Communication Currents

To Catch a Liar

June 1, 2009
Communication Ethics

Research consistently finds that people are not very good at distinguishing honest from deceptive messages, but the studies behind these conclusions are often not very realistic. With funding from the National Science Foundationa lab headed by Dr. Tim Levine at Michigan State University has created almost 300 high-quality video tapes of truthful and deceptive interactions for use in deception research. We are finding that subtle differences in communication style can make a big difference in deception detection accuracy, and that some previous ideas about deception detection may be wrong. Our findings are leading to the creation of a new theory of deception that we call Truth-Bias Theory.

The ability to distinguish truths from lies has an incredibly wide range of applications from calling bluffs in poker tournaments to the effective interrogation of terrorists. Lie detection is relevant to journalists and juries, voters, and employers. If we could not presume truth and honesty, we probably could not communicate, and we certainly could not communicate very efficiently. Yet, the ability to communicate makes deception possible, and everyone gets duped at least once and awhile. Since the failure to recognize deception can be costly, it makes sense to study deception, deception prevention, and deception detection.

There have been more than 200 published studies that examine people's ability to detect deception. Lie detection experiments have people read or watch--sometimes live, but more often videotaped--a series of statements some of which are honest and some of which are outright lies. Deception detection accuracy is scored as the percentage of truths and lies people get correct. It is much like taking a true-false test.

On average, research finds that people are correct about 54% of the time. By comparison, blind guessing has a 50% chance of being right. This 54% accuracy finding is very consistent, and virtually all studies find results within plus or minus 10% of this value. It does not seem to matter who participates in the research. Police, for example, do not do any better than college students. Neither age, sex, professional experience, or intelligence affect results much. Other findings show that training people in nonverbal communication is of little help, and that people tend to believe others more often than they disbelieve, regardless of their actual honesty. We call this last finding truth-bias, and it is a very consistent finding.

The realistic study of deception detection requires a realistic set of truths and lies that people can judge. Ideally, the truths and lies used in research need to meet three conditions. First, the researchers need to know for certain what the truth really is, otherwise it cannot be known if people's judgments are really correct. Second, people need to decide for themselves to lie or not. A researcher telling someone to lie is less realistic. Finally, the truths and lies need to be consequential. It must matter for the person who is lying or honest whether or not they are believed. Outside the lab, being thought a liar is something people avoid. These criteria are difficult to satisfy, and most deception experiments come up short.

In our National Science Foundation experiments at Michigan State, research participants come into our lab for what they believe is a study of teamwork. Together with a partner, who they believe is another subject but who is really working for us, they play a 10-question trivia game for a cash prize of $5 per right answer. They need to work together with their partner to come up with the right answer. When they get an answer right, both partners get $5 each. We put a stack of $5 bills in front of each of them, and pull a $5 bill away each time a question is missed.

Most teams miss most of the questions. The questions are very difficult. For example, “Who discovered X-rays?” Few subjects know the answer, and the money gets pulled away bill by bill. Part way through the game, the experimenter gets a cell phone and has to leave briefly for an emergency. The answers are left in a folder in easy reach. The subject's partner suggests cheating. If the subject does not want to cheat, the partner backs off. After several minutes, the experimenter returns and the game resumes. Following the game, the two partners are interviewed separately, and the interviews are videotaped. The questions vary, but research participants are usually always asked if cheating occurred.

Because the partner was working for us, we know if cheating occurred or not. The research participants decide for themselves to cheat or not, and if they cheated, they decide to lie about it or to confess. About one-third of cheaters confess during the interview, but that depends on how the questions are asked. Finally, because the subjects were cheating in federally funded research at the university, they have reason to believe that they might be in big trouble if they are caught. It is also the case that the honest non-cheaters very much want their denials to be believed. So, our interviewers provide a very realistic set of truths and lies for use in deception research that meet the criteria for realistic deception.

In a second step of the research, we take the tapes of the interviews and show them to another set of research participants who have to decide if each of the people interviewed answered honestly about cheating or not. These tapes are also being analyzed by our partners at University of Arizona who are using automated linguistic and nonverbal systems to look for reliable differences between truths and lies.

Our research using the new deception tapes is not yet published, but we have done more than a dozen experiments to date, and have several interesting findings. Perhaps the most interesting findings relate to the importance of how the interview questions are asked. For example, consider these two different approaches. Question: “Did cheating occur when the researcher left the room?” Follow up: “Why should I believe you?” Or, as an alternative: Question: “Did cheating occur when the researcher left the room?” Follow up: “When I interview your partner, what are they going to say?” The first set of questions yields 50% accuracy or worse while the second set results in people being correct 70% of the time. It appears that “why should I believe you?” makes the honest people look guilty. The question catches them off guard and they often have a hard time answering. But, the opposite is true for the “What is your partner going to say?” question. The honest people know the answer to that question. The lying cheaters do not. These findings show the importance of interaction in deceptive communication.

In another experiment, research participants watched the new Fox show “Lie to Me.” The show involves a psychologist who solves crimes by reading nonverbal communication for signs of deception. Two control groups either did not watch the show, or watched a different crime drama. Then we showed subjects our tapes, and tested their ability to distinguish truths from lies. The subjects who just watched “Lie to Me” did least well. They did no better at catching lies, but they were wrong about the honest interviewees more often than the other groups. The reason, we think, is because the key to detecting lies is more about listening for verbal content than nonverbal demeanor, especially if the right questions are asked in the right way.

About the author (s)

Timothy R. Levine

Michigan State University

Professor