J. Peter Rosenfeld

A Modified, Event-Related Potential-Based Guilty Knowledge Test

Subjects chose and pretended to steal one object from a box of nine. They then watched a visual display of verbal representations of objects including their chosen object or one of eight novel objects on each trial. They were told to count one of the novel objects and that although they were welcome to try to beat our test, they would be unable to avoid noticing the chosen object. P3 responses were obtained only to counted and to chosen objects in 7 of 10 subjects not eliminated for artifact or noncooperation.

Lying in the scanner: covert countermeasures disrupt deception detection by functional magnetic resonance imaging.

Functional magnetic resonance imaging (fMRI) studies have documented differences between deceptive and honest responses. Capitalizing on this research, companies marketing fMRI-based lie detection services have been founded, generating methodological and ethical concerns in scientific and legal communities. Critically, no fMRI study has examined directly the effect of countermeasures, methods used by prevaricators to defeat deception detection procedures. An fMRI study was conducted to fill this research gap using a concealed information paradigm in which participants were trained to use countermeasures. Robust group fMRI differences between deceptive and honest responses were found without, but not with countermeasures. Furthermore, in single participants, deception detection accuracy was 100% without countermeasures, using activation in ventrolateral and medial prefrontal cortices, but fell to 33% with countermeasures. These findings show that fMRI-based deception detection measures can be vulnerable to countermeasures, calling for caution before applying these methods to real-world situations.

Peak-to-peak measurement of P300 recorded at 0.3 Hz high pass filter settings in intraindividual diagnosis: complex vs. simple paradigms.

We compared effects of 0.3 Hz with 0.01 Hz settings of the high pass amplifier filter, and baseline-to-peak with peak-to-peak measurements of the P300 event-related potential. The key dependent variable of interest was intraindividual rate of accuracy in discrimination of oddball vs. frequent evoked P300 responses, in various paradigms. In Experiment 1 (a lab deception paradigm), we found that the combination of the 0.3 Hz filter setting and the peak-peak measurement of P300 correctly diagnosed oddball vs. frequent in 26 of 26 (100%) cases. This parameter combination outperformed all others. In a second, more field-like experiment (in that the participant knew that the experimenter was blind to ground truth), the peak-peak index again outperformed the base-peak index. It was also observed that the pre-stimulus EEG baseline variability exceeded that of the negative peak (NEG) following P300, i.e. the peak to which the peak-peak index refers P300 for computation. We also observed that the base-peak measurement of P300 is uncorrelated with NEG, and that NEG, seen only in 0.3 Hz channels, correlates highly (-0.67) with the duration of recovery of P300 to the pre-stimulus baseline EEG level as seen in the 0.01-Hz channel. However, in a final experiment using two simple visual and auditory oddball tasks, the base-peak measurement was as diagnostic as the peak-peak measurement.

Further investigation of traumatic brain injury versus insufficient effort with the California Verbal Learning Test

The present study replicates and attempts to extend previous research using the California Verbal Learning Test (CVLT) to identify malingerers. Documented moderate and severe traumatic brain injury patients (n = 42) were compared with clinical malingerers identified by criteria other than the CVLT (n = 21), malingering simulators instructed in realistic potential injury sequelae (n = 25), and normal controls (n = 21). Results of discriminant function analyses for high and low base rates are reported, showing similar results. Also, the four individual cutoff scores (Recognition Hits, Discriminability, Total Words Recalled, Long Delay Cued Recall) from Millis, Putnam, Adams, and Ricker (1995) were evaluated with these groups. Similar specificity rates were found with all four variables, while sensitivity rates were slightly lower than that of Millis. Adjusted cutoffs derived from the new samples resulted in slightly improved overall classification rates. Overall, present findings support those of Millis et al. (1995) with regard to the use of the CVLT in detection of malingering. Exploratory use of Total Intrusions and Recognition Hits Compared to Long Delay Free Recall was not promising. Simulators were found to be fairly comparable in performance to actual malingerers, affirming their use in malingering research.

P300 scalp amplitude distribution as an index of deception in a simulated cognitive deficit model

Truth-telling (Truth) and simulated malingering (Malinger) groups were tested in a matching-to-sample procedure in which each sample three-digit number was followed by a series of nine test numbers, only one of which matched the sample. P300 was recorded during test-number presentation. Group analyses revealed differences between the P300s of the groups in unscaled amplitude, but not latency, in response to match and mismatch stimuli. P300 amplitudes at Fz, Cz, and Pz were scaled to remove possible confounding effects of amplitude in tests of the interactions of site with other variables. Significant interactions of both stimulus-type (match vs. mismatch) and group (Truth vs. Malinger) with site were obtained. Within the Malinger group, a significant interaction was obtained (scaled data) between site and response type (honest vs. dishonest). These interactions suggest that deceptive and honest responding are associated with different neurogenerator sets or different sets of P300-overlapping components. In within-individual analyses, 100% of the Truth participants and 87% of the Malinger participants were found to have larger P300 responses at Pz to match stimuli than to mismatch stimuli on the basis of intra-individual bootstrap tests. This represents an improvement in comparison with our related, previous report on a matching-to-sample test using only one test stimulus per sample.

Detecting simulated amnesia for autobiographical and recently learned information using the P300 event-related potential.

To investigate whether the P300 (P3) event-related potential (ERP) can be used as an index of the intactness of recognition memory in subjects trying to simulate amnesia, two groups of subjects (n = 12 and n = 15) were instructed to simulate amnesia and one group of control subjects (n = 14) did not simulate amnesia while taking three recognition tests, during which ERPs were recorded. The three tests consisted of three different types of memory items: (1) the subjects birthday (birth), (2) the experimenters name (name), (3) a word list of 14 nouns (words). The memory item was presented in a random series with other, similar in type, non-memory items. In group tests, memory items evoked larger amplitude P3s than non-memory items (p < 0.001). Within-subjects tests were used to determine whether the P3 amplitude in response to memory items was larger than the P3 amplitude in response to non-memory items for each individual. There was no difference between the sensitivity of the best within-subjects tests for amnesia simulators (birth = 0.9, name = 0.85, words = 0.53) versus non-simulators (birth = 1.0, name = 0.81, words = 0.5) averaged across the three test types. This suggests that P3 used as an index of the intactness of recognition memory may be useful in cases of suspected malingering.

Preliminary evidence that daily changes in frontal alpha asymmetry correlate with changes in affect in therapy sessions.

Frontal EEG alpha asymmetry was recorded from five depressed outpatients during early EEG biofeedback sessions. Mood was assessed prior to and after each session, and affect change scores were also derived by subtracting pre-session from post-session scores. Alpha magnitude was obtained via Fast Fourier Transforms. All scores (EEG alpha asymmetry and affect) were converted to deviation scores by subtracting each patients daily score from that patients mean across all available sessions for that patient. Pearson correlations were then computed between asymmetry and affect scores using the deviation scores combined over patients. There was little evidence of correlation between day-to-day asymmetry score and any single affect score. Strong correlations were obtained, however, between asymmetry score and affect change score and, in particular, between asymmetry score and change in positive affect.

A mock terrorism application of the P300‐based concealed information test

Previous studies examining the P300-based concealed information test typically tested for mock crime or autobiographical details, but no studies have used this test in a counterterrorism scenario. Subjects in the present study covertly planned a mock terrorist attack on a major city. They were then given three separate blocks of concealed information testing, examining for knowledge of the location, method, and date of the planned terrorist attack, using the Complex Trial Protocol (Rosenfeld et al., 2008). With prior knowledge of the probe items, we detected 12/12 guilty subjects as having knowledge of the planned terrorist attack with no false positives among 12 innocent subjects. Additionally, we were able to identify 10/12 subjects and among them 20/30 crime-related details with no false positives using restricted a priori knowledge of the crime details, suggesting that the protocol could potentially identify future terrorist activity.

Comparison of two EEG asymmetry indices in depressed patients vs. normal controls

In 11 non-depressed, age-matched controls, and in 13 depressed patients, we compared the frontal alpha asymmetry mean for a baseline session with the percentage of the time in the session when the asymmetry score > 0. It was found that the percent index was a better discriminator of the two groups than was the asymmetry score.

Late vertex positivity in event-related potentials as a guilty knowledge indicator: A new method of lie detection

Subjects were allowed to choose an item to keep from nine items in a box. They then were shown one of nine words randomly selected on a display screen. One of these words described the chosen item, the others described novel items. The subjects were told to try not to react emotionally to any of the words, but to try to defeat this test of deception. It was found that large positive waves with latencies between 400 and 700 ms poststimulus were present in the ERPS to the chosen but not to the novel words.