prior against truth!

A question from X validated had interesting ramifications, about what happens when the prior does not cover the true value of the parameter (assuming there ? In fact, not so much in that, from a decision theoretic perspective, the fact that that π(θ⁰)=0, or even that π(θ)=0 in a neighbourhood of θ⁰ does not matter [too much]. Indeed, the formal derivation of a Bayes estimator as minimising the posterior loss means that the resulting estimator may take values that were “impossible” from a prior perspective! Indeed, taking for example the posterior mean, the convex combination of all possible values of θ under π may well escape the support of π when this support is not convex. Of course, one could argue that estimators should further be restricted to be possible values of θ under π but that would reduce their decision theoretic efficiency.

An example is the brilliant minimaxity result by George Casella and Bill Strawderman from 1981: when estimating a Normal mean μ based on a single observation xwith the additional constraint that |μ|<ρ, and when ρ is small enough, ρ≤1.0567 quite specifically, the minimax estimator for this problem under squared error loss corresponds to a (least favourable) uniform prior on the pair {−ρ,ρ}, meaning that π gives equal weight to −ρ and ρ (and none to any other value of the mean μ). When ρ increases above this bound, the least favourable prior sees its support growing one point at a time, but remaining a finite set of possible values. However the posterior expectation, 𝔼[μ|x], can take any value on (−ρ,ρ).

In an even broader suspension of belief (in the prior), it may be that the prior has such a restricted support that it cannot consistently estimate the (true value of the) parameter, but the associated estimator may remain admissible or minimax.

non-local priors for mixtures

[For some unknown reason, this commentary on the paper by Jairo Fúquene, Mark Steel, David Rossell —all colleagues at Warwick— on choosing mixture components by non-local priors remained untouched in my draft box…]

Choosing the number of components in a mixture of (e.g., Gaussian) distributions is a hard problem. It may actually be an altogether impossible problem, even when abstaining from moral judgements on mixtures. I do realise that the components can eventually be identified as the number of observations grows to infinity, as demonstrated for instance by Judith Rousseau and Kerrie Mengersen (2011). But for a finite and given number of observations, how much can we trust any conclusion about the number of components?! It seems to me that the criticism about the vacuity of point null hypotheses, namely the logical absurdity of trying to differentiate θ=0 from any other value of θ, applies to the estimation or test on the number of components of a mixture. Doubly so, one might argue, since a very small or a very close component is undistinguishable from a non-existing one. For instance, Definition 2 is correct from a mathematical viewpoint, but it does not spell out the multiple contiguities between k and k’ component mixtures.

The paper starts with a comprehensive coverage of l’état de l’art… When using a Bayes factor to compare a k-component and an h-component mixture, the behaviour of the factor is quite different depending on which model is correct. Essentially overfitted mixtures take much longer to detect than underfitted ones, which makes intuitive sense. And BIC should be corrected for overfitted mixtures by a canonical dimension λ between the true and the (larger) assumed number of parameters  into

2 log m(y) = 2 log p(y|θ) – λ log O(n) + O(log log n)

I would argue that this purely invalidates BIG in mixture settings since the canonical dimension λ is unavailable (and DIC does not provide a useful substitute as we illustrated a decade ago…) The criticism about Rousseau and Mengersen (2011) over-fitted mixture that their approach shrinks less than a model averaging over several numbers of components relates to minimaxity and hence sounds both overly technical and reverting to some frequentist approach to testing. Replacing testing with estimating sounds like the right idea.  And I am also unconvinced that a faster rate of convergence of the posterior probability or of the Bayes factor is a relevant factor when conducting

As for non local priors, the notion seems to rely on a specific topology for the parameter space since a k-component mixture can approach a k’-component mixture (when k'<k) in a continuum of ways (even for a given parameterisation). This topology seems to be summarised by the penalty (distance?) d(θ) in the paper. Is there an intrinsic version of d(θ), given the weird parameter space? Like one derived from the Kullback-Leibler distance between the models? The choice of how zero is approached clearly has an impact on how easily the “null” is detected, the more because of the somewhat discontinuous nature of the parameter space. Incidentally, I find it curious that only the distance between means is penalised… The prior also assumes independence between component parameters and component weights, which I think is suboptimal in dealing with mixtures, maybe suboptimal in a poetic sense!, as we discussed in our reparameterisation paper. I am not sure either than the speed the distance converges to zero (in Theorem 1) helps me to understand whether the mixture has too many components for the data’s own good when I can run a calibration experiment under both assumptions.

While I appreciate the derivation of a closed form non-local prior, I wonder at the importance of the result. Is it because this leads to an easier derivation of the posterior probability? I do not see the connection in Section 3, except maybe that the importance weight indeed involves this normalising constant when considering several k’s in parallel. Is there any convergence issue in the importance sampling solution of (3.1) and (3.3) since the simulations are run under the local posterior? While I appreciate the availability of an EM version for deriving the MAP, a fact I became aware of only recently, is it truly bringing an improvement when compared with picking the MCMC simulation with the highest completed posterior?

The section on prior elicitation is obviously of central interest to me! It however seems to be restricted to the derivation of the scale factor g, in the distance, and of the parameter q in the Dirichlet prior on the weights. While the other parameters suffer from being allocated the conjugate-like priors. I would obviously enjoy seeing how this approach proceeds with our non-informative prior(s). In this regard, the illustration section is nice, but one always wonders at the representative nature of the examples and the possible interpretations of real datasets. For instance, when considering that the Old Faithful is more of an HMM than a mixture.

approximations of Markov Chains [another garden of forking paths]

James Johndrow and co-authors from Duke wrote a paper on approximate MCMC that was arXived last August and that I missed. David Dunson‘s talk at MCMski made me aware of it. The paper studies the impact of replacing a valid kernel with a close approximation. Which is a central issue for many usages of MCMC in complex models, as exemplified by the large number of talks on that topic at MCMski.

“All of our bounds improve with the MCMC sample path length at the expected rate in t.”

A major constraint in the paper is Doeblin’s condition, which implies uniform geometric ergodicity. Not only it is a constraint on the Markov kernel but it is also one for the Markov operator in that it may prove impossible to… prove. The second constraint is that the approximate Markov kernel is close enough to the original, which sounds reasonable. Even though one can always worry that the total variation norm is too weak a norm to mean much. For instance, I presume with some confidence that this does not prevent the approximate Markov kernel from not being ergodic, e.g., not irreducible, not absolutely continuous wrt the target, null recurrent or transient. Actually, the assumption is stronger in that there exists a collection of approximations for all small enough values ε of the total variation distance. (Small enough meaning ε is much smaller than the complement α to 1 of the one step distance between the Markov kernel and the target. With poor kernels, the approximation must thus be very good.) This is less realistic than assuming the availability of one single approximation associated with an existing but undetermined distance ε. (For instance, the three examples of Section 3 in the paper show the existence of approximations achieving a certain distance ε, without providing a constructive determination of such approximations.) Under those assumptions, the average of the sequence of Markov moves according to the approximate kernel converges to the target in total variation (and in expectation for bounded functions). With sharp bounds on those distances. I am still a bit worried at the absence of conditions for the approximation to be ergodic.

“…for relatively short path lengths, there should exist a range of values for which aMCMC offers better performance in the compminimax sense.”

The paper also includes computational cost into the picture. Introducing the notion of compminimax error, which is the smallest (total variation) distance among all approximations at a given computational budget. Quite an interesting, innovative, and relevant notion that may however end up being too formal for practical use. And that does not include the time required to construct and calibrate the approximations.

how individualistic should statistics be?

Keli Liu and Xiao-Li Meng completed a paper on the very nature of inference, to appear in The Annual Review of Statistics and Its Application. This paper or chapter is addressing a fundamental (and foundational) question on drawing inference based a sample on a new observation. That is, in making prediction. To what extent should the characteristics of the sample used for that prediction resemble those of the future observation? In his 1921 book, A Treatise on Probability, Keynes thought this similarity (or individualisation) should be pushed to its extreme, which led him to somewhat conclude on the impossibility of statistics and never to return to the field again. Certainly missing the incoming possibility of comparing models and selecting variables. And not building so much on the “all models are wrong” tenet. On the contrary, classical statistics use the entire data available and the associated model to run the prediction, including Bayesian statistics, although it is less clear how to distinguish between data and control there. Liu & Meng debate about the possibility of creating controls from the data alone. Or “alone” as the model behind always plays a capital role.

“Bayes and Frequentism are two ends of the same spectrum—a spectrum defined in terms of relevance and robustness. The nominal contrast between them (…) is a red herring.”

The paper makes for an exhilarating if definitely challenging read. With a highly witty writing style. If only because the perspective is unusual, to say the least!, and requires constant mental contortions to frame the assertions into more traditional terms.  For instance, I first thought that Bayesian procedures were in agreement with the ultimate conditioning approach, since it conditions on the observables and nothing else (except for the model!). Upon reflection, I am not so convinced that there is such a difference with the frequentist approach in the (specific) sense that they both take advantage of the entire dataset. Either from the predictive or from the plug-in distribution. It all boils down to how one defines “control”.

“Probability and randomness, so tightly yoked in our minds, are in fact distinct concepts (…) at the end of the day, probability is essentially a tool for bookkeeping, just like the abacus.”

Some sentences from the paper made me think of ABC, even though I am not trying to bring everything back to ABC!, as drawing controls is the nature of the ABC game. ABC draws samples or control from the prior predictive and only keeps those for which the relevant aspects (or the summary statistics) agree with those of the observed data. Which opens similar questions about the validity and precision of the resulting inference, as well as the loss of information due to the projection over the summary statistics. While ABC is not mentioned in the paper, it can be used as a benchmark to walk through it.

“In the words of Jack Kiefer, we need to distinguish those problems with `luck data’ from those with `unlucky data’.”

I liked very much recalling discussions we had with George Casella and Costas Goutis in Cornell about frequentist conditional inference, with the memory of Jack Kiefer still lingering around. However, I am not so excited about the processing of models here since, from what I understand in the paper (!), the probabilistic model behind the statistical analysis must be used to some extent in producing the control case and thus cannot be truly assessed with a critical eye. For instance, of which use is the mean square error when the model behind is unable to produce the observed data? In particular, the variability of this mean squared error is directly driven by this model. Similarly the notion of ancillaries is completely model-dependent. In the classification diagrams opposing robustness to relevance, all methods included therein are parametric. While non-parametric types of inference could provide a reference or a calibration ruler, at the very least.

Also, by continuously and maybe a wee bit heavily referring to the doctor-and-patient analogy, the paper is somewhat confusing as to which parts are analogy and which parts are methodology and to which type of statistical problem is covered by the discussion (sometimes it feels like all problems and sometimes like medical trials).

“The need to deliver individualized assessments of uncertainty are more pressing than ever.”

 A final question leads us to an infinite regress: if the statistician needs to turn to individualized inference, at which level of individuality should the statistician be assessed? And who is going to provide the controls then? In any case, this challenging paper is definitely worth reading by (only mature?) statisticians to ponder about the nature of the game!

minimaxity of a Bayes estimator

Today, while in Warwick, I spotted on Cross Validated a question involving “minimax” in the title and hence could not help but look at it! The way I first understood the question (and immediately replied to it) was to check whether or not the standard Normal average—reduced to the single Normal observation by sufficiency considerations—is a minimax estimator of the normal mean under an interval zero-one loss defined by

where L is a positive tolerance bound. I had not seen this problem before, even though it sounds quite standard. In this setting, the identity estimator, i.e., the normal observation x, is indeed minimax as (a) it is a generalised Bayes estimator—Bayes estimators under this loss are given by the centre of an equal posterior interval—for this loss function under the constant prior and (b) it can be shown to be a limit of proper Bayes estimators and its Bayes risk is also the limit of the corresponding Bayes risks. (This is a most traditional way of establishing minimaxity for a generalised Bayes estimator.) However, this was not the question asked on the forum, as the book by Zacks it referred to stated that the standard Normal average maximised the minimal coverage, which amounts to the maximal risk under the above loss. With the strange inversion of parameter and estimator in the minimax risk:

which makes the first bound equal to 0 by equating estimator and mean μ. Note however that I cannot access the whole book and hence may miss some restriction or other subtlety that would explain for this unusual definition. (As an aside, note that Cross Validated has a protection against serial upvoting, So voting up or down at once a large chunk of my answers on that site does not impact my “reputation”!)

reading classics (#9,10)

Today was the very last session of our Reading Classics Seminar for the academic year 2013-2014. We listened two presentations, one on the Casella and Strawderman (1984) paper on the estimation of the normal bounded mean. And one on the Hartigan and Wong’s 1979 K-Means Clustering Algorithm paper in JRSS C. The first presentation did not go well as my student had difficulties with the maths behind the paper. (As he did not come to ask me or others for help, it may well be that he put this talk together at the last minute, at a time busy with finals and project deliveries. He also failed to exploit those earlier presentations of the paper.) The innovative part in the talk was the presentation of several R simulations comparing the risk of the minimax Bayes estimator with the one for the MLE. Although the choice of simulating different samples of standard normals for different values of the parameters and even for both estimators made the curves (unnecessarily) all wiggly.

By contrast, the second presentation was very well-designed, with great Beamer slides, interactive features and a software oriented focus. My student Mouna Berrada started from the existing R function kmeans to explain the principles of the algorithm, recycling the interactive presentation of last year as well (with my permission), and creating a dynamic flowchart that was most helpful. So she made the best of this very short paper! Just (predictably) missing the question of the statistical model behind the procedure. During the discussion, I mused why k-medians clustering was not more popular as it offered higher robustness guarantees, albeit further away from a genuine statistical model. And why k-means clustering was not more systematically compared with mixture (EM) estimation.

Here are the slides for the second talk

uniformly most powerful Bayesian tests???

“The difficulty in constructing a Bayesian hypothesis test arises from the requirement to specify an alternative hypothesis.”

Vale Johnson published (and arXived) a paper in the Annals of Statistics on uniformly most powerful Bayesian tests. This is in line with earlier writings of Vale on the topic and good quality mathematical statistics, but I cannot really buy the arguments contained in the paper as being compatible with (my view of) Bayesian tests. A “uniformly most powerful Bayesian test” (acronymed as UMBT)  is defined as

“UMPBTs provide a new form of default, nonsubjective Bayesian tests in which the alternative hypothesis is determined so as to maximize the probability that a Bayes factor exceeds a specified threshold”

which means selecting the prior under the alternative so that the frequentist probability of the Bayes factor exceeding the threshold is maximal for all values of the parameter. This does not sound very Bayesian to me indeed, due to this averaging over all possible values of the observations x and comparing the probabilities for all values of the parameter θ rather than integrating against a prior or posterior and selecting the prior under the alternative with the sole purpose of favouring the alternative, meaning its further use when the null is rejected is not considered at all and catering to non-Bayesian theories, i.e. trying to sell Bayesian tools as supplementing p-values and arguing the method is objective because the solution satisfies a frequentist coverage (at best, this maximisation of the rejection probability reminds me of minimaxity, except there is no clear and generic notion of minimaxity in hypothesis testing).