Dear Sir, Greetings. I am a PhD ... student based in [country]. I work in the area of [...] and much of the new models in this area are based on Bayesian approach. I too feel this approach needs to [be] adopted ... . I Sir would too like to learn this approach. Unfortunately here in [country] we in Social Science are [not] trained too well in methodology. This mail is to request you to guide me as to how I can learning Bayesian Modeling relevant to [...] by self-study. There are currently no courses in [country] (to the best of my knowledge) that offer Bayesian Statistics/Bayesian Modeling. Please suggest resources (books, articles, videos) and methodology. I sincerely thank you for going through my mail and taking out you valuable time to help me. Thanks and Regards,
My reply:
Dear [...]:
Thank you for your interest.
Well, there's a vast literature out there, but since you're asking me, I'll recommend the following sequence from my own work.
For an introduction to Bayesian data analysis:
1. Read Chapter 2 of the 2nd Edition of my book. You can preview the book at various sites listed here.
2. Read the article, "Bayesian estimation supersedes the t test," linked here.
3. Read the article, "The time has come: Bayesian methods for data analysis in the organizational sciences," linked here.
For an introduction to Bayesian models of perception and cognition:
1. Read the blog post, "Bayesian models of mind, psychometric models, and data analytic models," linked here.
2. Read the article, "Bayesian learning theory applied to human cognition," linked here.
3. Read the article, "Bayesian approaches to associative learning: From passive to active learning," linked here.
Good luck with your work, and thanks again for your interest.
John
Hi John,
ReplyDeleteI just bought the second edition. I have one major criticism of this work and you post above: you spend far too much time criticizing frequentist methods. The major problem with frequentist methods is that they are abused. The only reason Bayesian methods are not abused widely is that not enough people are using them.
If one succeeds in converting the whole world to using Bayesian methods exclusively, you will have to replace the opening chapters with a discussion of all the ways in which one can abuse Bayesian methods. Unless we can get emotion and the desire to be right all along and pure conflict of interest out of the game when a data analysis is done, abuse of statistical tools will always be with us. And I write this as an enthusiastic user of Bayesian methods.
I would like to see Bayesian methods presented as an important part of the toolkit of a data analyst. Fervour for Bayesian methods should be reserved for things like religion and God.
Even in this post, you suggest to a beginner to first spend time understanding the deficiencies of the frequentist method. It is not necessary to reject the false God of frequentist methodology to change one's denomination. Sure, your criticisms are all correct, but they are unnecessary and distract the reader from the major task of understanding this approach.
If your book were to stick to just getting the story across, it would be much cheaper and shorter. I paid 66 Euros for the E-edition, which I find outrageous. Gelman et al 3rd edition is 20 Euros cheaper and has MUCH more content. What did I pay for? Also, the book is so big that it crashes my pdf viewer (preview). I can't actually read the book without having to install a different pdf viewer.
SV> I just bought the second edition.
ReplyDeleteThank you, I appreciate it.
SV> I have one major criticism of this work and you post above: you spend far too much time criticizing frequentist methods.
Actually, in the book I try to isolate my criticisms of frequentist methods to one chapter. In the opening chapter I mention that there are criticisms and point to the relevant chapter. Scattered throughout there are allusions to criticisms, but the only real space devoted to a critique of p values and confidence intervals is a single chapter.
Should there be such a chapter at all? I think it's valuable for people to be aware of the issue, especially because for a lot of readers it's the first (and maybe only) time they will see such criticisms. I think that having a clear understanding of exactly what a p value (and confidence interval) is crucial for understanding frequentist methods.
SV> The major problem with frequentist methods is that they are abused.
Well, yes, that is a major problem. But I don't think it's the only major problem. I think most users of frequentist methods don't understand what a p value and confidence interval really are.
SV> The only reason Bayesian methods are not abused widely is that not enough people are using them. If one succeeds in converting the whole world to using Bayesian methods exclusively, you will have to replace the opening chapters with a discussion of all the ways in which one can abuse Bayesian methods. Unless we can get emotion and the desire to be right all along and pure conflict of interest out of the game when a data analysis is done, abuse of statistical tools will always be with us. And I write this as an enthusiastic user of Bayesian methods.
I completely agree that Bayesian methods can be abused, like any method. If a person's goal is to manipulate an analysis so that s/he can declare a desired effect to be present, then there are all sorts of ways to bend an analysis toward the desired result. In a Bayesian context, I think it is easier to abuse Bayes factors than estimates of continuous parameter values, which is one of many reasons I stress caution when using Bayes factors.
SV> I would like to see Bayesian methods presented as an important part of the toolkit of a data analyst. Fervour for Bayesian methods should be reserved for things like religion and God.
I always find it strange when religion is brought into these discussions. If anything, I find that adherence to frequentist methods require more blind faith than Bayesian methods, which to me just make rational sense. To the extent there is any tone of zealotry in my writing, it's only because the criticisms of p values and confidence intervals can come as a bit of a revelation after years of using p values without really understanding them. But I do not think of Bayesian methods as the only way to analyze data. It's just the current best alternative to frequentist approaches, at least in principle.
[Reply continues in next comment]
ReplyDeleteSV> Even in this post, you suggest to a beginner to first spend time understanding the deficiencies of the frequentist method.
Sorry, I just don't see that. Chapter 2 of DBDA2E summarizes Bayesian methods and barely mentions frequentism. The articles first summarize Bayesian methods, and then present criticisms of frequentist methods in secondary isolated sections. Again, I think it's important for beginners to see the contrast with frequentist methods, so that they know why to bother with Bayesian methods.
SV> It is not necessary to reject the false God of frequentist methodology to change one's denomination. Sure, your criticisms are all correct, but they are unnecessary and distract the reader from the major task of understanding this approach.
Hmmm. I'm not sure that the criticisms are unnecessary. The vast majority of users of traditional frequentist statistics don't know why they should bother with taking the effort to learn Bayesian methods. I think it's important to present the criticisms. You might want them to be presented with a different "tone."
SV> If your book were to stick to just getting the story across, it would be much cheaper and shorter. I paid 66 Euros for the E-edition, which I find outrageous. Gelman et al 3rd edition is 20 Euros cheaper and has MUCH more content. What did I pay for? Also, the book is so big that it crashes my pdf viewer (preview). I can't actually read the book without having to install a different pdf viewer.
Again, there is not much space spent on criticizing frequentism, and the book would not be much shorter if it were left out. In fact, what makes the book long is two other things: (i) Lots of patient explanation of introductory concepts, and (ii) lots of graphics and diagrams. The book is an introduction, with extensive chapters on basics such as probability density, R programming, etc., none of which is in Gelman et al. The eBook is 26MB (or something like that) because of all the graphics. Sorry you've had trouble viewing it, but I've had no trouble viewing it on any of my devices.
Finally, with respect to the touchy issue of cost, two points: First, the price is controlled primarily by the publisher, and even so there are occasional discounts. Second, perceived outrageousness is largely controlled by what one is comparing with, not by absolute price. Should this book be compared with, say, introductory calculus textbooks? It might look pretty cheap compared with those. Should the value of this book be compared with, say, the price of a ticket to a Broadway show? The reader of the book gets a huge increase in his/her knowledge that lasts a lifetime; the visitor to a show gets an evening's entertainment. If the book at least provides an evening's entertainment, maybe it's not so bad.
In any case, I truly appreciate that you purchased the book and took the time to write your comment. I hope that you find value in the book despite your criticisms.
Hi John,
ReplyDeletemy response is here (your blog limits the length of a comment so I posted it on my own blog):
http://vasishth-statistics.blogspot.de/2014/11/response-to-john-kruschke.html