Anirban DasGupta writes:
Recently, I read the 1966 Neyman Festschrift volume, edited by Evelyn Fix and F.N. David. In it, I found a nice overview of generalized inverses by C.R. Rao, and a lovely article by P.V. Sukhatme, with a long section on bias correction of ratio estimates by subsampling. My mind drifted off to the many other instances of long and productive connections between Berkeley and Indian statisticians, that have often surpassed the boundaries of professional collaboration and have turned into abiding friendships. I want to reminisce a little, a kind of history looking forward.
Early instances that I recall are the PhD dissertation of Ashok Maitra under David Blackwell, Prem Puri with Neyman, Sudhakar Dharmadhikari with Barankin, and D. Basu’s voyage to Berkeley on a Fulbright in 1953. The magnificent Maitra-Sudderth text on Discrete Games (1996) is clearly influenced by David Blackwell’s seminal contributions to stochastic games. And Basu’s entire professional development was directly influenced by his association with Neyman, and later with David Blackwell and Lucien Le Cam.
Some other early examples of the Berkeley–India relationship are the articles by Rao, Bahadur, and Basu in the Berkeley symposium proceedings (1965). Although not a joint collaboration, the Rao-Blackwell theorem is a household name, and such cornerstones of inference such as the Lehmann-Scheffé theorem were published in Sankhyá. It is tempting to conjecture that Sankhyá attracted these revolutionary articles because of the Berkeley-ISI personal bond. Much later, Peter Bickel and J.K. Ghosh wrote a well-known paper on Bartlett corrections (1990) and the 1982 Blackwell-Ramamoorthi note settled the conjecture that Bayes sufficiency is in general weaker than ordinary sufficiency. P.K. Sen and J.K. Ghosh contributed an article to the Neyman-Kiefer proceedings (1985) on the LRT for the finite mixture model (Hartigan, 1985). On the other hand, Peter Bickel and Anat Sakov contributed a survey on Richardson extrapolation and bootstrap (Bickel and Yahav, 1988; Politis et al., 1999) and Terry Speed and Yee Hwa Yang wrote on microarrays for the Basu Memorial Issue of Sankhyá (2002), all worthy examples of that Berkeley–India connection.
Then, too, there has been a steady flow of exchange of students and visitors. B.V. Rao was invited to Berkeley after he solved Ulam’s problem in his thesis. Ashok Maitra visited Berkeley several times. Few know that Terry Speed went to ISI and taught for ISEC, the international wing of ISI. Three of Peter Bickel’s earliest students were Hira Koul, D.P. Gokhale, and R.K. Aiyar. Erich Lehmann had (nearly) countless Indian students—Gouri Bhattacharya, M.L. Puri, M. Raghavachari, to name but a few. Rabi Bhattacharya did some of his most influential work at Berkeley. Most recently, Antar Bandyopadhyay, Smarajit Bose, Probal Chaudhuri, and Manjunath Krishnapur have returned to India after doctoral work at Berkeley, and Ani Adhikari and Sourav Chatterjee are current members of the Berkeley faculty. Such a long history!
Visits, seminars, and personal conversations are extremely helpful for exchange of ideas, confirmation, falsification, and crystallization of what is only a thought. I know that Neyman, Elizabeth Scott, David Blackwell, Peter Bickel, and Terry Speed have gone to the ISI; Jeff Wu and Jianqing Fan, both Berkeley PhDs, have too. Extraordinarily influential work on machine learning, high-dimensional inference, genomics, and random matrices is now going on at Berkeley, and students and faculty at the ISI ought to listen to this work face-to-face. These would be timely topics for the Mahalanobis lectures.
A few personal memories. I was Terry’s student on his sufficiency course at the ISI, and just this year he and I worked on putting together Basu’s most influential work in a Springer volume. It was a lovely period of my life. Sandrine Dudoit and I just worked on a survey of sufficiency. David Freedman was advising me on models for fractional data up until six days before his tragic death. I first listened to Peter Bickel in 1980 at a conference at the Delhi ISI. The responsible ISI official greeted a large contingent of us at the Calcutta central train station with a confident toothy smile and said, “Board this train.” Some twenty minutes later, we were all detained for ticketless travelling and given a hefty fine. On our return, ISI reimbursed us for the fines, but it had to be shown as taxi fare. Peter was also taking the train from New Delhi to Calcutta; he was a little late, and at the stairs, I only had time to shake his hand. I properly met Peter for the first time in 1991 at a conference in Ottawa: I gave a very simple talk on extremum efficiencies in some nonparametric problems and he made some comments to me after the talk. Peter invited me to come to Berkeley that Fall, and I spent a month at the MSRI. I was interested in some problems on convolutions at that time and I recall Peter coming to my office at Evans Hall to help me. The little work on t-intervals, Basu and DasGupta (1995), also benefitted from that MSRI visit.
At Purdue, I learned of the close professional and personal relationship between Shanti Gupta and Peter Bickel and Lucien Le Cam. I can vouch that Shanti earnestly counted on counsel from Peter and Lucien Le Cam. In the January of 2002, Shanti passed away most unexpectedly after a brief illness. I informed Peter the same evening. I recall Peter writing back “This is a great loss for Purdue. I am very sorry. It is a personal loss for me. He was a friend.”
That’s what I mean: there is a long and treasured friendship between Indian statistics and Berkeley, and I so much wish to see it prosper and last.