Archive for self-study

Solution manual to Bayesian Core on-line

Posted in Books, Statistics, University life with tags , , , , , , on October 25, 2009 by xi'an

The solution manual to Bayesian Core was initially intended only for instructors using the book as a textbook, so that these instructors could entirely rely on the exercises to provide graded homeworks, as I do in Dauphine. However, this policy of ours induced repeated criticisms and urgent requests from readers unable to complete some of the essential exercises. Since some exercises were stepping stones for following sections and also because of the lack of evidence that the exercises were truly used to set homeworks by instructors, Jean-Michel Marin and myself came to realise (albeit late in the day!) that some solutions were needed by some (self-study or not) readers. From this realisation, the move to make the whole set of solutions available to all readers was a rather natural step that we undertook yesterday. (In addition, this will prevent some unsavoury characters from making money by selling the manual to Bayesian Core online!) This specially makes sense when we are contemplating an incoming revision of Bayesian Core towards a Use R! oriented version, with an attempt at reducing the math complexity, which is another reproach found in some of the published criticisms. Therefore, lo and behold!, and truly “by popular request”, the solution manual to Bayesian Core is now available for free use and duplication by anyone, not only by instructors, on the book webpage as well as on Springer Verlag’s website. And it should also be posted on arXiv tomorrow the day after tomorrow (Tue 27 Oct 09 00:00:00 GMT), which means that the LaTeX code will be accessible as well.

However, there is (at least) one caveat to the opening of the manual to all: since this solution manual was first intended (and written) for instructors using Bayesian Core, some self-study readers will undoubtedly come to the realisation that the solutions provided here are too sketchy for them because the way we wrote those solutions assumes some minimal familiarity with the maths, with the probability theory and with the statistics behind the arguments. There is unfortunately a limit to the time and to the efforts we can dedicate to this solution manual and studying Bayesian Core requires some prerequisites in maths (such as matrix algebra and Riemann integrals), in probability theory (such as the use of joint and conditional densities) and some bases of statistics (such as the notions of inference, sufficiency and confidence sets) that we cannot cover here. Casella and Berger’s (2001) Statistical Inference is a good reference book in case a reader is lost with the “basic” concepts or sour ketchy math derivations. In connection with this point, we both came to realise that describing the book as “self-contained” was a dangerous add as readers were naturally inclined to always relate this term to their current state of knowledge, a bias resulting in inappropriate expectations. (For instance, some students unfortunately came to one of our short courses with no previous exposure to standard distributions like the t or the gamma distributions.)

We obviously welcome comments and questions on possibly erroneous solutions, as well as suggestions for more elegant or more complete solutions: since this manual is distributed both freely and independently from the book, it can be updated and corrected [almost] in real time! Note however that the R codes given in the solution manual to Bayesian Core are not optimised because we prefer to use simple and understandable codes, rather than condensed and efficient codes, both for time constraints (this manual took me about a whole week of August 2007 to complete!) and for pedagogical purposes: the readers must be able to grasp the meaning of the R code with a minimum of effort since R programming is not supposed to be an obligatory entry to the book. In this respect, using R replaces the pseudo-code found in other books since it can be implemented as such but does not restrict understanding. Therefore, if you find better [meaning, more efficient/faster] codes than those provided along those pages, I would be glad to hear your comments, but that does not mean that we will automatically substitute your R code for the current one, because readability is also an important factor. (This post is adapted from the manual preface.)

%d bloggers like this: