Abstract
Background and aims
The shift from goal-directed to habitual control is a well-documented phenomenon in addiction research and is considered a critical factor in the development and maintenance of addictive behaviors. Whether Internet Gaming Disorder (IGD) is also associated with such a shift is not yet clear. The current study investigated the imbalance between goal-directed and habitual control in individuals with IGD.
Methods
Goal-directed and habitual control, as informed by model-based (MB) and model-free (MF) learning, were assessed with a two-step sequential decision-making task during functional magnetic resonance imaging (fMRI) in 33 young adults with IGD and 32 healthy controls (HCs). Self-report data regarding addictive symptoms, game craving, and impulsivity were also collected.
Results
Individuals with IGD relied more heavily on habitual control to guide subsequent choices compared to HCs. According to a hybrid reinforcement learning model, individuals with IGD also exhibited a reduced MB weight related to HCs, which was correlated with more severe addictive symptoms. fMRI results revealed that individuals with IGD showed increased MF reward prediction error (RPEMF) signals in the right triangular part of the inferior frontal gyrus (IFG). No significant group differences were found in the contrast of RPEMB maps.
Discussion and conclusions
Our study provides both behavioral and neural evidence highlighting an imbalance between goal-directed and habitual control, favoring habitual control in individuals with IGD. This imbalance is associated with the severity of addictive symptoms, suggesting an indication of habit inclination in IGD could potentially contribute to the development or maintenance of the addiction.
Introduction
Internet Gaming Disorder (IGD) has been recognized as a type of behavioral addiction in both the 5th edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) and the 11th of the International Classification of Diseases (ICD-11). It is characterized by excessive gaming and impaired control over gaming (APA, 2013; WHO, 2018). The worldwide prevalence of IGD is estimated to be 3.05%, with a confidence interval (CI) of [2.38, 3.91] (adjusted to 1.96%, CI [0.19, 17.12], in studies employing rigorous sampling methods), making it a significant public health concern (Stevens, Dorstyn, Delfabbro, & King, 2021). IGD can have severe impacts on individuals' mental health and well-being. For example, excessive gaming has been linked to attention deficits, increased impulsivity, and reduced social competence (Gentile, Choo, Liau, Sim, Li, Fung, Khoo, 2011). In severe cases, individuals may neglect personal responsibilities such as work or education, and experience disrupted sleep patterns, leading to a decline in overall quality of life (Weaver, Gradisar, Dohnt, Lovato, & Douglas, 2010). Moreover, IGD is highly comorbid with mental disorders such as depression (Mihara & Higuchi, 2017), anxiety (Gentile et al., 2011; Van Rooij et al., 2014), as well as other types of behavioral and substance addictions (Walther, Morgenstern, & Hanewinkel, 2012). These negative impacts of IGD can be long-lasting. For instance, longitudinal studies found that heightened depression, anxiety, and impulsivity, as well as matter volume alterations in IGD can persist for nearly two years (Lee, Namkoong, Lee, & Jung, 2021). Similarly, a longitudinal study found no improvement in depression and anxiety symptoms, or impulsiveness despite a reduction in IGD severity after 6 months of treatment (Lim et al., 2016).
One key characteristic of IGD, and other types of addictions, is compulsivity, which refers to perseveration in the face of adverse consequences (Koob & Le Moal, 2005; Liu et al., 2024). For instance, a high school student struggling with IGD may repeatedly engage in gaming, despite being aware that it is seriously affecting his academic performance. A prominent theoretical framework attributes the emergence of compulsivity to dysfunctional habit formation. Initially, substance use or addictive behaviors are voluntary but, through repetition, transition into habits and eventually become compulsive (Everitt & Robbins, 2005, 2016). According to the Interaction of Person-Affect-Cognition-Execution (I-PACE) model (Brand et al., 2019), generalized deficits in inhibitory control may already be present in the early stages of addictive behavior, while stimulus-specific deficits may develop in later stages, making an individual behave habitually or seemingly automatically when encountering addiction-related stimuli. This transition is hypothesized to involve two parallel systems of instrumental control: the goal-directed and habitual systems (Daw, Niv, & Dayan, 2005). The goal-directed system infers which actions will yield the best outcomes based on response-outcome associations with current goals. The habitual system, on the other hand, relies on stimulus-response associations to produce responses that have led to favorable outcomes in similar past situations (Daw et al., 2005; Decker, Otto, Daw, & Hartley, 2016). Over time, disrupted balance between these systems may lead to compulsive behaviors overriding goal-oriented decision-making (Everitt & Robbins, 2005, 2016).
Computational models have proposed that goal-directed and habitual systems arise from two different reinforcement learning mechanisms, known as model-based (MB) and model-free (MF) learning (Daw, Gershman, Seymour, Dayan, & Dolan, 2011). The MB (goal-directed) system learns the value of actions based on their historical rewards or punishments, considering the transition dynamics of the environment, while the MF (habitual) system learns the value of actions based solely on historical rewards or punishments (Voon et al., 2015). While both systems rely on action value representations, the habitual system encodes values in a much more simplified manner. Overreliance on the habitual system may underlie the compulsive behaviors observed in disorders like addiction and obsessive-compulsive disorder (Koob & Volkow, 2016).
The two-step task is a sequential decision-making task that is commonly used to study goal-directed and habitual control in laboratory setting. In this task, participants choose between options at two stages and receive rewards based on stochastic randomization. Goal-directed and habitual learning can be distinguished based on the pattern of staying or switching choices following rewards (Daw et al., 2011). Clinical studies using the two-step task have found imbalanced goal-direct and habitual control (lower MB weights) in subjects with various substance use disorders and behavioral addictions, including individuals with detoxicated alcohol use disorder (Sebold et al., 2014), individuals who engage in binge drinking (Doñamayor, Strelchuk, Baek, Banca, & Voon, 2018), those experiencing gambling disorder (Wagner, Mathar, & Peters, 2022; Wyckmans et al., 2019, 2022), individuals with abstinent methamphetamine dependence (Voon et al., 2015), and people with binge eating disorders (Voon et al., 2015).
A recent study using two-step task found no significant goal-directed/habitual imbalance in IGD behaviorally, compared to controls, but reported greater MB reward prediction error (RPE) signals in the bilateral insula for IGD, indicating aberrant goal-directed processing in IGD (Kwon, Choi, Park, Ahn, & Jung, 2024). Additionally, a large-scale online study (n = 1,961) with a community sample using the two-step task found that deficits in goal-directed control were strongly associated with a symptom dimension encompassing compulsive behaviors and intrusive thoughts, including symptoms of eating disorders, impulsivity, obsessive-compulsive disorder, and alcohol-dependent (Gillan, Kosinski, Whelan, Phelps, & Daw, 2016).
At the neural level, the valuation processes of goal-directed and habitual systems can be measured through RPE signals (Daw et al., 2011). Previous functional magnetic resonance imaging (fMRI) studies have shown that the goal-directed (MB) system is primarily associated with the medial prefrontal cortex (MPFC), OFC, caudate, and nucleus accumbens (NAc), while the habitual (MF) system is primarily associated with the putamen and NAc (Balleine & O'Doherty, 2010; Daw et al., 2011; Yin, Knowlton, & Balleine, 2004). Our previous meta-analysis of fMRI decision-making tasks revealed bilateral striatum activation in both goal-directed (MB) and habitual (MF) learning, with the dorsal part of the striatum (i.e. the globus pallidus and caudate head) showing greater concordance for habitual learning, while the goal-directed learning was found to activate executive control-related regions, such as the prefrontal cortex and cingulate cortex (Huang, Yaple, & Yu, 2020). In sum, neuroimaging studies have collectively highlighted the significance of the fronto-striatal circuit in processing the RPE of goal-directed and habitual learning.
Addiction is often viewed as the transition from initially goal-directed drug-seeking to becoming a more habitual and compulsive pattern of behaviors (Lüscher, Robbins, & Everitt, 2020). This transition is thought to involve dynamic shifts in neural control over approach behaviors, especially a shift from the ventral striatum to the dorsal striatum, a process known as the ventral-to-dorsal striatal transition. This shift has been observed in various substance addictions (Everitt & Robbins, 2005; Lüscher et al., 2020), and has also been observed in IGD during the processing of game-related stimuli (Liu et al., 2017). Computational models have been developed to better understand this transition. These models propose that individuals balance goal-directed and habitual control by comparing how reliable or uncertain the predictions of each system are (Daw et al., 2005; Lee, Shimojo, & O'Doherty, 2014). Based on this proposal, the imbalance between goal-directed and habitual control is suggested to reflect a compromised uncertainty calculation, leading to an overreliance on habitual control (Lee et al., 2014). Previous studies have identified that the inferior frontal gyrus (IFG) may act as an “arbitrator” between the two systems. Specifically, the IFG assesses the reliability or uncertainty of goal-directed and habitual strategies and allocates control to the more reliable one (Kim et al., 2024; Lee et al., 2014). When this arbitration system is compromised, it may contribute to the imbalance between goal-directed and habitual behaviors seen in addiction. This perspective highlights the potential importance of the “arbitrator” in the imbalance of goal-directed and habitual behaviors.
However, the role of the imbalance between goal-directed and habitual control in addiction remains a subject of ongoing debate. Hogarth (2020) posits that habit or compulsion do not serve as the central mechanisms of addiction; instead, he argues that human addiction is predominantly influenced by excessive goal-directed drug choice, particularly in the context of negative affect. He contends that compulsivity––defined as the persistence of drug-seeking behavior despite punishment, as proposed by the habitual theory––can be better explained by the value ascribed to the drug outweighing the punisher, rather than insensitivity to drug value (Everitt & Robbins, 2016). Similarly, Vandaele and Ahmed (2020) argue that when considering the role of habitual control in addiction, we need to consider more complex frameworks, taking into account continuous interactions between goal-directed and habitual systems, and alternative decision-making models more representative of real-world conditions. They highlight specific conditions in animal studies that may influence the occurrence of habitual control. For example, reinforcement schedules, training settings, task contingencies, and extinction conditions influence habitual control by reducing response-reinforcement contingency, altering training variables, affecting reward predictability, and shifting behavior during reacquisition tests (Vandaele & Ahmed, 2020). The learning theory of addiction suggests that addiction could arise from “a maladaptive stimulus-response (S-R) habit” (Everitt & Robbins, 2005, 2016). However, it also proposes that a gain of function in the OFC–striatal pathway could lead to an overestimation of the value of the drug experience relative to punishment, biasing instrumental behavior toward drug consumption. These findings might represent a form of compulsive responding with a goal-directed basis (Lüscher et al., 2020). Some researchers suggest that drug-related behaviors may not be exclusively habitual or goal-directed (Schreiner, Renteria, & Gremel, 2020). Therefore, rather than categorizing drug-seeking behavior as purely goal-directed or habitual, it may be more meaningful to view goal-directed control as existing on a gradient and assess the balance between the two systems (Vandaele & Ahmed, 2020). Given the inconclusive body of evidence, further research is needed to clarify the role of habits in addictions, such as IGD.
While traditional tasks assessing individual sensitivity to outcome devaluation typically provide binary (yes-or-no) answers, the two-step task is more suitable for measuring the relative strength of both systems (Feher da Silva & Hare, 2020). Studies using the two-step task consistently suggest that goal-directed and habitual control are engaged in parallel and with individuals relying on both systems to make decisions, further support the idea that habitual and goal-directed processes are intermingled under a hierarchical decision-making structure (Otto, Gershman, Markman, & Daw, 2013). In the current study, we employed a two-step task and fMRI to examine the potential imbalance between goal-directed and habitual control and related RPE signal changes in individuals with IGD. We hypothesize that individuals with IGD may exhibit an imbalance between goal-directed and habitual control systems, favoring the latter. We also propose that the RPE neural signals may change in one or both systems between individuals with IGD and controls. The neural activity may particularly occur in the fronto-striatal circuit.
Methods
Participants
Seventy-four young adults were recruited from Luzhou, China, between September 2022 and June 2024 through online and poster advertisements. Participants were diagnosed for IGD according to DSM-5 criteria following a clinical interview conducted by trained clinical psychiatrists (KZL, KG, YYZ). These psychiatrists are clinical doctors with a minimum of two years of experience in diagnosing and treating mental disorders including IGD. A participant was diagnosed as IGD if he/she experienced five or more of the symptoms listed in the DSM-5 criteria within a year; otherwise, he/she would be deemed as a healthy control (HC). The Mini-International Neuropsychiatric Interview (MINI) was used to ensure that all participants had no history of any major psychiatric disorder as listed in DSM-IV Axis I (including schizophrenia, substance use disorder, mood disorders, anxiety disorders, etc.) (Sheehan et al., 1998). Although excluding individuals with IGD comorbid with other mental disorders may somewhat reduce the ecological validity of this study––given that IGD is frequently accompanied by conditions like depression and anxiety––we chose to focus on participants diagnosed exclusively with IGD to avoid confounding effects from other mental disorders. Initially, 39 individuals with IGD and 35 HCs were recruited. Nine of them were excluded after quality control (>20% missing trials n = 2; poor computational model fitting performance, as indicated by neff/iterations <0.01 or Rhat >1.1 n = 1; excessive head motion during fMRI scan, i.e. translational movement >3 mm or rotation >3°, n = 6). Data from the remaining 65 participants, including 33 IGD and 32 HCs were analyzed.
The two-step task
We adapted a two-step task from Daw et al. (2011) to dissociate goal-directed and habitual control strategies. The task was modified to enhance participants' engagement and improve their understanding of the narrative (Fig. 1A). Participants were asked to play a game called “finding the egg”. First, they choose between two cars, one blue and one green (first-stage choice). The blue car had a 70% probability (the common transition) of arriving at the duck farm and a 30% probability (the rare transition) of arriving at the chicken farm, while the green car had the opposite probability. Participants were informed about the transition structure, but not the exact probabilities. In the second stage choice, participants chose between two ducks or two chickens on the farm. They were then rewarded with an egg or with nothing (a red cross) according to a slowly drifting probability (0.2–0.75). These shifting reward probabilities encouraged participants to constantly update the utilities of the second-stage stimuli and explore different choices throughout the task to maximize rewards. Participants were given 2 s to make each choice, followed by a 1 s of reward feedback, and a jittered (1–3 s) intertrial interval. Following 20 practice trials, participants completed the formal task in the MRI scanner. The full game consisted of 200 trials in four blocks separated by breaks, which lasted for about 30 min. Participants were compensated for their participation, and extra bonuses were awarded according to the total amount of earned points in the two-step task at the exchange rate of 10 points for 1 RMB.
(A) The two-step task. Each trial began with a fixation presented randomly for 1–3 s. Then, in the first-stage decision, two cars were presented for 2 s, during which subjects determined which car to take using one of two buttons on a response box (maximum RT = 2 s; the selected car was highlighted with a green box once a response was made and the stimuli remained on the screen for the rest of the 2 s). Following a first-stage decision (the car would drive to a particular farm), the second-stage stimuli (either 2 hens or 2 ducks) were present for 2 s and subjects were required to select a stimulus using one of two buttons (maximum RT = 2 s; the selected hen/duck was highlighted with a green box once a response was made and the stimuli remained on the screen for the rest of the 2 s). Finally, during the outcome stage, an egg (reward) or a red “X” (no reward) was presented for 1 s. Each car had common (70%) or rare (30%) transition probabilities to the corresponding farms. (B) Mean first-stage stay probabilities under different reward and transition types in both IGD and HC groups. The error bars represent the standard error. (C) Model fitting results indicated that IGDs showed lower model-based weight (ω) than HCs. *p < 0.05. (D) In the IGD group, the model-based weight (ω) was negatively correlated with Internet Gaming Disorder Scale (IGDS) scores (E) and daily gaming time (DGT)
Citation: Journal of Behavioral Addictions 2025; 10.1556/2006.2025.00037
Stay probability of the first-stage choices under each condition were calculated and compared between groups to reflect the extent to which the choice under different conditions was affected by the imbalance between goal-directed and habitual control. The overall stay probability was also calculated and used as an index of preservation. The total eggs earned was taken as an indicator of overall performance.
Model fitting of the behavioral data
To further explore the underlying mechanisms of choices within the model-based and model-free reinforcement learning frameworks, we fitted our data using a hybrid reinforcement-learning model similar to the original model proposed by Daw et al. (2011). This hybrid model considers the entire preceding history of choices and rewards and assumes that choices are determined by a weighted combination of goal-directed (MB) and habitual (MF) values. The model consists of seven free parameters, including (1) a weighting parameter ω that determines the relative contribution of the MB and MF system (with a higher value indicating a higher weight of MB); (2) learning rate α1 for stage 1 and (3) α2 for stage 2, which govern the degree to which action values were updated after an outcome was received; (4) inverse-temperature parameters β1 for stage 1 and (5) β2 for stage 2, which control the exploitation-exploration trade-off at each stage; (6) an eligibility trace parameter λ that governs the relative importance of the second-stage reinforcers, reward or second-stage values, in updating first-stage values; and (7) choice stickiness parameters π that capture perseveration on the choice irrespective of reward and transition conditions. The model was fitted using rstan (Carpenter et al., 2017) (see Supplementary Information for details).
Measures
Chinese version 9-item dichotomous Internet Gaming Disorder Scale (IGDS) (Lei et al., 2020; Lemmens, Valkenburg, & Gentile, 2015) was used to assess game addictive symptoms. The Internet Addiction Test (IAT) (Young, 1998) was adopted to capture information on broader Internet usage in participants. The Barratt Impulsiveness Scale (BIS-11) was used to assess impulsivity in participants (Patton, Stanford, & Barratt, 1995). The BIS-11 contains 30 items that measure three main domains - attentional impulsivity, motor impulsivity, and non-planning impulsivity. Scores of these three subscales and a total score were calculated. The Beck's Depression Inventory (BDI) (Beck, Steer, & Carbin, 1988), and the Beck's Anxiety Inventory (BAI) (Beck, Epstein, Brown, & Steer, 1988) were used to assess depression and anxiety symptoms separately. All the scales showed acceptable internal consistency in this study (for IGDS, IAT, BIS-11, BIS-attentional, BIS-motor, BIS-non-planning, BDI, and BAI, Cronbach's α = 0.848, 0.960, 0.795, 0.504, 0.586, 0.722, 0.897, 0.909, separately). We also collected data on daily gaming time over the past year and the degree of craving for games (rated once on a 0-10 scale) using a custom-designed questionnaire. Participants report their retrospective craving experience as part of the clinical interview before the fMRI scanning.
fMRI data acquisition and preprocessing
The fMRI images were acquired using a Phillips Achieva 3T scanner with a 16-channel head-neck coil. Task-based fMRI data were collected in two sessions using echo-planar images sequence: slice thickness = 3 mm, gap = 1 mm, 38 axial slices in interleave order, repetition time = 2,000 ms, echo time = 30 ms, flip angle = 90°, matrix = 64 × 64, field of view = 240 mm, inplane resolution = 3.75 × 3.75 mm. A T1-weighted high-resolution anatomical scan was also obtained for registration: repetition time = 7.6 ms, echo time = 3.8 ms, field of view = 256 mm, matrix = 256 × 256, inplane resolution = 1 × 1 mm, slice thickness = 1 mm with no gap, 170 sagittal slices.
The preprocessing of fMRI images was conducted using SPM12 (http://www.fil.ion.ucl.ac.uk/spm). The following steps were performed: discard the first four volumes to achieve magnet-steady images; slice timing; realignment; manually reorient EPI and T1 images of each subject; co-registration; spatially normalization to the Montreal Neurological Institute space using the mapping parameters derived from segmentation of T1 images and interpolated to 3 mm × 3 mm × 3 mm voxels using DARTEL toolbox; and spatial smooth using a 6 mm Gaussian kernel. Participants showing excessive head movement (translational movement >3 mm or rotation >3°) during fMRI scanning were excluded.
Statistical analyses
Logistic mixed-effects regression
A significant main effect of reward indicates a habitual (MF) strategy, whereas a significant interaction of reward and transition type indicates a goal-directed (MB) strategy.
Statistical analysis of model fitting parameters
Firstly, model comparisons were carried out to assess if the hybrid reinforcement learning model would outperform alternative models including a null model where choices were assumed to be made by chance, a model with temporal-difference (TD) algorithm SARSA(λ) alone, a model-based RL algorithm alone, as well as two hybrid models with λ restricted to 0 or 1. Secondly, the group differences in the fitted parameters were tested using the Exact Mann-Whitney U-test. Individual estimates of ω were used as indicators for the balance of goal-directed and habitual control.
Correlations between model parameters and clinical measurements
To investigate the relationships between measures of goal-directed/habitual control and gaming behaviors, Spearman's rho was calculated between the fitted parameters of the hybrid model and the scores in clinical measures, including daily gaming time, craving, IGDS, BIS-11, and IAT. The threshold was set at p < 0.05.
Whole brain fMRI analysis
The first-level analysis was set up according to Daw et al. (2011). We aimed to investigate generalized RPE signals in the brain by including two RPE values, named model-free RPE (RPEMF) and model-based RPE (RPEMB), as parametric regressors modulating impulse events at the onset of the second-stage and the reward receipt. The RPEMF corresponds to the generalized model-free RPE computed under the assumption of purely MF control (ω = 0), while the RPEMB corresponds to the difference between MF and MB RPEs. As the RPEMB is zero at outcome presentation, we mean-centred RPEMB at the reward receipt stage to zero mean within-subject. Two additional nuisance regressors were also defined. One corresponds to the time of reward receipt to capture any difference in mean activity between the choice and outcome. Another nuisance was time-locked to the onset of the first-stage stimuli, modulated by two additional parametric regressors: P (a1, t |sA), a normalized measure of the first-stage action value, and its partial derivative with respect to ω (see Supplementary Information for details).
In the second-level analysis, separate one-sample t-tests of fMRI contrasts for RPEMF and RPEMB were performed in HC and IGD groups. Then the case-control differences in RPEMF and RPEMB maps were identified using two-sample t-tests, with age and sex included as covariates. A grey matter mask, derived from the SPM prior template (TPM.nii) by setting the threshold at > 0.2, was used to restrict the analysis within the grey matter area. To determine the significance of differences in regional activity, we carried out small volume correction (SVC) on a priori regions of interest (ROIs). The ROIs were related to the previously defined regions implicated in goal-directed and habitual systems, including the bilateral NAc, MPFC, OFC, caudate, putamen, NAc, and anterior cingulate cortex (ACC). To explore if the IFG attributor plays a role in the goal-directed/habitual imbalance, masks of bilateral IFG were also included in SVC. The mask of MPFC was confined to BA10 and BA32 (Haber & Knutson, 2010). All other ROIs were anatomically defined from the Automated Anatomical Labelling Atlas 3 (AAL3) (Rolls, Huang, Lin, Feng, & Joliot, 2020). Activations in other areas were reported if they survived at a voxel-level threshold of p < 0.001 and cluster-wise AlphaSim correction with a threshold of p < 0.05 (cluster size >11 voxels) (Song et al., 2011).
Ethics
After a detailed introduction of the study procedures, participants provided their written informed consent. The study was approved by the Institutional Review Board of the Southwest Medical University (Ethical approval number: KY2022081) and was conducted in accordance with the latest revision of the Declaration of Helsinki. All participants provided informed consent to participate in this study.
Results
Behavioral results
Table 1 shows demographic information, scale measurements, and performance in the two-step task for the two groups. The IGD and HC groups did not reveal significant differences in age, education years, and gender distribution (all p values >0.1). Individuals with IGD scored significantly higher than HCs on measures of BAI, BDI, daily gaming time, craving, IGDS, BIS-11 and its three subscales, and IAT (all p values <0.05). Regarding performance in the two-step task, there were no significant differences between IGDs and HCs in terms of the total first-stage stay probability, stay probabilities under the four reward and transition types (rewarded-common, rewarded-rare, unrewarded-common, and unrewarded-rare, Fig. 1B), as well as the total eggs earned in the game (all p values >0.1) (Table 1).
Demographic information, scores in scale measures, and performance in the two-step task for both IGD and HC groups
| Variables | IGD n = 33 | HC n = 32 | t/χ2 | p | Cohen's d |
| Demographic information | |||||
| Gender (F/M) | 6/27 | 2/30 | 2.14 | 0.143 | 0.18 |
| Age | 20.85 ± 2.06 | 21.38 ± 1.41 | −1.20 | 0.235 | −0.30 |
| Years of education | 14.88 ± 2.13 | 15.47 ± 1.39 | −1.32 | 0.193 | −0.33 |
| BAI | 8.94 ± 8.59 | 4.88 ± 5.28 | 2.29 | 0.025 | 0.57 |
| BDI | 11 ± 9.83 | 5.59 ± 5.34 | 2.74 | 0.008 | 0.68 |
| DGT | 4.39 ± 2.39 | 1.58 ± 1.7 | 5.46 | <0.001 | 1.36 |
| Craving | 5.64 ± 1.82 | 2.31 ± 2.18 | 6.69 | <0.001 | 1.66 |
| IGDS | 6.67 ± 1.71 | 1.94 ± 1.56 | 11.63 | <0.001 | 2.89 |
| BIS-attentional | 19.15 ± 2.98 | 17.19 ± 2.6 | 2.83 | 0.006 | 0.70 |
| BIS-motor | 23.45 ± 3.97 | 20.22 ± 3.21 | 3.61 | <0.001 | 0.89 |
| BIS-NP | 27.7 ± 4.67 | 23.75 ± 4.23 | 3.57 | <0.001 | 0.89 |
| BIS-11 | 70.3 ± 8.57 | 61.16 ± 7.8 | 4.49 | <0.001 | 1.12 |
| IAT | 62.97 ± 15.38 | 37.03 ± 9.23 | 8.21 | <0.001 | 2.04 |
| DSM-5 | 6.77 ± 1.18 | 1.34 ± 1.43 | 16.45 | <0.001 | 4.15 |
| Performance in the task | |||||
| Total stay probability | 0.83 ± 0.15 | 0.8 ± 0.14 | 0.84 | 0.406 | 0.21 |
| Stay probability-rewarded common | 0.87 ± 0.14 | 0.82 ± 0.13 | 1.40 | 0.166 | 0.35 |
| Stay probability-rewarded rare | 0.83 ± 0.17 | 0.79 ± 0.15 | 0.96 | 0.341 | 0.24 |
| Stay probability-unrewarded common | 0.8 ± 0.16 | 0.78 ± 0.15 | 0.57 | 0.574 | 0.14 |
| Stay probability-unrewarded rare | 0.79 ± 0.19 | 0.79 ± 0.16 | 0.12 | 0.902 | 0.03 |
| Total eggs earned | 91.42 ± 9.12 | 93.16 ± 10.79 | −0.70 | 0.487 | −0.17 |
Note: IGD: Internet Gaming Disorder; HC: healthy controls; BAI: Beck Anxiety Inventory; BDI: Beck Depression Inventory; DGT: daily gaming time of the past year (in hours); IGDS: Internet Gaming Disorder Scale; BIS-11: Barratt Impulsiveness Scale (11th version); BIS-NP: Non-planning subscale of BIS-11; BIS-Attentional: attentional subscale of BIS-11; BIS-Motor: motor subscale of BIS-11; IAT: Internet Addiction Test. IAT: Internet Addiction Test. Significant values were highlighted in bold fonts.
The mixed-effects logistic regression analysis revealed a significant main effect of reward in both groups (IGD: B = 0.54, SE = 0.13, p < 0.001; HC: B = 0.25, SE = 0.11, p = 0.030), indicating that both groups adopted habitual control strategies in the task. The reward-by-transition interaction effects, which indicates goal-directed behavioral control, were also significant in both groups (IGD: B = 0.52, SE = 0.24, p = 0.032; HC: B = 0.51, SE = 0.25, p = 0.043) (Table 2). Direct comparisons of the regression coefficients for each participant revealed a significantly higher reward effect in IGDs than in HCs (0.54 ± 0.19 vs. 0.24 ± 0.12, t = 7.49, p < 0.001), while both groups showed comparable coefficients in reward-transition interaction (0.51 ± 0.31 vs. 0.50 ± 0.53, t = 0.16, p = 0.876). Both groups showed significant random effects for the main effect of reward and the reward-by-transition interaction effect, demonstrating that the random factor (subject) had a significant influence on the models (Table S1).
Regression coefficients of the mixed-effects logistic regression
| B | SE | z value | Pr (>|z|) | |
| HC (n = 32) | ||||
| (Intercept) | 1.89 | 0.26 | 7.34 | <0.001 |
| Reward | 0.25 | 0.11 | 2.17 | 0.030 |
| Transition | 0.14 | 0.12 | 1.17 | 0.242 |
| Reward × transition | 0.51 | 0.25 | 2.03 | 0.043 |
| IGD (n = 33) | ||||
| (Intercept) | 2.09 | 0.25 | 8.35 | <0.001 |
| Reward | 0.54 | 0.13 | 4.23 | <0.001 |
| Transition | 0.23 | 0.18 | 1.31 | 0.190 |
| Reward × transition | 0.52 | 0.24 | 2.14 | 0.032 |
Note: Significant values were highlighted in bold fonts.
The model comparisons confirmed that the hybrid model was superior to the other five alternative models (Table S2). Therefore, the fitted parameters of the hybrid model were taken for further analysis. The model fitting further revealed a lower ω in the IGD group relative to the HC group (p = 0.02), suggesting reduced model-based learning in participants with IGD (Table 3). We further found that the ω values were negatively correlated with IGDS scores (Spearman's rho = −0.419, p = 0.015) and daily gaming time (rho = −0.419, p = 0.015) in the IGD group. The negative correlation between ω and IGDS scores remains significant in the merged sample (rho = −0.241, p = 0.053). However, the correlation between ω and IAT was not significant (rho = −0.055, p = 0.762) in the IGD group. Additionally, negative correlations were observed between the exploitation-exploration trade-off for the two choice options at the second stage (i.e., β2) and both game craving (rho = −0.371, p = 0.034), IAT total scores (rho = −0.392, p = 0.024), BIS-11 scores (rho = −0.390, p = 0.025), and BIS-attentional subscale scores (rho = −0.493, p = 0.004) in the IGD group. In the HC group, while the correlation between ω and IGDS scores was not significant (rho = 0.342, p = 0.056), ω was found to be positively and significantly correlated with IAT (rho = 0.515, p = 0.003) (Table 4).
Fitted model parameters and group differences
| α1 | α2 | β1 | β2 | λ | ω | π | |
| IGD | |||||||
| 25% percentile | 0.08 | 0.19 | 1.58 | 1.12 | 0.51 | 0.19 | 0.65 |
| Median | 0.30 | 0.34 | 3.45 | 1.75 | 0.56 | 0.28 | 1.03 |
| 75% percentile | 0.53 | 0.60 | 6.87 | 3.31 | 0.69 | 0.48 | 2.26 |
| HC | |||||||
| 25% percentile | 0.27 | 0.11 | 0.91 | 1.07 | 0.45 | 0.27 | 0.65 |
| Median | 0.44 | 0.21 | 1.88 | 1.95 | 0.51 | 0.38 | 1.12 |
| 75% percentile | 0.49 | 0.62 | 6.92 | 4.05 | 0.57 | 0.59 | 1.94 |
| Group differences test | |||||||
| U | 451 | 377 | 402 | 445 | 360 | 329 | 446 |
| Z | −0.606 | −1.624 | −1.280 | −0.688 | −1.858 | −2.284 | −0.674 |
| p | 0.545 | 0.104 | 0.201 | 0.491 | 0.063 | 0.022 | 0.500 |
Note: Significant group differences were highlighted in bold fonts. U: Mann-Whitney U.
Correlations between model parameters and scale measurements
| α1 | α2 | β1 | β2 | π | ω | λ | |
| IGD group | |||||||
| DGT | −0.302 | −0.100 | 0.165 | −0.141 | −0.11 | −0.419 | −0.083 |
| BAI | 0.109 | −0.141 | −0.069 | −0.191 | 0.104 | 0.036 | −0.242 |
| BDI | 0.229 | 0.065 | −0.155 | −0.202 | 0.153 | 0.007 | −0.237 |
| Craving | −0.18 | 0.093 | 0.010 | −0.371 | −0.196 | −0.142 | −0.200 |
| IGDS | −0.124 | −0.029 | 0.197 | −0.07 | −0.102 | −0.419 | −0.037 |
| IAT | −0.005 | 0.176 | −0.085 | −0.392 | 0.046 | −0.055 | −0.198 |
| BIS-11 | −0.043 | 0.076 | 0.069 | −0.390 | −0.112 | −0.016 | −0.16 |
| BIS-attentional | −0.152 | 0.012 | 0.021 | −0.493 | −0.293 | −0.143 | −0.283 |
| BIS-motor | 0.136 | 0.083 | −0.082 | −0.19 | −0.12 | 0.273 | −0.123 |
| BIS-NP | −0.128 | 0.037 | 0.219 | −0.218 | 0.03 | −0.3 | −0.067 |
| HC group | |||||||
| DGT | 0.099 | 0.1 | −0.109 | 0.176 | 0.378 | 0.344 | 0.084 |
| BAI | −0.094 | −0.024 | −0.068 | 0.018 | 0.053 | 0.073 | 0.054 |
| BDI | −0.01 | 0.194 | −0.057 | −0.049 | 0.315 | 0.169 | −0.114 |
| Craving | −0.033 | −0.147 | 0.214 | 0.394 | 0.279 | 0.338 | 0.221 |
| IGDS | −0.091 | 0.098 | −0.163 | 0.174 | 0.268 | 0.342 | 0.038 |
| IAT | 0.093 | 0.218 | −0.263 | 0.046 | 0.258 | 0.515 | 0.092 |
| BIS-11 | −0.061 | 0.205 | −0.266 | 0.143 | 0.074 | 0.344 | −0.185 |
| BIS-attentional | −0.379 | −0.145 | 0.189 | 0.556 | 0.117 | 0.05 | −0.098 |
| BIS-motor | −0.104 | 0.128 | −0.293 | 0.058 | 0.028 | 0.094 | −0.032 |
| BIS-NP | −0.052 | 0.242 | −0.27 | −0.084 | 0.012 | 0.410 | −0.264 |
| All participants merged | |||||||
| DGT | −0.108 | 0.153 | 0.112 | −0.038 | 0.131 | −0.15 | 0.149 |
| BAI | 0.01 | −0.051 | 0.012 | −0.089 | 0.074 | −0.041 | −0.029 |
| BDI | 0.087 | 0.125 | 0.012 | −0.128 | 0.225 | −0.053 | −0.113 |
| Craving | −0.086 | 0.095 | 0.228 | 0.009 | 0.047 | −0.084 | 0.212 |
| IGDS | −0.105 | 0.15 | 0.15 | −0.044 | 0.066 | −0.241 | 0.201 |
| IAT | −0.026 | 0.248 | 0.01 | −0.176 | 0.107 | −0.076 | 0.111 |
| BIS-11 | −0.096 | 0.175 | 0.001 | −0.125 | −0.015 | −0.016 | −0.054 |
| BIS-attentional | −0.204 | −0.027 | 0.135 | 0.007 | −0.087 | −0.163 | −0.074 |
| BIS-motor | 0.034 | 0.124 | −0.109 | −0.089 | −0.029 | 0.074 | 0.027 |
| BIS-NP | −0.13 | 0.172 | 0.044 | −0.146 | 0.024 | −0.067 | −0.08 |
Note: BAI: Beck Anxiety Inventory; BDI: Beck Depression Inventory; DGT: daily gaming time of the past year (in hours); IGDS: Internet Gaming Disorder Scale; BIS-11: Barratt Impulsiveness Scale (11th version); BIS-NP: Non-planning subscale of BIS-11; BIS-Attentional: attentional subscale of BIS-11; BIS-Motor: motor subscale of BIS-11; IAT: Internet Addiction Test. All participants merged: merged sample with participants of both groups (n = 65); Significant correlations (p < 0.05) were highlighted in bold fonts.
fMRI results
The fMRI results are shown in Table 5, Figs 2 and 3. We first tested the RPE maps in the two groups separately. In the HC group, the fMRI signals for RPEMF showed activations in all our ROIs, including the bilateral NAc, putamen, caudate, ACC, MPFC, OFC, and IFG (Fig. 2). Several other regions were found to be significantly associated with RPEMF signals as well, most notably the post cingulate cortex extending to precuneus and angular gyrus, and Cerebellum Posterior Lobe. The fMRI signals for RPEΔMB were found in the right MPFC (BA10). The IGD group showed similar maps of RPEMF and RPEMB to HCs (Fig. 2).
RPEMF and RPEMB in IGD and HC groups, and group differences
| Brain regions | Voxels | T | SVC* | MNI coordinates | ||
| RPEMF: HC | ||||||
| ACC-L | 273 | 7.35 | <0.001 | 0 | 51 | 9 |
| ACC-R | 178 | 6.76 | <0.001 | 3 | 54 | 9 |
| IFG-L | 261 | 6.09 | <0.001 | −42 | 36 | −12 |
| IFG-R | 94 | 5.01 | 0.002 | 39 | 36 | −12 |
| Caudate-L | 25 | 6.84 | <0.001 | −12 | 21 | −9 |
| Caudate-R | 78 | 7.17 | <0.001 | 12 | 21 | −6 |
| Nac-L | 47 | 8.35 | <0.001 | −12 | 12 | −9 |
| Nac-R | 40 | 8.89 | <0.001 | 9 | 12 | −9 |
| Putamen-L | 220 | 7.11 | <0.001 | −15 | 12 | −9 |
| Putamen-R | 223 | 6.35 | <0.001 | 15 | 12 | −6 |
| MPFC-L | 382 | 7.35 | <0.001 | −3 | 51 | 9 |
| MPFC-R | 373 | 7.15 | <0.001 | 3 | 57 | 6 |
| OFC-L | 281 | 6.05 | <0.001 | −39 | 36 | −15 |
| OFC-R | 201 | 6.48 | <0.001 | 27 | 21 | −21 |
| PCC extending to angular gyrus, posterior cerebellum lobe etc. | 13,740 | 10.42 | –a | −3 | −36 | 36 |
| RPEMF: IGD | ||||||
| ACC-L | 305 | 8.45 | <0.001 | −6 | 48 | 3 |
| ACC-R | 221 | 8.27 | <0.001 | 3 | 48 | 9 |
| IFG-L | 338 | 6.73 | <0.001 | 45 | 39 | −3 |
| IFG-R | 327 | 6.2 | <0.001 | −36 | 39 | −9 |
| Caudate-L | 43 | 6.19 | <0.001 | −12 | 21 | −9 |
| Caudate-R | 100 | 6.38 | <0.001 | 9 | 12 | −3 |
| Nac-L | 49 | 7.67 | <0.001 | −12 | 12 | −12 |
| Nac-R | 40 | 7.86 | <0.001 | 9 | 12 | −9 |
| Putamen-L | 218 | 7.26 | <0.001 | 30 | −9 | −6 |
| Putamen-R | 224 | 7.02 | <0.001 | −30 | −9 | −6 |
| MPFC-L | 640 | 8.45 | <0.001 | −6 | 48 | 3 |
| MPFC-R | 559 | 8.27 | <0.001 | 3 | 48 | 9 |
| OFC-L | 223 | 6.48 | <0.001 | −27 | 30 | −15 |
| OFC-R | 216 | 6.46 | <0.001 | 24 | 18 | −18 |
| PCC extending to angular gyrus, posterior cerebellum lobe etc. | 21,824 | 8.65 | –a | 0 | −33 | 33 |
| RPEMB: HC | ||||||
| MPFC-R | 21 | 4.1 | 0.032 | 12 | 54 | 3 |
| RPEMB: IGD | ||||||
| MPFC-R | 28 | 4.49 | 0.01 | 6 | 51 | −6 |
| RPEMF: IGD > HC | ||||||
| Triangular part of IFG-R | 23 | 4.26 | 0.018 | 42 | 27 | 0 |
| RPE MB: IGD > HC: no significant activations | ||||||
Note: * SVC: FWE small volume corrected p value; a Threshold at AlphaSim corrected p < 0.05 (p < 0.001 and voxels ≥ 11).
(A) RPEMB maps in individuals with IGD and HCs. (B) RPEMF maps in individuals with IGD and HCs. The green-coloured regions indicate the ROIs
Citation: Journal of Behavioral Addictions 2025; 10.1556/2006.2025.00037
(A) Individuals with IGD showed increased RPEMF signals in the right inferior frontal gyrus (IFG), compared to HCs. And (B) the RPEMF values extracted from the right IFG were positively correlated with Internet Gaming Disorder Scale (IGDS) scores
Citation: Journal of Behavioral Addictions 2025; 10.1556/2006.2025.00037
For group comparisons, compared to HCs, participants with IGD showed increased RPEMF signals in the triangular part of the right IFG (Table 5, Fig. 3A). The group difference showed a large effect size (t63 = 3.98, p < 0.001, Cohen’ s d = 0.99). However, voxel-wise comparison of RPEΔMB maps did not reveal any significant group differences. Consistent with the regression analysis results, these findings indicate that participants with IGD were more inclined to rely on model-free learning strategies during the task.
In the IGD group, we found that the RPEMF signals in the right IFG were positively correlated with the scores of IGDS (rho = 0.468, p = 0.006) (Fig. 3B, Table 6). None of these correlations were significant in the HC group. In the HC group, we observed a positive correlation between the RPEMF signals in the right IFG and the IAT scores (rho = 0.376, p = 0.034), which was not significant in the IGD group. These correlations were all evident in the merged sample. Additionally, we noticed positive correlations between the RPEMF signals in the right IFG and craving, daily gaming time, and scores of BIS-non-planning in the merged groups (Table 6).
Correlations between RPEMF signals in the right IFG and clinical measures
| All participants merged | HC group | IGD group | |
| DGT | 0.394 | 0.153 | 0.037 |
| BAI | 0.138 | 0.016 | 0.089 |
| BDI | 0.097 | 0.181 | −0.2 |
| Craving | 0.434 | 0.193 | 0.116 |
| IGDS | 0.507 | 0.130 | 0.468 |
| IAT | 0.435 | 0.376 | 0.106 |
| BIS-11 | 0.228 | 0.025 | 0.10 |
| BIS-attentional | 0.136 | −0.09 | 0.008 |
| BIS-motor | 0.096 | 0.008 | −0.204 |
| BIS-NP | 0.275 | 0.006 | 0.281 |
Note. BAI: Beck Anxiety Inventory; BDI: Beck Depression Inventory; DGT: daily gaming time of the past year (in hours); IGDS: Internet Gaming Disorder Scale; BIS-11: Barratt Impulsiveness Scale (11th version); BIS-NP: Non-planning subscale of BIS-11; BIS-Attentional: attentional subscale of BIS-11; BIS-Motor: motor subscale of BIS-11; IAT: Internet Addiction Test. All participants merged: merged sample with participants of both groups (n = 65); Significant correlations (p < 0.05) were highlighted in bold fonts.
Discussion and conclusions
The current study aimed to investigate the imbalance of goal-directed/habitual control and related RPE neural signals in individuals with IGD. Analysis of choice patterns during the two-step decision-making task showed that compared to HCs, the individuals with IGD relied significantly more on reward history––a marker of habitual control––to guide subsequent choices. Consistent with this, the model fitting results revealed significantly lower goal-directed weight (ω) in the IGD group, related to HCs, and the ω value was associated with more severe problematic gaming behaviors (higher IGDS scores and more gaming time) in the IGD group. At the neural level, individuals with IGD exhibited increased model-free reward prediction error (RPEMF) signals, indicative of habitual control, in the right triangular part of the IFG compared to HCs. Moreover, the RPEMF signals in the right IFG cluster were positively correlated with the severity of gaming disorder in the IGD group. No significant group differences were found for the model-based RPE signals at the neural level. These results collectively suggest an imbalance in the goal-directed/habitual control system, with a bias toward habitual control in individuals with IGD.
Consistent with previous studies using the two-step task (Obst et al., 2018; Wyckmans et al., 2019), we found that participants in both the IGD and HC groups employed a mixed learning strategy, incorporating both habitual and goal-directed reinforcement learning. Importantly, group comparison showed a significantly stronger reward effect in IGDs than in HCs, while the two groups showed comparable coefficients in reward-transition interaction. These results support the notion that individuals with IGD may struggle to regulate their gaming behaviors based on long-term goals and may instead be more influenced by automatic, habitual responses (Everitt & Robbins, 2016). Our results in IGD are in line with previous research that has shown lower goal-directed weights using a similar two-step task in individuals with substance use disorders, such as alcohol dependence (Sebold et al., 2014; Sjoerds et al., 2013), those with binge-drinking problems (Doñamayor et al., 2018), and individuals with cocaine dependence (Ersche et al., 2016). Our results are also consistent with the findings of lower goal-directed weight in individuals with gambling disorder (Wagner et al., 2022; Wyckmans et al., 2019, 2022), and the findings of a greater inclination toward the habitual system in people with Internet addiction using the instrumental learning task (Zhou, Wang, Zhang, Li, & Nie, 2018). Our results are, however, inconsistent with the findings of Kwon et al. (2024) where behaviorally no sign of goal-directed/habitual imbalance was found in individuals with IGD. This discrepancy may be attributed to the larger sample size in our study and several methodological differences in task design. For example, Kwon et al. (2024) used the original version of the two-step task, which involves learning associations among abstract fractal images that was proposed to be too abstract to elicit a substantial MB strategy (Feher da Silva & Hare, 2020), while we used an adaptive version with a cover story to help participants better understand the task structure and potentially improve the MB weight.
It should be noted that, although the IGD group demonstrated a stronger tendency toward habitual control in decision-making, both groups exhibited a significant main effect of reward, a reward-transition interaction, and overall comparable task performance (total earned eggs) in this study. This suggests that, on the whole, both groups employed a combination of goal-directed and habitual control in their decision-making processes. Given that participants were unaware of the task's payoff structure, these findings cannot be attributed to a deliberate adoption of habitual control by individuals with IGD. Instead, this outcome may be linked to the design of the two-step task utilized in this study, which does not include a trade-off between accuracy and computational demands (Kool, Cushman, & Gershman, 2016).
The negative correlations between ω and both IGDS scores and daily game time indicate a direct link between over-reliance on habitual control and the severity of addictive symptoms, as well as the time spent gaming in the IGD group. The negative correlation between ω and IGDS scores remains significant in the merged sample, but not in the HC group, which may indicate that the imbalance between goal-directed and habitual control may be particularly relevant to IGD symptoms. This study revealed a significant positive correlation between ω and IAT scores in the HC group, but not in the IGD group, which could suggest that the greater inclination toward habitual control is restricted to gaming-related behaviors rather than general Internet usage in individuals with IGD. These findings are consistent with previous research that has linked habitual control with the severity of addictive behaviors (Sjoerds et al., 2013; Voon et al., 2015). Together, our results suggest that the tendency toward habitual behavior may contribute to the development or maintenance of addictive patterns in IGD, aligning with the perspective that habit formation plays a role in addiction (Brand, 2022; Lüscher et al., 2020).
There were numerous literatures demonstrated the phenomenon called “drug-induced facilitation of habit” in addiction (Vandaele & Ahmed, 2020). That is (1) the transition to habit is faster for the drug of abuse compared to a nondrug reward, such as food (Dickinson, Wood, & Smith, 2002; Loughlin, Funk, Coen, & Lê, 2017). (2) drug of abuse could promote habitual responding (insensitive to devaluation) for other drugs and non-drug rewards. For instance, when rats were pretreated with injections of amphetamine or nicotine, they acquired a cocaine self-administration habit much quicker than control animals (Horger, Giles, & Schenk, 1992). Pre-exposure to amphetamine, relative to the saline-pretreated group, could also facilitate habits in subsequent self-administration training of amphetamine (Mendrek, Blaha, & Phillips, 1998). Additionally, amphetamine sensitization was found to accelerates the shift from goal-directed to habit-based responding in rats trained to press a lever for a food reward (Nelson & Killcross, 2006). This study thus demonstrated that, certain behavioral patterns (gaming) could also promote habits in instrumental task with monetary rewards (i.e. the two-step task), and this change is associated with addictive symptoms in individuals with IGD.
At the neural level, we confirmed that RPE maps of the habitual system consist of the key regions related to reward processing and executive control, such as NAc, MPFC, OFC, caudate, putamen and ACC (Shiflett & Balleine, 2010; Yin, Ostlund, Knowlton, & Balleine, 2005). The RPEMB maps consist of the right MPFC in both IGD and HC groups, which is also in line with previous literature on the neural basis of the goal-directed system (Balleine & O'Doherty, 2010; Daw et al., 2011; Yin et al., 2004). Importantly, we found hyperactive RPEMF signals in the right triangular part of IFG in IGDs relative to HCs, but no case-control differences were found in the REPMB maps, again highlighting an over-reliance on habitual control (but not impaired goal-directed control) in IGD.
In line with our results, a meta-analysis revealed significant hyperactivation in the right IFG in IGD during the cue-reactivity task, indicating a link between hyperactive right IFG and cue-induced game craving (Zheng et al., 2019). Similarly, hyperactivity in the right IFG has been found in individuals with various types of addiction, including cocaine, cannabis, and methamphetamine dependence, and was associated with a failure to suppress the urges to use or seek drugs (Klugah-Brown et al., 2020; Morein-Zamir & Robbins, 2015). The right triangular part of IFG is heavily implicated in the inhibition control of behaviors, which is one core of executive functions (Aron, Robbins, & Poldrack, 2004). Thus, the increased RPEMF in the right IFG observed in our study could indicate impaired behavioral control in IGD. Particularly, the hyperactivity in the right IFG could reflect an altered attribution between goal-directed and habitual behavioral control. Computational neuroscience studies showed that the balance between goal-directed and habitual control is mediated by uncertainty-related attribution signals from the IFG, which represents the reliability of the strategies and was modulating the activity of the right putamen that was involved in habitual (MF) learning (Kim et al., 2024; Lee et al., 2014). From this perspective, the hyperactivity in the right IFG could reflect an altered attribution process in IGD in favoring habitual control. The aberrant uncertainty coding in the IFG arbitrator could give rise to an overreliance upon the habitual strategy that leads to thoughtless and impulsive decision-making. Our correlation results support this notion, as RPEMF in the right IFG was negatively correlated with IGDS scores, suggesting an association among the altered attribution process, impulsivity, and addictive symptoms in IGD.
A few caveats about the present study should be mentioned. Firstly, due to the cross-sectional design of the current study, we cannot make causal inferences regarding changes in goal-directed and habitual behaviors and addiction characteristics in IGD. Secondly, it should be noted that other variables, such as working memory capacity, that could affect the balance of the goal-directed/habitual systems (Otto, Raio, Chiang, Phelps, & Daw, 2013; Vandaele & Ahmed, 2020), were not assessed in this study. Further research may benefit from including additional variables to rule out alternative explanations for the findings. Thirdly, it was argued that the trade-off between computational demands and accuracy benefits plays an important role in the arbitration between goal-directed and habitual control (Kool, Gershman, & Cushman, 2017). However, the two-step task used in this study may not adequately capture this trade-off, making the goal-directed strategy less favorable for this task and potentially reducing the task's sensitivity to assess model-based learning (Kool et al., 2016). Future studies could explore the balance of goal-directed and habitual control in IGD using different tasks. Lastly, as the goal-directed and habitual behaviors were measured on the same scale (where the choices are determined by a weighted combination of goal-directed (MB) and habitual (MF) values), the two-step task used in this study could have difficulty to pinpoint the specific contribution of each of the two hypothesized controllers and the interaction between them. Further studies using separate measures of goal-directed and habitual behaviors, such as such as Frolich, Esmeyer, Endrass, Smolka, and Kiebel (2022) and Luque, Molinero, Watson, López, and Le Pelley (2020), could bring more information on the interaction of these two systems.
In summary, our study provides behavioral and neural evidence of an imbalance between goal-directed and habitual control that favors habitual strategies in IGD. This imbalance is linked to the severity of addictive symptoms and impulsivity, suggesting that an inclination toward habitual control in IGD could potentially contribute to the development or maintenance of the addiction. Future research could conduct longitudinal studies to better understand the causal relationships between control imbalances and the trajectory of IGD. For example, although we are inclined to think that the imbalance between goal-directed and habitual control arose from repeated exposure to gaming-related rewards, our results cannot rule out the possibility that the imbalanced goal-directed and habitual control in individuals with IGD may be a proposed susceptibility characteristic of this condition (Brand et al., 2019). A prospective study starting before a gamer becomes addicted to gaming could thus provide important information on this matter. Additionally, exploring neural mechanisms and comparing IGD with other behavioral addictions could enrich also our understanding of this phenomenon (Wong et al., 2020). Clinically, further research should focus on evaluating the effectiveness of interventions that could modulate this imbalance, which could involve manipulating conditions that affect the occurrence of habitual control (Vandaele & Ahmed, 2020), or more comprehensive therapies that allow individuals with IGD to “keep distance with one's automatic (habitual) response”, such as cognitive-behavioral therapy (Stevens, King, Dorstyn, & Delfabbro, 2019).
Funding sources
This work was partly supported by the Humanities and Social Science Fund of Ministry of Education of China (23YJA190004); the National Science Foundation of China (32200882); Sichuan Science and Technology Department (23ZDYF2557); Sichuan Applied Psychology Research Center (CSXL-22102); Southwest Medical University (2022ZD004); and Social Sciences Federation of Southwest Medical University (SMUSS202220).
Authors' contribution
WL: writing – original draft, visualization, methodology, formal analysis, conceptualization. YH: Writing – review and editing, task, modeling, supervision. YP: investigation, formal analysis. GC: Data collection. KG: investigation, recruitment. KL: investigation, recruitment. DW: investigation, data collection. CQ: investigation, data collection. XC: funding acquisition. MT: investigation, data collection. LZ: investigation, data collection. YZ: investigation, data collection. RY: writing – review and editing. JC: Writing – review and editing, supervision, funding acquisition, conceptualization.
Conflict of interest
The authors declare no conflict of interest.
Data availability
The datasets used in the current study are available from the corresponding author upon reasonable request.
Supplementary material
Supplementary data to this article can be found online at https://doi.org/10.1556/2006.2025.00037.
References
APA (2013). Diagnostic and statistical manual of mental disorders (DSM-5®). American Psychiatric Pub.
Aron, A. R., Robbins, T. W., & Poldrack, R. A. (2004). Inhibition and the right inferior frontal cortex. Trends in Cognitive Sciences, 8(4), 170–177. https://doi.org/10.1016/j.tics.2004.02.010.
Balleine, B. W., & O'Doherty, J. P. (2010). Human and rodent homologies in action control: Corticostriatal determinants of goal-directed and habitual action. Neuropsychopharmacology, 35(1), 48–69. https://doi.org/10.1038/npp.2009.131.
Beck, A. T., Epstein, N., Brown, G., & Steer, R. A. (1988). An inventory for measuring clinical anxiety: Psychometric properties. Journal of Consulting and Clinical Psychology, 56(6), 893–897. https://doi.org/10.1037/0022-006X.56.6.893.
Beck, A. T., Steer, R. A., & Carbin, M. G. (1988). Psychometric properties of the beck depression inventory: Twenty-five years of evaluation. Clinical Psychology Review, 8(1), 77–100. https://doi.org/10.1016/0272-7358(88)90050-5.
Brand, M. (2022). Can internet use become addictive? Science, 376(6595), 798-799. https://doi.org/10.1126/science.abn4189.
Brand, M., Wegmann, E., Stark, R., Müller, A., Wölfling, K., Robbins, T. W., & Potenza, M. N. (2019). The interaction of person-affect-cognition-execution (I-PACE) model for addictive behaviors: Update, generalization to addictive behaviors beyond internet-use disorders, and specification of the process character of addictive behaviors. Neuroscience and Biobehavioral Reviews, 104, 1–10.
Carpenter, B., Gelman, A., Hoffman, M. D., Lee, D., Goodrich, B., Betancourt, M., … Riddell, A. (2017). Stan: A probabilistic programming language. Journal of Statistical Software, 76. https://doi.org/10.18637/jss.v076.i01.
Chung, Y., Rabe-Hesketh, S., Dorie, V., Gelman, A., & Liu, J. (2013). A nondegenerate penalized likelihood estimator for variance parameters in multilevel models. Psychometrika, 78(4), 685–709. https://doi.org/10.1007/s11336-013-9328-2.
Daw, N. D., Gershman, S. J., Seymour, B., Dayan, P., & Dolan, R. J. (2011). Model-based influences on humans' choices and striatal prediction errors. Neuron, 69(6), 1204–1215. https://doi.org/10.1016/j.neuron.2011.02.027.
Daw, N. D., Niv, Y., & Dayan, P. (2005). Uncertainty-based competition between prefrontal and dorsolateral striatal systems for behavioral control. Nature Neuroscience, 8, 1704. https://doi.org/10.1038/nn1560.
Decker, J. H., Otto, A. R., Daw, N. D., & Hartley, C. A. (2016). From creatures of habit to goal-directed learners: Tracking the developmental emergence of model-based reinforcement learning. Psychological Science, 27(6), 848–858. https://doi.org/10.1177/0956797616639301.
Dickinson, A., Wood, N., & Smith, J. W. (2002). Alcohol seeking by rats: Action or habit? The Quarterly Journal of Experimental Psychology Section B, 55(4), 331–348. https://doi.org/10.1080/0272499024400016.
Doñamayor, N., Strelchuk, D., Baek, K., Banca, P., & Voon, V. (2018). The involuntary nature of binge drinking: Goal directedness and awareness of intention. Addiction Biology, 23(1), 515-526. https://doi.org/10.1111/adb.12505.
Ersche, K. D., Gillan, C. M., Jones, P. S., Williams, G. B., Ward, L. H., Luijten, M., …, & Robbins, T. W. (2016). Carrots and sticks fail to change behavior in cocaine addiction. Science, 352(6292), 1468–1471.
Everitt, B. J., & Robbins, T. W. (2005). Neural systems of reinforcement for drug addiction: From actions to habits to compulsion. Nature Neuroscience, 8(11), 1481–1489.
Everitt, B. J., & Robbins, T. W. (2016). Drug addiction: Updating actions to habits to compulsions ten years on. Annual Review of Psychology, 67, 23–50.
Feher da Silva, C., & Hare, T. A. (2020). Humans primarily use model-based inference in the two-stage task. Nature Human Behavior, 4(10), 1053–1066. https://doi.org/10.1038/s41562-020-0905-y.
Frolich, S., Esmeyer, M., Endrass, T., Smolka, M. N., & Kiebel, S. J. (2022). Interaction between habits as action sequences and goal-directed behavior under time pressure. Frontiers in Neuroscience, 16, 996957. https://doi.org/10.3389/fnins.2022.996957.
Gentile, D. A., Choo, H., Liau, A., Sim, T., Li, D., Fung, D., & Khoo, A. (2011). Pathological video game use among youths: A two-year longitudinal study. Pediatrics, 127(2), e319–e329. https://doi.org/10.1542/peds.2010-1353.
Gillan, C. M., Kosinski, M., Whelan, R., Phelps, E. A., & Daw, N. D. (2016). Characterizing a psychiatric symptom dimension related to deficits in goal-directed control. eLife, 5, e11305. https://doi.org/10.7554/eLife.11305.
Haber, S. N., & Knutson, B. (2010). The reward circuit: Linking primate anatomy and human imaging. Neuropsychopharmacology, 35(1), 4–26. https://doi.org/10.1038/npp.2009.129.
Hogarth, L. (2020). Addiction is driven by excessive goal-directed drug choice under negative affect: Translational critique of habit and compulsion theory. Neuropsychopharmacology, 45(5), 720–735. https://doi.org/10.1038/s41386-020-0600-8.
Horger, B. A., Giles, M. K., & Schenk, S. (1992). Preexposure to amphetamine and nicotine predisposes rats to self-administer a low dose of cocaine. Psychopharmacology, 107, 271–276.
Huang, Y., Yaple, Z. A., & Yu, R. (2020). Goal-oriented and habitual decisions: Neural signatures of model-based and model-free learning. Neuroimage, 215, 116834. https://doi.org/10.1016/j.neuroimage.2020.116834.
Kim, T., Lee, S. W., Lho, S. K., Moon, S. Y., Kim, M., & Kwon, J. S. (2024). Neurocomputational model of compulsivity: Deviating from an uncertain goal-directed system. Brain, 147(6), 2230–2244. https://doi.org/10.1093/brain/awae102.
Klugah-Brown, B., Di, X., Zweerings, J., Mathiak, K., Becker, B., & Biswal, B. (2020). Common and separable neural alterations in substance use disorders: A coordinate-based meta-analyses of functional neuroimaging studies in humans. Human Brain Mapping, 41(16), 4459–4477. https://doi.org/10.1002/hbm.25085.
Koob, G. F., & Le Moal, M. (2005). Plasticity of reward neurocircuitry and the ‘dark side’ of drug addiction. Nature Neuroscience, 8(11), 1442.
Koob, G. F., & Volkow, N. D. (2016). Neurobiology of addiction: A neurocircuitry analysis. The Lancet Psychiatry, 3(8), 760–773. https://doi.org/10.1016/S2215-0366(16)00104-8.
Kool, W., Cushman, F. A., & Gershman, S. J. (2016). When does model-based control pay off? PLoS Computational Biology, 12(8), e1005090. https://doi.org/10.1371/journal.pcbi.1005090.
Kool, W., Gershman, S. J., & Cushman, F. A. (2017). Cost-benefit arbitration between multiple reinforcement-learning systems. Psychological Science, 28(9), 1321–1333. https://doi.org/10.1177/0956797617708288.
Kwon, M., Choi, H., Park, H., Ahn, W. Y., & Jung, Y. C. (2024). Neural correlates of model-based behavior in internet gaming disorder and alcohol use disorder. Journal of Behavioral Addictions, 13(1), 236–249. https://doi.org/10.1556/2006.2024.00006.
Lee, D., Namkoong, K., Lee, J., & Jung, Y. C. (2021). Dorsal striatal functional connectivity changes in internet gaming disorder: A longitudinal magnetic resonance imaging study. Addiction Biology, 26(1), e12868. https://doi.org/10.1111/adb.12868.
Lee, S. W., Shimojo, S., & O'Doherty, J. P. (2014). Neural computations underlying arbitration between model-based and model-free learning. Neuron, 81(3), 687–699. https://doi.org/10.1016/j.neuron.2013.11.028.
Lei, W., Liu, K., Zeng, Z., Liang, X., Huang, C., Gong, K., … Chen, J. (2020). The psychometric properties of the Chinese version internet gaming disorder scale. Addictive Behaviors, 106, 106392. https://doi.org/10.1016/j.addbeh.2020.106392.
Lemmens, J. S., Valkenburg, P. M., & Gentile, D. A. (2015). The internet gaming disorder scale. Psychol Assess, 27(2), 567–582. https://doi.org/10.1037/pas0000062.
Lim, J. A., Lee, J. Y., Jung, H. Y., Sohn, B. K., Choi, S. W., Kim, Y. J., … Choi, J. S. (2016). Changes of quality of life and cognitive function in individuals with internet gaming disorder: A 6-month follow-up. Medicine (Baltimore), 95(50), e5695. https://doi.org/10.1097/MD.0000000000005695.
Liu, L., Yao, Y. W., Fang, X. Y., Xu, L. X., Hu, M. J., Zhang, J. T., & Potenza, M. N. (2024). Compulsivity-related behavioral features of problematic usage of the internet: A scoping review of paradigms, progress, and perspectives. Journal of Behavioral Addictions, 13(2), 429–449. https://doi.org/10.1556/2006.2024.00023.
Liu, L., Yip, S. W., Zhang, J. T., Wang, L. J., Shen, Z. J., Liu, B., … Fang, X. Y. (2017). Activation of the ventral and dorsal striatum during cue reactivity in internet gaming disorder. Addiction Biology, 22(3), 791–801.
Loughlin, A., Funk, D., Coen, K., & Lê, A. (2017). Habitual nicotine-seeking in rats following limited training. Psychopharmacology, 234, 2619–2629.
Luque, D., Molinero, S., Watson, P., López, F. J., & Le Pelley, M. E. (2020). Measuring habit formation through goal-directed response switching. Journal of Experimental Psychology: General, 149(8), 1449.
Lüscher, C., Robbins, T. W., & Everitt, B. J. (2020). The transition to compulsion in addiction. Nature Reviews Neuroscience, 1–17.
Mendrek, A., Blaha, C. D., & Phillips, A. G. (1998). Pre-exposure of rats to amphetamine sensitizes self-administration of this drug under a progressive ratio schedule. Psychopharmacology, 135, 416–422.
Mihara, S., & Higuchi, S. (2017). Cross‐sectional and longitudinal epidemiological studies of internet gaming disorder: A systematic review of the literature. Psychiatry and Clinical Neurosciences, 71(7), 425–444.
Morein-Zamir, S., & Robbins, T. W. (2015). Fronto-striatal circuits in response-inhibition: Relevance to addiction. Brain Research, 1628, 117–129. https://doi.org/10.1016/j.brainres.2014.09.012.
Nelson, A., & Killcross, S. (2006). Amphetamine exposure enhances habit formation. Journal of Neuroscience, 26(14), 3805–3812. https://doi.org/10.1523/jneurosci.4305-05.2006.
Obst, E., Schad, D. J., Huys, Q. J. M., Sebold, M., Nebe, S., Sommer, C., … Zimmermann, U. S. (2018). Drunk decisions: Alcohol shifts choice from habitual towards goal-directed control in adolescent intermediate-risk drinkers. Journal of Psychopharmacology, 32(8), 855–866. https://doi.org/10.1177/0269881118772454.
Otto, A. R., Gershman, S. J., Markman, A. B., & Daw, N. D. (2013). The curse of planning: Dissecting multiple reinforcement-learning systems by taxing the central executive. Psychological Science, 24(5), 751–761.
Otto, A. R., Raio, C. M., Chiang, A., Phelps, E. A., & Daw, N. D. (2013). Working-memory capacity protects model-based learning from stress. Proceedings of the National Academy of Sciences, 110(52), 20941–20946.
Patton, J. H., Stanford, M. S., & Barratt, E. S. (1995). Factor structure of the Barratt impulsiveness scale. Journal of Clinical Psychology, 51(6), 768–774.
R Core Team. (2021). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. https://www.R-project.org/.
Rolls, E. T., Huang, C. C., Lin, C. P., Feng, J., & Joliot, M. (2020). Automated anatomical labelling atlas 3. Neuroimage, 206, 116189. https://doi.org/10.1016/j.neuroimage.2019.116189.
Schreiner, D. C., Renteria, R., & Gremel, C. M. (2020). Fractionating the all-or-nothing definition of goal-directed and habitual decision-making. Journal of Neuroscience Research, 98(6), 998–1006. https://doi.org/10.1002/jnr.24545.
Sebold, M., Deserno, L., Nebe, S., Schad, D. J., Garbusow, M., Hägele, C., … Smolka, M. (2014). Model-based and model-free decisions in alcohol dependence. Neuropsychobiology, 70(2), 122–131.
Sheehan, D. V., Lecrubier, Y., Sheehan, K. H., Amorim, P., Janavs, J., Weiller, E., … Dunbar, G. C. (1998). The mini-international neuropsychiatric interview (MINI): The development and validation of a structured diagnostic psychiatric interview for DSM-IV and ICD-10. Journal of Clinical Psychiatry, 59(20), 22–33.
Shiflett, M. W., & Balleine, B. W. (2010). At the limbic-motor interface: Disconnection of basolateral amygdala from nucleus accumbens core and shell reveals dissociable components of incentive motivation. European Journal of Neuroscience, 32(10), 1735–1743. https://doi.org/10.1111/j.1460-9568.2010.07439.x.
Sjoerds, Z., de Wit, S., van den Brink, W., Robbins, T. W., Beekman, A. T., Penninx, B. W., & Veltman, D. J. (2013). Behavioral and neuroimaging evidence for overreliance on habit learning in alcohol-dependent patients. Translational Psychiatry, 3, e337. https://doi.org/10.1038/tp.2013.107.
Song, C., Kanai, R., Fleming, S. M., Weil, R. S., Schwarzkopf, D. S., & Rees, G. (2011). Relating inter-individual differences in metacognitive performance on different perceptual tasks. Consciousness and Cognition, 20(4), 1787–1792. https://doi.org/10.1016/j.concog.2010.12.011.
Stevens, M. W., Dorstyn, D., Delfabbro, P. H., & King, D. L. (2021). Global prevalence of gaming disorder: A systematic review and meta-analysis. Australian and New Zealand Journal of Psychiatry, 55(6), 553–568. https://doi.org/10.1177/0004867420962851.
Stevens, M. W., King, D. L., Dorstyn, D., & Delfabbro, P. H. (2019). Cognitive–behavioral therapy for internet gaming disorder: A systematic review and meta‐analysis. Clinical Psychology and Psychotherapy, 26(2), 191–203.
Van Rooij, A. J., Kuss, D. J., Griffiths, M. D., Shorter, G. W., Schoenmakers, T. M., & Van De Mheen, D. (2014). The (co-) occurrence of problematic video gaming, substance use, and psychosocial problems in adolescents. Journal of Behavioral Addictions, 3(3), 157–165. https://doi.org/10.1556/JBA.3.2014.013.
Vandaele, Y., & Ahmed, S. H. (2020). Habit, choice, and addiction. Neuropsychopharmacology, 46(4), 689–698. https://doi.org/10.1038/s41386-020-00899-y.
Voon, V., Derbyshire, K., Rück, C., Irvine, M. A., Worbe, Y., Enander, J., … Sahakian, B. J. (2015). Disorders of compulsivity: A common bias towards learning habits. Molecular Psychiatry, 20(3), 345–352.
Wagner, B., Mathar, D., & Peters, J. (2022). Gambling environment exposure increases temporal discounting but improves model-based control in regular slot-machine gamblers. Computational Psychiatry, 6(1), 142–165. https://doi.org/10.5334/cpsy.84.
Walther, B., Morgenstern, M., & Hanewinkel, R. (2012). Co-occurrence of addictive behaviours: Personality factors related to substance use, gambling and computer gaming. European Addiction Research, 18(4), 167–174.
Weaver, E., Gradisar, M., Dohnt, H., Lovato, N., & Douglas, P. (2010). The effect of presleep video-game playing on adolescent sleep. Journal of Clinical Sleep Medicine, 06(02), 184-189. https://doi.org/10.5664/jcsm.27769.
WHO. (2018). International classification of diseases for mortality and morbidity statistics (11th revision). Retrieved from https://icd.who.int/browse/2024-01/mms/en#1448597234.
Wong, H. Y., Mo, H. Y., Potenza, M. N., Chan, M. N. M., Lau, W. M., Chui, T. K., … Lin, C.-Y. (2020). Relationships between severity of internet gaming disorder, severity of problematic social media use, sleep quality and psychological distress. International Journal of Environmental Research and Public Health, 17(6), 1879.
Wyckmans, F., Banerjee, N., Saeremans, M., Ross, O., Kornreich, C., Vanderijst, L., … Buchanan, T. (2022). The modulation of acute stress on model-free and model-based reinforcement learning in gambling disorder. Journal of Behavioral Addictions, 11(3), 831–844. https://doi.org/10.1101/2022.05.05.490735.
Wyckmans, F., Otto, A. R., Sebold, M., Daw, N., Bechara, A., Saeremans, M., … Noel, X. (2019). Reduced model-based decision-making in gambling disorder. Scientific Reports, 9(1), 19625. https://doi.org/10.1038/s41598-019-56161-z.
Yin, H. H., Knowlton, B. J., & Balleine, B. W. (2004). Lesions of dorsolateral striatum preserve outcome expectancy but disrupt habit formation in instrumental learning. European Journal of Neuroscience, 19(1), 181–189.
Yin, H. H., Ostlund, S. B., Knowlton, B. J., & Balleine, B. (2005). The role of the dorsomedial striatum in instrumental conditioning. European Journal of Neuroscience, 22(2), 513–523.
Young, K. S. (1998). Caught in the net: How to recognize the signs of internet addiction--and a winning strategy for recovery. New York, NY: John Wiley and Sons.
Zheng, H., Hu, Y., Wang, Z., Wang, M., Du, X., & Dong, G. (2019). Meta-analyses of the functional neural alterations in subjects with internet gaming disorder: Similarities and differences across different paradigms. Progress in Neuro-Psychopharmacology and Biological Psychiatry, 94, 109656. https://doi.org/10.1016/j.pnpbp.2019.109656.
Zhou, B., Wang, W., Zhang, W., Li, Y., & Nie, J. (2018). Succumb to habit: Behavioral evidence for overreliance on habit learning in internet addicts. Computers in Human Behavior, 89, 230–236.
