Let It Grow: The Garden Cress Challenge - Announcing our Winner

[#1]

Two of my kids were with me in the Kay family research kitchen working on an idea for a competition to encourage people to try design of experiments (DOE). We were experimenting to find out what factors affect the growth of cress…

Some important work being done in the Kay family research kitchen

The conversation went something like this:

8-year old kid: “It would have been easier if we just put all the soil ones in the dark and all the cotton wool ones in the light.”

15-year old kid: “But then we wouldn’t know if it was the soil or the dark that had the effect.”

I was so proud that it brought a tear to my eye. A simple explanation of what we call confounding in the statistical design and analysis of multifactor experiments. As they say, the apple doesn’t fall far from the tree!

I have been obsessed with DOE since I first learned about it as a lab development chemist. I had some big successes using the method back then, and it has been a huge part of my career ever since.

If you don’t know what DOE is, think about an experiment to find out how cress growth is affected by 2 factors:

• light conditions, sunlight or dark
• growing medium, soil or cotton wool

My guess is you would grow cress in 3 different conditions:

1	Soil in sunlight	“control”
2	Soil in the dark	test the effect of light conditions
3	Cotton wool in sunlight	test the effect of growing medium

But there is another combination that DOE would tell you to test also:

4

Cotton wool in dark

?

Why? First of all, it is an extra data point for more evidence of the effects. Secondly, it means you can see if the effect of light is the same for soil and cotton wool.

An example of an interaction between factors

In our experiment we found that there was a slightly bigger difference between dark and sunlight when soil was used as the growing medium. These interactions between factors are often really important when you are experimenting to understand how the real world works. You can find them only using DOE. Unfortunately many very clever and educated scientists and engineers still don’t know about DOE, which can make their jobs really hard.

I think people need to learn about DOE at a young age. My JMP colleagues and I came up with a summer challenge for people to try DOE by growing cress using a multifactor statistical design. I tested it with my kids to make sure it could work for experimenters of all ages. All it requires is some empty egg boxes, soil, cotton wool, cress seeds, and a measuring ruler. You can find out more here.

Cress growth under different conditions - from the Kay family research kitchen

We had some great entries. Our judging panel was looking for evidence of clear thought, effort, learning, and a sense of fun. Special mention goes to Theresa and Malik for bringing the fun with a beautiful and clear report of their work.

A nice visual from Theresa and Malik

The winning entry was from Weronika Wrazen, a PhD student at the Hasso Plattner Institute, Potsdam, Germany. The judges loved how this simple home experiment had been written as though it was for a serious academic journal. (You can see a small excerpt at the end of this post). I am looking forward to co-presenting with Weronika at a JMP event soon – for more information follow #DOEbyPhilKay on LinkedIn.

I have published some interactive reports of Weronika’s work that you can play with using JMP Public. This profiler plot is an interactive visual of a model fitted to the data to understand how the factors affect cress height at days 3, 5, and 7:

Profiler plot

So go and try DOE by growing cress. Try it with your friends and family. Try it with your school students. The world will be a better place when more people know about DOE.

Thanks to the judging panel: Shirley Coleman (Newcastle University), Simon Fuchs (2bind GmbH),  Sonja Kuhnt (Fachhochschule Dortmund), Volker Kraft (JMP), and Ryan Lekivetz (JMP).

.