Design
This longitudinal study is a secondary analysis of data from the 10th to 11th wave of the Panel Study on Korean Children (PSKC).
Setting and Subjects
In the PSKC, children born at medical institutions nationwide in 2008 will be followed up until 2027 to track their growth and development. The PSKC uses stratified multi-step sampling, and data are collected by dividing the whole country into six regions. The number of births per year in each region is proportionally allocated. PSKC data were acquired from a balanced panel surveyed at regular intervals. The subjects of this study participated in the 10th and 11th years of the PSKC, which started the investigation on smartphone addiction among preteens. And among 1,286 children aged 9-10 who responded parents, children, and school teachers 342 people who answered all the questions were selected as the subjects of this study.
Measurements
Sociodemographic and smartphone-related characteristics. Sociodemographic characteristics included gender, residence area, household monthly income, and mother’s educational level and employment status. Smartphone-related characteristics included whether preteens owned a smartphone and the duration of daily smartphone use.
Smartphone addiction. Smartphone addiction was assessed using the K-Scale for Adolescent Observers, an internet addiction diagnostic scale provided by the Internet Addiction Response Center (iapc.co.kr) of the National Information Society Agency (2019) [28]. The scale was modified for use in preteens. Answered by the subjects’ mothers, the instrument consists of 15 items: five items on factor 1 (daily life disorder), four items on factor 3 (withdrawal), four items on factor 4 (tolerance), and two items for an unclassified factor. The questionnaire consisted of evaluated on a four-point Likert scale from “not at all” (1 point) to “a great deal” (4 points), with higher scores indicating a greater risk of addiction. Using the total score and summation score for each factor, subjects were classified as general users, potential-risk users, and high-risk users.
General users had a total score of < 27 points, and the criteria for each factor were as follows—factor 1: < 12 points, factor 3: < 10 points, and factor 4: < 9 points. Potential-risk users had a total score of 28–29 points with a score of > 13 points for factor 1, > 11 points for factor 3, and > 10 points for factor 4. Lastly, high-risk users had a total score of > 30 points, with a score of > 14 points for factor 1, > 12 points for factor 3, and > 11 points for factor 4. Potential- and high-risk users were classified as risk users.
Child factors.
Overall happiness. This variable was measured using the Millennium Cohort Study Child Paper Self-Completion Questionnaire (2008) [29]. Overall happiness indicated happiness in each area of daily life (academics, appearance, family, friends, school life). The questionnaire consisted of six items evaluated on a four-point Likert scale from “not happy at all” (1 point) to “very happy” (4 points). A higher score indicated higher overall happiness.
Self-esteem. This variable was evaluated using five items from the Millennium Cohort Study, which, in turn, were taken from Rosenberg’s (1965) [30]. 10-item Self-Esteem Scale according to the children’s age [29]. In the PSKC, some items were modified to improve comprehension. The items were evaluated on a five-point Likert scale from “not happy at all” (1 point) to “very happy” (5 points). A higher score indicated higher self-esteem.
Strengths and difficulties. This variable was evaluated using the Korean translation of the Strengths and Difficulties Questionnaire for teachers available online [31]. The subjects’ teachers provided the necessary data by responding to the online questionnaire. In the provided translations, two items were partially revised to suit the context following discussions with PSKC researchers. For example, 'stealing from home, school, or somewhere' was partially modified to 'stealing things at home, at school, or somewhere'. Children’s strengths and difficulties may be divided into prosocial behavior and total difficulties. Total difficulties consisted of internalization difficulties (emotional symptoms, peer problems) and externalization difficulties (behavioral problems, hyperactivity/inattention). The tool consisted of 25 items evaluated on a four-point Likert scale from “not at all” (1 point) to “a great deal” (4 points).
School adjustment. This variable was evaluated using a tool developed by Ji and Jeong [32]. Data were collected from the subjects’ teachers through the online questionnaire. The tool consisted of 35 items in four sub-domains: 11 items on adjustment to school life, 11 items on adjustment to academic performance, eight items on peer adjustment, and five items on teacher adjustment. Each item was evaluated on a five-point Likert scale from “strongly disagree” (1 point) to “strongly agree” (5 points). A higher score indicated greater school adjustment.
Peer factors.
Peer attachment. This variable was assessed through a nine-item questionnaire used by extracting only the peer attachment scale from Armsden and Greenberg’s (1987) Parent and Peer Attachment Scale [33]. The assessed sub-domains were communication, trust, and alienation. Each item was rated a four-point Likert scale ranging from “strongly disagree” (1 point) to “strongly agree” (4 points), with higher scores indicating a stronger degree of each sub-domain.
Parental factors.
Parental supervision. Huo’s tool [34] was modified and supplemented by Kim et al. [35] in the Korea Institute for Health and Social Affairs’ Comprehensive Children’s Survey to evaluate parenting behavior using a total of eight items, including four items on praise and encouragement for children. In the PSKC, a total of four items related to parental supervision were used in consideration of the balance with the variables of the previous survey. Responses were provided by the main caregiver and rated on a five-point Likert scale from “strongly disagree” (1 point) to “strongly agree” (5 points).
Parenting behavior. This variable was evaluated using a tool developed by Robinson et al. [36] and modified by PSKC researchers. The tool consisted of three dimensions: authoritative, authoritarian, and permissive parenting behaviors. Authoritative parenting behavior included affection, involvement, reason/induction, democratic relationships, and kindness/comfort. Authoritarian parenting behavior included anger, punishment, and irrational punishment strategies and directives. Lastly, permissive parenting behavior included lack of consistency, neglect, and conviction. The items were evaluated on a five-point Likert scale from “strongly disagree” (1 point) to “strongly agree” (5 points). A higher score indicated higher parenting behavior in each dimension.
Data Collection
For this secondary data analysis, we requested the Childcare Policy Research Institute for the original national survey data. The purpose of use and researchers’ affiliations were explained through the PSKC website. The original data and questionnaire were provided and used in this study.
Data Analysis
STATA/BE 17 was used for longitudinal analysis of the two-year data. Data from the 10th and 11th years of the survey (2017 to 2018) were case merged using the sample ID. Then, the data were converted to long-type panel data for analysis. Frequency/percentage and mean/standard deviation were calculated to compare subject characteristics and differences between general and risk users. A panel logit model analysis was conducted using STATA/BE 17 to identify factors affecting changes in problematic smartphone users.
If Xit was an independent variable affecting yit, ui was the intrinsic unobserved heterogeneity of the panel object that does not change with time, and εit was pure error, the regression equation of the panel logit model was as follows:
yit = α + β'Xit + ui + εit
for i= 1,2,…..,N and t=1,2,....., T
N = number of individuals or cross section
T = number of time periods
A fixed effect and random effect model were estimated sequentially, and the Hausman test was conducted to evaluate the endogeneity problem of explanatory variables. The Hausman test of the panel logit model showed that the null hypothesis, that there was no correlation between individual unobserved heterogeneity and independent variables, was not rejected at a significance level of 5%. Thus, a random effect model was used instead of a fixed effect model (χ2(13) = (b-B)΄〔V_b-V_B〕^(-1)〕(b-B) = 13.37, Prob > χ2 = 0.4194).
Ethical Considerations
PSKC, the primary data collection institution for this study, was approved by the Korea Institute of Child Care and Education, Institutional Review Board (KICCEIRB-2017-No.05, KICCEIRB-2018-No.02). This data is publicly accessible and written consent was obtained from all participants prior to participating in the survey. The respondent's information was anonymized for research purposes and was not identified prior to analysis. This study was a secondary analysis of publicly available data without state-supplied personal information.