The reasoning is simple: to validate that you’ve set-up and run the test properly — and that the data coming back is clean, accurate, and reliable.
According to Instapage, it’s estimated 80% of A/B test results are imaginary. They’re based on false positives — which is a fancy, statistical way of saying, the results aren’t accurate. If you’re making optimization decisions on inaccurate results, you’ve got a problem.
The only way to truly validate accuracy is to test the same variant against itself. If you get markedly different results – and one version emerges a clear winner – you know there’s an issue. You might have noise in your data, or have not set things up properly.
It’s slightly counter-intuitive. But, with an A/A test, you’re actually looking to ensure there is no difference in results, between variants.
There are a couple of advantages to running an A/A test. Doing so:
1. Gives you more certainty you’ve set-up and run the test properly — assuming there is little difference in the results, between variants. This verification is especially valuable if you’re new to setting-up and running A/B tests.
2. Rules out the novelty effect. If you want to accurately test a change you’ve made on a website that people frequently go to, and are used to, like Facebook, implementing an A/A test helps you better understand how people are reacting. If one version of the same variant performs markedly different than the other, you know there’s something with the way your sample is reacting – not the variant itself.
2. Running an A/A test can distract you from running real, valid tests that bring in new, additional revenue.
My advice to you is this: if you’ve never set-up and run an A/B test before, start with an A/A test. Doing so is low-risk and will give the confidence to move forward with setting-up a real test the next time round.
If you’ve run tests before, but feel uncertain about some aspects, consider running an A/A/B test in which you split traffic three ways, to version 1 and 2 – which are the same – and to version 3, which is different. When analyzing results, you should see no major difference between versions 1 and 2 (A/A), but will, hopefully, see a difference in results when comparing versions 1 and 2 (A/A) to version 3 (B).
So, for example, if you’re running a test in a testing platform like VWO, it’s a good practice to also set-up and cross check the data in an analytics platform, like Google Analytics, to ensure the results are consistent across both platforms. If there’s large discrepancies in the data, across platforms, you know there’s an issue, and the results should be further assessed.
My recommendation is to set-up custom dimensions in Google Analytics. Doing so will enable you to:
1. Be more confident in the test results you obtain
2. Allow you to perform further data analysis, beyond what’s provided in the testing platform.
Testing platforms, like VWO and Optimizely, have built-in Google Analytics integrations, so you can easily set-up custom dimensions in Google Analytics. Depending on the testing platform you’re using, set-up will vary. Here’s some great resources to help you set-up custom dimensions in Google Analytics:
A/A tests have utility. They can help you confirm your sample groups are being split, or randomized properly, and your tests are set-up properly.
A/A tests essentially test your tests.
As this real-life GuessTheTest A/A test case study shows, A/A tests can help you uncover testing problems, and save you from implementing inaccurate tests.
Hope you’ve found this information useful and informative. Please share your thoughts and comments in the section below.