|
Abstract
|
OBJECTIVE
Child welfare involvement begins with unscripted or semi-scripted narratives reported to intake specialists by professional and nonprofessional reporters. These narratives determine how to respond to reports of child abuse and neglect. Child welfare reporting varies drastically based on geographic region and is known to have a disproportionate impact on children of color. This paper explores how textual analysis can be used to study language differences between urban and rural jurisdictions, report source, and how language use differs according to the child's race.
Methods
Reports on child abuse and neglect from State Fiscal Year 2019 in a Midwestern state's public child protection services agency were used. Analysis was conducted on 145,423 reports using sentiment analysis. Every word was assigned a sentiment score. Report scores were calculated using the average sentiment of all words within the report. Each report's overall score was used to evaluate whether sentimental language differed across geographic location, report source, and child race/ethnicity.
Results
Textual analysis shows that language differs between urban and rural reports (p = 0.033) and the language used by report sources to report on children of different races/ethnicities is significantly different across groups (p = 0.00). However, these variables only explained 4% of the variation in sentiment value with an effect size of D = 0.01.
Conclusions
Textual analysis offers child welfare jurisdictions a way to explore variation within reports across different geographic areas, report sources and children's race/ethnicity. For sentiment analysis to be more applicable to child welfare, a child welfare specific dictionary should be developed to ensure each word is considered within the context of child abuse and neglect.
Language: en |