LESSWRONG
LW

Comment Permalink
amitpamin00

Unfortunately I have no idea. My statistics knowledge is many years unused.... I have some brushing up to do.

But it seems you're right - if each person gave 50 statements, the sample size would just be 250 (the # of liers x the number of statements they give).

Reply
See in context
by Benquo
1 min read
9

17

I'm considering taking Ekman's microexpressions training because it's cheap in both time and money. Has anyone here taken it? Did it work for you? How do you know?

 

The course does seem to come with tests included (both before and after), but if anyone has any ideas for some cheap tests I can do before and after to see if it really works, I'd be happy to do those as well, and report the results. Cheap tests should cost me less than three hours total and less than $100 total.

 

Alternately, if enough people here have done it we could pool our "before" and "after" scores to independently verify whether there's an effect.

Personal Blog

17

Ekman Training - Reviews and/or Testing
New Comment
9 comments, sorted by
top scoring
Click to highlight new comments since:
[-]Benquo40

One idea for a cheap plan:

Before the training, pay attention to people's faces for a few days. Write down how many times I learn something useful or interesting. After a while (if I observe any learning effect I wait until it levels off), I take the training, and do the same test afterwards.

Reply
[-]amitpamin40

I have Ekman's training on my computer, and I quickly abandoned it. The low number of recommended training hours and low number of different training samples to me seemed an indication of poor quality. In addition, the only scientific study of efficacy I could find at the time was sponsored by a microexpressions company. Some other evidence I read elsewhere seemed to suggest that the lie-detection bump from microexpressions training was tiny, in the order of a few percentage points.

That said, it would be interesting to me to actually test the training.

We have two groups, one that get's Ekman's training and one that doesn't. We also have a group of liers. Let's say 5 people. Each person in both groups talks to each person in the group of liers. Each person in the group of liers tells the other people a consistent set of truths and lies. The people in the test groups record to each line if they think it's a truth or lie. They don't find out if they are wrong/right. After the training, both groups go through the process again. The set of truths and lies told is the same.

If the training group increased their accuracy significantly more than the control group, it worked. Of course, this method requires too much effort and isn't likely to be that accurate. Hm...

Reply
[-]Vaniver10

Let's say 5 people.

What is the smallest effect size you would be able to reliably detect with a test that small?

Reply
[-]amitpamin00

Unfortunately I have no idea. My statistics knowledge is many years unused.... I have some brushing up to do.

But it seems you're right - if each person gave 50 statements, the sample size would just be 250 (the # of liers x the number of statements they give).

Reply
[-]Jon Chui10

Interestingly... just found a study from 2019 showing little to no correlation with Ekman's study a control group & a bogus training group, in detecting lies in videos they watched 

https://web.williams.edu/Psychology/Faculty/Kassin/files/Jordan%20et%20al%20(2019)%20-%20METT.pdf

I came to this after watching the pilot of lie to me again - gosh, it's a great show though 

Reply
[-]katydee10

I am extremely skeptical of Paul Ekman's training products and research on lie detection in general. He no longer publishes studies in peer-reviewed journals, as he claims that to do so might jeopardize national security. I assess a significant chance that the actual reason for this is so that he can make money from selling expensive consultancy/training services without having to face critiques from the mainstream scientific community.

Reply
[-]buybuydandavis10

Thanks for reminding me! I was meaning to give that a try. I think his prices have come down since I last saw them.

Another way to train yourself would be to review videos in slow motion after you've made a guess.

If it's lie detection you're after, maybe just a basic lie detection game with someone, where you both had videos you could rewind and review, would work.

I had the same feeling as amitpamin, that the corpus he'd supply was too small. But it might give you enough of an idea to move on to looking at videos in the wild yourself.

Reply
[-]zntneo00

"Emotions Revealed" has a test in the back and is recommends to read along with doing the training (I am currently reading it)

Reply
[-]buybuydandavis00

This place would seem to be useful as a tool to help recognize emotions.

http://www.artnatomia.net/uk/artnatomiaIng.html

http://www.artnatomia.net/uk/artnatomiaIng.html The tab is really interesting - animation of the action of the muscle.

https://developer.valvesoftware.com/wiki/Flex_animation

http://en.wikipedia.org/wiki/Facial_Action_Coding_System

Reply
Moderation Log
Curated and popular this week
9Comments