Web Analytics and Stat Geeks

Why you shouldn't rely on web analytics for specific numbers and how even some "professional" SEO people get caught up in the statistics.

So SEOmoz has an interesting article on "Indexation for SEO" and why you shouldn't rely on the site: tag in Google. While I agree with Rand Fishkin in his article it is interesting to read the comments that follow and see how differently people view the web analytics. There is even one comment that is clearly from someone who has not been doing this very long. I won't name names, but they point out a clear issue with any kind of web analytics tool and this inherent discrepancy is what I plan to cover in this entry.

The main challenge with web analytics is visitors are somewhat anonymous in most cases and some visitors (albeit a small amount) will intentionally take steps to further hide their identity and surfing behavior online. There are also a dizzying number of things happening from the time a visitor types a URL or clicks a link and the browser displays the requested web page. I could go into detail here, but I'll spare you the info on dynamic IP assignments, content delivery networks, caching, etc. Suffice it to say there is a lot going on and different web analytics tools have different collection methods that carry different pros and cons.

The point: Web statistics should be used primarily for trending, secondarily for site profiling, and never for absolute numbers.

Now please don't go out and say "Chuck is off his rocker and doesn't think you can tell what is going on with your site." The key is understanding your baseline and how the changes you make to your site affect that baseline. It also doesn't really matter what analytics tool you use for most analysis and the main KPI's, as long as you use the same tool that you used when you established the baseline. You can't compare statistical results from two different packages and expect to get the same exact numbers, unfortunately.

Let's look at a theoretical example and then I'll briefly explain why the numbers differ. The following table lists some visitation numbers that you might see from Google Analytics and Webtrends for the same site and the same time periods:

Notice how the exact numbers are all different between the two packages. The easy answer for this could be that Google Analytics relies on a call to the Google Analytics tracking code to register the page view while Webtrends analyzes the log. There is a lot more going on, but in this simplest example you can imagine a page loading and being counted by Webtrends, but the call to Google Analytics failing so it doesn't show up in Google Analytics. Trying to explain a discrepancy between the two tools is a fruitless exercise. Does it really matter if there were 41,289 visits or 39,234 visits? No. What matters is the growth in visits over time (and ultimately how many of those visits converted into something actionable, but that's another discussion and I'm already long winded today).

Let's look at the same information as percentage growth:

When looking at percentages the numbers are fairly close and tell the same story every month. Whatever happened in March was good for visits and probably should be repeated if possible. Overall visitation improved about 40% over the 6 months shown. Assuming constant market demand and other external factors both sets of results show decent improvement in website visibility.

Second point: Comparing data from the same web analytics package yields useful information. Comparing data across web analytics packages is dangerous and suspect.

Of course, that is a generalization and we look at data from multiple sources here at IPG. What you have to understand is what the data as a whole is really telling you, why the data is what it is, and what you can do to positively affect the results in the future.