Vale Bob Rosenthal, Statistical Reform Leader and Much Else

I was much saddened to read of the death last month of Bob Rosenthal. See this obituary; and another in the New York Times.

I met him first in 1996 when I called on him at Harvard to discuss statistical reform. What a gentle, encouraging, and thoroughly nice person! What a giant intellect! He loved nothing better than to find innovative solutions to tricky problems.

Considering statistical reform and Open Science:

  • He was an early proponent of a focus on effect sizes, especially his favourite, Pearson correlation, r.
  • He was a pioneer of meta-analysis and identified what he called the file-drawer effect.
  • Rosenthal and Gaito (1963) reported evidence that researchers’ confidence in an effect drops sharply as the p value increases past .05; they labelled this the cliff effect. This was an early example of statistical cognition–the empirical study of how people understand statistical concepts and reports. We still need much more of that, imho.
  • Around 2009 Jerry Lai wanted to investigate the cliff effect as part of his PhD. He sent Bob a very polite request for any further information about the original study. Promptly, back came an encouraging message to Jerry and a scan of several hand-written pages of the original data. From almost 50 years earlier! A wonderful example of Open Data (well, available data), with no excuses about hard disk crashes and superseded storage formats.
  • He advocated analysis of well-chosen contrasts as better than the customary reliance on Anova and p values (*, **, ***, or ns) to interpret omnibus main and interaction effects. He stated that “the problem is that omnibus tests … do not usually tell us anything we really want to know”. Contrast Analysis: Focused Comparisons in the Analysis of Variance (1985) by Rosenthal and Rosnow remains an accessible and powerful explanation. UTNS, and both editions of ITNS take this planned contrast approach (these days, with preregistration) to the analysis of complex designs.
  • In 2008 Fiona Fidler and I were working on Confidence Intervals : Better Answers to Better Questions. We sent a draft to Bob who was working on an accompanying article Effect Sizes : Why, When, and How to Use Them. Bob responded with enthusiasm, saying he loved our article and also offering valuable suggestions.

Bob’s nickname among his students was “Prof ARRRZZZental“, recognising his love of correlation r.

I salute his memory and his enduring contribution to improving how we do things.


Leave a Reply

Your email address will not be published. Required fields are marked *