Now that you have an orthogonal matrix, you know that you need 9 sets of ads to conduct the test, and you know which test elements to put in each set of ads, you can start the experiment. We recommend that when you conduct ad testing, you remember to mark the ad combination as belonging to the experiment , so that it will be convenient for subsequent analysis.
Analysis of experimental results
After the experiment, we will get 9 groups of advertising conversion rates. At that time, we can use Python to quickly find the best advertising combination. We can use the following code to achieve this:
First, organize the data into a DataFrame:
Finally, we can directly find the best advertising combination:
The experiment is almost complete at this point. How to find the You can find the advertising combination with the best conversion rate using the least number of experimental groups at the lowest cost. However, the above experiments are all based on the premise that all elements are independent of each other and do not interact with each other . Therefore, if you want to check whether each element is truly independent, you can use ANOVA to do it, but we will not go into details here.
Use a simple case to verify the usability of orthogonal experimental design
Finally, we will use a simple case to understand the mathematical principles behind orthogonal experimental design and how the benefits of each element are calculated.
First, let’s assume that there are different lunch combinations, costa rica phone number library and the goal is to find the best lunch combination:
- Staple food: rice, noodles
- Side dishes: chicken, fish
- Beverages: juice, soda
How to find the Under normal make your number work for you circumstances, we need to try 8 lunch combinations (2x2x2) to find out which lunch combination is the most delicious. However, through the logic of the above orthogonal experimental design, we can select the following 4 lunch combinations to try and find the most delicious lunch combination:
- Rice + Chicken + Juice
- Rice + fish + soda
- Noodles + Chicken + Soda
- Noodles + Fish + Juice
What is the computational logic of orthogonal experimental design?
Suppose we tried the above 4 lunch combinations and got the following delicious scores:
- Rice + Chicken + Juice = 8 points
- Rice + fish + soda = 6 points
- Noodles + chicken + soda = 7 points
- Noodles + fish + juice = 5 points
At this time we can calculate the switzerland leads effects of various elements.
