The design of experiments is a task or a set of functions that aims to change the inputs to identify the output change. This tool is handy when we work on the unconscious needs of the client.



When a customer explains his needs, we can have two main possibilities:

  • The client can consciously present his needs, so for instance, the estimation time is too long because it is bigger than 30 days.
  • The client unconsciously presents his needsso for instance, he doesn’t like the website’s graphics. 

A test for experiments helps find the variable connected to his unconscious needs, but this is not the only use of this tool.


In designing an experiment, we need to:

  • Define the objectives of the experiment;
  • Define the experiment methods and consideration: choose your variable and consider the interactions;
  • Execute: Execute the experiment and analyze the result.


Defining the experiment’s objective helps to choose the correct methods. So, why are you running this experiment?

  • Do you want to compare two factors to find what best impacts the result?
  • Do you want to screen between many factors to find the most important one?
  • Do you want to find the regression model among factors to find the correlation?
  • Do you want to discover how some independent variables influence the output?


To define the experiments, we need to choose the variable we want to test and how we want to try it. For instance, we can wish to:

  • Test one factor at the time: in this case, we change only one factor at a time, and we look at the result. If the result is good for us, the test is ended. In another case, we change only one of the factors, and we run another test. With this approach, we don’t gain a full understanding of all the possible combinations;
  • Full factorial: we make a test where we test all the combinations of factors. It is applicable only when we have a low number of the combination.
  • Partial factorial design: is a hybrid of the full factorial where we consider only a combination subset.


At the end of the test, we need to analyze the result.

Example: we start from the need of our client to improve the usability of his website because he wants to attract more final customers. We identify three variables: loading time, a different menu organization, and a site map. 
Loading time (x1) can assume values like less than 1 second, more than 1 second;
The different menu organizations (x2) we assume are three;
The presence of the site map (x2) is only true or false;
We try all the possibilities of this value:
- Loading time less than 1 | Menu 1 | Site map true
- Loading time less than 1 | Menu 1 | Site map false
- Loading time less than 1 | Menu 2 | Site map true
- Loading time less than 1 | Menu 2 | Site map false
- Loading time less than 1 | Menu 3 | Site map true
- Loading time less than 1 | Menu 3 | Site map false
- Loading time more than 1 | Menu 1 | Site map true
- Loading time more than 1 | Menu 1 | Site map false
- Loading time more than 1 | Menu 2 | Site map true
- Loading time more than 1 | Menu 2 | Site map false
- Loading time more than 1 | Menu 3 | Site map true
- Loading time more than 1 | Menu 3 | Site map false
For each input combination of the variables (x1,x2,x3), we can look at how a group of people reacts, giving maybe a score from 1 to 10 (Y result). So, for example, we can look at loading time less than one and site map true, give a ten result. Instead, changing the organization of the menu doesn't affect the score.

For the exam you need to remember that:

  • experiment design is a helpful tool to look at how Causes and Effects are correlated;
  • You can test one factor at a time or make a full factorial test trying all the possible combinations of values.

References:

Share on: