WITHDRAWN: Fixed Effects Thinking in a Variable Effects World and What It Means for How We Study Child Welfare Disparities

Disparity within the child welfare system is something we need to better understand from a scientific perspective. Because science has a contribution to make to the larger conversation about disparity (e.g., placement rate disparity) and its causes, it is important to get the science right. In this workshop, we will unpack the ways in which disparity is typically studied, discuss the reasons why from a causal perspective the traditional approach is too narrow, and then illustrate how the study of disparity might be broadened empirically. This broadened view will deepen our causal understanding of why disparity persists thereby opening the way to new, previously untried remedies.

The focus of our workshop is on causal models that invoke the structural roots of disparity. If, as many people believe, system structures differentially affect people because of their racial and ethnic identities, then those structures have to be identified, measured, and examined for their impact on outcomes. This is a substantial empirical challenge if, in the end, the goal is to understand how system forces such as structural bias generate disparate outcomes within the child welfare system.

As a research problem, the key to understanding structural explanations rests with seeing disparity as both a dependent and independent variable in the context of the same statistical model. As an independent variable, disparity is associated with what might be called the race effect. In traditional studies of disparity, race and/or ethnicity is seen as a way to stratify outcomes. Statistical equations that follow this logic allow for observations of this type: All else being equal, Black children are more (or less) likely to be placed following a substantiated report of maltreatment. When disparity is on the dependent variable side of the equation, the structure of the argument changes. Rather than acting as the cause of an effect, when it shifts to the other side of the model, disparity becomes the outcome of structural processes or the effect of causes. Equations that follow this logic allow for this type of observation: In the presence of certain structural conditions, disparity tends to be more prominent. This is the logic of structural explanation. Disparity is the result of these structural conditions.

Therein lies the challenge. How does one set up a single statistical model that treats disparity as both an independent and dependent variable? In our workshop, we will illustrate how this is done. We start with the data structure. In order to accomplish the substantive goal, the data must be structured in a particular way. Once the data are properly structured, we will describe the statistical models that treat disparity as both a left- and right-side variable. We will close the workshop with real-world examples using both longitudinal and cross-sectional designs. Because our interests are organized around structural explanations, we will use the study designs to highlight how system structures including poverty and supply-induced service demand are related to spatial and temporal variation in disparity rates.