After a very, very long delay, we eventually re-revised our paper about necessary and sufficient conditions on summary statistics to be relevant for model choice (i.e. to lead to consistent tests). Reasons, both good and bad, abound for this delay! Some (rather bad) were driven by the completion of a certain new edition… Some (fairly good) are connected with the requests from the Series B editorial team, towards improving our methodological input. As a result we put more emphasis on the post-ABC cross-checking for the relevance of the summary choice, via a predictive posterior evaluation of the means of the summary statistic under both models and a test for mean equality. And re-ran a series of experiments on a three population population genetic example. Plus, on the side, simplified some of our assumptions. I dearly hope the paper can make it through but am also looking forward the opinion of the Series B editorial team The next version of Relevant statistics for Bayesian model choice should be arXived by now (meaning when this post appears!).
Archive for Series B
A few days ago, Dennis Prangle, Paul Fernhead, and their co-authors from New Zealand have posted on arXiv their (long-awaited) study of the selection of summary statistics for ABC model choice. And I read it during my trip to England, in trains and planes, if not when strolling in the beautiful English countryside as above.
As posted several times on this ‘Og, the crux of the analysis is that the Bayes factor is a good type of summary when comparing two models, this result extending to more model by considering instead the vector of evidences. As in the initial Read Paper by Fearnhead and Prangle, there is no true optimality in using the Bayes factor or vector of evidences, strictly speaking, besides the fact that the vector of evidences is minimal sufficient for the marginal models (integrating out the parameters). (This was a point made in my discussion.) The implementation of the principle is similar to this Read Paper setting as well: run a pilot ABC simulation, estimate the vector of evidences, and re-run the main ABC simulation using this estimate as the summary statistic. The paper contains a simulation study using some of our examples (in Marin et al., 2012), as well as an application to genetic bacterial data. Read more »
We have now completed our revision of the paper Relevant statistics for Bayesian model choice, written with Judith Rousseau, Jean-Michel Marin, and Natesh Pillai. It has been resubmitted to Series B and reposted on arXiv. The major change in the paper is the inclusion of a check about the relevance of a given summary statistics, as already explained in the talks I presented in Bristol and Glasgow. We also ran a realistic (and, I think, illuminating!) experiment to assess the impact of using one or two (δμ)² statistics as summaries in a simple population experiment, along with a theoretical explanation of the difference between both cases. This methodological addition answers in my opinion the major criticism contained in the review and I thus hope we can envision the eventual publication of this paper… In any case, the reviews have been tremendously helpful in improving the paper.
Here are the (revised) slides of my talk this afternoon at the Confronting Intractability in Statistical Inference workshop in Bristol, supported by SuSTain. The novelty is in the final part, where we managed to apply our result to a three population genetic escenario using one versus two δμ summary statistics. This should be the central new example in the incoming revision of our paper to Series B.
More generally, the meeting is very interesting, with great talks and highly relevant topics: e.g., yesterday, I finally understood what transportation models meant (at the general level) and how they related to copula modelling, saw a possible connection from computer models to ABC, got inspiration to mix Gaussian processes with simulation output, and listened to the whole exposition of Simon Wood’s alternative to ABC (much more informative than the four pages of his paper in Nature!). Despite (or due to?) sampling Bath ales yesterday night, I even woke up early enough this morning to run over and under the Clifton suspension bridge, with a slight drizzle that could not really be characterized as rain…
Statisfaction made me realise I had missed the latest ISBA Bulletin when I read what Julyan posted about Larry’s tribune on a World without referees. While I agree on many of Larry’s points, first and foremost on his criticisms of the refereeing process which seems to worsen and worsen, here are a few items of dissension…
The argument that the system is 350 years old and thus must be replaced may be ok at the rethoretical level, but does not carry any serious weight! First, what is the right scale for a change: 100 years?! 200 years?! Should I burn down my great-grand-mother’s house because it is from the 1800′s and buy a camping-car instead?! Should I smash my 1690 Stradivarius and buy a Fender Stratocaster?! Further, given the intensity and the often under-the-belt level of the Newton vs. Leibniz dispute, maybe refereeing and publishing in the Philosophical Transactions of the Royal Society should have been abolished right away from the start. Anyway, this is about rethoric, not matter. (Same thing about the wine store ellipse. It is not even a good one: Indeed, when I go to a wine store, I have to rely on (a) well-known brands; (b) brands I have already tried and appreciated; (c) someone else’s advice, like the owner, or friends, or Robert Parker…. In the former case, it can prove great or disastrous. But this is the most usual way to pick wines as one cannot hope [dream?] to sample all wines in the shop.)
My main issue with doing away with referees is the problem of sifting through the chaff. The amount of research documents published everyday is overwhelming. There is a maximal amount of time I can dedicate to looking at websites, blogs, twitter accounts like Scott Sisson’s and Richard Everitt’s, and such. And there clearly is a limited amount of trust I put in the opinions expressed in a blog (e.g., take the ‘Og where this anonymous X’racter writes about everything, mostly non-scientific stuff, and reviews papers with a definite bias!) Even keeping track of new arXiv postings sometimes get overwhelming. So, Larry’s “if you don’t check arXiv for new papers every day, then you are really missing out” means to me that missing arXiv for a few days and I cannot recover. One week away at an intense workshop or on vacations and I am letting some papers going by forever, even though I carry them in my bag for a while…. Noll’s suggestion to publish only on one’s own website is even more unrealistic: why should anyone bother to comment on poor or wrong papers, except when looking for ‘Og’s fodder?! So the fundamental problem is separating the wheat from the chaff, given the amount of chaff and the connected tendency to choke on it! Getting rid of referees and journals to rely on depositories like [the great, terrific, essential] arXiv forces me to also rely on other sources for ranking, selecting, and eliminating papers. Again with a component of arbitrariness, subjectivity, bias, variation, randomness, peer pressure, &tc. In addition, having no prior check of papers means reading a new paper a tremendous chore as one would have to check the references as well, leading to a sort of infinite regress… and forcing one to rely on reputation and peer opinions, once again! And imagine the inflation in reference letters! I already feel I have to write too many reference letters at the moment, but a world without (good and bad) journals would be the Hell of non-stop reference letters. I definitely prefer to referee (except for Elsevier!) and even more being a journal editor, because I can get an idea of the themes in the field and sometimes spot new trends, rather than writing over and over again about an old friend’s research achievements or having to assess from scratch the worth of a younger colleague’s work…
Furthermore, and this is a more general issue, I do not believe that the multiplication of blogs, websites, opinion posts, tribunes, &tc., is necessarily a “much more open, democratic approach”: everyone voicing an opinion on the Internet does not always get listened to and the loudest ones (or most popular ones) are not always the most reliable ones. A complete egalitarian principle means everyone talks/writes and no one listens/reads: I’d rather stick to the principles set by the Philosophical Transactions of the Royal Society!
Anyway, thanks to Larry for launching a worthwhile debate into discovering new ways of making academia a more rational and scientific place!