As enablers of change, we sometimes find ourselves navigating the interesting landscape of evaluation and data analysis. Within this space, two foundational concepts emerge: causation and correlation. In this episode, we explore the subtle yet critical difference between these concepts.
Causation is a critical concept for us as enablers of change. Causation involves establishing a direct cause-and-effect connection between two variables. In other words, when one variable changes, it triggers a change in another. A simple example is that when it gets hot, the thermostat turns on the air conditioner to cool the air. The increase in temperature is the cause and the resultant cool air is the effect. In agriculture, think about the relationship between water availability and crop yield. When water is limited, crop yield tends to decrease. Here, the availability of water is the cause, and the subsequent change in crop yield is the effect.
However, there’s another concept we need to keep in mind—correlation. Correlation is a statistical relationship between two variables. Correlation suggests that changes in one variable tend to coincide with changes in another. But correlation does not mean causation. While there might be a correlation between two variables, it doesn’t necessarily mean that changes in one are driving changes in the other.
Let’s illustrate this distinction between causation and correlation, by considering the often-used example of ice cream sales and drowning incidents. There appears to be a positive correlation between ice cream sales and the number of drownings. When ice cream sales go up, drownings also rise. However, it’s evident that ice cream consumption is not causing people to drown. The common factor here is the weather: both ice cream sales and swimming activities increase during the hot, sunny summer months. So the hot weather causes increased ice cream sales and more people to be swimming, hence the increased number of drownings. While there is a casual relationship between the hot weather and ice cream sales, there is only a correlation between ice cream sales and drownings.
Let’s imagine that our colleagues have come to us excitedly saying they have realised that when a particular pest is present in a crop, the crop yield decreases. They’d like us to start working with growers to eliminate the pest. They think that all we’d need to do is tell growers we can help them increase their crop yields! Well, it sounds nice in theory but as many enablers of change will tell you, there’s a whole range of factors that affect crop yield. Weather conditions, soil health, and farming practices could contribute to both lower yields and increased pest presence. We’d want to ask our colleagues to do some more work on the problem before we started a pest management extension programme.
Understanding the subtle difference between causation and correlation is important for our decision-making as enablers of change. So how do we tell what is causation and what is correlation? To establish causation, three criteria must be met. Firstly, temporal precedence dictates that the cause must precede the effect. In other words, any change in the cause variable must happen before any change in the effect variable. Secondly, a clear association should exist between the cause and effect variables. Lastly, alternative explanations for the observed relationship must be carefully considered and eliminated. These criteria form the foundation upon which causation is established.
For correlation, we need to understand the connection between variables. There are three common types of correlation: positive, negative and zero correlation. With positive correlation, both variables increase or decrease in tandem. For instance, as the amount of sunlight exposure increases, plant growth might also increase. However, this doesn’t imply that sunlight directly causes plant growth. With negative correlation, one variable increases as the other decreases, and vice versa. For instance, as the amount of pesticide usage decreases, insect damage to crops might increase. Yet, this negative correlation doesn’t automatically establish a causal link between pesticide reduction and increased insect damage. Finally, zero correlation occurs when variables show no apparent connection. For instance, there might be no discernible correlation between the number of farm advisors in a region and the average crop yield.
In our role as enablers of change, understanding causation and correlation helps us use data well. Correlation, with its ability to highlight potential associations, is a valuable tool for highlighting what we might want to ask more questions about or spend more time exploring or investigating. Causation helps us with the ability to make predictions and can highlight important next steps. If you’re interested to read more, check out the Australian Bureau of Statistics (n.d.).
Well, we’ve shared our insights on causation and correlation, and now we’re eager to hear from you! How have you encountered these concepts in your work? Do you have tips or thoughts to contribute? Join the conversation by leaving a comment below.
As we wrap up this Enablers of Change episode, we appreciate your engagement. To stay updated on future episodes, subscribe to our newsletter and spread the word among your network. Is it correlation or causation if you tell more people about us and we get more subscribers?!
We’ll let you decide on the answer to that question! Until next time, keep the conversation flowing and keep inspiring change!
Australian Bureau of Statistics (n.d.). Correlation and causation. Available online.