Contemporary Causal Frameworks and Methods: Innovative Strategies to Infer Causality in Social Work Research

Significance. Social work research is conducted in the service of social change. Causal inference—assessing the effect that social interventions have on their designated outcomes—is central to this mission. Because randomized designs cannot always be used, social work researchers should be fluent in a number of quasi-experimental methods. To successfully implement these methods, a clear understanding of causal frameworks is required.  Connecting frameworks to methods aids the researcher in identifying the assumptions, strengths, and limitations of these methods.

Content. The workshop will focus on two elements of causal analysis: frameworks and methods. I will focus on three frameworks, each of which clarifies the assumptions needed to infer causality from the methods: Campbell’s model of validity, which enumerates nine threats to credible causal inference; the Rubin potential outcomes model, which uses a statistical framework to describe assumptions; and Pearl’s directed acyclic graph (DAG), a graphical approach based on a set of rules. I will discuss the important differences between these three frameworks, but I will also show how they are complementary.

For the methods, I will first review randomized control trials, which will provide a standard to aim for. The frameworks previously described provide a structure for understanding how these methods address the problems typical of quasi-experimentation, and therefore help us achieve this standard (or alternatively, how they may fall short in a given context). I will discuss six quasi-experimental methods that can replicate randomization if their assumptions are realistic for a given set of data. I will first review propensity score analysis, which is well known in social work. This will be followed by methods used more commonly in other disciplines: instrumental variables, regression discontinuity, interrupted time series, fixed effects, and difference-in-difference. I will show how each method can address typical problems of quasi-experimental design, but I will also show the weaknesses that, depending on the data and design, can represent serious challenges to causal inference. I will show how these methods can be supported by good study planning such as by collecting data over multiple time points, and obtaining rich mediator and covariate data, and how designs can be strengthened by combining methods. In addition, a brief step-by-step guide will be provided for each method; examples of studies will be provided to illustrate these concepts; and I will guide the audience through a typical regression discontinuity analysis (as an example). 

Implications. Using causal frameworks, social work researchers can better explain the assumptions, strengths, and limitations of methods used to draw causal inferences from studies given the design and data. These skills are important when designing and carrying out a study, as well as peer review of grant proposals and manuscripts. This workshop will support these needs by: describing several innovative quasi-experimental methods; providing a blueprint to selecting an appropriate method for a given design; describing a set of frameworks that provide the basis for the untestable assumptions involved in causal inference; and providing instructions on how to conduct each of the methods in social work research.