The purpose of this study was to measure the bias on a binary option's effect estimate that appeared in the types of questions asked and in the placement changes of public service…
Abstract
Purpose
The purpose of this study was to measure the bias on a binary option's effect estimate that appeared in the types of questions asked and in the placement changes of public service users.
Design/methodology/approach
The author designed Monte Carlo simulations with the analytical strategy of latent trait theory leveraging a probability of care-placement change. The author used difference-in-difference (DID) method to estimate the effects of care settings.
Findings
The author explained the extent of discrepancy between the estimates and the true values of care service effects in changes across time. The time trend of in-home care for the combined effect of in-home care, general maturity, and other environmental factors was estimated in a biased manner, while the bias for the estimate of the incremental effect for foster care could be negligible.
Research limitations/implications
This study was designed based on individual child-unit only. Therefore, higher-level units, such as care setting or cluster, county, and state, should be considered for the simulation model.
Social implications
This study contributed to illuminating an overlooked facet in causal inferences that embrace disproportionate selection biases that appear in categorical data scales in public management research.
Originality/value
To model the nuance of a disproportionate self-selection problem, the author constructed a scenario surrounding a caseworker's judgment of care placement in the child welfare system and investigated potential bias of the caseworker's discretion. The unfolding model has not been widely used in public management research, but it can be usefully leveraged for the estimation of a decision probability.