Debates and Challenges in Causality Analysis: What Key Messages Have We Learned in the Past Half Century?

Objective: Marking the publication of Donald Rubin's seminal article on the potential outcome model in 1974, the field of causality analysis has witnessed half a century of significant development. Concurrently, social work research has undergone transformative changes, driven by the adoption of sophisticated causal inference methods in response to the increasing demand for evidence-based practice. Essential to all scientific inquiries, causality analysis has experienced profound advancements, notably marked by the contributions of three Nobel laureates. This workshop will explore the complex debates and evolving challenges within causality analysis, highlighting its significant implications for advancing social work science.

Methods: We have carefully selected 16 seminal papers from distinguished scholars in econometrics, statistics, and artificial intelligence, including Nobel laureates James Heckman, Joshua Angrist, and Guido Imbens, as well as pioneers Donald Rubin and Judea Pearl. Our discussion will critically assess how these influential contributions address core issues in social work research, with particular emphasis on selection bias in social justice evaluations, "what-if" scenarios in intervention research, and the integration of AI into contemporary social work research and practice.

Results: Our comprehensive literature review offers nine key insights: 1. Angrist's instrumental-variable approach is fundamental in addressing endogeneity, underscoring the importance of selecting robust instruments for social work research. 2. Heckman's sample selection model is vital for correcting hidden selection bias, highlighting the need for meticulous testing of hidden selection threats in social work evaluations. 3. Rubin's potential outcome model remains a dominant framework for causality studies, emphasizing the construction of valid counterfactuals to effectively address "what-if" questions. 4. Abadie and Imbens's matching estimators are particularly effective in empirical social work projects focusing on causality, providing robust solutions to complex data challenges. 5. Kernel-based matching and the "difference-in-differences" estimator by Heckman and colleagues address temporally invariant biases, crucial for studies with geographically or demographically distinct samples. 6. Hirano & Imbens's generalized propensity score model expands the range of research questions in social work, facilitating the evaluation of programs with non-binary treatment conditions. 7. The innovative methods recommended by Imbens and Wooldridge, including propensity score subclassifications, weighting, and matching estimators, offer invaluable tools for tackling a broad spectrum of challenges in social work evaluations. 8. The diversity of models illustrates that no single approach is universally applicable; it is essential for researchers to critically evaluate the assumptions and potential violations in practical applications to ensure accurate interpretation of findings. 9. Pearl's structural causal model, with its seven tools, provides a comprehensive framework for developing new AI tools, demonstrating the convergence of causality analysis and AI applications, which promises substantial benefits for social work research.

Conclusion and Implications: Causality analysis has not only advanced significantly but also created vast opportunities for social work science to effectively address contemporary challenges. This workshop will promote a critical examination of these advancements, encouraging a nuanced understanding and application of advanced methods in social work research, thereby enhancing the impact and relevance of social work interventions in real-world settings.