# It’s World Statistics Day! Death To P-Values, Hypothesis Tests, And False Ascriptions Of Cause! #StatsDay15

Who said the UN doesn’t do anything useful? Courtesy of that august body, it’s World Statistics Day!

Official statistics help decision makers develop informed policies that impact millions of people. Improved data sources,

sound statistical methods, new technologies and strengthened statistical systems enable better decisions that eventually result in better lives for all of us. On 20 October 2015, the global statistical community will showcase their achievements and their ongoing work to help this vision come true. [my emphasis]

Amen to sound statistical methods, and may unsound methods be consigned to the flames (as Hume might say).

Book update. I received an email from a colleague, much more well known than I, who said Wiley asked him to read/review my book. He’s on the outs with Wiley and said no (they weren’t offering any moola, either). But I still took it as good news that the thing is being circulated. Springer also is reviewing.

To have a small—only a small—flavor of the book, these classic posts give some idea. Though the book, rejoicing (thus far) under the title *The Philosophy of Uncertainty: An Introduction*, is much more than these scanty posts and papers.

- Unsignificant Statistics: Or Die P-Value Die Die Die;
- Hypothesis Testing Relies On The Fallacy Of False Dichotomy;
- The Third Way Of Probability & Statistics: Beyond Testing and Estimation To Importance, Relevance, and Skill
- The Crisis Of Evidence: Why Probability And Statistics Cannot Discover Cause
- The Problem Of Grue Isn’t
- On Probability Leakage
- It is Time to Stop Teaching Frequentism to Non-statisticians
- On the non-arbitrary assignment of equi-probable priors

Our friend Mademoiselle Deborah Mayo has the right idea with this title: “Statistical ‘reforms’ without philosophy are blind” We disagree about p-values, causation, hypothesis testing, the importance of parameters (I say they aren’t), and the nature of probability. But besides that, we’re sympatico that philosophy needs to be involved.

Here’s another false dichotomy for you: if you’re not a frequentist, you must be a Bayesian. I have much sympathy with the latter approach, but Bayesian Statistics Isn’t What You Think.

What’s that you ask? You’d like to see the Table of Contents of the book? Well, okay. Here:

1 Truth, Argument, Realism

1.1 Truth

1.2 Realism

1.3 Epistemology

1.4 Necessary \& Conditional Truth

1.5 Science \& Scientism

1.6 Faith

1.7 Belief \& Knowledge

2 Logic

2.1 Language

2.2 Logic is Not Empirical

2.3 Syllogistic Logic

2.4 Syllogisms

2.5 Informality

2.6 Fallacy

3 Induction \& Intellection

3.1 Metaphysics

3.2 Types of Induction

3.3 Grue

4 What Probability Is

4.1 Probability Is Conditional

4.2 Relevance

4.3 The Proportional Syllogism

4.4 Details

4.5 Assigning Probability

4.6 Weight Of Probability

4.7 Probability Usually Is Not A Number

4.8 Probability Can Be A Number

5 What Probability Is Not

5.1 Probability Is Not Physical

5.2 Probability \& Essence

5.3 Probability Is Not Subjective

5.4 Probability Is Not Only Relative Frequency

5.5 Probability Is Not Always a Number Redux

6 Chance \& Randomness

6.1 Randomness

6.2 Not a Cause

6.3 Experimental Design \& Randomization

6.4 Nothing is Distributed

6.5 Quantum Mechanics

6.6 Simulations

6.7 Truly Random \& Information Theory

7 Causality

7.1 What is Cause Like?

7.2 Causal \& Deterministic Models

7.3 Paths

7.4 Once A Cause, Always A Cause

7.5 Falsifiability

7.6 Explanation

7.7 Under-determination

8 Probability Models

8.1 Model Form

8.2 Relevance \& Importance

8.3 Independence Versus Irrelevance

8.4 Bayes

8.5 The Problem And Origin Of Parameters

8.6 Exchangeability And Parameters

8.7 Mystery Of Parameters

9 Statistical \& Physical Models

9.1 The Idea

9.2 The Best Model

9.3 Second-Best Models

9.4 Relevance And Importance

9.5 Measurement

9.6 Hypothesis Testing

9.7 Die, P-value, Die, Die, Die

9.8 Implementing Statistical Models

9.9 Model Goodness

9.10 Decisions

10 Modelling Goals, Strategies, \& Mistakes

10.1 Regression

10.2 Risk

10.3 Epidemiologist Fallacy

10.4 Quantifying The Unquantifiable

10.5 Time Series

10.6 The Future

Comes to about 300 pages in a trade paperback.

Don’t forget I’m ready and rarin’ to go to lecture on these topics. Time is finite, so get your orders in early for the pieces of it I have to offer.

I would like to order a copy. When is it available for pre-order?

John Z,

Well, assuming it gets through the initial review of the publishers who have it now, then I’m guessing it would still take six months to a year, perhaps more, before it comes out. I’m sure I’ll get comments which will lead to some editing on my part.

And that’s if it’s accepted. If it isn’t, add more time!

Self-publishing has its charms. I could charge a lot less for the book and get it out quickly, but I have no sales staff, so it would die a quick death. Probably.

I like the ToC. All very orderly and logical…until you can’t contain your rage against the wee-p. But then you recover and move on.

Have you considered sly cartoons to start off each chapter? Chapter One might be a take on the

“you can’t handle the truth”meme which seems to be your subversive overall theme.Gary,

Being a traditionalist, I use quotes. For instance, this one leads the chapter on probability:

Q: Was the the captain talking about Norris?

How to use statistics to your advantage:

“Now, four years later, Japan has confirmed the first case of cancer stemming from that dangerous work.”

“About 45,000 workers have been involved in cleanup work at the Fukushima plant since August 2011.”

The reported rate of leukemia in Japan is 3.15 in 100,000. One diagnosis among 45,000 workers is hardly unexpected even without Fukushima. The good news is statistics in this case can get you big bucks by suing someone. Pity the guy with leukemia who can’t sue anyone—he loses the cancer lottery.

Gotta love that certainty where there is none and the profitability of not understanding statistics for personal injury lawyers. Overcertainty rocks. 100% certainty of the future—the victim would not have gotten leukemia if he were not at Fukushima, so he wins the lottery. Who knew science could foretell the future with 100% accuracy. Long live statistics.

If you come to central Florida, let me know. I’d love to hear you lecture. I have serious disagreements with you on political issues, but I do sincerely appreciate your approach to statistics. There’s way too much playing around with stats these days and calling it science.

JMJ

TOC is missing my hobby horse.

“How to tell the difference between an outlier and corrupt data”

A.ka. “When do you know you have been Lowendowskied”

Will the book contain fully-worked examples? For instance, the regression in the sample you provided? It seems to me that if the reader can’t start using these better methods after finishing the book then you won’t have started your revolution.

Rich,

Yep. There are not endless examples; only the last Chapter has full, self-contained ones. A separate cookbook could be written on examples. I won’t be the one to write it. If I had students, which I don’t, I’d give the problems to them. The best things to work on are finite-discrete parameterless (

notnon-parametric) models, as these comprise all real things.JMJ,

I’m in central FL usually in March. Tigers spring training at my folks’ place in Lakeland.

Well! That’s what I’m talking about!

The grue paradox has been spotted in the Guardian.