How to effectively reduce the number of experimental groups?

At this time, if you want to conduct a test to find out which set of ads has the best conversion rate, you usually need to do 3x3x3, a total of 27 sets of ads to complete the test. But when the advertising colombia phone number library budget is limited and the number of people reached is insufficient, are there other more cost-effective ways to find out which set is the best advertising combination?

The answer is: Yes, you can try “orthogonal experimental design”

What is orthogonal experimental design?

Orthogonal experimental design is an experimental method that uses scientific methods to reduce the number of necessary test groups. We can find representative advertising combinations through the designed orthogonal table and remove unnecessary combinations to reduce the number of experimental groups. At the same time, we can monitor the effect of each factor on the results and find the best experimental combination.

In the above example, if you want to complete the experiment, you need 27 sets of advertisements, but if you use orthogonal experimental design, you only need to use 9 representative sets of advertisements to find the best advertising combination. This greatly reduces experimental costs and improves the efficiency of experimental testing.

I believe the above explanation may still be marketing that starts with a call confusing. Next, let’s use a case to explain how this experimental method reduces the number of experimental groups from 27 to 9!

Steps in orthogonal experimental design

The general orthogonal experimental design includes the following 4 steps:

1. Confirm the variables and switzerland leads levels to be tested.
2. The orthogonal experimental table selected and made.
3. Conduct experiments.
4. Analyze the experimental results.

Below we will use the aforementioned advertising test to explain how to design a securities exchange experiment.

1. Confirm the factors and levels of testing

In an ad test, there are three variables: ad image, ad title, and CTA. In each variable, there are three test contents, which we call levels here. The chart is as follows:

• Variable A: Advertising images (A1, A2, A3)
• Variable B: Ad title (B1, B2, B3)
• Variable C: CTA (C1, C2, C3)

2. Orthogonal Experiment Table Selected and Made

You can find a suitable orthogonal array by searching the Internet, asking ChatGPT (but sometimes they bluff, so be careful), or on websites like SPSSAU . The orthogonal array will list all the representative combinations so that the experimental levels are evenly distributed, thus reflecting the overall experimental results. I won’t go into too much detail about the derivation process. If you are interested, you can study the relevant articles on your own .

In the above experiment, we have 3 variables, each with 3 levels, so we can use the orthogonal array L9(3^3). The L9 in L9(3^3) means that 9 groups of experiments need to be done, and 3^3 means that it can be used to process 3 variables with 3 levels. The orthogonal array L9(3^3) looks like this:

Generally speaking, an orthogonal array needs to satisfy the following two characteristics:

1. The number of occurrences of different numbers in each column is the same : taking the above orthogonal array as an example, 1, 2, and 3 appear an average of 3 times in each column.
2. In any two columns of horizontal numbers, each pair of numbers appears the same number of times : Taking the above orthogonal array as an example, if you choose the two columns of factor 1 and factor 2, you can see 9 pairs of numbers, including (1. 1), (1. 2), (1. 3)…, and each pair of numbers appears the same number of times. If you select the Factor 2 and Factor 3 columns, you will see the same situation.

After understanding the above two characteristics, we can also check by ourselves whether the orthogonal array design is correct.