girls, while Internalizing is slightly higher.
* Child age: All three scores increase significantly with age.
* Underweight child and Severe Malaria episode: No differences were significant .
* Child not at school: Shows positive trends for all scores, with Total Difficulties being significantly higher.
* Single or Married polygamous mother: Most markers are positive but cross the zero line.
* 3 plus children at home: All scores are significantly lower.
* No maternal education and Aizo spoken: Markers are generally positive but mostly non-significant.
* S E S Q 1 and Q 4: Q 1 shows near-zero effects, while Q 4 shows significant lower scores for Total and Externalizing behavior.
* Maternal depression (E P D S greater than 12): Shows the strongest significant positive association across all three categories, with Total Difficulties peaking near 3.5 over non-depressive mothers.
* H O M E score: Markers are tightly clustered just below zero.
* Informant (Father or Other vs Mother): Markers are generally negative, indicating lower reported scores than mothers.
* Perceived difficulties (Impact): Shows significant positive associations for all the SDQ scores.
Among the parents/caregivers who answered the first impact question (N = 559 out of 566), only 10% considered that their child had behavioral or emotional difficulties. Reporting difficulties was significantly associated with higher SDQ Total Difficulties, Externalizing, Hyperactivity/Inattention and Emotional scores (Supplementary Table S3). The items most significantly associated with perceived difficulties were: “Many fears, easily scared,” “Thinks things out before acting,” “Often complains of headaches, stomach-aches or sickness,” “Sees tasks through to the end, good attention span,” “Picked on or bullied by other children” and “Constantly fidgeting or squirming”) (Supplementary Table S4).
Discussion
This study is the first to examine the psychometric properties of the SDQ-P scales in a community-based sample of 566 children (mean age: 6.3 years) from Benin, in Western Sub-Saharan Africa. Our results show some measurement issues that suggest a cautious use of the SDQ-P in similar contexts and further work on cross-cultural adaptation.
Internal consistency and factorial structure
As reported in previous studies in children from SSA of similar ages, the internal consistency of the SDQ-P scores in our population was low to moderate, which reflects an overall lack of correlation between items. The Conduct and Peer relationship scales were particularly affected, while the Hyperactivity/Inattention scale was the only one close to a satisfactory level of consistency. Expectedly, a higher consistency was obtained in participants from higher SES and education, while maternal depression reduced the consistency of most of the scales severely.
We did not replicate the expected five- or three-factor structures of the SDQ in our cohort and identified some inconsistencies in the factor loadings, as observed in previous studies conducted in sub-Saharan African children of similar ages (Skinner et al., Reference Skinner, Sharp, Marais, Serekoane and Lenka2014; Nakigudde et al., Reference Nakigudde, Bauta, Wolf and Huang2016; Mellins et al., Reference Mellins, Xu, Nestadt, Knox, Kauchali, Arpadi, Kvalsvig, Shrout and Davidson2018; Lai et al., Reference Lai, Small, Parchment and Lau2023; Brown et al., Reference Brown, Weiss, Suntheimer, Appiah, Aurino and Wolf2025). They mainly reported problematic loadings for the Peer relationships scale but also for the “Hyperactivity/Inattention” and “Conduct” scales. Contrarily, Emotion has been identified as a reliable and valid scale (Mellins et al., Reference Mellins, Xu, Nestadt, Knox, Kauchali, Arpadi, Kvalsvig, Shrout and Davidson2018; Brown et al., Reference Brown, Weiss, Suntheimer, Appiah, Aurino and Wolf2025). In our study, the most consistent scales were Hyperactivity/Inattention and Emotion. The high consistency of the Hyperactivity/Inattention scale in our population was supported by convergent reporting of the behaviors “Restless, overactive, cannot stay still for long,” “Constantly fidgeting or squirming” and “Easily distracted, concentration wanders.” These items were also the most frequently reported by the parents. As in other cultures, externalized behaviors are easier to identify, are more disturbing and caregivers do not hesitate to mention them. In many epidemiological studies using the SDQ, the Internalizing and Externalizing aggregated scales are preferably used instead of the five initial scales. In our cohort, the Externalizing scale was consistent, thanks to its Hyperactivity/Attention components that also correlate highly with the item “Often fights with other children or bullies them” from the Conduct scale. However, three other items from the Conduct scale (“Generally obedient,” “Steals from home, school or elsewhere” and “Often lies or cheats”) loaded highly negatively. These items were also the least frequently reported by the parents. The Internalizing scale presented erratic loadings, mixing moderate associations with items depicting socio-emotional difficulties such as “Picked on or bullied by other children,” “Many fears, easily scared” and “Often unhappy or tearful” and highly negative loadings with two items reflecting popularity and ability to have friends. As a consequence, the Internalizing score may be a “noisy” measure of socio-emotional difficulties, and using the Emotional subscale may be preferred to focus on emotional difficulties.
External validity
Despite measurement issues, the SDQ scales presented expected associations with the socioeconomic context and characteristics of the children that support its validity in our population. Consistently with the literature, externalizing difficulties were more reported in boys (Kovess-Masfety et al., Reference Kovess-Masfety, Husky, Keyes, Hamilton, Pez, Bitfoi, Carta, Goelitz, Kuijpers, Otten, Koç, Lesinskiene and Mihova2016), and internalizing difficulties at later ages. Also, as expected, school attendance was associated with lower total and externalizing scores, and children from higher SES backgrounds, those exposed to stimulating home environments, had lower externalizing symptom scores, while children of mothers without formal education showed elevated total difficulties and externalizing symptoms (Mieloo et al., Reference Mieloo, Raat, van Oort, Bevaart, Vogel, Donker and Jansen2012; Cortina et al., Reference Cortina, Fazel, Hlungwani, Kahn, Tollman, Cortina-Borja and Stein2013; Goemans et al., Reference Goemans, van Geel, Vedder and Bradley2016). We also found a strong association between maternal depression and higher SDQ-P scores that confirmed the central role of caregiver mental health in shaping children’s behavioral and emotional outcomes, but also in perceiving and reporting their children’s difficulties (Beck, Reference Beck1999; Nadeem et al., Reference Nadeem, Rafique and Chachar2018).
Overall, children in the Beninese cohort presented high SDQ scores, compared with the original UK norms (Goodman, Reference Goodman1997). The median Total Difficulties score (16) corresponded to the UK 90th percentile, and the medians of both Internalizing and Externalizing scores were above the 75th UK percentile. Similar upward shifts in the distributions have been consistently reported in previous Sub-Saharan African studies with children from the general population at similar ages (Kashala et al., Reference Kashala, Elgen, Sommerfelt and Tylleskar2005; Nakigudde et al., Reference Nakigudde, Bauta, Wolf and Huang2016; Mellins et al., Reference Mellins, Xu, Nestadt, Knox, Kauchali, Arpadi, Kvalsvig, Shrout and Davidson2018), or adolescents (De Vries et al., Reference De Vries, Davids, Mathews and Aarø2018; Nkuba et al., Reference Nkuba, Hermenau, Goessmann and Hecker2018). Higher scores may reflect a real higher prevalence of behavioral problems in children, in contexts where poverty, community violence, familial instability, under-nutrition or epidemics are commonly present (Cortina et al., Reference Cortina, Sodha, Fazel and Ramchandani2012). However, compared to the many studies conducted within clinical (mainly HIV+) or disadvantaged populations (orphans and violence-affected children), our cohort was recruited from the general population in a region characterized by political stability over the last decades and the absence of social violence related to ethnicity or religion, but poverty is present. However, despite high scores derived from the SDQ items, 90% of the parents did not perceive any emotional or behavioral difficulty for their child (as reported in the impact question), which suggests some discrepancies between their perception of behavioral difficulties and their use of the SDQ-P items (and their answer modalities) to describe them. The Hyperactivity/Inattention and Emotion scales were the most correlated with the impact question, suggesting their higher relevance for families to describe behavioral difficulties. As observed in other cultures, they correspond to behaviors that are more visible and disturbing (hyperactivity) for the families or notably distressing (somatic complaints and emotional symptoms) around 6 years of age, with also gender-differential expectations (Mieloo et al., Reference Mieloo, Raat, van Oort, Bevaart, Vogel, Donker and Jansen2012).
Altogether, our exploration of psychometric properties of the SDQ points to the Peer relationship and Conduct scales as more problematic subscales. Both are highly culture-dependent, and some measurement inconsistencies observed with cross-culturally adapted tools may be explained by differences between societies based on individualism (mostly high-income Western countries where tools have been developed) versus collectivism (more present in traditional low- and middle-income countries) (Henrich et al., Reference Henrich, Heine and Norenzayan2010). The rigorous methodologies that have been developed to guide cross-cultural adaptations tend to consider socio-emotional behaviors as universal (etic constructs), whereas the culture-specific relevance of the latent traits measured should be questioned, shifting to an emic approach (Berry, Reference Berry1969; Kleinman et al., Reference Kleinman, Eisenberg and Good2008). The potential content issues we observed in our cohort, through a lack of correlations within some of the originally defined scales, suggest further qualitative investigations about caregivers’ perceptions of child normative and maladaptive social and emotional behavior in the Beninese context, through interviews, as done in Zambia or South Africa by Menon et al. (Reference Menon, Glazebrook, Campain and Ngoma2007) and Skinner et al. (Reference Skinner, Sharp, Marais, Serekoane and Lenka2014).
Beyond construct issues, the lack of reliability we observed in our cohort may also be caused by the insufficient proximity of the informants to the child they assessed (mother/father vs. family members, or legal guardians). Moreover, some caregivers reported to the interviewers that they did not know much about their child’s social interactions occurring outside the home, at school or within communities where children navigate autonomously under the collective responsibility of several families. Communal child rearing and free-range social networking are common in SSA societies, while tools have been developed in Western high-income countries where nuclear family is the reference, impacting their content. A more comprehensive multi-informant assessment including teacher’s version of the SDQ would thus improve the reliability of the assessment of children’s difficulties (Hoosen et al., Reference Hoosen, Davids, de Vries and Shung-King2018).
Strengths and limitations
This study is the first conducted in Benin to explore several key psychometric properties of the SDQ. It included a large sample size of young children from a semi-rural region. The detailed interviewer-administered SDQ assessment was done by professionally trained local staff and provides critical reference data for both research and clinical application of the SDQ-P in school-age children from that region.
However, several limitations must be considered. Despite the availability of authorized adaptations of the SDQ in many languages, no version in local Beninese languages was available and a common translation from French to Fon had to be done. Although validated by field professionals, the version we used did not result from a formal cross-cultural adaptation process. Moreover, due to low caregiver literacy, all instruments were administered orally by a trained interviewer. This mode of administering the SDQ questionnaire was not anticipated by the developers, and its potential impact was not assessed. An interviewer-induced social desirability bias may occur, orienting answers that conform to perceived social norms. Despite efforts to harmonize the items’ formulations and administration, some residual interviewer effect exists that cannot be disentangled from site and language effects: interviewers each worked at a different site , which also differed in terms of language (Fon was spoken by 90% of the participants in Attogon, while by only 30.8% in Sekou, with 66.4% speaking Aïzo) and socioeconomic context (Attogon being a wealthier environment). The differences in scores we observed between both centers likely reflect the cumulative impact of socioeconomic disparities. They also suggest that our results in these semi-rural communities in the South of Benin may not be generalizable to the whole country, especially because multiple cultural and linguistic influences are present.
Finally, other psychometric properties of the SDQ-P should be tested in this population, such as reproducibility, concurrent validity with other reference tools and sensitivity to change, depending on its expected use (e.g., cross-sectional comparisons of subpopulations or evaluating changes in scores after an intervention). If the objective is to screen for emotional and behavioral dysfunction in children, the UK-based norms should not be applied and regionally relevant reference thresholds should be developed, as Kashala et al. (Reference Kashala, Elgen, Sommerfelt and Tylleskar2005) proposed in Congo.
Conclusion
Our study provides important data to inform the appropriate use and interpretation of the SDQ in sub-Saharan child populations. The Total Difficulties, Externalizing Behavior and Hyperactivity/Inattention scales demonstrated acceptable reliability, whereas the aggregated Internalizing scale showed major inconsistencies because of low convergence between the Emotional Symptoms and Conduct subscales. The cross-cultural relevance of the Peer Relationship items warrants further investigation. Our results suggest that measurement errors are likely present when using the SDQ scores, in particular for the Internalizing behavior aggregated scale. As a consequence, epidemiological studies including the Internalizing SDQ scale in similar populations may suffer from reduced statistical power due to bias toward the null in estimated associations. UK reference norms should not be used for screening purposes and context-specific thresholds have to be developed.
Open peer review
To view the open peer review materials for this article, please visit http://doi.org/10.1017/gmh.2026.10344.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/gmh.2026.10344.
Acknowledgments
The authors would like to thank the entire staff of the health centers in Benin involved in the study (Allada, Attogon and Sékou) and all the participating families along the follow-ups of this cohort study.
AI statement
While finalizing their manuscript, the authors used ChatGPT (OpenAI, 2023) and Claude (Anthropic) to assist with the formulation and clarification of certain sentences. The AI played no role in the study design, data collection, analysis or interpretation of the data.
Author contributions
Conceptualization: F.B.L., N.C.; Formal analysis: J.K., N.C.; Funding acquisition: F.B.L.; Investigation: R.Z., M.J.B., N.C.; Methodology: M.J.B., N.C.; Project administration: F.B.L.; Software: J.K., N.C.; Supervision: M.J.B., J.W., F.B.L., N.C.; Validation: R.Z., M.J.B., J.W.; Writing – review and editing: R.Z., J.K., M.J.B., J.W., F.B.L., N.C.
Competing interests
The authors declare none.
Data availability statement
The MiPPAD datasets analyzed in the present study are not publicly available as they contain information that could compromise the research participant’s privacy/consent and are property of the research institutes. However, they are available on reasonable request to the cohort steering committee, after examination according to the French ethical rules and the European individual data protection regulations.
Ethics and approvals
Ethical approval of the study was obtained from the Ethical Research Committee of the Institute of Applied Biomedical Sciences (CER-ISBA), No. 87 of May 17, 2016 in Benin, and the Ethics Advisory Committee of the Institute of Research for Development (IRD) in France.
Funding statement
This research was funded by the Fondation de France (Grant # 2015 00060746).
Patient consent statement
All caregivers of the children provided informed consent. Illiterate mothers gave their consent by fingerprint.
Open access
Comments
Dear editor,
We are pleased to submit our manuscript as an original research article in Global Mental Health.
This manuscript reports the assessment of the reliability and validity of the Strength and Difficulties Questionnaire (SDQ) within a mother-child cohort conducted in Benin (sub-Saharan, western Africa). Though widely used all around the world to assess children’s behavior, the psychometric properties of the SDQ remain insufficiently documented for its use in populations that differ massively from the culture where it has been developed by Goodman et al. in UK (1997). Only a few studies reported formally psychometric properties and most of them have been conducted within elder children (adolescents), in clinical contexts (HIV patients) or political contexts (civil war, orphans, …).
We assessed the consistency and validity of the SDQ in a sample of 566 6-year-old children. The internal consistency of the SDQ was tested analysis of the inter-items correlations and their structure (with Cronbach, and confirmatory and explanatory factor analyses). The associations of the scores with socio-economic characteristics of the child and its family allowed to document some validity of the instrument.
Our main results are as follows:
• The Total Difficulties, and Externalizing Behavior scores achieved acceptable reliability and external validity, despite moderate inter-item correlations overall.
• The Internalizing score showed weaker scale consistency, partly because it reflects less visible constructs such as emotions and peer-relationships that are harder to translate across different cultures
• For screening purposes, specific thresholds for normality should be redefined using local reference samples of children
• In epidemiological studies, potential measurement errors should be considered when interpreting the Internalizing score
We feel that all these elements support the interest of our study for a publication in Global Mental Health. Access to reliable and culturally appropriate screening tools for mental health disorders in children is a key challenge, particularly in low- and middle-income countries. This work adds useful information for the future users of the SDQ in similar populations.
This work has not previously been published, either in whole or in part, and no similar paper is in press or under review elsewhere.
None of the co-authors has any conflict of interest relating to this work.
Yours sincerely,
Nathalie Costet