Advertisement

Structural Equation Model of Smartphone Addiction Based on Adult Attachment Theory: Mediating Effects of Loneliness and Depression

Open AccessPublished:May 19, 2017DOI:https://doi.org/10.1016/j.anr.2017.05.002

      Summary

      Purpose

      This study investigated the mediating effects of loneliness and depression on the relationship between adult attachment and smartphone addiction in university students.

      Methods

      A total of 200 university students participated in this study. The data was analysed using descriptive statistics, correlation analysis, and structural equation modeling.

      Results

      There were significant positive relationships between attachment anxiety, loneliness, depression, and smartphone addiction. However, attachment anxiety was not significantly correlated with smartphone addiction. The results also showed that loneliness did not directly mediate between attachment anxiety and smartphone addiction. In addition, loneliness and depression serially mediated between attachment anxiety and smartphone addiction.

      Conclusion

      The results suggest there are mediating effects of loneliness and depression in the relationship between attachment anxiety and smartphone addiction. The hypothesized model was found to be a suitable model for predicting smartphone addiction among university students. Future study is required to find a causal path to prevent smartphone addiction among university students.

      Keywords

      Introduction

      The Korea Internet and Security Agency reported in 2014 that smartphone usage amongst Korean university students in their twenties was 99.8%, with a daily average of 152.1 minutes, which was the highest rate of any age group [
      • Korea Internet and Security Agency
      2015 survey on internet use [Internet].
      ]. Compared to Korean high school students or office workers, it is easier for Korean university students to access their smartphones because they have more free time and are less under the authority of parents and teachers. In addition, using smartphones might relieve their anxiety about the unpredictable future and stress from interpersonal conflicts, such as romantic relationship breakups [
      • Jeon H.S.
      • Jang S.O.
      A study on the influence of depression and stress on smartphone addiction among university students: focusing on moderating effect of gender.
      ]. Due to excessive use, therefore, the high smartphone usage rate among university students is apt to lead to smartphone addiction. According to the “Internet Addiction Survey of 2014” from the Ministry of Science, Information and Communication Technology (ICT), and Future Planning, 20.5% of university students are at risk of smartphone addiction, with 3.2% at high risk, and 17.3% at potential risk [
      • Korea Internet and Security Agency
      2014 survey on mobile internet use [Internet].
      ]. In other words, about one in five university students are at risk of smartphone addiction, which is a serious issue that may interfere with their abilities to live healthy lives. Although both mobile phones and the internet also have a high probability for addiction, the portability of smartphones might put them at greater risk of becoming addictive than either basic mobile phones or the internet.
      Some researchers who have examined the relationship between attachment styles and internet or smartphone addiction report that Korean youth or university students who have an insecure attachment style are more likely to become addicted to the internet or smartphone than those with a secure attachment style [
      • Choi Y.Y.
      • Seo Y.S.
      The relationship between insecure adult attachment and smartphone addiction: the mediation effect of impulsiveness moderated by social support.
      ,
      • Schimmenti A.
      • Passanisi A.
      • Gervasi A.M.
      • Manzella S.
      • Fama F.I.
      Insecure attachment attitudes in the onset of problematic Internet use among late adolescents.
      ]. These young people are apt to use smartphones in order to meet their need for attachment, thus regarding their smartphones as objects of attachment. Other studies have determined that depressed Korean university students have a high tendency to be addicted to smartphones [
      • Jeon H.S.
      • Jang S.O.
      A study on the influence of depression and stress on smartphone addiction among university students: focusing on moderating effect of gender.
      ]. For Korean university students, depression was found to be the highest predicting factor among other risk factors for addiction, presenting a significantly positive correlation with smartphone addiction [
      • Kim B.N.
      • Ko E.J.
      • Choi H.I.
      A study on factors affecting smartphone addiction in university students: a focus on differences in classifying risk groups.
      ].
      Yet other studies have found that loneliness plays the role of mediator to any addiction [
      • Park W.K.
      An explorative study on college students' addiction: using psychological variables as predictors.
      ,
      • Byun H.S.
      • Kweon S.Y.
      The effect of elementary school students' loneliness on school life adaptation: the mediating effect of smartphone addiction.
      ]. In a study by Park [
      • Park W.K.
      An explorative study on college students' addiction: using psychological variables as predictors.
      ], smartphone addiction and loneliness presented a significantly positive correlation. The study also reported, through multiple regression analysis, that loneliness was a major predictive factor for social networking service addiction [
      • Park W.K.
      An explorative study on college students' addiction: using psychological variables as predictors.
      ]. Furthermore, loneliness has been found to be a mediating factor between attachment style and smartphone or internet addiction [
      • Byun H.S.
      • Kweon S.Y.
      The effect of elementary school students' loneliness on school life adaptation: the mediating effect of smartphone addiction.
      ].
      Flores [
      • Flores P.J.
      Addiction as an attachment disorder.
      ], in exploring the relationship between insecure attachment style and addiction, suggested that addiction resulted from attachment problems. According to the attachment theory of Bowlby [
      • Bowlby J.
      The making and breaking of affectional bonds.
      ], attachment is established between an infant and the infant's primary caregiver through emotional bonds facilitated by physical contact and familiarity. Insecurely attached persons have a negative image of self or others in their internal working models due to their early experiences. As a result, insecurely attached people might either become dependent on or avoidant of others [
      • Ainsworth M.D.
      • Blehar M.C.
      • Waters E.
      • Wall S.
      Patterns of attachment: a psychological study of the strange situation.
      ]. Brennan et al [
      • Brennan K.A.
      • Catherine C.L.
      • Phillip R.S.
      Self-report measurement of adult attachment: an integrative overview.
      ] define “attachment anxiety” as having a fear of rejection and “attachment avoidance” as a fear of closeness and dependency [
      • Wei M.
      • Vogel D.L.
      • Ku T.
      • Zakalik R.A.
      Adult attachment, affect regulation, negative mood, and interpersonal problems: the mediating roles of emotional reactivity and emotional cutoff.
      ].
      According to Flores, the reason why substance abusers become addicts is because they experience difficulties with emotional regulation [
      • Flores P.J.
      Addiction as an attachment disorder.
      ]. Addiction is regarded as an alternative to overcoming the difficulties in their relationships, which worsens their attachment capabilities. Those who have insecure attachment styles are likely to become addicted due to their deficiency in regulating their emotions effectively [
      • Flores P.J.
      Addiction as an attachment disorder.
      ]. In a similar vein, Catanzaro and Mearns [
      • Catanzaro S.J.
      • Jack M.
      Measuring generalized expectancies for negative mood regulation: initial scale development and implications.
      ] insist that people who are able to cope with or regulate their negative emotions well are less prone to becoming depressed.
      In Korea, it has been shown that differences in attachment styles have a high correlation with depression or the lack of depression [
      • Kim B.J.
      • Oh K.J.
      Exploration of mediator between adult attachment and depressive mood in college students: brooding and emotion suppression.
      ,
      • Yi J.Y.
      • Choi H.C.
      Hope as a mediator between adult attachment and depression in university students.
      ]. In particular, Kim and Oh [
      • Kim B.J.
      • Oh K.J.
      Exploration of mediator between adult attachment and depressive mood in college students: brooding and emotion suppression.
      ] found that those who had anxious and avoidant attachment styles were more likely to suffer from depression. Wei et al [
      • Wei M.
      • Vogel D.L.
      • Ku T.
      • Zakalik R.A.
      Adult attachment, affect regulation, negative mood, and interpersonal problems: the mediating roles of emotional reactivity and emotional cutoff.
      ] found that both anxious and avoidant attachment styles had strong associations with increased loneliness and difficulties in interpersonal relationships. Yet another study [
      • Yi J.Y.
      • Choi H.C.
      Hope as a mediator between adult attachment and depression in university students.
      ] discovered that an anxious attachment style had direct and indirect effects on loneliness and depression. Based on these previous studies, this study explored the mediating effects of depression and loneliness in the relationship between smartphone addiction and attachment style.
      This study sought to build a conceptual framework based on a premise rooted in Flores' theory and attachment theory [
      • Flores P.J.
      Addiction as an attachment disorder.
      ,
      • Bowlby J.
      The making and breaking of affectional bonds.
      ]: insecurely attached university students tend to feel loneliness and depression, and cannot regulate their emotions properly. They therefore choose smartphones as self-objects to reduce their pain. By applying Flores' assertion in conjunction with the findings of previous studies [
      • Flores P.J.
      Addiction as an attachment disorder.
      ], the present study aimed to suggest the paths between the risk factors and the mediating effects of depression and loneliness in the relationship between smartphone addiction and attachment style.

      Purpose

      The purpose of this study was to propose a structural equation model between attachment style and smartphone addiction based on Flores' argument that people become addicted because they experience difficulties with emotional regulation [
      • Flores P.J.
      Addiction as an attachment disorder.
      ]. To do this, we verified the hypothesized structural equation model and examined the mediating effect of loneliness and depression in the relationship between attachment style and smartphone addiction.

      Methods

      Study design

      A cross-sectional research design was employed to explore the causal pathways linking attachment to smartphone addiction in university students.

      Study sample

      The participants for this research were university students enrolled in a major or elective class at one of the four universities located in cities designated as A, C, S, and P. The age range was 20–30 years old. The reason for choosing this age limit was that the rate of smartphone usage in the twenties was the highest and two times longer than any other working adult group [
      • Korea Internet and Security Agency
      2014 survey on mobile internet use [Internet].
      ]. From the cross-sectional survey, 210 students were recruited, but only data from 200 students were used for the final analysis, due to inconsistent answers. This number satisfied the minimum sample size for the maximum likelihood method, which was commonly used in structural equation modeling analysis [
      • Bae B.R.
      Structural equation modeling with Amos 21.
      ].

      Measurements

      Attachment anxiety

      Attachment anxiety was assessed using the Experiences in Close Relationships-Revised-Korean instrument, which was developed by Brennan et al [
      • Brennan K.A.
      • Catherine C.L.
      • Phillip R.S.
      Self-report measurement of adult attachment: an integrative overview.
      ] and translated by Kim et al [
      • Kim S.H.
      • Sherry A.R.
      • Lee Y.S.
      • Kim C.D.
      Psychometric properties of a translated Korean adult attachment measure.
      ]. The attachment anxiety measure was composed of 18 items, each rated on a 7-point Likert scale. Higher scores suggested an increased severity of attachment anxiety. The instrument demonstrated good reliability in the research of Brennan et al's [
      • Brennan K.A.
      • Catherine C.L.
      • Phillip R.S.
      Self-report measurement of adult attachment: an integrative overview.
      ] (α = .91) and Kim et al [
      • Kim S.H.
      • Sherry A.R.
      • Lee Y.S.
      • Kim C.D.
      Psychometric properties of a translated Korean adult attachment measure.
      ] (α = .85). In the current study, internal consistency was strong (α = .89). For the study, we followed the instructions of Bae [
      • Bae B.R.
      Structural equation modeling with Amos 21.
      ], who used the exploratory factor analysis, and did item parceling with three factors extracted from the exploratory factor analysis.

      Attachment avoidance

      Attachment avoidance was also measured using the Experiences in Close Relationships-Revised-Korean. The attachment avoidance assessment was composed of 18 items rated on a 7-point Likert scale. Higher scores represented an increased severity of attachment avoidance. The instrument had strong reliability in the study of Brennan et al [
      • Brennan K.A.
      • Catherine C.L.
      • Phillip R.S.
      Self-report measurement of adult attachment: an integrative overview.
      ] (α = .91) and the study of Kim et al [
      • Kim S.H.
      • Sherry A.R.
      • Lee Y.S.
      • Kim C.D.
      Psychometric properties of a translated Korean adult attachment measure.
      ] (α = .85). In our study, internal consistency was satisfactory (α = .79). As with attachment anxiety, we followed the instructions of Bae [
      • Bae B.R.
      Structural equation modeling with Amos 21.
      ].

      Depression

      Depression was assessed using the Center for Epidemiologic Studies Depression Scale (CES-D) developed by Radloff [
      • Radloff L.S.
      The CES-D scale: a self-report depression scale for research in the general population.
      ] and translated by Chon et al [
      • Chon K.K.
      • Choi S.C.
      • Yang B.C.
      Integrated adaptation of CES-D in Korea.
      ]. The CES-D is a questionnaire composed of 20 items, each rated on a 4-point Likert scale. Higher scores indicated a greater severity of depression. The instrument had strong reliability (α = .91) in Chon et al's [
      • Chon K.K.
      • Choi S.C.
      • Yang B.C.
      Integrated adaptation of CES-D in Korea.
      ] research, and in the present study internal consistency was strong (α = .89). We followed the same instructions of Bae [
      • Bae B.R.
      Structural equation modeling with Amos 21.
      ] referenced previously.

      Loneliness

      The Revised UCLA Loneliness Scale (RULS) was a 20-item Likert scale questionnaire used to measure loneliness. It was developed and revised by Russell et al [
      • Russell D.
      • Letitia A.P.
      • Mary L.F.
      Developing a measure of loneliness.
      ] and translated by Park and Doh [
      • Park S.Y.
      • Doh H.S.
      The relations between family functioning and self-esteem and loneliness in adolescents.
      ]. RULS was divided into two subscales: satisfaction with relationships (10 items) and dissatisfaction with relationships (10 items). Scores ranged from 20 to 80. Higher scores showed an increased severity of loneliness. RULS used a 5-point Likert scale with answers ranging from 0 (never) to 4 (very often). Cronbach α was reported in Park and Doh's [
      • Park S.Y.
      • Doh H.S.
      The relations between family functioning and self-esteem and loneliness in adolescents.
      ] study as .90. In this study, Cronbach α for the whole scale was .91, and the two subscales (satisfaction and dissatisfaction) each had a coefficient of .82. In this study, the two subscales (satisfaction and dissatisfaction) were used as factors for item parceling.

      Smartphone addiction

      Smartphone addiction was measured using the Smartphone Addiction Proneness Scale developed by the National Information Society Agency [
      • Shin K.W.
      • Kim D.I.
      • Chung Y.J.
      Development of Korea smartphone addiction proneness scale for youth and adults.
      ]. It consisted of four subdomains: disturbance of adaptive functioning (5 items), virtual life orientation (2 items), withdrawal (4 items), and tolerance (4 items), equaling a total of 15 items. Each item was scored on a 5-point Likert scale. A higher score represented an increased severity of smartphone addiction proneness. The current study yielded a Cronbach α coefficient of .92. Item parceling was conducted with the four subdomains as factors, as the instrument developers suggested.

      Data collection

      The researchers received permission from the developers and translators via email to use all instruments of data collection. The participants were university students from Seoul, Chungchung Province, Kangwon Province, and Kyungsang Province. In order to protect the participants, the researchers received permission from the Institutional Review Board (IRB no. PIRB-2015-E006) of Pohang University of Science and Technology before data collection commenced. Data collection lasted from March 15th to May 16th, 2015. In order to participate in the research, participants were informed of the study purpose, the participation procedures, the dangers and benefits of the research, confidentiality, and the option to withdraw from the study at any time.

      Data analysis

      The collected data were analyzed utilizing SPSS 21.0 (IBM Corp., Armonk, NY, USA) and AMOS 21.0 (IBM Corp., Armonk, NY, USA). In order to examine the characteristics of the participants and the measured variables, descriptive statistics were used. Pearson correlation was used to evaluate the correlations between the latent variables. In order to evaluate the fitness of the hypothesized and modified models, χ2, root mean square error of approximation (RMSEA), the goodness of fit index (GFI), the comparative fit index (CFI), the normal fit index (NFI), and the Tucker Lewis index (TLI) were used. Bootstrapping was conducted to examine the statistical significance of the modified model's indirect effects, total effects, and mediated effects. Since the missing values did not have a regular pattern, we decided to use the missing at random method instead of the systematic missing rate. Furthermore, the missing rate was lower than 0.1 per variable, so mean imputation was conducted with the missing data.

      Results

      General characteristics of study sample

      The participants in this study consisted of 200 university students, 126 female (63.0%) and 74 male (37.0%). Their ages ranged from 19 years old to 28 years old (M = 21.6, SD = 2.0). In terms of religious affiliation, there were 104 nonreligious (52.0%), 60 Protestant (30.0%), 15 Catholic (7.5%), and 21 (10.5%) unidentified students. As for grade, there were 23 freshmen (11.5%), 83 sophomores (41.5%), 56 juniors (28.0%), and 38 seniors (19.0%) (Table 1).
      Table 1General Characteristics of Study Sample (N = 200).
      CharacteristicsCategoriesn (%)
      GenderMale74 (37.0)
      Female126 (63.0)
      Age (yr)M ± SD21.6 ± 2.0
      1939 (19.5)
      2034 (17.0)
      2131 (15.5)
      2229 (14.5)
      2334 (17.0)
      2411 (5.5)
      2517 (8.5)
      264 (2.0)
      281 (0.5)
      ReligionProtestant60 (30.0)
      Catholic15 (7.5)
      Atheist104 (52.0)
      Unidentified21 (10.5)
      GradeFreshmen23 (11.5)
      Sophomore83 (41.5)
      Junior56 (28.0)
      Senior38 (19.0)

      Descriptive statistics of variables, test of normality and multicollinearity

      This analysis had the expectation that the observable variables of the hypothesized structural equation model would follow a normal distribution [
      • Song T.M.
      • Kim G.S.
      Structural equation modeling for health welfare research.
      ,
      • Yu J.P.
      Structural equation models: concepts and understanding.
      ]. Skewness and kurtosis were to be used in the examination of the assumption of normal distribution if the skewness of the variables was less than the absolute value 3 and the kurtosis was less than the absolute value 10, since the distribution would not be far from the assumption of normal distribution [
      • Song T.M.
      • Kim G.S.
      Structural equation modeling for health welfare research.
      ]. In this study, all the variables fell under the skewness with an absolute value of 1.72 and the kurtosis with an absolute value of 3.4. Therefore, as expected the results of skewness and kurtosis met the assumption of normal distribution.
      The examination of the assumption of multicollinearity was executed based on tolerance and the variance inflation factor. In this study, tolerance was .42–.62 (> .10), and variance inflation factor was 1.68–2.38 (< 10). Therefore, there was no problem with multicollinearity.

      Modification and verification of hypothesized model

      Verifications of the hypothesized model's fitness are shown in Table 2. Since χ2 of the hypothesized model was 210.66, which was twice greater than the degree of freedom (GFI = .88, CFI = .93, TLI = .91, NFI = .89, RMSEA = .09), it did not reach the recommended standard. Therefore, on the basis of modification indices, the model's fitness was improved through the covariance set between the measurement error of endogenous variables and the measurement error of exogenous variables. As a result, since χ2 was 145.76, which was lower than twice the degree of freedom (GFI = .91, NFI = .93, CFI = .96, TLI = .94, RMSEA = .07), it satisfied the model acceptance criteria.
      Table 2Fit Index of Hypothetical Model and Modified Model.
      Modelχ2 (p)χ2/dfGFIRMSEANFICFITLI
      Criteriap ≥ .05≤ 2≥ .9.08–.10 (mediocre fit)

      ≤ .08 (reasonable fit)

      ≤ .05 (close fit)
      ≥ .9≥ .9≥ .9
      Hypothetical model210.66 (< .001)2.63.88.09 (LO .07, HI .11).89.93.91
      Modified model145.76 (< .001)1.99.91.07 (LO .05, HI .08).93.96.94
      Note. CFI = comparative fit index; GFI = goodness fit index; NFI = normal fit index; RMSEA = root mean square error of approximation; TLI = Tucker-Lewis index.

      Effect analysis of modified model

      Among the nine paths of the modified model, five paths were significant (Figure 1, Table 3). The direct paths from attachment avoidance to smartphone addiction (β = −.37, t = −2.56) and attachment avoidance to loneliness (β = .46, t = 5.75) were statistically significant. The direct path from attachment anxiety to loneliness (β = .55, t = 5.99) was significant, and the paths from loneliness to depression (β = .54, t = 2.46) and from depression to smartphone addiction (β = .34, t = 3.06) were also valid.
      Figure 1
      Figure 1Path diagram of the modified model. Note. e1–e15 means measurment error and it is common symbols in AMOS; x1–x3 refer to parcels of attachment anxiety; x4–x6 refer to parcels of attachment avoidance; y1–y3 refer to parcels of depression; y4–y5 refer to parcels of loneliness; y6–y9 refer to parcels of addiction; *p < .05, **p < .01.
      Table 3Effects of Exogenous Variables in Modified Model.
      Endogenous variablesExogenous variablesβtR2Standardized direct effects (p)Standardized indirect effects (p)Standardized total effects (p)
      LonelinessAttachment avoidance.465.75.78.46 (< .001).46 (.012)
      Attachment anxiety.555.99.55 (< .001).55 (.012)
      DepressionAttachment avoidance−.12−0.95.50−.12 (.344).25 (.035).13 (.270)
      Attachment anxiety.291.91.29 (.057).30 (.018).59 (.007)
      Loneliness.542.46.54 (.014).54 (.037)
      Smartphone addictionAttachment avoidance−.37−2.56.31−.37 (.011).11 (.360)−.26 (.004)
      Attachment anxiety.281.70.28 (.089).28 (.139).56 (.009)
      Loneliness.150.62.15 (.534).18 (.039).33 (.272)
      depression.343.06.34 (.002).34 (.043)
      The total effect was computed by adding the direct effect of the predictor variable to the outcome variable and the indirect effect, which had rounded at least one mediating variable and had an impact on the variable [
      • Bae B.R.
      Structural equation modeling with Amos 21.
      ]. The total effect of attachment avoidance on smartphone addiction was β = −.26, which was significant (p = .004), and the total effect of attachment anxiety on smartphone addiction was .56, which was statistically significant (p=.009). Attachment avoidance, attachment anxiety, loneliness, and depression explained 31.0% of the variance. The result of bootstrapping on the mediating effect between attachment avoidance and depression was significant (β = .25, p = .035), and the mediating effect between attachment anxiety and depression was also significant (β = .30, p = .018).

      Examination of dual mediating effect

      The model of this research found a dual mediating effect of loneliness and depression on the relationship between attachment avoidance and smartphone addiction, and between attachment anxiety and smartphone addiction. In the case of a multiple mediating model, which has more than two mediating effects, bootstrapping provides total effects, including mediating effects but not specific indirect effects, according to the path of each mediating effect. Thus, the setting of a conversion model using phantom variables and examination was employed [
      • Bae B.R.
      Structural equation modeling with Amos 21.
      ,
      • Macho S.
      • Thomas L.
      Estimating, testing, and comparing specific effects in structural equation models: the phantom model approach.
      ]. Phantom variables as virtual variables do not have an effect on a model's fitness or the parameters of the original model. In this research, as presented in Figure 2, the path from attachment anxiety to smartphone addiction via loneliness and depression was set as p1→p2→p3. The indirect effect of the path showed that the coefficient of the nonstandardized indirect effect was .135; it was computed by multiplying the coefficients from the adjusted model of the nonstandardized direct effect of three paths: attachment anxiety on loneliness, loneliness on depression, and depression on smartphone addiction (Figure 2, Table 4). In addition, the path from attachment avoidance to smartphone addiction via loneliness and depression was set as p4→p5→p6. The result of this indirect effect was .061, and it was computed by multiplying the coefficients from the adjusted model of the nonstandardized direct effect of three paths: attachment anxiety on loneliness, loneliness on depression, and depression on smartphone addiction (Figure 2, Table 4). Also, the fitness between the adjusted model and the phantom conversion model was identical, indicating that the phantom conversion model was converted from the adjusted model properly.
      Figure 2
      Figure 2Mediation model by using phantom variables. Note. e1–e15 means measurment error and it is common symbols in AMOS; x1–x3 refer to parcels of attachment anxiety; x4–x6 refer to parcels of attachment avoidance; y1–y3 refer to parcels of depression; y4–y5 refer to parcels of loneliness; y6–y9 refer to parcels of addiction; p1–p6 refer to phantom variables; a denotes path from attachment anxiety to loneliness; b denotes path from attachment avoidance to loneliness; c denotes path from loneliness to depression; d denotes path from depression to smartphone addiction.
      Table 4Specific Indirect Effects.
      EffectEstimateSE95% CI (Bias corrected)p
      Indirect effect of attachment anxiety on smartphone addiction through loneliness & depression.140.090.02, 0.52.026
      Indirect effect of attachment avoidance on smartphone addiction through loneliness & depression.060.040.01, 0.22.028
      Note. CI = confidence interval; SE = standard error.
      As a result of examining the dual mediating effect through bootstrapping analysis, the dual mediating effect of loneliness and depression on the relationship between attachment anxiety and smartphone addiction was statistically significant (b = .14, p = .026), and their effect on the relationship between attachment avoidance and smartphone addiction was also found to be statistically significant (b = .06, p = .028) (Table 4).

      Discussion

      In this study, we examined the significance of the dual mediating effects of depression and loneliness in relationship to a hypothesized structural equation model based on Flores' claim [
      • Flores P.J.
      Addiction as an attachment disorder.
      ] and findings from previous studies. Our results show that attachment anxiety and attachment avoidance have a direct effect on loneliness and an indirect effect on depression through loneliness. These findings support Flores' claim [
      • Flores P.J.
      Addiction as an attachment disorder.
      ] that those who have insecure attachments cannot regulate their emotions completely, and they may feel lonely since they do not have any healthy attachments. In addition, attachment anxiety and attachment avoidance both have effects on smartphone addiction through the mediating effects of loneliness and depression. This result is consistent with the findings of previous studies [
      • Choi Y.Y.
      • Seo Y.S.
      The relationship between insecure adult attachment and smartphone addiction: the mediation effect of impulsiveness moderated by social support.
      ,
      • Schimmenti A.
      • Passanisi A.
      • Gervasi A.M.
      • Manzella S.
      • Fama F.I.
      Insecure attachment attitudes in the onset of problematic Internet use among late adolescents.
      ,
      • Flores P.J.
      Addiction as an attachment disorder.
      ] that have determined that insecure attachment has an effect on smartphone addiction. Attachment anxiety and attachment avoidance are closely related to smartphone addiction because, according to Flores' assertion [
      • Flores P.J.
      Addiction as an attachment disorder.
      ], smartphone usage is an alternative for a person's lack of secure attachment.
      The results of the present study also show that attachment avoidance has a direct negative effect on smartphone addiction. In other words, higher attachment avoidance scores correlated with lower smartphone addiction scores. This finding did not fully support the results of previous studies [
      • Choi Y.Y.
      • Seo Y.S.
      The relationship between insecure adult attachment and smartphone addiction: the mediation effect of impulsiveness moderated by social support.
      ,
      • Schimmenti A.
      • Passanisi A.
      • Gervasi A.M.
      • Manzella S.
      • Fama F.I.
      Insecure attachment attitudes in the onset of problematic Internet use among late adolescents.
      ], although insecure attachment was shown to have a positive correlation with smartphone addiction. As Wei et al [
      • Wei M.
      • Vogel D.L.
      • Ku T.
      • Zakalik R.A.
      Adult attachment, affect regulation, negative mood, and interpersonal problems: the mediating roles of emotional reactivity and emotional cutoff.
      ] stated, attachment avoidance was a fear of closeness and dependency, so those who were attachment avoidant might not be interested in other people or objects, such as smartphones. In contrast, the main characteristics of people with attachment anxiety were that they had negative images of themselves and positive images of others. Those who had the trait of attachment anxiety strived for an object to serve as a “secure base”, were always anxious, and wished to gain the approval of others. Therefore, they might be overly attached to other people or objects like smartphones. However, having a tendency toward attachment anxiety and feeling lonely did not always lead to smartphone addiction. Only when people had attachment anxiety, felt lonely, and had negative thoughts about themselves and their environments that led to depression were they highly likely to become addicted to smartphones.
      In our modified structural equation model, attachment anxiety, attachment avoidance, loneliness, and depression explained 30.7% of smartphone addiction among the university students we studied. Our findings showed that attachment, depression, and loneliness were the predictive variables of smartphone addiction. However, other variables also might predict smartphone addiction. For example, Park and Lee [
      • Park N.S.
      • Lee H.J.
      Nature of youth smartphone addiction in Korea.
      ] suggested that shyness and self-esteem might be predictive variables for smartphone addiction in addition to depression and loneliness. Wei et al [
      • Wei M.
      • Vogel D.L.
      • Ku T.
      • Zakalik R.A.
      Adult attachment, affect regulation, negative mood, and interpersonal problems: the mediating roles of emotional reactivity and emotional cutoff.
      ] explained the mediating effect of self-efficacy in the relationship between attachment anxiety, loneliness, and depression and the mediating effect of self-exposure in the relationship between attachment avoidance, loneliness, and depression. In addition, Wei et al [
      • Wei M.
      • Vogel D.L.
      • Ku T.
      • Zakalik R.A.
      Adult attachment, affect regulation, negative mood, and interpersonal problems: the mediating roles of emotional reactivity and emotional cutoff.
      ] examined the relationship between interpersonal difficulties and loneliness. Therefore, interpersonal relationships, shyness, self-esteem, self-efficacy, and self-exposure might be confounding factors having an impact on the mediating variables (loneliness and depression) and on the outcome variable (smartphone addiction). Since these potentially confounding factors can function as estimation errors, future research is needed for verification of their influences.
      To date, there has been no research on the mediating effect of loneliness and depression in the relationship between attachment and smartphone addiction for university students in Korea. However, there have been recent studies examining the mediating effects of loneliness and impulsivity in the relationship between attachment anxiety and smartphone addiction [
      • Yeo J.Y.
      • Kang S.Y.
      • Kim D.H.
      The impact of attachment anxiety on smartphone addiction with a mediating effect of impulsivity and loneliness: focused on the group of SNS and game.
      ], and the mediating effects of loneliness and university adjustment in the relationship between attachment and smartphone addiction [
      • Kim H.J.
      • Kim E.H.
      The relationship between anxious attachment and smartphone addiction: the mediating effects of loneliness and college adjustment.
      ]. Although there is recognition that loneliness affects addiction, there are limited studies that examine the dual mediating effect of loneliness and depression on the relationship between adult attachment and smartphone addiction. The results of the current study suggest that those who have insecure attachment may feel depressed due to their negative views toward self and others. Their depression and negative emotions may lead them to retreat into their own world and feel lonely. Then, they may fall into smartphone addiction.
      Furthermore, rarely do studies target university students with regard to this subject. Since the number of university students addicted to smartphones is increasing, study of the risk factors for and the protective factors preventing smartphone addiction is timely and appropriate in order to offer meaningful suggestions and ways to thwart addiction.
      In terms of smartphone addiction among university students, determining who is in a high-risk group and offering preventive intervention, solutions, and recovery paths is necessary. The present study confirms that those who have attachment anxiety want to feel security through involvement in cyberspace. Therefore, those who have traits of attachment anxiety need help establishing healthy and secure relationships in the offline world, rather than the online world, which is an artificial space.

      Conclusion

      This research examined the mediating effects of loneliness and depression in the relationship between attachment anxiety and smartphone addiction. In addition, the hypothesized model in the present study was indicated as a suitable model for predicting smartphone addiction among university students. Since smartphone addiction affects various areas of people's lives, including interpersonal relationships, studies, emotions, and health, the establishment of strategies to preclude loneliness and depression is absolutely necessary.
      The limitations of this study and recommendations for further research are as follows: First, since the participants in this study were located in Seoul, Chungchung Province, Kangwon Province, and Kyungsang Province, the results of this research cannot be generalized to other university students in other locations. While this study indicated that smartphone addiction can be reduced through interventions to prevent loneliness and depression, it was only a cross-sectional study. Thus, longitudinal research is recommended to confirm the causal path of this study's findings.

      Conflicts of interest

      The authors have no conflicts of interest to declare.

      References

        • Korea Internet and Security Agency
        2015 survey on internet use [Internet].
        Author, Seoul2015 ([cited 2015 Dec 30]. Available from:)
        • Jeon H.S.
        • Jang S.O.
        A study on the influence of depression and stress on smartphone addiction among university students: focusing on moderating effect of gender.
        Korean J Youth Stud. 2014; 21 (Korean): 103-129
        • Korea Internet and Security Agency
        2014 survey on mobile internet use [Internet].
        Author, Seoul2014 ([cited 2014 Dec]. Available from:)
        • Choi Y.Y.
        • Seo Y.S.
        The relationship between insecure adult attachment and smartphone addiction: the mediation effect of impulsiveness moderated by social support.
        Korean J Psychol Couns Psychother. 2015; 27 (Korean): 749-772https://doi.org/10.23844/kjcp.2015.08.27.3.749
        • Schimmenti A.
        • Passanisi A.
        • Gervasi A.M.
        • Manzella S.
        • Fama F.I.
        Insecure attachment attitudes in the onset of problematic Internet use among late adolescents.
        Child Psychiatry Hum Dev. 2014; 45: 588-595https://doi.org/10.1007/s10578-013-0428-0
        • Kim B.N.
        • Ko E.J.
        • Choi H.I.
        A study on factors affecting smartphone addiction in university students: a focus on differences in classifying risk groups.
        Stud Korean Youth. 2013; 24 (Korean): 67-98
        • Park W.K.
        An explorative study on college students' addiction: using psychological variables as predictors.
        Soc Psychol Res. 2014; 27 (Korean): 95-125
        • Byun H.S.
        • Kweon S.Y.
        The effect of elementary school students' loneliness on school life adaptation: the mediating effect of smartphone addiction.
        Korean J Child Educ. 2014; 23 (Korean): 303-317
        • Flores P.J.
        Addiction as an attachment disorder.
        Jason Aronson, Lanham (MD)2004
        • Bowlby J.
        The making and breaking of affectional bonds.
        Tavistock, London (UK)1979
        • Ainsworth M.D.
        • Blehar M.C.
        • Waters E.
        • Wall S.
        Patterns of attachment: a psychological study of the strange situation.
        Erlbaum, Hillsdale (NJ)1978
        • Brennan K.A.
        • Catherine C.L.
        • Phillip R.S.
        Self-report measurement of adult attachment: an integrative overview.
        in: Simpson J.A. Rholes W.S. Attachment theory and close relationships. Guilford, New York (NJ)1998: 46-76
        • Wei M.
        • Vogel D.L.
        • Ku T.
        • Zakalik R.A.
        Adult attachment, affect regulation, negative mood, and interpersonal problems: the mediating roles of emotional reactivity and emotional cutoff.
        J Couns Psychol. 2005; 52: 14-24https://doi.org/10.1037/0022-0167.52.1.14
        • Catanzaro S.J.
        • Jack M.
        Measuring generalized expectancies for negative mood regulation: initial scale development and implications.
        J Pers Assess. 1990; 54: 546-563https://doi.org/10.1080/00223891.1990.9674019
        • Kim B.J.
        • Oh K.J.
        Exploration of mediator between adult attachment and depressive mood in college students: brooding and emotion suppression.
        Korean J Sch Psychol. 2009; 6 (Korean): 357-375
        • Yi J.Y.
        • Choi H.C.
        Hope as a mediator between adult attachment and depression in university students.
        Korea J Couns. 2012; 13 (Korean): 819-838
        • Bae B.R.
        Structural equation modeling with Amos 21.
        Cheonglam, Seoul (South Korea)2014: 1-668
        • Kim S.H.
        • Sherry A.R.
        • Lee Y.S.
        • Kim C.D.
        Psychometric properties of a translated Korean adult attachment measure.
        Meas Eval Couns Dev. 2011; 44: 135-150
        • Radloff L.S.
        The CES-D scale: a self-report depression scale for research in the general population.
        Appl Psychol Meas. 1977; 1: 385-401https://doi.org/10.1177/014662167700100306
        • Chon K.K.
        • Choi S.C.
        • Yang B.C.
        Integrated adaptation of CES-D in Korea.
        Korean J Psychol Health. 2001; 6 (Korean): 59-76
        • Russell D.
        • Letitia A.P.
        • Mary L.F.
        Developing a measure of loneliness.
        J Pers Assess. 1978; 42: 290-294https://doi.org/10.1207/s15327752jpa4203_11
        • Park S.Y.
        • Doh H.S.
        The relations between family functioning and self-esteem and loneliness in adolescents.
        Korean J Child Stud. 1998; 19 (Korean): 229-244
        • Shin K.W.
        • Kim D.I.
        • Chung Y.J.
        Development of Korea smartphone addiction proneness scale for youth and adults.
        National Information Society Agency, Seoul (South Korea)2011 November (Report no.: IV-RER-11051)
        • Song T.M.
        • Kim G.S.
        Structural equation modeling for health welfare research.
        Hannarae Publishing, Seoul (South Korea)2012
        • Yu J.P.
        Structural equation models: concepts and understanding.
        Hannarae Publishing, Seoul (South Korea)2012
        • Macho S.
        • Thomas L.
        Estimating, testing, and comparing specific effects in structural equation models: the phantom model approach.
        Psychol Method. 2011; 16: 34-43https://doi.org/10.1037/a0021763
        • Park N.S.
        • Lee H.J.
        Nature of youth smartphone addiction in Korea.
        J Commun Res. 2014; 51 (Korean): 100-132
        • Yeo J.Y.
        • Kang S.Y.
        • Kim D.H.
        The impact of attachment anxiety on smartphone addiction with a mediating effect of impulsivity and loneliness: focused on the group of SNS and game.
        Korea J Youth Couns. 2014; 22 (Korean): 47-69
        • Kim H.J.
        • Kim E.H.
        The relationship between anxious attachment and smartphone addiction: the mediating effects of loneliness and college adjustment.
        Korea J Couns. 2016; 17 (Korean): 157-161