The Development and Testing of an Implicit Lie Detection System



A series of five experiments were conducted to explore whether Greenwald, McGhee, & Schwartz’s (1998) Implicit Association Test (IAT), which purportedly measures implicit affective evaluations, could be modified to differentiate between honest and deceptive responding to forced-choice questioning. Experiments 1 and 2 demonstrated that a dual-discrimination task can in fact be useful in deception detection but that the relative reaction time differences run opposite in direction from those expected from the typical IAT bias pattern. Subsequent experiments assessed the procedure’s susceptibility to simple countermeasures (Experiment 4) and tested variations to its trial sequence (Experiment 3) and stimulus presentation (Experiment 5). Neither of the two procedure variants was successful in producing above-chance predictions and instructions to delay reactions times to a constant latency sufficiently undermined the original procedure’s efficacy. The applied limitations notwithstanding, the present research extends the relevance of dual-discrimination methodologies and supports the idea that biographical information is cognitively represented such that what is known to be true or false is implicitly associated with one’s general concepts of “truth” and “lie” respectively.

Deception, by definition, is the “deliberate attempt… to create in another a belief which the communicator considers to be untrue” (Vrij, 2000, p.6). It should not be surprising that psychologists have taken considerable interest in deception, especially given the potentially negative interpersonal consequences of being deceived. Early research attempted to identify the individual characteristics of children that lie, and instead found that the decision to deceive is determined more by the social situation than personality (Hartshorne & May, 1928). Nonetheless there have been many appear to have little predictive validity (Goldberg, Grenier, Guion, Sechrest & Wing, 1991). Deception, especially in the context of psychological research, can often take the form of socially desirable responses (Rosenberg, 1965) and self-serving inferences (Johnson, Feigenbaum, & Weisbeg, 1964). As such, psychologists have developed strategies for circumventing deception, including the use of behavioral and “behavioroid” (Aronson & Carlsmith, 1969) measures, projective (e.g., Morgan & Murray’s Thematic Apperception Test) and implicit measures (e.g., Greenwald et al., 1998), and by establishing the pretense that deception will be detected and is therefore futile (Jones & Sigall, 1971). The majority of psychological research has focused on detecting deception, giving a great deal of attention to non-verbal behavior.

Methodological, Theoretical, & Applied Contributions

The contribution of this project is largely methodological in that it expands the parameters of both the social-cognitive and deception detection literature. Specifically it extends the relevance of implicit association procedures to include not only evaluative association but to beliefs about factual statements. Within the context of deception detection, above-chance predictions based on response latencies suggest that episodic information, both actual and fictional, is respectively linked to our cognitive representations of “the truth” and “our lies.” In an extensive review of the literature Nosek, Greenwald, and Banaji (2007) insist that the IAT was never intended as a lie detector and cite the lack of any evidence for such an application. That is, whereas the IAT was intended to be a measure free of the social desirability and selfserving biases that plague explicit attitude measures (Clark & Tifft, 1966; Jones & Sigall, 1971; Paulhaus, 1984) Nosek et al. (2007) caution that discrepancies between implicit and explicit measures cannot necessarily be interpreted as evidence of deception on the latter. It is more likely, given their conceptualization of implicit attitudes, that individuals may simply be unaware of their ambivalence towards the target attitude object. Nonetheless many have argued that such a discrepancy is evidence that individuals are attempting to conceal socially undesirable attitudes and, in effect, lie about their true feelings (see Fazio & Olson, 2003). Such debate notwithstanding, the IAT’s application to objectively true or false responses has yet to be explored, nor has any theory been advanced regarding the cognitive process by which such a procedure might work.

Cognitive Deception Detection

There is a small body of literature that has attempted to detect deception via reaction times based on the premise that lying requires more cognitive effort than telling the truth. There is some support for the prediction that dishonest responses take longer to put forward (e.g., Walczyk, Roper, Seemann, & Humphrey, 2003), but other studies have yielded inconsistent or contradictory results (e.g., Hsu, Santelli, & Hsu, 1989). Gregg (2007), in what is perhaps the most conceptually similar study to the present experiments, had participants categorize stimuli as either true or false. Those statements included factual and inaccurate statements about the world (e.g., “Grass is green” or “Grass is blue”) and about the participant (e.g., “I am a male” or “I am a female”). After completing a block of trials in which the two categories of statements were intermixed, participants were then instructed to complete a block in which they accurately categorized worldly statements but dishonestly categorized personal ones. Consistent with Gregg’s (2007) predictions, latencies were longer when participants were forced to respond dishonestly to questions about their first name, age, gender, address, birthplace, marital status, citizenship, and diet. Two important limitations to Gregg’s (2007) study are relevant to the present proposal. First, participants were not attempting to conceal information about themselves with the goal of being believed, but rather were following instructions to categorize incorrectly statements that had already been established as true or false. That is, the participants were fully aware that the answers they were providing in later blocks of trials were already known to be false by virtue of the instructions given.


Participants first completed a questionnaire asking them to report, among other things, whether they had ever used an illegal drug. The matter of one’s drug history was chosen because pilot testing indicated that roughly 50% of students have used an illegal substance at least once while 50% have not. We can be reasonably certain that the information provided on the initial survey was accurate because it was explicitly explained that all data would be tracked only by a random subject number not in any way associated with their identities. Further, participants were asked to complete the anonymous survey privately, seal it in an envelope themselves, and place it in a sealed box to be opened only after the study was complete. Given the extent to which their anonymity was protected there is no reason to suspect that any substantial number of participants were motivated to lie out of concern of legal ramifications or social desirability.

Fully aware that the purpose of the study is to evaluate a new lie detection technique, participants were then randomly assigned either to confess or deny illegal drug use upon any further questioning by the researcher. By doing so the participants were, in effect, assigned to respond honestly or dishonestly without the researcher being aware of their condition. This helped create the desired impact in that only the participant was aware of whether he or she was lying or telling the truth. Additionally, the participants’ actual drug history was not confounded with whether they were asked to lie or tell the truth.

Reference :

National Research Council. (2003). THe polygraph and lie detection. Washington DC: National Academy Press.

Nosek, B., Greenwald, A., & Banaji, M. (2007). The Implicit Association Test at Age 7: A Methodological and Conceptual Review. New York: Psychology Press.

Paulhaus, D. (1984). Two-component models of socially desirable responding. Journal of Personality and Social Psychology, 46, 598-609.

Dissertation Done by Scott Peter Roberts, University of Maryland






<< back

Copyright © Dissertationideas.co.uk 2012 through 2014