In many studies of attention-deficit hyperactivity disorder (ADHD), stimulus encoding and processing (perceptual function) and response selection (executive function) have been intertwined. To dissociate deficits in these functions, we introduced a task that parametrically varied low-level stimulus features (orientation and color) for fine-grained analysis of perceptual function. It also required participants to switch their attention between feature dimensions on a trial-by-trial basis, thus taxing executive processes. Furthermore, we used a response paradigm that captured task-irrelevant motor output (TIMO), reflecting failures to use the correct stimulus-response rule. ADHD participants had substantially higher perceptual variability than controls, especially for orientation, as well as higher TIMO. In both ADHD and controls, TIMO was strongly affected by the switch manipulation. Across participants, the perceptual variability parameter was correlated with TIMO, suggesting that perceptual deficits are associated with executive function deficits. Based on perceptual variability alone, we were able to classify participants into ADHD and controls with a mean accuracy of about 77%. Participants’ self-reported General Executive Composite score correlated not only with TIMO but also with the perceptual variability parameter. Our results highlight the role of perceptual deficits in ADHD and the usefulness of computational modeling of behavior in dissociating perceptual from executive processes.
In ADHD, the behavioral deficits captured by self-reports and collateral reports have been attributed to differences in attention, executive function, and lower level processes, including perceptual function. In the realm of visual attention, differences in accuracy or reaction time (RT) have been found in some visual search tasks but not in others (for a review, see Mullane & Klein, 2008). No consistent deficits have been found when probing selective attention with visuo spatial orienting tasks (Cubillo et al., 2010; C. L. Huang-Pollock & Nigg, 2003; Roberts, Ashinoff, Castellanos, & Carrasco, 2017; Rubia et al., 2010). ADHD patients tend to have worse executive function than controls (Boonstra, Oosterlaan, Sergeant, & Buitelaar, 2005; Castellanos & Tannock, 2002; Kofler et al., 2013; Willcutt, Doyle, Nigg, Faraone, & Pennington, 2005) predominantly in response execution and inhibition (Barkley, 1997; Booth et al., 2005; Casey et al., 1997) but also in working memory and switching between stimulus-response rules (Cepeda, Cepeda, & Kramer, 2000; Halleland, Haavik, & Lundervold, 2012; Homack, 2004; King, Colla, Brass, Heuser, & von Cramon, 2007).
While some researchers believe executive function impairments to be primary in ADHD, others acknowledge that they are neither necessary nor sufficient to cause the disorder (Boonstra et al., 2005; Willcutt et al., 2005). More specifically, yet others suggest that ADHD impairments are a combination of deficits in high-level and “low-level processes” (Castellanos et al., 2008; Gonen-Yaacovi et al., 2016; Killeen, Russell, & Sergeant, 2013; Rommelse et al., 2007; Sergeant, Geurts, & Oosterlaan, 2002; Sonuga-Barke & Castellanos, 2007). These low-level processes entail arousal (Sergeant, 2005), and relatedly, accumulation of evidence (Karalunas & Huang-Pollock, 2013), timing (Nigg & Casey, 2005), or reward sensitivity (see I. Ma, van Duijvenvoorde, & Scheres, 2016; Sonuga-Barke, 2003, for reviews). It should be kept in mind that ADHD might be a heterogenous disorder (Fair, Bathula, Nikolas, & Nigg, 2012; Nigg, Willcutt, Doyle, & Sonuga-Barke, 2005), and different causes might apply to different deficits.
Here, we extend the examination of low-level processes to perceptual encoding. Behavioral studies that examined the quality of perceptual encoding in ADHD in the absence of attentional or executive involvement have found small and inconsistent differences (see Fuermaier et al., 2017, for a review). On the other hand, other investigations have found evidence for self-reported impairments in perceptual function in ADHD participants (Bijlenga, Tjon-Ka-Jie, Schuijers, & Kooij, 2017; Micoulaud-Franchi et al., 2015) or in the general population with ADHD traits (Panagiotidi, Overton, & Stafford, 2018), as well as deficits in color processing and self-reported visual function in ADHD (Kim, Chen, & Tannock, 2014). These findings are not necessarily contradictory, as perceptual deficits might emerge when attention or executive function is simultaneously taxed.
Therefore, we believe it is important to use a task that taxes both perceptual function and either attention and/or executive function but that allows for a dissociation of the respective processes. This dissociation is difficult, as has been described in the study of autism (Robertson & Baron-Cohen, 2017). In ADHD, there have been a few attempts to dissociate perceptual function from attention within a single task (Kim, Al-Haj, Fuller et al., 2014; McAvinue et al., 2012; Stevens et al., 2012). For example, Stevens et al. (2012) compared performance on digit reports with or without distractors (letters surrounding the digits) and found that ADHD participants had lower performance only when distractors were present. However, spatial covert attention was similar across ADHD and controls, leading the authors to suggest that perceptual interference or crowding is increased in ADHD.
It is still unknown whether perceptual function is impaired when executive function is simultaneously taxed. A study by Friedman-Hill et al. (2010) used a face discrimination task where they probed perceptual noise by manipulating distractor saliency and probed top-down executive control by parametrically manipulating discrimination difficulty. In difficult discriminations, the reaction time difference between high-salience and low-salience distractors was comparable in the children with ADHD to that in the healthy children and adults; however, in easy discriminations, children with ADHD were slower to respond when presented with low-salience distractors. These results suggest similar perceptual interference due to distractor salience in ADHD and controls but a higher threshold in ADHD for activating executive control of attention. A complication in the design of Friedman-Hill et al. (2010) is that face stimuli are high-dimensional and have content at many levels, complicating the separation between perceptual, attentional, and executive function. Another complication is that if the observer uses only two response keys in a task-switching paradigm, an error could be either due to a failure to switch or to a successful switch followed by a perceptual or attentional error (Ravizza & Carter, 2008).
Here, we attempted to characterize performance in early processes of perceptual encoding in ADHD and dissociate them from later response selection (executive processes) using a visuo-motor decision-making paradigm with task-switching, which avoids the complications listed above. By using a total of eight possible buttons out of which only two were relevant on a given trial, our response paradigm allowed for task-irrelevant motor output (TIMO), a new measure of executive control deficits. We defined a perceptual error as a press of the wrong button among the two relevant ones. We optimized the quantitative characterization of perceptual function by (a) using simple stimuli with feature dimensions orientation and color, thus minimizing high-level cognitive effects; (b) varying stimuli parametrically along a continuum to estimate psychometric curve parameters (standard in perceptual psychophysics but still relatively rare in the study of ADHD, Chen & Niemeier, 2017; Friedman-Hill et al., 2010; Kim, Al-Haj, Chen et al., 2014; Kim, Al-Haj, Fuller et al., 2014; Roberts et al., 2017; Stevens et al., 2012); and (c) using an efficient stimulus selection method to minimize the number of trials needed for accurate estimation of parameters (Acerbi, 2016). Broadly, our work follows a recent proposal to apply four levels of analysis to computational psychiatry: development of behavioral tasks, fitting of computational models, estimating parameters, and classification for diagnosis (Wiecki, Poland, & Frank, 2015).
Twenty ADHD and 20 control adult participants took part in our experiment. Stimuli were two colored ellipses; each display contained one stimulus on the right of the fixation dot and one on the left. The participants performed yes–no discrimination of one of the ellipses (see Klein, 2001, for comparison with other psychophysical tasks). Specifically, the participants performed either fine orientation discrimination (was the cued ellipse clockwise or counterclockwise relative to vertical?) or fine color discrimination (was the cued ellipse more yellow or more blue relative to mid-level green?). The cue was 100% valid. In this task, participants had to rely on their internal memorized references, here for vertical and respectively the mid-level green in between the specific isoluminant yellow and blue values chosen.
Every trial started with a symbolic feature dimension cue, informing the participant which feature dimension was relevant on that trial. Simultaneously presented was a spatial cue (a line segment), informing the participant which side of the screen was relevant on that trial (Figure 1A). To better detect failures of spatial or feature switching, we used a response paradigm in which, on each trial, only two of eight response keys were relevant, depending on the spatial and the feature cue; any other key press counted as TIMO. Separately in each condition and for each participant, we used a Bayesian adaptive method to select maximally informative stimuli (see “Target Stimulus Generation”). This method allowed us to estimate the psychometric curve parameters with relatively few trials.
Each participant experienced three types of blocks: Ori, Col, and Switch. In Ori blocks, the feature dimension cue was always orientation. The spatial cue was randomly chosen on each trial, yielding two possible trial types: Ori-Left and Ori-Right (Figure 1B). We analyzed the Ori-Left and Ori-Right trials together as the Ori condition. In Col blocks, the feature dimension cue was always color, and again the spatial cue was randomly chosen on each trial, yielding two possible trial types, Col-Left and Col-Right, which we grouped together for analysis into the Col condition. In Switch blocks, all four trial types were possible. We will refer to the orientation and color trials in switch blocks as the OriS and ColS conditions, respectively, and to the difference between no-switch and switch blocks as a difference in (executive) load.
An observer’s sequence of computations in the task can be conceptualized as a perceptual decision-making stage (stimulus encoding, affected by attention, and inference), followed by executive processing (rule retrieval and response execution, Figure 2). The parametric variation of stimulus strength allowed us to estimate perceptual variability σ (or noise, the inverse of slope/sensitivity) as a main metric of perceptual function, and the eight-button response paradigm allows us to estimate task-irrelevant motor output as a main metric of executive function. In addition, we characterized behavior using other psychometric curve parameters, median reaction time, and reaction time variability.
While usually a noise parameter (equivalent to our perceptual variability) in psychometric curves reflects a mix of sensory and decision noise (Gold & Ding, 2013), we believe that here the perceptual variability parameter for orientation and color is likely additionally modulated by attention. The Ori and Col conditions attempt to engage endogenous covert spatial attention, and the Switch conditions additionally engage attention to feature dimension. Previous studies showed modulation of psychometric curve parameters by attention, though either with larger stimulus eccentricies (usually equal or larger than 4 dva vs 2.5 dva here), in different tasks such as target detection (Bashinski & Bacharach, 1980), 2AFC orientation discrimination (Downing, 1988), color-change detection (Herman, Bogadhi, & Krauzlis, 2015), examined exogenous attention (Fuller & Carrasco, 2006), or with other stimulus strength manipulation, such as contrast (Ling & Carrasco, 2006; Pestilli, Viera, & Carrasco, 2007; for reviews, see Carrasco, 2011, 2014).
We recruited all participants through local advertisements, including flyers and newspaper and radio advertisements. Information on the participants is presented in Mihali, Young, Adler, Halassa, and Ma (2018, appendix). Participants in both groups were matched as much as possible by age, sex, and education (see Mihali et al., 2018, Appendix, Table A1). Twenty ADHD participants (12 female) of mean age 35.3 (SD 10.0, range 21 to 55) and 20 control participants (11 female) of mean age 32.5 (SD 6.1, range 19 to 44), with no statistical difference between their ages (Wilcoxon rank-sum test, p = 0.78), participated. Seventeen out of the 20 ADHD participants presented the combined subtype and three the inattentive subtype. All participants spoke English and had normal or corrected-to-normal vision. We asked every participant before they started if they were color-blind. One participant was excluded because of color-blindness. All participants provided informed consent. The study conformed to the Declaration of Helsinki and was approved by the Institutional Review Board of New York University School of Medicine.
None of the participants with ADHD were prescribed or took stimulant medication within 2 months of participating in the study. Participants with comorbid anxiety or unipolar depressive disorders were included as long as the symptoms at the time of evaluation were mild or in remission. Participants with bipolar disorders, psychotic disorders, substance use disorders, and neurologic disorders were excluded. For all adults, the diagnostic procedure included both clinician-administered and self-administered scales. A trained clinician assessed every participant using the Adult ADHD Clinician Diagnostic Scale (ACDS) v. 1.2, the Adult ADHD Investigator Symptom Rating Scale (AISRS), the Clinical Global Impressions–Severity of Illness (CGI-S) Scale, and the MINI International Neuropsychiatric Interview. All participants also completed the Adult ADHD Self-Report Scale (ASRS v. 1.1), the Adult ADHD Quality of Life (AAQoL) Scale, the World Health Organization Disability Assessment Schedule (WHODAS–II), and the Behavior Rating Inventory of Executive Function Adult Version (BRIEF-A). These scales have been extensively validated (L. Adler & Cohen, 2004; Kessler et al., 2005, 2006; Silverstein et al., 2018).
We displayed stimuli on a 23-inch (58.42 cm) Acer T232HL LCD monitor of resolution: 1920 × 1080 pixels and 60 Hz refresh rate (1 frame lasting 16.7 ms). We used a Kinesis Freestyle2 split keyboard. Participants used a head rest located at approximately 55 cm from the screen; this meant that 1 degree of visual angle (dva) subtended approximately 34 pixels. Stimulus presentation and response collection were controlled by a Windows computer running Matlab 7.1 (MathWorks, Massachusetts, USA) with Psychtoolbox3 (Brainard, 1997; Kleiner et al., 2007; Pelli, 1997) and EyeLink (Cornelissen, Peters, & Palmer, 2002).
For 10 out of 20 ADHD participants and 10 out of 20 control participants, we monitored their fixation and recorded their eye movements. The rationale for not eye tracking all participants was a mixture of lack of sufficient time on the participants’ side and balanced design on the experimenters’ side. The eye tracker was calibrated using the five-point calibration routine before every block. We recorded eye movements using a remote infrared video-oculographic system (EyeLink 1000 Plus; SR Research, Ltd, Mississauga, Ontario, Canada) with a 1 kHz sampling rate. We set the heuristic filtering option to “OFF.”
The background was mid-level gray (28.7 cd/m2). The stimuli were ellipses with area of 1,600 pixels2 and 0.55° eccentricity, and thus with a major axis of 50 pixels and minor axis of 41 pixels. For the nontarget ellipse, the orientation was randomly drawn from a von Mises distribution centered at 0 with κ = 30 and then divided by 2, approximately equivalent to a Gaussian distribution with mean 0 and a standard deviation of about 5°. The color of the nontarget ellipse was based on a uniformly drawn sample that was used to linearly interpolate between blue and yellow in CIE L*a*b* (CIELAB) color space, with blue as [78 −30 −40], corresponding in RGB space to [0 167 255], and yellow as [78 0 80], corresponding in RGB to [200 130 0]. For each color, lightness was always kept constant at L = 78. Indeed, measured luminance was ∼39 cd/m2. The target stimulus was specified on a trial-to-trial basis using the Bayesian algorithm described below.
The orientation and color of the target stimulus were based on the participant’s previous responses according to an adaptive procedure, a type of Bayesian staircase, applied separately for each of the four conditions. We used the Psybayes algorithm (Acerbi, 2016), based on Kontsevich and Tyler (1999), with extentions to include the lapse rate (Prins, 2012). This procedure maintains a posterior distribution over the parameters and updates it after each trial on which the participant pressed one of the two task-relevant buttons. The next stimulus value is chosen to minimize the entropy of the updated posterior given the stimulus, averaged over the participant’s possible responses weighted by their respective probabilities (Kontsevich & Tyler, 1999). Each one of these four Bayesian staircases generated on every trial a unitless value w within the range [−0.5, 0.5] that was converted to stimulus values: to degrees in orientation trials and to [L, (w + 0.5)ayellow + (0.5 − w)ablue, (w + 0.5)byellow + (0.5 − w)bblue] in color trials. Thus, target stimulus values fell within an orientation range of −30° to 30° and within a blue to yellow range from [78 −30 −40] to [78 0 80]. We defined the space of parameters that Psybayes constructs the posterior on: For μ, we used a linear grid of 51 points from −0.5 to 0.5; for σ, a logarithmic grid of 25 points from 0.002 to 0.8; and for λ, a linear grid of 25 points from 0 to 0.3.
A trial sequence started with the simultaneous appearance of a feature dimension cue and a spatial cue, presented for 500 ms. The feature dimension cue for orientation consisted of two white line segments, each of length approximately 1 dva, crossing at the center, with orientations tilted ±26.6° with respect to vertical; for color, it consisted of two semicircles (divided vertically, right one yellow, left one blue) joined to form a circle of radius approximately 0.3 dva. Simultaneously, a spatial cue was presented, which consisted of a horizontal line segment of length approximately 0.5 dva emanating from the center of fixation to the left or to the right. We chose 500 ms to ensure sufficient time for the deployment of endogenous feature-based attention (Liu, Stevens, & Carrasco, 2007). Following a delay of 250 ms consisting of the presentation of a central fixation circle of radius 0.12 dva, two ellipses appeared at 2.5 dva to the right and left of a central fixation circle. The stimuli were presented on the screen for 117 ms, followed by another delay period of 250 ms.
After the poststimulus delay, the participant had to respond about the target ellipse via a specific key press out of a total of eight keys (Figure 1A). On any given trial, six of these eight keys are irrelevant. For orientation, the participants were instructed to press one of the two labeled keys for clockwise (CW) or counterclockwise (CCW), using the left keypad for the left spatial cue and the right one for the right spatial cue. For color, they had to press one of two labeled keys to indicate whether the ellipse was more yellow or more blue, also using the left or respectively right keypad depending on the spatial cue. Figure 1B shows all the four possible cue-response mappings. The direction of the spatial cue was randomly drawn on each trial, so participants used their right hand approximately half the time. After the response, auditory feedback was provided for 200 ms: a 1200 Hz tone if the participant had pressed the correct key and a 500 Hz tone if the participant had pressed any of the seven incorrect keys.
Before they began the experiment, participants were guided step by step through the different parts of instructions. The experimenter read the instructions on the screen (presented in Mihali et al., 2018, Appendix, Figure A1A) out loud. To remind subjects of the stimulus–response pairings, a sheet with these pairings was posted on the wall of the psychophysics room (Mihali et al., 2018, Appendix, Figure A1B). In total, participants performed 40 training trials: a short orientation-only block (“O”) of 10 trials, a short color only block of 10 trials, and a short switch block (“S”) of 20 trials. The experimenter was present with the participants during the training to observe responses, provide further feedback, and answer questions. Participants repeated the set of all 40 training trials until they achieved a performance greater than 65%.
After the training, participants performed eight blocks of about 100 trials each in the order O-C-S-S-S-S-C-O or C-O-S-S-S-S-O-C, with 30-s breaks in between blocks. Changes in block type were signaled with a screen with the instruction “In this block, your job is to report ORIENTATION” for O blocks or “In this block, your job is to report COLOR” for C, or “In this block, your job is to report either ORIENTATION or COLOR” for S, with each feature dimension word followed by its associated symbol. In total, participants completed 800 nonaborted trials, approximately 200 in each one of the four conditions, Ori, Col, OriS, and ColS (from S blocks).
For most metrics, we report median values and 95% bootstrapped confidence intervals. Across 50,000 iterations, we took samples with replacement from and of the same size as the original data with Matlab’s randsample and calculated the median of each of those sets of samples. The the 2.5th and 97.5th quantiles of the distribution of medians across iterations were taken as the 95% confidence intervals.
To determine the differences between groups and the two experimental conditions of load and feature, we used three-way mixed-design ANOVA with two repeated measures, since we have one “between-participants” variable (group) and two “within-participants” factors (feature, Ori vs. Col and load, No-switch vs. Switch). Beforehand, we log-transformed the measures that were lower bounded by 0. When we assumed shared parameters between No-switch and Switch and thus we had only one “within-participants” factor, we used two-way mixed-design ANOVA. We implemented the ANOVAs in SPSS with “General linear model: repeated measures.” For post hoc comparisons, we adjust the significance level according to the Sidak correction to . For the three-way mixed-design ANOVA, we performed, unless otherwise specified, 12 planned pairwise comparisons in Matlab: Wilcoxon rank-sum tests between groups (one for each condition, four in total) and Wilcoxon signed-rank tests for conditions within a group (four per group, eight in total). We used the Sidak correction for multiple comparisons, decreasing the significance level to α = 0.0043 for post hoc comparisons following the three-way mixed-design ANOVA or, respectively, α = 0.0127 following the two-way mixed-design ANOVA.
To correct for multiple comparisons when examining the pairwise correlation matrix of the performance measures, we used a method from Nyholt et al. (2004). If M is the total number of measures, the number of effective comparisons will be decreased more if the measures are more highly correlated, as captured in a higher variance of the eigenvalues λobs of the correlation matrix, which we calculated with Matlab’s function eig. Then, . As in Nyholt (2004), Meff is used in the Sidak correction (a slightly less conservative alternative to the Bonferroni correction), modifying the significance level to .
We implemented multivariate linear regression with Matlab’s fitlm.
We fit the logistic regression coefficients with Matlab’s glmfit with input “binomial” and the link parameter “logit.” For a given participant, we used the task metrics and the fitted coefficients with glmval to get p(Diagnosis), which was then thresholded at 0.5 to predict the 0 or 1 ADHD diagnosis.
To assess the use of this logistic regression classifier for out-of-sample prediction, we calculated the cross-validated accuracy. We did stratified 10-fold cross-validation, in which each fold had four participants, two ADHD and two controls; thus, we trained the classifier to find the coefficients over 36 participants and tested over 4 and calculated the mean accuracy across folds. We did 1,000 runs of this stratified 10-fold cross-validation to allow for different random assignments of participants into folds and took the mean accuracy over runs.
We fitted psychometric curves to trials on which a participant pressed one of the two relevant buttons; s denotes the normalized stimulus value on a given trial (ranging between [−0.5, 0.5]). We use the following form of the psychometric curve (Wichmann & Hill, 2001):
We performed maximum-likelihood estimation of the psychometric curve parameters μ, σ, and λ. The likelihood of a parameter combination is the probability of the data given that parameter combination; we denote the log-likelihood by LL. We assumed that trials are independent of each other, and thus we summed the log-likelihoods across all trials. We fitted orientation and color trials separately; thus the following log-likelihoods apply to either set of trials. In the main model, we assumed that μ and λ are shared across both load conditions (No-switch and Switch), whereas σ might differ. These assumptions had both a practical and a principled motivation. Assuming that parameters are shared between conditions reduced the number of parameters to eight and made parameter estimates more reliable. Moreover, if μ reflects an overall bias and λ a generic lapsing process, we did not expect them to change with load. For a model without these assumptions, and a model comparison, see Mihali et al. (2018, Appendix, Figures A7 and A8). The log-likelihood for trials in a given feature dimension becomes
For fitting ex-Gaussian distributions to reaction times, we used a custom-made script modeled after an existing software package (Zandbelt, 2014).
Clinical data are not available beyond diagnosis labels, experiment code is available upon request, and behavioral data and analysis code are available at https://github.com/lianaan/Perc_Var.
We attempted to dissociate perceptual from executive deficits in ADHD with a new visuomotor decision-making task with a task-switching component. This task yielded two main measures: TIMO and perceptual variability.
TIMO refers to the trials when participants pressed one of the six irrelevant keys, and hence such responses most likely reflect a failure of proper stimulus-response rule retrieval. TIMO was quite low overall (mean ± sem: 0.06 ± 0.01); ADHD participants produced a higher proportion of TIMO (0.079 ± 0.018) relative to controls (0.041 ± 0.008). Figure 3A presents a breakdown of TIMO by condition. A three-way mixed-design ANOVA on log-TIMO with between-participants variable group and within-participants factors load (No-switch and Switch) and feature (Ori and Col) reveals a significant effect of group, F(1, 38) = 8.83, p = 0.005, ηp2 = 0.19, a significant effect of load, F(1, 38) = 101.4, p < 0.0001, ηp2 = 0.73, and no significant effect of feature, F(1, 38) = 1.62, p = 0.21, ηp2 = 0.04. Neither of the two-way interactions nor the three-way interaction were significant, p > 0.06. In particular, the Group × Load interaction was not significant, F(1, 38) = 3.72, p = 0.06, ηp2 = 0.09; thus, we did not find that switching between feature dimensions carries a higher cost in ADHD. Next, we performed 12 post hoc planned comparisons: within each group, Wilcoxon signed-rank tests for Ori versus OriS, Col versus ColS, Ori versus Col, and OriS versus ColS, and between groups, Wilcoxon rank-sum tests for Ori, OriS, Col, and ColS. After Sidak correction (α = 0.0043), no between-group comparisons were significant, p > 0.0046. The within-group load comparisons were all significant, p < 0.002. No within-group feature comparisons were significant, p > 0.07. Taken together, these results validate TIMO as a metric of interest for executive control.
In the OriS and ColS conditions, the majority of TIMO seemed to be feature errors (Mihali et al., 2018, Appendix, Figure A2B). Relative to the instructions on a given trial, the six irrelevant keys subdivide into two that represent spatial errors, two feature errors, and two both spatial and feature errors. We did not delve into these distinctions since overall TIMO was quite low.
ADHD participants showed longer reaction times (RTs; Figure 3B). Three-way mixed-design ANOVA on log RTs revealed a significant effect of group, F(1, 38) = 4.72, p = 0.036, ηp2 = 0.11, and significant effects of load, F(1, 38) = 84.92, p < 0.0001, ηp2 = 0.69, and feature, F(1, 38) = 70.29, p < 0.0001, ηp2 = 0.65. In addition, we found significant Group × Feature, F(1, 38) = 4.63, p = 0.038, ηp2 = 0.11, and Load × Feature interactions, F(1, 38) = 12.37, p = 0.001, ηp2 = 0.25, but not a significant Group × Load interaction, F(1, 38) = 0.08, p = 0.77, ηp2 = 0.002. After Sidak correction (α = 0.0043), none of the between-group comparisons were significant, p > 0.019. The effects of within-group load and feature on log RTs were all significant both within control and within ADHD, p < 0.001. Higher RTs for Col than Ori could be due to the fact that the Ori responses are intuitively mapped to left and right, while the Col responses are arbitrarily mapped as blue to left and yellow to right.
Higher RT variability (or intra-individual variability) in ADHD has been found consistently (Kofler et al., 2013) and has been generally thought to reflect cognitive processes separate from higher median RTs (Castellanos, Sonuga-Barke, Milham, & Tannock, 2006; Kofler et al., 2013, but see Karalunas, Huang-Pollock, & Nigg, 2012, for an opposing account). The term RT variability has been used to refer to different aspects of RT distributions (Kofler et al., 2013); here we fitted ex-Gaussian distributions (Leth-Steensen, Elbaz, & Douglas, 2000) and used the τ parameter as a measure of RT variability. The τ parameter has been shown to capture the tendency of ADHD participants to have a higher proportion of abnormally slow responses (Castellanos et al., 2006; Kofler et al., 2013; Leth-Steensen et al., 2000). Before committing to the ex-Gaussian, we verified that it captures the data better than the log-normal and gamma distributions (see Mihali et al., 2018, Appendix, Figure A4). Three-way mixed- design ANOVA on log τ revealed a significant effect of group, F(1, 38) = 7.72, p = 0.008, ηp2 = 0.17, an effect of load, F(1, 38) = 9.32, p = 0.004, ηp2 = 0.20, and an effect of feature, F(1, 38) = 18.85, p < 0.001, ηp2 = 0.33. The only significant interaction was between load and feature, F(1, 38) = 14.96, p < 0.001, ηp2 = 0.28. After Sidak correction (α = 0.0043), none of the between-group comparisons were significant, p > 0.006. Within controls, the effects of load and feature on log RT τ were significant for Ori versus OriS and Ori versus Col, p < 0.001. Within ADHD, no effects of load or feature were significant, p > 0.02. We confirmed the pattern of higher RT variability in ADHD with a nonparametric measure, RT iqr (see Mihali et al., 2018, Appendix, Figure A5).
Overall, we found that ADHD participants had longer and more variable reaction times, consistent with previous work (Douglas, 1999; Kofler et al., 2013; Leth-Steensen et al., 2000). However, RT-related differences across groups are usually difficult to interpret because they might encompass multiple processes, including sensory encoding, decision time, speed–accuracy trade-offs, stimulus–response rule retrieval, response preparation, and response execution (unless some of these processes are disentangled with drift diffusion models; C. Huang-Pollock et al., 2016, Karalunas et al., 2012). The effect of load on RT and RT τ does seem to suggest that on Switch trials, more time is spent on executive processes, here stimulus–response rule retrieval, response preparation, and response execution, relative to No-switch trials.
We confined the following analysis to the trials in which participants pressed one of the two relevant keys. Because of the Bayesian stimulus selection method, each participant received a different set of stimuli for each condition (see Mihali et al., 2018, Appendix, Figure A6), and thus proportion correct is largely stable across conditions and participants (mean ± sem: 0.811 ± 0.007; Mihali et al., 2018, Appendix, Figure A2A) and thus not informative. Instead, we fitted a psychometric curve to non-TIMO trials in each condition (Kingdom, 2009). Thus, the parameters of the psychometric curves captured the differences in performance across conditions and participants. The normalized orientation and color continua spanned the interval [−0.5, 0.5]. Each psychometric curve has three parameters: a point of subjective equality μ, perceptual variability σ, and a lapse rate λ (Figure 4B, 4C; Mihali et al., 2018, Appendix, Figure A8D). Nonzero μ represents a tendency to choose one option more than the other. The parameter σ is a composite of sensory noise and decision noise and might also reflect the quality of the allocation of spatial attention and of feature attention in switch blocks. Higher σ denotes a reduced ability to discriminate between small variations within a feature. The parameter λ reflects trials with lapses in attention or erroneous motor output. In our main model, we assumed that μ and λ are independent of load; we confirmed this assumption by comparing to a model without these assumptions (“full” model) in Mihali et al. (2018, Appendix, Figures A7 and A8). The parameters σ and λ might trade off against each other, although this is less of a concern in our main model than in the “full” model.
Three-way mixed-design ANOVA on log σ showed a significant effect of group, F(1, 38) = 10.56, p = 0.002, ηp2 = 0.22, and a significant effect of feature, F(1, 38) = 38.3, p < 0.001, ηp2 = 0.50, but not of load, F(1, 38) = 0.97, p = 0.33, ηp2 = 0.025. The effect of Group × Load did not reach significance, F(1, 38) = 3.97, p = 0.054, ηp2 = 0.09, and neither did the other two-way interactions, p > 0.1, but the effect of Group × Load × Feature was significant, F(1, 38) = 6.75, p = 0.013, ηp2 = 0.15. Upon exclusion of the two salient outliers from the ADHD group (Figures 4A and 5A), the three-way mixed-design ANOVA results were highly similar: significant effect of group, F(1, 36) = 10.97, p = 0.002, ηp2 = 0.23, and significant effect of feature, F(1, 36) = 34.2, p < 0.001, ηp2 = 0.49, but not of load, F(1, 36) = 1.79, p = 0.18, ηp2 = 0.05. None of the two-way interactions reached significance, p > 0.1, but the effect of Group × Load × Feature was significant, F(1, 36) = 6.12, p = 0.018, ηp2 = 0.145. Because the normalization to the (arbitrary) stimulus range is specific to each feature dimension, the values of σ cannot be meaningfully compared between orientation and color: A different stimulus range would have changed the σ values without changing the observer’s true perceptual variability. Therefore, only the within-feature post hoc comparisons are meaningful, giving a corrected significance level of α = 0.0065. Then, the between-group comparisons were significant for both Ori and OriS, p < 0.0005, but not for Col, p = 0.0083, or ColS, p = 0.28. No post hoc comparisons with load were significant either within controls or within ADHD, p > 0.01. Higher σ for orientation in ADHD could result from worse low-level sensory encoding (e.g., higher neural noise), lower covert endogenous attention, higher decision noise, or even noise in the inference process about the perceptual category. The lapse rate reflects responses that are independent of the stimulus, such as lapses of attention, but could also trade off with the σ parameter. Two-way mixed-design ANOVA on log λ showed a large effect of feature, F(1, 38) = 28.08, p < 0.0001, ηp2 = 0.43, but no significant effect of group, F(1, 38) = 1.72, p = 0.19, ηp2 = 0.04, and no significant Group × Feature interaction. After Sidak correction (α = 0.0127), we found that control, p < 0.0001, and ADHD, p = 0.011, participants tended to lapse more for color than for orientation, possibly because the stimulus–response mapping is less intuitive. Results for μ are in Mihali et al. (2018, Appendix, Figure A8B). In conclusion, the parametric variation of low-level stimulus variables combined with stimulus optimization revealed robust perceptual deficits in ADHD, especially for orientation.
A possible cause of the increased perceptual variability in ADHD could be that ADHD participants were slower to learn the task. To check for learning, we fitted two psychometric curves for each condition, one to the first half of the trials and one to the second half. The σ parameters across participants, conditions, and time are presented in Mihali et al. (2018, Appendix, Figure A9). Visually, we notice a slight improvement in perceptual variability in the second half of the trials (Mihali et al., 2018, Appendix, Figure A9). To quantify it, we performed a four-way mixed-design ANOVA on log σ with time as an additional factor. We found an effect of time, F(1, 38) = 12.7, p = 0.001, ηp2 = 0.25. We did not find a significant Group × Time interaction, F(1, 38) = 0.42, p = 0.52, ηp2 = 0.01, and thus we have no evidence for a differential learning pattern for ADHD relative to controls.
Next, we asked whether behavioral metrics are correlated with each other (Table 1). For this analysis, we collapsed across groups; per participant, we averaged each behavioral metric across all four conditions. We found that the perceptual variability parameter σ is significantly correlated with TIMO, RT median, and RT τ, with high effect sizes. Note that the perceptual variability parameter and TIMO were computed from different sets of trials, therefore reducing the probability that their correlation is spurious. In addition, a breakdown of some of these correlations by group, symptom type, and condition is presented in Mihali et al. (2018, Appendix, Breakdown of correlations).
|TIMO||TIMO||RT||RT τ||Perceptual variability (σ)||Lapse rate (λ)||GEC|
|RT||ρ = 0.46|
|p = 0.003|
|RT τ||ρ = 0.42||ρ = 0.84|
|p = 0.007||p < 0.0001|
|Perceptual variability (σ)||ρ = 0.41||ρ = 0.55||ρ = 0.57|
|p = 0.0085||p = 0.0003||p = 0.0002|
|Lapse rate (λ)||ρ = 0.46||ρ = 0.23||ρ = 0.17||ρ = 0.28|
|p = 0.003||p = 0.15||p = 0.30||p = 0.08|
|GEC||ρ = 0.53||ρ = 0.25||ρ = 0.34||ρ = 0.50||ρ = 0.30|
|p = 0.0005||p = 0.12||p = 0.03||p = 0.0009||p = 0.06|
|ACDS||ρ = 0.40||ρ = 0.31||ρ = 0.45||ρ = 0.51||ρ = 0.18||ρ = 0.80|
|p = 0.01||p = 0.05||p = 0.004||p = 0.0008||p = 0.26||p < 0.0001|
So far, we have characterized behavioral differences between ADHD and controls in our task. Next, we asked if behavioral metrics relate to common clinical ones, namely, the General Executive Composite score (GEC), as assessed by the Brief-A questionnaire (self-reported; Roth, Lance, Isquith, Fischer, & Giancola, 2013), as well as the ACDS scores (clinician interview; L. Adler & Cohen, 2004). The GEC and ACDS (Mihali et al., 2018, Appendix, Table A2) are meant to be continuous measures of executive control and symptom severity, respectively. Both GEC and ACDS revealed strong correlations with TIMO, suggesting that TIMO could serve as a behavioral marker of executive deficits. GEC and ACDS were both also strongly correlated with perceptual variability. In addition, ACDS (but not GEC) was correlated with RT τ, which provides a graded counterpart of the robust finding of increased RT variability in ADHD (Kofler et al., 2013). A linear regression of GEC as a function of behavioral metrics (R2 = 0.38) showed only TIMO as statistically significant (Mihali et al., 2018, Appendix, Table A7A), reinforcing our interpretation of TIMO as reflecting failures of executive function. A linear regression of ACDS as a function of the same metrics (R2 = 0.33) only showed significance for RT τ (Mihali et al., 2018, Appendix, Table A7B), suggesting that GEC and ACDS, despite being strongly correlated (Mihali et al., 2018, Appendix, Figure A11), could capture distinct aspects of impairment (L. A. Adler et al., 2017). However, the determinant of the correlation matrix of these measures is 0.11, nearing 0 and thus signaling multicollinearity (Dormann et al., 2012). Therefore, we have to be cautious in interpreting the individual contributions of these regressors. Nevertheless, these results suggest that our behavioral metrics capture to some extent the same processes as clinical metrics, while having the advantage of avoiding the potential subjectivity inherent in questionnaires.
Finally, we asked how accurately we can classify a given participant as ADHD or control based purely on behavioral task metrics. Figure 5 depicts these results. A logistic regression using only the perceptual variability parameter yielded a classification accuracy of 77.5%, with a hit rate (sensitivity) of 75% and a false-alarm rate (1 minus specificity) of 20% (Figure 5A). A logistic regression classifier based on TIMO only had an accuracy of 62.5%, with a hit rate of 65% and a false-alarm rate of 40%; using both perceptual variability and TIMO improved the accuracy to 82.5%, with a hit rate of 80% and a false-alarm rate of 15% (Figure 5A). Of note, while these variables are correlated, the determinant of their correlation matrix is 0.82, far enough from 0 that collinearity should not be a problem (Dormann et al., 2012). Adding more regressors (RT, RT τ, and lapse) did not yield further improvement (80.0%); this is not surprising in light of multicollinearity. Hence, we consider perceptual variability and TIMO as the main regressors of interest. To assess the use of this logistic classification for out-of-sample prediction, we did stratified 10-fold cross-validation and found mean accuracies of 77.1% with perceptual variability as the only regressor, 63.1% with TIMO only, 77.8% with both perceptual variability and TIMO, and 70.0% with all metrics. The relatively high classification performance suggests that our task has potential as a diagnostic tool.
In addition to thresholding at 0.5 to get diagnosis and, subsequently, accuracy as above, we also thresholded p(Diagnosis) at linearly spaced values between 0 and 1 and plotted the resulting receiver operating characteristic (ROC) curves, both without (Figure 5B) and with stratified 10-fold cross-validation (Figure 5C). As expected, the ROC curve for the classifier all metrics shows the best performance (highest area under the curve) without cross-validation, but its performance degrades for out-of-sample predictions in the cross-validated case.
In this study, we dissociated stimulus encoding (perceptual, early) from response selection (executive, late) deficits in ADHD with a new visuomotor decision-making task with a task-switching component. To better separate executive deficits from perceptual and attentional failures, we used eight response keys, six of which were irrelevant on any given trial (TIMO). To assess perceptual precision, we used simple stimuli that varied continuously along one dimension. We used an adaptive stimulus selection method (Acerbi, 2016) that reduced the number of trials needed for accurate parameter estimation (relative to, for instance, the method of constant stimuli); reducing the number of trials is crucial when running the ADHD population. We found differences between ADHD and controls in our main task metrics, TIMO (Figure 3A) and perceptual variability (Figure 4B), as well as median reaction times and reaction time variability (Figure 3C, 3D). We found correlations of these behavioral metrics with clinical metrics (Table 1) and were able to classify participants into ADHD and controls with high (≈77%) accuracy solely on the basis of our main behavioral metrics (Figure 5).
Our finding of higher TIMO in ADHD could be due to more spatial switching errors or more feature switching errors, but it is hard to quantify these contributions since TIMO was overall relatively low. It is conceivable that a less intuitive stimulus–response mapping for orientation distrimination (stimulus oriented toward left/respond with key on the left), or types of stimuli that require spatial integration (Greenberg, Esterman, Wilson, Serences, & Yantis, 2010; Liu, 2003; Mante, Sussillo, Shenoy, & Newsome, 2013; Siegel, Buschman, & Miller, 2015) or cross-modal switching (Haigh et al., 2016), or more complex forms of task switching would produce larger differences on a TIMO-like executive function measure, in line with the executive function impairments widely reported in ADHD (Boonstra et al., 2005; Willcutt et al., 2005).
In line with previous work (Douglas, 1999; King et al., 2007; Kofler et al., 2013; Leth-Steensen et al., 2000), we found that ADHD participants had longer and more variable reaction times. While accuracy was maintained to be approximately stable in all participants, perceptual variability was higher in ADHD, and thus the increased reaction times are not reflective of speed–accuracy trade-offs. In addition, our paradigm allowed us to analyze correlations across individuals between reaction time metrics and other metrics. The correlation between the perceptual variability parameter σ and median reaction time is consistent with a drift-diffusion model, in which slower accumulation of evidence simultaneously leads to lower accuracy and longer reaction times. Indeed, many studies have found slower drift rates in ADHD (C. Huang-Pollock et al., 2016; Karalunas & Huang-Pollock, 2013; Karalunas et al., 2012; Lúcio et al., 2016; Ziegler, Pedersen, Mowinckel, & Biele, 2016).
We found higher σ in ADHD than in controls. This parameter—which we called the perceptual variability parameter—could be affected both by sensory encoding (affected by attention) and decision processes. Could the differences in σ be attributed to either type of process? Sensory and decision noise are usually confounded in the parameters derived from behavior in common discrimination tasks (Gold & Ding, 2013). However, tasks exist in which the influences of sensory and decision noise can potentially be separated (Drugowitsch, Wyart, Devauchelle, & Koechlin, 2016; Keshvari, van den Berg, & Ma, 2012; Lam et al., 2017). Additionally, neural data with high temporal resolution, such as EEG or MEG, could separate perceptual from decision-related variability as early versus late activity relative to stimulus onset (Gonen-Yaacovi et al., 2016; Mostert, Kok, & de Lange, 2015). Decision noise in perceptual decision-making might be related to decision noise on action selection in reinforcement learning models of high-level cognitive tasks. Hauser et al. (2014) found increased decision noise (temperature parameter) in ADHD in a probabilistic reversal learning task and later proposed this to underlie behavioral variability found in ADHD more generally (Hauser, Fiore, Moutoussis, & Dolan, 2016). Our result of increased perceptual variability in ADHD is consistent with this general proposal and extends it to include the possibility of an even lower level correlate of behavioral variability.
Earlier studies examining perceptual function in isolation did not find differences between ADHD and controls (see Fuermaier et al., 2017, for a review). Our result of higher perceptual variability in the ADHD group suggests that the encoding of visual stimuli is less precise than in controls, at least when experimental conditions simultaneously tax other processes. In our case, participants had to allocate either spatial attention or both spatial and feature-based attention as well as employ executive function by maintaining and acting on either two (No-switch) or four (Switch) stimulus–response rules. Earlier studies examining covert spatial attention while attempting to minimize executive load did not find differences between ADHD and controls (Cubillo et al., 2010; C. L. Huang-Pollock & Nigg, 2003; Roberts et al., 2017; Rubia et al., 2010). While perceptual precision and attention might be comparable between ADHD and controls when studied in isolation, it is possible that asking ADHD participants to simultaneously devote brain resources to other processes might allow for differences in perceptual variability to emerge.
Our results could speak to the question of low-level perceptual components interacting with measured executive control deficits, as we found a significant correlation between the perceptual variability parameter and the executive control metric TIMO. In particular, our results raise the possibility of a shared neural source of perceptual and executive function deficits, such as a lower signal-to-noise ratio in early brain areas. While ideas of lowered signal-to-noise ratio implemented through impaired dopamine and noradrenaline signaling in ADHD have been put forward before, they have been mainly confined to cerebellar, striatal, and prefrontal regions (del Campo, Chamberlain, Sahakian, & Robbins, 2011; Frank, Santamaria, OReilly, & Willcutt, 2006; Hauser et al., 2016). Beyond that, one study found higher neural noise in the visual and auditory cortex of ADHD participants (Gonen-Yaacovi et al., 2016). ADHD participants could have higher perceptual variability in orientation by having less selective orientation tuning of cells in V1; this was the mechanism proposed to underlie decreased orientation discrimination with aging in monkeys (Leventhal, Schmolesky, Wang, & Pu, 2000). The list of regions with lower signal-to-noise ratio in ADHD could include deeper brain structures with roles in selecting relevant sensory stimuli and maintaining stimulus–response rule representations, such as the thalamus (Halassa & Kastner, 2017; Schmitt et al., 2017; Wells, Wimmer, Schmitt, Feng, & Halassa, 2016; Wimmer et al., 2015; Young & Wimmer, 2017), or even lower regions with roles in orienting of attention and behavioral flexibility, such as the superior colliculus (Krauzlis, Lovejoy, & Zénon, 2013; Overton, 2008) or the locus coeruleus (Aston-Jones, Rajkowski, & Cohen, 1999; Devilbiss & Berridge, 2006). However, these regions do not just modulate cortical representations but also receive substantial top-down inputs, so the reduced signal-to-noise ratio could originate from either lower or higher level brain regions.
Based on our data, we cannot establish whether the proposed low-level correlate of behavioral variability is reflective of a diffuse deficit, of frontal-based executive function, or of impairments in endogenous attention reliant on frontoparietal circuits. Nevertheless, our results make the case that low-level perceptual function in ADHD deserves further investigation and that future task designs can easily include assessments of perceptual function—both as behavioral tasks and as questionnaires (Bijlenga et al., 2017; Kim, Chen, & Tannock, 2014; Micoulaud-Franchi et al., 2015; Panagiotidi et al., 2018)—in conjunction with attention and executive function. Using simple rather than high-level cognitive stimuli has the advantage that they can be used in parallel human and animal studies. Studies on animal models of ADHD, such as mouse (Leo & Gainetdinov, 2013; Majdak et al., 2016) and rat (Clements, Devonshire, Reynolds, & Overton, 2014), will provide further insight into the neural circuits implicated in ADHD and how medications can alter these circuits (Hetherington et al., 2017; Mueller, Hong, Shepard, & Moore, 2017).
ADHD diagnosis still relies predominantly on self and sometimes collateral reports, and widely accepted “psychomarkers” (also called “neurocognitive endophenotypes”) and biomarkers are lacking (Thome et al., 2012). For our findings to have implications for clinical practice, it is necessary that our task metrics be predictive of clinical metrics. We found that this was indeed the case. First, based on perceptual variability alone, we were able to classify participants into ADHD and control with cross-validated mean accuracy of 77.0% (including TIMO, 77.7%). Beyond binary classification, we also found strong correlations between behavioral metrics (σ, TIMO, and RT τ) and clinical ones (GEC and ACDS). Based on these correlations, the behavioral metrics in our task could be considered candidate psychomarkers for ADHD, similar to the performance on the CPT (Ogundele, Ayyash, & Banerjee, 2011), response variability (Castellanos & Tannock, 2002; Henríquez-Henríquez et al., 2015), and drift rate (Salum et al., 2014) and along with potential oculomotor markers like saccade patterns (Munoz, Armstrong, Hampton, & Moore, 2003), microsaccade rate in specific tasks (Dankner, Shalev, Carrasco, & Yuval-Greenberg, 2017; Fried et al., 2014; Panagiotidi, Overton, & Stafford, 2017), pupil size (Wainstein et al., 2017), or eye vergence (Casal et al., 2018). Psychomarkers and oculomotor markers are substantially easier and quicker to test for in large populations relative to other candidate biomarkers, for instance, based on neuroimaging or EEG data (Castellanos & Aoki, 2016; Faraone, Cristian, & Scassellati, 2014; Lenartowicz, Mazaheri, Jensen, & Loo, 2018). While there is a long pipeline from task and metric to clinically useful assay (Hitchcock, Radulescu, Niv, & Sims, 2017; Paulus, Huys, & Maia, 2016), simple behavioral paradigms and modeling applied to ADHD and other disorders could in the long term help refine diagnostic categories and inform and quantify the efficacy of treatment, as is the goal in computational psychiatry more broadly (Montague, Dolan, Friston, & Dayan, 2012; Redish & Gordon, 2016; Wiecki et al., 2015).
We thank Terry Leon, Michael Silverstein, Saima Milli, and Jonathan Yuh for assistance with protocol preparation, participant recruitment, and clinical data management and analysis. We thank Ili Ma and two anonymous reviewers for comments on earlier versions of this manuscript. In addition, we thank Eero Simoncelli, Marisa Carrasco, Roozbeh Kiani, Mariel Roberts, and members of the Ma lab, especially Maija Honig, Luigi Acerbi, Will Adler, Bas van Opheusden, and Aspen Yoo, for useful conversations. Lenard A. Adler makes the following disclosures for the last three years: Grant/Research: Purdue Pharma, Sunovion Pharmceuticals, Enzymotec, Shire Pharmaceuticals, Lundbeck; Consultant: Sunovion Pharmaceuticals, Shire Pharmaceuticals, Enzymotec, Otsuka Pharmaceuticals, National Football League, Major League Baseball, Rhodes Pharmaceuticals; Shire Pharmaceuticals, Alcobra Pharmaceuticals; Royalty payments (as inventor) from NYU for the license of adult ADHD scales and training materials since 2004. No ownership interests and no speaking fees.
Andra Mihali: Conceptualization: Equal; Data curation: Lead; Formal analysis: Lead; Investigation: Equal; Methodology: Equal; Software: Lead; Visualization: Lead; Writing – original draft: Lead. Allison Young: Conceptualization: Equal; Data curation: Supporting; Formal analysis: Supporting; Investigation: Equal; Methodology: Equal; Writing – review & editing: Equal. Lenard A. Adler: Conceptualization: Supporting; Funding acquisition: Equal; Methodology: Equal; Project administration: Equal; Resources: Equal; Supervision: Supporting; Writing – review & editing: Equal. Michael M. Halassa: Conceptualization: Equal; Formal analysis: Equal; Funding acquisition: Equal; Methodology: Equal; Supervision: Equal; Visualization: Supporting; Writing – review & editing: Equal. Wei Ji Ma: Conceptualization: Equal; Formal analysis: Lead; Funding acquisition: Equal; Investigation: Equal; Methodology: Lead; Supervision: Lead; Visualization: Equal; Writing – original draft: Lead.
Acerbi, L. (2016). Bayesian adaptive stimulus placement of psychometric function for Matlab. Retrieved from https://github.com/lacerbi/psybayes
Adler, L. A. , Faraone, S. V. , Spencer, T. J. , Berglund, P. , Alperin, S. , & Kessler, R. C. (2017). The structure of adult ADHD. International Journal of Methods in Psychiatric Research, 26(1), e1555.
Aston-Jones, G. , Rajkowski, J. , & Cohen, J. (1999). Role of locus coeruleus in attention and behavioral flexibility. Biological Psychiatry, 46(9), 1309–1320. https://doi.org/doi:10.1016/s0006-3223(99)00140-7
Bashinski, H. S. , & Bacharach, V. R. (1980). Enhancement of perceptual sensitivity as the result of selectively attending to spatial locations. Perception & Psychophysics, 28(3), 241–248. https://doi.org/doi:10.3758/BF03204380
Booth, J. R. , Burman, D. D. , Meyer, J. R. , Lei, Z. , Trommer, B. L. , Davenport, N. D. , … Mesulam, M. M. (2005). Larger deficits in brain networks for response inhibition than for visual selective attention in attention deficit hyperactivity disorder (ADHD). Journal of Child Psychology and Psychiatry, 46(1), 94–111.
Carrasco, M. (2014). Spatial covert attention ( A. C. K. Nobre & S. Kastner , Eds.). Oxford, England: Oxford University Press. https://doi.org/doi:10.1093/oxfordhb/9780199675111.013.004
Casal, P. V. , Esposito, F. L. , Martínez, I. M. , Capdevila, A. , Puig, M. S. , de la Osa, N. , … Cañete, J. (2018). Clinical validation of eye vergence as an objective marker for diagnosis of ADHD in children. Journal of Attention Disorders. https://doi.org/https://doi.org/10.1177/1087054717749931
Casey, B. J. , Castellanos, F. X. , Giedd, J. N. , Marsh, W. L. , Hamburger, S. D. , Schubert, A. B. , … Rapoport, J. L. (1997). Implication of right frontostriatal circuitry in response inhibition and attention-deficit/hyperactivity disorder. Journal of the American Academy of Child & Adolescent Psychiatry, 36(3), 374–383.
Castellanos, F. X. , & Aoki, Y. (2016). Intrinsic functional connectivity in attention-deficit/hyperactivity disorder: A science in development. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, 1(3), 253–261. https://doi.org/doi:10.1016/j.bpsc.2016.03.004
Castellanos, F. X. , Margulies, D. S. , Kelly, C. , Uddin, L. Q. , Ghaffari, M. , Kirsch, A. , … Milham, M. P. (2008). Cingulate-precuneus interactions: A new locus of dysfunction in adult attention-deficit/hyperactivity disorder. Biological Psychiatry, 63(3), 332–337. https://doi.org/doi:10.1016/j.biopsych.2007.06.025
Chen, J. , & Niemeier, M. (2017). Altered perceptual pseudoneglect in ADHD: Evidence for a functional disconnection from early visual activation. Neuropsychologia, 99, 12–23. https://doi.org/doi:10.1016/j.neuropsychologia.2017.02.022
Clements, K. , Devonshire, I. , Reynolds, J. , & Overton, P. (2014). Enhanced visual responses in the superior colliculus in an animal model of attention-deficit hyperactivity disorder and their suppression by d-amphetamine. Neuroscience, 274, 289–298. https://doi.org/doi:10.1016/j.neuroscience.2014.05.054
Cornelissen, F. W. , Peters, E. M. , & Palmer, J. (2002). The eyelink toolbox: Eye tracking with MATLAB and the psychophysics toolbox. Behavior Research Methods, Instruments & Computers, 34(4), 613–617.
Cubillo, A. , Halari, R. , Ecker, C. , Giampietro, V. , Taylor, E. , & Rubia, K. (2010). Reduced activation and inter-regional functional connectivity of fronto-striatal networks in adults with childhood attention-deficit hyperactivity disorder (ADHD) and persisting symptoms during tasks of motor inhibition and cognitive switching. Journal of Psychiatric Research, 44(10), 629–639.
Dankner, Y. , Shalev, L. , Carrasco, M. , & Yuval-Greenberg, S. (2017). Prestimulus inhibition of saccades in adults with and without attention-deficit/hyperactivity disorder as an index of temporal expectations. Psychological Science, 28(7), 835–850.
del Campo, N. , Chamberlain, S. R. , Sahakian, B. J. , & Robbins, T. W. (2011). The roles of dopamine and noradrenaline in the pathophysiology and treatment of attention-deficit/hyperactivity disorder. Biological Psychiatry, 69(12), e145–e157.
Devilbiss, D. M. , & Berridge, C. W. (2006). Low-dose methylphenidate actions on tonic and phasic locus coeruleus discharge. Journal of Pharmacology and Experimental Therapeutics, 319(3), 1327–1335. https://doi.org/doi:10.1124/jpet.106.110015
Dormann, C. F. , Elith, J. , Bacher, S. , Buchmann, C. , Carl, G. , Carré, G. , … Lautenbach, S. (2012). Collinearity: A review of methods to deal with it and a simulation study evaluating their performance. Ecography, 36(1), 27–46. https://doi.org/doi:10.1111/j.1600-0587.2012.07348.x
Downing, C. J. (1988). Expectancy and visual-spatial attention: Effects on perceptual quality. Journal of Experimental Psychology: Human Perception and Performance, 14(2), 188–202. https://doi.org/doi:10.1037/0096-1522.214.171.124
Fair, D. A. , Bathula, D. , Nikolas, M. A. , & Nigg, J. T. (2012). Distinct neuropsychological subgroups in typically developing youth inform heterogeneity in children with ADHD. Proceedings of the National Academy of Sciences, 109(17), 6769–6774.
Faraone, S. V. , Cristian, B. , & Scassellati, C. (2014). Biomarkers in the diagnosis of adhd—promising directions. Current Psychiatry Reports, 16(11). https://doi.org/doi:10.1007/s11920-014-0497-1
Frank, M. J. , Santamaria, A. , OReilly, R. C. , & Willcutt, E. (2006). Testing computational models of dopamine and noradrenaline dysfunction in attention deficit/hyperactivity disorder. Neuropsychopharmacology, 32(7), 1583–1599.
Fried, M. , Tsitsiashvili, E. , Bonneh, Y. S. , Sterkin, A. , Wygnanski-Jaffe, T. , Epstein, T. , & Polat, U. (2014). ADHD subjects fail to suppress eye blinks and microsaccades while anticipating visual stimuli but recover with medication. Vision Research, 101, 62–72.
Friedman-Hill, S. R. , Wagman, M. R. , Gex, S. E. , Pine, D. S. , Leibenluft, E. , & Ungerleider, L. G. (2010). What does distractibility in ADHD reveal about mechanisms for top-down attentional control? Cognition, 115(1), 93–103.
Fuermaier, A. B. M. , Hupen, P. , Vries, S. M. D. , Muller, M. , Kok, F. M. , Koerts, J. , … Tucha, O. (2017). Perception in attention deficit hyperactivity disorder. ADHD Attention Deficit and Hyperactivity Disorders, 10, 21–47.
Fuller, S. , & Carrasco, M. (2006). Exogenous attention and color perception: Performance and appearance of saturation and hue. Vision Research, 46(23), 4032–4047. https://doi.org/doi:10.1016/j.visres.2006.07.014
Greenberg, A. S. , Esterman, M. , Wilson, D. , Serences, J. T. , & Yantis, S. (2010). Control of spatial and feature-based attention in frontoparietal cortex. Journal of Neuroscience, 30(43), 14330–14339.
Haigh, S. M. , Heeger, D. J. , Heller, L. M. , Gupta, A. , Dinstein, I. , Minshew, N. J. , & Berhmann, M. (2016). No difference in cross-modal attention or sensory discrimination thresholds in autism and matched controls. Vision Research, 121, 85–94.
Halassa, M. M. , & Kastner, S. (2017). Thalamic functions in distributed cognitive control. Nature Neuroscience, 20(12), 1669–1679. https://doi.org/doi:10.1038/s41593-017-0020-1
Hauser, T. U. , Fiore, V. G. , Moutoussis, M. , & Dolan, R. J. (2016). Computational psychiatry of ADHD: Neural gain impairments across marrian levels of analysis. Trends in Neurosciences, 39(2), 63–73.
Hauser, T. U. , Iannaccone, R. , Ball, J. , Mathys, C. , Brandeis, D. , Walitza, S. , & Brem, S. (2014). Role of the medial prefrontal cortex in impaired decision making in juvenile attention-deficit/hyperactivity disorder. JAMA Psychiatry, 71(10), 1165–1173.
Henríquez-Henríquez, M. P. , Billeke, P. , Henríquez, H. , Zamorano, F. J. , Rothhammer, F. , & Aboitiz, F. (2015). Intra-individual response variability assessed by ex-Gaussian analysis may be a new endophenotype for attention-deficit/hyperactivity disorder. Frontiers in Psychiatry, 5, Article 197.
Herman, J. P. , Bogadhi, A. R. , & Krauzlis, R. J. (2015). Effects of spatial cues on color-change detection in humans. Journal of Vision, 15, 1–16. https://doi.org/doi:10.1167/15.6.3
Hetherington, L. , Dommett, E. , Turner, A. , Riley, T. , Haensel, J. , & Overton, P. (2017). Effect of methylphenidate on visual responses in the superior colliculus in the anaesthetised rat: Role of cortical activation. Journal of Psychopharmacology, 31(10), 1347–1361. https://doi.org/doi:10.1177/0269881117730661
Hitchcock, P. , Radulescu, A. , Niv, Y. , & Sims, C. R. (2017). Translating a reinforcement learning task into a computational psychiatry assay: Challenges and strategies. Proceedings of the 39th Annual Conference of the Cognitive Science Society, 39, 2217–2222.
Huang-Pollock, C. , Ratcliff, R. , McKoon, G. , Shapiro, Z. , Weigard, A. , & Galloway-Long, H. (2016). Using the diffusion model to explain cognitive deficits in attention deficit hyperactivity disorder. Journal of Abnormal Child Psychology, 45(1), 57–68.
Huang-Pollock, C. L. , & Nigg, J. T. (2003). Searching for the attention deficit in attention deficit hyperactivity disorder: The case of visuospatial orienting. Clinical Psychology Review, 23(6), 801–830.
Karalunas, S. L. , & Huang-Pollock, C. L. (2013). Integrating impairments in reaction time and executive function using a diffusion model framework. Journal of Abnormal Child Psychology, 41(5), 837–850.
Karalunas, S. L. , Huang-Pollock, C. L. , & Nigg, J. T. (2012). Decomposing attention-deficit/hyperactivity disorder (ADHD)-related effects in response speed and variability. Neuropsychology, 26(6), 684–694.
Keshvari, S. , van den Berg, R. , & Ma, W. J. (2012). Probabilistic computation in human perception under variability in encoding precision. PLoS ONE, 7(6), e40216. https://doi.org/doi:10.1371/journal.pone.0040216
Kessler, R. C. , Adler, L. , Ames, M. , Delmer, O. , Faraone, S. , Hiripi, E. , … Walters, E. E. (2005). The World Health Organization adult ADHD Self-Report Scale (ASRS): A short screening scale for use in the general population. Psychological Medicine, 35(2), 245–256.
Kessler, R. C. , Adler, L. , Barkley, R. , Biederman, J. , Conners, C. K. , Demler, O. , … Zaslavsky, A. M. (2006). The prevalence and correlates of adult ADHD in the United States: Results from the national comorbidity survey replication. American Journal of Psychiatry, 163(4), 716–723.
Kim, S. , Al-Haj, M. , Chen, S. , Fuller, S. , Jain, U. , Carrasco, M. , & Tannock, R. (2014). Colour vision in ADHD: Part 1—testing the retinal dopaminergic hypothesis. Behavioral and Brain Functions, 10(1), Article 38.
Kim, S. , Al-Haj, M. , Fuller, S. , Chen, S. , Jain, U. , Carrasco, M. , & Tannock, R. (2014). Color vision in ADHD: Part 2—does attention influence color perception? Behavioral and Brain Functions, 10(1), Article 39.
King, J. A. , Colla, M. , Brass, M. , Heuser, I. , & von Cramon, D. Y. (2007). Inefficient cognitive control in adult ADHD: Evidence from trial-by-trial Stroop test and cued task switching performance. Behavioral and Brain Functions, 3(1), Article 42.
Klein, S. A. (2001). Measuring, estimating, and understanding the psychometric function: A commentary. Perception & Psychophysics, 63(8), 1421–1455. https://doi.org/doi:10.3758/BF03194552
Kofler, M. J. , Rapport, M. D. , Sarver, D. E. , Raiker, J. S. , Orban, S. A. , Friedman, L. M. , & Kolomeyer, E. G. (2013). Reaction time variability in ADHD: A meta-analytic review of 319 studies. Clinical Psychology Review, 33(6), 795–811.
Krauzlis, R. J. , Lovejoy, L. P. , & Zénon, A. (2013). Superior colliculus and visual spatial attention. Annual Review of Neuroscience, 36(1), 165–182. https://doi.org/doi:10.1146/annurev-neuro-062012-170249
Lam, N. H. , Borduqui, T. , Hallak, J. , Roque, A. C. , Anticevic, A. , Krystal, J. H. , … Murray, J. D. (2017). Effects of altered excitation-inhibiton balance on decision making in a cortical circuit model. Unpublished manuscript. https://doi.org/doi:10.1101/100347
Lenartowicz, A. , Mazaheri, A. , Jensen, O. , & Loo, S. K. (2018). Aberrant modulation of brain oscillatory activity and attentional impairment in attention-deficit/hyperactivity disorder. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, 3(1), 19–29. https://doi.org/doi:10.1016/j.bpsc.2017.09.009
Leth-Steensen, C. , Elbaz, Z. K. , & Douglas, V. I. (2000). Mean response times, variability, and skew in the responding of ADHD children: A response time distributional approach. Acta Psychologica, 104(2), 167–190.
Ling, S. , & Carrasco, M. (2006). Sustained and transient covert attention enhance the signal via different contrast response functions. Vision Research, 46(8–9), 1210–1220. https://doi.org/doi:10.1016/j.visres.2005.05.008
Lúcio, P. S. , Salum, G. A. , Rohde, L. A. , Swardfager, W. , Gadelha, A. , Vandekerckhove, J. , … Cogo-Moreira, H. (2016). Poor stimulus discriminability as a common neuropsychological deficit between ADHD and reading ability in young children: A moderated mediation model. Psychological Medicine, 47(2), 255–266. https://doi.org/doi:10.1017/s0033291716002531
Majdak, P. , Ossyra, J. R. , Ossyra, J. M. , Cobert, A. J. , Hofmann, G. C. , Tse, S. , … Rhodes, J. S. (2016). A new mouse model of ADHD for medication development. Scientific Reports, 6(1). https://doi.org/doi:10.1038/srep39472
McAvinue, L. P. , Vangkilde, S. , Johnson, K. A. , Habekost, T. , Kyllingsbaek, S. , Bundesen, C. , & Robertson, J. H. (2012). A componential analysis of visual attention in children with ADHD. Journal of Attention Disorders, 19(10), 882–894.
Micoulaud-Franchi, J.-A. , Lopez, R. , Vaillant, F. , Richieri, R. , El-Kaim, A. , Bioulac, S. , … Lancon, C. (2015). Perceptual abnormalities related to sensory gating deficit are core symptoms in adults with ADHD. Psychiatry Research, 230(2), 357–363.
Mihali, A. , Young, A. G. , Adler, L. A. , Halassa, M. M. , & Ma, W. J. (2018). Appendix for “A low-level perceptual correlate of behavioral and clinical deficits in ADHD.” Computational Psychiatry, 2, 141–163. https://doi.org/https://doi.org/10.1162/cpsy_a_00018
Mostert, P. , Kok, P. , & de Lange, F. P. (2015). Dissociating sensory from decision processes in human perceptual decision making. Scientific Reports, 5(1). https://doi.org/doi:10.1038/srep18253
Munoz, D. P. , Armstrong, I. T. , Hampton, K. A. , & Moore, K. D. (2003). Altered control of visual fixation and saccadic eye movements in attention-deficit hyperactivity disorder. Journal of Neurophysiology, 90(1), 503–514. https://doi.org/doi:10.1152/jn.00192.2003
Nigg, J. T. , & Casey, B. J. (2005). An integrative theory of attention-deficit/hyperactivity disorder based on the cognitive and affective neurosciences. Development and Psychopathology, 17(3), 785–806.
Nigg, J. T. , Willcutt, E. G. , Doyle, A. E. , & Sonuga-Barke, E. J. (2005). Causal heterogeneity in attention-deficit/hyperactivity disorder: Do we need neuropsychologically impaired subtypes? Biological Psychiatry, 57(11), 1224–1230. https://doi.org/doi:10.1016/j.biopsych.2004.08.025
Panagiotidi, M. , Overton, P. G. , & Stafford, T. (2018). The relationship between ADHD traits and sensory sensitivity in the general population. Comprehensive Psychiatry, 80, 179–185. https://doi.org/doi:10.1016/j.comppsych.2017.10.008
Paulus, M. P. , Huys, Q. J. , & Maia, T. V. (2016). A roadmap for the development of applied computational psychiatry. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, 1(5), 386–392. https://doi.org/doi:10.1016/j.bpsc.2016.05.001
Pestilli, F. , Viera, G. , & Carrasco, M. (2007). How do attention and adaptation affect contrast sensitivity? Journal of Vision, 7(7), Article 9. https://doi.org/doi:10.1167/7.7.9
Ravizza, S. M. , & Carter, C. S. (2008). Shifting set about task switching: Behavioral and neural evidence for distinct forms of cognitive flexibility. Neuropsychologia, 46(12), 2924–2935. https://doi.org/doi:10.1016/j.neuropsychologia.2008.06.006
Rommelse, N. N. J. , Altink, M. E. , de Sonneville, L. M. J. , Buschgens, C. J. M. , Buitelaar, J. , Oosterlaan, J. , & Sergeant, J. A. (2007). Are motor inhibition and cognitive flexibility dead ends in ADHD? Journal of Abnormal Child Psychology, 35(6), 957–967. https://doi.org/doi:10.1007/s10802-007-9146-z
Roth, R. M. , Lance, C. E. , Isquith, P. K. , Fischer, A. S. , & Giancola, P. R. (2013). Confirmatory factor analysis of the behavior rating inventory of executive function-adult version in healthy adults and application to attention-deficit/hyperactivity disorder. Archives of Clinical Neuropsychology, 28(5), 425–434.
Rubia, K. , Halari, R. , Cubillo, A. , Mohammad, A.-M. , Scott, S. , & Brammer, M. (2010). Disorder-specific inferior prefrontal hypofunction in boys with pure attention-deficit/hyperactivity disorder compared to boys with pure conduct disorder during cognitive flexibility. Human Brain Mapping, 31(12), 1823–1833.
Salum, G. A. , Sonuga-Barke, E. , Sergeant, J. , Vandekerckhove, J. , Gadelha, A. , Moriyama, T. S. , … Rohde, L. A. P. (2014). Mechanisms underpinning inattention and hyperactivity: Neurocognitive support for ADHD dimensionality. Psychological Medicine, 44(15), 3189–3201.
Schmitt, L. I. , Wimmer, R. D. , Nakajima, M. , Happ, M. , Mofakham, S. , & Halassa, M. M. (2017). Thalamic amplification of cortical connectivity sustains attentional control. Nature, 545(7653), 219–223. https://doi.org/doi:10.1038/nature22073
Silverstein, M. J. , Faraone, S. V. , Alperin, S. , Leon, T. L. , Biederman, J. , Spencer, T. J. , & Aldler, L. A. (2018). Validation of the expanded versions of the adult ADHD Self-Report Scale v1.1 symptom checklist and the adult ADHD Investigator Symptom Rating Scale. Journal of Attention Disorders. https://doi.org/doi:10.1177/1087054718756198
Sonuga-Barke, E. J. , & Castellanos, F. X. (2007). Spontaneous attentional fluctuations in impaired states and pathological conditions: A neurobiological hypothesis. Neuroscience & Biobehavioral Reviews, 31(7), 977–986. https://doi.org/doi:10.1016/j.neubiorev.2007.02.005
Stevens, A. A. , Maron, L. , Nigg, J. T. , Cheung, D. , Ester, E. , & Awh, E. (2012). Increased sensitivity to perceptual interference in adults with attention deficit hyperactivity disorder. Journal of the International Neuropsychological Society, 18, 511–520.
Thome, J. , Ehlis, A.-C. , Fallgatter, A. J. , Krauel, K. , Lange, K. W. , Riederer, P. , … Gerlach, M. (2012). Biomarkers for attention-deficit/hyperactivity disorder (ADHD): A consensus report of the WFSBP Task Force on Biological Markers and the World Federation of ADHD. World Journal of Biological Psychiatry, 13(5), 379–400.
Wainstein, G. , Rojas-Líbano, D. , Crossley, N. A. , Carrasco, X. , Aboitiz, F. , & Ossandón, T. (2017). Pupil size tracks attentional performance in attention-deficit/hyperactivity disorder. Scientific Reports, 7(1), Article 8228.
Willcutt, E. G. , Doyle, A. E. , Nigg, J. T. , Faraone, S. V. , & Pennington, B. F. (2005). Validity of the executive function theory of attention-deficit/hyperactivity disorder: A meta-analytic review. Biological Psychiatry, 57(11), 1336–1346.
Zandbelt, B. (2014). Exgauss: A Matlab toolbox for fitting the ex-Gaussian distribution to response time data. https://doi.org/doi:10.6084/m9.figshare.971318.v2
Ziegler, S. , Pedersen, M. L. , Mowinckel, A. M. , & Biele, G. (2016). Modelling ADHD: A review of ADHD theories through their predictions for computational models of decision-making and reinforcement learning. Neuroscience & Biobehavioral Reviews, 71, 633–656.