This article provides a guidepost for those who have rarely studied statistics to be able to brilliantly talk about "Bayesian statistics", now and young statistics.
The goal of this article is to be able to talk about Bayesian statistics ** correctly ** with a doy face, and to be able to have a thorough discussion with people who are familiar with statistics.
Keep in mind that studying this article does not mean that you will be able to master Bayesian statistics. To be able to use it in the field, you need to be able to do some programming, and you also need the actual data to be targeted. Before you really do statistical processing, you also need preprocessing ...
Data analysis using statistics requires not only the theory of statistics but also various tools ... tears (Reviving sleepless nights ...)
Basically, I will introduce the study method using books. If you know statistics to some extent, you can skip the books in the middle!
What is Bayesian statistics in this article? I will not go into detail about that. It's an article that you can read if you somehow know that statistics have frequencyism and Bayesianism (and more specifically), and that they are fundamentally different. I will.
For details, please refer to the article below!
-I want to be Bayesian! Bayesian statistics starting from almost zero 1 (Probability and Bayes' theorem) -Now, what is Bayesian statistics?
For the time being, the authors are also vaguely writing, and there are some techie points.
That doesn't mean that these articles are bad. The goal of this article I'm writing now is to be able to understand where and how I'm blurring when I read these articles, or if I'm really writing without knowing.
The only very important thing about Bayesian statistics is Bayesian statistics are neither absolutely better than non-Bayesian statistics nor upward compatible.
It's just an idea. It is a religion.
Absolutely correct statistics are just illusions. There is no dream story in this beautiful and ruthless world that Bayes can solve all the other statistical problems.
But what principle / position do you take? It's just a deep question.
As the name Bayesian ** principle ** suggests, Bayesian statistics is a principle / position, as mentioned above. However, "principle" has become fashionable and obsolete with each era.
In an easy-to-understand example, democracy is the best example. People in the developed world today see democracy as the best political method, but it's not always the best. There must have been a time when principles other than democracy worked in their own way? is. Perhaps democracy will be ridiculed 1000 years later.
In the world of statistics, frequencyism has been criticized in recent years.
-[Statistical layman's opinion on the ASA statement that announced the six principles of p-value, worried about misunderstanding and abuse of statistics and "p-value supremacy"](https://y-mattu.hatenablog.com/entry/ 2016/03/09/03 4248) -Amateur () statement to a statement from a data scientist's blog to a statement from the American Statistical Association that "don't worry too much about p-values and significance, let's end the era when p <0.05 determines everything" -Statistical significance test meaningless
If you read this article, you've probably heard the phrase that there is a "significant difference." The mainstream of statistics in the 20th century was frequency. The "p-value" and "significance" of the "statistical hypothesis test", which is the most convenient tool for this mainstream frequency-based statistics, has been criticized.
So Bayesian statistics are gradually spreading as a frequency-based antithesis.
In addition, handling Bayesian statistics requires a great deal of computing power, and in the last 10 or 20 years, everyone has finally become able to handle Bayesian statistics.
Two years ago when I was in graduate school, an increasing number of researchers insisted not to rely too much on p-values in agriculture (biostatistics), medical statistics, psychometrics, etc., which I specialized in. I felt like I was there.
I think it's better to talk about Bayes if you want to create a cool breeze (laughs) in the future.
To understand Bayesianism properly, first study frequencyism properly lol
Bayesianism and frequencyism are in different positions, so if you don't have to study Bayesianism in order to understand Bayesianism, the wholesaler will not let you down. If you didn't want to study frequencyism and were trying to escape to Bayes, please change your mind secretly lol If you don't want to rehash that rotten guts, gently close this article lol
It ’s better to study frequencyism In order to understand Bayesianism deeply, it is recommended to study while being aware of the difference from frequencyism.
Many Bayesian introductory books and beginner articles write about the differences from frequencyism. Frequent knowledge is always useful when reading such things. This is unavoidable in the history of statistics as I explained earlier.
So, first of all -[Introduction to Statistics Learned with R](https://www.amazon.co.jp/R%E3%81%A7%E5%AD%A6%E3%81%B6%E7%B5%B1%E8% A8% 88% E5% AD% A6% E5% 85% A5% E9% 96% 80-% E5% B6% 8B% E7% 94% B0-% E6% AD% A3% E5% 92% 8C / dp / 4807908596) -[Easy statistics by R](https://www.amazon.co.jp/R%E3%81%AB%E3%82%88%E3%82%8B%E3%82%84%E3%81 % 95% E3% 81% 97% E3% 81% 84% E7% B5% B1% E8% A8% 88% E5% AD% A6-% E5% B1% B1% E7% 94% B0-% E5% 89 % 9B% E5% 8F% B2 / dp / 4274067106 / ref = pd_sbs_14_img_2 / 355-1889098-6063963? _Encoding = UTF8 & pd_rd_i = 4274067106 & pd_rd_r = 6ce8e615-705c-40fb-9e27-91005ac854c3 & pd_rd_w = 1Nune -c84a20b3f3a8 & pf_rd_r = 6WXVVV50XKS626EZJAJ4 & psc = 1 & refRID = 6WXVVV50XKS626EZJAJ4)
It is better to learn the basics with a book that allows you to learn statistics while actually moving your hands. There are rotten technical books detailing frequency-based statistics. However, even if you suddenly tackle a difficult lecture, your heart will only break.
The above book is a content that even beginners of programming can work on, so you can easily and enjoyably imagine statistics.
And next, let's move on to a little theoretical study. personally, -[Basics of Psychological Statistics-For Integrated Understanding](https://www.amazon.co.jp/%E5%BF%83%E7%90%86%E7%B5%B1%E8%A8 % 88% E5% AD% A6% E3% 81% AE% E5% 9F% BA% E7% A4% 8E% E2% 80% 95% E7% B5% B1% E5% 90% 88% E7% 9A% 84 % E7% 90% 86% E8% A7% A3% E3% 81% AE% E3% 81% 9F% E3% 82% 81% E3% 81% AB-% E6% 9C% 89% E6% 96% 90% E9% 96% A3% E3% 82% A2% E3% 83% AB% E3% 83% 9E-% E5% 8D% 97% E9% A2% A8% E5% 8E% 9F-% E6% 9C% 9D% E5% 92% 8C / dp / 4641121605 / ref = sr_1_2? __mk_ja_JP =% E3% 82% AB% E3% 82% BF% E3% 82% AB% E3% 83% 8A & keywords =% E5% BF% 83% E7% 90% 86% E7% B5% B1% E8% A8% 88% E5% AD% A6% E3% 81% AE% E5% 9F% BA% E7% A4% 8E & qid = 1576586725 & s = books & sr = 1-2)
Thank you for this book.
It is very helpful even if you are not a psychologist. In fact, psychology is a university question that laid the foundation for the advancement of statistics.
Is it because it is not clear at a glance whether the operation is really effective, and there is no choice but to experimentally confirm it. To put it the other way around, psychology would have had to be helped by an objectively visible tool called "statistics" in order to be recognized as "science". This is the same, for example, in medicine where you want to check the effect of a drug.
In the above book, it is explained based on how statistics are actually used in psychology, and I think that it is fun to read.
At this point, you should have some idea of frequencyism. Next, read a bridging book between frequencyism and Bayesianism.
-Introduction to Statistical Modeling for Data Analysis-Generalized Linear Model, Hierarchical Bayes Model, MCMC (Science of Probability and Information) E3% 83% BC% E3% 82% BF% E8% A7% A3% E6% 9E% 90% E3% 81% AE% E3% 81% 9F% E3% 82% 81% E3% 81% AE% E7% B5% B1% E8% A8% 88% E3% 83% A2% E3% 83% 87% E3% 83% AA% E3% 83% B3% E3% 82% B0% E5% 85% A5% E9% 96% 80% E2% 80% 95% E2% 80% 95% E4% B8% 80% E8% 88% AC% E5% 8C% 96% E7% B7% 9A% E5% BD% A2% E3% 83% A2% E3% 83% 87% E3% 83% AB% E3% 83% BB% E9% 9A% 8E% E5% B1% A4% E3% 83% 99% E3% 82% A4% E3% 82% BA% E3% 83% A2% E3% 83% 87% E3% 83% AB% E3% 83% BBMCMC-% E7% A2% BA% E7% 8E% 87% E3% 81% A8% E6% 83% 85% E5% A0 % B1% E3% 81% AE% E7% A7% 91% E5% AD% A6-% E4% B9% 85% E4% BF% 9D-% E6% 8B% 93% E5% BC% A5 / dp / 400006973X / ref = pd_bxgy_14_img_3 / 355-1889098-6063963? _encoding = UTF8 & pd_rd_i = 400006973X & pd_rd_r = d6348a6c-3a95-4ca8-9c1f-f98017a68392 & pd_rd_w = 5Rdyl & pd_rd_wg = KsUqF & pf_rd_p = b25bd748-082b-4f2a-b724-125316a35a9c & pf_rd_r = DAJ3VED9BEE81TBADM7R & psc = 1 & refRID = DAJ3VED9BEE81TBADM7R)
In the positioning of bridging between frequency principle and Bayesian principle, there is no other book (commonly known as Midori book). This book has been acclaimed in many articles other than this one.
You can easily and enjoyably feel the expansion of the world of statistics. I also read this book and became more fond of statistics. Statistics, which I thought was not fun because it was just mathematical formulas, seemed to be colored at once when I encountered the Midori book.
It is no exaggeration to say that ** a masterpiece that changed my life **.
The great thing about Midori books is
Understanding MCMC is essential to understanding Bayes. The Markov Chain Monte Carlo method, a very difficult name, was absolutely necessary for Bayes to be put to practical use. At the same time, he is also the person who firmly defended Bayesian chastity until recent years when the power of computers increased.
After this book
-[Bayes Statistics from the Basics: A Practical Introduction to the Hamiltonian Monte Carlo Method](https://www.amazon.co.jp/%E5%9F%BA%E7%A4%8E%E3%81%8B%E3 % 82% 89% E3% 81% AE% E3% 83% 99% E3% 82% A4% E3% 82% BA% E7% B5% B1% E8% A8% 88% E5% AD% A6-% E3% 83% 8F% E3% 83% 9F% E3% 83% AB% E3% 83% 88% E3% 83% 8B% E3% 82% A2% E3% 83% B3% E3% 83% A2% E3% 83% B3% E3% 83% 86% E3% 82% AB% E3% 83% AB% E3% 83% AD% E6% B3% 95% E3% 81% AB% E3% 82% 88% E3% 82% 8B% E5% AE% 9F% E8% B7% B5% E7% 9A% 84% E5% 85% A5% E9% 96% 80-% E8% B1% 8A% E7% 94% B0-% E7% A7% 80% E6% A8% B9 / dp / 4254122128 / ref = pd_sbs_14_7? _encoding = UTF8 & pd_rd_i = 4254122128 & pd_rd_r = 1d0bdb0a-431d-4c19-a64f-0f82f4396152 & pd_rd_w = UWGob & pd_rd_wg = nVX9V & pf_rd_p = 1585d594-d9d0-474b-8a4e-69eca1566911 & pf_rd_r = VW5FXY54YJ2Y1Q7JXYJY & psc = 1 & refRID = VW5FXY54YJ2Y1Q7JXYJY) -Introduction to Bayesian Statistics for Complete Self-study
Let's read a little more Bayes Bayes (laughs) book. Neither is so difficult, and it is recommended that it is written so that you can understand something like "the meaning of using Bayes" rather than getting into a technically difficult story.
Read the Midori books and these and you'll be at the entrance to the deep world of Bayes!
After going through the above study process, it's a good idea to actually move your hands while studying!
-Bayesian statistical modeling with Stan and R -[Introduction to MCMC with Bayesian reasoning PyMC experienced in Python](https://www.amazon.co.jp/Python%E3%81%A7%E4%BD%93%E9%A8%93%E3%81%] 99% E3% 82% 8B% E3% 83% 99% E3% 82% A4% E3% 82% BA% E6% 8E% A8% E8% AB% 96-PyMC% E3% 81% AB% E3% 82% 88% E3% 82% 8BMCMC% E5% 85% A5% E9% 96% 80-% E3% 82% AD% E3% 83% A3% E3% 83% A1% E3% 83% AD% E3% 83% B3 -% E3% 83% 87% E3% 83% 93% E3% 83% 83% E3% 83% 89% E3% 82% BD% E3% 83% B3-% E3% 83% 94% E3% 83% AD % E3% 83% B3 / dp / 4627077912 / ref = sr_1_5? __mk_ja_JP =% E3% 82% AB% E3% 82% BF% E3% 82% AB% E3% 83% 8A & keywords =% E3% 83% 99% E3 % 82% A4% E3% 82% BA + python & qid = 1576588497 & s = books & sr = 1-5)
It is a step to actually use Bayes. Best of all, these books help you understand MCMC by moving your hands.
There is no one who actually uses Bayes and cannot program. That's because, as I explained at the beginning, there are many other things you need to do to work with data in Bayes. Collecting data, sticking it together, cleaning it up, doing things like that and doing things like this.
These books are written on the premise that you can program in their own way, but if you have reached this point, you will surely be able to have a fun Bayesian life with a book in your hand.
It's a personal impression, but Bayes is very interesting. The seemingly illusory scholarship of statistics reveals a soft and adorable side to learning Bayes.
But I can see some deep darkness in that smile ...
Would you like to take a peek into such a mysterious and fascinating world of Bayesian statistics?
Recommended Posts