To grow the perfect strawberry, you need sun, water, soil … and statistics! In this video, strawberry researchers Gene Galletta and Olivia Mageau explain how they breed and evaluate new strawberry plants for commercial growers.
An important step comes at the end, when they compare new plants to established commercial varieties. Because growing conditions are often uneven across a field – for instance, some parts of the field might have better soil or more water – the team needs a way to take these differences into account.
They do this by using a statistical method called a “randomized complete block design.” In this scheme, they divide the field into sections, or blocks. Within each block, they plant all the varieties they’re testing (including the established ones) in a random order. They then sample strawberries of each variety from all of the blocks. That way, they can be pretty sure that any differences they find are due to the quality of the plants, not the conditions in which they were grown.
In testing their strawberries, Gene Galletta and Olivia Mageau faced a question that confronts almost every scientist: how do you design an experiment so that it clearly answers the question you are asking?
It’s a surprisingly thorny problem. In our messy world, it’s very rare that you can measure something and know immediately what your result means. For instance, Dr. Galletta could try a new strawberry and tell immediately that it tastes better than older varieties. But he still wouldn’t know why – was it because this new variety is naturally better, or because it grew in better soil?
Factors like the strawberry plants’ growing conditions are called confounding variables, and they’re extremely common across science. Doctors see them every time they try a new medicine: it’s hard to know whether patients are getting better because of the new drug, or because of some other influence. Even physicists run into them in their experiments.
Fortunately, knowledge of statistics plus some clever set-up can help scientists around this problem. The specifics can vary a lot from experiment to experiment, but they all have three pieces in common:
Controls: These are extra experiments that the scientist runs to catch any confounding variables in action. For instance, a doctor testing a drug on one group of patients will also gather a second group of patients and give them something inactive – usually a placebo, or sugar pill. If a confounding variable is working, it should affect both groups the same way.
Randomization: Scientists pick their experiment and control groups at random to make sure that the makeup of each group doesn’t affect the outcome.
Repetition: In order to further decrease the chance that something random creeps in and skews the results, scientists repeat their experiments many times.