For the second installment of our StatXact 25th Anniversary Retrospective Series, Professor Joan Hilton (UC San Francisco) reflects on her pioneering work on exact conditional inferences.
(For our first story in this series, see StatXact 25th Anniversary: A Horizon for the Stars)
I met Cyrus Mehta in 1986 while taking a course as a doctoral student at the Harvard School of Public Health's Department of Biostatistics. I was immediately inspired by his enthusiasm for the topic he was teaching and, soon after, asked if he would serve as my thesis advisor. We got off to a great start with a paper on how to calculate power using the exact distribution of the Wilcoxon rank-sum statistic, assuming the treatment effect could be modeled by proportional odds. Since the network algorithm underlying this work built on earlier algorithms for non-ordinal data, I often visited the Cytel office to work with the start-up programming team, Pralay Senchaudhuri and Yogesh Gajjar. At the office, I met Cyrus's young family and his partner, Nitin Patel. I recall asking Cyrus's daughter her age, to which she promptly replied, "Three and three quarters!" Wow! Facile with numbers at that age! I also remember all of us, including my little sister during a visit, pitching in to prepare a booth for JSM conferences and the delicious chocolate cake we shared in celebration of the first anniversary of Cytel Software Corporation.
Another of my thesis papers addressed the issue of conditional versus unconditional inference in the two sample binary-response setting. Such questi ons could not be addressed with high accuracy before Cytel's StatXact software became available. Consequently, this breakthrough paper generated considerable attention, and led to a debate between Cyrus Mehta and Jonathan Shuster at a JSM conference.
After graduation, I took a faculty position at UCSF medical school that provided little protected time for me to continue conducting statistical research. Nonetheless, I published a few additional papers using exact conditional inference before moving on to new interests. My Harvard doctoral thesis work, based on Cytel software, provided me with a valuable foundation as a fledgling biostatistician. To this day, the Mehta family and my Cytel colleagues remain dear friends.
I congratulate Cyrus, Nitin, and all the team at Cytel on 25 years of success!
Related Items of Interest
Hilton, Joan F., and Cyrus R. Mehta. "Power and sample size calculations for exact conditional tests with ordered categorical data." Biometrics 49.2 (1993): 609-616. (Requires JSTOR access)
Mehta, Cyrus R., and Joan F. Hilton. "Exact power of conditional and unconditional tests: going beyond the 2× 2 contingency table." The American Statistician 47.2 (1993): 91-98. (Requires access to Taylor & Francis)
Professor Chris Corchoran's Webinar Slides on Exact Nonparametric Inference for Correlated Data (Access available to all)
