Multivariate Testing – Taguchi Method

In the earlier post, we talk about multivariate testing.

Here’s a more advanced of multivariate testing that I prefer – the Taguchi testing.

Let me explain what’s Taguchi testing is all about.

Taguchi testing is a statistical testing method devised by a Japanese engineer, Dr Genichi Taguchi.

It uses the orthogonal array model instead of the traditional full fractional model (used by Google Website Optimizer). Now that’s all math geek speak, so I’m not going to elaborate on that.

What’s most important is that with Taguchi testing, you can reduce the number of combinations needed for testing.

For example, if you have 4 elements with 2 variations each, you will need to test 16 (2 x 2 x 2 x 2) combinations using the traditional multivariate method.

With the Taguchi method, the number of combinations needed is reduced to only 8.

That means, you can reach a reasonable conclusion in a much faster time. And with that, you also save more on advertising fees because you only need a smaller sample size.

Let’s not forget, the faster you find a successful combination, the faster you get profitable. It’s all about time and money!

Introduction To Multivariate Testing

In split testing, we test one element of your sales page or website. For example, we can test headline A vs headline B (or vs C, D and so on), and we see which is the winner.

Split testing is pretty easy and straightforward to set up and implement.

But there are several drawbacks.

It’s very time consuming and costly. That’s because you can only test 1 element at a time.

Say we want to also test different images, location of image, sub headlines and so on…we can only do so one element at a time – one after another.

Because of this, by the time you are finished with all the elements, it could be weeks or months later!

You waste more time. And money as well as you incur more cost when the duration of the test drags.

There is another important point to consider. With split testing, it doesn’t tell us how each elements affect one another.

You might independently test the headline, and find one that performs really badly – so you don’t use it. Then you test your image, and find one that does really bad – so again, you don’t use it.