FEATURE ARTICLE
Winning the Accuracy Game
Three statistical strategies—replicating, blocking and modeling—can help scientists improve accuracy and accelerate progress
Hugh G. Gauch, Jr.
In a scientist's life, data that closely fit a hypothesis are
exceedingly rare. In fact, it can be almost impossible to
distinguish whether a new treatment is actually more effective than
an old one. Most experiments or observations produce noisy and
imperfect data that must be subjected to statistical analysis in
order to tease out results. Three strategies are most effective in
improving the accuracy of this process: replication, blocking and
modeling. Unfortunately, these strategies are underused: Many
scientists fail to design their experiments or model the results in
ways that yield the greatest accuracy. Replication has a diminishing
payoff, but special blocking techniques (such as the
"randomized complete block" design) and modeling can give
scientists impressive gains in accuracy at low cost.
Go to Article