Average position is one of the easiest metrics in AdWords to understand. That simplicity of concept lures campaign managers and clients into misunderstanding what they're really being told.
Average position is a mean average but the natural inclination is to think of it as a mode average. Ask most campaign managers what it signifies and the answer you will get will be along the lines of "it tells me the position is was mostly in." But it doesn't, it tells you the mean of some of your positions (not all).
Let's take a look at some of the reasons these are different.
AdWords Impression Share
Your average position is weighted by the impressions in each position. If you get twice as many impressions in position 5 than position 4, that will be reflected in the mean average. The calculation looks something like:
Where: M = Impressions P = Position Mp = Impressions in position P
This is pretty easy stuff still. Where the complication arises is that not all of the searches on your keywords got an impression.
Check your impression share lost due to ad rank. If that value is greater than 0 then it means in some auctions you placed below the first page.
If you placed below the first page you usually didn't get an impression. No impression means no contribution to the calculation for your average position.
So you had some (possibly large) amount of auctions in which your actual auction result was very low, but it wasn't reflected in your average position. Whoops. This is why you will sometimes see the "keyword below first page bid estimate" message on a keyword with an average position obviously on the first page.
The average position is only reporting those auctions in which you reached the first page, and ignoring the others. Raising your bid to the suggested value might see your impression volume rise considerably.
The second factor in average position's deceptiveness is that they don't give you any idea of standard deviation. It takes several factors to describe a distribution function adequately. Mean is one, but standard deviation is just as crucial. Without it you just don't know how spread out your actual positions were.
Imagine the following scenarios:
- 100 impressions in position 3
- 50 impressions in position 2 and 50 impressions in position 4
- 30 impressions each in positions 1, 3 and 5
Each of the above situations will give you exactly the same average position: 3. But only one of those actually tells you a story about your ads.
Because your brain looks at average position and thinks "cool, that's where my ads appeared" you could potentially be tricked. In the first scenario that would be an accurate analysis, but in the second it's entirely false. It might be that your ads normally follow a distribution where the vast majority of ad positions are quite close to the mean. But they might not, and crucially you don't know.
AdWords doesn't tell you the standard deviation and therefore you're working from a position of zero knowledge.
Skew and Kurtosis
Skew and kurtosis are statistical concepts for describing the shape of distribution functions, but luckily for us the former is actually very similar to how it sounds, and the two are similar for these purposes.
If your distribution is skewed, it means that you don't have half your data points on each side of the mean, but more on one than the other. This can happen when one side is more skewed towards a long tail than the other.
Kurtosis represents a fat tail or thin tail for your data. Ad position only has one tail - downwards. You can't move further up the page than position one, but you can come lower and lower in the auction.
Sixty-six impressions in position 2 and 33 impressions in position 5 gives you an average position of 3. But more of your impressions came in a higher position than your mean, and fewer in a lower position. So the average position of 3 has lied to you again.
In common practice you're likely to have some impressions in every position through the day (maybe not on brand terms). But there are more positions below your average than there are above your average. So they're pulling your average down below you actual most common positions.
What's the Problem?
Bias is introduced to your data. Because your data will generally skew upwards due to impression share being below 100 percent, and downwards due to the one-tailed effect, you really don't know exactly how this is affecting your data.
Average position becomes a tool that represents only one thing: aggressiveness. You need to remove its relationship in your mind to actual ad positions.
If your average position rises then you are doing well in more auctions. If it falls, you're doing more poorly. It's a proxy for how well you do in auctions. It's not a proxy for where you showed on the page most commonly.
Once you're comfortable that average position is just a proxy for aggressiveness, we can introduce a slightly better one: top of page rate. Your average position is quite linear and doesn't really relate to your click volumes too closely. At least not to what is happening in each auction.
CTR does not gently rise as you move up the page. Instead it leaps as you get into the banner positions.
Check your statistics with the "top vs other" segment applied. You will typically find a jump in CTR of 10x-15x by getting into the banner, compared to the right hand side or bottom of the page.
So let's drop average position and go to something more closely allied to CTR. Top of page rate (e.g., the proportion of your impressions that occurred in the banner positions) is a better estimate of what is going to happen to your traffic. It's closer to the real driver of whether you get a click (banner or not, instead of position) and it's subject to fewer biases because it is a simple division, not a mean average.
What You Should Do
Take average position out of your reports. Educate your stakeholders and clients about top of page rate.
Every time a client tells you "I want to be in position 2 because that's best for me" is a chance for you to improve their knowledge and make sure they understand that ROI is what's important, not position. Reporting to them on position is going to include sketchy stats and disambiguates them from the figures that really count: money.
Meet Your Favorite Search Engine Watch Contributors
Many of SEW's leading expert contributors will be at ClickZ Live, the new online and digital marketing event kicking off in New York (March 31-April 3). Hear from the likes of: Thom Craver, Josh Braaten, Lisa Barone, Simon Heseltine, Josh McCoy, Lisa Raehsler, Greg Jarboe, Dan Cristo, Joseph Kerschbaum, John Gagnon, Eric Enge and more!