Xiao-Li says yes:
The most compelling reason for having highly visible awards in any field is to enhance its ability to attract future talent. Virtually all the media and public attention our profession received in recent years has been on the utility of statistics in all walks of life. We are extremely happy for and proud of this recognition—it is long overdue. However, the media and public have given much more attention to the Fields Medal than to the COPSS Award, even though the former has hardly been about direct or even indirect impact on everyday life. Why this difference? . . . these awards arouse media and public interest by featuring how ingenious the awardees are and how difficult the problems they solved, much like how conquering Everest bestows admiration not because the admirers care or even know much about Everest itself but because it represents the ultimate physical feat. In this sense, the biggest winner of the Fields Medal is mathematics itself: enticing the brightest talent to seek the ultimate intellectual challenges. . . .
XL invites speculation on “NP-hard problems and (hence) NP-worthy figures in statistics.” My first reaction was No: I like honoring great work but I don’t like the personalization of the Nobel or the competitive aspect of it. Also, I’m guessing that such a prize would create more unhappiness than happiness (see discussion here and here).
Xian agrees: “I do not think we should start on a similar prize by raising money for that purpose. Better support young researchers and international projects. In addition, I fear it would appear as a negative compared with the existing math prizes.”
So I am not convinced by XL’s arguments. On the other hand, I accept that he is much wiser than I am regarding the “real world.” So I’m guessing that he is probably right.
If we do have it, let’s make it about the work, not about the people
If XL is behind something, it’s probably a good idea and it’s probably going to happen. Setting aside the nuts and bolts of actually creating such an award, if we have speculation, I’d be much less interested in speculating who would or should get the prize (indeed, I am turned off by much of the hero-worship surrounding the economics and math prizes), and much more interested in speculating on what work would receive or deserve the prize.
XL talks about Nobel-worthy open problems. That’s fine but I think it might make sense to start by taking a cue from the existing Nobel prizes in chemistry, physics, and biology, and considering what work that has already been done, by people who are currently alive, that might deserve the prize. That is, if this prize were to be given out annually starting tomorrow, what research would or should get it?
Here are some important contributions that come to mind (in no particular order): bootstrap, lasso, generalized estimating equations (I actually hate that stuff but it’s undeniably had lots of inference), false discovery rates (ditto), multiple imputation, various methods in imaging.
Then there’s the big one (to me): hierarchical Bayes. It’s hard for me to single out a particular contribution here, but I’m thinking of the cluster of papers from about 1970 to 1980 where a bunch of researchers demonstrated that multilevel models can work in a general way in many different application areas.
What about computing? Arguably, separate awards could be given to three different packages: Stata, Sas, Bugs, and R. Sas is sort of horrible but it’s doubtless enabled lots and lots of statistics. And Bugs is getting replaced by Stan and various alternatives but it was an important research contribution and indeed is still in use. One might also add Glim but it’s associated with Nelder who is no longer alive so we can’t include it in this list.
What else? Various theoretical ideas, I suppose, although it’s hard for me to weigh the importance of these, as compared with methods and applications. For example, posterior predictive checks are a big deal and getting bigger, but the none of the theoretical papers on the topic (including those of XL and myself) really do the idea justice. Another example would be the work of Berger and others on Lindley’s paradox: I think this work would have to be included because it is so influential and has affected practice as well, even though it did not directly lead to any model or method.
Then there’s probability theory. I’ll let others judge what is the most important work in that area.
What about graphics? The exploratory data analysis revolution is huge, and indeed it is continuing to spread, both on its own and within the world of model-based inference. It’s tough to pinpoint any particular work in this area as representing a key contribution. Even Tukey’s classic work published in the 1970s is more of an inspiration than a contribution—I mean, really: Stem-and-leaf plots? Rootograms? The January temperature in Yuma Arizona?? This work changed the face of statistics but I’d feel uncomfortable giving an award for a set of methods that nobody ever actually uses, or should use. Anyway, it wouldn’t count since Tukey is no longer around, but you get the point.
What about statistical computing? I may stand too close to this subject to be a good judge, but I’m thinking that data augmentation, Hamiltonian Monte Carlo, variational inference, and expectation propagation are four great ideas that have made a difference. We could keep going down the list to include things like slice sampling, simulated tempering, bridge sampling, etc etc. One difficulty is that I don’t quite know where to stop. Should preference be given to ideas such as slice sampling that seem to stand alone as unique contributions? Or does it make more sense to give awards to general areas such as tempering and multigrid methods that have been hammered out by dozens of computational physicists working on lots of problems? Does Gibbs sampling get an award? From the standpoint of statistical computing it’s really nothing special compared to stuff that physicists were doing in the 1940s. On the other hand, the method’s very simplicity has contributed to its wide use among statisticians.
There also must be some important statistical computing that has nothing to do with Bayes. I don’t know if hadoop or whatever deserves a Nobel prize, but you get the idea.
Hmm, what else? The central application areas of statistics are survey sampling and causal inference. Both have seen lots of progress in the past few decades. Again it can be difficult to pick out specific contributions, partly because the theory and applications are often in separate places. It’s not difficult to find examples of theoretical work (for example, by Rubin, Imbens, and Pearl) making fundamental contributions to causal inference, but I think applied work in the area deserves a prize too, perhaps the work of Greenland and Robins. It could be difficult to pick out a single applied paper but the body of work is important. Just as, in biology, the development of a useful lab technique can have important scientific implications, similarly, in statistics, a series of successful applications can lead the way to important methodological developments.
Survey sampling, that’s more difficult to isolate the contributions. The big thing here is Mister P (or so I think), and, again, progress has been slow, starting with work on small-area estimation in the 1970s and then continuing through the 1990s and today with a gradual integration of model-based and design-based approaches. Perhaps the weighting-based and poststratification-based approaches deserve a single shared prize. Experimental design is another hugely important topic in statistics but I don’t know if there have been any really important contributions in this area by researchers who are still alive. I think if experimental design as an old, classical topic (and a topic that’s important enough that it continues to be rediscovered by outsiders), but that might just be my own ignorance.
To go in an entirely different direction: I’m not sure how to consider work that’s tied to specific application areas. If Jun Liu or whoever develops some brilliant method to solve a problem in protein folding, and this method does not generalize at all, is it still prize-worthy? I’d say yes. As the saying goes, statistics is applied statistics. The only trouble with giving prizes within application areas is: (a) the two biggest application areas of statistics are, I’d guess, biology and economics, and these fields already have their own Nobel prizes; and (b) it can be difficult to judge the importance of applied work in an unfamiliar area. I speak as someone whose colleagues (in the early 1990s) ignorantly and rudely dismissed my work in social science. Things have changed, and social science is now a hot area in statistics (“big data” and all that) but the general point remains. I also don’t know how to rank the importance of specific methods. For example, proportional hazard models have had a huge impact in biostatistics. Is that enough to be worth an award all on its own.
Finally, I suspect there are some big ideas I’ve missed, either because I forgot them or just because there are big areas of statistics that I don’t know much about. My main point here is to take the topic that Xiao-Li launched, and to steer it away from discussion of personalities and toward a discussion of ideas and research contributions. Less “who,” more “what.”
Feel free to give your suggestions of Nobel-worthy statistical ideas in the comments.
P.S. Some thoughts from Christian Robert here.