One of the most common mistakes in web analytics is assuming that all numbers can simply be added together. This works in many cases but there are a few common metrics where you can easily run into a duplication problem. Some common ones are:

Visits/Visitors
Lets say that a visitor (in a singe visit) comes to your site and performs a search for “dress” and then for “skirt.” If you were then to look at your Internal Search report you would expect to see one visit for “dress” and one visit for “skirt.” This is correct because, in fact, each of these keywords got a visit. However, what if you wanted to know simply how many visits performed a search? It sounds simple but, in the absence of any other reporting, some are tempted to add the one visit for “dress” and the one visit for “skirt” in order to get a grand total of two. With this information they can then report to their boss that of the one visit to their site a total of two visits engaged in search. Doesn’t make much sense does it? In this case the one visit for “dress” and the one visit for “skirt” should still be one because it it the same visit. This example is a little absurd but keep this in mind because with real data, where only a portion of your visits engage with search, the mistake may not be as obvious.

Orders
This is another one that sometimes gets people. Let’s say that I’m looking at a products report and there are two products (A and B) that were purchased on my site, each with one order. What would my total orders be? Someone that is too quick to add may say two. Someone that read the Visits/Visitors example may be too quick to say one. In the end, though, you really don’t know by just looking at this report. These products may have been each purchase in two separate orders or in the same order. To get to this information you will need to have a different report that gives you de-duplicated orders (a total orders report.)

Rule of thumb
The easy way to determine if the metrics that you are using are additive or not is to ask yourself if the metric that you are viewing (revenue, orders, page views, visits …) could potentially span multiple values of the dimension (product names, keywords, page names …). For example, In the case of products and orders, the question would be “Could a single order potentially span multiple products?” The answer here is most likely “Yes” (not always, though) and you need to be careful to not fall into the trap of duplicating metrics.

Another interesting case to think about is visits, entries, or single-page visits in a pages report. If you were to just sum the visits in this report then you would have a serious duplication issue being that a single visit can span multiple pages. However, you are safe summing entries and single-page visits because these metrics are limited to only a single page. If you were limited to just this report you could, though, get a de-duplicated visit count by using the sum of entries across all pages. This is because the ratio of total visits and total entries is always one and summing the entries does not put you at risk of duplication. Really, though, if you want this number it’s easiest to just use a totals report but it’s something to think about.

This entry was posted on Saturday, August 2nd, 2008 and is filed under Whatever, Analytics. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.