Weeds are big business – the global herbicide market is over $35 billion annually. Weeds are also big government (think “invasive species”). California’s listing of weeds is called Encycloweedia, and the state publishes a quarterly newsletter called Noxious Times. Colorado publishes a similar periodical, Invader.
The weed-killer Roundup is the focus of lawsuits that illustrate the role that statistics plays (or doesn’t play) in scientific proof. A few weeks ago, a former groundskeeper with non-Hodgkins lymphoma, which he attributed to his use of Roundup, won $289 million in damages (later reduced by a judge). There are now over 12,000 lawsuits pending against Monsanto, Roundup’s maker. Most of them were filed after Monsanto was acquired by the German pharmaceutical company, Bayer, which was worth nearly $100 billion at the time. For Bayer, this was like acquiring a time bomb – its market capitalization has fallen by half since the post-acquisition peak.
The Roundup case illustrates the various “standards of proof” that come into play in high stakes liability cases.
On the one hand, numerous studies based on solid statistical methods have failed to find evidence that Roundup is a threat to human health. Regulatory agencies in the U.S., Europe and other areas, relying on these studies and their own reviews, have repeatedly cleared Roundup for use.
On the other hand there is the compelling evidence of human suffering presented to juries which have, in two cases, awarded hundreds of millions of dollars to plaintiffs. Solid evidence that the cancers are caused by Roundup is lacking, but courtroom juries, responsive to emotion and suffering, have never been an effective arbiter of science.
One factor in the jury awards is the financial ties that exist between industry and those who conduct research. It has long been the case that industries seeking regulatory approval of pharmaceutical products, medical devices and potentially toxic agents must prove that they are safe and effective. So, of course they must pay researchers to conduct the studies. In the pharmaceutical industry, a strict protocol has evolved involving outside monitoring committees, but the process is not as formal for pesticide review.
A wild card is the role played by the International Agency for Research on Cancer. This group applies much looser standards in labeling something as carcinogenic. Evidence that most statisticians would regard as falling below the standard of statistical significance is given weight, as are animal studies that are conducted at dosages that are considerably higher than would be encountered by students. Over 1000 substances have been studied by the IARC and over half were found to be definitely, probably, or possibly carcinogenic. Coffee, power lines and cell phones have all been classified by the IARC as “possibly carcinogenic.” Only one substance, out of the 1000+ tested, was found to be non-carcinogenic.
The most powerful contribution of statistics to scientific advancement in the 20th century was designs for studies (experiments) that have the ability to prove scientific hypotheses through empirical evidence. These designs were initially developed and applied in agriculture. Among the first such studies were those undertaken by W. S. Gossett (his employer, Guinness, had agricultural as well as brewing interests), and R. A. Fisher in the fields of Rothamsted, an agricultural research station. It was Fisher’s agricultural research that led to his famous 1925 work Statistical Methods for Research Workers.
In the 21st century, the statistical techniques of big data analytics, originally developed and honed in finance and retail, have come back to agriculture, where statistically-based research started. Data generated by farmers can be collected and analyzed, and the contributions of highly specific soil, weather and drainage conditions can be factored in, supplementing the more general and global knowledge derived from agricultural research. he average U.S. farm today feeds more than 165 people. “Precision agriculture” and will be a major factor in driving further improvements in yield as less and less land remains available for farming. Companies that provide analytical services to farmers will be central to the process. In the normal course of their activities, they collect more and more data over time, allowing their statistical models to gain a deeper and richer understanding of just how the many different factors in farming affect current yield, cost, and future crop fertility.