A few stories from graduate school at the University of Minnesota School of Statistics where Seymour Geisser was head. I didn't have as many interactions with Seymour as some folks did, but his philosophy certainly influenced me. Perhaps I'll detail his influence on my research in another post. 

The most common Seymour story involves seminar. Each year he told the graduate students that attending seminar was obligatory, but not mandatory. Or was it mandatory but not obligatory? We interpreted this to mean we really ought to go, but if we could live with ourselves, we could skip. We were obliged to go, but not mandated to go. Me, I attended every seminar except one where I missed not one, but two buses and arrived so late that I would have greatly disturbed seminar, so I left rather than open the closed seminar room door at the front of the room. The next time I missed seminar, I was out of town on job interviews. 

Seymour sat in the front row at seminars, and would crane his head around to inspect the audience. We assumed he was checking on grad student attendance, though we didn't really know. I now do the same thing at seminars, looking around at the audience to see how/whether the graduate students are attending seminar in the numbers they are supposed to be. And yes, it's something I picked up from Seymour. 

Seymour was very enamored with algebra. In his multivariate class we were oft treated to his love of algebraic manipulation. Some proofs/calculations could get complex enough that Seymour would start recycling notation, defining a symbol $a$ to be one thing early in the calculation, then later redefining the symbol to be something different. Unfortunately this offended my righteous sense of rightness. Towards the end of day one in a two day proof (one day = one class lecture), I had a premonition we were to be treated to more redefining of notation. Somewhat imperiously, I requested that Seymour not reuse notation. Actually, it's possible I instructed him not to or insisted that he not redefine notation. He looked startled (that's the closest I can get to describing his facial expression), and I can't recall his verbal response. But sure enough, at the beginning of the second day, $a$ got a new definition. 

Seymour could be quite generous. I opted to work with Dennis Cook for my thesis. Originally we had agreed I would work on optimal design, specifically on optimal design in the presence of non-constant variance. I told plenty of people that that was my research area, but over a 9 month period I spent almost no time on the subject. It was clear I wasn't going to extract a dissertation out of optimal design and I began to look for another topic. Eventually Dennis and I discussed perhaps working on diagnostics, and I think I was particularly interested in Bayesian diagnostics. Well, Bayesian diagnostics was Seymour's purview and he had a grant on the subject. So I went over to meet with him. He showed me his grant, and even allowed me to make a copy to read over to see if there was something in there that was of interest to me. Grants often hold proprietary information on a professor's research program; sharing a grant isn't a given among researchers. 

I eventually settled into a dissertation on Bayesian diagnostics, though not something that derived from Seymour's grant. And eventually I had a final defense and Seymour was on my committee. After presenting a nerve-wracking seminar in front of a large crowd, my dissertation committee and I retired to the library to grill me. There was an interaction with my thesis topic and having Seymour on my committee. My thesis involved a set of measures of case influence along with a graphical influence tool. I knew how to assess influence on the posterior. But as is well known, Seymour was a predictivist; supposedly he did not even believe in parameters except as intermediaries to get to a prediction. And I did not know how to use my measures to assess predictive influence, influence on predictions. The committee members asked me questions in turn. When Seymour's turn came, he started asking me about my time in the consulting clinic. I was a bit startled, because I wasn't expecting this direction. It hadn't previously occurred to me that Seymour might know (and why would he care?) that I had been in the consulting clinic. 

He asked: What are consulting clinic clients most concerned with? I responded with some cant about F-tests and p-values. He pushed: what else are they interested in? And I mentioned they sometimes want estimates and standard errors but not as often. And what else, he asked again. Design sometimes. Sample size. And again he queried. I stonewalled. We danced for half an hour. I answered truthfully, because no one coming in to the consulting clinic had ever once asked about predictions. I knew and everyone in the room knew what Seymour wanted. But because I didn't know how to extend my dissertation to predictions, I wasn't going to volunteer the idea of predictions and wasn't going to make it easy for him to ask the obvious and perfectly sensible question. And Seymour was equally stubborn and wanted me to mention predictions first. Eventually Dennis intervened with a gentle 'Seymour, we talked about this' comment. And the defense continued. 

These days I have a better understanding of what a dissertation defense is about and what the possible outcomes involve. But at the time I thought I was under real threat of flunking the final defense. 

The class of influence measures that I used in my dissertation involved a set of divergence measures between densities. Turns out someone else had investigated them previously, though not for purposes of influence diagnostics. Seymour knew of the paper (Csiszar, 1967). Once I had the reference from Seymour, it still took me a while to find the paper. Pre-internet, this counted as a pretty obscure reference, but Seymour knew it. Eventually I got my thesis published (Weiss and Cook 1992, Biometrika). And a few years after that, I figured out how to do predictive inference as well (Weiss 1996, JRSS-B). Seymour didn't get a citation in the 1992 paper, but I did send a couple citations his way in the 1996 paper.