Item Response Theory Analysis of the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition Criteria for Internet Gaming Disorders
Article information
Abstract
Objective
The nine diagnostic criteria for Internet Gaming Disorder (IGD) proposed in Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) have been widely adopted, but have also faced significant criticism. This study evaluated the psychometric properties of the nine DSM-5 criteria for IGD using item response theory (IRT).
Methods
Four DSM-5-based scales, including the Game Addiction Scale, the Internet Gaming Disorder Scale Short-Form, the Ten- Item Internet Gaming Disorder Test, and the Internet Gaming Disorder Scale, were administered to 1,530 Chinese adolescents and young adults. IRT models were used to assess the psychometric properties of these criteria. The potential for differential item functioning (DIF) related to sex and between adolescents and adults was also evaluated.
Results
All nine DSM-5 criteria demonstrated very high to perfect discrimination (a >1.38), providing substantial diagnostic information (item information >0.476) for diagnosing IGD. These scales, despite variations in wording and scoring methods, exhibited strong correlations in total scores (r≥0.502, p<0.001) and high consistency (Kendall’s W ≥0.656, p<0.05) concerning the fitted IRT parameters. Among the nine criteria, escape showed the lowest cross-scale discrimination, item information, and difficulty. In contrast, withdrawal showed the highest item information and discrimination, and the second highest difficulty. Additionally, there was no evidence of DIF related to sex or age (adolescents vs. adults) across nearly all scale items.
Conclusion
These findings indicate that the nine DSM-5 criteria generally possess acceptable psychometric properties for diagnosing IGD. Notably, withdrawal may represent a core symptom of IGD. Conversely, escape demonstrated inferior performance compared to the other eight criteria, indicating a potential need for further revision.
INTRODUCTION
Online gaming has become a staple of our daily lives, serving as a popular form of entertainment. However, excessive gaming can lead to impairment in academic performance, work responsibilities, and both physical and mental well-being. In 2013, Internet Gaming Disorder (IGD) was included in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5), as a condition that requires further evaluation [1]. The DSM-5 recommended nine tentative diagnosis criteria for IGD, including: 1) preoccupation, 2) withdrawal, 3) tolerance, 4) loss of control, 5) give up other activities, 6) persistence, 7) deception, 8) escape, and 9) negative outcomes. A diagnosis of IGD requires the endorsement of at least five of these nine criteria.
The DSM-5 criteria have been widely adopted in research and clinical practices [2-4]. Many instruments have been developed based on these criteria. For example, the Ten-Item Internet Gaming Disorder Test (IGDT-10) [5], the 9-item Internet Gaming Disorder Scale Short-Form (IGDS9-SF) [6], the Internet Gaming Disorder Scale (IGDS) [7], 7-item Game Addiction Scale (GAS) [8], and the Chinese IGD Scale (C-IGDS) [9]. The DSM-5 standard is currently the most influential diagnostic standard for IGD, which has undoubtedly greatly promoted research in IGD-related fields.
However, the DSM-5 criteria have also faced considerable criticism. Firstly, the diagnostic criteria for IGD were mainly drawn from those of gambling and substance use disorder [10]. Yet, the theoretical rationality of this approach has been questioned by some researchers. For instance, Pies [11] claims that some well-established features of substance use disorders, such as withdrawal and tolerance, can be difficult to define in the context of IGD as it does not involve chemical consumption. Likewise, some criticisms suggest that give up other activities lacks a clear connection to functional impairment and can also be a symptom of depression (which is highly comorbid with IGD) [12]. Secondly, empirical studies have demonstrated low diagnostic accuracy for certain criteria for IGD. Research by Ko et al. [13] highlighted concerns about the reliability of deception and escape, while other studies, including those by Ko et al. [14] and Lemmens et al. [7], have noted similar issues with the escape criterion. These findings suggest that the current diagnostic criteria may not effectively capture the complexities of gaming behaviours, raising questions about the overall validity of these criteria in diagnosing IGD.
Item response theory (IRT) is a commonly adopted psychometric model that aims to determine the relationship between latent traits (e.g., IGD severity) and responses [15]. IRT enables precise measurement of difficulty, discrimination, and item information of each item, and allows direct comparison of these features between items [16]. IRT has been frequently adopted in studies of IGD [5,17-19]. For example, Király et al. [5] carried out an IRT analysis on the IGDT-10 and found that IGD manifests through a different set of symptoms depending on the severity level. Specifically, persistence, preoccupation, negative outcomes, and escape were associated with lower severity of IGD, while tolerance, loss of control, give up other activities, and deception were associated with more severe levels. They also found that preoccupation and escape provided very little information to estimate IGD severity, related to other criteria. Likewise, Gomez et al. [17] applied the IRT to the items of IGDS9-SF and found that escape had a relatively lower discrimination compared to the other eight criteria. It also showed low reliability at all trait levels. However, these studies were all conducted with a single scale, and the conclusions are limited to the given scale rather than the diagnostic criteria themselves. As a result, it remains unclear whether these findings, for example, the low discrimination in escape, were due to the definition of symptoms per se or the wording of the scales.
The study aimed to examine the psychometric properties of the nine DSM-5 criteria of IGD by conducting IRT analyses on four commonly used DSM-5-based self-report scales: GAS, IGDT-10, IGDS9-SF, and IGDS. Based on previous studies, we hypothesize that escape may have lower IRT parameters (discrimination, difficulty, and item information) compared to other diagnostic criteria.
METHODS
Participants and procedures
A combined online and offline survey using a cross-sectional design was conducted to test the psychometric properties of four scales. Particularly, online questionnaires were distributed through an online survey platform (https://www.wjx.cn/). The offline survey was conducted by researchers-issued questionnaires. Previous studies have found approximate general equivalence of paper-and-pencil and Internet self-report survey-based questionnaires [20]. Additionally, to further assess the impact of the survey method, we tested measurement invariance using a multi-group confirmatory factor analysis (CFA) method [21], which demonstrated strong measurement invariance across subsamples surveyed online and offline (Supplementary Material and Supplementary Table 1).
A total of 1,722 potential participants initiated the survey. Respondents who missed any item or had overt perfunctory responses (e.g., consistently selecting the same options) were excluded. Consequently, data from 1,530 respondents were analyzed. This study was carried out in accordance with the Declaration of Helsinki and was approved by the Clinical Experiment Ethics Committee of Affiliated Hospital of Southwest Medical University (No. KY2024181). The participants provided informed consent to participate in this study. For participants under 16 years of age, parental (or legal guardian) informed consent was also obtained.
Measures
Four scales developed based on the DSM-5 criteria, including GAS, IGDT-10, IGDS9-SF, and IGDS, were administered. These scales were selected because they are commonly used in this field and have demonstrated good psychometric properties. In a systematic review, King et al. [22] evaluated 32 tools assessing IGD symptoms across 320 empirical studies and recommending five instruments with relatively great evidential support, including GAS, IGDT-10, IGDS9-SF, IGDS, and Assessment of Internet and Computer Addiction Scale-Gaming [23]. Good internal consistency and test-retest reliability of these instruments were further confirmed in a later meta-analysis by Yoon et al. [24].
7-item GAS
This self-report scale assesses the severity of IGD and its harmful effects that occurred in the last 6 months. Each item is scored on a 1–5 Likert scale (1=never, 5=very often), with higher scores indicating a greater level of game addiction. GAS showed good concurrent validity across samples [8]. Although the GAS was developed before the publication of DSM-5, it covers most (seven out of nine) of the DSM-5 criteria (Table 1).
Correspondence of items from the four scales with DSM-5 diagnostic criteria
IGDT-10
The IGDT-10 is a ten-item self-report scale that assesses the game-play activities during the last 12 months [5]. Each item was scored on a 1–3 Likert scale (1=never, 2=sometimes, 3=often).
IGDS9-SF
The IGDS9-SF is a nine-item self-report scale that assesses the severity of IGD and its detrimental effects over the past 12 months [6]. Each item is rated on a 1–5 Likert scale (1=never, 5=very often), with higher scores indicating severe symptoms.
IGDS
The IGDS is a self-report scale that assesses the presence of nine IGD symptoms over the past 12 months, with each item answered in a dichotomous “yes/no” manner [7]. The total score ranges from 0 to 9, with higher scores representing higher levels of IGD. Our previous study found good reliability and validity of the Chinese version of the IGDS [25].
Based on the semantic understanding of the items, all items of the four scales corresponded to the nine DSM-5 criteria in a one-to-one manner, except that items 9 and 10 in IGDT-10 were combined in one criterion, and GAS covers seven of the nine criteria (Table 1). The internal consistency of the four scales in this study was good (Cronbach’s α ≥0.779) (Table 2), indicating strong reliability for application.
Internal consistency and Spearman’s correlations among scales
IRT models
According to the scoring methods of items, graded response models (GRM) [26] were established for GAS, IGDT-10, and IGSF9-SF separately, while a two-parameter logistic (2PL) model was specified for IGDS.
The 2PL model is a commonly used IRT model in analyzing binary data, which can be expressed as:
Where θ represents the latent trait level of the subject; Pi(θ) represents the probability that a subject with a latent trait level of θ has a correct/positive response on item i; ai represents the discrimination of item i, and bi represents the difficulty of item i.
In this study, θ was defined as a range from -5 to 5, with a higher value indicating a higher level on the latent trait (i.e., IGD level in our case). The IRT analysis estimates two key parameters of items: the discrimination parameter (a) and the difficulty parameter (b). The discrimination parameter refers to the accuracy of the criterion in distinguishing between respondents with different latent traits [27]. The difficulty parameter represents a point on the latent trait where the probability of a positive/correct response is 50% [27]. In addition to discrimination and the difficulty, IRT models can also generate item information, which refers to the amount of information provided by an individual item about the respondents’ trait level [28].
The GRM is widely used for analyzing ordinal response data. Similar to the 2PL model, the likelihood of a respondent choosing a specific category depends on their latent trait level, discrimination, and difficulty. However, unlike the 2PL, in the GRM, each score level for every item has a distinct difficulty parameter. Specifically, the probability of a subject attaining a particular score level (k) on an item (i) is defined by:
Where θ represents the subject’s latent trait level, ai represents the discrimination parameter for item i, and bik represents the difficulty parameter for score level k of item i.
Note that, as items 9 and 10 of IGDT-10 are related to the same criterion, negative outcomes, the scores of these two items were combined by taking the maximum score of items 9 and 10 to facilitate cross-scale comparisons. This approach aligns with the original paper of Király et al. [5], which suggests that a positive response to either item indicates an endorsement of negative outcomes. The combined data (nine items) were used in the IRT analysis.
Statistical analysis
All analyses were conducted using R (R Project for Statistical Computing). We began by calculating Spearman’s correlations between the scale scores to verify consistency across the instruments.
Next, we tested the assumptions for IRT, including 1) unidimensionality of the measured trait, 2) local independence, and 3) monotonicity. The unidimensionality assumption posits that the correlation among these items can be explained by a single latent factor. We performed CFA with a one-factor structure, using the Lavaan package [29], to test unidimensionality and local dependence assumptions. According to the criteria proposed by Evren et al. [30], the chi-square freedom ratio (χ2/df) ≤5, the Goodness of Fit Index (GFI), the Comparative Fit Index (CFI), Normed Fit Index (NFI) and Incremental Fit Index (IFI) >0.90, and Root Mean Square Error of Approximation (RMSEA) <0.05 were considered indicators of good fits. Local independence assumes that the response to one item should not influence the response to another when the latent trait is held constant [31]. Local independence is closely related to unidimensionality, with some authors suggesting that the presence of unidimensionality ensures local independence [27,32]. In this study, local independence was tested by calculating Yen’s Q3 statistics through the residual correlation generated by the CFA model. A Q3 >0.3 is considered a possible local dependence of the test [33,34]. To evaluate the extent of monotonicity violations, we used the Mokken package [35], considering a ratio of violations to acceptable items (#vi/#ac) ≤0.3 and a Diagnostic Crit Value (crit) ≤80 as indicators of no significant monotonicity violations. Additionally, a visual inspection of the rest-score graphs was also conducted to assess the monotonicity of each item.
After confirming that the IRT assumptions were tenable, the Mirt package was used to fit the IRT models and estimate the discrimination, difficulty, and maximum item information for each item [36].
For fitted IRT parameters, we calculated Kendall’s W as a measure of rater consistency across scales, specifically to assess whether the different wording of these symptoms is essentially in agreement with one another. Moreover, to facilitate cross-instrumental comparisons, we ordered the IRT parameters (discrimination, difficulty, and information) and calculated the mean rank of these parameters for each symptom. Note that, since the GRM provides multiple b values for each item, we first calculated the average rank of each item within its scale (for example, for item #1 in GAS, its within-scale average rank is the arithmetic mean of the ranks of its four b values, b1–b4). Then, the cross-scale average rank was calculated based on the within-scale average rank of each item.
Differential Item Functioning (DIF) analysis was carried out to test the measurement invariance of these criteria. Participants were categorized by sex (males vs. females) and age group (adolescents who were less than 18 years old vs. adults who were 18 years old or above). The DIF test was executed using the generalized logistic regression method in the difR package [37]. Nagelkerke’s R2 was utilised as a measure of the DIF effect. According to the criteria proposed by Jodoin and Gierl [38], R2<0.035 is considered a small and negligible DIF effect.
RESULTS
Demographic information and correlations
Among the 1,722 subjects surveyed, 1,530 valid questionnaires were collected, resulting in a response rate of 88.85%. The final sample included 685 males (44.77%) and 845 females (55.23%). Of these participants, 619 participated in the online survey (40.46%) and 911 participated in the offline survey (59.54%). The mean age was 18.35±3.66 years, ranging from 11 to 30 years old, with 826 adults and 704 adolescents.
The four scales demonstrated high (with a large effect size according to Cohen’s guidelines [39]) positive correlations with each other (r≥0.502, p<0.001), indicating high consistency across the scales.
IRT assumptions tests
Our analyses confirmed that the data from all four scales met the model assumptions of unidimensionality (Supplementary Table 2), local independence (max absolute Q3 for GAS, IGDT-10, IGSF9-SF, and IGDS were -0.268, -0.287, -0.235, and -0.198 separately), and monotonicity (#vi/#ac ≤0.3 and crit ≤80 for all items). Therefore, it is appropriate to conduct IRT analysis on these data.
IRT results
Discrimination parameter
For the discrimination parameter of each DSM-5 criterion, we found a significant cross-scale consistency (Kendall’s W=0.696, p=0.033, df=8). This suggests that, despite variations in wording across different scales, the items related to the same symptom exhibited essentially consistent discrimination parameters.
According to the standard proposed by Baker [16], a discrimination parameter equal to 0, 0.01–0.34, 0.35–0.64, 0.65–1.34, 1.35–1.69, and >1.70 indicates non, very low, low, moderate, high, very high, and perfect discriminative power, respectively. Based on this classification, all nine criteria demonstrated very high or perfect discriminative power, with values ranging from 1.380 to 3.721. Among the nine DSM-5 criteria, escape exhibited the lowest cross-scale discrimination, whereas withdrawal showed the highest discrimination, followed by tolerance and loss of control (Table 3).
Discrimination parameters of each Internet Gaming Disorder diagnostic criterion
Difficulty parameter
Similar to discrimination, we observed a significant cross-scale consistency for the mean rank of difficulty parameters (Kendall’s W=0.656, p=0.046, df=8), indicating an essentially consistent pattern of difficulty across scales.
As shown in Table 4, escape exhibited the lowest level of difficulty compared to other criteria across scales. Conversely, deception demonstrated the highest difficulty among the criteria, followed by withdrawal and give up other activities.
Cross-scale ranks of difficulty for each Internet Gaming Disorder diagnostic criterion
Item information
We found significant cross-scale consistency for the item information parameter (Kendall’s W=0.719, p=0.028, df=8), indicating an essentially consistent pattern of item information across scales.
The nine criteria exhibited item information ranging from 0.476 to 4.168 (Table 5). Similar to the discrimination parameter, escape had the lowest item information, while withdrawal had the highest, followed by tolerance and loss of control.
Item information of each Internet Gaming Disorder diagnostic criterion
DISCUSSION
The present study aimed to examine the psychometric properties of the nine criteria of IGD through IRT analyses of DSM-5-based scales. All nine DSM-5 criteria demonstrated high discrimination, providing substantial item information for diagnosing IGD. The used scales, despite varying in wording and scoring methods, exhibited strong correlations in total scores (r≥0.502, p<0.001) and high consistency (Kendall’s W ≥0.656, p<0.05) concerning the fitted IRT parameters. Among the nine criteria, escape showed the lowest cross-scale discrimination, item information, and difficulty. In contrast, withdrawal showed the highest item information and discrimination, and the second highest difficulty. Additionally, there was no evidence of DIF related to sex or age (adolescents vs. adults) across nearly all scale items. Refers to the accuracy of the criterion in different latent traits [27].
All item discrimination values exceeded 1.35 (1.38–3.72), indicating very high to perfect discrimination according to Baker’s standard [16]. This finding indicates that all nine criteria, despite variation in difficulty and item information, can correctly distinguish respondents with and without IGD [5,17]. Moreover, our DIF analyses suggest that these criteria can be applied fairly in different sex and age groups. Together, our results seem to support validating the nine DSM-5 criteria of IGD.
Importantly, we found that despite variations in wording and scoring methods, these scales exhibited high consistency in the fitted IRT parameters (Kendall’s W ≥0.656). These results seem to indicate that each diagnostic criterion could possess a certain degree of tolerance. In other words, what is measured does not have to align precisely with the core pathological changes described by the symptoms, but rather with related variations. This consistency among scales also suggests that, at least for these nine symptoms, their relative importance, in relation to the “core” pathological changes of IGD, may be essentially stable. For instance, the distance of escape from the “core” of IGD could be greater than that of withdrawal. This may be more intuitively illustrated through network analysis [40]. Moreover, all four scales met the unidimensionality assumption, which not only indicates suitability for IRT analysis but also suggests that IGD likely has only one “core” pathological change. This conclusion is consistent with existing research on individual scales [5,8,17,25]. Therefore, our results seem to indicate that IGD has a singular core pathological change, encompassing the aspects described by the nine symptoms, and the relative importance of these symptoms according to the core pathological change is likely stable across scales.
The escape criterion exhibited the lowest discrimination, difficulty, and item information among all nine diagnostic criteria for IGD. These results align with multiple previous IRT studies, which found that escape was less effective than other criteria in distinguishing between individuals with IGD and those who are healthy [5,17,41]. Our results expanded these findings by demonstrating that the escape performs poorly across different scales with varying wording and scoring methods, i.e., IGDS9-SF, IGDT-10, IGDS, and GAS, suggesting a cross-scale generalization of this finding. Indeed, studies using different instruments or methods have also reported low diagnosis accuracy of the escape criterion. For example, Ko et al. [13] evaluated the diagnostic validity of the nine DSM-5 criteria and the criteria of craving and irritability based on the Diagnostic Criteria of Internet Addiction for College Students, and found that, except for the deception and escape criteria, all DSM-5 criteria had diagnostic accuracy. This finding was reiterated in a study conducted in 2020 [14]. Likewise, in a Delphi study, 29 international experts on IGD rated the escape and tolerance as inadequate to distinguish problematic from normal gaming [42]. Additionally, a factorial analysis on data based on the Compulsive Internet Use Scale indicated that the item on escape had no association with the other items [43], a finding echoed in a recent IGDS factor analysis [44]. Together with these findings, our results suggest that the underperformance of escape in IRT is likely to be due to the definition of this criterion itself, rather than to differences in the wording of different questionnaires. The escape could be controversial for several reasons. Firstly, escape is one of the main motivations of game-playing in general gamers, but not just addicted ones [41,45]. A systematic review suggested that escapism and avoidance coping represent both a motivational factor mostly associated with disordered gaming [46]. Secondly, IGD is often co-morbid with depression [47], and escapism may also occur under a depressed mood. Existing studies have demonstrated that depressed mood may be a mediator between IGD and other psychological factors, such as mood control [48] and causality orientations [49]. Kim et al. [50] found that escape was the only significant item associated with depression among symptoms of IGD, and underscored this connection. However, it should be stressed that the escape criterion demonstrated acceptable levels of discrimination, difficulty, and item information in diagnosing IGD, suggesting its relevance to addiction-related pathological changes. In line with this perspective, mood modification (including “high” and “escape”) was considered one of the core components of addiction [51]. In summary, although the escape criterion met the minimum statistical threshold for acceptability on some metrics, its consistently inferior performance relative to all other criteria across four different instruments raises significant concerns about its utility and validity.
Contrary to escape, withdrawal showed the highest discrimination and item information, and the second highest difficulty among the nine criteria. These results were similar to several previous IRT studies, highlighting the significant role of withdrawal in diagnosing IGD [19,41]. Indeed, multiple studies have demonstrated that withdrawal is a core presentation in IGD [52,53]. A recent network analysis of IGD symptoms suggested that withdrawal, together with loss of control, persistence, and tolerance, may form the core of the network [40]. Our results align with these studies, suggesting that withdrawal could reflect the core pathology of IGD, which can provide more information, better discrimination, and, with high reliability, distinguish between IGD and healthy individuals. Withdrawal may be central to addiction for two reasons: 1) Withdrawal is a diagnostic symptom in all existing addictive disorders, including substance use disorder, gambling disorder, and IGD [1], highlighting its commonality in addiction. 2) The significance of withdrawal is emphasized in the negative reinforcement theory of addiction [54], which suggests that negative reinforcement driven by withdrawal symptoms plays a crucial role in the development and maintenance of addiction. While some debate whether withdrawal exclusively drives addiction [55], few studies dispute its importance. Thus, withdrawal may represent a core symptom of IGD. Importantly, the significant role of withdrawal in IGD contributes to ongoing discussions about the relevance of substance-use criteria in behavioural addictions [11], suggesting that chemical consumption—and related physical discomfort—might not be central to withdrawal symptoms. The typical withdrawal symptoms in IGD are irritability, restlessness, boredom, anhedonia, and gaming craving [52,56]; physical withdrawal symptoms are rarely reported [57]. However, this does not mean that IGD can be easily quiet. In fact, the irritability, anxiety, and boredom resulting from forced abstinence are precisely the reasons many individuals with IGD seek treatment. Most participants (84%) of IGD indicated that they had to play games because of these withdrawal symptoms [52]. However, despite its widespread recognition, there is a lack of qualitative studies providing detailed clinical descriptions of this symptom.56 Further studies with more specific qualitative descriptions, maybe with more objective measurements such as heart rate, neuroimaging techniques, are warranted [52].
In addition to withdrawal, tolerance and loss of control also showed high item information and discrimination. In line with our results, an IRT study of IGDS-SF, where withdrawal, tolerance, and persistence, along with preoccupation, were identified as the strongest discrimination items [41]. Our results also align with the findings of Gomez et al. [40], which suggested that withdrawal, persistence, tolerance, and loss of control form the core of the network of IGD symptoms. In sum, our results indicate that withdrawal, tolerance, and loss of control could play a significant role in distinguishing between individuals with IGD and those who are healthy.
This study has some limitations. First, our sample covered a small age interval of 11–30 years old, given that most IGD occurs during adolescence and early adulthood. This sample does not allow for generalization to another age interval, such as the middle-aged. Secondly, our data were derived from self-report scales, while DSM-5 diagnoses are typically administered through clinical interviews. Future studies directly assessing the DSM-5 criteria using data from clinical interviews or evaluation scales (e.g., the Chinese IGD scale) may provide more information on the diagnostic efficacy of these criteria. Thirdly, to facilitate cross-scale comparisons of IRT parameters, we combined items 9 and 10 of the IGDT-10 by taking the maximum score of these two items. Additional IRT analyses with items 9 and 10 combined and treated separately demonstrated good fit for both methods (Supplementary Tables 3 and 4). Furthermore, the combined item exhibited comparable discrimination and item information, but slightly lower difficulty (ranges of b value changed from 0.659–2.685 to 0.554–2.010) (Supplementary Table 5). This combination could make the criterion of negative outcome easier to endorse in the IGDT-10.
Conclusions
This study contributed to the debate regarding the usefulness and validity of the nine DSM-5-based IGD criteria. Our cross-scale IRT analysis indicates that the nine DSM-5 criteria generally possess acceptable psychometric properties for diagnosing IGD. Notably, the criterion of withdrawal may represent a core symptom of IGD. Conversely, the criterion of escape demonstrated inferior performance compared to the other eight criteria, indicating a potential need for further revision. These conclusions may be generalised to different self-reported instruments of IGD and to respondents of both genders, as well as to both adolescents and adults.
Supplementary Materials
The Supplement is available with this article at https://doi.org/10.30773/pi.2025.0082.
Measurement invariance analysis
Measurement invariance results for online and offline subsample
Confirmatory factor analysis results of scales
Global absolute fit statistics when items 9 and 10 were combined or not combined
Global relative fit statistics when item 9 and 10 were combined or not combined
Item response theory parameters of items in IGDT-10 when items 9 and 10 were combined or not combined
Notes
Availability of Data and Material
The datasets used in the current study are available from the corresponding author upon reasonable request.
Conflicts of Interest
The authors have no potential conflicts of interest to disclose.
Author Contributions
Conceptualization: Wei Lei. Data curation: all authors. Formal analysis: Luoya Zhang, Xingzi Lu. Funding acquisition: Jing Chen. Investigation: Yanjie Peng, Wenyuan Wang, Zhen Zeng, Juan Deng, Shuang Feng, Yuxiang Wang, Maomao Zhang, Yuwen Chen. Methodology: Wei Lei. Project administration: Jing Chen. Writing—original draft: Wei Lei, Luoya Zhang, Xingzi Lu. Writing—review & editing: Wei Lei, Jing Chen, Kezhi Liu, Ke Gong.
Funding Statement
This work was supported by the National Science Foundation of China (32200882); Sichuan Science and Technology Department (23ZDYF2557); the joint project of Hejiang people’s Hospital & Southwest Medical University (2022HJXNYD13, 2021HJXNYD16); and the joint project of Luzhou Science and Technology Bureau & Southwest Medical University (2019LZXNYDJ39); Sichuan Applied Psychology Research Center (CSXL-22102, CSXL-212A17); Joint Innovation Project of Sichuan Provincial Science and Technology Plan (2022YFS0616); Southwest Medical University (2022ZD004, YJG202289); and Social Sciences Federation of Southwest Medical University (SMUSS202220).
Acknowledgments
The authors would like to thank all participants who contributed to the study.