The Quarterly Journal of Experimental Psychology Impaired Pavlovian Conditioned Inhibition in Offenders with Personality Disorders

distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.

Thus some have argued that a unifying feature in the psychological profile of forensic populations is poor impulse control, similarly recognized as a feature of psychopathy (Johansson, Kerr, & Andershed, 2005;Lesch & Merschdorf, 2000;Prichard, 1837) and BPD (Stein, Hollander, & Liebowit 1993;Stein, Towney, & Hollander, 1995). Impulsivity, a complex and multidimensional concept, spans lack of inhibitory control, actions without forethought, novelty seeking, and inability to delay gratification (Barratt, 1985(Barratt, , 1994. Impulsivity in psychiatric disorders is obviously more extreme than that in healthy individuals and has been described as: "a predisposition toward rapid, unplanned reactions to internal or external stimuli without regard to the negative consequences of these reactions to the impulsive individual or to others" (Moeller, Barratt, Dougherty, Schmitz, & Swann, 2001, p. 1784. But despite its obvious importance, the exact nature of impulsivity has not been unambiguously specified, making further analysis difficult. For example, some have argued that impulsive behaviour results from lack of inhibitory control (Buss & Plomin, 1975), others that it stems from an inability to tolerate delays of reinforcement (e.g., Logue, 1988;Logue et al., 1992;Thiébot, Le Bihan, Soubrié, & Simon, 1985). In any event, individuals with high impulsivity fail to inhibit unwanted actions, and thus behavioural measures of impulsivity include a range of established laboratory behavioural tasks designed to measure this deficit, such as the go/no-go task, the stop-signal task, and antisaccadic eye movement procedures. These tasks can be characterized by the fact that the subject is normally required to make a specific response, but on certain occasions is required to withhold it, and a failure to do so is taken as an index of impulsivity; for example, in a stop-signal task the subject is required to respond unless the stop signal is presented. Moreover, deficits in the performance of such tasks have been demonstrated in participants with BPD and psychopathy (e.g., Grootens et al., 2008;Kiehl, Smith, Hare, & Liddle, 2000;Newman & Schmitt, 1998;Nigg, Silk, & Stavor, 2005;Rentrop et al., 2007;Rubio et al., 2007;Ruchsow et al., 2008).
However, inhibition is a broad construct and should not be too narrowly identified with any one behavioural paradigm. A response may occur not only because it has been explicitly rewarded, as in the traditional tests of impulsivity, but also as a conditioned response to an environmental stimulus that has consistently signalled some outcome of motivational value (as a result of Pavlovian stimulus-stimulus, S-S, contingencies rather than response-stimulus, R-S, learning). Moreover, in the chain of cause and effect that ultimately results in unwanted actions, environmental cues that trigger associated thoughts and emotions through such S-S associations can be primary. The so-called weapons effect-that the presence of a weapon can heighten the level of aggression displayed-provides the classic example of how potentially antisocial responses can be triggered by environmental cues, in this case guns (Berkowitz & LePage, 1967). Moreover, such situations are typically constituted by more complex configural cues: For example, the sight of the gun generally precedes the sensory stimuli provided by grasping the weapon. Thus, the inhibition of S-S associations could play a critical role in suppressing unwanted behaviours. As it cannot be assumed that the same psychological mechanisms inhibit S-S and R-S associations (Nigg, 2000), the inhibition of responding based on Pavlovian S-S contingencies-formally "conditioned inhibition" (CI)-should also be examined.
In CI procedures, the expectation of an outcome is inhibited by the presence of a qualifying stimulus. Building on the basic classical conditioning procedure in which a conditioned stimulus (CS) signals an outcome (unconditioned stimulus, US), a further stimulus (the conditioned inhibitor) signals the omission of the expected US. For example if, after a number of pairings of a CS with a US (A+), A is paired with another stimulus B and the US is omitted (AB-), subjects learn that B signals the absence of the US-that is, it becomes a conditioned inhibitor (Pavlov, 1927). While a few studies have successfully demonstrated CI in humans (Grings, Carey, & Schell, 1974;Migo et al., 2006;Neumann, Lipp, & Siddle, 1997;Wilkinson, Lovibond, Siddle, & Bond, 1989), to date very few studies have explored CI in relation to individual differences, and these used quite different tasks and measures (Kantini, Cassaday, Hollis, & Jackson, 2011;Migo et al., 2006); moreover, to our knowledge, to date none have demonstrated CI deficits in relation to psychological or psychiatric disorder. If the effective disinhibition of responding based on undesirable S-S contingencies (such as the sight followed by the feel of a weapon; Berkowitz & LePage, 1967) contributes to the well-documented inability of offenders to inhibit punished responses, we might expect to see a deficit in CI in this subject group. Thus the present study was devised as a first test of the prediction that offenders with PD (in the absence of comorbid schizophrenia or learning disability) would show impaired CI.
To date, learning differences and attentional anomalies have been most extensively investigated specifically in relation to psychopathy, rather than PDs as defined by DSM-IV criteria. Such studies have also tended to use learning tasks in which outcomes can be negative, as a defining feature of impulsivity is heedless action despite aversive consequences (Cleckley, 1964;Flor, Birbaumer, Hermann, Ziegler, & Patrick, 2002;Gorenstein & Newman, 1980;Kosson, Smith, & Newman, 1990;Lykken, 1957;Newman, 1987;Newman, Patterson, & Kosson, 1987;Newman & Schmitt, 1998;Schmauk, 1970;Thornquist & Zuckerman, 1995;Vitale & Newman, 2001). However, as discussed above, psychopathy has been argued to be on a continuum with ASPD (Blackburn & Coid, 1998;Coid & Ullrich, 2010;Kosson et al., 2006). Thus, since a high proportion of the present sample were expected to be psychopathic and therefore poor at aversively motivated learning, we used a positively motivated learning task. This also allows us to rule out any contribution of insensitivity to aversive outcomes-a true deficit in CI would be evident regardless of the valence of the omitted outcome. Finally, in forensic populations of this type, poor impulse control may contribute to general learning deficits, whenever unwanted actions interfere with task performance (Ávila & Parcet, 2001;Gullo, Jackson, & Dawe, 2010). Therefore the task developed for use in this population measured both inhibitory and excitatory learning within the same procedure, to simultaneously monitor any nonspecific, general learning deficits.

Method
Participants A total of 26 offenders volunteered to participate in the experiment, 2 of whom were subsequently excluded for a priori procedural reasons (see below). They were all male inpatients at Rampton Hospital, a high-security psychiatric hospital in the UK. All the participants recruited to the present study had a history of criminal offending behaviour, including convictions for violent and sexual offences.
Assessment of violence was based on offending history using a Severity of Violence Rating Scale (SVRS) adapted from that originally developed (Gunn & Robertson, 1976) and validated in hospitalized forensic patients (Wong, Lumsden, Fenton, & Fenwick, 1993). The SVRS was modified in house for use at Rampton by adding a measure of institutional behaviour. Thus, there were three subscales-Violence in Index Offence, Violence in Criminal Record, and Violence in the Institution. Each was rated on a 4-point scale (0 = minimal/ no violence, 4 = severe violence, indicating someone's life or health was seriously endangered), yielding a total score out of 12. Patients' history of violent offending was derived from their case files, and this information was supplemented by selfreport data.
Of those who completed the study, 8 participants were in the Personality Disorder Unit (PDU) and 16 in the Dangerous and Severe Personality Disorder (DSPD) Unit. To meet the criteria for severe PD justifying admission to the DSPD unit, an offender must: (a) score 30 or more on the PCL-R; (b) score between 25 and 30 on the PCL-R plus have at least one DSM-IV PD diagnosis other than ASPD; or (c) have two or more DSM-IV PD diagnoses (Howells et al., 2007).
According to the available IPDE categorical diagnoses (n = 24), of those participants that were finally included, there were definite (n = 18) or probable (n = 4) diagnoses of ASPD and definite (n = 13) or probable (n = 1) diagnoses of BPD, and 1 participant had a confirmed obsessive compulsive personality disorder but was comparable with others in the sample in that this was comorbid with BPD. The one offender without any PD diagnosis according to the IPDE had nonetheless met the criteria for admission to the DSPD (he was diagnosed as ASPD with conduct disorder according to DSM-IV assessment). Psychiatric assessments confirmed that none of the offenders approached in connection with the study had comorbid schizophrenia or learning disability. Fourteen of the offenders were on psychotropic medication, including typical (n = 1) and atypical antipsychotics (n = 6), antidepressants (n = 7), anxiolytics (n = 7), and anticonvulsants (n = 1).
The healthy controls were a community-based sample of 27 participants, 3 of whom were subsequently excluded for a priori procedural reasons (see below), leaving 24 to complete the study. These controls lived in the same county and were recruited at the University of Nottingham (ancillary staff), Nottingham National Ice Centre, and the Nottingham Trent FM Arena. They were all without a higher education; some were unemployed, and others reported having jobs such as swimming instructor, driver, waiter, shop assistant, and university support staff. None reported or showed any indication of mental illness or substance abuse. Table 1 shows the details of participants' age, gender, ethnicity, and educational level. Controls were matched in age to the offender group, t(46) = 1.35, p = .18. Allocation to the counterbalanced experimental conditions was identical for the offender and control groups. Other than the fact that they were not incarcerated, control participants were tested under comparable quiet environmental conditions in the same way by the same experimenter (Z.H. In the UK, the number of years in education required to achieve A level is 14. THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2011, 64 (12) participants received an inconvenience allowance of £5 cash to cover their travel expenses. No such payment was possible in the case of the offender group, but they had no travel expenses or loss of earnings in consequence of participation.

Stimuli
Lego block pictures (n = 9) were used as the CSs in the present study. The USs were selected by a pilot study from the International Affective Picture System (IAPS; Lang, Bradley, & Cuthbert, 2005). The IAPS provides a set of images, standardized on the basis of participants' ratings, on the dimensions of valence and arousal from 1 to 9, 1 representing a low rating on each dimension and 9 a high rating (i.e., 1 as low pleasure, low arousal). Based on standardized ratings, the USs in the present study were provided by 10 IAPS "positive" images and 10 IAPS "neutral" images; because of the nature of the sample, pictures depicting children and erotic images were excluded from the present study. 1 Whenever a "reinforced" trial was required, a positive US was presented, whereas neutral USs followed the CS on "nonreinforced" trials. Conditioning was measured using a rating scale: Participants were asked to guess or predict what kind of picture would follow presentation of the Lego blocks using a rating scale from 1 (neutral) to 9 (positive), with the rating 5 to reflect uncertainty.

Design and analysis
The design of the experiment is shown in Table 2. There were three stages: (a) preexperimental, (b) training with elemental and compound stimuli, and (c) test. In the preexperimental stage, the participants rated the stimuli and stimulus compounds to be used in the test stage, to preclude the possibility that differences in responding to the stimuli at test could be due to biases present before the start of training.
In the elemental training stage, two of the neutral CSs, A and C, were paired with reinforcement, positive IAPS pictures (A+ and C+ trials), while a further two, U and V, were paired with nonreinforcement, neutral IAPS pictures (U-and V-trials). This training provided a measure of participants' simple associative learning. It also established A and C as excitatory CSs (i.e., signalling a positive outcome), which facilitated the subsequent establishment and detection of CI.
During the compound training stage, the AZ compound signalled reinforcement (AZ+), whereas the AP compound signalled nonreinforcement (AP-). As A had been paired with reinforcement in the previous stage, presenting AP allowed P to signal the absence of the reinforcement otherwise indicated by A and was thus expected to establish P as a conditioned inhibitor. Two additional stimulus compounds, CY and BX, were reinforced and nonreinforced, respectively.
Although successful discrimination between AZ and AP would be consistent with the proposal that P was a conditioned inhibitor, it was not sufficient; for example, participants might respond more to AZ simply because Z was reinforced on every trial. In order to establish unequivocally that P was a conditioned inhibitor, we conducted a summation test-more specifically, we examined whether P would suppress responding to a different excitatory stimulus more than would a suitable control stimulus (cf. Rescorla, 1969). The continued excitatory training with C on CY+ trials (C had also been reinforced in the previous stage) means it provided an excitatory test stimulus against which the inhibitory effects of P could be evaluated. The BX-trials were designed to establish X as a control stimulus, which was presented the same number of times as P and in a similar manner (in compound with another stimulus and paired with nonreinforcement); however, the stimulus with which X was presented was novel, so that X, unlike P, did not signal the absence of reinforcement during this training stage and so should not have acquired any inhibitory properties.
The test stage compared the ratings given to the trained stimuli and stimulus compounds that had signalled reinforcement (A, C, AZ, CY) and non-reinforcement (AP, BX), but the critical comparison was between two additional test compounds, CP and CX. C was excitatory and was predicted to elicit high ratings indicating expectation of reinforcement. If P was a conditioned inhibitor, it should reduce this high rating to C, whereas the critical comparison stimulus, X, should not. CI would therefore be evident as lower ratings to CP than to CX. The identities of the stimuli used as P and X were counterbalanced across the participants, as were those of A and B (and C and V, see above).
The dependent variable was the mean rating given for each particular trial type, which was assessed at each stage of the experiment. Statistical analyses were by a mixed-design analysis of variance (ANOVA), with group (offender vs. control) as a between-subjects factor, and discrimination (e.g., A+ vs. U-and C+ vs. V-), reinforcement (reinforced or not), and trial block as within-subjects factors. Significant interactions were explored with simple main effects analysis using the pooled error term. Planned comparisons of the assessment score data were by t test, or Kruskal-Wallis where the data were not normally distributed (i.e., to compare the IPDE scores). The measure of effect size given for mixeddesign ANOVAs was Pearson's correlation coefficient, r.
Correlational analyses (Pearson's or Spearman's, as appropriate) were used to compare overall learning and questionnaire scores. A summary measure of excitatory learning was provided by the difference in mean ratings on C and V trials during the initial training stage-that is, C -V. As C was the excitatory stimulus, the greater the C -V score, the higher the level of excitatory learning. A summary measure of CI was provided by the difference between the mean ratings on CP and CX trials given during the test stage-that is, CX -CP. P was the putative inhibitor and thus was supposed to suppress evaluation of C more than X; thus the higher this score, the greater the inhibitory learning.
An a priori exclusion criterion was applied to the elemental training performance: Five participants (two offenders and three controls) who failed to learn the simple discrimination between C+ and V-trials-that is, rating scores (C -V) ≤ 0-were excluded from the CI analysis (the C + /V-discrimination was fully counterbalanced for stimulus identity for this purpose).

Procedure
Participants were invited to take part in a research study on learning using a computer-based task. Before the task, each participant had to read the information sheet and sign a consent form.
Note: CS = conditioned stimulus. Letters denote the 9 conditioned stimuli (pictures of Lego blocks); these were counterbalanced (see text). With respect to unconditioned stimulus presentations, " + " represents presentation of a positive IAPS (International Affective Picture System) picture and "-" presentation of a neutral IAPS picture.
The task instructions were that a cat "Mogwai" would bring participants either a positive picture or a neutral, boring picture, depending on what kind of Lego blocks she found in her basket.
Participants were asked to guess or predict what kind of picture would follow presentation of the Lego blocks using a rating scale from 1 (neutral) to 9 (positive), with the rating 5 to reflect uncertainty. Reminder instructions were presented on screen at each stage of the procedure. Before the start of the preexperimental phase, participants were shown some example CSs and USs, and further explanation was given as necessary. The samples of CS and US pictures were individually colour printed on a 4.5 × 6-cm card, and these pictures were representative of, but not subsequently used as, stimuli during the experiment. Participants were told that the whole computerbased experimental session would last about 20 min and comprise three stages. At the same time, they were shown an example of CS presentations with the rating scale and were told that during the experiment they would need to click the corresponding number to guess or predict the valence of the US (a positive or a neutral picture) according to the different Lego blocks that had been presented. Participants were encouraged to ask questions at this stage. The three stages of the computerbased experimental session then followed.
Preexperimental stage. During the first (preexperimental) stage of the experiment, participants were told they must guess what kind of picture the cat might bring based on the Lego blocks presented, although the instructions specified that no pictures would follow. After participants clicked on a number button to guess the US valence, the next presentation followed in a semirandom sequence. In this and all subsequent stages of the experiment, CS presentations were counterbalanced for right/left position on the screen across participants, and the various trial types were presented in a semirandom sequence (constrained by the total number of trials of a particular type scheduled in each stage). There was a total of 16 presentations, 2 of each stimulus or stimulus combination presented (these being A, C, AZ, AP, BX, CY, CP, and CX; see Table 2).
Training stages. On completion of the preexperimental ratings, the conditioning trials commenced, and US presentations were introduced. The instructions were as before, but with the exception that participants were advised that following their guess they would be shown the picture that the cat had brought. The first training stage used the CS elements and comprised six training blocks, each with two of each of the four kinds of trial (A + , U-, V-, and C+). After the participant had clicked a number button to predict the valence of the US to follow, a US, randomly selected from the pool of positive or neutral USs as appropriate, was shown on the screen for 1 s. This was followed by a 1-s gap before the next trial started, during which a picture of the cat Mogwai (around 6 × 6 cm) was presented in the middle of the screen on a white background (see Supplementary Materials Online). The second, compound training stage followed directly after this training with the CS elements and comprised four kinds of trial (AZ + , AP-, BX-, and CY+). There were 8 training trials with each of the reinforced compounds (AZ and CY) and 12 with each of the nonreinforced compounds (AP and BX). This proportion of reinforced to nonreinforced trial types has produced robust CI in previous studies; the different trial types were analysed in two blocks of 4 and 6, respectively.
Test stage. The test stage was exactly the same as the preexperimental stage, except that the number of presentations of the critical test compounds CP and CX was increased from 2 to 4. According to the CS compound on the screen, participants were asked to predict the valence of the US that would follow. However, no USs were presented in the test stage. As in the earlier stages of the experiment, there were on-screen reminders of the task instructions.
Throughout the experiment, whenever participants asked questions or made comments they were asked to try to focus on the task and to try to remember or guess which outcome (positive or neutral picture) was predicted by the Lego blocks.

Preexperimental stage
The preexperimental stage established that there were no preexisting differences between the two critical stimulus compounds CP and CX. ANOVA with stimulus (CP vs. CX) and group (offender vs. control) as factors revealed no significant effects or interactions, largest F(1, 46) = 2.22, p = .14.
Training stage Excitatory training. In the initial training stage, participants learned a simple discrimination between the reinforced A and C and the nonreinforced U and V; this provided a measure of simple excitatory learning in the two groups. It is evident from Figure 1 that both groups came to respond accurately on this task, although there was some indication that the offender group showed a slightly poorer discrimination, especially at the start of the training phase.
This impression was confirmed by the results of an ANOVA with group, discrimination (A+ vs. U-, C+ vs. V-), reinforcement, and training block. This showed a main effect of reinforcement, F(1, 46) = 260.66, p , .001, r = .92, which interacted with group, F(1, 46) = 4.40, p = .04, r = .295. Simple main effects analysis revealed that the groups differed on reinforced trials, F(1, 92) = 4.14, p = .04, but not on nonreinforced trials, F(1, 92) = 3.08, p = .08. However, the difference between reinforced and nonreinforced trials was significant in both control and offender groups: F(1, 46) = 166.38, and F(1, 46) = 98.68, respectively, ps , .001. Thus, although both groups learned to respond differently on reinforced and nonreinforced trials, the offender group responded slightly less on reinforced trials than did the control participants, which could indicate a general learning impairment in this group.
Reinforcement also interacted significantly with training block, F(5, 230) = 26.69, p , .001, r = .75, confirming that the discriminations developed over the course of training. Simple main effects confirmed that responding on reinforced trials increased, and that on nonreinforced trials decreased-there were significant effects of block on both types of trial, F(5, 460) = 16.24, p , .001, and F(5, 460) = 16.44, p , .001, respectively. In addition, the difference between reinforced and nonreinforced trials was significant on all training blocks, smallest F(1, 276) = 20.32, p , .001. These results demonstrate that the discrimination developed over training-and there was no sign that this differed between the two groups: The interaction between reinforcement, training block, and group was not significant, F , 1.
Significant interactions were also found between training block and discrimination, and between training block, discrimination, and group, F(5, 230) = 2.49, p = .03, r = .27, and F(5, 230) = 2.87, p = .02, r = .33, respectively. However, as none of these interactions involved the factor reinforcement, they did not bear on the ease with which the discriminations were learned and so were not considered further. Nothing else was significant, the largest F(1, 46) = 1.65, p = .21.
Inhibitory training. During the second stage, participants were trained on the key discrimination between the reinforced AZ and the nonreinforced AP, which was designed to turn P into a conditioned inhibitor. They were also trained to discriminate between the reinforced CY and the nonreinforced BX. It is clear from Figure 2 that both groups learned this task, although the offender group showed a less clear discrimination between the two types of trial, especially at the start of training. An ANOVA with group (offender vs. control), and discrimination (AZ+ vs. AP-, and CY+ vs. BX-), reinforcement (reinforced or not), and training block (1-2) as factors, revealed no main effect of either group, F , 1, or discrimination, F(1, 46) = 2.25, p = .14. However, the main effects of training block and reinforcement were significant, F(1, 46) = 5.65, p = .02, r = .33, and F(1, 46) = 168.28, p , .001, r = .89, respectively, as was the interaction between reinforcement and group, F(1, 46) = 11.01, p = .002, r = .44, suggesting that the offenders might have learned less effectively than the control participants. This was confirmed by simple main effects analyses, which revealed that the offenders responded less on reinforced trials and more on nonreinforced trial than did the controls, F(1, 92) = 10.55, p = .002, and F(1, 92) = 5.35, p = .02, respectively. Nonetheless, the difference between reinforced and nonreinforced trials was significant in both groups, F(1, 46) = 46.60 and 132.70 for offender and control groups, respectively, ps , .001, confirming that both groups had learned the discrimination.
A significant interaction was also found between discrimination and reinforcement, F(1, 46) = 19.52, p = .001, r = .55, suggesting there was a difference between the difficulty of the AZ +/ AP-and CY + /BX-discriminations. This might be expected: The nonreinforced AP compound contained element A that was also present on the reinforced AZ trials, whereas the CY + /BX-discrimination did not involve reinforcement of either element of the nonreinforced BX compound. The simple main effects accordingly revealed that ratings were higher on AP-trials than on BXtrials, F(1, 92) = 15.70, p , .001, but that there was no difference in ratings between CY+ and AZ+ trials, F(1, 92) = 2.70, p = .10. However, the difference between reinforced and nonreinforced trials was significant for both discrimination types, F(1, 92) = 64.83, and F(1, 92) = 172.83, for AZ/AP and CY/BX, respectively, ps , .001, suggesting that both discriminations had nonetheless been learned effectively. Nothing else was significant, largest F(1, 46) = 1.92, p = .17.

Test stage
The ratings for the training stimulus compounds were maintained in the test stage-an impression confirmed by statistical analysis-and so these data are not considered further. Figure 3 shows the rating scores for the two critical stimulus compounds, CP and CX, at the preexperimental and test stages. It is clear that during the preexperimental stage, the ratings of the two compounds were similar in both groups, but that at test, although the rating of CP was markedly lower than that of CX in control participants, no such difference was evident in the offender group. This description was confirmed by the results of an ANOVA with group, stage (preexperimental vs. test), and stimulus compound (CP vs. CX) as factors; this revealed a significant three-way interaction, F(1, 46) = 10.03, p , .01, suggesting that the groups differed in the way in which the CP/ CX ratings changed as a result of training. In order to explore this three-way interaction further, two further ANOVAs, with stage and stimulus compound as factors, were performed on the data from each group. In the control group, this revealed no main effect of either stage, F , 1, or stimulus compound, F(1, 23) = 8.00, p = .10, but a significant interaction between these two factors, F(1, 23) = 27.89, p , .001, r = .75. Simple main effects revealed that there was no difference between CP and CX at the preexperimental stage, F(1, 23) = 2.40, p = .14, but that at test, participants gave significantly lower ratings to CP than to CX, F(1, 23) = 30.80, p , .001. The fact that P was able to suppress the positive ratings of C substantially more than was the control stimulus X is consistent with P being a conditioned inhibitor. No such pattern was evident in the offender group: Here the corresponding ANOVA revealed nothing significant, all Fs , 1. There was thus no evidence of CI in the offender group.
Although this apparent abolition of CI in the offenders was a striking effect, one could argue that the offenders were showing signs of impaired learning throughout the experiment, so that this test effect might reflect no more than a general learning deficit. Indeed, there was statistical evidence in the first stage of training that the offenders were worse at simple excitatory conditioning, and the associated effect size was moderate (r = .295). However, this suggestion is weakened by an examination of the individual differences that participants displayed in this task.

Individual differences
In the offender group, there was no significant correlation between any of the IPDE subscales and the summary measure of inhibitory or excitatory learning, largest r(23) = .34, p = .10. This was also true when the IPDE dimensional scores were subgrouped as three clusters according to DSM-IV, largest r(23) = .17, p = .43. Similarly, there was no correlation between PCL-R scores, either overall or by subscale, and the summary measure of excitatory or inhibitory learning, largest r (24) = .29, p = .17, for Factor 2.
However, there did appear to be differences between participants in the DSPD and PD units, despite the fact that levels of learning during the excitatory and inhibitory training phases were relatively similar in these two clinical subgroups (Figures 4  and 5). ANOVA of rating scores from the excitatory and inhibitory training phases, but comparing only the PD or DSPD subgroups, revealed no effects or interactions involving group, smallest p = .14. The test data for the two units are shown in Figure 6 (for comparison with Figure 3). Although the figure suggests that the DSPD showed relatively less inhibition at test than the PD participants, analysing these data in the same manner as those in Figure 3 did not support this conclusion, possibly in part because of the apparent difference between the two groups at the preexperimental stage (though this was not significant, Fs , 1).
However, the additional analyses of the summary scores of excitatory and inhibitory  learning (as used for the other analyses in relation to individual differences) confirmed the impression that the level of CI was indeed further reduced in the DSPD participants. The summary excitatory learning scores (C -V) were 4.05, 3.89, and 5.12 for the PD, DSPD, and control participants, respectively; the corresponding scores for the CX -CP measure of CI were 1.91, -1.03, and 2.69. ANOVA with group (PD, DSPD, and control) and measure (excitatory vs. inhibitory) was performed on these data and revealed a significant interaction between these two factors, F(2, 45) = 10.41, p , .001, r = .43; simple main effects showed that there was an effect of group on the CI measure, F(2, 90) = 11.63, p , .0001, but not on the excitation measure, F(2, 90) = 1.43, p = .25. Tukey's test showed that whereas DSPD differed from the control participants p = .001, PD participants did not, p = .78. A further analysis on these summary scores, but for PD and DSPD groups only, revealed an almost significant interaction, F(1, 22) = 4.18, p = .053, r = .399; simple main effects revealed that the groups differed on the inhibitory but not the excitatory learning measure, F(1, 44) = 6.40, p = .015, and F , 1, respectively.
We therefore conclude that the deficit in CI was greater in the DSPD participants. With respect to the source of the difference in performance between these two subgroups of offenders, the IPDE dimensional scores were not significantly different in the PD and DSPD units on any of the subscales, with one exception: There was a significant difference in avoidant PD between the two units, χ 2 (1, N = 24) = 4.04, p = .05 (PD unit mean scores = 7.13, SD = 4.39; DSPD unit mean scores = 2.75, SD = 3.36). Nothing else was significant, the largest χ 2 (1, N = 24) = 2.20, p = .14. Furthermore, when the IPDE dimensional scores were subgrouped as the three DSM-IV clusters, there was no significant difference between the two units, largest t(22) = 1.84, p = .10. However, consistent with the admissions criteria, participants in the DSPD unit had significantly higher PCL-R scores (by both the individual factors and the total scores) than those in the PD unit, smallest t (22) = 2.53, p = .02, r = .47. Thus, participants with sufficiently high PCL-R scores to warrant DSPD admission showed overall less CI than those with lower PCL-R scores. Moreover, PCL-R Factor 1 and Factor 2, as well as the total scores, were significantly correlated with the dimensional scores for ASPD, r(24) = .53, p = .01, r(24) = .41, p = .04, and r(24) = .61, p = .001, respectively, but not BPD, largest r (24) = .24, p = .27. In addition, although they did not differ in age, t(22) = 0.28, p = .78, the PD group had been incarcerated significantly longer than the DSPD group, t = -2.43, p = .03. This direction of effects is opposite to what we  would expect if incarceration were the cause of the CI deficit.
Thus, differences in ASPD could modulate the level of CI shown in these violent offenders. This suggestion is supported by further analyses using the SVRS. There was no significant difference by the Violence in Criminal Record subscale, t(22) = 1.70, p = .10, or by the Violence Total score, t(22) = 1.32, p = .20. There was a significant difference in participants recruited from the PD and the DSPD units in the level of Violence in Index Offence subscale, t(22) = 2.55, p = .02, but this took the direction that PD unit patients showed greater severity of violence in their index offence (PD unit mean scores = 3.00, SD = 0.54; DSPD unit mean scores = 2.12, SD = 1.15). It was the severity of Violence in the Institution that reflected the dangerousness of the DSPD compared to PD patients, t(22) = 2.19, p = .05 (PD unit mean scores = 1.62, SD = 1.41; DSPD unit mean scores = 2.94, SD = 1.34). With respect to mediation of the CI deficit, scores on the Violence in Criminal Record subscale were positively correlated with the measure of inhibitory learning, r(24) = .41, p = .04. Thus, the higher the violence ratings in relation to criminal record, the worse the CI learning performance.

Discussion
This experiment provided evidence that, while CI was clearly demonstrated in the community control participants, it was effectively absent in a group of offenders, particularly in the DSPD subgroup. The critical test comparison, ratings of CP and CX, involved fully counterbalanced stimulus compounds and took into account preexperimental differences in their ratings. In addition, the training history of P and X was matched in every respect apart from the fact that P signalled the absence of a reinforcing picture, whereas X did not.
There was some evidence of a more general learning deficit in the offender group, so one must consider the possibility that our results do not represent a specific impairment in CI, but a more nonspecific effect on learning. For example, it may be that the positive pictures had less salience for the offender group, which could have lessened both excitatory and inhibitory learning in these participants. However, although we have no direct evidence against this possibility, the impairment in excitatory conditioning in the offender group was statistically less consistent, as well as being numerically more modest, than the impairment in CI, and the CI deficit was significantly greater than the effect on excitatory learning in offenders. This pattern became clearer when the PD and DSPD subgroups were considered separately: Here analysis of the summary measures of excitation and inhibition showed clearly that, while excitatory learning did not differ among the three groups, inhibitory learning did, and this difference stemmed from a selective deficit in inhibitory learning in the DSPD group. This pattern of results makes it unlikely that the inhibitory learning deficit was an indirect result of a more general learning deficit.
A curious feature of the results was that performance in the inhibitory learning phase appeared to be less impaired than performance on the summation test. This was especially evident in the comparison of the PD and DSPD groups, who showed essentially identical performance throughout training, but differed significantly at test. It might be that the critical difference between these two groups had less to do with inhibitory learning than with a differential tendency in the DSPD participants to process compound cues in a configural manner-such that they were unable to identify, or respond appropriately to, elements of the compound when they were presented in a different compound, as occurred in the summation test (cf., Pearce, 1987). Unfortunately at present we do not have any data that would allow us to discriminate between these two possibilities.
Any firm conclusion that CI is impaired in this forensic population of course depends on the adequacy of the control condition. Although the offenders all had PD, while the community control participants did not, it must be acknowledged that the differences between forensic patients in a high-security hospital and healthy community controls are doubtlessly vast across many other social, economic, and psychological dimensions. Nonetheless, control participants were deliberately matched as closely as possible with offenders in terms of general factors, including educational level and socioeconomic status, and participants who did not learn in the first training stage were excluded. Moreover, the inevitable limitations of the control condition were mitigated by the fact that the inhibitory learning deficit was relatively confined to the DSPD subgroup of offenders. Still, differences in general intelligence or motivational factors among the groups cannot be ruled out; however, although such differences might be able to account for a generally different level of performance, it is less clear how they could explain the relatively specific deficit in CI.
It is also important to consider the possibility that medication might be sufficient explanation of the loss of CI in the offenders, as a relatively large proportion of this sample, especially of the DSPD participants, was on benzodiazepines or antipsychotics of some description, and such medications generally impair cognitive function. Moreover, we have previously found that CI was impaired under medication with clonidine in a subset of participants with Tourette syndrome, even though there was no overall impairment in the diagnosed group (Kantini et al., 2011). Nonetheless, a number of arguments may be made against the possibility that medication mediated the difference in CI observed in the present study. First, statistical analysis did not reveal any difference in performance on either the excitatory or inhibitory tasks according to medication, either in general or by specific type. Second, the diversity of medication regime makes it unlikely that this could provide a systematic account of the failure of offenders to show CI. Nonspecific effects of medication-for example, on arousal, attention, or motivation to engage with the task-would be expected to depress performance throughout, and this was not the pattern of effects observed in the present study. For example, as noted above, by the end of training the offender group was performing as well as controls on the inhibitory discrimination, and a clear deficit only emerged in the transfer test.
Finally, there are other confounded factors that could contribute to the observed difference in CI. For example, the controls could not be matched for incarceration. Because of its admissions criteria, it was not possible-within the same institutionto recruit offenders without PD, psychiatric illness, or learning disability, as a further comparison group to allow us to examine the role of incarceration in a secure hospital. Nonetheless, given the significant difference in CI between the participants from the PD and DSPD units, we would argue that incarceration per se does not seem to be the critical issue-especially as the DSPD group, who showed worse CI than the PD group, had been incarcerated for less time, rather than more. We would suggest that a more natural conclusion is that inhibitory learning deficits may contribute to the cognitive profile of an individual whose behaviours result in incarceration-rather than suggest that incarceration per se has a selective cognitive effect. In summary, the pattern of results obtained is not obviously explicable in terms of the nature of our control condition.
It is not immediately apparent how such a selective effect on expression of CI in our offender group might be mediated. We have previously identified a role for individual differences in CI (using a different variant of the CI task) in a larger sample of normal participants (Migo et al., 2006), although the scales used in this earlier study measured variation in normal function, and so some discontinuity can be expected in the socially deviant sample who participated in the present study (Howard & Duggan, 2009). However, although a high proportion of the sample had confirmed ASPD and/ or BPD according to the IPDE categorical diagnoses, there was no correlation between any of the IPDE dimensional measures and CI scores.
Similarly, there was no correlation between CI scores and psychopathy levels as measured by the PCL-R, either overall or by either of the subscales (Hare, 1991;Lykken, 1995). Certainly one should not place too much weight on this null result as, in common with other experimental studies of this type, the sample size was underpowered for such correlational analyses, and so the failure to find correlations here does not mean none could be found in a larger study.
Moreover, direct comparison of the PD and DSPD subgroups, using the same summary learning scores, showed that the DSPD group showed worse CI than the PD participants. Critically, these subgroups of offenders differed significantly in their PCL-R scores, an index of psychopathysuggesting that the CI impairment could stem from this difference in the personality profile of these two participant groups. But it is possible that the true picture is more complex. DSPD unit patients are typically characterized by the co-occurrence of high PCL-R scores and an ASPD diagnosis and, frequently, a BPD diagnosis (Howard & Duggan, 2009). This pattern of comorbidity is associated with a significant degree of serious, in particular violent, offending (Coid & Ullrich, 2010;Kosson et al., 2006) and with high scores on a dimension of hostile impulsivity, characterized by aggression, resentment, deviance, and paranoid beliefs together with affective dyscontrol (Blackburn, 2009). This may provide a further explanation of the lack of correlation between the CI deficit and the PCL-R scores taken in isolation, since it is the co-occurrence of psychopathy with ASPD (and frequently BPD as well) that characterizes these deviant and disinhibited patients, rather than simply a high PCL-R score. Moreover, patients in the DSPD unit were found to have significantly lower ratings of avoidant personality disorder-that is, less social inhibition, feelings of inadequacy, hypersensitivity to negative evaluation, and avoidance of social interaction. To the extent this constellation of traits relates also to levels of cognitive inhibition, this profile is consistent with the finding of the present study, that their CI was relatively worse than that demonstrated in the PD unit.
We therefore conclude that the demonstration that the participants in the DSPD showed less CI than those in the PD unit is consistent with a relationship between CI and the level of deviancy and disinhibition within the offender group. In this sense it is significant that higher PCL-R scores, as was found in the DSPD patients, went together with elevated ASPD but not BPD scores, suggesting that higher ASPD could also be driving the poorer CI in the DSPD group. Conversely a diagnosis of psychopathy did not affect the level of CI in participants with comorbid BPD/ASPD, making it less likely that psychopathy in isolation was a critical factor. Importantly, correlational analyses showed that the higher the SVRS scores in relation to criminal record (Gunn & Robertson, 1976;Wong et al., 1993), the worse the CI learning performance. Although not a direct measure of personality, this finding points to the role of long-standing individual characteristics in relation to CI deficits.
Irrespective of how the effect was mediated, in terms of implications for our understanding of cognitive control in offenders, the loss of CI reflects a selective effect on an aspect of associative learning that could potentially be relevant to both the symptom profile and the ability to control unwanted behaviours that have identified triggers. Stimuli provided by environmental events are recognized to trigger associations that generate habitual thoughts and feelings (Ferguson & Cassaday, 1999;Lishman, 1987;Siegel, 1977;Stewart, de Wit, & Eikelboom, 1984;Watson, 1924), which can in turn generate unwanted behaviours. Similarly, other environmental events and thoughts provide contextual stimuli with the potential to become learned inhibitors of these unwanted associations. However, in offenders, precisely where the need to inhibit such associations is greater, impaired inhibition of responding based on S-S contingencies may leave some individuals less able to inhibit the unwanted thoughts that can lead to unwanted actions. Thus, therapeutic interventions to improve learning by inhibiting antecedent S-S associations could be an effective behavioural approach to symptom control (via an action on the associative chain that generates triggers).
It should be noted that the S-S association being inhibited in the present study was between the Pavlovian CS and a motivationally significant outcome (a positive picture) rather than a specific image: The inhibited CR was elicited by a stimulus that had predicted a variety of different images that shared only their positive motivational value. We are also unable to specify exactly what the participants learned in order to be able to inhibit their behaviour. For example, the CI could have suppressed the ability of the conditioned excitor to elicit expectation of the outcome, but it might equally have suppressed performance of the conditioned response directly (Rescorla, 1993;Rescorla & Holland, 1977). The present study is not able to discriminate between these, or other, possibilities. But the critical point is that, by whatever mechanism the response was suppressed, it was the product of learning based on Pavlovian S-S contingencies, not an operant response that had been rewarded directly. This critical theoretical distinction sets the present task apart from other tests of impulsivity and broadens the operational and theoretical criteria by which disorders in impulsivity may be identified.
Theoretical accounts of ASPD, BPD, and psychopathy do not specifically address CI. However, the results of the present study are broadly consistent with predictions arising from a model of deficits in emotionally impulsive PD (Howard, Fenton, & Fenwick, 1982). Although we find CI deficits in association with offending behaviour and level of violence in participants' criminal record rather than psychopathy specifically, the results of the present study are similarly consistent with a recent reformulation of the response modulation deficit hypothesis of psychopathy (Gorenstein & Newman, 1980;Vitale & Newman, 2009). Newman and colleagues have suggested that the insensitivity to inhibitory cues that characterizes psychopathy is a result of anomalous attentional modulation of responding (Newman & Lorenz, 2003). This attentional anomaly has the consequence that contextual information has less effect on prepotent S-R responding (Newman & Wallace, 1993;Patterson & Newman, 1993;Zeier, Maxwell, & Newman, 2009). We already know that bad actions are a defining feature of offenders; the present study provides a direct investigation of the cognitive mechanisms necessary to dampen certain kinds of trigger. Specifically, the results of the present study show that the contextual information provided by conditioned inhibitors has virtually no effect on the expression of prepotent associations based on Pavlovian S-S contingencies.
Original manuscript received 21 February 2011 Accepted revision received 5 August 2011 First published online 18 November 2011