March 15, 2005
In describing the evolution of statistical analysis in his book The Lady Tasting
, David Salsburg illuminates the ways in which the development of statistical analysis
was shaped by the personal and political circumstances surrounding its pioneers.
chapter focuses on the unique story of a particular statistical method and its discovery,
providing the reader with a vivid sense of its temporal relationship and importance to the
growing body of statistical knowledge.
Though these discoveries are not always
presented chronologically, a clear path emerges from abstract mathematics in the late
nineteenth century to current applied statistics.
Salsburg’s personal knowledge of many
of the founding fathers of statistical analysis draws the reader into this tight knit
community and adds to an understanding of the genius that brought the science out of the
air and into the hands of millions around the world.
He continually brings to light the
pervasive nature of statistical analysis and its crucial role in so many, often unexpected,
realms of contemporary society.
Today, many statisticians work in epidemiology, the biomedical sciences or in the
business sector, but the first statisticians were, in fact, mathematicians interested in
solving abstract problems.
While earlier discoveries in statistics were made by
eighteenth century mathematicians like Gauss and Bernoulli (Salsburg, 16), it was not
until the late nineteenth century that Karl Pearson and his rival, R.A. Fisher, began to
elucidate the fundamentals.
Pearson began by collecting massive amounts of data and
analyzing them without a physical problem to solve, developing the first statistical model
of a skew distribution and its four key parameters.
Pearson believed, correctly, that his
observations were a random sampling from the actual distribution that existed in nature,
and this distribution, not the data themselves, was the important result (17).
error lay in the assumption that his four parameters could actually be determined, that
they were the same for the distributions of both the sample and the population.
later pointed out, these parameters could only be estimated for the population
distribution, never known (35).
Yet, this did not diminish the importance of Pearson’s