Everything wrong with statistics (and how to fix it)

A crisis has emerged across a number of research fields with the discovery that many published results are not reproducible, and applied statistics has been assigned a substantial share of the blame. But if you ask LLNL statistician Kristin Lennox, the problem does not lie with statistical methods, but rather from misleading training for non-statisticians.
Lennox, director of statistical consulting in the Computational Engineering Division, recently drew about 150 LLNL employees for a talk titled “Everything wrong with statistics (and how to fix it).” In it, she described the cause of the crisis and three golden rules researchers should apply ensure their data analysis is robust.
The slides are available on Slideshare: bit.ly/1PoqYJw.
For more information at statistics at LLNL, visit the data analytics website: data-analytics.llnl.gov/

[@Misuci]

Kristin Lennox, PhD Data Science Consultant, Managing Scientist at Exponent San Francisco Bay Area 500+ connections > Great presentation ! What a great sense of HUMOUR !

[@rammsteinmaiky1929]

Excellent talk, what she says at minute 29:35 is key for us, the non-statisticians who use statistics very often. Transformed data: "Reasonable normal in the neighborhood of the mean".

[@VaShthestampede2]

Great talk. Thanks for sharing!

[@Sharpodiplomat]

"Randomization is pretty much the only thing we have to protect us from unknown unknowns."

[@igordeocosta]

Amazing talk! I've learned a lot from it! Thank you!

[@lm58142]

27:28 "We know they're very tiny cause we've been using the same drop-hammer for years."

[@alpa1410]

Thank you so much, your lectures are very interesting

[@zaibatsu8582]

43:21 - Let me introduce you to 2016 and 2020...

[@auto_ego]

51:00 "starps working"? Is that shorthand for a brief stop followed immediately by a start?

[@noneya8100]

Starps indicate the beginning of the End. The Start of the Stoppage.

[@pieterverhoeven1642]

“Somethings that happens *a lot* to statistics consultants…” I was dissappointed there was no interjection from the audience.

[@valuesim]

Great!

[@namehidden8854]

I'd argue the problem lies both in training and in the methods themselves.

[@rajinfootonchuriquen]

The problem is always in the assumption. Now with metalog distribution, you don't need to asume anything, but it become difficult to "analize statistically" just a buch of coeffcient from a multilinear regression.

[@zoozolplexOne]

interesting !!!

[@EvaSlash]

6:00 Why did she say statistics is banned in social psychology?

[@EvaSlash]

+OpiatedBliss So by that do you mean they are not true statisticians and must abide by the scientific community? I always wondered how much mathematical/statistical training psychologists have in actually understanding the "under the hood" theoretical nature of the methods and techniques they use.

[@rajinfootonchuriquen]

Because the abuse of p-values

[@RichardAlsenz]

A wonderful observation. If you would like to improve the model let me know, I can show you how Mother Nature does it. And, it is has nothing to do with the same plagued physics used today.

[@rkpetry]

1. Why is the unicorn labeled with a lowercase sigma [σ]? Is sigma [σ] a 'myth'? 2. I asked a friend working toward her actuarials the following: "Suppose in a place there's a 30% chance of rain, and an east wind is blowing so there's a 50% chance of rain, Which is it, 30%, or 50%, (depends on what we agree we know)..." 3. Now, while that may seem 'tricky', let's try a real-world example that LLNL scientists know: In order for a special relative to get up to 10% the speed of light he accelerates one gee for 35 days... but... all during that time the far ends of his simultaneity train are running late so he calculates and sees and they-are creeping closer or further away: the difference being very-nearly proportional to the velocity, there's no such thing as an inertial equivalence principle, and, the zeroeth-plus-second order effect because creeping exponentially adjusts the lateness to the distance, is as big as Einstein's gamma... So, What, is it, that, we agree, we know... now...

[@jonasmaugust]

+Raymond K Petry The sigma is famous to non-statisticians as a statistical symbol, so by putting it on a unicorn, it reinforces the point she's making (at around 9:50) that statisticians are rare, like unicorns.

[@noneya8100]

You are beautiful, intelligent, well-spoken, and both interesting and entertaining. If you are straight, I want to marry you. ♥

[@nighthawkviper6791]

This is fun, I like cave-dwelling troglodytes; I hear they keep the keenest of numbers.

[@ahvigier]

Completely hollow talk