One of the basic analysis angles for any digital marketer will be to see if there are any related patterns. For example, it’s safe to assume that if you up your CPC bid, the click through rate is also going to go up. When two data points like CPC bid and CTR both go up at the same time, we described this as having a positive correlation. And so, we’re going to focus on what correlation is, and how to interpret it.
Let’s first begin with an explanation of how correlation is actually measured. This will provide a bit more context. In the last paragraph, we gave an example of positive correlation – it’s also possible to have relationships with negative correlation. In other words, when one of the variables goes up, the other one moves in the other direction. Putting this into a marketing context, you may find that if your CTR goes up, then the number of impressions drops, because of limited budget.
Measuring correlation uses a scale that ranges from -1 to 1. If the two variables you’re analysing have a strong correlation, either positive or negative, the result will be closer to the extremes. With this figure, it’s possible to determine the strength of the relationship, and also the direction that it goes in. A positive result of 1 represents a positive correlation, whereas the negative equivalent of -1 shows a negative correlation. Results that are in and around 0 suggest that there is no correlation between the two variables. In these cases, all it means is that there’s no consistent pattern when either of the variables fluctuates – the second variable will change in a much less predictable way.
Be careful when interpreting correlation, though. A strong correlation is simply an indication that two variables are moving in certain directions at the same time. What it doesn’t tell you is whether one variable is a cause, and the other an effect. Of course, with extended knowledge of the environment and the variables, you can use correlation as a step towards inferring causation, but it is not statistically proven to be the case.
Let’s think about this from another perspective of your campaigns. You might say that there’s a correlation between the impressions of the ads and the number of clicks. But it’s not right to say that the high number of impressions is a direct cause of the number of clicks. There should be other factors that impact the clicks from impressions, but statistically speaking, the two metrics simply have a relationship.
So to finish, correlation is a great way to prove that there’s a relationship between two variables. You can clearly state something about performance, to suggest that there’s a pattern involved. This can be powerful, especially when you can combine it with your extended knowledge of the campaigns, and this can help you in your analysis which leads to inference of some of the activity. However, correlation on its own should not be used to imply causation between variables. It is limited only to proving that variables show the same fluctuations.