What if it were the other way around? Could it also work in that way?

These are the questions you should always ask once you have developed a hypothesis on the relationship between two variables. Reverse your thinking: is there a possibility that the association you’re thinking about is generated by a chain of events that runs in the exact opposite direction? Usually there is. As a challenge, take a correlation and try to come up with a line of reasoning that explains why the association came about with a chain of events that is the other way around.

Here’s an example: the relationship between the homicide rate due to firearms and the level of generalized trust in the US (Wu, 2020). The analyses show a negative relationship: citizens who have been threatened with a gun are less trusting, and people residing in states with higher levels of homicide in a given year have lower levels of trust in that same year. The key figure is below.

What if we switch this graph? Perhaps people carry guns because they think that most people cannot be trusted. Then they may think it is better to protect oneself with a gun, and use it if they feel threatened.

The article does not mention the possibility that low trust increases the likelihood that citizens carry firearms and use them. In fact, the article does not discuss the possibility of reverse causality at all. The article makes the implicit argument that the survey question “Have you ever been threatened by a gun or shot at?” refers to events in the past, while the question “Do you think most people can be trusted” is about the current day. The chronological order of the target periods in these two questions is that the threat comes first, and the distrust came later.

Also the article reports analyses that respondents ever having been shot or threatened with a gun are less trusting regardless of when the threat occurred: in childhood or adulthood. This finding suggests that exposure to gun violence has long lasting effects that do not dissipate over time. These findings, however, do not rule out reverse causality. Even when gun violence has long term negative effects on trust that do not dissipate over time, it is likely that people with less trust are more likely to carry guns themselves, have friends and family with guns, live in areas with more guns, and are more likely to encounter gun violence, precisely because they are less trusting.

Neither does the possibility of reverse causality mean that the causality must run in the other way, or that the association in the first graph must entirely be due to the reverse pathway. The point is that it cannot be ruled out, and that the reverse pathway also makes sense. Without an effective strategy for causal inference we just don’t know.

This is just one example. Next time you see a correlation, or when you are developing a set of hypotheses for a study, switch sides. Reverse the perspective to your advocate of the devil, who thinks things are exactly the other way around. For instance when you see a correlation between the murder rate of a country and the proportion of people agreeing that most people can be trusted.

Related posts: Design your test to be as stringent as possible, A critical approach to the quality of previous research, Composing a causal model, and Omitted variables

Reference

Wu, C. (2020). How does gun violence affect Americans’ trust in each other? Social Science Research, 91, 102449. https://doi.org/10.1016/j.ssresearch.2020.102449