Abstract
Biases in information gathering are common in the general population and reach pathological extremes in paralysing indecisiveness, as in obsessive–compulsive disorder (OCD). Here we adopt a new perspective on information gathering and demonstrate an information integration bias whereby there is over-weighting of most recent information via evidence strength updates (ΔES). In a crowd-sourced sample (N = 5,237), we find that a reduced ΔES weighting drives indecisiveness along an obsessive–compulsive spectrum. We replicate this attenuated ΔES weighting in a second lab-based study (N = 105) that includes a transdiagnostic obsessive–compulsive spectrum encompassing OCD and generalized anxiety patients. Using magnetoencephalography (MEG), we trace ΔES signals to a late neural signal peaking at ~920 ms. Critically, highly obsessive–compulsive participants, across diagnoses, show an attenuated neural ΔES signal in mediofrontal areas, while other decision-relevant processes remain intact. Our findings establish biased information weighting as a driver of information gathering, where attenuated ΔES is linked to indecisiveness across an obsessive–compulsive spectrum.
Similar content being viewed by others
Main
Knowing when to decide is difficult1,2. From selecting a movie on a streaming platform to buying a car, whenever there is uncertainty about the ‘right’ choice, gathering more information usually helps us make better decisions. But being indecisive and spending too much time gathering information is also disadvantageous and carries substantial opportunity and energy costs3.
In humans, information gathering biases are common, often expressed as gathering too little or too much information4,5. Dramatically inflated information-gathering biases are a feature of psychopathology and believed to drive core symptoms4,6,7,8,9,10,11. For example, indecisiveness and persistent doubt can be debilitating aspects of obsessive–compulsive disorder (OCD)12,13. Excessive deliberation also extends beyond core OCD symptoms and can impact everyday wellbeing and function14.
While indecisiveness in OCD is traditionally assessed using clinical interviews15, recent literature has provided a more quantitative approach based on using information gathering tasks4,10,16,17,18,19,20,21,22,23. Studies using the latter have shown that patients with OCD, as well as non-clinical participants with high obsessive–compulsive (OC) scores, manifest elevated levels of information gathering before committing to a decision. While similar effects have been observed in naturalistic settings23, the precise neurocognitive mechanisms underlying the expression of indecisiveness remain unknown.
A relevant field of study, largely disconnected from information gathering, is that of evidence accumulation in decision making24. Here it is considered that evidence for different options increases gradually with incoming information, and a decision is made when evidence for an option hits a decision threshold24,25,26,27,28,29. The latter is believed to collapse with time, meaning that decisions become increasingly liberal, which is often framed in terms of an urgency signal30,31,32,33. These effects are observed in a range of information-gathering contexts despite striking differences in tasks and timescales34,35,36. Importantly, recent studies suggest that evidence does not accumulate linearly as a function of decision-relevant information, but in a recency-biased manner such that the most recent evidence is over-weighted37,38,39,40. Determining whether such biases also exist in information gathering and whether they act as drivers for indecisiveness is an important question.
In this study, we investigate the neurocognitive contributors to information gathering across two samples and tasks which we previously collected in the lab. To pre-empt the results, we find that, in addition to urgency-like signals, evidence is integrated non-monotonically, with what we term as evidence strength updates (ΔES) critically determining when a participant decides to decide. Furthermore, across both studies, we find that participants along an OC spectrum rely less on these ΔES both in clinical and non-clinical samples. Using magnetoencephalography (MEG), we show that this attenuation of ΔES is mirrored in the brain, with neural ΔES representations in mediofrontal areas being less evident in OC whereas other critical contributors to information gathering remain intact. We conclude that indecisiveness along an OC spectrum is driven by a reduced reliance on most recent information.
Results
Cognitive contributors to information gathering
To first ascertain a link between information gathering and OC symptoms in the general population, we conducted a crowd-sourced study (N = 5,237 at the time of analysis). To this end, we implemented an information-gathering task on an app for handheld electronic devices (www.brainexplorer.net), where participants had to decide which of two stimuli was more plentiful across 25 hidden locations (Fig. 1a). Participants were free to sample as many of the locations as they wanted before committing to a decision, such that we could use the number of draws before a decision as an index of indecisiveness.
a, The participants’ task was to determine which of two possible gems was more abundant (screenshot of the task on the Brain Explorer app, https://brainexplorer.net/). They were free to uncover as much evidence as they wished, by tapping on locations on the grid, until they decided to commit to a binary choice between one of two gems. An example sequence of samples and corresponding evidence accumulation process for cumulative evidence strength at the previous draw ESd−1 and the evidence strength update ΔES is depicted on the right. Here evidence strength at draw d ESd is quantified as the cumulative difference in evidence for the two gems with respect to the gem that is more abundant at draw d (for example, here 3 diamonds minus 2 yellow gems constitutes an evidence strength of 1 at draw 5). Cumulative evidence strength at the previous draw ESd−1 is defined as the lagged ES by one draw, and evidence strength update ΔES is quantified as the signed difference between the cumulative ES at draw d−1 and draw d. Participants rated their confidence after every binary choice, then received 100 points for correct responses or lost 100 points for incorrect responses. b, Within a sample of the general population (N = 5,237), individuals with higher OC symptoms sampled more information before committing to a decision (ρs = 0.040, P = 0.004). Data points pertaining to individuals with a self-reported OCD diagnosis are shown in orange. c, The probability of making a decision p(decide) was predicted from experimental factors as well as individual differences in OC symptoms using a general linear mixed model (GLMM, N = 5,237; data are presented as beta coefficients ± standard error (s.e.), scale adjusted to show main effects of experimental factors on the left and effects associated with individual differences in OC on the right y axis). The experimental factors (left y axis) comprised current draw number, cumulative evidence from the start of the game until the previous draw ESd−1, and the evidence strength update ΔES. More draws and higher ESd−1 and ΔES increased the probability of making a decision p(decide) (all P < 0.001). Individual differences in obsessive–compulsive symptoms (right y axis) were quantified as the main effect of the OCI-R score and the interactions of the OCI-R score with the experimental factors. The analysis showed that OC symptoms per se did not have a significant effect on p(decide) (P = 0.265), and that people with high OC symptoms did not weight the current draw number differently in deciding to commit to a choice (P = 0.406); instead, they weighted both previous and current evidence less (P < 0.001, highlighted by shaded area). The latter effect is illustrated in d. d, To obtain estimates of sensitivity to evidence at a single-participant level, we fit a general linear model (GLM) predicting p(decide) separately for each participant. Convergent with the significant interaction effect in the GLMM, those with higher OC symptoms had reduced beta coefficients for ΔES, that is, they weighed ΔES less in deciding to commit (ρs = −0.121, P < 0.001, N = 3,901 after removing outliers, defined as any participant with any beta values 1.5× interquartile range (IQR) above the third quartile or below the first quartile). e, Individuals with higher OC symptoms also rated their confidence in their decisions as being lower on average (ρs = −0.054, P = 0.002, N = 3,293, see exclusion criteria in Methods). f, A mediation analysis suggests that sensitivity to ΔES partially mediates the association between OC symptoms (OCI-R total score) and confidence ratings, such that lower confidence is partially explained by altered evidence integration (N = 3,293). All tests are two-sided and not corrected for multiple comparisons.
First, to better understand the cognitive contributors to participants’ information-gathering behaviour, we used general linear mixed models (GLMMs) to predict on each draw whether a participant would continue sampling information or make a decision (p(decide)) using a series of cognitive and individual differences predictors (see Methods for the full model). In this task, information accumulates non-monotonically, allowing us to distinguish the contribution of different cognitive factors (for example, current information, total evidence, number of draws) to commitment to a decision. To test whether a recency bias41,42,43 is present in an information-gathering context, we distinguished total evidence strength on the previous draw (ESd−1; absolute cumulative evidence difference across draws [1 to d−1] for the current majority) from the most recent change in evidence strength or evidence strength update (ΔES; Fig. 1a). This is analogous to a distinction between previous expectation and prediction error, that is, the mismatch between the previous expectation and the actual observation, which is central to frameworks and models on processes spanning from perception to learning across domains44,45.
Model comparison favoured the model with ΔES and ESd−1 as separate predictors over a model where all evidence was aggregated in ESd as a single predictor (difference in Akaike information criterion (ΔAIC) = 1.709 × 106; difference in Bayesian information criterion (ΔBIC) = 1.709 × 106), supporting the idea of a (recency-) weighted information integration. We found that both the current update in signal strength (ΔES, β = 1.105, P < 0.001, 95% confidence interval (CI) 1.086–1.123; Fig. 1c) and previous evidence strength (ESd−1; β = 1.042, P < 0.001, 95% CI 1.029–1.055) positively predicted whether participants would commit to a decision, but that the former predicted it more strongly. A separate GLMM predicting p(decide) from the current evidence at draw d, d−1 and d−2 relative to the chosen option showed a decreasing contribution of current evidence for lagged draws (ESd: β = 0.648, P < 0.001, 95% CI 0.635–0.662; ESd−1: β = 0.253, P < 0.001, 95% CI 0.245–0.262; ESd−2: β = 0.195, P < 0.001, 95% CI 0.185–0.204; Supplementary Fig. 1a). This accords with a recency bias in information gathering, analogous to that found across other decision-making domains37,38,44,46. It is particularly striking as this task was designed so that all information remained present on the screen irrespective of when it was gathered, such that this recency bias is not accounted for by working memory limitations or related explanations.
In keeping with the idea of collapsing decision boundaries or urgency signals4,30,47,48, we also included the draw number as a predictor. Draw number additionally predicted whether participants would make a decision (β = 1.789, P < 0.001, 95% CI 1.752–1.826), such that participants were more likely to stop gathering information the longer they had already spent gathering information (irrespective of how strong the evidence is) and, in addition, if they (recently) received stronger evidence favouring one option over the other.
OC-linked indecisiveness is due to attenuation of evidence weighting
Next, we investigated whether an OC spectrum was linked to indecisiveness. Taking the number of draws before making a decision as an indicator of indecisiveness4,34,36,49,50, we found, as hypothesized, that participants with higher OC symptoms (as determined by the Obsessive–Compulsive Inventory-Revised (OCI-R) total score51) gathered more evidence before making a decision (Fig. 1b; ρs = 0.040, P = 0.004, 95% CI 0.012–0.067), while not showing any difference in accuracy for these decisions (ρs = −9.998 × 10−4, P = 0.942, 95% CI −0.028 to 0.026, BF₀₁ = 38.81, robustness region RR > 0.036). Thus, in this information-gathering dataset, we find that indecisiveness relates to OC symptoms across the general population.
Next, we investigated which of the above identified cognitive contributors explained differences in task behaviour in high-OC participants. We found that OC symptoms were linked to reduced weighting of previous evidence ESd−1 (ESd−1 × OCI-R: β = −0.053, P < 0.001, 95% CI −0.067 to −0.040) and particularly of ΔES (ΔES × OCI-R: β = −0.085, P < 0.001, 95% CI −0.104 to −0.067). There was no significant main effect of OC symptoms on sampling decisions (p(decide); Fig. 1d; β = 0.033, P = 0.265, 95% CI −0.025 to 0.092, ΔBIC = 2,205 comparing GLMM with and without main effect of OC), which would imply a simple tendency not to decide; and, contrary to our initial assumptions4,52, there was no significant effect on urgency (as captured by Ndraws × OCI-R: β = −0.016, P = 0.406, 95% CI −0.053 to 0.021, ΔBIC = 1,786 comparing GLMM with and without Ndraws × OCI-R interaction effect). We found similar but weaker effects of OC symptoms on evidence weighting when comparing a subset of participants with a self-reported OCD diagnosis with those who explicitly declared no lifetime psychiatric disorder (see Supplementary Information ‘Self-reported OCD diagnosis effects on information gathering in smartphone population sample’). To follow up on the recency effect suggested by these results, we targeted this by fitting a separate GLMM predicting p(decide) from the current evidence relative to the chosen option at draw d, d−1 and d−2 per participant (see Methods). We found that a difference in the beta weight for the evidence at draw d and the beta weight for the evidence at draw d−1 correlated negatively with the OCI-R score (ρs = −0.069, P < 0.001, 95% CI −0.099 to −0.038), supporting the notion of a decreased recency effect. This means that high-OC participants were no more likely to decide per se, or express an altered urgency signal, but instead showed an attenuated impact of the most recent information sample. In other words, weaker evidence strength updates seem to be driving indecisiveness along the OC spectrum.
Lowered confidence in high-OC participants mediated by attenuated ΔES
Having identified key drivers of indecisiveness, we next assessed whether this was a purely behavioural effect or whether it also affected participants’ perception of their performance or insight. To assess this, we asked participants to rate their post-choice confidence (after committing to a decision, but before viewing the outcome; Fig. 1a). We found that OC symptoms were also linked to confidence, with high-OC participants having lower confidence (Fig. 1e; ρs = −0.054, P = 0.002, 95% CI −0.088 to −0.020, N = 3,293, after excluding additional participants based on confidence data criteria, see Methods). This means that the OC spectrum was not only linked to indecisive information gathering but also to a lack of confidence, a finding that aligns with evidence of impoverished confidence in other domains53. It also chimes with the notion that indecisiveness is closely linked to subjective doubt.
Lastly, using mediation analysis (https://github.com/canlab/MediationToolbox (refs. 54,55)), we evaluated the link between the reduced ΔES weight from our information-gathering analysis and confidence findings. We found that the ΔES weighting (using beta estimates from non-hierarchical models; Fig. 1d) was associated with task confidence (b = −0.066, Z = −3.721, P < 0.001, 95% CI −0.100 to −0.031; Fig. 1f). Specifically, we found that ΔES partially mediated the influence of OC symptoms on confidence (ab = 0.010, Z = 3.775, P < 0.001, 95% CI 0.005–0.015; c = −0.043, Z = −2.335, P = 0.020, 95% CI −0.079 to −0.008; c′ = −0.052, Z = −2.809, P = 0.005, 95% CI −0.088 to −0.016; Fig. 1f). This means that the lower confidence in high-OC participants is, at least in part, influenced by the same underlying process as that of attenuated ΔES.
Cognitive contributors to information gathering replicate in a lab-based sample
As the above dataset was primarily derived from a self-selected population of non-clinical participants, we next asked whether the findings also held true in a more targeted patient sample. Thus, we analysed a previously collected independent sample that included patients assessed in the lab (N = 105), using a novel adaptation of the above information-gathering paradigm (Fig. 2a).
a, Participants’ goal was to determine which of two possible stimuli was more abundant. They were presented with a sequence of draws until they decided to declare for a stimulus. Each draw consisted of five stimuli at a time in varying proportions, allowing for the sequences to be constructed so as to reduce the collinearity between the number of draws, current and previous evidence, which in turn enables us to study their specific representations in the brain. Here again, evidence strength (ES) at draw d is quantified as the cumulative difference in evidence for the 2 gems with respect to the gem that is more abundant at draw d (for example, here 4 blue gems minus 1 yellow gem constitutes an evidence strength of 3 in the first draw, and 8 yellow gems minus 7 blue gems yields an evidence strength of 1 in the third draw). Cumulative evidence strength in the previous draw ESd−1 is defined as the lagged ES by one draw, and evidence strength update ΔES is quantified as the signed difference between the cumulative ES at draw d−1 and draw d. The maximum number of draws varied between 4 and 8 in the short-horizon condition and between 10 and 14 in the long-horizon condition. The horizon condition was cued by the colour of the frame (different colours were presented in the task). Participants received 2 points for a correct response, lost 2 points for an incorrect response, and lost 1 point for non-decisions. Points were later translated into bonus payments. b, To pool across clinical and non-clinical groups, we reduced the dimensionality of 7 questionnaires using an exploratory factor analysis (N = 105). This revealed a 3-factor solution, where the second factor mapped primarily onto items in questionnaires assessing OCD (OCI-R and Padua Inventory-Washington State University Revision (PI-WSUR)). We used this OC factor to quantify individual differences across all participants, plotted here by categorical group, that is, for participants from the general population with low and high obsessive–compulsive scores (‘low C’ and ‘high C’, respectively), healthy controls (‘Control’), participants with generalized anxiety disorder (‘GAD’) and participants with OCD (‘OCD’). Boxplots show the median (centre line) and IQR; whiskers extend to the largest value within 1.5× IQR. c, The number of draws to a decision varied across trials and participants in the short- and long-horizon condition (N = 105). Note that the number of draws to a decision per se was not associated with the transdiagnostic OC factor (ρs = −0.117, P = 0.236), yet this is not surprising given the highly constraining horizon manipulation. d, The probability to make a decision was predicted in a GLMM from experimental factors as well as variability in the OC factor (N = 105, data are presented as beta coefficients ± s.e., scale adjusted to show main effects of experimental factors on the left and effects associated with individual differences in OC on the right y axis). Experimental factors (left y axis) comprised the number of draws, the cumulative evidence strength from the start of the game until the previous draw ESd−1, ΔES and the horizon condition (controlling for interactions of the horizon condition with all other experimental factors). More draws, higher previous and current evidence strength, and a short-horizon condition all increase the probability to make a decision (all P < 0.001). Individual differences in the OC factor (right y axis) are quantified as the main effect of the OC factor score and its interaction with the experimental factors. The GLMM shows no significant effect of the OC factor per se (P = 0.318), or differences in weighting the current number of draws (P = 0.155), the horizon condition (P = 0.320), or the previous evidence (P = 0.272) in making a decision. Conversely, it shows that those with higher OC factor scores weight ΔES less in making a decision (P = 0.013, highlighted by shaded area). e, Illustration of the association between ΔES weighting and the OC factor (using individual GLMs per participant; N = 92 after outlier exclusions were applied to all beta coefficients in the GLM, defined as any participant with any beta values 1.5× IQR above the third quartile or below the first quartile). f, OCD patients with stronger obsession symptom severity had more attenuated weighting of ΔES (ρs = −0.475, P = 0.012, N = 27). All tests are two-sided and not corrected for multiple comparisons.
We first replicated the findings related to the cognitive contributors to information gathering per se (Fig. 2c). In a GLMM similar to the one above (see Methods for the equation), we found that both ΔES (β = 0.830, P < 0.001, 95% CI 0.791–0.869) and previous evidence strength (β = 0.516, P < 0.001, 95% CI 0.475–0.557) positively predicted when participants would make a decision. Likewise, we replicated the effect of current draw number (β = 2.367, P < 0.001, 95% CI 2.232–2.503), supporting the notion of an emerging decision urgency. A noteworthy extension here was that this lab version featured two different finite horizons, meaning that each game would end after a probabilistically predetermined number of draws (either 4–8 in the ‘short’ or 10–14 in the ‘long’ horizon condition). We found that this horizon condition significantly predicted participants’ decision-making (β = 1.602, P < 0.001, 95% CI 1.457–1.747). In other words, participants are not only more likely to commit to a decision if they (recently) received more evidence for one of the options and if they had already gathered information for longer, but also if they are incentivized to trade off speed for accuracy.
Attenuated ΔES is also linked to OC symptoms in a transdiagnostic clinical sample
Next, we investigated how these predictors link to an OC spectrum in a sample consisting of a clinical group of OCD patients (N = 29) and generalized anxiety disorder (GAD) patients (N = 17), as well as non-patient cohorts with varying degrees of OC symptoms (high OC: N = 20, low OC: N = 20, controls: N = 19). This recruitment strategy allowed for a broad sampling of OC symptoms not only in OCD patients but also in related disorders as well as non-clinical high-OC participants. A similar approach has previously revealed striking dissociations between transdiagnostic dimensions and traditional diagnostic categories56. To derive a transdiagnostic and dimensional OC symptom measure, we factor analysed key psychiatric questionnaires collected across the entire sample, yielding a 3-factor solution where one factor primarily reflected OCD and related symptoms (Methods and Supplementary Fig. 3). On average, OCD patients ranked highest on this factor, followed by high-compulsive non-patients and GAD patients, while low-compulsive non-patients and controls ranked the lowest (Fig. 2b).
Investigating how this transdiagnostic OC factor linked to cognitive contributors of information gathering, we again replicated the effect of ΔES (ΔES × OC: β = −0.049, P = 0.013, 95% CI −0.088 to −0.010; Fig. 2c,d), such that high-OC participants weighted the update in evidence strength less in their decision to declare for a stimulus. Similar, albeit less robust, effects were seen when comparing OCD patients to matched controls alone (β = −0.102, P = 0.046, one-sided test, 95% CI −0.221 to 0.017). In addition, by fitting a separate GLMM predicting p(decide) from the evidence at draw d, d−1 and d−2 relative to the chosen option per participant, we again found a decreasing contribution of current evidence for lagged draws (ESd: β = 0.531, P < 0.001, 95% CI 0.500–0.562; ESd−1: β = 0.232, P < 0.001, 95% CI 0.205–0.259; ESd−2: β = 0.246, P < 0.001, 95% CI 0.224–0.267; Supplementary Fig. 1b), and that the difference in the beta weight for the evidence at draw d and the beta weight for the evidence at draw d−1 correlated negatively with OC symptoms (ρs = −0.187, P = 0.034, one-sided test, 95% CI −0.395 to 0.042).
As was the case for the previous population sample, we find no significant main effect of OC symptoms (β = 0.055, P = 0.318, 95% CI −0.053 to 0.162, ΔBIC = 57.641, comparing the GLMMs with and without the OC main effect), or any significant interaction with the current number of draws (Ndraws × OC: β = −0.098, P = 0.155, 95% CI −0.234 to 0.037, ΔBIC = 90.411, comparing the GLMMs with and without any 2- or 3-way Ndraws × OC interaction effects) or with the horizon condition (Horizon × OC: β = −0.074, P = 0.320, 95% CI −0.219 to 0.072, ΔBIC = 233.312, comparing the GLMMs with and without any 2- or 3-way horizon × OC interaction effects). In this sample, the interaction of OC symptoms with previous evidence was not significant (ESd−1 × OC: β = −0.023, P = 0.272, 95% CI −0.064 to 0.018, ΔBIC = 203.803, comparing the GLMMs with and without any 2- or 3-way ESd−1 × OC interaction effects), in contrast to the first smartphone sample. In sum, this further supports the notion that an over-reliance on most recent information during information gathering is attenuated in high-OC participants, including patients.
Attenuated ΔES is primarily linked to obsessions in OCD patients
Lastly, we asked whether an attenuated weighting of ΔES was linked to symptom severity, especially in patients with OCD. To this end, we correlated the scores from the clinical Y-BOCS interview (only conducted in OCD patients by trained researchers) with the ΔES effect on information gathering. Interestingly, we found that attenuated weighting of ΔES was significantly associated with symptom severity for obsessions (ρs = −0.475, P = 0.012, 95% CI −0.696 to −0.166, Bayes factor for the alternative hypothesis BF10 = 0.101, robustness region RR > 0.138), but not for compulsions (ρs = −0.205, P = 0.306, 95% CI −0.548 to 0.198, Bayes factor for the null hypothesis BF01 = 5.457, RR > 0.268; total score trended at ρs = −0.342, P = 0.081, 95% CI −0.616 to −0.016, BF10 = 0.133, RR > 0.187). This difference in effect size suggests that the attenuated weighting of recent information may be primarily linked to an obsession dimension, in line with the idea that underweighting of information about the unlikely nature of the obsessions promotes their perpetuation.
To assess how generalizable this effect was, we went back to the complete samples for both in-person and online datasets, and found that ΔES attenuation related to obsessions from the OCI-R questionnaire both in the smartphone population sample (ρs = −0.083, P < 0.001, 95% CI −0.1144 to −0.0535) as well as in the lab-based sample (including patients and non-patients: ρs = −0.226, P = 0.030, 95% CI −0.416 to −0.013). However, these effects were less specific (Supplementary Fig. 4), possibly due to range effects across the wide spectrum of patients and non-patients.
Thus, our findings indicate that attenuated ΔES is not only associated with OC symptoms across a wide OC spectrum but also linked to symptom severity of obsessions within OCD patients.
The brain processes decision-relevant factors sequentially
To better understand the brain processes relevant to information gathering, we concurrently acquired MEG data from all participants in our in-person study. Here, the validity of our behavioural model leads to the prediction that we could decode (5-fold cross-validated iterative lasso regression with whole-brain MEG57, see Methods) each of the relevant cognitive factors used to predict information gathering, as described above, namely: ΔES, total evidence strength at draw d−1 ESd−1, horizon and number of draws.
We found evidence for a representation of each of the aforementioned factors in our MEG data (Fig. 3a; P < 0.05, cluster-based permutation tests at t > 3.1), supporting the idea that the brain tracks these variables. Interestingly, these representations were instantiated at different times within a trial following a sequential order, where pre-existing latent factors were represented first, and novel information that needed to be computed on the basis of these latent representations was represented later in a trial. Specifically, we found that the representations of the number of draws, horizon and the total evidence at draw d−1 preceded the representation of ΔES (comparison of maximum decodability timepoint; number of draws vs ΔES: t(104) = −14.773, P < 0.001, Cohen’s d = −1.983, 95% CI −2.311 to −1.651; horizon vs ΔES: t(104) = −5.387, P < 0.001, Cohen’s d = −0.777, 95% CI −1.056 to −0.497; ESd−1 vs ΔES: t(104) = −6.568, P < 0.001, Cohen’s d = −0.947, 95% CI −1.231 to −0.662; one-sided tests; Fig. 3b). Importantly, the representation of the currently shown evidence (quantified as the relative evidence for the two alternative response options in the current draw) also preceded the representation of ΔES (t(104) = −8.711, P < 0.001, Cohen’s d = −1.202, 95% CI −1.494 to −0.907; one-sided test; Supplementary Fig. 5c), as would be expected when considering that ΔES is computed on the basis of both previous total evidence and current evidence. In addition, we observed that latent variables that are not dependent on currently presented information manifest as a more sustained representation (for example, horizon), consistent with the fact that they do not need to be computed on the fly after each draw. Thus, our findings demonstrate a cascade of neural information processing and representation that aligns with the evolution of cognitive processes we deciphered from participants’ behaviour.
a, Cognitive factors are predicted from MEG activity using lasso regression following an iterative method described previously57 (N = 105). The decoded variables of current number of draws, horizon condition, cumulative evidence strength from the start of the game until the previous draw ESd−1 and ΔES were all significantly decodable (P < 0.05, two-sided cluster-based permutation test at threshold t > 3.1; denoted by horizontal lines). Peak decodability timepoints, quantified as the group average of individual maximum values, are indicated by lollipops. Correlation values were normalized by dividing these by the maximum correlation value for visualization purposes. Note that all statistical tests were performed before normalization. Shaded areas represent the standard error. b, Maximum decodability is achieved earlier in the trial for factors independent of the currently presented stimulus (current number of draws, horizon condition, the cumulative evidence from the start of the game until the previous draw) than for those dependent on current evidence processing (ΔES, all P < 0.001, N = 105). Boxplots show the median (centre line) and IQR; whiskers extend to the largest value within 1.5× IQR. Pairwise t-tests are one-sided and not corrected for multiple comparisons.
Reduced neural signatures of ΔES in high-OC participants mirror behavioural findings
Having established the decodability of cognitive contributors to information gathering, we next tested whether these representations were altered in high-OC participants. More concretely, given a reduced effect of ΔES on behaviour in the latter group, we hypothesized a reduced strength of this representation in the brain. On this basis, we assessed how transdiagnostic OC symptoms relate to the decodability of ΔES by predicting decodability at each timepoint from the three factor scores in individual GLMs. Using cluster-based permutation testing, we found that ΔES representations were attenuated in high-OC participants, whereby two time periods reached significance, namely, 420–470 ms and 530–560 ms post stimulus onset (Fig. 4a,b; P < 0.025 for a two-sided test, cluster-based permutation tests at t > 2.1). Neither decodability of the variable nor individual differences in its decodability appear to be driven exclusively by processes related to the draw wherein participants actually declare, as opposed to the longer-term evidence accumulation process, as both decodability and individual differences associated with the OC factor remained consistent after omitting the data for the draw immediately preceding the response (Supplementary Fig. 6a). Likewise, the results held for the cumulative evidence strength for the current majority ESd, as opposed to the update in evidence strength ΔES (Supplementary Fig. 6b). When analysing other variables, we did not find any links with the OC spectrum. This means that, based on behavioural findings, the ΔES representation is attenuated in participants with high OC symptoms.
a, ΔES can be decoded from MEG activity using an iterative multivariate lasso regression (left y axis: decodability, measured as the correlation between trained and tested data at each timepoint in the epoch, plotted in blue as group-level mean decodability, with shaded area representing the standard error of the mean, N = 105; horizontal lines denote significant time clusters (P < 0.05 in a two-sided cluster-based permutation test at threhold t > 3.1)). Individual differences in OC factor predict significantly decreased decodability within 420–470 ms and 530–560 ms post stimulus (right y axis: beta coefficient for the OC factor predicting the decodability of ΔES at each timepoint in the epoch from the 3 factor scores in a GLM per timepoint, plotted in orange, N = 105; horizontal lines denote significant time clusters (P < 0.05 in a two-sided cluster-based permutation test at threshold t > 2.1)). b, Illustration of the attenuated ΔES decodability within 420–470 ms (N = 105), plotted here by categorical group, that is, for participants from the general population with low and high obsessive–compulsive scores (‘low C’ and ‘high C’, respectively), healthy controls, participants with GAD and participants with OCD. Note that this is merely a visualization of the significant difference found in a, showing individual data points. c, To investigate the spatial distribution of the sensors contributing to the decodability of ΔES within the time windows where we find individual differences (420–470 ms post stimulus onset), we used searchlight analyses (N = 105). ΔES are represented in a wider network comprising frontal, mediocentral and occipital areas (left). The attenuation of ΔES representations in high-OC participants was primarily driven by mediofrontal sensors, supporting the idea of altered ΔES processing in mediofrontal areas in high-OC participants (right).
Attenuation of ΔES representations arises from mediofrontal areas
Lastly, we investigated the likely regional origins of an attenuated OC spectrum-related ΔES signal. While we found a relatively widespread activation pattern for ΔES per se (Fig. 4c, left panel), the attenuation in high-OC participants was pronounced in mediofrontal areas (Fig. 4c, right panel; see Supplementary Fig. 7 for the results for the second time window and Supplementary Figs. 8 and 9 for alternative methods). This suggests that the observed indecisiveness may arise from altered ΔES processing in mediofrontal areas.
Discussion
Indecisiveness and doubt are common experiences in everyday life but can reach debilitating extremes in mental health conditions, notably OCD. Some aspects of these symptoms can be quantitatively captured as excessive information gathering, yet their roots remain unclear4,10,16,18,19,21,22,23. Here, using a crowd-sourced behavioural dataset in combination with data from a novel neuroimaging task, we show that attenuated weighting of most recent information by means of evidence strength updates (ΔES) drives altered information gathering across an OC spectrum. Our finding of attenuated ΔES was evident at the level of both behaviour and brain, and was also evident dimensionally and transdiagnostically across both patients and those with non-clinical symptoms.
Excessive information gathering has been extensively investigated in relation to OCD and OC-like symptoms4,10,16,18,19,21,22,23, yet the precise mechanisms remain unclear. Much previous work on information gathering (including our own) has focused on altered decision criteria, such as collapsing decision thresholds or urgency signals4,34, ignoring the possibility that evidence integration (that is, how information gathered at different timepoints is combined) might be implicated. We show that information integration shows a strong recency effect, where most recent information (by means of ΔES) exerts a much bigger impact on deciding, and when to decide, than more remote information. A recency weighting has been argued to confer an advantage under certain signal strength conditions58 or in dynamic contexts where the latest information is most representative59. It is also well aligned with similar recency biases found across domains that span perception, memory and mood37,38,44,46. Interestingly, the nonlinear accrual of evidence has been widely accepted in the reward learning domain60,61, and the present results indicate that similar mechanisms are also at play during information gathering. Crucially, ΔES is also represented in the brain, as a late electrophysiological component peaking at ~920 ms after new information is revealed. Importantly, this ΔES representation peaks following the representation of other decision variables from which it is computed (specifically, newly revealed information and previous evidence), further supporting a putative sequential cascade of information processing62,63.
A mediofrontal localization of ΔES decodability is strongly reminiscent of that found in previous studies on information sampling during value-based decisions, where consistency of current evidence with a participant’s belief is signalled by anterior cingulate cortex (ACC) activation in single-cell recordings from monkeys64 and in an equivalent human magnetic resonance imaging paradigm65. In line with the strong effect of ΔES on the probability of committing to a decision, ACC signals peak immediately before the final decision in monkeys64. While the signal we show here does not depend crucially on the brain activity immediately before commitment to a decision (Supplementary Fig. 6a), it should be noted that this possibility would be best addressed using a response-locked analysis, as opposed to the stimulus-locked approach we chose, and remains a question for future studies. More broadly, the localization of the ΔES decodability effect is in line with accumulating evidence for alterations in fronto-striatal circuitry in OCD, whereby these findings are often related to heightened error monitoring and altered reward prediction error responses48,66,67.
The ΔES measure may also fit within a broader belief-updating or predictive coding framework, provided the evidence strength for the current majority maps onto a participant’s current belief. If this assumption holds, ΔES can be cast as a type of prediction error (PE), in that it captures the deviation from the expectation formed in the previous draw. While different PEs have been described beyond the classic reward prediction errors, from perceptual to cognitive44,45, ΔES is different as it is not based on explicit feedback for a completed action but rather on novel information available during the decision-making process. As such, its relationship to a traditional conceptualization of PEs needs to be established. This also applies to its implementation in the brain—while classic reward PEs have been associated with dopamine signalling, the extent to which dopamine influences information gathering is less clear34,68,69, and other candidate neurotransmitter systems, such as noradrenaline or serotonin, may conceivably contribute to the formation of ΔES34,36.
Critically for this study, we find attenuation of ΔES in people with (clinically and non-clinically) high OC symptoms and we propose that it is the likely driver underlying excess information-gathering behaviour found by draws to a decision4,10,18,19,20,21,22,23 (although others have reported fewer draws to a decision in OCD16,17). Importantly, we find a closer association between the ΔES weighting and the OC dimension, than with draws to a decision alone, supporting the notion that this is the driving factor, while the latter behavioural marker is probably also under the influence of many other cognitive contributors, such as stringent constraints, urgency signals and clear speed–accuracy trade-offs (as found in our MEG experiment). The finding of attenuated ΔES in OC, without a significant effect on draws to a decision, in the more constrained MEG experiment (cf. horizon conditions) also speaks to this, and accords with previous literature showing intact adjustments in speed-accuracy trade-offs in OCD patients4,22. Our finding of an attenuated neural representation of ΔES in high-OC participants also suggests that it may provide a sensitive biomarker. Indeed, the mediofrontal regions which appear most implicated in the attenuation of the ΔES representation, provide a tantalizing target for more focused future investigations, as well as for potential therapeutic interventions (for example, using electrical or magnetic stimulation70).
A diametrically opposite information-gathering bias prominent in mental health disorders is jumping to conclusions (JTC), particularly seen in patients with psychosis9,71,72,73. Whether this JTC bias is driven by similar or distinct neurocognitive mechanisms remains unexplored. It is interesting, however, that similar imbalanced weighting hypotheses have been proposed for JTC74,75,76, along with the notion of altered PE processing in both psychosis and schizophrenia77.
In this study, we focus on transdiagnostic OC effects within the general population as well as across different clinical diagnoses. While many of our (primarily behavioural) effects hold in more traditional patient vs controls subsamples, these tended to be less strong (for example, MEG differences did not reach significance, data not shown), possibly due to a much smaller sample size. Conversely, we find that OC symptoms are not exclusive to OCD patients but are also prominent in some GAD patients and non-clinical, high-OC participants in our sample (Fig. 2b). While data on indecisiveness in clinically diagnosed GAD is, to the best of our knowledge, lacking, our findings align with previous reports of high indecisiveness in undiagnosed individuals with probable GAD78 as well as in those with high levels of symptoms of anxiety and depression79,80. Thus, we propose that indecisiveness and excessive information gathering may represent transdiagnostic symptoms not limited to OCD.
Nevertheless, it is yet to be determined whether these information-gathering biases are best described using transdiagnostic approaches, as adopted in previous studies where dimensional measures outperformed traditional diagnostic categorical comparisons56. Specifically, it remains to be established, first, whether our findings hold in traditional OCD patient case–control studies, and second, to what extent these hold across diagnoses. Indecisiveness can also be a symptom of both depression and dependent personality disorder12, yet understanding an across-diagnosis overlap in symptom phenomenology and underlying mechanisms will require future studies that include these populations alongside those with OCD and GAD.
Further, indecisiveness manifests in different ways and with different degrees of severity, and our highly abstract task does not capture this complexity. The strength of this abstraction is that it enables the isolation of fundamental evidence accumulation processes and meets neuroimaging constraints, yet it comes at the cost of ecological generalizability. While previous work has linked increased real-life information gathering to OC symptoms, and the current paradigm shows robust and replicable effects in relation to OC symptoms, it will be important to establish how well our findings translate to real-life settings. Nevertheless, we previously found that information gathering in a version of this task without any sampling constraints is linked to indecisiveness (as rated in the clinician-rated Y-BOCS52). At the opposite end of the spectrum, JTCs in a similar paradigm are consistently linked to clinically reported delusions, indicating good ecological validity for this symptom. However, a more targeted ecological validation of the paradigm could help refine our understanding of the throughline from our findings to differences in real-life information gathering in OCD.
In sum, within a crowd-sourced sample as well as a lab-based sample including clinical patients, we provide evidence of an attenuated weighting of ΔES, both at the behavioural and neural level, linked to transdiagnostic OC symptoms. This novel neurocognitive marker highlights the potential contribution of altered evidence integration to excessive information gathering and indecisiveness across health and mental illnesses, potentially opening avenues to improve behavioural and brain-based interventions for these pervasive symptoms.
Methods
We collected two types of data: a non-clinical sample completed a gamified information-gathering task using the Brain Explorer app for handheld devices as well as the OCI-R. A second in-person sample including OCD and GAD patients together with matched controls and low- and high-compulsive non-patients completed an analogous information-gathering task while their MEG activity was recorded. In both online and in-person studies, human participants completed variants of information-gathering tasks81 in which they were asked to indicate which of two possible stimuli was more plentiful. They had the choice to either continue sampling additional information or stop and make a binary choice between the two stimuli. Classically, the key metric in this task is ‘draws to a decision’, that is, the number of draws or samples the participant chooses to view before deciding for either stimulus. We experimentally varied the amount of current and past information and the maximum allowed draws to a decision to assess their impact on information gathering and the relationship with OC symptoms.
Participants
Smartphone population sample
At the time of analysis, 8,670 participants completed the gamified information-gathering task on Brain Explorer (www.brainexplorer.net), of whom 6,743 had also completed the OCI-R (data collected between November 2020 and August 2023). We analysed each participant’s first complete game (N = 15 trials) that fulfilled the inclusion criteria, which comprised a mean number of draws between 2 and 23 over the 15-trial game, accuracy in the binary choice between the two stimuli above 80%, and a minimum of 3 unique values in the total set of 15 numbers of draws to a decision, resulting in a final sample size of 5,237.
Additional exclusion criteria were applied to conduct a further analysis on participants’ confidence ratings in their binary choice between the two possible stimuli, which they provided immediately after each decision, on a slider ranging from 0 to 100. These criteria comprised a mean confidence rating over the 15-trial game between 15 and 98, and a minimum of 3 unique values in the total set of 15 confidence ratings, yielding a sample size of 4,302. Here, participants with any outlier beta coefficients in the GLM (see below), defined as values 1.5× interquartile range (IQR) above the third quartile or below the first quartile, were removed, resulting in a final sample size of 3,293.
In-lab MEG sample
A total of 115 participants completed the MEG study (data collected between April 2015 and August 2018). OCD and GAD patients were recruited through NHS services, charities and advertisements. To ensure comparable backgrounds, controls were recruited in areas of similar socioeconomic status to the patients’. Exclusion criteria were current use of antipsychotic medication, severe learning disability and comorbidities including current psychosis or bipolar disorder, substance abuse disorder and personality disorders other than obsessive–compulsive personality disorder, as well as self-disclosed autism spectrum, tic or Tourette disorder. As part of the recruitment process through NHS clinics, clinicians pre-screened potential participants on the basis of the inclusion criteria. Those considered eligible were subsequently screened by a study staff member, focusing on existing diagnoses and MEG inclusion criteria. All participants underwent an extensive clinical interview, conducted by experienced researchers at UCL, comprising a structured clinical interview (SCID)82, the Y-BOCS interview15 (which assesses clinician-rated symptom severity, only for OCD patients) and 7 questionnaires, namely: the Barratt Impulsiveness Scale (BIS83), Beck Depression Inventory II (BDI-II84), Intolerance of Uncertainty Scale (IUS85), OCI-R51, PI-WSUR86, State and Trait Anxiety Inventory (STAI87) and Frost Multidimensional Perfectionism Scale (FMPS88). The final decision to include a participant was based on a discussion of the results of the interview among the research group members. All participants with OCD and GAD fulfilled the corresponding diagnostic criteria for only one of these disorders12.
Data from the high- and low-OC groups have been reported previously89. Briefly, participants were recruited from a population-based sample of young people (U-CHANGE study; www.nspn.org.uk)90,91 based on the revised Padua Inventory–Washington State University Revision (PI-WSUR86).
Two participants were excluded because they did not meet the clinical criteria for GAD, one participant was excluded because of comorbid OCD and GAD, one was excluded due to fully missing questionnaire data. One additional participant was excluded due to an excessive number of non-decision trials (~55%, compared to a mean of ~20% across the remaining participants), resulting in below-chance performance. Five further participants were excluded due to technical difficulties and/or poor MEG data quality. A total of 105 participants were thus included in the analysis (see Supplementary Fig. 2 for a visualization of the exclusion process). The final groups comprised OCD patients (N = 29), GAD patients (N = 17), matched controls (N = 19), and non-clinical high-compulsive (N = 20) and low-compulsive groups (N = 20). Group demographics, medication percentages and comorbidities for the final sample included in the analysis are reported in Supplementary Table 1.
The study was approved by the Research Ethics Committees at University College London (UCL; protocol number 16711/002) and the NHS (protocol number 15/LO/1361), and all participants provided written informed consent.
Task
Participants in both online and in-person studies completed an information-gathering task where their goal was to determine which of two possible stimuli was more plentiful. Each draw consisted of a new sample of the stimuli, after which the participant made the decision to either continue sampling additional information or stop and make a binary choice between the two stimuli.
Smartphone population sample
During each game (Fig. 1a), participants viewed a grid of 25 locations, each of which could contain one of two possible stimuli, here gems of different shape and colour. They could view which of two possible gems was hidden at a given position by tapping on it on the grid, or make a binary choice indicating which of the two gems was more plentiful by tapping on the corresponding gem at the bottom of the screen. Uncovered gems remained on the screen until the participant made a binary choice. There were no constraints or penalties on the response time or the number of draws. After each decision, participants gave a confidence rating using a slider and then received feedback on their performance in the form of a reward or a loss. They won 100 points for every correct choice and lost 100 points for every incorrect choice. Each session consisted of interactive instructions, which included a simplified mock game, followed by the 15 games that were included in the analysis.
In-lab MEG sample
To better disentangle cognitive contributing factors and their neural representations, we designed an entirely novel task optimized for MEG. In this version, the two possible products were always blue zircons or gold nuggets, represented respectively by blue and yellow squares on a grey background. Evidence samples consisted of 5 squares at a time which varied in their proportion of yellow to blue, and were presented automatically every 1,250 ms, that is, participants did not actively gather information as in the Brain Explorer task. Participants were instructed that samples were recovered by a robot which may run out of batteries, such that here there was a variable finite number of draws per game, termed the horizon. The maximum number of draws varied between 4 and 8 in the short-horizon condition, and between 10 and 14 in the long-horizon condition. These conditions were cued by a different coloured frame (green or pink, assigned randomly for each participant). There were 80 games in each horizon condition, randomly intermixed. Once participants committed to a choice, they made their response by pressing one of two buttons on a MEG-compatible button box. The length of the draw sequence was presented in full independently of when the participant made their decision; in other words, if the participant made a decision before the last draw, the draw sequence continued until it reached its predetermined end. Participants received thorough instructions and completed 12 games as part of a practice block that included the horizon manipulation before beginning the experiment, such that they acquired a good intuition about the horizon length already during the practice block. They received 2 points for a correct response, lost 2 points for an incorrect response, and lost 1 point for non-decisions. They could track their performance and corresponding rewards by means of a blue bar at the bottom of the screen, which started at the 50% mark. They were told that each time the bar reached the maximum threshold of 100%, they would receive an additional £0.50 at the end of the experiment.
OC symptoms assessment
Smartphone population sample
Participants completed the OCI-R51 within the Brain Explorer app.
In-lab MEG sample
Participants completed seven self-report questionnaires assessing psychiatric symptoms: BIS83, BDI-II84, IUS85, OCI-R51, PI-WSUR86, STAI87 and FMPS88. Participants additionally completed the matrix and vocabulary subtests of the Wechsler Abbreviated Scale of Intelligence (WASI92), and Edinburgh inventory93 data were collected to assess handedness. Missing responses in the questionnaires were imputed with the mean response, provided the number of missing items did not exceed 20% of the items in the corresponding scale.
We conducted an exploratory factor analysis (EFA) on the 210 individual items that compose the seven psychiatric symptom questionnaires to reduce the dimensionality of the data. The EFA solution was estimated using maximum likelihood and oblimin rotation based on the heterogeneous correlation matrix to account for both continuous and binary correlations using the psych package in R94. The number of factors was determined on the basis of the Cattell–Nelson–Gorsuch index95, that is, the eigenvalue at which the greatest difference between two successive slopes occurs, visible as a sharp elbow in a scree plot, implemented with the nFactors package in R96. Although the sample size resulted in a relatively low ratio of observations to items, adequate sample size for factor analyses ultimately depends on the correlation structure in the data. Indeed, in this case the Kaiser–Meyer–Olkin (KMO) Measure of Sampling Adequacy was 0.94 and Bartlett’s Test of Sphericity was significant (P < 0.001), in line with published recommendations97. More detailed results of the EFA can be found in Supplementary Information under ‘In-lab MEG sample factor analysis’.
Behavioural analysis
Smartphone population sample
We correlated the primary index of information gathering, namely, the number of draws to a decision, as well as accuracy, with the OCI-R scores using Spearman’s correlations.
We then investigated the underlying processes of information gathering by fitting participants’ draw-wise binary choice to sample information or not to a logit-linked binomial general linear mixed model (GLMM) predicting the probability to make a decision from the experimental factors, OCI-R scores and the interactions of OCI-R scores with the experimental factors, as well as random intercepts and slopes for each experimental factor.
To quantify the evidence strength (ES), we define the following variables. The cumulative ES for stimulus A at a given draw is the cumulative sum of instances of A over all draws, for example, all yellow gems seen until the current draw d:
and analogously for stimulus B:
We define the evidence strength for the current majority stimulus, that is, that which is more abundant on the current draw, ESd,majority as ESd,A when ESd,A > ESd,B and as ESd,B when ESd,B > ESd,A. Vice versa, ESd,minority is defined as ESd,A when ESd,A < ESd,B and as ESd,B when ESd,B < ESd,A.
The total evidence strength at a given draw is defined as the difference between the evidence for the stimulus in the current majority and that for the stimulus in the current minority:
Note that ESd,total is therefore by definition always positive.
ESd−1 is thus the cumulative relative evidence strength for the current majority stimulus until the previous draw d−1. Finally, we define the update in evidence strength as
Thus, experimental factors consisted of the cumulative evidence strength on the previous draw ESd−1, the update in evidence strength ΔES, and the current number of draws Nd.
The GLMM was thus defined as
with a logit link function (main effects were included in the analysis but are omitted here for legibility).
In a follow-up analysis, we tested the recency effect suggested by these results in a more targeted manner by fitting a GLMM predicting p(decide) from the current evidence at draw d, d−1 and d−2:
with a logit link function, whereby the evidence ESd,total is defined relative to the chosen option as opposed to the current majority:
To corroborate the individual differences we found in the recency effect associated with OC symptoms, we derived a metric of recency bias for each participant by fitting a separate generalized linear model (GLM) to each participant analogous to the GLMM
with a logit link function, and taking the relative difference between the beta coefficients for ESd,total and ESd−1,total. We then correlated this index of recency bias with the OCI-R score.
Please note that for simplicity, we omit the subscript ‘total’ from ESd,total and ESd−1,total in the main text and figures.
Lastly, we conducted a mediation analysis evaluating how the observed correlation between OCI-R and sensitivity to current evidence relates to the correlation between OCI-R and confidence ratings, as opposed to the probability to commit to a decision, using the M3 mediation toolbox in MATLAB (https://github.com/canlab/MediationToolbox)54,55, as in previous work in our research group98,99,100. To perform this analysis, we obtained a measure of sensitivity to ΔES in deciding to commit to a choice for each individual participant by fitting a separate GLM to each participant analogous to the GLMM above
with a logit link function.
We then tested whether the individual beta coefficients for ΔES mediated the negative association between mean confidence and OCI-R scores. In other words, we take OCI-R as the independent variable, confidence as the outcome variable, and individual beta coefficients for ΔES as the mediating variable. The c path represents the total effect of OCI-R on confidence (without controlling for the mediating variable); c′ is the direct effect of OCI-R on confidence (controlling for the individual beta coefficients for ΔES); b is the effect of the individual beta coefficients for ΔES on confidence, controlling for OCI-R; a is the effect of OCI-R on the individual beta coefficients for ΔES; and the product ab is the indirect effect of OCI-R on confidence through the individual beta coefficients for ΔES. We tested for statistical significance using a bootstrap test with 10,000 bootstrap samples (https://github.com/canlab/MediationToolbox)54,55.
The rationale for the particular model we tested (OCI-R → ΔES → confidence) was motivated by evidence from the field of metacognition showing a directional link between evidence strength and confidence (that is, stronger evidence strength preceding higher confidence ratings, as reviewed recently101). Likewise, there is also evidence for low confidence in OCD53. With respect to the directionality assumptions, note that here we measure confidence in the restricted context of this particular task as opposed to a trait-like measure of confidence. We therefore asked whether the observed lowered confidence associated with high OCI-R scores is mediated through ΔES, rather than whether confidence in a task would be causal with respect to a stable mental health trait (OCI-R).
In-lab MEG sample
We performed an analogous analysis, while accounting for sample and task differences. As stated above, we used the OC factor score to quantify variability in OC symptoms. In addition, we adapted the GLMM by including the horizon condition and the interactions between the horizon and all other regressors. The interaction of ΔES with the horizon was omitted in the participant-level GLMs, as the beta coefficient was not significantly different from zero.
The GLMM was thus defined as:
with a logit link function (main effects were included in the analysis but are omitted for legibility).
MEG analysis
Data acquisition and preprocessing
Participants sat upright in the scanner while MEG was recorded continuously using a 275-channel system, whereby sensor MLO42 was missing for all participants, and in addition, sensor MRC12 was missing for 59 participants. Empty channels were added in these cases to keep the dimensionality of the MEG data consistent across participants. Bad-quality channels were removed and not interpolated. Physiological and technical artefact removal was performed manually using Fieldtrip. The data was downsampled to 100 Hz and high-pass filtered at 0.5 Hz to remove slow drift effects. Data were segmented into 1,350-ms segments from −100 to 1,250 ms relative to stimulus onset. MEG data were z-scored by block and channel (univariate normalization).
Decoding analysis
Whole-brain MEG activity was used to decode ΔES using an iterative multivariate lasso regression approach57,102,103. Here we estimated the lasso regression 2,000 times, using a random subset of 50 sensors on each iteration. While the method was originally developed to determine which sensors contribute most to the decoding, its primary benefit in this study is that it increases reliability of the individual differences analysis relative to the results from a single iteration of a multivariate lasso regression, by leveraging the many iterations of the lasso (see also Supplementary Information ‘Alternative analyses of the localization of the ΔES representation’). The regularization parameter λ was set at 0.01 using 5-fold cross-validation. The value for the regularization parameter was tuned by optimizing the cross-validation performance of an orthogonal variable, namely, the current evidence for yellow (correlation between current evidence for yellow and ΔES: r = 2.755 × 10−4, P = 0.920), and was fixed for all subsequent whole-brain analyses. The initial optimization procedure was conducted using a non-iterative multivariate lasso regression using all 273 channels, yet note that the decodability using the iterative method remains extremely similar for a range of regularization parameter values in follow up analyses (Supplementary Fig. 4d).
The correlation between the tested and trained data constituted the measure of decodability. Non-parametric cluster-based permutation testing (CBPT104) was used to assess significant differences in decodability timecourses. CBPT is commonly used in M/EEG analysis to test for significant differences in a given timeseries while accounting for multiple comparisons (for example, in previous work105,106). Briefly, CBPT takes the temporal structure of the data into account and uses permutation tests (here, N = 1,000) to build a null distribution of clusters above a given size (here, threshold t = 3.1) within the same temporal structure, enabling the evaluation of whether an effect of interest exceeds the expected cluster size under the null.
Individual differences in decodability were assessed by predicting the decodability at each timepoint from the 3 factor scores in individual GLMs. Significant differences in the beta regressor timecourses were assessed by non-parametric CBPT (here, threshold t = 2.1). Correlation values were normalized by dividing these by the maximum correlation value for visualization purposes, yet note that all statistical tests were performed prior to normalization.
To investigate which brain areas underpin the decoding, we used a searchlight analysis, such that the lasso regression was estimated 273 times, that is, for each sensor and its direct neighbours, determined using the Fieldtrip ‘ft_prepare_neighbours’ function. We chose this method instead of the iterative multivariate lasso regression described above for the localization of the effect because we and others102 have found sensor contribution maps derived from the iterative multivariate lasso regression to be very diffuse (see Supplementary Information ‘Alternative analyses of the localization of the ΔES representation’). Given the stark difference in the number of sensors used in the searchlight analysis, we first fine-tuned the value of the regularization parameter λ in the same way as above and set it to 0.001. We then calculated the mean decodability per sensor within a given time window of interest to generate sensor contribution maps. To explore individual differences in the sensor contribution maps, we fit a GLM predicting the sensor contribution per sensor averaged across the time window of interest from the 3 factor scores. To keep the sensor subsets constant across participants, sensors MLO42 and MRC12 were omitted for all.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
Behavioural and summary neuroimaging data are publicly available on OSF at https://osf.io/fks97 (ref. 107). Source data are provided with this paper.
Code availability
Analysis scripts are publicly available on OSF at https://osf.io/fks97 (ref. 107).
References
Bogacz, R., Wagenmakers, E.-J., Forstmann, B. U. & Nieuwenhuis, S. The neural basis of the speed–accuracy tradeoff. Trends Neurosci. 33, 10–16 (2010).
Schouten, J. F. & Bekker, J. A. M. Reaction time and accuracy. Acta Psychol. 27, 143–153 (1967).
Bogacz, R., Hu, P. T., Holmes, P. J. & Cohen, J. D. Do humans produce the speed–accuracy trade-off that maximizes reward rate? Q. J. Exp. Psychol. 63, 863–891 (2010).
Hauser, T. U. et al. Increased decision thresholds enhance information gathering performance in juvenile obsessive-compulsive disorder (OCD). PLoS Comput. Biol. 13, e1005440 (2017).
Phillips, L. D. & Edwards, W. Conservatism in a simple probability inference task. J. Exp. Psychol. 72, 346–354 (1966).
Hemsley, D. R. & Garety, P. A. The formation of maintenance of delusions: a Bayesian analysis. Br. J. Psychiatry 149, 51–56 (1986).
Moritz, S. et al. A two-stage cognitive theory of the positive symptoms of psychosis. Highlighting the role of lowered decision thresholds. J. Behav. Ther. Exp. Psychiatry 56, 12–20 (2017).
Ward, T. & Garety, P. A. Fast and slow thinking in distressing delusions: a review of the literature and implications for targeted therapy. Schizophr. Res. 203, 80–87 (2019).
McLean, B. F., Mattiske, J. K. & Balzan, R. P. Association of the jumping to conclusions and evidence integration biases with delusions in psychosis: a detailed meta-analysis. Schizophr. Bull. 43, 344–354 (2017).
Fear, C. F. & Healy, D. Probabilistic reasoning in obsessive–compulsive and delusional disorders. Psychol. Med. 27, 199–208 (1997).
Koutsouleris, N., Hauser, T. U., Skvortsova, V. & Choudhury, M. D. From promise to practice: towards the realisation of AI-informed mental health care. Lancet Digit. Health 4, e829–e840 (2022).
Diagnostic and Statistical Manual of Mental Disorders: DSM-5 (American Psychiatric Association, 2013).
The ICD-10 Classification of Mental and Behavioural Disorders: Clinical Descriptions and Diagnostic Guidelines (World Health Organization, 1992).
Taillefer, S. E., Liu, J. J. W., Ornstein, T. J. & Vickers, K. Indecisiveness as a predictor of quality of life in individuals with obsessive and compulsive traits. J. Obsessive Compuls. Relat. Disord. 10, 91–98 (2016).
Storch, E. A. et al. Development and psychometric evaluation of the Yale–Brown Obsessive-Compulsive Scale—Second Edition. Psychol. Assess. 22, 223–232 (2010).
Grassi, G. et al. Think twice: impulsivity and decision making in obsessive–compulsive disorder. J. Behav. Addict. 4, 263–272 (2015).
Jacobsen, P., Freeman, D. & Salkovskis, P. Reasoning bias and belief conviction in obsessive-compulsive disorder and delusions: jumping to conclusions across disorders? Br. J. Clin. Psychol. 51, 84–99 (2012).
Pélissier, M.-C. & O’Connor, K. P. Deductive and inductive reasoning in obsessive-compulsive disorder. Br. J. Clin. Psychol. 41, 15–27 (2002).
Volans, P. J. Styles of decision-making and probability appraisal in selected obsessional and phobic patients. Br. J. Soc. Clin. Psychol. 15, 305–317 (1976).
Voon, V. et al. Decisional impulsivity and the associative-limbic subthalamic nucleus in obsessive-compulsive disorder: stimulation and connectivity. Brain 140, 442–456 (2017).
Erhan, C. et al. Disrupted latent decision processes in medication-free pediatric OCD patients. J. Affect. Disord. 207, 32–37 (2017).
Banca, P. et al. Evidence accumulation in obsessive-compulsive disorder: the role of uncertainty and monetary reward on perceptual decision-making thresholds. Neuropsychopharmacology 40, 1192–1202 (2015).
Loosen, A. M., Skvortsova, V. & Hauser, T. U. Obsessive–compulsive symptoms and information seeking during the Covid-19 pandemic. Transl. Psychiatry 11, 309 (2021).
Cisek, P. & Thura, D. in Oxford Research Encyclopedia of Neuroscience (Oxford Univ. Press, 2022); https://doi.org/10.1093/acrefore/9780190264086.013.346
Gold, J. I. & Shadlen, M. N. The neural basis of decision making. Annu. Rev. Neurosci. 30, 535–574 (2007).
Brody, C. D. & Hanks, T. D. Neural underpinnings of the evidence accumulator. Curr. Opin. Neurobiol. 37, 149–157 (2016).
Shadlen, M. N. & Shohamy, D. Decision making and sequential sampling from memory. Neuron 90, 927–939 (2016).
Krajbich, I., Armel, C. & Rangel, A. Visual fixations and the computation and comparison of value in simple choice. Nat. Neurosci. 13, 1292–1298 (2010).
Busemeyer, J. R. & Townsend, J. T. Decision field theory: a dynamic-cognitive approach to decision making in an uncertain environment. Psychol. Rev. 100, 432–459 (1993).
Cisek, P., Puskas, G. A. & El-Murr, S. Decisions in changing conditions: the urgency-gating model. J. Neurosci. 29, 11560–11571 (2009).
Drugowitsch, J., Moreno-Bote, R., Churchland, A. K., Shadlen, M. N. & Pouget, A. The cost of accumulating evidence in perceptual decision making. J. Neurosci. 32, 3612–3628 (2012).
Milosavljevic, M., Malmaud, J., Huth, A., Koch, C. & Rangel, A. The drift diffusion model can account for the accuracy and reaction time of value-based choices under high and low time pressure. Judgm. Decis. Mak. 5, 437–449 (2010).
Thura, D., Beauregard-Racine, J., Fradet, C.-W. & Cisek, P. Decision making by urgency gating: theory and experimental support. J. Neurophysiol. 108, 2912–2930 (2012).
Hauser, T. U., Moutoussis, M., Purg, N., Dayan, P. & Dolan, R. J. Beta-blocker propranolol modulates decision urgency during sequential information gathering. J. Neurosci. 38, 7170–7178 (2018).
Bowler, A. et al. Children perform extensive information gathering when it is not costly. Cognition 208, 104535 (2021).
Michely, J., Martin, I. M., Dolan, R. J. & Hauser, T. U. Boosting serotonin increases information gathering by reducing subjective cognitive costs. J. Neurosci. 43, 5848–5855 (2023).
Wyart, V., Myers, N. E. & Summerfield, C. Neural mechanisms of human perceptual choice under focused and divided attention. J. Neurosci. 35, 3485–3498 (2015).
Cheadle, S. et al. Adaptive gain control during human perceptual choice. Neuron 81, 1429–1441 (2014).
Osth, A. F. & Farrell, S. Using response time distributions and race models to characterize primacy and recency effects in free recall initiation. Psychol. Rev. 126, 578–609 (2019).
Carland, M. A., Thura, D. & Cisek, P. The urge to decide and act: implications for brain function and dysfunction. Neuroscientist 25, 491–511 (2019).
Bogacz, R. Optimal decision-making theories: linking neurobiology with behaviour. Trends Cogn. Sci. 11, 118–125 (2007).
Hogarth, R. M. & Einhorn, H. J. Order effects in belief updating: the belief-adjustment model. Cognit. Psychol. 24, 1–55 (1992).
Usher, M. & McClelland, J. L. The time course of perceptual choice: the leaky, competing accumulator model. Psychol. Rev. 108, 550–592 (2001).
Kiyonaga, A., Scimeca, J. M., Bliss, D. P. & Whitney, D. Serial dependence across perception, attention, and memory. Trends Cogn. Sci. 21, 493–497 (2017).
den Ouden, H. E., Kok, P. & de Lange, F. P. How prediction errors shape perception, attention, and motivation. Front. Psychol. 3, 548 (2012).
Eldar, E., Rutledge, R. B., Dolan, R. J. & Niv, Y. Mood as representation of momentum. Trends Cogn. Sci. 20, 15–24 (2016).
Bogacz, R., Brown, E., Moehlis, J., Holmes, P. & Cohen, J. D. The physics of optimal decision making: a formal analysis of models of performance in two-alternative forced-choice tasks. Psychol. Rev. 113, 700–765 (2006).
Hauser, T. U. et al. Increased fronto-striatal reward prediction errors moderate decision making in obsessive–compulsive disorder. Psychol. Med. 47, 1246–1258 (2017).
Moutoussis, M., Bentall, R. P., El-Deredy, W. & Dayan, P. Bayesian modelling of jumping-to-conclusions bias in delusional patients. Cognit. Neuropsychiatry 16, 422–447 (2011).
Huq, S. F., Garety, P. A. & Hemsley, D. R. Probabilistic judgements in deluded and non-deluded subjects. Q. J. Exp. Psychol. A 40, 801–812 (1988).
Foa, E. B. et al. The Obsessive-Compulsive Inventory: development and validation of a short version. Psychol. Assess. 14, 485–496 (2002).
Hauser, T. U., Moutoussis, M., Dayan, P. & Dolan, R. J. Increased decision thresholds trigger extended information gathering across the compulsivity spectrum. Transl. Psychiatry 7, 1296 (2017).
Dar, R., Sarna, N., Yardeni, G. & Lazarov, A. Are people with obsessive-compulsive disorder under-confident in their memory and perception? A review and meta-analysis. Psychol. Med. 52, 2404–2412 (2022).
Wager, T. D., Davidson, M. L., Hughes, B. L., Lindquist, M. A. & Ochsner, K. N. Prefrontal-subcortical pathways mediating successful emotion regulation. Neuron 59, 1037–1050 (2008).
Wager, T. D. et al. Brain mediators of cardiovascular responses to social threat. Part I: reciprocal dorsal and ventral sub-regions of the medial prefrontal cortex and heart-rate reactivity. NeuroImage 47, 821–835 (2009).
Gillan, C. M. et al. Comparison of the association between goal-directed planning and self-reported compulsivity vs obsessive-compulsive disorder diagnosis. JAMA Psychiatry 77, 77–85 (2020).
Kurth-Nelson, Z., Barnes, G., Sejdinovic, D., Dolan, R. & Dayan, P. Temporal structure in associative retrieval. eLife 4, e04919 (2015).
Summerfield, C. & Parpart, P. Normative principles for decision-making in natural environments. Annu. Rev. Psychol. 73, 53–77 (2022).
van Bergen, R. S. & Jehee, J. F. M. Probabilistic representation in human visual cortex reflects uncertainty in serial decisions. J. Neurosci. 39, 8164–8176 (2019).
Rescorla, R. A. & Wagner, A. R. A theory of Pavlovian conditioning: variations in the effectiveness of reinforcement and non-reinforcement. Class. Cond. Curr. Res. Theory 2, 64–69 (1972).
Emanuel, A. & Eldar, E. Emotions as computations. Neurosci. Biobehav. Rev. 144, 104977 (2023).
Parés-Pujolràs, E., Kelly, S. P. & Murphy, P. R. Dissociable encoding of evolving beliefs and momentary belief updates in distinct neural decision signals. Nat. Commun. 16, 3922 (2025).
Diaconescu, A. O. et al. Hierarchical prediction errors in midbrain and septum during social learning. Soc. Cogn. Affect. Neurosci. 12, 618–634 (2017).
Hunt, L. T. et al. Triple dissociation of attention and decision computations across prefrontal cortex. Nat. Neurosci. 21, 1471–1481 (2018).
Kaanders, P., Nili, H., O’Reilly, J. X. & Hunt, L. Medial frontal cortex activity predicts information sampling in economic choice. J. Neurosci. 41, 8403–8413 (2021).
Perera, M. P. N., Bailey, N. W., Herring, S. E. & Fitzgerald, P. B. Electrophysiology of obsessive compulsive disorder: a systematic review of the electroencephalographic literature. J. Anxiety Disord. 62, 1–14 (2019).
Perera, M. P. N., Mallawaarachchi, S., Bailey, N. W., Murphy, O. W. & Fitzgerald, P. B. Obsessive–compulsive disorder (OCD) is associated with increased engagement of frontal brain regions across multiple event-related potentials. Psychol. Med. 53, 7287–7299 (2023).
Ersche, K. D. et al. Peripheral biomarkers of cognitive response to dopamine receptor agonist treatment. Psychopharmacology 214, 779–789 (2011).
Ermakova, A. O., Ramachandra, P., Corlett, P. R., Fletcher, P. C. & Murray, G. K. Effects of methamphetamine administration on information gathering during probabilistic reasoning in healthy humans. PLoS ONE 9, e102683 (2014).
Fitzsimmons, S. M. D. D. et al. Repetitive transcranial magnetic stimulation for obsessive-compulsive disorder: a systematic review and pairwise/network meta-analysis. J. Affect. Disord. 302, 302–312 (2022).
Dudley, R., Taylor, P., Wickham, S. & Hutton, P. Psychosis, delusions and the “jumping to conclusions” reasoning bias: a systematic review and meta-analysis. Schizophr. Bull. 42, 652–665 (2016).
Ross, R. M., McKay, R., Coltheart, M. & Langdon, R. Jumping to conclusions about the beads task? A meta-analysis of delusional ideation and data-gathering. Schizophr. Bull. 41, 1183–1191 (2015).
Fine, C., Gardner, M., Craigie, J. & Gold, I. Hopping, skipping or jumping to conclusions? Clarifying the role of the JTC bias in delusions. Cognit. Neuropsychiatry 12, 46–77 (2007).
Ashinoff, B. & Horga, G. Evidence-order effects in probabilistic inference: recency bias and delusion-like ideation across the psychosis continuum. Biol. Psychiatry 87, S392–S393 (2020).
Scheunemann, J., Fischer, R. & Moritz, S. Probing the hypersalience hypothesis—an adapted judge–advisor system tested in individuals with psychotic-like experiences. Front. Psychiatry 12, 612810 (2021).
Ashinoff, B. K., Buck, J., Woodford, M. & Horga, G. The effects of base rate neglect on sequential belief updating and real-world beliefs. PLoS Comput. Biol. 18, e1010796 (2022).
Sterzer, P. et al. The predictive coding account of psychosis. Biol. Psychiatry 84, 634–643 (2018).
Koerner, N., Mejia, T. & Kusec, A. What’s in a name? Intolerance of uncertainty, other uncertainty-relevant constructs, and their differential relations to worry and generalized anxiety disorder. Cogn. Behav. Ther. 46, 141–161 (2017).
Rassin, E., Muris, P., Franken, I., Smit, M. & Wong, M. Measuring general indecisiveness. J. Psychopathol. Behav. Assess. 29, 60–67 (2007).
Lauderdale, S. A. & Oakes, K. Factor structure of the revised indecisiveness scale and association with risks for and symptoms of anxiety, depression, and attentional control. J. Ration. Emot. Cogn. Behav. Ther. 39, 256–284 (2021).
Clark, L., Robbins, T. W., Ersche, K. D. & Sahakian, B. J. Reflection impulsivity in current and former substance users. Biol. Psychiatry 60, 515–522 (2006).
First, M. B. in The Encyclopedia of Clinical Psychology (eds Cautin, R. L. & Lilienfeld, S. O.) https://doi.org/10.1002/9781118625392.wbecp351 (John Wiley & Sons, 2015).
Stanford, M. S. et al. Fifty years of the Barratt Impulsiveness Scale: an update and review. Personal. Individ. Differ. 47, 385–395 (2009).
Beck, A. T., Steer, R. A., Ball, R. & Ranieri, W. F. Comparison of Beck Depression Inventories-IA and -II in psychiatric outpatients. J. Pers. Assess. 67, 588–597 (1996).
Buhr, K. & Dugas, M. J. The intolerance of uncertainty scale: psychometric properties of the English version. Behav. Res. Ther. 40, 931–945 (2002).
Burns, G. L., Keortge, S. G., Formea, G. M. & Sternberger, L. G. Revision of the Padua Inventory of obsessive compulsive disorder symptoms: distinctions between worry, obsessions, and compulsions. Behav. Res. Ther. 34, 163–173 (1996).
Spielberger, C., Gorsuch, R., Lushene, R., Vagg, P. & Jacobs, G. Manual for the State-Trait Anxiety Inventory (Form Y1–Y2) (Consulting Psychologists Press, 1983).
Frost, R. O., Marten, P., Lahart, C. & Rosenblate, R. The dimensions of perfectionism. Cogn. Ther. Res. 14, 449–468 (1990).
Hauser, T. U., Allen, M., Rees, G. & Dolan, R. J. Metacognitive impairments extend perceptual decision making weaknesses in compulsivity. Sci. Rep. 7, 6614 (2017).
Whitaker, K. J. et al. Adolescence is associated with genomically patterned consolidation of the hubs of the human brain connectome. Proc. Natl Acad. Sci. USA 113, 9105–9110 (2016).
Vértes, P. E. et al. Gene transcription profiles associated with inter-modular hubs and connection distance in human functional magnetic resonance imaging networks. Phil. Trans. R. Soc. B 371, 20150362 (2016).
Wechsler, D. Wechsler Abbreviated Scale of Intelligence–Second Edition (APA PsycTests, 2011).
Oldfield, R. C. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia 9, 97–113 (1971).
Revelle, W. psych: procedures for psychological, psychometric, and personality research. R package version 2.5.6 https://CRAN.R-project.org/package=psych (2024).
Cattell, R. B. The scree test for the number of factors. Multivar. Behav. Res. 1, 245–276 (1966).
Raîche, G., Walls, T. A., Magis, D., Riopel, M. & Blais, J.-G. Non-graphical solutions for Cattell’s scree test. Methodology 9, 23–29 (2013).
Williams, B., Onsman, A. & Brown, T. Exploratory factor analysis: a five-step guide for novices. Australas. J. Paramed. 8, 1–13 (2010).
Chew, B. et al. Endogenous fluctuations in the dopaminergic midbrain drive behavioral choice variability. Proc. Natl Acad. Sci. USA 116, 18732–18737 (2019).
Habicht, J., Bowler, A., Moses-Payne, M. E. & Hauser, T. U. Children are full of optimism, but those rose-tinted glasses are fading—reduced learning from negative outcomes drives hyperoptimism in children. J. Exp. Psychol. Gen. 151, 1843–1853 (2022).
Moses-Payne, M. E., Habicht, J., Bowler, A., Steinbeis, N. & Hauser, T. U. I know better! Emerging metacognition allows adolescents to ignore false advice. Dev. Sci. 24, e13101 (2021).
Fleming, S. M. Metacognition and confidence: a review and synthesis. Annu. Rev. Psychol. 75, 241–268 (2024).
Rollwage, M. et al. Confidence drives a neural confirmation bias. Nat. Commun. 11, 2634 (2020).
Liu, Y., Dolan, R. J., Kurth-Nelson, Z. & Behrens, T. E. J. Human replay spontaneously reorganizes experience. Cell 178, 640–652.e14 (2019).
Maris, E. & Oostenveld, R. Nonparametric statistical testing of EEG- and MEG-data. J. Neurosci. Methods 164, 177–190 (2007).
Hauser, T. U. et al. Temporally dissociable contributions of human medial prefrontal subregions to reward-guided learning. J. Neurosci. 35, 11209–11220 (2015).
Hunt, L. T., Woolrich, M. W., Rushworth, M. F. S. & Behrens, T. E. J. Trial-type dependent frames of reference for value comparison. PLoS Comput. Biol. 9, e1003225 (2013).
del Rio, M. & Hauser, T. U. Indecision and recency-weighted evidence integration in non-clinical and clinical settings. OSF https://osf.io/fks97 (2025).
Acknowledgements
We thank the members of the Developmental Computational Psychiatry group who participated in a hackathon forming the basis of some of the findings presented here; the Central London OCD support group, the OCD Action charity, the North East London NHS foundation, especially the IAPT services at Waltham Forest, Redbridge and Barking & Dagenham, and the Camden and Islington NHS talking therapies service iCope for help with patient recruitment. T.U.H. was supported by a Sir Henry Dale Fellowship (211155/Z/18/Z; 211155/Z/18/B; 224051/Z/21) from Wellcome and Royal Society, a grant from the Jacobs Foundation (2017-1261-04), the Medical Research Foundation, a 2018 NARSAD Young Investigator grant (27023) from the Brain and Behavior Research Foundation, and a Philip Leverhulme Prize from the Leverhulme Trust (PLP-2021-040). T.U.H. was also supported by the Carl-Zeiss-Stiftung. A Wellcome Trust Cambridge–UCL Mental Health and Neurosciences Network grant (095844/Z/11/Z) supported this work. The Max Planck UCL Centre is a joint initiative supported by UCL and the Max Planck Society. The Wellcome Centre for Human Neuroimaging is supported by core funding from the Wellcome Trust (203147/Z/16/Z). This research was funded in whole, or in part, by the Wellcome Trust (211155/Z/18/Z). For the purpose of open access, the author has applied a CC BY public copyright license to any author-accepted manuscript version arising from this submission.
Author information
Authors and Affiliations
Contributions
M.d.R., N.T. and T.U.H. analysed the data. G.P. and T.U.H. performed the experiments. L.T.H., M.M., R.J.D. and T.U.H. conceived and designed the experiments. All authors wrote the paper.
Corresponding authors
Ethics declarations
Competing interests
T.U.H. consults for Limbic Ltd. and holds shares in the company, which is unrelated to the current project. The other authors declare no competing interests.
Peer review
Peer review information
Nature Human Behaviour thanks M. Prabhavi Perera and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Supplementary Information
Supplementary results, Table 1 and Figs. 1–9.
Source data
Source Data Fig. 1
Statistical source data.
Source Data Fig. 2
Statistical source data.
Source Data Fig. 3
Statistical source data.
Source Data Fig. 4
Statistical source data.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
del Río, M., Trudel, N., Prabhu, G. et al. Indecision and recency-weighted evidence integration in non-clinical and clinical settings. Nat Hum Behav (2026). https://doi.org/10.1038/s41562-025-02385-1
Received:
Accepted:
Published:
Version of record:
DOI: https://doi.org/10.1038/s41562-025-02385-1
