What actually works when it comes to detecting lies, and why are so many popular ideas about reading people and nonverbal behavior at odds with the scientific research? I talk with Aldert Vrij, one of the world’s leading deception-detection researchers, who has spent almost four decades studying these topics. We discuss why what people say is far more revealing than their body language; why the perceived “sexiness” of reading nonverbal behavior has contributed to widespread myths and misconceptions; whether vocal behaviors like pitch, volume, and hesitation contain meaningful clues; and whether there’s value in analyzing minor word choices and turns of phrase, as is done in so-called “statement analysis.” We discuss Paul Ekman’s legacy as a source of faulty ideas about deception, micro-expression-based deception detection training, and the strategies that popular “body language experts” use to appear scientifically credible while simultaneously promoting ideas that conflict with the science. We also discuss how focusing on body language can actually make someone a worse interviewer, why governments spend enormous sums on questionable automated lie-detection systems, and the value of artificial-intelligence-based approaches to detecting lies.
A transcript is available below.
Episode links:
- YouTube (coming soon)
- Apple Podcasts
- Spotify
Resources related to or mentioned in our talk:
- Aldert Vrij’s university page and Wikipedia
- 2023 TheConversation.com article by Vrij summarizing why verbal analysis is best for detecting lies
- Criticisms of language analysis lie detection systems:
TRANSCRIPT
(transcripts are partially generated automatically so will contain errors)
Aldert Vrij: Typically, lie tellers will avoid some issues. When they tell a story, very often it’s vague in terms of when they did something or vague in where they did something. If you just listen to the amount of details people give, it is far more diagnostic than any kind of behaviors you can think of. And now go to the internet and see how many people will say, “Just listen about the amount of information you will get.” Nobody will say that, because it doesn’t sound fanciful. But way more interesting, way more diagnostic than all those nonverbal cues.
Zach Elwood: That was a clip from my talk with the legendary deception detection researcher Aldert Vrij.
I’m Zach Elwood and this is the People Who Read People podcast. This is probably one of the top three most important episodes in the history podcast. I mean “important” in the sense that this podcast is about understanding behavior, and this is a talk with one of the most respected researchers into the topic of how nonverbal and verbal behavior can be used to detect deception. This talk contains a lot of important information on this topic in a pretty concise format; information that I think it would be good for many people to understand, especially because we are surrounded these days by so many myths and lies on the topic of nonverbal behavior, promoted to us by a slew of self-proclaimed “behavior experts.”
So Aldert and I get on a variety of interesting topics in this talk. We talk about what actually matters in detecting deception, which isn’t, as many people think, nonverbal behavior, but the verbal content — what people actually say. We talk about how the perceived “sexiness” of reading nonverbal behavior accounts for this faulty focus. I ask Aldert for his views on behaviors that are sometimes called paralanguage; verbal aspects apart from the content, things like volume and pitch and how long it takes people to respond. We talk about the practice of trying to find meaning in minor phrasing and word choices, as used in so-called “statement analysis,” – for example, someone avoiding the use of the pronoun “I.” We talk about the well known researcher Paul Ekman and his legacy as the source of many bad, faulty ideas about behavior and deception detection. We talk about micro-expression deception-detection training, as put out by Ekman’s company and David Matsumoto’s Humintell company. We talk about so-called “body language experts,” and the strategies they use to be perceived as intellectually legitimate while simultaneously spreading bad and wrong ideas that conflict with what the science says.
I’ll read a little bit about Aldert’s background from his page at the University of Portsmouth in England, where he is Professor of Social Psychology.
His main research interest is deception, resulting in more than 600 publications and more than 30,000 citations. He’s received grants totalling more than $11,500,000.
He’s worked closely with practitioners (police, security services and insurers) in terms of conducting research and disseminating its findings. He’s the author of the book Detecting Lies and Deceit: Pitfalls and Opportunities, which had a second edition in 2008; that book was an overview of research into (non)verbal and physiological deception and lie detection.
In 2016 he received the International Investigative Interviewing Research Group (iiiRG) Lifetime Achievement Award in recognition of his significant contribution to investigative interviewing.
After our talk, Aldert clued me into an academic book that had come out this year, which is titled Verbal Lie Detection, and which is edited by the researchers Pär-Anders Granhag and Galit Nahari. That book is dedicated to Aldert Vrij and his large contributions in this space. I’ve been reading that book, and I also have an upcoming interview of Galit Nahari about that book. I also have an upcoming talk with Mathew Gillings, who also researches verbal lie detection and is the author of the book “Corpus Linguistic Approaches to Deception Detection.”
If you want a really good introduction to some of the most important ideas Aldert has on these areas, I recommend reading a 2023 article of his titled “Lie detectors: body language tells us surprisingly little about whether someone is being honest” https://theconversation.com/lie-detectors-body-language-tells-us-surprisingly-little-about-whether-someone-is-being-honest-202405. That is just a great concise summary to why using nonverbal behavior for deception detection is useless, why analyzing speech content is so powerful, and what aspects of verbal content analysis matter the most. If you want a link to that, go to the entry for this episode at my site behavior-podcast.com. There is also a transcript there and links to resources we mention in this talk.
I’ll also mention that I’m currently working on a book about what actually matters when getting quick reads of people. It will be titled Read People For Real, with a possible subtitle of “Making Smart, Quick Behavioral Reads in High-Stakes Situations.” It will highlight the ideas Aldert and others in this space mention: the importance of focusing on the things people say. But it will also delve into my background in poker and my work on poker tells, and discuss the times when I think nonverbal behavior can be useful in high-stakes, adversarial domains. Some people are surprised that I often criticize the use of nonverbal behavior in this show, while also being someone who has written some well known books on poker tells. If you’re curious about that seeming contradiction, I have a frequently asked questions page on my poker tells site; you can read that at www.readingpokertells.com/faqs; that goes into detail about why I think making use of behavior in game scenarios is an entirely different domain from trying to do the same in real-world, non-game situations, for several reasons. If you want to stay in the loop about this book I’m working on, sign up on my site behavior-podcast.com, or on my poker tells site, either one.
Okay, here’s the talk with Aldert Vrij.
Aldert Vrij
Zach Elwood: Thanks for joining me, Aldert.
Aldert Vrij: You’re welcome.
Zach Elwood: Um, really appreciate you coming on. I’ve, I’ve been a big fan of your work for many years now. I’ve read many of your studies and papers, so, um, yeah, big honor to, to have you on. So maybe we could start with, um… You know, it might be a hard question to answer in a short format like this, but I was curious, you know, from your long career studying behavior and studying deception, uh, I was curious what stands out to you, top of mind, for the most important clues for, uh, determining if someone is being deceptive or, or hiding something.
What’s, what stands out to you in that area?
Aldert Vrij: Uh, the most important cues, I think is more general that, uh, how important speech is to detect deception rather than behavior. Uh, if you go, uh, on the internet, everything is about behavior. It is surprising how little there is about speech. But speech content tells you far more about deception than behavior.
In fact, uh, there’s not much in behavior at all. We can talk about it later if you want to, but it’s hardly anything in behavior, whereas speech tells you a lot about deception. But it is, it is, I think, less fanciful or something, so people don’t pick it up. But just by listening to what somebody says makes lie detection far easier.
Even already in normal, uh, interview settings, in normal conversation settings. If you only can see the person without hearing, uh, you’re guessing. It’s about 52%, uh, hit rate. If you only can, uh, hear the person without seeing, it becomes much more. It’s in the 60s.
Zach Elwood: Yeah. And, and to clarify, I think one of the things that makes this so hard to talk about is the, the word behavior covers so many things.
So, you know, and you’re, and you’re talking about, yeah, the fact that the nonverbal behavior, specifically the body language and facial expressions are so, uh, bad at, uh, helping people determine such things. And, and you, you had wr- you’ve written great pieces about this topic. You know, for example, you had pointed me to one 2023 piece where you were kind of summarizing your views about what matters in the, the speech content area.
But, uh, off the top of your head, what, what are some of the biggest clues in the, in the speech content area that stand out to you?
Aldert Vrij: Typically, lie tellers will, will avoid some issues. For them, the difficulty is that they should not incriminate themselves. So they should not say something that leads to clues for investigators to find out that they are lying.
Or maybe they say something that investigators already know it was a lie. That’s the difficulty lie tellers face. So when they tell a story, it is quite often it’s vaguer. Very often it’s vague in terms of when they did something or vague in where they did something, or vague in any kind of other, uh, details they, they can give.
Uh, whether somebody saw them, whether somebody got a receipt of something, that kind of stuff. It becomes vaguer. And, and that’s because there’s a difficulty they have. If they are more concrete about all kind of things, uh, it will give the lie away But the difficulty also makes, uh, lie tellers to give you less information than truth tellers do.
If you just do that, if you just listen to the amount of details people give, that’s a har- a rather diagnostic cue. It is far more diagnostic than any kind of behaviors you can think of. And now go to internet and see how many people will say, “Just listen, uh, about the amount of information you will get.”
Nobody will say that ’cause it doesn’t sound fanciful. But way more interesting, way more diagnostic than a- than all those nonverbal cues.
Zach Elwood: Yeah. There, like you said, there’s something very sexy about the nonverbal behavior and the body language thing, and there’s something that really draws people to that.
It, it’s somehow not as exciting to say examine what they say for logical and, and, uh, you know, various clues that, that are logical in nature. I don’t know why- Yeah … but there is something much less sexy and exciting about that to most people even though even common sense would tell you there’s so much more information in what people say.
But I’m curious, what do you think accounts, do, do you have any views on what accounts for this, the excitement of, that people have about the body language, reading body language things?
Aldert Vrij: Uh, be- because it sounds, it sounds exciting. It sounds, what they say is happening is that, uh, lie tellers, uh, try to, to, uh, control some behaviors and they’re not very good at doing that.
That’s the facial expressions. They try to control their expressions of fear or guilt, and they can’t do that very well, so they try to control themselves and they fail to do that. Or, uh, they show behaviors because they’re unaware that they show those behaviors. For example, fidgeting, li- like scratching your head, that kind of stuff, that people are not aware that they do that.
So they show cues that they cannot control or don’t know that they show, and you have to be very well-trained in order to see those cues. That’s the theory behind the whole thing. It’s called the leakage hypothesis. And maybe it sounds nice in, in theory, but in practice it doesn’t work. There are just no differences in that.
But the body language experts, that’s what they do. So when you go to the internet you will see, um, several cases in which they have a clip. It’s sometimes very short. It can be three or four or five seconds. And they’re telling you now in that five seconds are eight, nine, 10 cues of deception in those 10 seconds, in those five seconds.
And then they notice that the person moved a finger or the person indeed did this or something in those five seconds. What they don’t do is thinking and saying, “How does a person behave when he tells the truth?” ‘Cause they don’t have that. They only have the deceptive clip in it. And all kind of behaviors they then see, they say these are cues to deceit.
And if you do research on that, you will see that also truth tellers will show exactly those cues in similar situations.
Zach Elwood: Mm-hmm. Mm-hmm.
Aldert Vrij: The difficulty with what these, uh, these body language experts do, it’s pseudoscience in a way because you can’t … They, they cannot tell you beforehand what they’re looking at because all kind of leakage you will have.
That, that’s … You can’t, they can’t predict which leakage they’re looking for. So you cannot, uh, test what they are doing. It’s only lying. They have no truth teller control group. It’s, but it sounds absolutely fascinating. I agree. But you look at videotapes, I understand people think this is absolutely fascinating and how on earth did this body language expert see that behavior?
Yeah, because he played it in slow motion and so forth. That tells you nothing about deception whatsoever, and that’s the problem. And I started this work in, in late ’80s, and then everything was about [nonverbal] deception. So including what we did. Absolutely everything. Um, but then we found, we, we soon realized there’s not much in that behavior, that speech works far better.
And then late ’80s, beginning ’90s, I thought maybe we can change the world on that. And maybe we can make clear to all kind of people that indeed speech works better than behavior. But I’ve completely given up on that. Yeah. They become stronger and stronger.
Zach Elwood: Yeah. Uh, one thing I was curious to get your take on that’s kind of been confusing to me as someone, you know, pretty far from these areas is do you, I’m curious to get your thoughts on kind of like the paralanguage kind of things because it’s, it’s strange to me.
It seems strange to me that those kind of things like the, you know, pitch and hesitancy in talking and fillers are classified as like nonverbal when they have much more to do with the verbal to me. But, and I’ve also seen some people classify them as like a completely separate area from like nonverbal body language things and verbal content things.
But I’m curious, A, do you think those are entirely, should be seen as entirely different things than most body language and nonverbal things? And B, do you, I know you’ve written, you, you have papers about the paralanguage things. Do you see them as meaningful compared to the, you know, com- relatively compared to the content or would you prefer most people just focus entirely on the content?
Aldert Vrij: Um, I can see your confusion. But, um, verbal behavior refers to the speech content, so the meaning of what people say. There’s the paralinguistics, it’s not about the meaning. That’s called vocal and nonverbal behavior. So nonverbal behavior has visual, what you can see, vocal, what you can hear. But it’s not the speech content, so therefore the ums, and the uhs, and the stutters, and the pitch, and the loudness, that’s all vocal ’cause it doesn’t tell you anything about speech content.
Uh, do I think it’s, it’s diagnostic? No, it’s not. For example, the latency period, something else, is the time between question asked and answer given. That is something that’s also a, is a pause, so therefore that’s also vocal, uh, behavior. If you look at research, you can see that there is something in that, that often it takes liars longer to produce an answer.
But the difference is extremely small, and certainly you can’t do that in normal conversations. You need to do that in all kind of computer tests in which people have to react as fast as possible, and then you will see that they wait longer when they lie than tell the truth. They, they react slower when they lie than when tell the truth.
But the difference is extremely small, and the difference can only be measured in those kind of computer-type tasks and, and not where you just listen and speak to people.
Zach Elwood: Hmm.
Aldert Vrij: But that’s one of the most diagnostic nonverbal, uh, cues, one of the vocal cues, is the latency period.
Zach Elwood: Hmm. Mm-hmm. And I’ve also seen some studies about how the context can completely change things.
Like if they, if they’ve had time to prepare an answer, it might be completely, that’s a completely different scenario than like an off-the-cuff thing. But just to say, yeah, e- even if there are clues, it can be hugely context difference. And the, the important point to me is that you would focus on, you would recommend people f- just focusing on the content because that’s much more meaningful than these small, you know, trivial clues.
Aldert Vrij: Yes. But also that’s especially in interviewing and investigative interviewing, why do we interview somebody? Because you want to get information from that person. That’s why you interview someb- some people. And then, of course, is verbal lie detection far more relevant because you listen to the speech, and you try to get more information out of people.
And if you do nonverbal stuff, yeah, well, there’s not, not, there’s nothing about speech anymore in it. So what these nonverbal guys do is they, they focus very carefully on the behavior somebody shows. But what you can’t do in the interview, you cannot both focus on behavior and listen carefully to the speech and think about what to ask, what the next question will be.
So what will happen, those guys who focus on that behavior, the quality of the interview goes down ’cause they can’t just focus also on what people say and the next question. So you get also a lower quality interview.
Zach Elwood: Yeah.
Aldert Vrij: And you interview somebody because you’re trying to get information out of people.
Well, that is, that is verbal anyway. So verbal lie detection is far more in alignment with, with information gathering than the nonverbal stuff is. There’s so many arguments against that nonverbal stuff, and it continues.
Zach Elwood: Yeah. I mean, when I think about, ’cause I, I’m interested in these things, so I’ve watched a lot of interrogation footage and interviews thinking about these things.
And I mean, to, to your point, what stands out to me, e- even if I was not familiar with any of the research or anything, what stands out to me is the, the things that are important when, when in these interviews are when people are clearly not answering questions. You know, they’re, they’re clearly, like, not volunteering things, they’re, they’re changing the subject, they’re, they’re not being forthcoming with information, they’re being vague.
They’re, uh, to your point about verifiability, they’re not volunteering anything that can be verified and, uh, and, and then you also get clues that someone’s probably telling the truth because they’ll quickly come out with statements that could be verified. So even if you verify it or not, you, you have a sense of like, oh, they’re probably telling the truth ’cause they probably wouldn’t say those things if they couldn’t be verified.
So just to say- Yeah … ev- even, you know, examining these things at, at just a common sense level of like, when am I actually getting clu- valuable clues as a, as a layperson, it just stands out to me that those are where the most valuable clues are. Now, whether those are actionable now is another story because at the end of the day, investigators and interviewers are usually gonna ask the questions and actually dig into the, to the answers and such.
But in those, in the short term of like, “Oh, I wanna follow up with more questions,” tho- those are the kinds of clues that seem valuable from a practical level. It’s like they’re not answering the question, so I’m gonna ask more questions about that. Um, but yeah, yeah, I, I’m sure you, I’m sure we’re probably on the same page there, but if you have anything to say about that observation.
Aldert Vrij: Yeah, exactly, yeah. But by the way, it’s, it’s, uh, good interviewing is very hard to do. ‘Cause what you need to do in good interviewing is the person need to talk, and you just listen to what the person says. And based on what the person says, you ask follow-up questions, get more information and so forth.
And all kind of, um, view you have on what people say, whether it’s truthful or not, something like that, or plausible or not- Keep that with you and don’t say that too quickly, ’cause now you get an easy way out for the suspect almost.
Zach Elwood: Hmm.
Aldert Vrij: You know what I mean? But if you indeed, if you say, “Well, I don’t believe this,” yeah, well, that’s a nice excuse for somebody to say, “Well, if you don’t believe me, I stop talking.”
Because why, I just tell the truth, so why would I say more if you don’t believe me anyways? That’s-
Zach Elwood: I’ve seen these bad interviews and in my opinion where, yeah, there’s all these ways that they basically shut down the conversation or, or make them feel confronted and then they just stop talking and clam up and yeah.
Aldert Vrij: Yeah. But that’s of course the interview styles that were developed in, in the 1950s, 1960s, and they’re still rather dominant and rather prominent, uh, but to a lesser extent now. Uh, that kind of stuff was developed in, in the, in the US, uh, it was highly dominant there. It was abandoned quite quickly or not quite, yeah, but in the late ’80s in the UK, starting there, uh, because those interview styles resulted in false confessions in IRA terrorist, uh, uh, cases.
Some kind of IRA bombings going on in the UK, and some people were arrested, have confessed to those bombings although they were completely innocent. And then then the system realized this is just leading to false confessions if you do this kind of stuff, so we need to stop doing it.
And that’s triggered a, a completely different way of interviewing that has been developed in the UK where this, this confession-based coercive type of interviewing is not happening at all. It’s not allowed anymore. So now it’s far more a conversation you have, and you don’t put any pressure on people and so forth.
But the difficulty is that people think that will not work. People think in order to get somebody to talk you need to put a lot of pressure on the person, otherwise they will not talk. But research shows that’s not the case. And of course, UK was a real-life experiment, ’cause we had that kind of coercive style stuff until 1990.
No coercive stuff after that, and the confessions didn’t drop at all and all kind of things. It, it didn’t, it didn’t change. It, it didn’t get less. In fact, it only get better, ’cause we know now if you do that non-coer- coercive interviewing, it leads to more information, it leads to more, uh, true confessions, it leads to fewer false confession.
It is all kind of, uh, benefits for that. And now even, even, also in the US and Canada people start to realize that now, so that coercive interview style is getting out of fashion. But it’s still very fashionable, for example, in Asia. And I guess and also in South America, I know that. And I guess, I don’t know, but I think also in Africa and but I don’t know there.
Zach Elwood: Just a note here. My last guest for this podcast was Bob Pointer, who is a former London cop and he talked about the IRA bombing era and how the false confessions and convictions in that era led to an overhaul of British policing and interrogation approaches. Okay back to the talk.
Looking back at your career, are there certain findings and ideas that you’ve contributed in these areas that you’re most proud of?
Aldert Vrij: Most proud of? Mm. Well, I think we started by, uh, saying that nonverbal lie detection is not very useful. Uh, as I say, it still continues, but it’s out of fashion for, uh, scientists, academics.
Um, recently was a survey and hardly anyone says that that type of lie detection is useful. It’s because it’s not. Uh, I think we also contributed to, uh, uh, the idea that successful lie detection is not looking at how nervous people are. Completely forget about anxiety. Anxiety doesn’t work. Look at something else, and we do.
We look at cognitions. The thinking of truth tellers and lie tellers is completely different. That arousal-based stuff is completely out of fashion, and I think we contributed to that. And even interestingly also in the polygraph. In the polygraph, the lie detection, uh, machine that was also based on the idea that, uh, lie tellers are more nervous and aroused and therefore score higher on certain questions.
And supporters of that type of lie detection still like the polygraph lie detection, but do not anymore say that arousal is the reason why it works. Now they say we don’t know why it works, but even in that area it’s completely out of fashion. And that was very strong, that arousal is the reason why lie tellers and truth tellers respond differently.
Even there it’s out of fashion. So that’s, that’s certainly a contribution we had.
Zach Elwood: Yeah, you’ve played a big role in, in all of those areas, I would say from my understanding you’ve, you’ve played a major role in shifting the views on those things, right?
Aldert Vrij: No, I mean, you don’t do that on your own. You never do these kind of things.
You need other people to support you in that.
Zach Elwood: Right.
Aldert Vrij: And I think the—
Zach Elwood: But you, you’ve been a key player is, is what I’m saying, yeah.
Aldert Vrij: And because I’m older than some others that also makes a difference. I started earlier. Um, and the third contribution I think is so-called interviewing to detect deception.
To make lie detection easier you need to have interview protocols, and that’s now current lie detection is developing interview protocols that will make a difference of the lie tellers and truth tellers more pronounced. And we do research in that, others do research, but we pushed that agenda into that interviewing.
Zach Elwood: Like the, uh, strategic use of evidence, you’ve studied that and so on.
Aldert Vrij: That’s more my colleagues from, from Sweden. But that, that is an example of indeed an interview style that will facilitate lie detection.
Zach Elwood: Mm-hmm. I was curious, uh, I was curious to get your take on what’s called statement analysis, which is kind of like a separate thing.
Usually, uh, considered a separate thing from analyzing the content. It’s often- Mm … associated with these like minor turns of phrase. And it seems like from my rough understanding, it seems like the, the word itself gets a lot of hatred, unnecessary hatred because at the end of the day even content, analyzing content is about analyzing statements, so in some sense that’s statement analysis.
But I understand that it’s also linked to some of these things like SCAN which have been debunked, you know, the SCAN system. Uh, although it also seems like some of the things that SCAN w- would list have been studied in research and seem kind of okay. So it, it seems hard to, uh, cast aside the whole idea that there can be clues in small turns of phrase and small phrasings.
But I’m curious to get your, get your take on that overall, uh, and how do you view that as separate from some of the more meta level content, uh, clues, if that makes sense.
Aldert Vrij: Yeah, the, uh, statement analysis and also SCAN, SCAN, uh, scientific content analysis, that’s the most dominant one in, in that, that type of lie detection.
Uh, it doesn’t work. There’s absolutely no evidence whatsoever that SCAN or these kinds of methods work. It is highly popular. It’s highly popular amongst practitioners. It is developed by, by practitioners. Uh, other, uh, uh, protocols are developed by scientists. This is not developed by psychologists, but by practitioners, and it is powerful because it’s folk psychology.
There is no psychology behind the whole thing at all, but it sounds all kind of very plausible, and therefore it is so appealing to people. If you ask all kind of people some kind of kind of questions, “What do you think would be the difference between lying and truth, and truth telling?” You get very often these kind of things spontaneously people will say that.
That’s called folk psychology. For example, what you have in SCAN is a popular idea there is that in a statement, a short part is the introduction, a main part is about the event itself, and then a shorter part is about the end, uh, the exit of the statement, so to speak. There is no evidence for that. Or something else that’s, uh, even more bizarre maybe, that objective and subjective time should be reported in the same amount of detail.
So if people talk about a day, a day between 9:00 and 5:00, they should give the same amount of information about each hour. That is completely absurd ’cause how much you can say about something depends on what you did. If you did a lot of things in the morning, you can say a lot about that. If you had a nap in the afternoon, you can’t say anything about that.
So the idea that it will work like that is just, it’s based on nothing, and that SCAN has absolutely, has zero support in anything. People have done research and then other people, nobody finds anything. And that is done on statements in which other verbal tools indeed work. So we know the statements work.
SCAN is guessing; other tools work. Yeah, the SCAN is developed by, uh, practitioners and also very, it, it’s, uh, you can do a lot of training in that SCAN. And the stuff that, uh, scientists developed, uh, yet the scientists don’t do training. So the training that, that works, there’s far less training in that.
And that, of course, is a problem. If you really try, if you want practitioners to change their practice, you should offer them the training they need to do in order to improve. And that is largely lacking because it’s all academic work, and the scientists behind this are scientists and don’t, and not the trainers.
And that they give some training, but not enough. It’s not a worldwide, uh, training scale as the people who do SCAN and so forth do. And then that’s, that’s, that’s a difficulty.
Zach Elwood: I was curious, uh, it’s hard for me to get a handle on this, but some of the things in the statement analysis area, and I’m not, I’m not even sure if they’re associated with SCAN because I, I can’t remember offhand if this is covered in SCAN.
But I’ve seen some people that say that they’re, like, forensic linguists. Like for example, Kirsty King has a book about using, you know, analyzing language for, for lies. And I think her and other people have said some of these small clues have been studied. For example, like the u- you know, that lacking the use of the “I” pronoun- Mm-hmm
in a form of, like, distancing. Are there some- Yeah … of those little clues that are valid in your view? And, uh, do you have any takes on, you know, people or resources that can help us distinguish what are, that have actually been studied or seem statistically relevant versus not?
Aldert Vrij: Uh, well, the use of “I,” the self-references, in- indeed is a very popular cue.
A lot of research in that. Uh, the idea is truth tellers say “I” more than lie tellers do. They do. That’s true. But the difference is so extremely small, you can’t use that in real life. And so what you, what you need to do, not only should there be a difference, it’s the difference should be large enough to rely upon.
And that is not the case with “I”. And I can’t, well, won’t give any statistics in here, but the effect size, effect size tells you about the strength, about the size of the, of the difference is far too small for self-references to be useful in real life. If you just use that as an indicator, you get loads of answers wrong.
It’s just, it’s just above chance you get. You get about 55% of the statements correctly classified if you just listen to the self-references.
Zach Elwood: A note here. If you want to read more about this SCAN system we’re talking about here, a great article for understanding how faulty that system is, can be found in a propublic article from 2019 titled Why Are Cops Around the World Using This Outlandish Mind-Reading Tool? When you read that article, I think you’ll come away with a view of just how unreasonable and irresponsible SCAN’s creator, Avinoam Sapir, really is, and why this is such a big problem.
Aldert and I didn’t talk about it, but a related area of bullshit is overly confident 911 call analysis. A great read about that is another one from Propublica titled “How Jessica Logan’s Call for Help Became Evidence Against Her”. That article goes into detail about a 911 call analysis system’s creator Tracy Harpster, who, like the creator of SCAN, I think is irresponsible and unethical, in terms of making grand claims about what his system can find in people’s small verbal turns of phrase and word choices.
Just as there are people who promote outlandish and exaggerated and debunked uses of reading nonverbal behavior, there are people doing that for so-called “statement analysis” and language analysis. I think the most irresponsible uses come not necessarily from the ideas, as I think there can be a range of debatable ideas in the mix in these various discussions — but from the highly certain and confident framings that some people promote about these various ideas and systems. In other words, SCAN is a compilation of many ideas, and some of those ideas are defensible and some have even been studied and have some validity. Many of its ideas are bad and have been debunked. But regardless, the problem isn’t necessarily from the ideas; the problem and dangers really arise when people irresponsibly claim that these various ideas have a much larger effect size and predictive power than they do. That is where the danger comes in; people taking small effect size information and then saying, based on such things, “I am certain this person is lying.” Put another way: if people were ethical about the truth of these various ideas; that these things are highly ambiguous sources of information and, even when there is a signal, that it is far from 100% reliable, there would be little harm done, even from the worst ideas in the mix.
I’ll also throw in that Mark McClish’s book on statement analysis, I Know You Are Lying, was the inspiration for my book Verbal Poker Tells, which is the poker tells work I’m most proud of. One of my first interviews for this podcast was a talk with Mark on his ideas. Now, I’ll say that, looking back on things with a more informed perspective now, I still do like Mark McClish’s books. I think, as with a lot of these types of things, that it contains a mix of good observations and bad observations. I’ll say that what bugs me now about Mark is what bugs me about a lot of people who sell these types of trainings and systems: they speak with way too much confidence about what are highly unreliable clues. I’ve heard Mark take a single minor statement choice and frame that as basically, “that tells us he did it,” and I think that’s where the irresponsible framings come in. Anyway, that is what stands out to me in these areas, and I thought it was worth sharing — because I think it’s okay to have a broad range of ideas, and to form your own ideas about what works in various domains – the problems come in with the high confidence and certainty about things that really shouldn’t give you high confidence and certainty.
Okay back to the talk.
Zach Elwood: Yeah, no, that’s a great segue into something I wanted to ask you about because it, it does seem to me a lot of these debates in the behavior analysis use in security or policing environments, it does seem to me that a lot of these debates are really about, at the heart of them, about what you can do with really low effect size clues basically.
It’s like, what good is it to know something that gives you, like, a 55% instead of a chance clue, right? Like, so I think, you know, debates about the polygraph are largely about this because y- we can imagine even if the polygraph was, like, 95% reliable, you would still have the question of, like, is this actually useful in real world?
You know, you can’t base major life-changing decisions on something that even has a 5%, uh, margin of error, right? So it seems like- Mm-hmm … a lot of these debates are really about what can you do with imperfect information that has meaning but that is really small effect size. But I’m curious to get your take on that.
Do you see a lot of the debates coming down to that question?
Aldert Vrij: Indeed, it needs to be highly accurate for evidence in criminal courts, huh? Typically, it’s about 95%. That is not happening in, uh, in lie detection, and that’s the reason why lie detection methods are, are not used as evidence in criminal courts because it’s not accurate enough.
In other situations it can work. It can work if you have … You need to decide whether to continue investing in a certain, uh, person, whether you think, okay, it’s useful, continue with this. Uh, he denies, uh, the crime. We think he’s done it. Uh, listen to what he says. Let’s continue with this because we think there’s something more in it than just his denial.
Uh, source handlers and also kind of intelligence settings as well, same. Indeed somebody says something, okay, could it be true? Could it be indeed what the person says is going to happen in the future or not? And that is useful to have some kind of ideas or thoughts about that.
Zach Elwood: Especially when you lack better information, right?
In the, in the scenarios where you really are kind of blind and don’t know which way to go in
Aldert Vrij: Yes. But for example, what you will see what’s highly popular is this kind of lie detection at, uh, border crossings. Uh, the EU, uh, has spent a lot of money, in the billions, in the billions of, uh, of euros on that, on developing mostly nonverbal of course, nonverbal lie detection systems.
So with avatars, they ask now questions, and you answer that as a, as a passenger and whatever the re- responses you give, uh, you’re classified as a truth teller or a lie teller. That’s, that’s the idea behind that. The US also does that, has done research on that. In the billions of dollars goes in that direction.
But what can you achieve with that? It is the hit rate you get is now is about 70%, so 7 out of 10 the most you will get, but needs to be extremely accurate to be meaningful. There’s so many people at airports walking around. Even it’s 90% accurate, if every thousand passengers, 100 of them are now seen as lie tellers and need to be further in, in investigated, further interviewed, it’s not happening.
There’s just, there are no resources for that. It needs to be way and way and way more accurate than that to make any kind of meaningful decisions. It’s never going to happen. So why on earth we spent government pay- spent billions on it, I have absolutely no idea. I have absolutely no idea why. It is, I think, the fascinating that if you, you, um, of course the nonverbal stuff to come back to that again, what we did in the past is in so much, uh, detail.
We coded every single behavior we could see. We did it manually, and now engineers can do that via the computer in real time. That’s interesting. But of course, they’re coding exactly the same behaviors as we did manually, so you get the same results. And I think the idea there is if you simultaneously record all kinds of behaviors people show, you must be able to reveal a lie. [Wording uncertain.] And then the question is, what’s the theory behind this?
Why would a lie teller now indeed show a different behavior than truth tellers? And to, to be used in this machine, you need to have that all these passengers when they’re lying should have a certain pattern of behaviors. And why is that the case? There’s no theory behind that. And when you ask these guys, these engineers who developing the machines, they get very annoyed because they can’t answer it.
And there’s no answer to this. And okay, that they would try to make the machines fine, but that governments spend billions on it, that is absolutely is complete, it’s, well, bizarre to me that they do that. I don’t understand why. The EU does that. U- uh, US does it, and it will continue. There’s no doubt about that.
Mon- much money goes into those systems. It’s absolutely fruitless.
Zach Elwood: Yeah. It seems like there’s a big overlap between, we had talked about the perceived sexiness and exciting nature of reading people in the public, you know, in, in lay people’s, uh, views. There’s also this, I think there’s a similar demand for easy wins and easy solutions in like security environments.
So it’s kind of like across the board, everybody has this big demand for like, “Oh, we can, we can read people and get these shortcuts to reading people. That’s exciting, and also might save us a lot of trouble.” So it makes sense that there’s this huge demand in the public audience and, and security demand- demands across the board for these things, it seems like.
Aldert Vrij: Yeah. But you then also have the so-called black box problem you will have, huh? One of those guys, uh, a few years ago showed me, uh, the tapes of, of truth tellers and lie tellers, and had about 75% accuracy, something. He showed me a tape and I asked him, “Is this person we’re now looking at lying or telling the truth?”
And he couldn’t answer that question. So then I asked him the question, “Okay, but if I try to find that out now, which behaviors should I pay attention to?” He couldn’t answer it either. The machine could tell he is lying, he’s telling the truth, but he couldn’t tell based on what. How can you use that in real life?
You’re at an airport and y- you now, you, you try to fly, and now the officer says, “We are… You’re lying.” Why? The machine tells you so. It’s absolutely completely bizarre. It’s not working. Of course, nobody, nobody can accept that. The passengers will get very agitated in that. And if you go further, maybe in court cases, why?
How? You even can’t tell why the person is now, uh, now, uh, seen as a liar to everyone, ’cause they don’t know themselves which behaviors triggered it. It’s just, it’s absolutely useless.
Zach Elwood: Right. You need observability, like, for these machine learning kind of algorithms. It’s the same way in a lot of applications where it’s like, you know, i- if a car makes a decision, even it…
You can’t have it be a black box. Like, a self-driving car makes a decision, you, you can’t have… The manufacturer can’t be like, “Well, we don’t know why it, you know, decided to run into these people. It’s a black box.” You know, that, that only goes so far. But, uh, yeah, I, I was curious, uh, understand if you don’t wanna talk about, but would you be up for talking about Paul Ekman and your view of him?
Because it does seem to me, in digging into these areas over the last few years, it does seem to me like he’s the source of so many faulty myths and misinformation about nonverbal behavior. But I’m curious if you’d like to give your, your takes on him and what he’s done in the, in the nonverbal space.
Aldert Vrij: Uh, certainly did.
He was the man behind the leakage hypothesis. So the idea that the lie tellers, uh, show behaviors they can’t control, that they are unaware of some behaviors they show, and also kind of, uh, uh, behavior and speech can be out of sync. That’s another popular thing. So people can say, uh, yes, but they shake their head.
And shaking head means no, saying yes, so there’s out of sync. That these are three things according to the leakage hypothesis you need to pay attention to. Uh, according to Paul Ekman, he did [little] research on that. It’s hardly anything. It became really popular, uh, in his book Telling Lies that he wrote in 1985.
That is a, is a well-written book. Uh, he’s a good writer. He’s an excellent writer. He started the book saying, okay, uh, there was no empirical evidence yet for the things he was going to talk about, but he realized so much nonsense, uh, in the literature about nonverbal behavior and deception and so forth.
He thought, “I need to do something else. I need to, to pre- uh, present my views now in order to give a, a different view on this kind of stuff as what happened.” Uh, 1985, okay, but he hasn’t done any research in that. So in all those decades after that, and he has several times updated or new versions of the book, but not with research.
Not at all. So there’s just no research on that area. There’s no research on these microexpressions, no research on all the kind of things. And if there’s research, it doesn’t show what he says that will happen. So there, there’s lack of empirical evidence in that. But it sounds so intriguing. It sounds so interesting.
So you have the TV program Lie to Me, that’s based on that kind of stuff. So people like watching that on TV. It sounds so intriguing. And that, that’s the benefit, but there’s no evidence to support it. He’s never done it. He’s never published in a journal [what he’d need to publish to back up his own views].
He hasn’t done it
Zach Elwood: Yeah. And it seems like from talking to behavior researchers and for this podcast and also just off the podcast because I, I’ve gotten the sense that a lot of researchers don’t like to- they’d rather focus on, you know, criticizing the ideas and not focus on criticizing the person. But it does seem to me like from what I’ve gathered is that many people view him as really becoming less and less ethical and in the, in the research space and doubling down on his ideas despite what the research says and really, you know, leaning into like saying that his, his ideas were still valid and, and could detect deception.
You know, for example, in the, I think it was early 2010s, uh, congressional testimony, he was still saying like, “My- our, our methods can detect deception at like 80 to 90% accuracy,” which to me just seems obviously false at that time. So it does seem to me like… A- and then he would also say, “I can’t share my findings because the security implications are so serious.”
Yeah. “So that’s why I don’t put out things publicly.” But just to say there’s a number of things which make people view him as becoming less and less ethical and, uh, [less] willing to acknowledge what the evidence shows over time. That’s not to say his earlier work wasn’t good, but just to say that’s what I’ve gathered from talking to various people.
But I’m curious if you agree with that or care to share your thoughts on that.
Aldert Vrij: Uh, I do certainly think that the idea that his, uh, method leads to so high hit rates, accuracy rates, there’s no evidence for that. In part what he does are two different, different things. He gives a training in these microexpressions of emotions.
He gives a training how to detect those, uh, those emotions. And then in the post-test after training, he shows videotapes, of course his own videotapes, and now he asks people to detect those emotions they’ve just been taught. And you get about 90% in that. So that means that people who are taught those cues can detect those cues in his own training, uh, material.
But it doesn’t mean you detect lies with 90%, ’cause it doesn’t mean that those cues occur only in lie tellers and so forth. It doesn’t mean. It just means that what he teaches people he can do rather accurately. People, um, will, will be pretty good at these, his own tests after being trained. That’s exactly- Yeah.
Zach Elwood: I’ve taught you how to take my tests. Yeah, basically.
Aldert Vrij: Yeah. That’s what it is. And indeed, th- what he’s saying in Parliament there is about the, the SPOT program. Now the SPOT, these are these behavior detection officers walking around in airports and then looking for all kind of behaviors to see whether people are, uh, wrongdoers or not.
He indeed has some claims about 80, 90% accuracy, and that there is absolutely no evidence for that. The- some studies have been done at these airports, and they’re not very well conducted ’cause what they’re looking at, the, these, these BDOs are, they’re called, the behavior detection officers, they will, uh, select a person that they will interview and then, then whatever they do with the person to see whether the person indeed is smuggling something, for example.
And they have a control group, but the control group is completely random, whereas their, their group is not really random. They’re not a, uh, I don’t know, a very old pensioner. Even if that old pensioner can show the behaviors they’re interested in. They will not select that old pensioner. They more often select, uh, people, younger guys traveling on their own, traveling light and so forth.
So what you need to do for their stuff is you need to have the same type of passenger. And then those who show the cues and those who do not show the cues so that you have now the same people you compare the two on. And they haven’t done that. So you have not only a difference in whether people show those cues, it’s also a different type of person.
So it’s not good research. And it didn’t lead to 80, 90% accuracy. No way.
Zach Elwood: Yeah. Um, I wanted to get your take on something that I’ve been thinking about the last couple months. It does seem to me there’s some, you know, so-called body language experts that will pay lip service to the research. Like they’ll point out that the research shows that nonverbal behavior is not useful for detecting deception.
But I think they also, as the saying goes, kind of want to have their cake and eat it too because they’ll imply that they can still give you things that are basically deception detection skills. So they promote myths that people’s facial expressions and body language w- will give them away, and that there’s a tremendous amount of information even in high stakes adversarial situations, that there’s a tremendous amount of information that you can gain and that you can basically kind of see through people.
And then they’ll even use anecdotes that are deception detection related whilst w- while at the same time giving caveats about what the research shows about deception detection. Because I think in order to be savvy, you have to reference the research, right? You want to be perceived as being with it. But I think they’re basically trying to play both sides of the fence because they want to sell trainings and materials that make people think that you can do these amazing things with body language and that it’s really important and you can see through people while also, you know, giving the caveats, appropriate caveats.
But I’m curious, do, have you seen that too? Because that’s what really stands out to me is this kind of like straddling of the fence by body language experts.
Aldert Vrij: That is very interesting. And I think they do two things. First, they s- then talk about behaviors they’re not looking at. So to say behavior, uh, deception, uh, detection will not work if you look at A, B, C, and they will look at something else.
Or they will say, uh, it, it doesn’t work in all kind of d- situations, but it will work in our situation and what we are looking at. So that’s what they do. So they will say, “In general, it doesn’t work. Most cues don’t work. Most situations doesn’t work. But now I give you the examples where those cues will work in those specific situations.”
That’s what they do. And indeed, it’s never based on any kind of research in that area. Uh, they of course they don’t do research themselves. There’s a lot of research in many kind of situations for nonverbals, and it doesn’t show you anything, and they don’t do that. Or what they sometimes do is, um, cherry-picking, is, uh, okay, one study showed something.
But that can always happen. So many studies. If there so many studies done- Yeah … there’s always one study that will show something, and they will then highlight that study, say, “That study shows that.” But you need to do is look at the meta-analysis, at the overall pattern of research findings, and they will never refer to that ’cause that shows that indeed what they say that, that doesn’t work.
And for example, the idea that they very often use is low and high stakes. Low stakes is it, the consequences are very low. High stakes means it matters. It matters whether you’re, uh, caught or not. So at low stakes is if you lie in an, in an experiment at university because who cares when you be found out?
You just walk out anyway. High stakes is that, uh, you may be convicted if your, if your lie is caught. That’s high stakes. The idea is that research is low stakes, and they look at high stakes. But even if you look at that, there is no theoretical reason why high stakes would make a bigger difference in lying and truth telling ’cause the stakes are high also for the truth teller.
Also, the stakes are high for truth teller as well if indeed if that, if the police or something don’t believe their story and they can’t, uh, uh, um, demonstrate their alibi, they can’t demonstrate their innocence, then it becomes very high stake for them as well. So the idea that then a lie teller will show all kind of cues and the truth teller doesn’t, there’s no theory behind that.
And we also know that from research. We know from research if you compare low stakes and high stakes lies, the difference between lie tellers and truth tellers stays the same. And of course, they will never refer to that
Zach Elwood: Yeah, and I, and I should mention, like when I use the word high stakes there, I was actually thinking…
‘Cause I think there’s something interesting in the two different groups of like our everyday social situations that most of us know. There’s something, uh, very different about those environments than there are in any sort of environment where you, there’s an incentive to deceive or you can’t trust the other person.
So to me, it’s like it, it, that helps explain why some of these body language experts can escape criticism too, because I think in the everyday social situations, we are often reading other people, and we do rely on people’s body language and facial expressions because those are situations where we are trying to communicate with each other.
Mm-hmm. So it’s a very… And there’s not much incentive to hide things from each other. So just to say in those situations, it does make sense that we are often reading each other, but that’s a completely different area from like high stakes or adversarial, if you will, situations where we know that we can’t trust these things in any sort of like deception detection way or guilt detection way.
So I think that’s important because I think a lot of people, when I’ve seen people defend body language experts, they’ll point to all these everyday social situations where these things do apply. You know, like- Mm-hmm … we often get a sense that our, someone’s irritated, and we adjust our behavior based on that in a everyday social situation.
But that’s a, that’s a very different thing fr- from trying to use those things to determine guilt or determine deception. But I think in many people’s minds, all those things kind of get conflated, so they’re like, “Oh, we know every day, in our everyday lives we use these things, so therefore maybe we can use them in these high stakes environments too.”
But that’s the way… I, I don’t know if you have any thoughts on those two d- different domains.
Aldert Vrij: Well, we know that even those social situations, uh, lie detection ability-
Zach Elwood: We, we often get it wrong there too. Yeah, yeah …
Aldert Vrij: Yeah, I think there we go very often wrong. Yeah, yeah.
Zach Elwood: Yeah. So true that. And, and I should say, I should say I’m not even talking about deception detection.
I’m just talking about like every day, like you notice that your coworker is a little bit irritated at something, and you’re like- Yeah … “Oh, maybe I should ig- you know, ask if they’re okay, or maybe I should, uh, you know, leave them alone.” I’m not even talking about deception detection there. I, I’m just talking about like interpersonal stuff.
Aldert Vrij: Nonverbal behavior tells you indeed whether people are nervous, irritated. It tells you that. But it doesn’t tell you anything about deception. There’s no reason why only the lie teller will be nervous and, and not the truth teller. There’s no reason for that. There’s no theoretical explanation for that.
Zach Elwood: Yeah. Yeah. And I, I think it helps explain why, you know, it also plays into how body language experts sometimes try to frame these things and they’ll be like, “Well, I’m not telling you how to detect deception. I’m telling you how to look for signs of anxiety that could be tied to deception.” So they’re kind of like lengthening- Yeah
lengthening the distance as a way to make it, you know, they’re still trying to sell the deception detection idea, but it’s, it’s faulty because as you point out, I mean, that’s what the studies are trying to find. They’re trying to find indicators- Mm-hmm … nonverbal indicators of deception. The, the studies are trying to find those what, what you would call hotspots, and they’re not finding anything.
So the, the thing, this narrative that the body language experts are telling you that they’re gonna clue you into hotspots or clues that they could be deceptive is, is like a way to protect themselves because they’re length- lengthening the distance from deception detection. But they’re still trying to communicate what will help you detect deception.
At least that’s how it seems to me.
Aldert Vrij: And of course they use, uh, anecdotes, huh? So they carefully select the clips they will show you. But as I said earlier, you don’t have a, a comparable, uh, truth-telling clip of the same person, same situation. Do that. Get the same person in a kind of interview setting and the person now is telling the truth.
See how he behaves then in these situations. That’s far more interesting, and they never do that. It is, it is pseudoscience. It is, yeah, well, it’s, it’s, it’s poor. Yeah. And of course the problem with pseudoscience is the people who really want to know something about a topic, they’re hampered in their in- information gain ’cause they’re listening now to the body language experts.
I think it’s absolutely fascinating. So they go, they go that way and get now im- and get impressed by it, and they get convinced that what these guys are saying is, is truthful, and it is not