Experiments are the cornerstone of scientific research and most scientists rely heavily on their domain expertise to carry them out. However, this is only one side of the coin; the other is in the quality of the data generated in the experiment itself.
George Box stated that “All experiments are designed experiments, it’s just that some are poorly designed and some are well designed”. And here, “well designed” means that it provides the most information possible.
In this context, a broad statistical theory has been developed that deals with how to maximize the information extracted from experiments.
To address scientific questions, experimenters must define many details that affect the quality of the resulting data:
▪︎ How many experimental runs should I do to find clear answers?
▪︎ What are the risks if I reduce the number of experiments, and therefore the cost?
▪︎ Can I run the experiments in any order?
▪︎ How many individuals or experimental units will participate?
▪︎ Can I manage known/unknown confounders?
▪︎ Etc.
Of course, the questions are as broad as they are numerous.
At Codas Lab we make it a priority to extend the rich theory that exists on statistical experimental design to the complexities of modern experiments through computational tools. We look for clever ways to design experiments in order to achieve some data analysis advantage and, accordingly, we co-design the experiments and the data analysis itself.
We are fortunate to have participated in projects from a wide variety of domains (research areas), discovering differences and similarities between the experimental design practices of each of them.
This experience offers us an abstract vision of good practices that is usually valid in any domain and independent of data.
We understand data knowledge as an element of great value for those research groups that perceive Data Science as an essential piece in their own research.
Only through the duo of domain knowledge – data knowledge can we aspire to top-level research.
At Codas Lab we firmly believe that experiments should be designed by domain expert scientists in collaboration with data scientists.
If you want to receive more information about our Statistical design of experiments research line, do not hesitate to contact us.
