The march to data-driven advertising lately has been as relentless because the circulation of lava down the edges of an erupting volcano.
The usage of information in advertising is not at all new, however entrepreneurs now have entry to an unlimited quantity of knowledge concerning clients and potential patrons. Equally vital, additionally they have entry to highly effective and reasonably priced analytics applied sciences.
Right this moment, it is practically unattainable to discover a marketer who would not assume utilizing the best information in the best methods can enhance advertising efficiency.
A lot of the heavy lifting in advertising information evaluation includes correlation. In easy phrases, correlation is a relationship between phenomena or issues – “variables” within the lingo of math and statistics – that are inclined to range or happen collectively in ways in which aren’t as a result of probability alone.
It isn’t shocking that correlation performs such a central function in advertising analytics. A single information level can present helpful info, however the true energy of analytics is its capability to determine and quantify relationships between two or extra “variables” in your advertising information. Understanding these relationships can allow entrepreneurs to make choices that enhance advertising efficiency.
Correlation ≠Causation
One of many elementary ideas of knowledge evaluation is that correlation doesn’t set up causation. In different phrases, information evaluation could present that two occasions or circumstances are strongly correlated statistically, however this alone would not show that one of many occasions or circumstances induced the opposite.
The next chart offers an illustrative instance of why entrepreneurs should always remember the excellence between correlation and causation. It exhibits that from 1999 by way of 2009 there was a robust correlation ( r = 0.99789126 for you information geeks) between US spending on science, house, and expertise, and the variety of suicides by hanging, strangulation, and suffocation. (Notice:Â To see this and different nonsensical correlations check out Spurious Correlations.)
 |
| Supply:Â Tyler Vigen, Spurious Correlations |
I doubt any of us would argue that there is a causal relationship between these two variables (regardless of the robust correlation) as a result of they only do not have a believable relationship. In advertising, nevertheless, it is easy to come across occasions which might be strongly correlated and have a believable cause-and-effect relationship. The issue is, the causal relationship, whereas believable, could be weak or nonexistent.
When To Rely On Correlation
It is preferable, after all, to base advertising choices and actions on confirmed cause-and-effect relationships, however this will not at all times be practical and even potential. Proving the existence of a causal relationship sometimes requires the usage of a well-designed and tightly managed experiment. In advertising, such experiments could be simple to conduct in some conditions, however troublesome, if not unattainable, to run in others.
Underneath these circumstances, the true query is:Â When ought to entrepreneurs act based mostly on a correlation?
David Ritter with the Boston Consulting Group described a course of for answering this query in an article revealed on the Harvard Enterprise Assessment web site just a few years in the past. I’ve used Ritter’s course of – with a few minor modifications – quite a few instances in my work with shoppers, and I’ve discovered it to be efficient at focusing the eye of decision-makers on the best points.
The diagram beneath is my adaptation of Ritter’s framework.
Whether or not it is best to depend on a correlation relies upon totally on two components – your confidence within the correlation as an indicator of trigger and impact, and the stability of dangers and rewards.
Confidence within the correlation – The primary issue is your degree of confidence that the correlation factors to an actual cause-and-effect relationship. This issue is in flip a operate of two issues:
- How typically the correlation has occurred prior to now. The extra steadily occasions have occurred collectively, the extra doubtless it’s they’re causally associated.
- The variety of potential explanations for the impact into consideration. For instance, your information could present a robust correlation between the variety of advertising emails despatched and income development throughout a given interval. However, if there are a number of believable explanations for the elevated income, you’ve gotten much less motive to assume there is a causal connection between the variety of emails despatched and income development.
The stability of dangers and rewards – The second issue concerned in figuring out whether or not it is best to depend on a correlation is an analysis of dangers and rewards. Any resolution based mostly on a correlation ought to embody an evaluation of the potential dangers and advantages related to the motion.
The above diagram illustrates how these two components are used collectively that can assist you determine whether or not it is best to act based mostly on a correlation.
I must make two factors about utilizing this framework. First, it is vital to undergo this evaluation for every motion you are contemplating. While you determine a correlation, there’ll in all probability be a number of methods you might act on that correlation. Every possibility ought to be evaluated individually as a result of they’ll in all probability have completely different risk-reward profiles.
It is also vital to think about the scale of the “hole” between the potential dangers and rewards. For instance, if a possible motion has enormous potential advantages and really low dangers, you could wish to act even when your confidence that the correlation signifies a cause-and-effect relationship is not very excessive.
Prime picture courtesy of International Panorama through Flickr (CC).
