### BIOGRAPHY 13.2* Jerzy Neyman* (1894 -1981)

Jerzy Neyman was born in Benderey, Bessarabia. He studied at the University of Warsaw and, in the 1920's, achieved world-renown status by extending sampling theory to sampling from finite populations without replacement and carrying out complex stratified sampling schemes for the Polish government. In the 1930s, he joined the faculty at University College, London, where Ronald A. Fisher (Biography 13.1) and Egon S. Pearson (Biography 13.3) had just filled two positions created to replace the one long held by Karl Pearson (Biography 14.1). Neyman, who soon was to move on to a long and distinguished career at the University of California, Berkeley, shared many interests with Fisher -- agricultural experimentation, weather-modification experiments, genetics, astronomy, medical diagnosis -- yet the two ended up on opposites sides when the famous and bitter controversy arose concerning the nature of estimation and hypothesis testing. For years, however, Neyman collaborated fruitfully with Egon S. Pearson; jointly the two men were immortalized by the Neyman-Pearson theory of estimation and hypothesis testing that is now generally accepted.

Neyman and Pearson introduced the concept of *confidence intervals* into the theory of estimation at about the same time that Fisher wrote about *fiducial intervals*, and for a time the two concepts lived amiably side by side, appearing to be two names for the same thing. Eventually, however, it became clear that these were different concepts, indeed. Fisher's 95 percent fiducial interval, for example, would claim a 95 percent probability that a given parameter lay within the interval constructed around a sample statistic already calculated. Unlike Fisher, Neyman-Pearson would establish the interval *before* the sample was taken and before any statistic was calculated. A 95 percent confidence interval according to Neyman-Pearson only claims that use of their formula in the long run produces intervals such that 95 out of 100 of them contain the parameter, while any actual interval, constructed *after* sampling, was *certain* either to contain or to exclude the parameter. This important concept of confidence intervals is highlighted by text Figure 12.3.

For additional information on Neyman, see *Encyclopedia of Statistical Sciences* (New York: Wiley-Interscience, 1982-86), vol. 6, pp. 215-223.