The internet is a convenient technology that may be used across many devices, including tablets, laptops, smartphones, and desktop computers. The internet can be reached in various ways, including cable and Wi-Fi, which may facilitate use for leisure, work, social interactions, and entertainment (Nurmala et al., 2022). Although the purpose of the internet and related devices (e.g., smartphones) is designed to promote better lives for people, a minority of individuals exhibit problems of internet addiction (IA) or other associated behavioral addictions, such as smartphone addiction and social media addiction (K.-C. Chang et al., 2022; Saffari et al., 2022a, b; Xu et al., 2022). Multiple health concerns (e.g., musculoskeletal discomfort, sleep difficulties, and mental health problems) have been associated with IA (Alimoradi et al., 2019; Huang et al., 2023; Kukreti et al., 2023; Wong et al., 2020; Yang et al., 2017). Such health problems may relate to hardware features of a device (e.g., blue light in smartphones) (Wong et al., 2020), time spent on the internet leading to impairments in functioning (Kamolthip et al., 2022; Kukreti et al., 2021; Lin et al., 2021a, b; Liu et al., 2022), or other factors.

IA may be classified into generalized and specific forms (Chen et al., 2020a, b; Leung et al., 2020). Both forms often share similar features such as affected individuals having strong desires or cravings focused on internet use (Chen et al., 2020a, b; Lin et al., 2019). However, the two types of IA also differ. Specifically, generalized IA usually refers to the addiction toward a medium (e.g., internet) or device (e.g., smartphone) used for engagement (Chen et al., 2020a, b; Leung et al., 2020). Specific forms involve distinct activities such as gaming or the use of social media (Kamolthip et al., 2022; Montag et al., 2015). Given that different activities may be differentially associated with measures of poor health (Wong et al., 2020), it is important for healthcare providers to understand how each specific IA relates to different health domains.

The present study focused on one type of specific IA: social media addiction. Social media addiction involves excessive and interfering involvement with specific platforms (e.g., Facebook, Twitter, WhatsApp) designed for interactions with other people (Kamolthip et al., 2022; Montag et al., 2015). A widely used instrument for assessing social media addiction is the Bergen Social Media Addiction Scale (BSMAS), modified from the Bergen Facebook Addiction Scale (Cecilie Schou Andreassen et al., 2016). The BSMAS was developed using a theoretical framework (i.e., the component model of addiction; Griffiths, 2005) and is a validated instrument with promising psychometric properties (Chen et al., 2020a, b; Leung et al., 2020). However, it assesses general social media and does not consider specific features of different social media. For example, most social media like Facebook possess features of personal profiles and interactions with real-life friends or other unknown/less well-known individuals (de Bérail et al., 2019). YouTube does not have the feature of social networking functionalities but has features of content viewing (Balakrishnan & Griffiths, 2017; Daria J Kuss & Griffiths, 2017). Indeed, YouTube is viewed as a video streaming application and thus is likely to have a different construct of general social media like Facebook. In this regard, assessing specific social media addiction (e.g., YouTube addiction) may provide in-depth information for healthcare providers to make clinical decisions.

We were especially interested in YouTube use for the following reasons. First, YouTube is a popular site with the following characteristics: (i) having over 2.5 billion people using YouTube worldwide; (ii) 95% of teenagers use YouTube (the second most frequently used similar site is TikTok at 67%); and (iii) the most visited website (Geyser, 2023). Second, YouTube has a monetization program called the YouTube Partner Program (YPP) that allows content creators to earn money from their videos through advertisements (YouTube, 2023). This has created a competitive environment among YouTube participants to produce engaging or “viral” content that will attract more views and generate more revenue. The desire to become successful and earn money through YouTube may contribute to addictive engagement and excessive use of the platform. Some other social media have similar programs. For example, Twitch has a “Twitch Partner Program”; Facebook has a “Facebook Gaming Partner Program”; and Instagram and TikTok have their own creator funds that pay popular creators to produce content on their platforms. However, the YPP impacts are possibly greater because of the large number of people using YouTube. Additionally, the most important feature of YouTube is its video functions.

Given the specific features of YouTube, the general assessment of social media addiction may not apply to YouTube addiction. Accordingly, some researchers have studied YouTube addiction, using various scales to assess YouTube addiction and relate it to measures of social and academic functioning (Balakrishnan & Griffiths, 2017; Chan et al., 2022; de Bérail et al., 2019; Moghavvemi et al., 2017) . Some valid instruments assessing YouTube addiction have been developed, including Problematic YouTube Use Scale and the Internet Addiction Test—YouTube use Adaptation (Rahat et al., 2022). However, the Internet Addiction Test—YouTube use Adaptation contains 20 items (de Bérail et al., 2019), which may not be efficient for healthcare providers or researchers to quickly assess YouTube addiction. The Problematic YouTube Use Scale contains only six items and is developed using the component model of addiction (Kircaburun et al., 2021). However, the Problematic YouTube Use Scale was developed by simply changing the term “Facebook” from the Bergen Facebook Addiction Scale to “YouTube.” Therefore, using a more rigorous methodological method (e.g., involving expert committees and item generation process) to develop the YouTube Addiction scale is needed.

The study aimed to develop a new instrument to assess YouTube addiction; namely, the YouTube Addiction Scale (YAS). Apart from developing the YAS with the use of the component model of addiction, psychometric testing including advanced psychometric methods (e.g., Rasch models) was used to examine the properties of the YAS. We hypothesized that the YAS would be psychometrically sound in terms of its internal consistency, unidimensionality, network analysis results, and Rasch findings. A secondary aim of the present study was to use the YAS to classify people with different levels of YouTube addiction and to test if the different levels of YouTube addiction were associated with general social media addiction, psychological distress (including depression, anxiety, and stress), age, gender, and weekly hours on YouTube.

Methods

Participants and Recruitment Procedure

In this cross-sectional study, a convenient sampling approach was used to collect data from people using online social media. To be eligible for inclusion, participants needed to be at least 13 years old and use YouTube. All adult participants provided their online consent. Adolescents and one of their parents provided online consent before participation in the study. The study protocol was approved by the ethics committee of Qazvin University of Medical Sciences (IR.QUMS.REC.1401.097). Moreover, in order to thoroughly examine the factor structure of the YAS, two samples were collected: the first sample was collected for exploratory factor analysis (EFA), and the second sample for confirmatory factor analysis (CFA). For the procedure of the EFA and CFA, please see the “Data analysis” section for details. Moreover, the first sample (N = 530) completed the YAS together with another two measures (i.e., BSMAS and Depression, Anxiety, Stress Scale-21; please see the “Measures” section for details) and demographic information. The second sample (N = 512) only completed the YAS and demographic information.

The first sample included 530 participants with a mean age of 23.63 years (SD = 8.1 years). Nearly 20% of the sample was adolescent (n = 103; 19.4%) and approximately half was female (n = 268; 50.6%). Weekly YouTube use was 9.49 h (SD = 6.83 h). The second sample included 512 participants with a mean age of 26.00 years (SD = 11.45 years). Nearly half of the sample was adolescent (n = 233; 45.5%) and slightly less than half was female (n = 287; 56.1%). Weekly YouTube use was 11.27 h (SD = 5.54 h).

Measures

Development of the YouTube Addiction Scale (YAS)

The YAS was developed based on Griffiths’ component model of addiction (Griffiths, 2005). In the component model of addiction, addictive behavior is defined as having six core components, including salience (individuals consider that YouTube engagement is the most important daily activity), mood modification (individuals use YouTube to cope with mood difficulties), tolerance (individuals have used YouTube increasingly frequently to get the same desired effect over time), withdrawal (individuals feel unpleasant and somatic problems in the immediate setting of not using YouTube), conflict (individuals have personal or interpersonal conflicts with continued YouTube engagement), and relapse (individuals revert back to the previous levels of YouTube engagement when attempting to reduce the use of YouTube) (Daria J. Kuss et al., 2014). A series of questions related to each component were then developed to measure the severity of problematic use of YouTube. Two rounds of expert panel sessions were conducted to discuss the potential items. The expert team members included a psychologist, pediatrician, psychometrician, health psychologist, and nurse. The expert team members revised or omitted several items. Finally, six questions were retained. A cognitive interview with 10 adolescents and 5 young adults was then conducted to examine the validity of the YAS. We proceeded with having expert panel sessions before cognitive interview in the present study is because the committee members were all experts in this field and they are very familiar with the used theory (i.e., component model of addiction). In this regard, we could save the time in item generation and the cognitive interview were conducted to ensure the readability of the generated items.

The final YAS included one item from each component. The first question asks about the degree to which watching YouTube videos has subsumed or replaced other activities of daily life, which is related to the component of salience. The second question assesses the increased use of YouTube over time to achieve the same level of satisfaction, which is related to the component of tolerance. The third question asks if YouTube videos are used to alleviate negative feelings, which is related to the component of mood modification. The fourth question assesses whether individuals have tried to reduce their time spent viewing YouTube but failed, which is related to the component of relapse. The fifth question asks if individuals have felt restless or worried if they cannot access YouTube videos, which is related to the component of withdrawal. Finally, the sixth question assesses whether individuals have experienced problems with work, school, university, friends, or family due to watching YouTube videos, which is related to the component of conflict. Regarding responses, the YAS uses a five-point Likert-type scale to score items, with “never” corresponding to 1 and “very often” corresponding to 5. Thus, a higher YAS score indicates a greater addiction severity.

Bergen Social Media Addiction Scale (BSMAS)

The BSMAS contains six items to assess severity of addiction to social media (Cecilie Schou Andreassen et al., 2016). A sample item of the BSMAS is, “How often during the last year have you used social media to forget about personal problems?” All items were rated using a five-point Likert scale (1, very rarely; 5, very often) and higher BSMAS scores reflect greater addiction severity. The BSMAS has been validated in different language versions, including Persian (Lin et al., 2017), with satisfactory psychometric properties.

Depression Anxiety Stress Scale-21 (DASS-21)

The DASS-21 contains 21 items assessing psychological distress related to depression (seven items), anxiety (seven items), and stress (seven items) (Lovibond, 1995). A sample item of the DASS-21 for depression is, “I found it difficult to work up the initiative to do things”; for anxiety is, “I was aware of dryness of my mouth”; and for stress is, “I found it hard to wind down.” All items were rated using a four-point Likert scale (0, did not apply to me at all; 3, applied to me very much or most of the time), and higher DASS-21 scores reflect greater severity of depression, anxiety, or stress. The DASS-21 has been validated in different language versions including Persian (Kakemam et al., 2022), with satisfactory psychometric properties.

Demographics

Apart from the external criterion instruments, the participants were asked to provide demographic information, including their age (in years) and sex (male or female), and weekly YouTube use (in hours).

Data Analysis

In addition to the descriptive statistics, the present study used EFA, CFA, internal consistency, Rasch analysis, and network analysis to evaluate the psychometric properties of the YAS. The EFA, internal consistency, Rasch analysis, and network analysis were performed using the first sample (N = 530); the CFA and another set of internal consistency were performed using the second sample (N = 512). For descriptive statistics on the YAS items, skewness and kurtosis were checked. When a YAS item had skewness ranging between − 3 and 3 together with kurtosis ranging between − 8 and 8, the item was deemed to have an acceptable normal distribution (Kline, 1998; Lin et al., 2013).

For EFA, the extraction method of principal axis factoring with Promax rotation was used, and a factor was considered to exist when its eigenvalue was larger than 1 (i.e., Kaiser’s rule) (Kaiser, 1960). Moreover, the factor loadings in the EFA larger than 0.4 indicated satisfactory factor loading for each item (Maskey et al., 2018). For CFA, the diagonally weight least squares estimator was used to reexamine the factor structure found from the EFA results. Fit indices of comparative fit index (CFI)>0.9, Tucker-Lewis index (TLI)>0.9, standardized root mean square residual (SRMR) together with a nonsignificant χ2 indicate good data model fit. After confirming the factor structure of the YAS, measurement invariance of the YAS was examined for the following subgroups: gender (male vs. female), age (below mean age vs. above mean age), and weekly YouTube use (below mean use vs. above mean use). In the measurement invariance test, four nested models were constructed: Model 1 as a configural model assuming subgroups sharing the same YAS factor structure; Model 2 as a metric model constraining factor loadings to be equal across subgroups; Model 3 as a scalar model constraining factor loadings and item intercepts to be equal across subgroups; Model 4 as a strict model constraining factor loadings, item intercepts, and residuals to be equal across subgroups. The four nested models were compared using ∆CFI, ∆SRMR, and ∆RMSEA, where ∆CFI > 0.02 together with both ∆SRMR and ∆RMSEA < 0.02 indicating invariance across subgroups.

For internal consistency, Cronbach’s α was used, with values larger than 0.7 suggesting satisfactory internal consistency (Taber, 2018). Moreover, a corrected item-total correlation for each YAS item was computed, with values larger than 0.4 considered satisfactory (Poon et al., 2021).

For Rasch analysis, a partial credit model was applied, and each item was evaluated using inlier-sensitive (infit) mean square (MnSq) and outlier-sensitive (outfit) MnSq. When an item has both infit and outfit MnSq ranging between 0.5 and 1.5, the item was considered to be fit (Lin et al., 2021a, b; Saffari et al., 2022a, b). Apart from the infit and outfit MnSq, each item was assessed regarding their differential item functioning (DIF) across different subgroups: age (23.6>vs. 23.6 ≤), gender (male vs. female), and weekly YouTube use (9.49 > vs. 9.49 ≤). When an item had a DIF contrast across the subgroups less than 0.5, the item was considered not to have substantial DIF concerns (Lin et al., 2020; Nejati et al., 2021). Also, the discrimination property of each YAS item was estimated using the Rasch model. Apart from the item-level properties for YAS using Rasch analysis, Rasch analysis was also used for its scale-level properties. Specifically, Rasch separation reliability (including item separation and person separation) larger than 0.7 and Rasch separation indices (including item separation and person separation) larger than 2 were considered suggestive of good properties of the YAS (K.-C. Chang et al., 2014). Moreover, the variance explained by the Rasch factor (i.e., > 50%) and the eigenvalue of the first residual factor from the Rasch model (i.e., < 2) were used to examine if the YAS had a unidimensional factor structure (C.-C. Chang et al., 2015). For the network analysis, the EBICglasso (Extended Bayesian Information Criterion Graphical LASSO) algorithm with 500 bootstrapping was used to conduct the network analysis.

Afterward, the latent class analysis (LCA) was conducted to identify how different subgroups of individuals with YouTube addiction differed on measures of stress, anxiety, depression, social media addiction, time spent using YouTube, age, and sex. The number of subgroups was identified using the following statistics: Akaike information criterion (AIC), Bayesian information criterion (BIC), sample-size adjusted BIC (SSABIC), entropy, and bootstrap likelihood ratio test (BLRT). The smallest AIC, BIC, and SSABIC together with the largest entropy and significant BLRT were preferred.

Results

Item Properties of the YAS Using the First Sample

The item properties were used to understand the distribution and fit of each YAS item. Regarding the YAS item properties in classical test theory, all six items had strong factor loadings in EFA (range between 0.651 and 0.884), good item-total correlations (range between 0.625 and 0.833), and relatively normal distribution (skewness range between 0.80 and 1.97; kurtosis range between − 0.46 and 2.99). Regarding the YAS item properties in Rasch analysis, all six items had adequate infit MnSq (range between 0.73 and 1.39) and outfit MnSq (range between 0.60 and 1.23). There was no substantial DIF across age (range between − 0.46 and 0.29), sex (range between − 0.10 and 0.10), and weekly time spent on YouTube (range between − 0.35 and 0.32). Additionally, the six items shared similar discrimination properties between 0.58 and 1.30 (Table 1).

Table 1 Psychometric properties of the YouTube Addiction Scale (YAS) at the item level
Full size table

Scale Properties of the YAS Using Both the First and the Second Samples

The scale properties of the YAS help understand the YAS factor structure (i.e., if the YAS is a unidimensional structure). From the data of the first sample, the internal consistency of YAS was good (0.912) and all six YAS items loaded on the same construct (eigenvalue of the first factor = 4.25; eigenvalue of the second factor = 0.57) in the EFA. Moreover, Rasch separation reliability was good in both item separation (0.98) and person separation (0.70). The Rasch separation index was satisfactory in item separation (6.49) and relatively good in person separation (1.52). The variance explained by the Rasch factor was good (62.5%), with little residual (eigenvalue of the first residual factor = 1.49) (Table 2). From the data of the second sample, the CFA results confirmed the unidimensional structure of the YAS. Specifically, the internal consistency values were acceptable (0.852) and the fit indices of CFA were all good (CFI = 0.990, TLI = 0.984, RMSEA = 0.059, and SRMR = 0.051), except for the significant χ2 test (χ2 = 25.198; df = 9; p < 0.05). Moreover, measurement invariance of the YAS was supported to be across genders (ΔCFIs >  − 0.02; ΔSRMRs < 0.02; ΔRMSEAs < 0.02) (Table 3).

Table 2 Psychometric properties of the YouTube Addiction Scale (YAS) at the scale level
Full size table
Table 3 Measurement invariance of the YouTube Addiction Scale (YAS) across gender (male vs. female), age (below mean age [< 26.00 years] vs. above mean age [> 26.00 years]), and weekly YouTube use (below mean use [< 11.27 h] vs. above mean use [> 11.27 h])
Full size table

Network Analysis of the YAS Using the First Sample to Identify YAS Construct Validity

Network analysis showed a clear pattern between the YAS and the BSMAS, indicating that the YAS has a different construct from the BSMAS (Fig. 1). The strong association between item 4 from the YAS scale and item 5 from BSMAS suggests that both items may be measuring related aspects of problematic social media use. Specifically, item 4 from YAS (“Have you tried to reduce your YouTube watching time but failed?”) measures related to difficulty in reducing or stopping YouTube use, while item 5 from BSMAS (“Do you get restless or troubled when you attempt to cut down on your use of social media?”) measures withdrawal symptoms when attempting to decrease social media use. The strong association between these two items indicates that individuals who experience relapse related to their YouTube use may also withdraw when attempting to reduce their social media use.

Fig. 1
figure 1

EBICglasso model based on network analysis according to the Internet Addiction scale among 530 participants. Note: Social 1 to 6 are six items of the Bergen Social Media Addiction Scale. Y1 to Y6 are six items of the YouTube addiction

Full size image

This network results consisted of 12 nodes, representing the items from the YAS and BSMAS scales, and 47 non-zero edges (Fig. 2), indicating the strength of the associations between these items. The sparsity of the network is 0.288 (Fig. 3), indicating that there are still some weak connections between nodes that were not statistically significant enough to be included as edges (Fig. 4).

Fig. 2
figure 2

Centrality Plots for EBICglaaso network depicting the betweenness, closeness, and degree (strength) of each node (variable) among 530 participants. Note: Social 1 to 6 are six items of the Bergen Social Media Addiction Scale. Y1 to Y6 are six items of the YouTube addiction

Full size image
Fig. 3
figure 3

Edge stability

Full size image
Fig. 4
figure 4

Centrality stability

Full size image

The strong association between item 2 from YAS (“How much time do you spend watching videos on YouTube that you didn’t plan for?”) and item 3 from BSMAS (“Do you spend a lot of time thinking about social media or planning to use social media?”) could indicate that individuals who engage in excessive use of YouTube may also spend a lot of time thinking about and planning their social media use. This finding highlights the potential overlap between problematic use of YouTube and other forms of social media addiction.

LCA of the YAS Using the First Sample to Identify YAS Known Group Validity

The LCA results suggested that there should have two subgroups (AIC = 6033.797, BIC = 6243.168, SSABIC = 6087.628, entropy = 0.961, and significant BLRT [p < 0.01]). The differences of the two subgroups are presented in Table 4. Specifically, the two groups (n = 126 for the Cluster 1 group; 404 for the Cluster 2 group) were compared showing significant differences in all YAS item scores, BSMAS total score, depression score, anxiety score, stress score, age, and weekly hours on YouTube use (p < 0.033). No significant differences were found between the sexes (p = 0.534; Table 4).

Table 4 Bivariate associations between cluster membership, psychological distress, sociodemographic factors, and YouTube addiction items (n = 530)
Full size table

Discussion

The present study used the component model of addiction proposed by Griffiths (2005) to design the six-item YAS assessing YouTube Addiction, proposed as a specific type of social media addiction (Balakrishnan & Griffiths, 2017; Chan et al., 2022; de Bérail et al., 2019; Moghavvemi et al., 2017). After designing the six-item YAS, several psychometric testing methods were used to examine if the YAS possesses good properties. The findings consistently indicated that the YAS is a valid instrument for assessing YouTube Addiction. For example, the internal consistency was high, the unidimensionality was supported, every YAS item had strong factor loadings and good item properties, no DIF was displayed, and all YAS items were coherent with each other as shown by the network analysis. Moreover, the two groups with different levels of YAS had different levels of general social media addiction (measured using the BSMAS), psychological distress (measured using the DASS-21), and weekly hours on YouTube.

The good psychometric properties of the YAS found in the present study may be explained by the guided theoretical framework (i.e., the component model of addiction (Griffiths, 2005). The component model of addiction provides clear definitions of each component contributing to behavioral addiction and has been widely used in the field of behavioral addictions, including social media addiction (Cecilie Schou Andreassen et al., 2016; Jameel et al., 2019; Varona et al., 2022). In addition to the clear definitions, many instruments assessing behavioral addictions report strong evidence regarding the utility of the component model of addiction in explaining various behavioral addictions. That is, the BSMAS, Bergen Shopping Addiction Scale, and Smartphone Application-Based Addiction Scale developed using the component model of addiction all show satisfactory psychometric properties (Cecilie Schou Andreassen et al., 2016; C. S. Andreassen et al., 2015; Stănculescu, 2022; Tung et al., 2022; Yam et al., 2019). Therefore, the satisfactory psychometric properties of the YAS found in the present study echo the prior instruments developed using the component model of addiction.

Because YouTube addiction may be considered a specific behavioral addiction, the development of the YAS can help researchers and healthcare providers to obtain in-depth knowledge associated with YouTube addiction. Specifically, how the program YPP promoted by YouTube may trigger or promote YouTube addiction may be investigated. In addition, the YAS may be used to replicate and extend prior findings assessing associations between YouTube addiction and other phenomena like psychosocial health and academic performance (Balakrishnan & Griffiths, 2017; Chan et al., 2022; de Bérail et al., 2019; Moghavvemi et al., 2017).

The present findings of the LCA involving the YAS showed that the present sample could be classified into two groups: one having a higher level of YouTube addiction and another having a lower level of YouTube addiction. The group with more severe YouTube addiction had significantly greater social media addiction than that with less severe YouTube addiction, and this result resonates with prior findings (Geyser, 2023). Also similar to prior findings, we found that the group with more severe YouTube addiction had greater psychological distress (de Bérail et al., 2019) and was younger (Geyser, 2023) than the group having a low level of YouTube addiction. The aforementioned findings also support the utility of the YAS for assessing different research topics.

Study limitations warrant mention. First, the present sample was Iranian. Therefore, the generalizability of the present findings may be limited. Future studies are needed to verify if the YAS can be translated into other language versions while retaining robust psychometric properties. Second, the YAS was not examined for its test–retest reliability and responsiveness. Therefore, it is unclear if the stability and sensitivity of the YAS are satisfactory . Future studies may want to investigate these properties to strengthen the utility of the YAS. Third, the nonstandard procedure of item generation (i.e., lack of cognitive interviews before committee members’ discussion and the use of few item numbers for exploratory factor analysis) is a limitation. However, given that the psychometric results in the present study across two samples were all satisfactory, we consider that this may not be a serious limitation. Lastly, all measures used in the present study were self-reported. Therefore, social desirability bias and single-rater bias may have influenced the findings.

Conclusion

In conclusion, the YAS has promising psychometric properties for assessing the severity of YouTube addiction. With the use of YAS, healthcare providers may monitor individuals’ YouTube addiction and provide early intervention as indicated, and this should be examined directly in future studies. Moreover, researchers could use the YAS to investigate different topics. However, additional psychometric evidence, such as verifying its unidimensionality and validating its test–retest reliability, is needed for the YAS. In addition, different language versions of the YAS are needed to promote the comparisons of YouTube addiction across countries.