When an individual resists the temptation to stay out late in order to get a good night’s sleep, he or she is exercising what is known as “goal-directed control”. This kind of control allows individuals to regulate their behaviour in a deliberate manner. It is thought that a reduction in goal-directed control may be linked to compulsiveness or compulsivity, a psychological trait that involves excessive repetition of thoughts or actions. Furthermore, evidence shows that goal-directed control is reduced in people with compulsive disorders, such as obsessive-compulsive disorder (or OCD) and drug addiction. However, failures of goal-directed control have also been reported in other mental health conditions that are not linked to compulsivity, such as social anxiety disorder.

The fact that reduced goal-directed control is found across various mental health conditions highlights a core issue in modern psychiatric research and treatment. Mental health conditions are typically defined and diagnosed by their clinical symptoms, not by their underlying psychological traits or biological abnormalities. This makes it difficult to determine the cause of a specific disorder, as its symptoms are often rooted in the same psychological and biological traits seen in other mental health conditions.

To start to tackle this issue, Gillan et al. used a strategy that allowed them to look at compulsivity as a “trans-diagnostic dimension”; that is, as something that exists on a spectrum and is not specific to one disorder but involved in numerous different mental health conditions. Nearly 2,000 people completed an online task that assessed goal-directed control, and filled in questionnaires that measured symptoms of various mental health conditions. Gillan et al. showed that, as expected, people with reduced goal-directed control were generally more compulsive, and that this relationship could be seen in the context of both OCD and other compulsive disorders such as addiction.

Further, by leveraging the efficiency of online data collection to collect such a large sample, Gillan et al. were also able to examine how much different symptoms co-occurred in people. This enabled them to use a statistical technique to pick out three trans-diagnostic dimensions – compulsive behaviour and intrusive thought, anxious-depression and social withdrawal – and found that only the compulsive factor was associated with reduced goal-directed control. In fact, reduced goal-directed control was found to be more closely related to compulsivity than the symptoms of traditional mental health disorders including OCD. These findings show that research into the causes of mental health conditions and perhaps ultimately diagnosis and treatment – all of which have traditionally approached specific disorders in isolation – would benefit greatly from a trans-diagnostic approach.

Compulsivity is a theoretical clinical phenomenon that reflects the loss of control over repetitive self-deleterious behavior seen in a range of disorders, most notably obsessive-compulsive disorder (OCD) and addiction (Everitt and Robbins, 2005; Gillan and Robbins, 2014). But what are the underlying neural, computational, or psychological mechanisms? Researchers have suggested that compulsivity in these disorders may be partially explained by an imbalance between two different modes of control, which are more and less flexible (Everitt and Robbins, 2005; Graybiel and Rauch, 2000). In particular, a deficit in deliberative, ‘goal-directed’ control may leave individuals vulnerable to rely excessively on forming more rigid habits. Habits are behaviors that animals and humans learn to execute automatically when presented with familiar environmental cues (Dickinson, 1985). While habits are typically very useful, allowing us to efficiently perform routine actions while expending minimal cognitive effort, they cannot adapt flexibly to new situations. To override our habits, organisms are capable of ‘goal-directed behavior’. This refers our ability to make more considered choices, reflecting both (i) knowledge of the outcomes that our actions typically produce and (ii) our current motivation for those outcomes (Dickinson and Balleine, 1994). Consistent with the hypothesis that compulsion is linked to an imbalance between these modes of control, deficits in goal-directed learning have been observed across a range of putatively compulsive disorders such as drug addiction (Sjoerds et al., 2013; Voon et al., 2015), obsessive-compulsive disorder (OCD) (Voon et al., 2015; Gillan et al., 2011; 2014a; 2014b; 2015a) and also binge-eating disorder (Voon et al., 2015). These deficits in goal-directed control have been linked to abnormal structure and function of the caudate and medial orbitofrontal cortex (Voon et al., 2015; Gillan et al., 2015a), suggesting that they may be a promising target for understanding the etiology of these disorders and thus for future treatment development.

Critically, the scope of the relationship between goal-directed learning deficits and psychopathology, and particularly their specificity to compulsive versus non-compulsive aspects has not been established. In fact, a similar deficit in goal-directed control was recently reported in other patient groups (Alvares et al., 2014; Morris et al., 2015), including those diagnosed with social anxiety disorder and schizophrenia, at least the former of which is not characterized by repetitive compulsive acts. This casts serious doubt over the hypothesis that goal-directed deficits are a neurocognitive mechanism that is partly responsible for psychiatric compulsivity. This lack of specificity is unfortunately ubiquitous in psychiatry research (Lipszyc and Schachar, 2010; Bickel et al., 2012), a result, we suggest, of the broader issue that psychiatric diagnostic categories do not reflect the most discrete and neurobiologically informative phenomena. Of particular relevance to the present study are the high rates of co-morbidity between OCD and social anxiety disorder (Ruscio et al., 2010), the preponderance of OCD symptoms in the schizophrenia poopulation (Poyurovsky and Koran, 2005), and more broadly that the vast majority of patients diagnosed with obsessive-compulsive disorder (OCD) meet the criteria for another lifetime psychiatric disorder (Ruscio et al., 2010). Given these major overlaps, dissociating the neurocognitive bases for these respective diagnostic categories in their current form may be untenable.

Indeed, the Diagnostic and Statistical Manual of Mental Disorders (DSM), now in it’s fifth edition (American Psychiatric Association, 2013) was developed to provide a reliable, descriptive psychiatric taxonomy, rather than an etiologically valid one. As such it is difficult to clearly discriminate the diagnostic categories it defines on the basis of genetics, neuroimaging, or indeed any of the modern tools of cognitive neuroscience. These issues have been described in detail by others (Cuthbert and Kozak, 2013; Hyman, 2007; Robbins et al., 2012), and have been recognized by the National Institute of Mental Health (NIMH), which has launched the Research Domain Criteria (RDoC) initiative, aiming to identify biologically plausible, trans-diagnostic markers of psychiatric disturbances (Insel et al., 2010). Although progress towards this goal has already been made by studies examining dissociable clusters of patients within groups diagnosed with the same disorder (Brodersen et al., 2014; Fair et al., 2012), the identification of robust, generalizable and specific markers that contribute to psychiatric co-morbidity has been curtailed by the small sample sizes that are typical of patient studies.

Accordingly, we hypothesized that a dimensional approach leveraging the efficiencies of large-scale online data collection among healthy individuals could be used to determine the precise psychiatric phenotype associated with deficits in goal-directed control, and test the specificity of this relationship with respect to other aspects of psychopathology. We hypothesized that this phenotype would broadly relate to compulsive behavior, which is seen across multiple disorders, including OCD and addiction (Gillan et al., 2015b), but were interested to reveal the scope and generality of this, e.g. with respect to impulsivity, a putatively related clinical phenotype (Robbins et al., 2012). We also wished to study how any psychiatric correlates of goal-directed control relate to variation in age and IQ, two more general factors that have been shown to covary both with goal-directed control and with some aspects of psychopathology (Eppinger et al., 2013; Schad et al., 2014; Sandstrom et al., 1998). To this end, rather than diagnosed patients, we used two large general-population samples collected online via Amazon’s Mechanical Turk (AMT) to test (i) if compulsivity as indicated by self-report OCD symptoms is associated with individual differences in goal-directed learning, (ii) if this association generalizes to self-report symptoms of other DSM diagnostic categories that involve compulsivity, and (iii) if this association is specific to compulsive versus non-compulsive psychopathology.

Goal-directed control has recently been computationally formalized as arising from a form of reinforcement learning known as ‘model-based’ (Daw et al., 2011), which can be expressed as an individual difference measure that has been shown to predict how likely individuals are to form habits (Gillan et al., 2015c). Using this well-validated task (Figure 1) (Daw et al., 2011), we found support for all three postulates. In Experiment 1, we found that total scores on a self-report questionnaire measuring the severity of OCD symptoms were tracked by normal variation in model-based learning in the general population, but not by self-report anxiety or depression symptoms. In Experiment 2, we replicated the association with self-report OCD symptoms and showed that it generalized to a broader set of psychiatric symptoms that similarly involve failures in exerting control over self-deleterious behaviors, specifically alcohol addiction, eating disorders and impulsivity. Once again, we found tentative evidence for specificity with respect to non-compulsive aspects of psychopathology. Next, we conducted a factor analysis, which indicated the existence of three latent symptom dimensions that cut across the nine different questionnaires assessed in this study. Crucially, the second symptom dimension identified was characterized by ‘Compulsive Behavior and Intrusive Thought’, in which items were most consistently drawn from the questionnaires assessing symptoms of OCD, eating disorders and addiction, pertaining not just to repetitive compulsive behaviors (as was our prediction), but also to associated preoccupations and cognitive distortions. This factor, which was defined independently of task performance, was a significant predictor of deficits in model-based learning. Crucially, this effect was highly specific to this factor, when directly compared to the two other factors identified in this analysis, ‘Anxious-Depression’ and ‘Social Withdrawal’.

Figure 1

Download asset Open asset

In Experiment 1, we tested the hypothesis that individual differences in total scores on a questionnaire assessing the severity of OCD symptoms are associated with normal variation in goal-directed control, rather than necessitating the categorical comparison of OCD patient vs. control groups. Participants (N = 548) first completed a reinforcement-learning task that quantifies individual differences in goal-directed (‘model-based’) learning, which is operationalized as a parameter estimate from a logistic regression analysis predicting choices in the task (see Materials and methods and refs [Daw et al., 2011; Gillan et al., 2015c]). Next, we administered a short Intelligence Quotient (IQ) test, followed by self-report questionnaires assessing symptoms of OCD, along with depression and trait anxiety, which we did not expect to be associated with goal-directed deficits. In line with our hypothesis, there was a significant association between scores on the OCD questionnaire and goal-directed deficits (i.e. a negative relationship between OCD severity and model-based learning; β = −0.040, Standard Error (SE) = 0.02, p=0.049) when (as in all analyses reported henceforth) controlling for age, IQ and gender, which have been previously reported to covary with goal-directed behavior (Eppinger et al., 2013; Schad et al., 2014; Sandstrom et al., 1998). Specifically, for each increase of 1 standard deviation (SD) in the total score on the OCD questionnaire, model-based learning was reduced by 14%. No such relationship was observed for self-report depression (β = −0.016, SE = 0.02, p=0.439) or trait anxiety (β = −0.006, SE = 0.02, p=0.777) severity (Table 1, Figure 2A). Moreover, the relationship between total scores on the OCD questionnaire and goal-directed deficits survived inclusion of the depression and trait anxiety total scores in the same model as covariates (β = −0.048, SE = 0.02, p=0.04). These data indicate that deficits in goal-directed control are a marker of normal variation in OCD symptomatology in the general population.

Figure 2

Download asset Open asset

In Experiment 2, we aimed to test the reliability, generalizability and specificity of this finding in a larger cohort of task-naïve subjects (N = 1413, based on a power analysis given the aforementioned results). The procedure was identical to that in Experiment 1, except for the addition of several more clinical questionnaires. To test for generalizability, we assessed symptoms associated with other disorders that have been hypothesized to have compulsive features. In addition to the OCD questionnaire used in Experiment 1, these pertained to alcohol addiction, eating disorders, along with aspects of impulsivity and schizotypy (Everitt and Robbins, 2005; Poyurovsky and Koran, 2005; Robbins et al., 2012; Godier and Park, 2014). To test for specificity, in addition to the mood symptoms assessed previously (depression and trait anxiety) we added self-report measures assessing social anxiety and apathy; we also predicted that non-compulsive aspects of schizotypy and impulsivity might not be associated with goal-directed control. In this independent sample, we replicated the results from Experiment 1; scores on the OCD questionnaire were significantly associated with goal-directed deficits (β=−0.026, SE=0.01, p=0.020), while controlling for age, gender and IQ (Table 1, Figure 2B). As we hypothesized, this effect generalized to phenotypically disparate manifestations of psychiatric compulsivity: total scores on self-report measures of eating disorder severity (β=−0.037, SE=0.01, p<0.001), impulsivity (β=−0.034, SE=0.01, p=0.007) and alcohol addiction (β=−0.025, SE=0.01, p=0.029). Also as predicted, we found no significant associations between goal-directed deficits and total scores on the depression (β=−0.01, SE=0.01, p=0.385), apathy (β=−0.001, SE=0.01, p=0.953), trait anxiety (β=−0.008, SE=0.01, p=0.498) or social anxiety (β=0.008, SE=0.01, p=0.496) questionnaires. We found no significant association between self-report levels of schizotypy and goal-directed control (β=−0.017, SE=0.01, p=0.14), possibly reflective of the great deal of heterogeneity within this particular psychiatric construct.

Previous studies using this task have assessed an individual’s goal-directed learning in either of two ways: predicting their choices using either a regression model (as reported above) or the fit of a more elaborate computational learning model, which the regression model approximates. In separate analyses using fits of the computational learning model (see Materials and methods; Supplementary file 5A), all of the aforementioned results were recapitulated, with the exception that the relationship between OCD and model-based learning in Experiment 1 fell short of significance (but was significant in Experiment 2) and schizotypy reached significance in Experiment 2 as a negative predictor of model-based learning.

Given both the heterogeneity within, and the high correlation across these questionnaires (e.g., Depression and Trait Anxiety scores correlate at r=0.81) these questionnaires, assessing the statistical specificity of these effects by including their total scores in the same model is both methodologically and conceptually fraught. To address this issue, we conducted a factor analysis based on the 209 individual questionnaire items, thereby reducing the collinearity across scores on these psychiatric questionnaires. Note that this analysis was carried out on the questionnaire scores alone, without reference to the results on the reinforcement learning task. We found evidence for three dissociable factors (‘dimensions’) that cut across the nine questionnaires from which items were drawn, which we labeled ‘Anxious-Depression’, ‘Compulsive Behavior and Intrusive Thought’ and ‘Social Withdrawal’, based on the loadings of individual items (Supplementary file 2A–C, Table 2, Figure 3A). Although the labeling of factors is of course a subjective process, quantitatively speaking, ‘Compulsive Behavior and Intrusive Thought’ had high and consistent loadings from almost all items pertaining to eating disorders (Mean loading=0.36, SD=0.15), OCD (Mean loading=0.50, SD=0.06) and addiction (Mean loading=0.31, SD=0.07), which have all been couched as ‘compulsive’ disorders in the literature (Everitt and Robbins, 2005; Gillan and Robbins, 2014; Godier and Park, 2014) (Table 2). In addition to picking up every self-report item that pertained to compulsive behavior from our question pool, the loadings on Factor 2 were not confined to compulsive behaviors, but equally featured items pertaining to related patterns of thought, i.e. obsessions, preoccupations, or intrusive thoughts. We cannot speak to causality here, but this suggests that repetitive behavior and repetitive, irrational patterns of thought are not orthogonal symptom dimensions, but perhaps share a common neurobiological root. Items from the impulsivity scale (of which the total score was a significant predictor of goal-directed deficits) did not load as strongly or consistently on this factor (M=0.15, SD=0.15; significantly less than the former three questionnaires, Eating Disorders vs. Impulsivity: t(52)=5.178; OCD vs. Impulsivity: t(41)=11.379; Alcohol Addiction vs. Impulsivity: t(33)=4.342, all p<0.001) (Table 2).

Figure 3

Download asset Open asset

We next tested for an association between subjects’ scores on these three factors and their, separately measured, goal-directed performance. When tested alone, ‘Compulsive Behavior and Intrusive Thought’ was significantly associated with deficits in goal-directed learning (β=−0.046, SE=0.01, p<0.001), and this effect size was greater than that of any of the questionnaires used in this study, corresponding to a 17% reduction in model-based learning for an increase of 1 SD in ‘Compulsive Behavior and Intrusive Thought’ (Figure 3B, Table 3). There were no significant effects of Factor 1 (β=−0.001, (0.01), p=0.92) or Factor 3 (β=0.013, SE=0.01, p=0.24) on model-based learning. Finally, we directly compared the associations between goal-directed deficits and these factors by including them in the same model and conducting planned contrasts. We found that deficits in goal-directed control were highly specific to the ‘Compulsive Behavior and Intrusive Thought’ (vs. ‘Anxious-Depression’, β=−0.062, SE=0.02, p=0.001; vs. ‘Social Withdrawal’, β=−0.089, SE=0.02, p<0.001). Moreover, when included in the same model with the other factors, ‘Social Withdrawal’ (onto which addiction and aspects of impulsivity load negatively) emerged as a significant positive predictor of goal-directed control over action (β=0.031, SE=0.01, p=0.014). To test the extent to which the relationship between goal-directed deficits and ‘Compulsive Behavior and Intrusive Thought’ is truly continuous, we carried out a supplementary analysis in which this factor was entered as a quadratic term in our model, thereby testing for a nonlinear effect. We found no evidence for nonlinearity (beta=−0.0016, p=0.822), and the linear effect remained significant when included in this model (beta=−0.045, p=0.001). Similarly, we repeated our analyses in subsets of our population comprising either ‘putative patients’ (defined as those who scored in the top 25% on a given self-report measure) or subjects in the normal range (bottom 75%) and the results were broadly consistent across sub-samples (Materials and methods, Supplementary file 3).

Finally and in complement to the unsupervised factor analysis used to define ‘Compulsive Behavior and Intrusive Thought’, we carried out a fully supervised analysis (regression with elastic net regularization) to identify directly from the individual questionnaire items those most predictive of goal-directed learning, as assessed using the regression model. Supporting our previous conclusions, those items that predicted model-based deficits in the negative direction substantially overlapped with items with above-threshold loadings on ‘Compulsive Behavior and Intrusive Thought’ (75% overlap; Supplementary file 4). One noteworthy pattern arises among the exceptions. The supervised analysis also identified several additional items from the impulsivity questionnaire, which had not loaded on ‘Compulsive Behavior and Intrusive Thought’, but did predict goal-directed learning. Those were items that tracked subjects’ motivation to engage with the experimental paradigm, e.g. “I (do not) like to think about complex problems”. Other, more compulsivity-relevant items from the impulsivity scale, involving compulsive shopping and general loss of control over action, were identified in both analyses. The former items are likely of little clinical relevance, but can explain the strong association between impulsivity total scores and goal-directed deficits, despite the fact that impulsivity did not load strongly onto ‘Compulsive Behavior and Intrusive Thought’.

In addition to tracking one well-delineated aspect of psychopathology, we found that task performance was significantly related to other measures collected in this study. First, although individual variation in ‘model-free’ performance on the learning task did not track any of the scores from our psychiatric questionnaires (Supplementary file 1C), in Experiment 1, model-free performance did relate significantly to age (Supplementary file 1B, Reward*Age interaction). ‘Model-based’ learning was also related to age and IQ. In particular, higher IQ was associated with increases in goal-directed, ‘model-based’ learning. In contrast to the effect of age on ‘model-free’ learning, older people were significantly less ‘model-based’ compared to their younger counterparts. All of these results were replicated in Experiment 2. Additionally, the larger sample size in Experiment 2 allowed us to detect small but significant associations between gender and model-free and model-based learning. Males were significantly less model-free and more model-based relative to females tested in this study. Importantly, all of these effects are controlled for (by including age, IQ, and gender as additional covariates) in the analyses relating learning to psychiatric symptoms.

Here, we tested the utility of a dimensional approach to investigating the neurocognitive basis of compulsivity using two large-scale general population samples. Evidence from multiple complimentary analyses supported the conclusion that ‘Compulsive Behavior and Intrusive Thought’ is a symptom dimension associated with deficits in goal-directed control that links features of multiple psychiatric disorders; most notably symptoms of OCD, addiction, and eating disorders. Interestingly, this dimension goes beyond the uncontrolled behaviors that have been previously associated with compulsivity, to include obsessions, preoccupations and intrusive thoughts.

That self-report scores of OCD and addiction symptoms were associated with these deficits is consistent with previous research in patient populations (Sjoerds et al., 2013; Voon et al., 2015; Gillan et al., 2011; 2014a; 2014b; 2015a), and extends these results for the first time to a general population sample. Likewise, binge-eating disorder has also been previously associated with reduced goal-directed control in one patient study and an animal model (Voon et al., 2015; Furlong et al., 2014). Critically, the results of the present study extend this finding to self-report symptoms of other subtypes of eating disorders, suggesting that Compulsive Behavior and Intrusive Thought (and associated deficits in goal-directed control) are a key component of more aspects of eating disorders than previously documented. An entirely consistent exception was that items relating to exerting control over food intake (e.g. “I display self-control around food”) did not load strongly on the ‘Compulsive Behavior and Intrusive Thought’ factor.

A previous study reported an association between social anxiety disorder and deficits in goal-directed control (Alvares et al., 2014). Using self-report social anxiety symptom scores in our general population sample, we did not replicate this finding, and in fact observed a trend towards enhanced goal-directed control associated with social anxiety symptoms. Specifically, in most analyses social anxiety symptoms (both total scores and the ‘Social Withdrawal’ factor) was unrelated to task performance. We did however observe a significant positive association between the ‘Social Withdrawal’ factor and goal-directed control in one analysis, while controlling for the other factors in the same analysis. This result should be interpreted with caution, given that the association was not sufficiently robust to predict goal-directed control alone, but this serves to illustrate that ‘Social Withdrawal’ trended towards predicting better goal-directed control, not worse. Two explanations for the discrepant findings between the present study and the prior investigation with diagnosed social anxiety disorder patients are the differences in sample size between our respective studies and that the co-morbidities reported for the social anxiety disorder population of the study by Alvares and colleagues (2014) could not be controlled for and may have driven the reported association. This underscores the importance of a dimensional approach to psychiatric phenotyping.

Schizophrenia has also been previously associated with deficits in goal-directed control (Morris et al., 2015), a finding that was partially supported by the present study (to the limited extent that ‘schizotypy’, measured here, has implications for schizophrenia as a clinical condition). Consistent with the heterogeneous nature of schizophrenia, where two diagnosed patients can have entirely non-overlapping symptoms (American Psychological Association, 2013), we did not find a significant association between the total score on the schizotypy questionnaire and deficits in goal-directed control (although this was significant in a second analysis based on a full computational model). However, using our trans-diagnostic approach, we found that in particular ‘unusual experiences’ characteristic of schizotypy loaded onto the ‘Compulsive Behavior and Intrusive Thought’ factor, which in turn was a strong predictor of goal-directed deficits. This finding converges with studies highlighting that delusions are more closely linked to executive deficits than the negative symptoms of schizophrenia (Lysaker et al., 1998; 2003). In terms of clinical phenomenology, schizophrenia and OCD share a common pattern of abnormal beliefs and as DSM-5 and others have noted, the distinction between a delusion in schizophrenia and a strongly held belief in OCD is often blurred (Poyurovsky and Koran, 2005; American Psychological Association, 2013). These data suggest that ‘Compulsive Behavior and Intrusive Thought’, which comprises automatic behaviors as well as associated repetitive thoughts, may be common to both schizophrenia and OCD and explained by deficits in goal-directed control.

Earlier work investigating deficits in goal-directed learning in compulsive patient populations did not employ a positive clinical control (Voon et al., 2015), therefore until now the possibility that goal-directed deficits were non-specific, i.e. evident in all psychiatric populations, remained untested. For instance, prior studies have found a consistent association between stress and goal-directed learning deficits (Otto et al., 2013; Schwabe and Wolf, 2009), which might in principle mediate non-specific effects due to the considerable burdens of mental illness. Here, we tested this possibility rigorously in two independent samples. We found no association between ‘Anxious-Depression’ and deficits in goal-directed control, and moreover the specificity of goal-directed deficits to ‘Compulsive Behavior and Intrusive Thought’ was confirmed through direct statistical comparisons.

Prior work has shown that the model-based learning deficits predict the presence of habits using a devaluation probe (Gillan et al., 2015c), providing a tentative mechanism through which the goal-directed deficits observed in the present study might cause the development of compulsive behaviors. Indeed, this converges with prior work showing that when OCD patients are performing habits, they show dysfunctional hyperactivity in the caudate (Gillan et al., 2015a), a region associated with goal-directed control over behavior (Dolan and Dayan, 2013). An outstanding question, however, is the extent to which excessive stimulus-response habit learning also contributes to Compulsive Behavior and Intrusive Thought. The model-free component of the task we employed in the present study did not relate significantly to psychiatric symptomatology, as indeed we had hypothesized because it also does not appear to be sensitive to slow habitual learning (indeed, unlike the model-based component of the task, it does not predict devaluation [Gillan et al., 2015c]). Future work is needed to develop a computational marker of individual differences in stimulus-response habit formation, so that this possibility can directly be tested.

Another interesting question that emerges from these data is how deficits in goal-directed control might result in both cognitive distortions (which take the form of obsessions in OCD, such as a fear of germs) and compulsive behavior (e.g. repetitive hand-washing), which our factor analysis suggested are inextricably linked. One possibility was raised by a recent study, which demonstrated that just like low-level stimulus-response behaviors, more abstract goal selection can also be rendered habitual (Cushman and Morris, 2015). If these habitual cognitive actions can be conceived as a sort of ‘habit of thought,’ this might indicate a common mechanism for both compulsive behavior and the related repetitive patterns of thought (i.e. ‘habits of thought’). An alternative possibility posits that obsessive thoughts may develop as a result of compulsive behavior (Gillan and Robbins, 2014). Evidence for this idea comes from a study where OCD patients were found to engage in post-hoc rationalization in order to explain a series of habitual responses (Gillan et al., 2014b). The notion is that in OCD, experiencing a recurrent urge to wash one’s hands might cause a patient to infer that they are concerned about hygiene. Future, longitudinal work will be needed to dissect the temporal dynamics of these symptom features to test these hypotheses, which are not mutually exclusive.

Researchers have suggested that ‘Impulsivity’ and ‘Compulsivity’ are partially overlapping neurocognitive features relevant for many psychiatric disorders (Robbins et al., 2012). The present study offers some insights in this regard. While the total score of the impulsivity scale was a strong predictor of goal-directed deficits, it did not load significantly onto the ‘Compulsive Behavior and Intrusive Thought’ factor, suggesting it has an independent association with goal-directed deficits. The supervised analysis identified the items from the impulsivity scale that best predicted goal-directed deficits. In terms of the overlap between the impulsivity questionnaire items and Factor 2, the two above-threshold predictors of model-based deficits were “I spend or charge more than I earn” and “I do things without thinking”, each of which is qualitatively characteristic of compulsive, habitual behavior. Importantly, the three items that did not overlap with ‘Compulsive Behavior and Intrusive Thought’, but still predicted model-based learning, tracked subjects’ general interest in engaging with the task (e.g. “I do not like puzzles”, “I do not like to think about complex problems”). We suggest that these items may not be of particular clinical importance, but simply serve as a marker of how likely individuals are to engage with the task material. In summary, while a small subset of the impulsivity items contributed to ‘Compulsive Behavior and Intrusive Thought’, impulsivity as assessed by our scale was mostly distinct. Of course, impulsivity as a construct itself involves a broad range of potentially distinct behaviors, such as impatient inter-temporal choice preferences and premature responding (Dalley et al., 2011). Further work will be need to assess how such behaviors relate to the features measured here; notably, our large-scale online methodology is well suited for examining such questions.

As has been shown for other tests that broadly fall within the category of executive function (Arffa, 2007), model-based learning was also associated with IQ and age (and gender in experiment 2 only). Although these effects were controlled for in all analyses and therefore do not bias the interpretation of our results, they highlight the fact that the coupling between model-based learning and ‘Compulsive Behavior and Intrusive Thought’ is far from perfect. One particularly interesting observation is that as people get older, they show greater deficits in model-based learning (Supplementary file 1B), but fewer psychiatric symptoms on all nine questionnaires collected in the present study (Supplementary file 1A), in line with prior work with diagnosed patients (Kessler et al., 2005). This incongruence suggests that there may be multiple dissociable processes responsible for model-based learning. Future studies are needed to dissect this somewhat complex construct into its constituent parts (as has been already attempted for other executive tasks [Miyake et al., 2000]), with a view to identifying the simpler component that is specific to the compulsive phenotype. Relatedly, future work might test if working memory might conceivably contribute to this association observed in the present study (Otto et al., 2013). Also, the strength of the association between a clinical phenotype and an underlying mechanism is fundamentally limited by the accuracy with which we can assess that phenotype. Aside from issues of relatively low reliability of self-report clinical symptoms (e.g. self-report OCD, r=0.71 [Hajcak et al., 2004]), we are also limited by the questions we ask. For example, in the present study we did not account for pathological gambling or trichotillomania, which are similarly defined clinically by a loss of control over repetitive behavior (Potenza, 2008; Chamberlain et al., 2007) and therefore may contribute noise to our signal. It is clear that iterative improvements to both self-report assessment and behavioral testing are needed to increase effect sizes and further refine the neurobiological characterization of Compulsive Behavior and Intrusive Thought suggested by these data.

Although we have labeled the three factors that emerged from our unsupervised analysis based on theoretical considerations, we acknowledge that this is an inherently subjective process and that some may rightfully disagree with our choice of terminology. An important distinction to be made here is that although this labeling process was subjective, the way in which these clusters were identified was not. We first identified a heretofore-unrecognized collection of trans-diagnostic psychiatric symptoms based on their inter-correlations and then validated this clustering by demonstrating an association with neurocognitive performance in an independent task. ‘Compulsive Behavior and Intrusive Thought’ is not intended to be a fixed or final definition – rather it is hoped that future work can (i) use the clusters defined in this study to find closer links between biological markers and clinical and (ii) improve and augment these clusters through further data-driven evaluations. More broadly, we hope that this methodology can be employed in many other areas of psychiatry where the considerable issues of heterogeneity within and homogeneity across the existing diagnostic categories is curtailing efforts to delineate the precise neurobiological basis of psychiatric problems.

In the present study, we did not screen for psychiatric disorders, favoring the acquisition of a large sample within which we could leverage normal variation in psychopathology. Although our results converge with prior work using this neurocognitive marker in compulsive disorders (Voon et al., 2015), future studies will be needed to test if these dimensional results map onto clinically diagnosed patients. For example, based on the results of the present study, we hypothesize that the co-morbidity between OCD and addiction might be largely explained by a common deficit in goal-directed control. Conversely, the co-morbidity between OCD and anxiety disorders might be explained by an orthogonal (equally important) symptom dimension. This kind of exciting work should be coupled with studies aiming to use such trans-diagnostic markers to predict treatment response on an individual basis within the existing diagnostic categories.

Altogether, these data suggest that ‘Compulsive Behavior and Intrusive Thought’ together constitute a dimensional psychiatric phenotype that can be tracked in the general population and is linked to deficits in goal-directed control over action, which has a clear neurobiological foundation (Dolan and Dayan, 2013). These data highlight the utility of a computational approach to psychiatry (Montague et al., 2012) and specifically our novel approach of leveraging large datasets, online testing, and normal variation in psychopathology to isolate the neurocognitive basis of psychiatric dimensions that may be relevant for multiple disorders. More broadly, the results of this study constitute progress toward realizing the promise of the RDoC initiative, suggesting that dimensional markers of psychiatric disturbances may map more closely to underlying biological states than do the overlapping and heterogeneous definitions of DSM disorders.

1. 1. Alvares GA
   2. Balleine BW
   3. Guastella AJ

   (2014) Impairments in goal-directed actions predict treatment response to cognitive-behavioral therapy in social anxiety disorder

   PLoS ONE 9:e94778.

   https://doi.org/10.1371/journal.pone.0094778

   • Google Scholar
2. 1. American Psychiatric Association

   (2013)

   Diagnostic and Statistical Manual of Mental Disorders, (5 edn), American Psychiatric Publishing

   Diagnostic and Statistical Manual of Mental Disorders, (5 edn), American Psychiatric Publishing.

   • Google Scholar
3. 1. Arffa S

   (2007) The relationship of intelligence to executive function and non-executive function measures in a sample of average, above average, and gifted youth

   Archives of Clinical Neuropsychology 22:969–978.

   https://doi.org/10.1016/j.acn.2007.08.001

   • Google Scholar
4. 1. Baker F

   (1992)

   Item Response Theory: Parameter Estimation Techniques, Marcel Dekker

   Item Response Theory: Parameter Estimation Techniques, Marcel Dekker.

   • Google Scholar
5. 1. Bezanson J
   2. Karpinski S
   3. Shah VB
   4. Edelman A

   (2012)

   Julia: a fast dynamic language for technical computing

   MIT.

   • Google Scholar
6. 1. Bickel WK
   2. Jarmolowicz DP
   3. Mueller ET
   4. Koffarnus MN
   5. Gatchalian KM

   (2012) Excessive discounting of delayed reinforcers as a trans-disease process contributing to addiction and other disease-related vulnerabilities: emerging evidence

   Pharmacology & Therapeutics 134:287–297.

   https://doi.org/10.1016/j.pharmthera.2012.02.004

   • Google Scholar
7. 1. Birnbaum A

   (1969) Statistical theory for logistic mental test models with a prior distribution of ability

   Journal of Mathematical Psychology 6:258–276.

   https://doi.org/10.1016/0022-2496(69)90005-4

   • Google Scholar
8. 1. Brodersen KH
   2. Deserno L
   3. Schlagenhauf F
   4. Lin Z
   5. Penny WD
   6. Buhmann JM
   7. Stephan KE

   (2014) Dissecting psychiatric spectrum disorders by generative embedding

   NeuroImage: Clinical 4:98–111.

   https://doi.org/10.1016/j.nicl.2013.11.002

   • Google Scholar
9. 1. Cattell RB

   (1966) The scree test for the number of factors

   Multivariate Behavioral Research 1:245–276.

   https://doi.org/10.1207/s15327906mbr0102_10

   • Google Scholar
10. 1. Chamberlain SR
    2. Fineberg NA
    3. Blackwell AD
    4. Clark L
    5. Robbins TW
    6. Sahakian BJ

    (2007) A neuropsychological comparison of obsessive–compulsive disorder and trichotillomania

    Neuropsychologia 45:654–662.

    https://doi.org/10.1016/j.neuropsychologia.2006.07.016

    • Google Scholar
11. 1. Crump MJC
    2. McDonnell JV
    3. Gureckis TM

    (2013) Evaluating amazon's mechanical turk as a tool for experimental behavioral research

    PLoS One 8:e57410.

    https://doi.org/10.1371/journal.pone.0057410

    • Google Scholar
12. 1. Cushman F
    2. Morris A

    (2015) Habitual control of goal selection in humans

    Proceedings of the National Academy of Sciences of the United States of America 112:13817–13822.

    https://doi.org/10.1073/pnas.1506367112

    • Google Scholar
13. 1. Cuthbert BN
    2. Kozak MJ

    (2013) Constructing constructs for psychopathology: the NIMH research domain criteria

    Journal of Abnormal Psychology 122:928–937.

    https://doi.org/10.1037/a0034028

    • Google Scholar
14. 1. Dalley JW
    2. Everitt BJ
    3. Robbins TW

    (2011) Impulsivity, compulsivity, and top-down cognitive control

    Neuron 69:680–694.

    https://doi.org/10.1016/j.neuron.2011.01.020

    • Google Scholar
15. 1. Daw ND
    2. Gershman SJ
    3. Seymour B
    4. Dayan P
    5. Dolan RJ

    (2011) Model-based influences on humans' choices and striatal prediction errors

    Neuron 69:1204–1215.

    https://doi.org/10.1016/j.neuron.2011.02.027

    • Google Scholar
16. 1. Dickinson A

    (1985) Actions and habits: the development of behavioural autonomy

    Philosophical Transactions of the Royal Society B: Biological Sciences 308:67–78.

    https://doi.org/10.1098/rstb.1985.0010

    • Google Scholar
17. 1. Dickinson A
    2. Balleine B

    (1994) Motivational control of goal-directed action

    Animal Learning & Behavior 22:1–18.

    https://doi.org/10.3758/BF03199951

    • Google Scholar
18. 1. Dolan RJ
    2. Dayan P

    (2013) Goals and habits in the brain

    Neuron 80:312–325.

    https://doi.org/10.1016/j.neuron.2013.09.007

    • Google Scholar
19. 1. Drasgow F
    2. Lissak RI

    (1983) Modified parallel analysis: a procedure for examining the latent dimensionality of dichotomously scored item responses

    Journal of Applied Psychology 68:363–373.

    https://doi.org/10.1037/0021-9010.68.3.363

    • Google Scholar
20. 1. Eppinger B
    2. Walter M
    3. Heekeren HR
    4. Li S-C

    (2013) Of goals and habits: age-related and individual differences in goal-directed decision-making

    Frontiers in Neuroscience 7:.

    https://doi.org/10.3389/fnins.2013.00253

    • Google Scholar
21. 1. Everitt BJ
    2. Robbins TW

    (2005) Neural systems of reinforcement for drug addiction: from actions to habits to compulsion

    Nature Neuroscience 8:1481–1489.

    https://doi.org/10.1038/nn1579

    • Google Scholar
22. 1. Fair DA
    2. Bathula D
    3. Nikolas MA
    4. Nigg JT

    (2012) Distinct neuropsychological subgroups in typically developing youth inform heterogeneity in children with ADHD

    Proceedings of the National Academy of Sciences of the United States of America 109:6769–6774.

    https://doi.org/10.1073/pnas.1115365109

    • Google Scholar
23. 1. Foa EB
    2. Huppert JD
    3. Leiberg S
    4. Langner R
    5. Kichic R
    6. Hajcak G
    7. Salkovskis PM

    (2002) The obsessive-compulsive inventory: development and validation of a short version

    Psychological Assessment 14:485–496.

    https://doi.org/10.1037/1040-3590.14.4.485

    • Google Scholar
24. 1. Friedel E
    2. Koch SP
    3. Wendt J
    4. Heinz A
    5. Deserno L
    6. Schlagenhauf F

    (2014) Devaluation and sequential decisions: linking goal-directed and model-based behavior

    Frontiers in Human Neuroscience 8:587.

    https://doi.org/10.3389/fnhum.2014.00587

    • Google Scholar
25. 1. Furlong TM
    2. Jayaweera HK
    3. Balleine BW
    4. Corbit LH

    (2014) Binge-like consumption of a palatable food accelerates habitual control of behavior and is dependent on activation of the dorsolateral striatum

    Journal of Neuroscience 34:5012–5022.

    https://doi.org/10.1523/JNEUROSCI.3707-13.2014

    • Google Scholar
26. 1. Garner DM
    2. Olmsted MP
    3. Bohr Y
    4. Garfinkel PE

    (1982) The eating attitudes test: psychometric features and clinical correlates

    Psychological Medicine 12:871–878.

    https://doi.org/10.1017/S0033291700049163

    • Google Scholar
27. 1. Gillan CM
    2. Papmeyer M
    3. Morein-Zamir S
    4. Sahakian BJ
    5. Fineberg NA
    6. Robbins TW
    7. de Wit S

    (2011) Disruption in the balance between goal-directed behavior and habit learning in obsessive-compulsive disorder

    American Journal of Psychiatry 168:718–726.

    https://doi.org/10.1176/appi.ajp.2011.10071062

    • Google Scholar
28. 1. Gillan CM
    2. Robbins TW

    (2014) Goal-directed learning and obsessive-compulsive disorder

    Philosophical Transactions of the Royal Society B: Biological Sciences 369:20130475.

    https://doi.org/10.1098/rstb.2013.0475

    • Google Scholar
29. 1. Gillan CM
    2. Morein-Zamir S
    3. Kaser M
    4. Fineberg NA
    5. Sule A
    6. Sahakian BJ
    7. Cardinal RN
    8. Robbins TW

    (2014a) Counterfactual processing of economic action-outcome alternatives in obsessive-compulsive disorder: further evidence of impaired goal-directed behavior

    Biological Psychiatry 75:639–646.

    https://doi.org/10.1016/j.biopsych.2013.01.018

    • Google Scholar
30. 1. Gillan CM
    2. Morein-Zamir S
    3. Urcelay GP
    4. Sule A
    5. Voon V
    6. Apergis-Schoute AM
    7. Fineberg NA
    8. Sahakian BJ
    9. Robbins TW

    (2014b) Enhanced avoidance habits in obsessive-compulsive disorder

    Biological Psychiatry 75:631–638.

    https://doi.org/10.1016/j.biopsych.2013.02.002

    • Google Scholar
31. 1. Gillan CM
    2. Apergis-Schoute A
    3. Morein-Zamir S
    4. Urcelay GP
    5. Sule A
    6. Fineberg NA
    7. Sahakian BJ
    8. Robbins TW

    (2015a) Functional neuroimaging of avoidance habits in obsessive-compulsive disorder

    American Journal of Psychiatry 172:284–293.

    https://doi.org/10.1176/appi.ajp.2014.14040525

    • Google Scholar
32. 1. Gillan CM
    2. Robbins TW
    3. Sahakian BJ
    4. van den Heuvel OA
    5. van Wingen G

    (2015b) The role of habit in compulsivity

    European Neuropsychopharmacology.

    https://doi.org/10.1016/j.euroneuro.2015.12.033

    • Google Scholar
33. 1. Gillan CM
    2. Otto AR
    3. Phelps EA
    4. Daw ND

    (2015c) Model-based learning protects against forming habits

    Cognitive, Affective, & Behavioral Neuroscience 15:523–536.

    https://doi.org/10.3758/s13415-015-0347-6

    • Google Scholar
34. 1. Godier LR
    2. Park RJ

    (2014) Compulsivity in anorexia nervosa: a transdiagnostic concept

    Frontiers in Psychology 5:778.

    https://doi.org/10.3389/fpsyg.2014.00778

    • Google Scholar
35. 1. Gorsuch R
    2. Nelson J

    (1981)

    CNG screen test: an objective procedure for determining the number of factors

    • Google Scholar
36. 1. Graybiel AM
    2. Rauch SL

    (2000) Toward a neurobiology of obsessive-compulsive disorder

    Neuron 28:343–347.

    https://doi.org/10.1016/S0896-6273(00)00113-6

    • Google Scholar
37. 1. Hajcak G
    2. Huppert JD
    3. Simons RF
    4. Foa EB

    (2004) Psychometric properties of the OCI-r in a college sample

    Behaviour Research and Therapy 42:115–123.

    https://doi.org/10.1016/j.brat.2003.08.002

    • Google Scholar
38. 1. Hoerl AE
    2. Kennard RW

    (1970) Ridge regression: biased estimation for nonorthogonal problems

    Technometrics 12:55–67.

    https://doi.org/10.1080/00401706.1970.10488634

    • Google Scholar
39. 1. Huys QJM
    2. Cools R
    3. Gölzer M
    4. Friedel E
    5. Heinz A
    6. Dolan RJ
    7. Dayan P

    (2011) Disentangling the roles of approach, activation and valence in instrumental and pavlovian responding

    PLoS Computational Biology 7:e1002028.

    https://doi.org/10.1371/journal.pcbi.1002028

    • Google Scholar
40. 1. Hyman SE

    (2007) Can neuroscience be integrated into the DSM-v?

    Nature Reviews Neuroscience 8:725–732.

    https://doi.org/10.1038/nrn2218

    • Google Scholar
41. 1. Insel T
    2. Cuthbert B
    3. Garvey M
    4. Heinssen R
    5. Pine DS
    6. Quinn K
    7. Sanislow C
    8. Wang P

    (2010) Research domain criteria (rDoC): toward a new classification framework for research on mental disorders

    American Journal of Psychiatry 167:748–751.

    https://doi.org/10.1176/appi.ajp.2010.09091379

    • Google Scholar
42. 1. Kessler RC
    2. Chiu WT
    3. Demler O
    4. Walters EE

    (2005) Prevalence, severity, and comorbidity of 12-month DSM-IV disorders in the national comorbidity survey replication

    Archives of General Psychiatry 62:617–389.

    https://doi.org/10.1001/archpsyc.62.6.617

    • Google Scholar
43. 1. Kingsbury GG
    2. Zara AR

    (1989) Procedures for selecting items for computerized adaptive tests

    Applied Measurement in Education 2:359–375.

    https://doi.org/10.1207/s15324818ame0204_6

    • Google Scholar
44. 1. Liebowitz MR

    (1987) Social phobia

    Modern Problems of Pharmacopsychiatry 22:141–173.

    https://doi.org/10.1159/000414022

    • Google Scholar
45. 1. Lipszyc J
    2. Schachar R

    (2010) Inhibitory control and psychopathology: a meta-analysis of studies using the stop signal task

    Journal of the International Neuropsychological Society 16:1064–1076.

    https://doi.org/10.1017/S1355617710000895

    • Google Scholar
46. 1. Lysaker PH
    2. Bell MD
    3. Bryson G
    4. Kaplan E

    (1998) Neurocognitive function and insight in schizophrenia: support for an association with impairments in executive function but not with impairments in global function

    Acta Psychiatrica Scandinavica 97:297–301.

    https://doi.org/10.1111/j.1600-0447.1998.tb10003.x

    • Google Scholar
47. 1. Lysaker PH
    2. Bryson GJ
    3. Lancaster RS
    4. Evans JD
    5. Bell MD

    (2003) Insight in schizophrenia: associations with executive function and coping style

    Schizophrenia Research 59:41–47.

    https://doi.org/10.1016/S0920-9964(01)00383-8

    • Google Scholar
48. 1. Marin RS
    2. Biedrzycki RC
    3. Firinciogullari S

    (1991) Reliability and validity of the apathy evaluation scale

    Psychiatry Research 38:143–162.

    https://doi.org/10.1016/0165-1781(91)90040-V

    • Google Scholar
49. 1. Mason O
    2. Linney Y
    3. Claridge G

    (2005) Short scales for measuring schizotypy

    Schizophrenia Research 78:293–296.

    https://doi.org/10.1016/j.schres.2005.06.020

    • Google Scholar
50. 1. Miyake A
    2. Friedman NP
    3. Emerson MJ
    4. Witzki AH
    5. Howerter A
    6. Wager TD

    (2000) The unity and diversity of executive functions and their contributions to complex “frontal lobe” tasks: a latent variable analysis

    Cognitive Psychology 41:49–100.

    https://doi.org/10.1006/cogp.1999.0734

    • Google Scholar
51. 1. Montague PR
    2. Dolan RJ
    3. Friston KJ
    4. Dayan P

    (2012) Computational psychiatry

    Trends in Cognitive Sciences 16:72–80.

    https://doi.org/10.1016/j.tics.2011.11.018

    • Google Scholar
52. 1. Morris RW
    2. Quail S
    3. Griffiths KR
    4. Green MJ
    5. Balleine BW

    (2015) Corticostriatal control of goal-directed action is impaired in schizophrenia

    Biological Psychiatry 77:187–195.

    https://doi.org/10.1016/j.biopsych.2014.06.005

    • Google Scholar
53. 1. Otto AR
    2. Raio CM
    3. Chiang A
    4. Phelps EA
    5. Daw ND

    (2013) Working-memory capacity protects model-based learning from stress

    Proceedings of the National Academy of Sciences of the United States of America 110:20941–20946.

    https://doi.org/10.1073/pnas.1312011110

    • Google Scholar
54. 1. Patton JH
    2. Stanford MS
    3. Barratt ES

    (1995) Factor structure of the barratt impulsiveness scale

    Journal of Clinical Psychology 51:768–774.

    https://doi.org/10.1002/1097-4679(199511)51:6<768::AID-JCLP2270510607>3.0.CO;2-1

    • Google Scholar
55. 1. Potenza MN

    (2008) Review. the neurobiology of pathological gambling and drug addiction: an overview and new findings

    Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 363:3181–3189.

    https://doi.org/10.1098/rstb.2008.0100

    • Google Scholar
56. 1. Poyurovsky M
    2. Koran LM

    (2005) Obsessive–compulsive disorder (oCD) with schizotypy vs. schizophrenia with OCD: diagnostic dilemmas and therapeutic implications

    Journal of Psychiatric Research 39:399–408.

    https://doi.org/10.1016/j.jpsychires.2004.09.004

    • Google Scholar
57. 1. Raven J

    (2000) The raven's progressive matrices: change and stability over culture and time

    Cognitive Psychology 41:1–48.

    https://doi.org/10.1006/cogp.1999.0735

    • Google Scholar
58. 1. Robbins TW
    2. Gillan CM
    3. Smith DG
    4. de Wit S
    5. Ersche KD

    (2012) Neurocognitive endophenotypes of impulsivity and compulsivity: towards dimensional psychiatry

    Trends in Cognitive Sciences 16:81–91.

    https://doi.org/10.1016/j.tics.2011.11.009

    • Google Scholar
59. 1. Ruscio AM
    2. Stein DJ
    3. Chiu WT
    4. Kessler RC

    (2010) The epidemiology of obsessive-compulsive disorder in the national comorbidity survey replication

    Molecular Psychiatry 15:53–63.

    https://doi.org/10.1038/mp.2008.94

    • Google Scholar
60. 1. Sandstrom NJ
    2. Kaufman J
    3. A. Huettel S

    (1998) Males and females use different distal cues in a virtual environment navigation task

    Cognitive Brain Research 6:351–360.

    https://doi.org/10.1016/S0926-6410(98)00002-0

    • Google Scholar
61. 1. Saunders JB
    2. Aasland OG
    3. Babor TF
    4. De La Fuente JR
    5. Grant M

    (1993) Development of the alcohol use disorders identification test (AUDIT): WHO collaborative project on early detection of persons with harmful alcohol consumption-II

    Addiction 88:791–804.

    https://doi.org/10.1111/j.1360-0443.1993.tb02093.x

    • Google Scholar
62. 1. Schad DJ
    2. Jünger E
    3. Sebold M
    4. Garbusow M
    5. Bernhardt N
    6. Javadi A-H
    7. Zimmermann US
    8. Smolka MN
    9. Heinz A
    10. Rapp MA
    11. Huys QJM

    (2014) Processing speed enhances model-based over model-free reinforcement learning in the presence of high working memory functioning

    Frontiers in Psychology 5:1450.

    https://doi.org/10.3389/fpsyg.2014.01450

    • Google Scholar
63. 1. Schwabe L
    2. Wolf OT

    (2009) Stress prompts habit behavior in humans

    Journal of Neuroscience 29:7191–7198.

    https://doi.org/10.1523/JNEUROSCI.0979-09.2009

    • Google Scholar
64. 1. Sjoerds Z
    2. de Wit S
    3. van den Brink W
    4. Robbins TW
    5. Beekman ATF
    6. Penninx BWJH
    7. Veltman DJ

    (2013) Behavioral and neuroimaging evidence for overreliance on habit learning in alcohol-dependent patients

    Translational Psychiatry 3:e337.

    https://doi.org/10.1038/tp.2013.107

    • Google Scholar
65. 1. Skatova A
    2. Chan PA
    3. Daw ND

    (2013) Extraversion differentiates between model-based and model-free strategies in a reinforcement learning task

    Frontiers in Human Neuroscience 7:.

    https://doi.org/10.3389/fnhum.2013.00525

    • Google Scholar
66. Book

    1. Spielberger CD
    2. Gorsuch RL
    3. Lushene R
    4. Vagg PR
    5. Jacobs GA

    (1983)

    Manual for the State-Trait Anxiety Inventory

    Consulting Psychologists Press.

    • Google Scholar
67. 1. Tibshirani R

    (1996)

    Regression shrinkage and selection via the lasso

    Journal of the Royal Statistical Society Series B-Methodological 58:267–288.

    • Google Scholar
68. Book

    1. va der Linden W
    2. Pashley P

    (2000)

    Computerized adaptive testing

    In: van der Linden WJ, Glas C A W, editors. Theory and Practice. Kluwer. pp. 1–25.

    • Google Scholar
69. 1. Voon V
    2. Derbyshire K
    3. Rück C
    4. Irvine MA
    5. Worbe Y
    6. Enander J
    7. Schreiber LRN
    8. Gillan C
    9. Fineberg NA
    10. Sahakian BJ
    11. Robbins TW
    12. Harrison NA
    13. Wood J
    14. Daw ND
    15. Dayan P
    16. Grant JE
    17. Bullmore ET

    (2015) Disorders of compulsivity: a common bias towards learning habits

    Molecular Psychiatry 20:345–352.

    https://doi.org/10.1038/mp.2014.44

    • Google Scholar
70. 1. Zung WWK

    (1965) A self-rating depression scale

    Archives of General Psychiatry 12:63–70.

    https://doi.org/10.1001/archpsyc.1965.01720310065008

    • Google Scholar
71. 1. Zou H
    2. Hastie T

    (2005) Regularization and variable selection via the elastic net

    Journal of the Royal Statistical Society: Series B (Statistical Methodology) 67:301–320.

    https://doi.org/10.1111/j.1467-9868.2005.00503.x

    • Google Scholar

Author details

1. Claire M Gillan

   1. Department of Psychology, New York University, New York, United States
   2. Department of Psychology, University of Cambridge, Cambridge, United Kingdom
   3. Behavioural and Clinical Neuroscience Institute, University of Cambridge, Cambridge, United Kingdom

   Contribution

   CMG, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   claire.gillan@gmail.com

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0001-9065-403X

2. Michal Kosinski

   Stanford Graduate School of Business, Stanford University, Stanford, United States

   Contribution

   MK, Acquisition of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

3. Robert Whelan

   Department of Psychology, University College Dublin, Dulbin, Ireland

   Contribution

   RW, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

4. Elizabeth A Phelps

   1. Department of Psychology, New York University, New York, United States
   2. Center for Neural Science, New York University, New York, United States
   3. Nathan Kline Institute, New York, United States

   Contribution

   EAP, Conception and design, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

5. Nathaniel D Daw

   1. Department of Psychology, Princeton University, Princeton, United States
   2. Neuroscience Institute, Princeton University, Princeton, United States

   Contribution

   NDD, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

• 19,635

  views

• 2,531

  downloads

• 387

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.