Causation Killed the Cat (almost)

My cat almost died last week. He’s a pseudo-exotic breed with diabetes, bad teeth, and total hearing loss that my family carried with us from the other side of the planet, and we are attached to him. Almost as traumatic as the near-fatal event itself, and the accompanying bills, has been the unsuccessful attempt to understand the causes of this particular crisis, and thereby prevent a relapse. Forgive me for starting with a personal anecdote. I thought it might help to remind us that causation is not simply an esoteric topic of debate for philosophers but an issue of relevance to many situations, some of which are life and death. During this week’s Research Methods Café Conversation, academics from fields as diverse as anthropology, education, political science, exercise science, and archeology raised vital points for making sense of causation and for how we go about determining and demonstrating it. This blog post will of necessity touch on only a sampling of these issues.

In engaging a cross-disciplinary group, it was important to open by asking whether there is a shared conceptualization of causation and whether one is in fact necessary for productive work. Causation has been important from ancient times, but has recently come under criticism as reifying a positivist worldview. In the conversation, an anthropology researcher suggested that the nature of causation, or at least our understanding of it, may differ substantially based on our research framework. Experimental and statistical methods have been championed as providing the most “robust evidence” (i.e. evidence of causation), yet this interpretation was contested among the conversation participants, even those active in such methods. As demonstrated in this “statistical rethinking,” numerical procedures cannot replace judgement, especially when considering alternative causal models. Of course, human interpretation plays an even more visible role in qualitative research, and this article argues that localized, qualitative work can illuminate causal relationships, sometimes better than quantitative designs. In addition to epistemological and methodological commitments, the type of entity that acts or is acted upon might also affect our understanding of causation. Several participants mentioned the idea of agency, which has special relevance for causation in social contexts. How should individuals’ choices, for example, be taken into account in causal models when those choices may reinforce societal norms?   

Causation also involves timelines of events. This reading discusses the challenges in forward and reverse causal inference: How did we get here (reverse) and what will follow from present conditions (forward)? The timeframe in which we consider causation connects to specific research purposes mentioned by participants, for example explanation of a given state of things versus the ability to predict the outcomes of actions, such as interventions. While an anthropologist emphasized that understanding the mechanisms at work can lead to better predictions, an education researcher asserted that predictive ability alone can be sufficient, especially if predictions hold despite the ever-shifting nature of the world. One theme of the conversation was that experimental and statistical methods can be very useful for identifying relevant causal elements and for making predictions, but the more explanatory approach to causation requires theories of some kind.

Appropriate modelling of causation was another major theme of the conversation. Many critical issues humans face operate at a global as well as local level and necessitate multilevel solutions. In the conversation, participants referred to “generalisability” or the application of research findings to larger-scale groups, as a perennial problem, and one which “big data” approaches alone cannot solve (see previous blog post). An anthropology researcher made reference to the tension between idiographic and nomothetic models of causation, and there was a concern voiced that individual and local needs may not be served well by statistical approaches. A participant from sport and exercise sciences pointed out that the systems we study are often more than the sum of their parts, and it is not clear how computational models could represent or explain irreducible complexity. An education researcher asserted that even if a model could perfectly explain past situations, it would fail at prediction because the world is constantly changing.

Related to the issue of research application, stakeholders often need to understand both global and local causes and effects in order to positively impact their systems. This contributes to the rationale for rich qualitative descriptions and explanations, as well as larger scale research on the same questions. Both of these can feed into theories and plausible interventions, which can then be evaluated in experimental and mixed-method designs. The idea that research operates best as a cycle involving work at different scales is not new and has been argued eloquently elsewhere. One newer research initiative commissions multiple investigations around a single topic in diverse, international settings to investigate how local processes affect results and to increase generalisability. Such multilevel and multimethod approaches to demonstrating causation should boost research impact.

Active researchers may be reluctant to engage with the causation debate, but sitting it out may be a luxury we do not have. A researcher in anthropology asserted that some kind of causal understanding is generally sought in any research effort, and other participants noted that sponsors often want research to provide unambiguous causal models. A desire for reliable solutions to pressing issues, rather than simply adding to knowledge, seems to be central to the causation push. It makes sense that stakeholders and researchers would want to invest in such solutions. On the other hand, several participants mentioned that real-world systems, whether physical or social, can be genuinely unpredictable and chaotic. Rushing to causal explanations may oversimplify the nature of these systems (but see here for a discussion). I do not believe we should therefore abandon our search for causation, but rather acknowledge that all models, and interventions based on them, are to some degree interim ones. The researchers I most admire all share a commitment to this kind of honesty.

Just as I want very much to understand the combination of factors necessary for my cat’s well-being, many people today seek to better understand economic, environmental, and health challenges around the world. This desire is not enough, however. As humans, we need to believe that our actions matter, and that the outcomes can be changed. The assumption of causation is at the heart of efforts to apply knowledge to practice, and even where that knowledge is imperfect and incomplete these efforts are having important effects.

Researchers interested in moving the causation discussion forward were encouraged to reach out to the Durham University Institute of Advanced Study with project ideas.

This blog post expresses the author's views and interpretation of comments made during the conversation. Any errors or omissions are the author's.

Loraine Hitt is an Ed.D. candidate in Durham University's School of Education.