INTRODUCTION
Spatial cognition refers to the perception, organization, maintenance, and revision of information about the spatial relationships (i.e., distance, direction, and rotation) of the parts within an object, or between an object and its environment (de Vega, Intons-Peterson, Johnson-Laird, Denis, & Marschark, Reference de Vega, Intons-Peterson, Johnson-Laird, Denis and Marschark1996). These abilities are shared by humans and non-humans (Landau & Jackendoff, Reference Landau and Jackendoff1993; Thinus-Blanc, Reference Thinus-Blanc1996) and are critical to environmental learning and navigation, perspective taking, and the execution of locomotor and visually guided actions (Kessler & Thomson, Reference Kessler and Thomson2010; Pham & Hicheur, Reference Pham and Hicheur2009; Wolbers & Hegarty, Reference Wolbers and Hegarty2010). Given the regularity with which humans rely on these functions in daily life, understanding the mechanisms underlying those processes has long been a focal area of research.
Among the most widely used tasks to evaluate spatial cognition in research is Shepard and Metzler’s (Reference Shepard and Metzler1971) mental rotation task. In the most common version, participants are shown two unfamiliar shapes, each at different angular rotations, and are asked to decide if the two stimuli are identical. A large body of evidence suggests that individuals complete the task using analog spatial representations (Zacks, Reference Zacks2008). First, observers encode a stimulus in the form of an abstract representation which is maintained in memory via rehearsal (Prime & Jolicoeur, Reference Prime and Jolicoeur2010). The stored representation is then mentally rotated to match the orientation of the stimulus to which it is being compared. Finally, comparisons are drawn between the two stimuli, which informs the parity judgment. Dating back to Shepard and Metzler’s (Reference Shepard and Metzler1971) seminal study, research has consistently found that mean response time (RT), error rate, as well as activation in the superior parietal cortex (an area of the brain that codes target locations in space) all increase as a function of angular disparity between the stimuli (Zacks, Reference Zacks2008). This effect has been attributed to the longer period of time required to rotate the analog stimulus to alignment.
Since Shepard and Metzler’s early work, cognitive psychology has been interested in better understanding how experimental manipulations may impact one’s ability to perform mental rotations. For example, studies have examined the influence of: practice (Meneghetti, Borella, & Pazzaglia, Reference Meneghetti, Borella and Pazzaglia2016), cognitive load (Pannebakker et al., Reference Pannebakker, Jolicœur, van Dam, Band, Ridderinkhof and Hommel2011), and stimulus dimensionality and complexity (Bethell-Fox & Shepard, Reference Bethell-Fox and Shepard1988). However, using the standard metrics of mean RT (MRT) and mean accuracy makes it difficult to identify where in the stream of information processing (i.e., encoding/rehearsing, rotating, decision-making) that any observed effects may originate. This in turn poses a challenge for efforts to reliably quantify the independent contributions of rotation-specific and more general decision-making processes on performance.
One way of addressing this issue is through the use of the diffusion model (DM; Ratcliff & McKoon, Reference Ratcliff and McKoon2008; Voss, Rothermund, & Voss, Reference Voss, Rothermund and Voss2004), a computational model developed to explain rapid forced choice decision-making (e.g., do the stimuli match or not match?). The DM assumes that these types of decisions are made via three main information processing stages: encoding, accumulation of evidence toward a decision, and the planning/execution of a motoric response. After a stimulus is encoded, evidence accumulates toward one of the two possible responses (i.e., match vs. not-match). When the accumulation of evidence crosses either response boundary, the decision is reached and the appropriate behavioral response is then initiated. Drift rate (v) reflects the speed with which the information accumulation process occurs. The distance between the two response boundaries is referred to as the boundary separation (a). Boundary separation is wider when participants emphasize accuracy over speed (intrinsically or by instruction) and is narrower when speed is prioritized. There is also evidence that boundary separation increases when rotation angle increases, and when presented with mirrored (vs. identical) stimuli (Provost & Heathcote, Reference Provost and Heathcote2015). Finally, nondecision time (Ter) reflects the amount of time it takes to complete all other processes that are not involved in the decision process, namely, the time it takes to encode a stimulus and the planning and execution of motor responses. There is considerable evidence that motor imagery, including mental rotation, engages areas of the premotor and supplementary motor cortex, which are also involved in motor planning and execution (Kosslyn, Ganis, & Thompson, Reference Kosslyn, Ganis and Thompson2001; Wexler, Kosslyn, & Berthoz, Reference Wexler, Kosslyn and Berthoz1998). Given this, and that rotation is assumed to occur prior to the onset of the decision stage (Cooper & Shepard, Reference Cooper, Shepard and Chase1973; Heil, Reference Heil2002; Riečanský & Katina, Reference Riečanský and Katina2010), a DM framework would assume the rotation process itself to be captured by Ter.
The DM performance parameters are produced based upon the shape of the RT distribution for both correct and error responses for each individual. That is, compared to traditional analytic techniques that rely solely on either the mean reaction time or mean accuracy of responses, the DM uses the entirety of the data to provide a more complete description of performance. And, unlike other statistical approaches such as Shifted Wald or ex-Gaussian modeling, the DM parameters have clearly defined psychological interpretations that have been experimentally validated in several studies since the model was first conceived in the 1970s (Matzke & Wagenmakers, Reference Matzke and Wagenmakers2009; Ratcliff & McKoon, Reference Ratcliff and McKoon2008; Voss et al., Reference Voss, Rothermund and Voss2004; Wagenmakers, Reference Wagenmakers2009).
The DM has been particularly useful in understanding the cognitive correlates of childhood ADHD. Among the most consistent hallmarks of ADHD is slow, variable, and error-prone performance on experimental paradigms designed to measure many different types of executive and non-executive processes. Using this approach, the broad consensus has been that slow drift rate explains much of the ADHD-related variance observed in performance across a broad range of these cognitive paradigms (Fosco, White, & Hawk, Reference Fosco, White and Hawk2017; Huang-Pollock, Karalunas, Tam, & Moore, Reference Huang-Pollock, Karalunas, Tam and Moore2012; Karalunas & Huang-Pollock, Reference Karalunas and Huang-Pollock2013; Karalunas, Huang-Pollock, & Nigg, Reference Karalunas, Huang-Pollock and Nigg2012; Metin et al., Reference Metin, Roeyers, Wiersema, van der Meere, Thompson and Sonuga-Barke2013; Mulder et al., Reference Mulder, Bos, Weusten, van Belle, van Dijk, Simen and Durston2010; Weigard & Huang-Pollock, Reference Weigard and Huang-Pollock2014, Reference Weigard and Huang-Pollock2017; Ziegler, Pedersen, Mowinckel, & Biele, Reference Ziegler, Pedersen, Mowinckel and Biele2016). That is, slow drift rate has been found in ADHD regardless of the type of task on which it is measured.
That being said, small effects for timing-specific and sustained attention processes, independent of generally slow drift rate, have been documented (Huang-Pollock et al., Reference Huang-Pollock, Ratcliff, McKoon, Roule, Warner, Feldman and Wise2020; Shapiro & Huang-Pollock, Reference Shapiro and Huang-Pollock2019). Specific deficits in spatial short-term memory, spatial planning, and visual recognition have also been noted (Chiang & Gau, Reference Chiang and Gau2014; Ferrin & Vance, Reference Ferrin and Vance2012; Semrud-Clikeman, Reference Semrud-Clikeman2012) and can persist in adolescents who cease to meet full diagnostic criteria (Lin, Chen, & Gau, Reference Lin, Chen and Gau2014). In fact, meta-analytic effect sizes for spatial working memory (WM) deficits in ADHD are among the largest across all neurocognitive domains (.63–1.06), are greater than effect sizes for verbal WM (.43–.55) (Martinussen, Hayden, Hogg-Johnson, & Tannock, Reference Martinussen, Hayden, Hogg-Johnson and Tannock2005; Willcutt, Doyle, Nigg, Faraone, & Pennington, Reference Willcutt, Doyle, Nigg, Faraone and Pennington2005), and have been shown to persist even after controlling for IQ, spatial memory span, and other rapid processing tasks (Lin et al., Reference Lin, Chen and Gau2014). Nevertheless, specific weaknesses in manipulation of spatial information are less widely studied by comparison and often yield inconsistent interpretations across different tasks and metrics.
Three previous studies have utilized mental rotation tasks to evaluate the status of spatial cognition in ADHD (Silk et al., Reference Silk, Vance, Rinehart, Egan, O’Boyle, Bradshaw and Cunnington2005; Vance et al., Reference Vance, Silk, Casey, Rinehart, Bradshaw, Bellgrove and Cunnington2007; Williams, Omizzolo, Galea, & Vance, Reference Williams, Omizzolo, Galea and Vance2013), but all had unacceptably high error rates for RT to be confidently used as valid indices of performance (error rates ranging from 30 to 73% for ADHD and 8–44% for controls). When RTs are used as the sole index of performance, error trials are usually removed from analysis to prevent confounds to the interpretation of RT. However, when error rates are high, basic assumptions about the cognitive process that occurred to produce that RT can no longer be made confidently, including assumptions that participants were sufficiently engaged with the task (Shepard & Cooper, Reference Shepard and Cooper1982). A fourth study reported lower accuracy among children with ADHD versus non-ADHD controls, but the speed with which they made those decisions was not reported (Jakobson & Kikas, Reference Jakobson and Kikas2007). Thus, the status of visuospatial processing in ADHD is unresolved.
In the present study, we examine visuospatial cognition in children with and without ADHD using a mental rotation task. We selected this task because it is one of the most heavily researched tasks of visuospatial cognitive ability that allows both accuracy and RT to be measured in a trial-by-trial fashion. Mental rotation was also an attractive option for the current study because it has previously been studied in adults using evidence accumulation models similar to the DM (Larsen, Reference Larsen2014; Provost & Heathcote, Reference Provost and Heathcote2015). If specific difficulties in the mental rotation process exist, it would be captured by a significant ADHD × Rotation interaction for Ter, where group differences increase with angle of rotation. Otherwise, main effects of diagnostic group in the absence of an interaction with rotation would suggest that poor ADHD-related performance is better attributed to task or process non-specific performance deficits. We apply the DM, which produces parameters based on the RTs to error as well as correct responses, to evaluate these possibilities.
Methods
Participants
Three hundred and six children with (n = 207, 142 boys) and without (n = 99, 48 boys) ADHD participated. They were recruited through advertisement via local schools, newspaper and radio ads, and fliers distributed throughout Centre and Dauphin counties in Pennsylvania. All children were between 8 and 12 years of age and were ethnically representative of the region: 74% Caucasian, 6% Caucasian/Hispanic, 8% African American/non-Hispanic, 1% African American/Hispanic, 1% Other Hispanic, 1% Asian, and 8% mixed/unknown. Children who were prescribed a non-stimulant psychoactive medication; who had an estimated Full Scale IQ below 80 based on a two-subtest short form (i.e., vocabulary and matrix reasoning, predictive validity = .87) of the Wechsler Intelligence Scale for Children (Wechsler, Reference Wechsler2003); and those with a history of parent-reported head injuries, psychosis, neurological, developmental, intellectual, or sensorimotor disabilities were excluded from the study.
Children with ADHD
Children with ADHD met DSM criteria (any presentation type), including duration, age of onset, and multi-context impairment (American Psychiatric Association, 2013). Parent report of symptomology was obtained via the Diagnostic Interview Schedule for Children-IV (DISC-IV; Shaffer, Fisher, & Lucas, Reference Shaffer, Fisher and Lucas1997). To demonstrate cross-situational impairment, at least one parent and one teacher report of behavior on the Attention, Hyperactivity, or ADHD subscales of the Behavioral Assessment Scale for Children (BASC-2; Reynolds & Kamphaus, Reference Reynolds and Kamphaus2004) or the Conners’ Rating Scales (Conners’; Conners, Reference Conners2001) were required to exceed the 85th percentile (i.e., T-score ≥ 61), and at least 3 symptoms were required to be present at an impairing level at school. Both the BASC-2 and Conners’ are commonly used, well-validated measures for the evaluation of ADHD. Following DSM-IV field trials (Lahey et al., Reference Lahey, Applegate, McBurnett, Biederman, Greenhill, Hynd and Richters1994), diagnostic determination and final symptom counts followed an “or” algorithm to integrate parent responses on the DISC with teacher reports on the ADHD Rating Scale (DuPaul, Power, Anastopoulos, & Reid, Reference DuPaul, Power, Anastopoulos and Reid1998). Children prescribed stimulant medication (n = 68) were asked to discontinue medication use for 24–48 h (Mean washout = 103.43 h).
Non-ADHD controls
Controls did not meet ADHD criteria on DISC-IV, had T-scores below the 80th percentile (T-score < 58) on all ADHD-related parent and teacher rating scales, and had never been previously diagnosed or treated for ADHD. All had ≤4 total symptoms and ≤3 symptoms per ADHD dimension according to the “or” algorithm.
Descriptive statistics of groups are provided in Table 1. The presence of anxiety, depression, oppositional defiant, and conduct disorders was not exclusionary for either group. To control for the potential confounding effects of IQ on the high end of the spectrum, controls were required to have estimated IQs < 115. No upper IQ limit was set for children with ADHD. Compared to controls, children with ADHD had more symptoms of hyperactivity/impulsivity and inattention (both p ≤ .001, both η 2 ≥ .54). There were no group differences in estimated IQ or age (both p ≥ .61, both η 2 ≤ .001).
Table 1. Means (SD) of group descriptives
BASC-2 = Behavioral Assessment Scale for Children; Conners’ = Conners’ Rating Scales; H = Primarily hyperactive/impulsive subtype; I = Primarily inattentive subtype; C = Primarily combined subtype; ODD = Oppositional Defiant Disorder; CD = Conduct Disorder; GAD = Generalized Anxiety Disorder; MDD = Major Depressive Disorder; DD = Dysthymic Disorder; LD = Learning Disability.
Rating scales reported in T-scores.
Procedures
Informed written consent from parents and verbal assent from children were obtained prior to participation. Parents were provided with $100 and clinical feedback; children were given their choice of a small prize. The mental rotation paradigm completed by participating children was one of a battery of tasks associated with a larger study investigating neurocognitive deficits in childhood ADHD. The Pennsylvania State University Institutional Review Board (IRB#32126) approved all study procedures.
Mental Rotation
Task description
There were three blocks of 48 trials (144 total trials). For each trial, the word “Ready?” first appeared for 500 ms (see Figure 1(a)). Children were then shown the comparison stimulus alongside a target stimulus and asked to indicate whether the two were the same by pressing either “yes” or “no” on a response box. The comparison was always an upright stick figure with a yellow dot on the figure’s right hand and a blue dot on the figure’s left foot. Stimuli remained on the screen until a response was made, after which the child was given visual feedback (i.e., “Right!” or “Wrong!”) for 1000 ms. Five practice trials preceded the experimental trials.
Fig. 1. (a) The mental rotation paradigm and illustration of rotation angle groupings. (b) Illustration of rotation angle groupings.
Half of the targets were congruent (i.e., matching), and the other half were incongruent (i.e., mirror image) to the comparison. Each target was rotated at one of the eight angular rotations (i.e., 0°, 45°, 90°, 135°, 180°, 225°, 270°, or 315°) from the comparison (see Figure 1(b)), so that there were 18 trials per rotational angle. To allow enough trials for parameter recovery, trials with target stimuli oriented at 0°, 45°, and 315° were classified as “small” rotations, while those with 135°, 180°, and 225° angle rotation trials were classified as “large” rotations (72 trials per bin). Stimuli rotated at either 90° or 270° ultimately require the same amount of rotation to the upright position and did not differ from one another in either MRT, F(1,11014) = 3.22, η 2 < .001, p = .073, or accuracy, F(1,11014) = .63, η 2 < .001, p = .43. Therefore, the random case selection functionality in SPSS was used to classify half of each of the 90° and 270° trials (independently) into “small” and “large” rotations. Results did not vary when all 90° and 270° stimuli were classified as small and large, respectively, or when categorized in the reverse.
Data Analysis Plan
Anticipatory responses faster than 300 ms were discarded per DM convention (Ratcliff, Love, Thompson, & Opfer, Reference Ratcliff, Love, Thompson and Opfer2012; Ratcliff & Tuerlinckx, Reference Ratcliff and Tuerlinckx2002). In line with a prior accumulator analysis of mental rotation, trials slower than 7000 ms were also excluded (Provost & Heathcote, Reference Provost and Heathcote2015; Provost, Johnson, Karayanidis, Brown, & Heathcote, Reference Provost, Johnson, Karayanidis, Brown and Heathcote2013). Twenty-two children with ADHD and two non-ADHD controls were excluded from analysis based on task performance (i.e., if overall accuracy <50%, or if >25% of trials fell outside of the RT cut-offs). Children with ADHD who were removed from analysis were younger than children with ADHD who remained in analysis (M = 8.50 vs. 9.55 years, F(1,227) = 15.07, η 2 = .062, p < .001). Because there were no significant differences in IQ, F(1,227) = .003, η 2 < .001, p = .96, number of inattentive symptoms, F(1,227) = 1.01, η 2 = .004, p = .32, or number of hyperactivity/impulsivity symptoms, F(1,227) = .43, η 2 = .002, p = .51, among children with ADHD who were removed versus retained from analysis, we were reassured that our exclusionary criteria did not skew our sample toward greater or lesser severity and remained representative of children with ADHD. Additionally, children with ADHD and a comorbid learning disability who were retained in the final sample did not differ from those without a comorbid learning disability on any performance metric (all p > .066, all η 2 < .016).
The original sample included 229 children with ADHD and 101 children without ADHD, resulting in the final N reported earlier: n = 207 children with ADHD and n = 99 non-ADHD controls.
Diffusion modeling
Trial-by-trial RT and accuracy data for each subject were input into the Fast-dm modeling program (downloadable at http://www.psychologie.uni-heidelberg.de/ae/meth/fast-dm; Voss & Voss, Reference Voss and Voss2007) to provide individual estimates of drift rate, boundary separation, and nondecision time by block and rotation condition. Response bias (z) was fixed to a/2, and inter-trial variability parameters were fixed to zero, as they are difficult to estimate and contribute little to the shape of the distribution (Voss & Voss, Reference Voss and Voss2007). AIC model selection favored this model over others in which Ter was fixed across conditions and drift and response bias varied by match/nonmatch trials (Supplemental Table 1).
During the modeling process, and for each individual participant, the shapes of the observed RT distributions for both correct and error responses are statistically compared against the shapes of RT distributions that are predicted to occur given a set of DM parameter values. Multiple parameter sets are successively tested until optimal fit between the actual and predicted distributions is obtained (Voss et al., Reference Voss, Rothermund and Voss2004). With as few as 20 trials per condition, Fast-dm parameter estimates are consistent with those obtained from 2000-trial model simulations (Voss & Voss, Reference Voss and Voss2007). Within the current data set, Fast-dm was able to fit the data of all participants well, as indicated by nonsignificant p-values for all model fits (all p > .18).
Data analytic approach
The experiment generated a Rotation (2: small/large) × Block (3) × Diagnosis (2: ADHD/Control) mixed between- and within-subjects ANOVA. Dependent variables were MRT for correct trials, standard deviation of response time (SDRT) for correct trials, accuracy, drift rate, boundary separation, and nondecision time.
Results
Table 2 provides summary statistics for all conditions and dependent variables. Table 3 provides summary F, η 2 , p, and df values for all analyses.
Table 2. Mean (SD) for Response Times, SDRT, Accuracy, v, a, and Ter by Block, Rotation, and Diagnostic Group
Table 3. Diagnostic Group and Task Condition Effects for MRT, SDRT, Accuracy, v, a, and Ter
Task Validation
Using standard indices of performance, there was a main effect of Rotation in which children were faster, less variable, and made fewer errors on small (vs. large) rotations (all p < .001, all η 2 ≥ .094). There was also a main effect of Block, in which RTs became faster and SDRT became smaller with practice (both p < .001, both η 2 ≥ .030; see Figure 2). Finally, a Block × Rotation interaction was also found in which SDRT decreased over time on large, but not small rotations, F(2,608) = 3.67, η 2 = .012, p = .026.
Fig. 2. Plots of MRT, SDRT, accuracy, drift rate, boundary separation, and nondecision time (Ter) by group, block, and rotation. Black line = large rotations; Grey lines = small rotations; Dotted line = ADHD; Solid line = Controls.
These main effects of Rotation and Block were driven by changes in drift rate, boundary, and Ter (all p < .001, all η 2 > .097; see Figure 2). Compared to large rotations, small rotations had smaller boundary separations; drift rates and Ter were also faster. With practice, boundary separation became smaller; drift rate and Ter became faster. The Rotation × Block interaction was similarly influenced by drift rate, boundary separation, and Ter (all p ≤ .024, all η 2 ≥ .012). With practice, trials with small (vs. large) rotations showed greater increases in drift, greater decreases in boundary separation, but somewhat more shallow improvement in Ter.
ADHD Effects
With respect to diagnostic group differences, using standard indices of performance, there was a main effect of ADHD in which children with ADHD had more variable RTs and were less accurate than non-ADHD controls (both p ≤ .002, both η 2 ≥ .032). This was due to slower drift, F(1,304) = 9.13, η 2 = .029, p = .003, and faster Ter, F(1,304) = 4.83, η 2 = .016, p = .029.
An ADHD × Rotation interaction for both MRT and accuracy was also found (both p ≤ .041, both η 2 ≥ .014). This interaction was driven by changes in boundary separation, F(1,304) = 9.93, η 2 = .032, p = .002 (see Figure 3). Among non-ADHD controls, boundary separation was wider for large versus small rotations, F(1,98) = 38.96, η 2 = .28, p < .001. Although the same effect was seen for children with ADHD, F(1,206) = 9.30, η 2 = .043, p = .003, it was not as pronounced and children with ADHD generally maintained larger boundaries. This inflexibility in the strategic adjustment of speed/accuracy trade-off has also been observed in previous studies (Mulder et al., Reference Mulder, Bos, Weusten, van Belle, van Dijk, Simen and Durston2010; Weigard & Huang-Pollock, Reference Weigard and Huang-Pollock2014). The ADHD × Rotation effect was not significant for either drift or Ter (both p ≥ .49, both η 2 ≤ .002). The lack of an ADHD × Rotation effect for Ter suggests that performance differences were not due to difficulties in the rotational processes.
Fig. 3. Boundary separation by ADHD status and stimulus rotation. Dotted line = ADHD; Solid line = Controls.
Discussion
In the current study, we examined mental rotation among school-age children with and without ADHD. Traditionally, MRT to correct responses has served as the primary dependent variable of performance, and despite the fact that children with ADHD responded more variably and less accurately, no diagnostic-based differences in MRT were identified. Thus, the most basic interpretation would be that there is no evidence of a specific spatial processing/mental rotation atypicality in ADHD. However, this interpretive practice ignores the interdependency and distributional properties of RT and accuracy in such a way that renders them potentially misleading indicators of cognitive performance.
Applying the DM to the data, we found that greater variance and higher error rates observed in children with ADHD were primarily due to slower drift rates, a finding that has been extensively documented across a range of tasks and studies (Huang-Pollock et al., Reference Huang-Pollock, Karalunas, Tam and Moore2012; Karalunas & Huang-Pollock, Reference Karalunas and Huang-Pollock2013; Karalunas et al., Reference Karalunas, Huang-Pollock and Nigg2012; Merkt et al., Reference Merkt, Singmann, Bodenburg, Goossens-Merkt, Kappes, Wendt and Gawrilow2013; Metin et al., Reference Metin, Roeyers, Wiersema, van der Meere, Thompson and Sonuga-Barke2013; Mulder et al., Reference Mulder, Bos, Weusten, van Belle, van Dijk, Simen and Durston2010; Salum, Sergeant, et al., Reference Salum, Sergeant, Sonuga-Barke, Vandekerckhove, Gadelha, Pan and Rohde2014; Salum, Sonuga-Barke, et al., Reference Salum, Sonuga-Barke, Sergeant, Vandekerckhove, Gadelha, Moriyama and Rohde2014; Weigard & Huang-Pollock, Reference Weigard and Huang-Pollock2014, Reference Weigard and Huang-Pollock2017). Drift rate is conceptualized as a “signal” to “noise” ratio (i.e., accumulation of task-relevant vs. irrelevant information) in the neural systems involved in decision-making (Heekeren, Marrett, & Ungerleider, Reference Heekeren, Marrett and Ungerleider2008; Ratcliff, Philiastides, & Sajda, Reference Ratcliff, Philiastides and Sajda2009). As signal strength decreases (or as noise increases), drift rate slows.
Group effects were also partially due to faster Ter. Ter represents all parts of the RT that are not involved in the decision-making process. In the context of a mental rotation task, this would include the time spent encoding the stimulus, rotating the stimulus, and preparing/executing a motor response. Given evidence for fine motor control deficits (Kaiser, Schoemaker, Albaret, & Geuze, Reference Kaiser, Schoemaker, Albaret and Geuze2015; Rommelse et al., Reference Rommelse, Altink, Fliers, Martin, Buschgens, Hartman and Oosterlaan2009), and lack of evidence for visual stimulus encoding exceptionalities, a true temporal advantage in Ter for youth with ADHD seems unlikely. Instead, the shorter Ter might best be understood as the result of prematurely discontinuing the mental rotation of images. Such an incompletely rotated image would be expected to compromise the quality of the evidence on which decisions are based and would reduce the signal strength of the decisional process. Thus, the incomplete rotation of images would also contribute to slow drift rate in ADHD, over and above the task non-specific slower drift rate that is already expected. It bears mentioning that faster Ter in children with ADHD is not uncommon (Karalunas, Geurts, Konrad, Bender, & Nigg, Reference Karalunas, Geurts, Konrad, Bender and Nigg2014) and could also potentially be an artifact of a “trade-off” effect between correlated DM parameters. That is, slow drift rates in ADHD may induce faster Ter estimates because they impact RT in opposing directions. Nevertheless, the absence of a Rotation × ADHD interaction for Ter suggests the absence of a specific ADHD-related deficit in mental rotation abilities.
Interestingly, there was a significant Rotation × ADHD interaction in which increases in boundary separation between small and large rotations were less pronounced in children with ADHD. Similar inflexibility has also been noted in a contextual cueing paradigm (Weigard & Huang-Pollock, Reference Weigard and Huang-Pollock2014), as well as in a paradigm in which speed/accuracy trade-off was manipulated (Mulder et al., Reference Mulder, Bos, Weusten, van Belle, van Dijk, Simen and Durston2010). Boundary inflexibility has been linked to hypoactivation in the striatum (Forstmann et al., Reference Forstmann, Dutilh, Brown, Neumann, Von Cramon, Ridderinkhof and Wagenmakers2008; Ivanoff, Branning, & Marois, Reference Ivanoff, Branning and Marois2008), which is among the most highly implicated brain regions in ADHD (Cubillo, Halari, Smith, Taylor, & Rubia, Reference Cubillo, Halari, Smith, Taylor and Rubia2012; Sonuga-Barke, Cortese, Fairchild, & Stringaris, Reference Sonuga-Barke, Cortese, Fairchild and Stringaris2016). Here, and despite trial-by-trial feedback, children with ADHD continued to make more errors than non-ADHD controls, suggesting that they had difficulty using the corrective feedback and experience to strategically modify their behavior to improve performance. This in turn likely led to a general and somewhat less efficient strategy of indiscriminately increasing boundary separation or caution in an effort to reduce error rates.
As would be expected with time on task, across both diagnostic groups, there were main effects of Block on drift rate, boundary separation, and Ter, in which drift rate and Ter became faster and boundary separation narrowed over time. Previous research examining practice effects on speeded RT tasks have found that practice-related reductions in RT are due to a combination of changes to all three primary parameters (Dutilh, Vandekerckhove, Tuerlinckx, & Wagenmakers, Reference Dutilh, Vandekerckhove, Tuerlinckx and Wagenmakers2009). When stimuli are repeated over time (as in the case of the current study), drift rate and Ter become faster (Dutilh, Krypotos, & Wagenmakers, Reference Dutilh, Krypotos and Wagenmakers2011) and participants who narrow their boundaries during the course of an experiment (i.e., requiring less evidence to make decisions) also show greater reductions in RT (Dutilh et al., Reference Dutilh, Vandekerckhove, Tuerlinckx and Wagenmakers2009). Finally, a significant Block × Rotation interaction was also found, in which there were greater increases in drift rate, greater decreases in boundary separation, and smaller decreases in nondecision time for small versus large rotations, which may reflect the automatization of the rotation process. Prior studies using mental rotation tasks attribute the disappearance of angle effects on RT with extended practice to an apparent switch from the slow/algorithmic rotation strategy to a more rapid one based on memory retrieval (Kail, Reference Kail1986; Provost et al., Reference Provost, Johnson, Karayanidis, Brown and Heathcote2013).
Decades of research have documented the presence of performance deficits among children with ADHD on a wide range of neuropsychological tasks including, but not limited to, tasks of visuospatial cognition, timing, WM, inhibitory control, and sustained attention (Castellanos & Tannock, Reference Castellanos and Tannock2002; Frazier, Demaree, & Youngstrom, Reference Frazier, Demaree and Youngstrom2004; Martinussen et al., Reference Martinussen, Hayden, Hogg-Johnson and Tannock2005; Schoechlin & Engel, Reference Schoechlin and Engel2005; Willcutt et al., Reference Willcutt, Doyle, Nigg, Faraone and Pennington2005). The DM results reported here provide clearer evidence of intact rotational abilities in ADHD than is currently available in the literature, and our work contributes to mounting evidence that slow drift rate provides a more parsimonious explanation for cognitive weaknesses in ADHD across domains (Huang-Pollock et al., Reference Huang-Pollock, Karalunas, Tam and Moore2012; Karalunas et al., Reference Karalunas, Geurts, Konrad, Bender and Nigg2014; Merkt et al., Reference Merkt, Singmann, Bodenburg, Goossens-Merkt, Kappes, Wendt and Gawrilow2013; Metin et al., Reference Metin, Roeyers, Wiersema, van der Meere, Thompson and Sonuga-Barke2013; Salum, Sonuga-Barke, et al., Reference Salum, Sonuga-Barke, Sergeant, Vandekerckhove, Gadelha, Moriyama and Rohde2014; Shapiro & Huang-Pollock, Reference Shapiro and Huang-Pollock2019; Weigard & Huang-Pollock, Reference Weigard and Huang-Pollock2017).
These empirical advances also bear important implications for process-oriented clinical assessment, in which the separate contributions of different psychological processes to task completion are quantified (e.g., seminal work by Larry Jacoby in the field of memory; Jacoby, Reference Jacoby1991). As two descriptors of a single response, RT and accuracy are produced simultaneously and are non-independent. However, the standard in the field is to select one or the other for analysis, even if they yield important interpretive differences. For example, in some situations, an individual may intentionally sacrifice accuracy for speed. In others, the mean RT for an individual or group may not be significantly slower than another, as was the case in the present study, but the shape of the RT distribution could clearly show a pattern of frequent long RTs that is characteristic of children with ADHD (Antonini, Narad, Langberg, & Epstein, Reference Antonini, Narad, Langberg and Epstein2013). So, separately considering RT and accuracy leads to at best an incomplete understanding of performance, and at worst, erroneous interpretations of data.
Although clinicians often “eyeball,” or informally take differences in speed/accuracy/skew into consideration during interpretation, the DM provides an empirically supported method of integrating all of these descriptors into a single set of psychologically meaningful metrics. It has also shown promise in discriminating clinical populations beyond ADHD, including anxiety (White, Skokin, Carlos, & Weaver, Reference White, Skokin, Carlos and Weaver2016), depression (Pe, Vandekerckhove, & Kuppens, Reference Pe, Vandekerckhove and Kuppens2013), and schizophrenia (Moustafa et al., Reference Moustafa, Kéri, Somlai, Balsdon, Frydecka, Misiak and White2015). Currently, the lack of nationally representative normative data for DM parameters on commonly administered tests prevents clinicians in the field from personally adopting such an approach (Galloway-Long, Shapiro, & Huang-Pollock, Reference Galloway-Long, Shapiro and Huang-Pollock2016), but establishing these norms while developing future commercial tasks holds promise for enhancing clinical utility.
Among the many strengths of the current study was the use of a visuospatial task that does not use traditional alphanumeric stimuli, which could have been a confound given the well-documented academic difficulties among children with ADHD, including in this sample (Arnold, Hodgkins, Kahle, Madhoo, & Kewley, Reference Arnold, Hodgkins, Kahle, Madhoo and Kewley2020; Frazier, Youngstrom, Glutting, & Watkins, Reference Frazier, Youngstrom, Glutting and Watkins2007). There are of course limitations as well. First, although the mental rotation task is among the most widely accepted tests of visuospatial cognition and has been previously evaluated with evidence accumulation models (Larsen, Reference Larsen2014; Provost & Heathcote, Reference Provost and Heathcote2015), it does not capture all elements of spatial cognition. Future studies are encouraged to evaluate the generalizability of our findings by applying the DM to other spatial processes such as perspective taking, spatial memory, or other forms of mental transformation (e.g., 3-dimensional rotation, mental folding, brittle transformations) (Bayliss, Jarrold, Baddeley, Gunn, & Leigh, Reference Bayliss, Jarrold, Baddeley, Gunn and Leigh2005; Kozhevnikov & Hegarty, Reference Kozhevnikov and Hegarty2001; Moreau, Reference Moreau2013; Resnick & Shipley, Reference Resnick and Shipley2013; Rump & McNamara, Reference Rump and McNamara2013).
Second, despite its advantages over traditional statistical methods, the DM is optimized for binary decision tasks with a mean RT of approximately 1500 ms. However, this is unlikely to have impacted the reliability of our results, as the model’s use has been validated in tasks with average RTs of over 7000 ms (Lerche & Voss, Reference Lerche and Voss2017). Third, there were too few trials to calculate DM parameters on all angular rotations, leading to the choice to use two broader classifications of rotation angles. It is likely that some of the interactions with rotation may have been different had greater trial numbers been included. And finally, it bears mentioning that a common argument within the diffusion modeling literature is that because trial condition cannot be anticipated, and stimulus properties that inform processing adjustments are not identified until after evidence accumulation has begun, only the model parameter representing stimulus difficulty (i.e., drift rate) can change across manipulations on a trial-by-trial basis (Ratcliff, Reference Ratcliff1978). Thus, our findings for the influence of rotation on boundary separation or Ter might be questioned. However, in this particular task, stimuli are actively rotated before the onset of the accumulation stage, such that knowledge of rotation angle could theoretically be used to determine the appropriate level of caution for each trial (Larsen, Reference Larsen2014). Likewise, since a significant portion of Ter is rotation time, which necessarily varies by angular discrepancy, trial-by-trial differences are to be expected.
Conclusion
Performance on speeded reaction time tasks among children with ADHD is often marked by slow, variable, and inaccurate responding. Despite evidence for spatial difficulties in ADHD, and the importance of spatial cognition in general, this remains an understudied domain. In a mental rotation task, we found no evidence of a specific deficit in visuospatial cognition. However, there was evidence of ADHD-related slower drift rates, faster nondecision times, as well as ADHD-related difficulty adjusting response thresholds with changes in rotation.
ACKNOWLEDGEMENTS
This work was funded by the National Institute of Mental Health (grant number R01 MH084947). The authors thank the parents, teachers, and children who participated, and research assistants who helped conduct the study.
CONFLICTS OF INTEREST
There are no conflicts of interest declared.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S1355617720001198
