Should I set up my data as Events or Properties?


In general, the way to think about structuring your data looks like this:

  • Create Events to group people who do similar actions.
  • Record Properties for the criteria you are comparing/contrasting similar people.

Tradeoffs

  • Tracking with events could lead to too many events, and it becomes difficult to create aggregate reports without recording at least 2 events for each user action. But it’s still the best way to count the number of times Action X happened.
  • Tracking with properties is the best way to compare and contrast sets of people. The downside is that our reports do not make it elegant to list the number of times each Property Y was set – this is exclusive to the Power Report.

In practice, there’s a balance in the way you structure your data, mainly because the reports you’re looking for can change as time goes on. Let’s walk through three examples:

Example 1: Very generic event names (the 2nd/3rd columns) with one set of specific properties (name of the property is italicized and the values of the property are at the beginning of each row).

If you’re only running tests on a few Landing Pages, you’ll likely get less confused by a generic event name.

04-06-2013 Landing Page Test Received Email Viewed Landing Page
LP-Variation1 50 25
LP-Original-01-01-2013 50 10

Example 2: Specific event names with one set of specific properties.

If you run simultaneous tests on several Landing Pages, you might want to report on the results of this particular test in isolation.)

04-06-2013 Landing Page Test Received Email Participated in “04-06-2013 Landing Page Test”
LP-Variation1 50 25
LP-Original-01-01-2013 50 10

Example 3: Specific event names combined with two sets of specific properties.

If you are running simultaneous tests on two parts of the funnel – the email AND the landing page, each of those two should be their own Property.

04-06-2013 Introduction Email Test & 04-06-2013 Landing Page Test Received Email Participated in “04-06-2013 Landing Page Test”
Email Variation1 50 25
    LP-Variation1     25     15
    LP-Original-01-01-2013     25     10
Email Control 50 10
    LP-Variation1     25     6
    LP-Original-01-01-2013     25     4

Takeaways

In each of these cases, Properties are named to be specific (rather than “A/B Test”), so that you don’t mix data in from other tests.

Is anything on this page unclear? Suggest edits on Github!