Reverse Causality

The text below is a transcript of the video above, edited with AI assistance for clarity and flow. Where the two differ, the video is the authoritative version.

You’ll learn to

You have learned about confounding variables, one of the key reasons why correlation doesn’t imply causation. Another key concept for understanding why correlation doesn’t imply causation is reverse causality. In this video you will learn to:

  • Define reverse causality.
  • Identify relationships where the causal direction is ambiguous.
  • Explain why correlations alone cannot tell us which way causation runs.

A familiar correlation

Let’s get back to the correlation between healthcare spending and carbon dioxide emissions. We have seen that the two are positively correlated: countries that spend more on healthcare are also countries that emit more carbon dioxide. One interpretation could be that spending more on healthcare causes more emissions.

In the previous video, we saw that national wealth is a potential confounder that we should pay attention to before we draw this type of conclusion about causality between these two variables, healthcare spending and greenhouse gas emissions.

A second problem with the same correlation

A second problem with the same correlation between healthcare spending and carbon dioxide emissions is that we could have the direction of causation backwards. When we look at carbon dioxide emissions and healthcare spending, the causation could go the other way. That is reverse causality.

Definition

Reverse causality occurs when:

  • You see a correlation between X and Y, where X is the independent variable and Y is the dependent variable.
  • You hypothesise that causation goes from X to Y, or you identify the correlation and conclude that there is a causal relationship between the two variables.
  • But it could actually be Y, your dependent variable, that is causing the independent variable. Or the causal chain could run both ways, so that the two variables are mutually causing one another.

Causation running both ways is also something that can happen in social, political and economic phenomena.

The reversed pathway: healthcare and emissions

In the healthcare and emissions example, it is plausible to think that higher emissions would lead to more pollution-related illness. This would generate a greater demand for healthcare, and therefore countries would spend more on healthcare if they pollute more. So emissions could be driving spending, and not the other way around.

Both directions may be plausible

Both directions may be plausible. You could have wealthier, higher-emitting countries also investing more in healthcare, but you could also have higher emissions generating the healthcare needs. You could hypothesise either way, and the data alone cannot tell us which causal direction dominates in this case.

Example 1: Education and democracy

Let’s look at another example: education and democracy, a country-level correlation. Countries where more people have higher education degrees are more likely to be democratic than countries where fewer people have higher education.

More educated populations may demand democratic governance. But democracies may also invest more in public education. So when you observe that more educated countries are more democratic, that is also a situation where you can’t tell that there is a single causal pathway. Both causal directions could be at work in this relationship.

Example 2: Policing and crime

Take policing and crime, at the community level. You could have more police deterring criminal activity, but you could also have higher crime rates prompting increases in policing. It is really difficult to tell the causal direction between policing and crime.

The core difficulty

The core difficulty is that whenever we are looking at observational data, data that we observe and collect, such as data about countries’ levels of education and democracy, or their healthcare spending and carbon dioxide emissions, we can see that X and Y, the independent and the dependent variable, move together. But that type of data cannot tell us whether X drives Y, Y drives X, or both. Establishing a causal direction is going to require more than a correlation.

Recap

  • Reverse causality happens when you are assuming that X causes Y, that your independent variable causes your dependent variable, but in truth the causal direction could be different: Y causing X instead of X causing Y.
  • The healthcare spending and emissions example is one in which you could have reverse causality. The direction of causality between those two variables could run both ways.
  • Correlations alone cannot resolve which direction causation runs.

Why this matters

Causal claims based on correlations appear in public debate very often, and you should be able to question and assess those claims. Understanding confounding variables is one way of doing so. Understanding reverse causality, and asking yourself whether it could be happening whenever you encounter a causal claim, is also very important for evaluating quantitative evidence critically.