Contributed Poster Presentations: History of Statistics Interest Group

For more than a half century, the US pharmaceutical industry has employed statisticians to support the development and marketing of therapeutically important compounds. Although the earliest statisticians in the industry date back to the 1950s, separate administratively organized statistics departments began to be formally established in the mid to late 1970s. These early groups were mainly clinical statistics departments, supporting human dug trials, that formed because of the passage in 1962 of the Kefauver–Harris Amendment to the 1938 Food, Drug and Cosmetic Act. Hiring of nonclinical statisticians to the industry can be traced back to the publication of the 21 CFR Part 58 - Good laboratory practice for nonclinical laboratory studies regulation in 1978. This prompted many companies to hire statisticians to support drug safety and toxicology studies and marked the impetus for the creation of nonclinical statistics groups with a distinct administrative identity. These groups went on to expand the scope of their statistical services to early development and CMC studies. This presentation will examine the development and expansion of nonclinical statistics departments in the US pha 

Kurtosis continues to be misinterpreted as "peakedness/flatness"; large kurtosis "indicating" peakedness, and small "indicating" flatness. The "indicates" terminology appears in the first sentence of the abstract of DeCarlo, 1997 Psychological Methods, "On the Meaning and Use of Kurtosis."

A downside of using AI for research is that it can provide faulty information when its sources are wrong. A case in point is "indicates peakedness/flatness," which was shown false in statistical literature showing how kurtosis measures tails. However, AI reports this phrase, as do current scientific research articles.

In this presentation I show graphs of families of symmetric, unimodal distributions for which *smaller* kurtosis determines *greater* peakedness, and where *larger* kurtosis determines *flatness.*

The only logic ever given for "high kurtosis indicates peakedness; low kurtosis indicates flatness" is the existence of families of distributions for which this is true. By the same logic, based on the given families, one can as easily conclude "high kurtosis indicates flatness; low indicates peakedness."

(Both interpretations are wrong; examples do not generalize.)