Welcome back to Beyond the Patterns. So today I have the great pleasure to announce Dr.

Anton Butlerner, who is a former member of our lab. So he will present on his latest

insights on significance testing and how worshipping p-values can actually bring more trouble than

benefits. Anton has received his doctor degree in phonetics in 1978 at LMU Munich. He has been

with the Institute of Nordic Languages and the Institute for German Philology, both at LMU Munich.

The IMS at Stuttgart University, then the Pattern Recognition Lab at FAU Erlangen-Nürnberg,

the Institute for Human-Machine Communication at TU Munich, the Chair of Complex and Intelligent

Systems at the University of Passau and the Chair of Embedded Intelligence for Healthcare

and Well-being at University of Augsburg. He is co-editor, author of two books and co-author

of more than 300 technical articles with an age index of 48 and more than 11,000 citations.

His main interests are all cross-lingual aspects of prosody and computational

paralinguistics. He repeatedly served as co-organizer for workshops and sessions,

challenges on emotion and other paralinguistic events at ELRIG, ICPHS, speech prosody and

interspeech. He was guest editor for AHCI, CSL and speech communication, associated editor for IEEE

transactions on effective computing, as well as reviewer for numerous leading journals,

conferences and workshops. So we can't summarize all of the contributions by Anton to the field,

but of course I can take the opportunity to announce him here as our invited speaker and

today's presentation will be entitled Magic Numbers vs. Sound Reasoning Moving Towards a

World Beyond P Smaller Than 5%. Anton, without further ado, I'd like to hand the award over

and the stage is yours.

Okay, thank you Andy. Yeah, I will start with a very short history, some glimpses on history.

First, four years ago, I was very astonished that I could read a whole page in the Süddeutsche Zeitung

telling us the magical P in English, the right statisticians are often wide off the mark. The

magic P is only a statistical parameter that is not known outside science, yet it often decides

on life or death. Should we deprive it of its power? And it's really a good article, it's

worthwhile to read it. But one thing is wrong, it's not statisticians actually that push the

magic P, it's rather practitioners. So what are P values? I guess these are frequent opinions

about P values, they tell you how probable the result is, if they're very small, the result is

correct, if they are above a threshold, the result is not correct, all this is wrong.

In fact, they are important because people believe that they are important.

So I'm really asking myself whether the last 60 years have been in vain. This is a rather

well-known paper with almost a thousand citations before the time of the internet from 1960, the

fallacy of the null hypothesis significance test, it's NHST or NHD in short. And today,

two years ago, there is still a paper entitled Moving to a World Beyond P Below 5%, so P is

still existing. But the debate has been going on for the last 60 years, I'll just give you

two other papers. In my opinion, the third one was a Stein and Lazar, the ASUS statement,

American Statistical Association. This article is most important, and I'm really happy that

they published this article because now I have a really authoritative reference to refer to

and not just single scientists who don't like p-values and other two.

Now, a short anecdote from my scientific life, when I did my PhD some 40 years ago,

five-way manoeuvre was very fancy amongst planetations and I guess psychologists and

people like that as well. That's what today are GLMM, General Linear Mixed Models, and phonetics,

or in your field, Roberta plus attention plus you name it. So the phoneticians at the lab where I

did my PhD hired a statistician to tell them about statistics and they were especially interested in

how to compute p-values. So what told us these statisticians 40 years ago? This.

And I'll come back to these figures and what they mean. I mean, you can imagine it's explorative,

descriptive statistics, box plots, and some assessment of your results along these lines.

Let's now start with a hypothetical example. Imagine that you are a young student and want

to write a small paper correlation between pitch and age for males. So you collect some data from

your family, you end up with eight subjects, different age and a different pitch height.