During late November. I received, by e-mail from Michele Guindani an announcement of the forthcoming webinar (on Nov 27th) of the following article, which has already been published in Bayesian Analysis. [Michele is the editor in chief of this online journal]
Latent Nested Nonparametric Priors"
Federico Camerlenghi, David B. Dunson, Antonio Lijoi, Igor Prünster, and Abel Rodríguez
Upon reading the mind twisting paper, I, with the greatest of respect, found it to consist of a vast amount of high quality, relatively useless, blither, much too brilliantly complex in mathematical terms, and with scant chance of computing any Bayesian solution. Here is my subsequent slightly nutty correspondence with the Bayesian Analysis, Editor in Chief Michele Guindani:
Hi! I won't be here for the webinar discussion. Could you possibly insert this comment in the discussion? Thank you Tom
I helped initiate Bayesian semi-parametrics in the 1970s with my papers in Biometrika 1973 and JRSSB 1978, which complemented Ferguson's 1973 paper in the Annals of Statistics ,I was very interested to see how the subject had advanced during the last 46 years. Upon reading the paper I conclude that it has regressed at least 200 years!!!. While highly complicated and doubtlessly brilliant mathematically, in true Dunsonian style, the authors seem to simply be spinning their wheels without prospect of useful applications [ WITH POTENTIALLY PRECISE PRACTICAL BAYESIAN CONCLUSIONS]. However, I am 71 and getting a bit senile. Maybe I'm missing something profound, When I published my Personal History of Bayesian Statistics with Wiley in 2014, I thought to myself (and kept a bit quiet about it while praising too many people to bits) that DIC was the only useful addition to our previously wonderful subject during the current century, together with the brilliant work by Burkert and Evans on the resolution of Lindley's paradox which still seems to be ignored by everybody who still falls into the standard Bayes factor traps. Now I'm wondering why so Bayesian papers and non-convergent simulation procedures are getting churned out willy nilly, even now and without apparent purpose, But maybe it's something to do with this new-fangled Big Data Analysis. Is Big Data a catch word or is it just something else to justify your meal tickets?
Thu, Nov 21, 6:24 PM (20 hours ago)
Thank you very much. My apologies for the late reply, it has been quite hectic for me over the past week. It is a pity you will not be able to make the seminar discussion. Although I am relatively young, I realize that there has been indeed a change of perspective in many respects, -----[THE REMAINDER OF PROFESSOR GUINDANI'S REPLY HAS BEEN DELETED FOLLOWING HIS CURIOUS INTERVENTION ON THE J ISBA FACEBOOK PAGE. HE APPEARS TO HAVE AGREED WITH ME A TOUCH MORE THAN HE WISHED TO MAKE PUBLIC. IN THE MEANTIME ANOTHER BAYESIAN COLLEAGUE CONTACTED ME TO SAY THAT ONE OF THE CO-AUTHORS OF THE FORTHCOMING WEBINAR PAPER HAD BEEN HEAVILY CRITICISED AT A RECENT CONFERENCE FOR HIS CONTRIBUTIONS TO THE PAPER. THIS IS ALL HIGHLY PERPLEXING]
11:54 AM (2 hours ago)
Thank you for your reply.I guess that I indeed might have said something pertinent. I am taking a back seat now from ISBA, because of my declining heath and because the antics of Bernardo and Carlin and their likes have ruined my memories of conferences long ago, after I helped Arnold Zellner to devise ISBA in the first place. I am also scared of the (massive) way in which Bayesians are spreading their wings and its potential impact on the creation of misinformation and skewed information as the world spirals in its apparent irrational death throes, this followed the flawed leadership of Adrian Smith and Its acolytes during the 1990s and before. I hope this makes sense, Please feel free to repeat or publish any of my utterances which you might care to,
With all best wishes,
P.S. I don't know whether this poem is at all relevant. I composed it yesterday. https://
thomashoskynsleonardblog. blogspot.com/2019/11/bubble- blether-blither-abusive- pariahs.html
On the left ----ISBA's Three founders (Hotel Nikko, San Francisco 1993)
Tom Leonard, Arnold Zellner, and Gordon Kaufmann, with Carl Morris, secomd from right, and two other Bayesian Colleagues, Wesley Johnson and Shanti Gupta I thinlk
Most of Professor Guindani's methodologies invoke highly complex probabilistic models with the zillions of assumptions they entail. I doubt that that many computationally feasible and at all close to convergent procedures,e.g. variations of MCMC,are available for analysing these models except in very special cases. I therefore draw into question, for future consideration, the scientific usefulness of this and any similar Bayesian research
Given my knowledge of Bayesian research in general e.g. as reported on the Approximate Bayesian Computation Facebook Page, I tentatively conclude that there might be a 'circle jerk' in Bayesian Statistics whose participants have some tendency towards publishing articles which are scientifically virtually useless, but which by camouflage appear to be of prize winning quality unless distinguished otherwise by a top and suitably assertive expert in the field. If the editors of our journals happened to be part of this circle, then Society as a whole would be in great trouble indeed. This is of course solely my individual, and highly individualistic perception, which could, given the frailties of my mind, be ENTIRELY MISGUIDED AND INCORRECT,
I do however definitely believe that circle jerks have existed during the history of Bayesian Statistics, though the De Finetti-style 'coherence' circle jerk of the 1970s differed in nature from the 'not really convergent Metropolis algorithm' circle jerk which was initiated, in Bayesian Statistics in 1992.
COMMENT FROM DR, EWART SHAW (WARWICK)
I've always been worried by the shift towards high-dimensional models with parameters far removed from identifiable properties of observables, and from human experience in general. The title "Latent Nested Nonparametric Priors" alone puts me right off with each successive adjective! High-dimensional space is astoundingly 'empty' (witness e.g. the number of directions in Euclidean space no closer together than 60 degrees, as shown by sphere packing). The human mind - mine at least - finds it 'challenging' to come up with appropriate possible models, parametrisations, interrelationships, priors, and representations + interpretations + communication of the resulting posteriors. Maybe I'm just a grumpy old man.