Wait! New to multivariate testing? If so, we recommend you first read our article, Multivariate Testing: All you need to know about multivariate testing
During an A/B test, you must only modify one element at a time (for example, the wording of an action button) to be able to determine the impact. If you simultaneously change this button’s wording and color (for example, a blue “Buy” button vs. red “Purchase” button) and see an improvement, how do you know which of the wording or the color changes really contributed to this result? The contribution of one may be negligible, or the two may have contributed equally.
The benefits of multivariate tests
A multivariate test aims to answer this question. With this type of experiment, you test a hypothesis for which several variables are modified and determine which is the best combination of all possible ones. If you change two variables and each has three possibilities, you have nine combinations between which to decide (number of variants of the first variable X number of possibilities of the second).
Multivariate testing has three benefits:
- avoid having to conduct several A/B tests one after the other, saving you time since we can look at a multivariate test as several A/B tests conducted simultaneously on the same page,
- determine the contribution of each variable to the measured gains,
- measure the interaction effects between several supposedly independent elements (for example, page title and visual illustration).
Types of multivariate tests
There are two major methods for conducting multivariate tests:
- “Full Factorial“: this is the method that is usually referred to as multivariate testing. With this method, all combinations of variables are designed and tested on an equal part of your traffic. If you test two variants for one element and three variants for another, each of the six combinations will be assigned to 16.66% of your traffic.
- “Fractional Factorial“: as its name suggests, only a fraction of all combinations is actually subjected to your traffic. The conversion rate of untested combinations is statistically deduced based on that of those actually tested. This method has the disadvantage of being less precise but requires less traffic.
While multivariate testing seems to be a panacea, you should be aware of several limitations that, in practice, limit its appeal in specific cases.
Limits of multivariate tests
The first limit concerns the volume of visitors to subject to your test to obtain usable results. By multiplying the number of variables and possibilities tested, you can quickly reach a significant number of combinations. The sample assigned to each combination will be reduced mechanically. Where, for a typical A/B test, you are allocating 50% of your traffic to the original and the variant, you are only allocating 5, 10, or 15% of your traffic to each combination in a multivariate test. In practice, this often translates into longer tests and an inability to achieve the statistical reliability needed for decision-making. This is especially true if you are testing deeper pages with lower traffic, which is often the case if you test command tunnels or landing pages for traffic acquisition campaigns.
The second disadvantage is related to the way the multivariate test is brought into consideration. In some cases, it is the result of an admission of weakness: users do not know exactly what to test and think that by testing several things at once, they will find something to use. We often find small modifications at work in these tests. A/B testing, on the other hand, imposes greater rigor and better identification of test hypotheses, which generally leads to more creative tests supported by data and with better results.
The third disadvantage is related to complexity. Conducting an A/B test is much simpler, especially in the analysis of the results. You do not need to perform complex mental gymnastics to try to understand why one element interacts positively with another in one case and not in another. Keeping a process simple and fast to execute allows you to be more confident and quickly iterate your optimization ideas.
Conclusion
While multivariate tests are attractive on paper, note that carrying out tests for too long only to obtain weak statistical reliability can make them a less attractive option in some cases. In order to obtain actionable results that can be quickly identified, in 90% of cases, it is better to stick to traditional A/B tests (or A/B/C/D). This is the ratio found among our customers, including those with an audience of hundreds of thousands or even millions of visitors. The remaining 10% of tests are better reserved for fine-tuning when you are comfortable with the testing practice, have achieved significant gains through your A/B tests, and are looking to exceed certain conversion thresholds or to gain a few increments.
Finally, it is always helpful to remember that, more than the type of test (A/B vs. multivariate), it is the quality of your hypotheses – and by extension that of your work of understanding conversion problems – which will be the determining factor in getting boosts and convincing results from your testing activity.