Directed Acyclic Graph Construction through the Use of Theory and Research

Background and Purpose: Social work research is carried out with a focus on improving a person’s well-being. Towards this goal, causal inference is necessary for understanding how risk and protective factors influence functioning in intrapersonal and interpersonal contexts. Additionally, social work researchers develop interventions for social problems. In many of these situations, randomized controlled trials are not possible and we rely on observational data. Directed acyclic graphs (DAGs) offer a visual aid for identifying mediators and confounders for a hypothesized causal effect prior to analyzing observational data. To use the DAG framework, we must combine existing subject area knowledge, hypotheses about the causal effect, and knowledge about research design to exhaustively map out the confounding paths that can bias causal effect estimation. This is the first step of three in utilizing DAGs in social work research.

Methods: In this oral presentation, we present the assumptions upon which we graphed the causal relationship between a hypothesized cause and effect as well as how those variables are related to a set of hypothesized and known confounders for both a retrospective and a prospective research design. We grounded our DAGs in the conceptual understanding of the treatment and exposure and searched for possible confounders using relevant theories embedded in existing research. The first DAG comes from a retrospective study of secondary data and uses the Integrated Model of Minority Youth Development (IMMYD) to isolate the causal pathway of perceived racial-ethnic discrimination on academic maladjustment. The second DAG comes from a prospective study of the effect of precarious employment conditions on the health of immigrant workers in the informal service sector.

Results: In both DAGs, known associations between the exposure and outcome are identified from the literature to specify the minimal set of covariates necessary for a causal estimate. We also looked to rule out paths between variables, and to identify variables that might make bias worse if conditioned upon. For example, in the first DAG, guided by the IMMYD, peers are identified as a confounder of academic maladjustment. Conditioning on peer pressure blocks a confounding path between discrimination and academic maladjustment. Using existing data, we then identified variables associated with both discrimination and academic maladjustment. In the second DAG, precarious employment is understood in terms of a continuum of labor exploitation that immigrant workers are subjected to, especially those working in low-wage, informal employment arrangements, and there are numerous ways in which precarious employment is associated with health.

Conclusions and Implications: Conducting causal research is important for actualizing social work’s mission of improving the well-being of all people with particular attention to minoritized groups. From a social justice perspective, needs-based research approaches in observational studies are a necessary method of investigation. DAGs provide a valuable tool for estimating the causal effect in observational studies when randomization is not plausible due to practical or ethical limitations. The DAG framework enables casual inference with observational data by providing ways to adjust for relevant confounders in our model and minimizing the risk of inducing further bias.