Why Don't They Add Up? Understanding How Web Metrics Work Across Time Periods
It’s a common question, really. Inevitably someone will walk up to you and say, “There is something wrong with your reporting. When I add up your daily visitors they do not equal the weekly visitors you are reporting. Then, I add up the monthly visitors and they do not equal the yearly visitors. Why are you underreporting?”
Of course, you are not underreporting at all, and Visitors is not the only metric impacted by this phenomenon. This is a great time to grab a white board and explain.
Web analytics tools gather data on web activity 24/7 in a constant stream. This stream is then portioned into time buckets, for example: daily, weekly, monthly, quarterly and yearly.
To do this, the data is cut at the time cutoffs for each bucket and then totaled within the bucket to calculate metrics like: hourly visitors, daily visitors, weekly visitors, monthly visitors, quarterly visitors and yearly visitors. It is the cutting that could place a single visitor into two time buckets at the same time, which would double-count if the two time periods were added together, and when the data is uncut, the single visitor is counted only once.
The best way to explain is to illustrate with an example. It is Saturday, June 30th at 11:50PM and I visit karenlynnvincent.com. While I am there, I read the About Karen Vincent page and 3 of her blog entries, for a total of 4 pages. In the middle of the last entry I am reading, I notice that it is now Sunday, July 1 at 12:03AM. I shut down my computer and go to sleep.
Yearly Reports: I am 1 Visitor in 1 Visit with 4 Page Views. Yearly reports would reflect these metrics accurately.
Quarterly Reports: The quarterly report would split my session at 12:00AM, July 1. This means that the Q2 Quarterly Report would show 1 Visitor, 1 Visit and 4 Page Views. The Q3 Report would show 1 Visitor, 1 Visit, 1 Page View. These reports are accurate for activity within each quarter, but would over-inflate what happened if the reports were added together.
Monthly Reports: The monthly report would also split my session at 12:00AM, July 1. The June report would show 1 Visitor, 1 Visit and 4 Page Views. The July report would show 1 Visitor, 1 Visit, 1 Page View. While the monthly reports accurately show web activity for each month, but when added together they would over-count activity.
Weekly Reports: The weekly reports can be set to bucket weeks Sunday through Saturday or Monday through Sunday. If bucketed Sunday through Saturday, one week would carry 1 Visitor, 1 Visit and 4 Page Views and the next week would have 1 Visitor, 1 Visit, and 1 Page View. If bucketed Monday through Sunday, the report would keep the session together and would accurately reflect 1 Visitor in 1 Visit with 4 Page Views.
Daily Reports: Daily reports split at 12:00AM. This means that Saturday would hold 1 Visitor, 1 Visit and 4 Page Views and Sunday would hold 1 Visitor, 1 Visit, and 1 Page View.
Each report is accurate for its time period, but would provide inaccurate data if data for those time periods were to be added together manually in a spreadsheet. To safeguard the accuracy of the reports, be sure that reporting you do for any time period is based on reports from your Web Analytics software for that time period, and that no manual adding is done. Weekly reports should be based on Weekly blocks, Monthly on Monthly Blocks, etc to accurately reflect activity
Tell those who point out the difference to you that they are very clever to have found one of the lost secrets of data aggregation: that when it comes to aggregating, the whole is usually NOT the sum of its parts. Encourage them to stop adding and instead pull fresh and accurate data for each time period they need. This will provide a an accurate picture of activity on the site.