Limitations to the detection of deception: True and false recollections are poorly distinguished using an event-related potential procedure

Abstract

The utility of using indices of neural function to identify deception relies on finding highly reliable and valid approaches that adequately identify the guilty and exonerate the innocent. A class of approaches, based on the guilty knowledge technique (GKT), assume that guilty individuals will recognize specific crime-relevant details, whereas innocent individuals will not. Memory distortions, however, may limit the accuracy of such procedures. To investigate these limits, two studies were conducted to examine whether brain electrical activity could differentiate true from false recollections elicited by the Deese-Roediger-McDermott (DRM) paradigm. The design of each study maximized the opportunity of finding electrocortical differences between true and false recognition. Each study found very high rates of false recognition, with little evidence that brain electrical activity could differentiate true from false memories. Results suggested that under certain conditions both true and false recollections can produce a pattern of brain activity indicative of recognition.

Acknowledgements

The authors thank Veronica Rodriguez, Jeffrey Chang and Lori Anne Tracy for their invaluable help in recruiting and running participants for these studies. Portions of these data were presented at the annual meetings of the Society of Psychophysiological Research in Granada, Spain, October 1999 and San Diego, CA, October 2000.

Notes

¹The Department of Psychology does not permit the exclusion of study participants based on handedness or native language. To comply with these requirements, 38 undergraduate students expressing interest were enrolled in Study 1; 4 participants were excluded for not being exclusively right-handed and 6 excluded because they were non-native English speakers. One additional student was excluded after acknowledging having been aware of the paradigm after data collection was completed. Data for 5 participants were never analyzed due to equipment problems and excessive artifacts. Data from the remaining 22 students were initially analyzed, but 2 additional participants were excluded from the analyses because they failed to produce at least 10 behavioral responses indicating false recognition of lure words. Inclusion of these 2 participants changed the overall recognition rates to 73% for Lures, 66% for Learned, and 5% for Distractor items.

²The display software used for Study 1 was the DOS-based DMTG, and the display software for Study 2 was the Windows based DMDX, both developed at the University of Arizona by J. C. Forster and K. I. Forster. A detailed description of these systems can be found at www.u.arizona.edu/~jforster/dmdx.htm

³Available at http://www.psy.uwa.edu.au/MRCDataBase/uwa_mrc.htm

⁴Study 2 enrolled a total of 35 undergraduate students. Data from 4 participants were never collected due to equipment problems, 1 student was aware of the paradigm and data from 5 participants showed excessive EEG artifact. Data from the remaining 25 participants was initially analyzed, but 3 additional participants were excluded from the analyses because they failed to produce at least 10 behavioral responses indicating false recognition of lure words. Inclusion of these three participants changed the overall recognition rates to 69% for Lures, 69% for Learned and 7% for Distractor items.

⁵To assess whether latency jitter across trials may have contributed to the amplitude differences, the approach described by Woody (1967) was used to align single trials with predefined sine-wave template that was centered on 550 ms. The procedure shifted trials up to ±125 ms, selecting the shift that maximally correlated with the template. Subsequent iterations then used the average of latency-shifted trials from the previous iteration as a template, and again shifted epochs to maximize the correlation with the template. Iterations terminated when the average correlation of single trials with the template increased less than .005 from the previous iteration. The output of this procedure then allowed for an examination of the degree of latency shift required to fit individual trials to the grand average. Analysis of these latency shifts revealed no significant difference across lure and learned conditions in terms of either mean latency shift, F(2, 44) = 0.31, ns, or the absolute value of these latency shifts, F(2, 44) = 0.95, ns, suggesting that the extent and direction of latency jitter were not different between these conditions.