Correlation vs Causation: Understanding the Difference to Drive Uplift and Save Spend
People like to make connections between actions - it’s in our nature to try to organise the chaos. When looking at consumer behaviour, it’s particularly important for marketers to join the dots to uncover the actions that drove a conversion, but it is often done without considering whether the conversion took place despite the action or because of it.
To illustrate correlation (where two events occur together) vs causation (where an event happens as a direct result of another), let’s pretend that you are going to write a report on fire safety. You conduct the research and are surprised to discover that the more firemen that attend a fire, the more damage the fire causes.
Statistically this is accurate, but logically it makes no sense. If you took this research at face value you would probably decide the best course of action would be to reduce the amount of firefighters attending emergencies – a big mistake.
Without incorporating any other factors into the data, you find that the number of firefighters fighting a fire does indeed correlate with the amount of damage caused by the fire, simply because the fire is bigger, and more firemen are needed.
The conclusion is wrong because correlation was mistaken for causation.
What does this mean for my marketing and my budget?
The Holy Grail for marketers is to know exactly what touchpoints influenced a user, and how much they affected their decision, for every sale. If you have a user on your site who abandons their basket then returns ten minutes later and converts – after you’ve served them a retargeting ad – you could argue that they were always going to buy. Did they actually see it?
You could also argue that your advert was the catalyst that caused the conversion.
Understanding the incrementality of your channels, and exactly how different actions and touchpoints positively impact the path to purchase, is crucial. Was the increase in conversion rate because of a new feature on your website, or was it the new content affiliate you’re working with? If you’re mistaking correlation for causation, then you could be wasting precious budget on advertising that isn’t actually driving any uplift.
Identifying the difference will enable you to concentrate on the activity that has a direct impact on consumers.
Using causality to improve attribution
Although there are different ways of describing the process, most attribution models are created by analysing user journeys to determine the effect that the presence of a particular touchpoint has on users’ propensity to convert. Put very simply: the higher the uplift, the higher the weighting and the higher the value attributed to that touchpoint.
This approach works generally but doesn’t separate correlation from causality. For example, just looking at propensity will often present brand PPC as the most effective channel – because it is present in a high percentage of converting journeys – so the attribution model will assign it a high degree of sale value. However, further analysis will show it being used as a way to navigate back to the site – correlating with the larger number of sales rather than causing them.
Attribution models need to include allowances and understanding of causality, which can only be developed from specific testing and analysis. Building these results into the attribution model will ensure all channel valuations are based on their real life impact.
How to get started with attribution
Attribution is no longer a nice to have, it’s a necessity. It side-steps assumptions and enables more accurate, data-driven decisions.
When we first work with new clients we use these methods to identify wasted media spend, which typically ends up around 20%. Removing spend from ineffective channels (or moving it within channels) will free up budget that can be used for experimenting with new activity, or moved to where you know it will be effective.
A good first step would be to explore how attribution can help your business. We have a library of attribution resources available on our website and you can register for our next webinar on The future of retail and impact of emerging technology, featuring Julian Burnett, CTO at TH_NK (previously John Lewis) and experts from House of Fraser, trendwatching.com and Style Psychology.