1
Are individuals with higher psychopathic traits better learners at lying? Behavioural and neural evidence
Supplementary Information (Additional Materials and Methods, Data analyses and discussions, Legends for Supplementary Figures, and Supplementary Tables)
Robin Shao Ph.D1,2 and Tatia M.C. Lee Ph.D1,2,3,4*
1 Laboratory of Neuropsychology, The University of Hong Kong, Hong Kong
2 Laboratory of Cognitive Affective Neuroscience, The University of Hong Kong, Hong Kong
3The State Key Laboratory of Brain and Cognitive Sciences, The University of Hong Kong, Hong Kong
4 Institute of Clinical Neuropsychology, The University of Hong Kong, Hong Kong
Correspondence to:
Tatia M.C. Lee, Ph.D.
Rm 656, Jockey Club Tower
The University of Hong Kong
Pokfulam Road, Hong Kong
Tel: (852) 3917-8394
E-mail:
Additional Materials and Methods
Participant
Seven low-PPI and one high-PPI participants whose PPI-R scores satisfied the criteria for entering Stage 2 dropped out, leaving 23 low-PPI and 29 high-PPI participants for Stage 2 screening. Although official records of the participants’ crime histories were not available, the likelihood of such histories of serious crimes was very low given all participants were officially enrolled HKU students at the time of the study. Detailed exclusion criteria for stage-2 participant screening include the following: 1) <18 years old; 2) any personal or family (first-degree relative) history of major physical illness, neurological or psychological conditions, such as psychotic disorders, affective disorders or addictions; 3) unsuitability for entering functional magnetic resonance imaging (fMRI) scanner (e.g., mental implant, claustrophobic or pregnancy condition); 4) not primarily right-handed as assessed based on the Edinburgh Handedness Inventory1; 5) currently taking regular medications or 6) non-correctable visual impairments. The PPI-R items are grouped into 8 subscales (Machiavellian egocentricity, Rebellious nonconformity, Blame externalization, Carefree nonplanfulness, Social influence, Fearlessness, Stress immunity, and Coldheartnedness), and a contentious 3-factor structure has been proposed for PPI-R, namely ‘Fearless-dominance’, ‘Self-centred impulsivity’ and ‘Coldheartedness’2. No gender difference was observed for either the PPI-R total raw score or the sub-factor scores (ps>.37). The mean total PPI-R raw score of the included participant sample after Stage 2 screening (306. 23) was higher than previously reported mean total scores from university or community samples3,4 and modestly lower than those of prisoner samples5. Thus, the current sample was likely located at the higher spectrum of psychopathic traits. We did not measure participants’ IQ scores, as we expected their IQ levels to be reasonably similar across individuals given they were drawn from a homogenous student population confined to one university campus and of similar ages (SD=1.58 years). Additionally, no clear relationship between PPI total scores and IQ scores in community samples has been observed2,6. Two stage 2 participants were excluded from behavioral analyses due to poor task performance (see the section ‘Error and outlier frequency analysis’ below) or abnormal picture familiarity ratings (see the section ‘Task and Materials’ below), and two others were excluded from the imaging analyses due to excessive in-scanner movement (see the section ‘Image acquisition and analysis’ below). The current sample size (23 low-PPI and 29 high-PPI participants) was chosen based on previous research investigating the effect of psychopathy tendency on lying-related or social-defect-related behavioural and fMRI measures7-10. The principle investigator (RS) was blind to the participants’ PPI group status during conducting the experiment.
Task and Materials
During photo-taking, the participant was explicitly instructed to 1) look directly forward, 2) maintain neutral facial expressions, and 3) remove all facial accessories. Subsequently these photos were converted to grey scale and cropped to include only the face and the hair. All of the unfamiliar photos also faced and looked directly forward, maintained neutral facial expressions (as judged by 2 experimenters independently), and were converted to grey scale and cropped to reveal only the face and the hair. Photo processing was conducted using Adobe Photoshop 12.0 (San Jose, CA, USA). To further ensure the level of neutrality and non-familiarity of theunfamiliar faces, each of the unfamiliar photos was independently rated by 12 judges, and none of the 60 unfamiliar photos employed by the task received more than 2 points on a 7-point Likert scale (1=extremely low and 7=extremely high) on facial expression emotion intensity or familiarity. The familiar and unfamiliar photos were then matched for resolution (200 × 250 voxels displayed on a 640 × 480 PC monitor for the behavioural sessions, 480 × 600 voxels displayed on a 1600 × 1200 PC monitor for the imaging sessions), dots-per-inch (DPI = 72), brightness, the amount of space occupied by the face in the picture, and the position of the face in the picture.
To ascertain that our participants indeed perceived their friends’ photos as familiar and the unfamiliar faces as unfamiliar and perceived both familiar and unfamiliar faces as emotionally neutral, we asked participants to rate the levels of familiarity and emotion intensity of all the face stimuli. Familiarity ratings were made both before the initial session and before the testing session. No participant reported knowing any of the unfamiliar faces. As expected, almost every participant rated the familiar faces as highly familiar at initial session, reflected by scores >5 on a 7-point Likert scale (1=highly unfamiliar, 7=highly familiar), and the unfamiliar faces as highly unfamiliar, reflected by scores <3. Only one male high PPI participant showed abnormally low ratings to the familiar faces (mean score <5), who was thus excluded from all behavioural and imaging analyses. Comparing the familiarity ratings made at initial session and before testing revealed a general increase in familiarity from initial session to pre-testing and a familiarity × session effect (both ps<.001) resulting from a larger increase in familiarity with time for the unfamiliar faces than for the familiar faces, which was likely due to familiarity levels for the familiar faces already approaching the maximum level at initial session (Mean=6.48). However, familiar faces were clearly perceived as being more familiar than unfamiliar faces at both initial session (F (1, 47) = 1279.34, p<.001) and pre-testing (F (1, 47) = 1865.935, p<.001). PPI group did not interact with session, familiarity, or session × familiarity (ps>.742). Also as expected, participants rated both familiar (Mean = 1.36) and unfamiliar photos (Mean = 1.56) as of very low emotional intensity on a 7-point Likert scale (1=very weak emotion, 7=very strong emotion). No significant effect of familiarity or familiarity × PPI group was found on emotion ratings (ps>.08).
In the directed lie task, the 30 unfamiliar faces were randomly split into 2 sets of 15, with half the participants required to give honest responses to the first set and dishonest responses to the second set, which was reversed for the other half of participants. This was performed to minimize any unexpected characteristic differences across unfamiliar faces that may confound the effect of response type. No unfamiliar face was presented more than once in either the initial or the testing sessions, thus preventing any potential within-session learning effect. A similar arrangement of repeating familiar stimuli and/or non-repeating unfamiliar stimuli was adopted in previous studies10,11. The correspondence between the circle/triangle cue and the honest/dishonest response, between the key press and the ‘known’/’Unknown’ response, and the spatial positions (left/right) and colours (red/blue) of the ‘known’ and ‘Unknown’ text were all counterbalanced across participants but remained constant for each participant across sessions. The experimental task was delivered via E-prime software (Psychology Software Tools, Inc.).
Based on the presumption that thecognitive and motor processes required by the control task were primarily a subset of the processes involved in performing the directed lie task, we did not include the control task in the training sessions.
Error and outlier frequency analysis
We analyzed participants’ error frequency and outlier rate. Outlier trials were defined as those in which thereaction times (RTs) exceeded 5 seconds or 4 standard deviations from thatparticipant’s session mean RT. Both error and outlier trials were excluded from the behavioural analyses. One female low-PPI participant had >15% trials excluded in training session 2 and her data were thus discarded. For the remaining 50 participants, at least 85% of all trials were retained in each of the task sessions (Table 1). The error rates for neither the low-PPI nor the high-PPI groups differed across the initial and training sessions (ps>.34), minimizing the possibility that any difference after training was due to switching to a ‘gaining-speed-sacrificing-accuracy’ strategy. The total excluded trial frequency also did not change across the initial and testing sessions (F (1, 46) = 1.059, p>.3).
Levene’s test of equality of variance showed that the high-PPI and low-PPI groups were comparable in variance of RT across each of the response type, picture familiarity and session conditions (Ps>.18).
Image acquisition and analysis
Imaging data were acquired using a 3-T Phillips scanner equipped with an 8-channel SENSE head coil. EPI slices were acquired at 30o relative to AC-PC to optimize signals for orbitofrontal cortical regions. Anatomical T1-weighted images were acquired using a magnetization-prepared rapid gradient-echo (MPRAGE) sequence. Image normalization into MNI space was via unified segmentation of T1 images. Two participants who moved by more than 1 voxel in theX, Y or Z directions during either the initial or testing session were excluded, leaving 48 participants for the imaging analyses.
When carrying out Alphasim cluster-level correction, we ensured the correctness of the smoothness estimate through using estimates generated by the SPM, and the correctness of the voxel number computation included in the searching space by cross-checking with the number generated by the SPM.
On the participant level, besides the main contrast assessing the effect of response type, other contrasts that were computed included the comparisons of honest and dishonest response types to the control task type and those relevant to the main effect of familiarity (familiar>unfamiliar, unfamiliar>familiar) and response × familiarity (Honest (familiar>unfamiliar)>Dishonest (familiar>unfamiliar), Dishonest (familiar>unfamiliar)>Honest (familiar>unfamiliar)).These latter contrasts, however, were computed for complementary and exploratory purposes only.
Regions of interest (ROI) anatomical masks were constructed using WFU_Pickatlas software based on Talairach Daemon atlas. The ROIs for cue-elicited activities included the parietal cortices encompassing the inferior and superior parietal regions (IPL and SPL), the dorsolateral frontal cortex encompassing the frontal pole and supplementary motor area (SMA), the VMPFC, and the cerebellum. For picture-elicited activities, theROIs included the ventrolateral prefrontal cortex/insula (VLPFC-insula), the ACC, and the cerebellum. The dorsolateral frontal cortex mask encompassed both the superior and middle frontal gyri (BA 6, 8, 9, 46)12, and the VLPFC mask encompassed principally the inferior frontal gyrus (IFG), including the lateral OFC (BA 44, 45, 47)13 and the adjacent anterior insula (BA13)14.
Additional behavioural analysis
It could be seen from Figure 2 that high-PPI participants generally showed faster RTs than low-PPI participants. The reason for this is unclear. It could be that the high-PPI participants were generally more impulsive in response making2. However, their error rates were comparable to those of the low-PPI participants. Alternatively, the high-PPI participants may have better visuospatial motor capacity than the low-PPI participants. As we did not have separate measures for such capacity, it is difficult to know whether either, or both, of the above factors led to the faster overall RTs of high-PPI participants. However, given in the current study we were primarily interested in the behavioral and neural response difference between giving lying and truthful responses, and both high-PPI and low-PPI participants showed greater RT to lying responses than to truthful responses during the initial session, we believe that our current findings on the interactive effects of response type, session and PPI group are valid despite group difference in general RTs.
Participants’ RTs were longer when responding to unfamiliar faces than to familiar faces (β=.061, z=7.63, p<.001). This is consistent with previous literature showing that lying-related RTs are generally longer for stimuli that are ‘new’ or ‘personal’ relative to those that are ‘old’ or ‘impersonal’7,10,11,15-17. The familiarity of the face also modulated the RT increase when making dishonest responses versus honest responses (β=-.058, z=-3.63, p<.001), such that the increase was larger when responding to familiar faces (β=.287, z=16.11, p<.001) than to unfamiliar faces (β=.171, z=9.24, p<.001) (Supplementary Figure S1a). Similar interactive effects of response type and familiarity were observed in one previous studying that employeda directed lie task on face familiarity17, but not in other studies10,11,15, possibly dependent on the study design (e.g., trial number, stimulus type, familiarity manipulation). This response type × familiarity effect was further modulated by session (β=-.062, z=-1.96, p<.05), as the response type × familiarity effect was only observed during the testing session (β=-.078, z=-4.31, p<.001) but not during the initial session (β=-.02, z=-1.14, p>.1) (Supplementary Figure S1b). Such results indicate that, across the total participant sample, while lying speed associated with unfamiliar faces may be improved through practice, lying speed associated with familiar faces may not, consistent with previous accounts that processing familiar faces is largely automatic18,19.
Furthermore, the familiarity× session × PPI group effect was significant (β=.065, z=2.03, p<.05). Detailed analyses revealed that a familiarity × session effect was significant in the high PPI group (β=.039, z=2.26, p=.024), characterized by a marginally significant increase in the response RT for unfamiliar faces compared to familiar faces during the initial session (β=.047, z=1.96, p=.05) but a much greater increase in the response RT for unfamiliar versus familiar faces during the testing session (β=.134, z=5.45, p<.001) (Supplementary Figure S1c). No significant familiarity × session effect was found in the low PPI group (β=-.016, z=-.84, p>.1). Such reduced RTs to familiar versus unfamiliar faces following training in participants with higher psychopathic traits may indicate a particular decrease in the processing of familiar faces following repeated exposure and/or responses, which might relate to those individuals’ fundamental social affective characteristics20.
Finally, gender had no main effect on RTs, and it didn’t interact with response type, response type session, response type group or response type PPI group effects (all zs<1.94, ps>0.05). Nonsignificant effect of gender on lying RT was also observed in several previous experiments7,21,22. Further, previous evidence indicates that gender may have little moderating effect on the association of PPI scores and external personality constructs23, and high-PPI and low-PPI male and female participants in our study didn’t differ in either total PPI-R scores or sub-factor scores (P.25), both of which may be related to the lack of moderating effect of gender on PPI effects as observed by us. To our awareness, no study that looked at the training effect on lying RT has specifically examined gender effect. As such, our findings provide preliminary support for lack of gender differences in the trainability of lying RT.
Additional whole-brain analysis
Relative to responding to unfamiliar faces, responding to familiar faces activated widespread areas in the dorsal and ventral medial prefrontal cortex, the IFG, the PCC/precuneus, the STS, the medial temporal limbic regions, and the parietal and visual cortices. These results are consistent with previous studies that investigated lying on face familiarity10,24and on personal versus impersonal information7. The reverse contrast generated a sole cluster in the visual area involved in face processing, likely reflecting early-stage perceptual processing of the relatively novel unfamiliar faces25. Furthermore, high PPI relative to low PPI participants showed reduced activations to familiar versus unfamiliar faces within the IFG-insula, the striatum, the STG, and the motor and inferior parietal regions. Both the IFG-insula and STG are included in the ‘paralimbic’ circuitries where deficits have been proposed for high psychopathy20, and reduced reactivity in these regions along with the sensori-motor networks could indicate underlying social affective deficits of individuals with higher psychopathic tendencies, which influence their responses towards ‘known’ and ‘unknown’ faces. Finally, unlike a previous studying that employeda directed-lie task10, we found no interactive effect of response type and familiarity on the neural level. Various differences in the design, paradigm and participant characteristics between the current study and the previous study may explain the observed difference. In particular, aside from the cross-sectional versus longitudinal design, our current study separately modelled the cue-processing and face-responding phases, whereas the previous study treated the two phases as a single epoch10.
Control ROI analyses for familiarity rating change
We also tested whether the difference in the response type effect on neural responses before and after training was due to changes in the perceived familiarity of the face stimuli. We computed the average and difference in the familiarity rating change of familiar and unfamiliar photos and included them as covariates of no interest into the ANOVA models. The results showed that controlling for changes in the familiarity rating had very limited impact on the ROI results, such that almost all the response type × PPI group, response type × session and response type × PPI group × session effects were preserved. Only two effects became non- or less significant after adding in the familiarity change covariate terms: the response × session effect on the face-responding signals in the ACC became marginally significant (F (1, 42) =3.334, p=.075), and the marginally significant response × session effect in the cerebellum became non-significant (p>.1). Taking into account the loss of the degree of freedom due to adding in the extra covariate terms, we consider the actual impact of face familiarity change to be very limited in the current observed results.