Marketing's collinearity problem

(From this week's Partner letter at The Hook - our Mutiny Newsletter. Sign up here: https://www.mutiny.group/subscribe)

One of the key problems we see emerging across the digital landscape is over-investment in digital performance channels at the expense of brand building. The narrative goes a bit like this:

• Channels that are easily measured have naturally more to report with sales
• As sales go up, these channels performance metrics go up
• Finance and marketing types invest more behind those ‘winning’ channels
• Marketing performance stagnates

We’re often asked about this, and it’s worth talking about why this happens — the statistical phenomena known as co-linearity. 

Firstly, what the hell is co-linearity? 

Texas A&M statistics department has this neat definition:

“Collinearity is a condition in which some of the independent variables are highly correlated. Why is this a problem? Collinearity tends to inflate the variance of at least one estimated regression coefficient, β?j . This can cause at least some regression coefficients to have the wrong sign.”

Most marketers won’t know what the hell co-linearity is. But in marketing data, co-linearity is rife. Search data for example will typically go up when sales go up, but are likely to have limited value in driving the sale. 

In short, co-linearity is a problem.

The dangerous thing is lots of platforms exploit co-linearity. Pixels and basic attribution modelling tries to confuse the activity with impact on action, leveraging co-linearity as a dirty trick to inflate certain platform’s contribution to your sales.

Solving co-linearity means understanding which channels have high co-linearity, understanding the dependencies that drive those channels and ranking them accordingly. In short, it becomes a complex task for most marketers to handle. 

Done properly, though, and it can uncover where your marketing dollars might be wasted –even when the pixels are telling you that you’re raking in the dollars.