Ronald Fisher spent fourteen years, from 1919, as the statistician at Rothamsted Experimental Station, an agricultural research institute in England that had been recording crop yields since the 1840s. Faced with decades of messy field data, and the need to keep testing new fertilizers and crop varieties efficiently, Fisher developed the statistical tools that would come to define how scientific experiments are designed and analyzed across nearly every discipline, not just agriculture.
His 1935 book The Design of Experiments introduced the principle of randomization, assigning treatments to experimental plots or subjects by chance rather than by a researcher's judgment, as the foundation of a trustworthy experiment, without it, Fisher argued, no amount of statistical sophistication afterward could rule out the possibility that some unnoticed pattern in how treatments were assigned, rather than the treatments themselves, explained the results. Alongside randomization he developed analysis of variance, a method for splitting the variation observed in an experiment's results into the separate contributions of each factor being tested, and worked out the mathematics of maximum likelihood estimation, a general method for finding the parameter values that make observed data most probable, which remains one of the most widely used tools in all of statistics.
Fisher also popularized the p-value as a measure of how surprising a result would be if there were truly no effect, a tool he intended as one piece of evidence among several for a working scientist to weigh, not the automatic accept-or-reject rule it is often used as today. That gap between Fisher's own more nuanced use of significance testing and the mechanical way it is frequently applied in modern research has itself become a subject of methodological debate nearly a century later.