Modifications to the Devereux Early Childhood Assessment Factor Structure for Application to Child Welfare Populations

Background and Purpose

The Devereux Early Childhood Assessment (DECA) measure is frequently used across populations to assess early childhood social-emotional developmental needs and guide practice recommendations. One such population is that of infants, toddlers, and preschoolers in foster care. However, the DECA instruments were normed on the general population. Whether the originally derived factor structure is replicable with the child welfare population is unknown. This study utilizes practices similar to those employed for the original factor analysis study with data derived from the child welfare population. The primary objective is to discern whether the factor structure emerging from an analysis of child welfare data resembles the original factor structure to inform the use of the instrument with this population.

Methods

Data and Samples

The data utilized for this study represent 320 infants, 335 toddlers, and 502 preschool-age children. The DECA scales were completed for these children an average of 38 days after placement into foster care. In most cases, the instruments were completed by foster parents with guidance from child developmental assessment professionals. These data were collected in a total of 85 counties in a Midwest state between 2017 and 2019.

Methodological Procedures

MPlus was utilized to conduct an Exploratory Factor Analysis. Orthogonal rotation utilizing the varimax method was selected for two reasons. First, the factors derived based on the original analysis can be considered to be uncorrelated based on theoretical considerations. Second, this was the method utilized in the original analysis. Scree plots, eigenvalues, and item information were used to identify the number of factors present. Model fit was evaluated using Chi-square and RMSEA statistics. When the model fit was poor, items were deleted one by one based on factor loadings and conceptual considerations, and the model was re-run.

Results

The factor structure derived for each instrument departed from the published factor structures. Model fit indices for the derived structures were generally poor. For each instrument, individual items failed to load on any factor (Infants: 1 item failed to load and 8 items cross-loaded; Toddlers: 6 items failed to load and 9 items cross-loaded; Preschoolers: 6 items failed to load and 12 items cross-loaded). In some cases, items loaded on different factors than were initially identified. In the Infant model, subsetting the sample to include only infants older than two months and reducing items replicated the original factor structure. For the Toddler model, item reduction replicated the original factor structure. The fit for the Preschool model, however, departed significantly from the original structure. The Attachment/ Relatedness factor did not appear, while two new factors - Sadness and Attentional Difficulties - emerged.

Conclusions and Implications

In conclusion, the assumption that instrumentation normed on the general population can be expected to behave similarly in the child welfare population may not be warranted. This population may differ systematically from the general population due to experiences of abuse and neglect. Thus, attention to reformulation of measures for this population could improve the utility of these measures for screening and referral purposes.