Validation of The Infant/Todler HOME Inventory in Rural Rwanda: Dimensionality, Item Calibration, and Differential Item Functioning

• In Event: Agency and empowerment in ECD

Proposal

Background: The home is defined as not only the physical setting in which children live but also the people, as well as the physical and emotional resources available that support children’s holistic development (Elardo & Bradley, 1981; Lansford et al., 2023; Totsika & Sylva, 2004). Considering the relevance of early stimulation, parenting practices, and the quality of the home environment for child development, several tools have been developed to capture a child’s home environment. The most widely used tool is the Home Observation for the Measurement of the Environment (HOME) (Bradley & Caldwell, 1976; Caldwell & Bradley, 2016; Elardo et al., 1975). The HOME is a simple, summative measure that uses both observational and caregiver-reported information to measure the quality of the emotional and physical environment of a child (Caldwell & Bradley, 2016) and is widely used in applied research, both as a predictor of child development (e.g., cognition, language, motor development, and academic achievement) and as an outcome measure in the context of early childhood intervention studies. Despite its widespread use and the existence of multiple adaptations, no peer-reviewed papers have been published reporting on the validity and psychometric properties of the use of the HOME inventory in Rwanda.
Objectives: This paper aims to generate psychometric evidence for a culturally relevant and statistically valid version of the Infant/Toddler HOME Inventory that early childhood researchers and practitioners could use in the Rwandan and potentially sub-Saharan Africa contexts. Specific objectives are: 1) to evaluate the psychometric properties and instrument dimensionality of a culturally adapted version of the Infant-Toddler HOME Inventory; 2) Balancing psychometric information and instrument content, identify a suitable set of items to be used in the Rwandan context; 3) to evaluate Differential Item Functioning of the HOME inventory items, and 3) to assess convergent, concurrent, and known-group balidity of the Rwandan version of the HOME inventory.
Methods: Data was collected in Rwanda as part of a Cluster Randomized Trial designed to evaluate the effectiveness of a 12-week home-visiting parenting intervention, Sugira Muryango. The intervention is intended to develop parental skills through playful parenting, father engagement, care-seeking behaviors, and family functioning to promote early childhood development among the most vulnerable families in rural Rwanda (Betancourt et al., 2020; Jensen et al., 2021). The sub-sample of interest corresponds to the baseline assessments of 1084 children 6 to 36 months (M=21, SD=8.4, 50.2% Female) from three Rwandan districts (Nyanza, Ngoma, and Ruvabu). The item-level analysis were conducted by inspecting item means, variances, and point-biserial correlations. Scale reliability was assessed using the Kuder-Richardson Formula 20 (KR-20), scale dimensionality was evaluated using Horn’s Parallel Analysis, and Confirmatory Factor Analysis was also used to assess the existence of method effects. Instrument calibration was performed using a 2PL Item Response Theory (IRT) model, uniform and non-uniform Differential Item Functioning was evaluated using the Logistic Regression Approach (De Ayala, 2009). Convergent validity was evaluated using Pearson’s correlations between HOME IRT scores, OMCI, MICS-FCI, and MDAT scores. Known-group validity was tested using logistic regressions to evaluate if HOME IRT estimates influence the likelihood of developmental delays, as captured by ASQ-III western cut-off scores.
Results: The 43-item HOME inventory revealed acceptable shared variance (KMO = 0.75) and reliability estimates (KR − 20 = 0.74, average inter-item correlation = 0.06). However, some items showed extremely high mean values, low variability, and negative or very small point-biserial correlations. Seven items with corrected point-biserial values of < 0.1 were removed. Parallel analysis of the remaining 36 items suggested the extraction of only two factors, contradicting the original instrument's hypothesized six-factor structure. A two-factor Exploratory Factor Analysis (EFA) showed six items with factor loadings below 0.25 and cross-loadings; therefore, they were removed from further analysis. More importantly, the structure of factor loadings revealed potential method effects as items comprising factor 1 were either caregiver-reported or observed, while factor 2 comprised only observed items.
The reduced 30-item scale was subjected to further testing. The reliability of the reduced scale improved (KR-20 =0.78). No items showed point-biserial <0.1. Subsequent parallel analysis and EFA confirmed the two-factor structure and factor loadings also pointed to potential method effects. A CFA model (factor 1: caregiver-reported or observed items; factor 2: observational items only) was fitted to test for method effects by correlating error terms of items sharing the same assessment mode in factor 1. Results showed very good model-fit statistics, significant loadings, and a larger, significant correlation between factors. Results highlighted two issues. First, currently available psychometric evidence does not support the existence of six original subscales. Second, the 2-factor structure points to the potential impact of method effects from data collection as it appears to reflect the items’ administration mode rather than two different home-environment factors or dimensions.
Given the strong internal consistency of the reduced 30-item HOME inventory and the relatively high correlation between the factors, a unidimensional 2PL IRT model was used for calibration. Results revealed desirable item difficulty and discrimination parameters, very good item-level and model fit (RMSEA = 0.06; SRMR = 0.07), and acceptable empirical reliability (ρ = 0.81 ). DIF testing comparing children’s sex revealed minimal differences, with only one item showing uniform DIF and four items showing non-uniform DIF. However, according to common effect size measures, these differences are negligible. Validity testing of the reduced 30-item scale showed strong convergent validity evidence (significant positive associations between HOME IRT scores and MICS-FCI and the OMCI. Results also showed strong concurrent validity evidence, demonstrated by statistically significant positive associations between HOME and MDAT scores. Known-group validity evidence highlighted that HOME scores significantly predicted the likelihood of developmental delays across different domains.
Conclusion: Having a statistically validated instrument is not only critical from a psychometric standpoint but also from a policy perspective, as multiple policy efforts are currently in place to promote early childhood development in the country, reflected in strong governmental commitment through policy efforts such as the comprehensive ECD Policy established in 2016 and the National Early Childhood Development Program (NECDP) Strategic Plan 2018-2024 (Placencio-Castro et al., Under Review).

Authors

• Matias Placencio-Castro, Boston College Research Program on Children and Adversity

• Sarah Jensen, Boston College

• Theresa S. Betancourt, Boston College