Convenience sampling was adopted to recruit participants from junior high schools in Northeast China. This study followed and met the ethical requirements of the researchers’ institutions. A total of 856 online questionnaires were completed by Chinese junior school students, whom, along with their parents, gave their informed consents to participate in the study. Among the collected responses, 748 were kept for analysis (87.38% effective rate) after excluding those questionnaires that were deemed invalid for various reasons (e.g., choosing the same response for all questions or leaving too many missing values). A total of 382 (51.07%) male and 366 (48.93%) female participants comprised the sample. These participants were aged between 12 and 15 years (M = 12.86; SD = 0.77). In terms of educational level, those participants in grades 7, 8, and 9 comprised 32.62% (244), 38.77% (290), and 28.61% (214) of the sample, respectively. In terms of their smartphone use duration, 38 (5.08%), 181 (24.20%), 386 (51.60%), 128 (17.11%), and 15 (2.01%) participants admitted using their smartphones for less than an hour, 1 to 3 h, 4 to 6 h, 7 to 9 h, and more than 10 h each day, respectively.

This study adopted four scales for the data collection, namely, the Smartphone Addiction Inventory-Short Form (SPAI-SF) to categorize the smartphone addiction levels of the participants, the Smartphone Use Motivations Scale to determine the motivations behind their smartphone use, the Toronto Alexithymia Scale (TAS-20) to measure their alexithymia attributes, and the Relationship Structures Questionnaire (RSQ) to gauge their insecure attachment. The researchers also developed a Personal Information Form to collect the demographic data of the participants. These instruments, the validity and reliability of which have been proven in the literature, were utilized without any modification.

Developed by Lin et al. (2017), SPAI-SF contains ten items that evaluate the propensity of adolescents to develop smartphone addiction behavior. A sample item is “I try to spend less time on smartphones, but the efforts were in vain.” Each item is rated on a four-point Likert scale (“strongly disagree” to “strongly agree”). SPAI-SF has a composite score ranging from 10 to 40, with a higher score corresponding to a higher risk of developing smartphone addiction behavior. The SPAI-SF has been validated among Chinese adolescents and has demonstrated good internal consistency (Cronbach’s alpha = 0.77–0.81; Zhang et al., 2023). In this study, this scale reported an internal consistency of 0.889.

The six-item Smartphone Use Motivations Scale was developed based on the results of previous studies on the motivations for smartphone use (Leung and Wei, 2000; Li et al., 2021) to evaluate the escape motivation of smartphone users. A sample item is “It makes me feel less alone.” Each item is measured on a seven-point Likert scale (“strongly disagree” to “strongly agree”). This scale has a Cronbach’s alpha of 0.82. Five items for measuring social motivation, such as “I use a smartphone to socialize with others,” were adapted from the Smartphone Use Motivations Scale developed by Dong et al. (2017). The Smartphone Use Motivations Scale has also been proven to have strong reliability and validity among Chinese adolescents (Cronbach’s alpha = 0.781). In this study, the questions for escape and social interaction motivation obtained Cronbach’s alpha coefficients of 0.852 and 0.857, respectively.

Developed by Bagby et al. (1994) and adapted into Chinese by Zhu et al. (2007), TAS-20 is a 20-item scale for measuring alexithymia. Each item is measured on a five-point Likert scale (“strongly disagree” to “strongly agree”). TAS-20 is divided into the three sub-dimensions of DIF (e.g., “I often don’t know why I am angry.”), DDF (e.g., “I find it hard to describe how I feel about people.”), and EOT (e.g., “I prefer to just let things happen rather than to understand why they turned out that way.”). A higher composite score from these sub-dimensions corresponds to a higher degree of alexithymia. In the initial study, this scale has an overall Cronbach’s alpha of 0.87, whereas its sub-dimensions have internal consistency coefficients ranging from 0.65 to 0.80. In this study, these sub-dimensions reported an overall Cronbach’s alpha of 0.83.

Developed by Fraley et al. (2011) and adapted into Chinese by Zhang et al. (2022), the RSQ of the Experiences in Close Relationships-Revised (ECR-RS) contains nine items that evaluate the attachment styles of adolescents. These items offer prototypical descriptions of avoidance (6 items) and anxiety (3 items), with each item measured on a seven-point Likert scale (“strongly disagree” to “strongly agree”) to indicate the participants’ degree of intimacy with their parents. Sample items are “I worry a lot about my relationships” (attachment anxiety) and “I am nervous when parents get too close to me” (attachment avoidance). In Zhang et al. (2022), this scale obtained internal consistency coefficients ranging from 0.71 to 0.80. However, in this study, attachment avoidance and anxiety obtained Cronbach’s alpha values of 0.904 and 0.867, respectively.

The data were collected between November and December 2022 using sojump.com, an online survey platform in China. The participants submitted their voluntary participation forms, including their and their parents’ consent forms, prior to the survey. All participants were informed that the collected data would be kept confidential and would only be available to the researchers. The links to the Online Parent Consent Forms were sent to the participants via their social media accounts, and the link to the online survey was sent once receiving the signed consent form.

The relationships among use motivations, alexithymia, insecure attachment, and smartphone addiction were analyzed by inputting the collected data into SPSS 21 and AMOS 26. A descriptive analysis of the data was then conducted, followed by a validity check of the adopted instruments. The normally distributed data were processed using AMOS 26 for the mediation analysis and to evaluate both the measurement and structural models. Following the suggestions of Schreiber et al. (2006), the overall fit was assessed using several indices, including chi-square (χ²), χ²/df ratio, incremental fit index (IFI), comparative fit index (CFI), Tucker-Lewis index (TLI), and root mean square error of approximation (RMSEA) (Browne and Cudeck, 1989). A model is said to have an acceptable fit when its CFI and TLI values are greater than or equal to 0.80 and its RMSEA values are less than or equal to 0.08 (Schreiber et al., 2006).

The skewness and kurtosis of the data were initially evaluated to ensure the trustworthiness of the dataset. The sample had a nearly normal distribution as reflected in the skewness (−1.67 to 0.51) and kurtosis (−1.25 to 3.00) of all variables, which were within acceptable ranges (<2 for skewness and <4 for kurtosis) (Kline, 2005). Descriptive statistics for study variables are presented in Table 1. And none of these variables showed any significant association with gender (p > 0.05).

As indicated by a Kaiser-Meyer-Olkin value of 0.91 and Bartlett’s test of sphericity value of 19428.108 (p < 0.001), these 50 items were sufficiently correlated to warrant a component analysis. Given their poor factor loadings, items SM3 and SM4 in social motivation and EM2 and EM3 in escape motivation were removed. Moreover, one item of the escape motivation variable (EM6) had multiple loadings on both factors and was thus excluded from further analysis. Then, the eight factors obtained Cronbach’s alpha coefficients ranging from 0.75 to 0.92, which exceeded the 0.70 thresholds (Sanchez, 2013).

After rotation, all items obtained statistically significant (p < 0.001) factor loadings ranging from 0.61 to 0.91. Supplementary Table 1 presents the confirmatory factor analysis (CFA) results. The factor loadings of all variables exceeded 0.7. Meanwhile, their average variance extracted (AVE) and CR values exceeded 0.511 and 0.715, respectively, which are above their respective thresholds (0.5 and 0.7). These results confirm the sufficient reliability and convergent validity of the selected factors. Moreover, as shown in Table 2, the structural equation modeling (SEM) correlation coefficients of these dimensions were below their corresponding squared AVEs, thus indicating that the selected factors have satisfactory discriminant validity.

Structural equation modeling was performed using AMOS to test the hypotheses, and the resulting path model (Figure 2 and Table 3) reported a good fit (χ²/df = 2.729, IFI = 0.916, CFI = 0.916, TLI = 0.911, RMSEA = 0.048). Among the 17 hypotheses, only 13 were supported as the coefficient beta for the respective hypothesized paths reached statistical significance.

Five factors, namely, EM (β = 0.042; p < 0.05), SM (β = 0.178; p < 0.001), DIF (β = 0.11; p < 0.001), EOT (β = 0.179; p < 0.001), AAV (β = 0.072; p < 0.05), and AAN (β = 0.062; p < 0.001) were significantly and positively associated to smartphone addiction, thus supporting hypotheses H1 to H3 and hypotheses H5 and H7. However, smartphone addiction was not significantly related to DDF (β = 0.033; p = 0.118 > 0.05). The effects of EM (β = 0.133; p < 0.01), SM (β = 0.177; p < 0.001), DIF (β = 0.184; p < 0.01), DDF (β = 0.357; p < 0.001), and EOT (β = 0.363; p < 0.001) on AAV were also confirmed, thus supporting hypotheses H8a and H8b and hypotheses H10a to H10c. Meanwhile, ANN was confirmed to be influenced by both SM (β = 0.192; p < 0.001) and EOT (β = 0.233; p < 0.001), thus supporting hypotheses H9b and H11c. However, ANN was not significantly affected by DIF (β = 0.108; p = 0.128 > 0.05) and DDF (β = 0.069; p = 0.229 > 0.05), thus rejecting hypotheses H11a and H11b.

Considering the insignificant path coefficients among EM, DDF, DIF and AAN, and between DDF and SA, the mediating influence of AAV on the relationship among SM, EM, DIF, EOT, and SA was tested along with that of AAN on the relationship among SM, EOT, and SA. AMOS was used to simulate the bootstrapped 95% confidence intervals (CI) with 5,000 bias corrections, as shown in Table 4. Since 95% confidence intervals (95% CI) did not comprise zero, the relationships of smartphone addiction with DIF [indirect effect = 0.013, 95% CI (0.004, 0.028), p < 0.01], EOT [indirect effect = 0.026, 95% CI (0.014, 0.047), p < 0.01], EM [indirect effect = 0.001, 95% CI (0.003, 0.020), p < 0.001], and SM [indirect effect = 0.013, 95% CI (0.006, 0.023), p < 0.01] were significantly mediated by AAV, while its relationships with SM [indirect effect = 0.012, 95% CI (0.005, 0.021), p < 0.01] and EOT [indirect effect = 0.014, 95% CI (0.005, 0.028), p < 0.01] were mediated by AAN.

First, the results for the relationships between alexithymia traits and insecure attachment styles were in line with those obtained by Scigala et al. (2022). Moreover, consistent with the findings of Xiao et al. (2021), smartphone addiction was positively predicted by DIF and EOT. While supporting the findings of Ridout et al. (2021), smartphone addiction did not show any significant correlation with the DDF subscales of TAS-20, hence rejecting the research hypotheses. This result could be ascribed to the relationship of DDF with mindfulness facets, leading to less contact with other people. According to Scigala et al. (2021), individuals exhibiting elevated levels of DDF might be perceived as distant or antagonistic by their peers. Despite this, they tend to resist conforming to external expectations, resulting in unfavorable responses from others. Notably, it has been shown that mindfulness facets negatively mediated the relationship between alexithymia and smartphone addiction (Yuchang et al., 2017). As a result, individuals who have difficulty describing their feelings would act with self-awareness and also show a tendency toward less connection with others, which may lead to less frequent smartphone use. As mentioned earlier, having both DIF and EOT can lead to challenges in emotion regulation, thus increasing the tendency for individuals to adopt unhealthy coping strategies, and eventually trigger smartphone addiction (Zhou et al., 2022).

Second, consistent with previous research (e.g., Kim and Koh, 2018; Zhang and Wu, 2022), smartphone addiction was predicted by motivations for using smartphones. Although smartphone use among adolescents is driven by different motivations, their motivations for internal escape and external social interaction are internal psychological factors that depend on their individual needs and define the recurrence of targeted behaviors in their daily lives (Schaller et al., 2017). According to Zhang et al. (2022), an individual’s level of motivation to use their smartphone is directly linked to the frequency and duration of smartphone usage, which ultimately increases the risk of developing smartphone addiction. Therefore, personal needs define the smartphone usage behavior of individuals (Eighmey and McCord, 1998; Busch and McCarthy, 2021). Those adolescents that frequently satisfy their personal motivation by using their smartphones are more likely to become overdependent on these devices (Wen et al., 2022).

Third, this study incorporates insecure attachment styles (i.e., attachment avoidance and attachment anxiety) as a mediating variable to explore the potential mechanisms that underlie the relationship among alexithymia, smartphone use motivations, and smartphone addiction. Consistent with our assumption, attachment avoidance is confirmed to mediate the association of smartphone addiction with several factors, including DIF, EOT, escape motivation and social motivation. People showing attachment avoidance tend to have the interpersonal functioning of avoidance and apathy; therefore, these people tend to leave their feelings unexplored and not care about the feelings of others (Jauregui and Estevez, 2020; Ghinassi and Casale, 2023). Other scholars have shown that users with smartphone addiction behavior consider their smartphone as an instrument that helps them escape reality (Boldrini et al., 2021; Gao et al., 2022). Therefore, one may argue that the tendency for an individual to become addicted to smartphones depends on the strength of his/her avoidance attachment. The findings of this study also highlight the mediating effect of attachment anxiety on the relationships between smartphone addiction with EOT and social motivation. However, this variable does not mediate the relationships of smartphone addiction with DIF, and escape motivation. This finding can be attributed to the fact that attachment anxiety pertains to an individual’s intense desire to acquire and sustain the attention of others (Zhou et al., 2022). In contrast, people with DIF tend to experience difficulties with social interaction, avoid intimacy, and reject the idea of seeking help or closeness from others (Fraley et al., 2011; Kim and Koh, 2018). As a result, they display a greater degree of attachment avoidance rather than attachment anxiety. Additionally, the direct path from an avoidance motivation to attachment anxiety was less significant among individuals who use their smartphones to escape reality because of their greater desire to limit their contact with others and their tendency to avoid frustrations associated with social belonging (Fu et al., 2020; Musetti et al., 2022).

This study contributes to the literature in two ways. First, this study offers a highly nuanced discussion of the relationship among smartphone addiction, alexithymia, use motivations, and insecure attachment for adolescents. While previous studies of smartphone addiction among adolescents have mostly focused on a single category of predictors (Busch and McCarthy, 2021), this work comprehensively explores the relationship of smartphone addiction with multiple factors. Specifically, results show that the increasing prevalence of smartphone addiction among adolescents is related to two types of motivation, three dimensions of alexithymia, and two types of insecure attachment. Second, consistent with the I-PACE model (Brand et al., 2019), the current study reveals that alexithymia and use motivations are influenced differently by attachment insecurities. While earlier studies have examined the overall role of various dimensions, a clear explanation of the influence of each variable is lacking as these studies have ignored the sub-dimensions (Xiao et al., 2021). By incorporating the effects of anxiety and avoidance attachment, smartphone addiction, and use motivations, the findings of this work offer a groundwork for further explorations of the mechanisms that underlie the development of smartphone addiction. In response to the increasing number of scholars that use the I-PACE model to understand smartphone addiction (Chatterjee et al., 2022; Wen et al., 2022), this study opens up a new avenue for investigating the mechanisms by which smartphone addiction is predicted by various personal and affective factors.

For practical implications, the findings of this work are supported by the I-PACE model, which suggests that smartphone addiction is directly influenced by attachment insecurities, use motivations, DIF, and EOT. Parents and teachers alike should determine how adolescents can be guided in regulating their negative feelings and balancing their interactions with the virtual and real worlds (Eighmey and McCord, 1998; Boldrini et al., 2021). Adolescents need to be taught how to interact positively with their real environments; moreover, they need to be provided an atmosphere where emotional support is readily available, especially at times when they face challenges (Chen, 2019; Shutzman and Gershy, 2023). Parents and teachers should also strive toward closing the emotional gap between them and adolescents by designing additional interventions that can compensate for the unmet needs of the latter, such as team counseling and leisure activities (Xiang et al., 2022). They also need to identify those adolescents with narrative disorders and insecure attachment styles in a timely manner and offer them much-needed support accordingly (Scigala et al., 2022).

This study is not without its limitations. First, this work is limited by subjective self-report questionnaires (Zhang and Wu, 2022). Future studies can further validate the findings of this work by utilizing more objective techniques, such as smartphone apps, to monitor smartphone usage analysis and thus minimize reporting bias (Busch and McCarthy, 2021). Second, as our data were cross-sectional, it may limit the causal inference (Yuchang et al., 2019). To further validate the potential temporal association between each smartphone addiction and different variables among adolescents, future studies can consider conducting a longer follow-up period through longitudinal studies. Third, the mediators or outcome variables in this study were not related to the demographic variables of the participants and hence may affect the reliability of the findings regarding smartphone addiction among adolescents (Chen et al., 2023). Future studies can address this gap by collecting more relevant demographic information, such as family income and parents’ educational background, and adopting a larger sample size. Furthermore, the intricate correlations among the research variables were explored in this study using SEM. However, the SEM results could not be extrapolated to causality due to the cross-sectional nature of this research (Schaller et al., 2017). Future research can further validate the findings of this work by using longitudinal observations to understand how the development of smartphone addiction can be predicted by specific variables.