We conducted a school-based cross-sectional analytical study in the Colombo and Gampaha districts of Sri Lanka. The cross-sectional design was chosen as the most appropriate for several reasons. This is the first systematic study of internet addiction within a large adolescent population in Sri Lanka. While this design provides essential baseline data for future longitudinal and interventional studies, it cannot establish causality. In comparison to other districts, Colombo and Gampaha were selected for several reasons: (1) the highest population density in Sri Lanka, providing efficient access to a large sample; (2) the greatest internet penetration rates in the country (>70%), ensuring adequate numbers of internet users; (3) diversity of urban and semiurban areas; (4) a range of school types (government, private, and international) and socioeconomic backgrounds; and (5) representativeness of broader trends in internet adoption and use patterns among Sri Lankan adolescents. The cross-sectional design enables understanding of multiple potential correlates concurrently.

Our primary objectives were to evaluate the prevalence of internet addiction in a Sri Lankan student population and determine personal characteristics associated with internet addiction, including age, sex (male and female), language, and year of study. We explored patterns of internet use, such as social media, gaming, and web surfing, associated with internet addiction.

Students of both sexes were eligible for inclusion if they were aged between 15 and 19 years and studying in grades 10, 11, 12, and 13 in the Gampaha and Colombo districts of Sri Lanka. This age group was selected because adolescence is a critical developmental stage characterized by high internet use and vulnerability to behavioral addictions. As the study involved validation of the Sinhala translation of the IAT, students without adequate Sinhala proficiency were excluded. This constitutes a considerable limitation as exclusively Tamil-speaking students were not included in the study.

A multistage stratified cluster sampling method was selected. A classroom was considered a cluster. Considering the diversity of the school population, the sample was stratified at 2 stages. First, the sample (clusters) was divided equally into 2 strata according to school grades. The strata were defined as grades 10 and 11 (Ordinary Level classes) and grades 12 and 13 (Advanced Level classes). Within each of these strata, the sample was divided to include students from government, private, and international schools. Students from government schools were stratified according to the functional classification of schools adopted by the Ministry of Education, namely, types 1AB, 1C, and type 2 [39].

Schools within each district were organized according to the type of school, and the number of students to be sampled from each grade was determined using probability proportional to size based on the student population in the specific grades in the different types of schools (government—1AB, 1C, and type 2; private; and international). To enhance the distribution of the sample over the district, only one class from Ordinary Level classes and one from each type of Advanced Level classes were randomly selected from a single school.

The required sample size was calculated using the following formula: sample size (N) = Z² × p(1–p)/d², where Z is the value corresponding to the CI (1.96 for 95%), p is the anticipated population proportion (p=0.1), and d is the absolute precision (d=0.025).

On the basis of the literature, 4% to 10% of users have internet addiction [27]. Studies in the general population and adolescents report a lower prevalence of internet addiction, ranging from 0.3% to 0.07% [32]. To ensure a conservative and sufficient sample size, a prevalence of 10% of internet users was taken as the cutoff. With a 95% CI and a margin of error set at 2.5%, the minimum number of positive cases required was calculated as 554 as follows:

Sample size (n)=Z² × p(1–p)/d²

n=(1.96)² × 0.10(0.90)/(0.025)²

n=3.84 × 0.09/0.000625

n=553.9, or approximately 554 internet users with addiction.

As of July 2014, approximately 19.9% of the Sri Lankan population was estimated to be internet users having access to the internet at home via any type of device (computer or mobile) [34,35]. Therefore, to obtain a sample of 139 adolescents with frequent internet use (assuming 20% of the student sample are frequent users), a total of 2800 students were targeted. Our study sample comprised 2835 students. Nonresponse was minimal; incomplete questionnaires were excluded from the final analysis.

Thereafter, the study was divided into 2 phases to allow systematic data collection: initial validation of the translated Sinhala version of the IAT (phase 1) and subsequent data collection to evaluate the prevalence of internet addiction (phase 2).

It was necessary to ensure that the tool was validated before being used to measure internet addiction. Previous validations of the IAT have offered several models, ranging from 1 factor to 6 factors [32-35,37-39]. Recent studies have identified that a 1-factor or 2-factor model can be used effectively in validating the IAT and have shown no clear benefit over one another [23]. With the above evidence, we proceeded with a single-factor validation model, adopting a unidimensional approach.

On the basis of the guidelines proposed by George and Mallery [39] for interpreting Cronbach α values [40,41], internal consistency was interpreted using the criteria shown in Table 1.

To measure test-retest reliability, the same test was given to the same students on 2 occasions and the scores were correlated using Pearson correlation coefficient. The score obtained the first time the test was given was T1, and the score obtained the second time was T2. The scores on the 2 occasions were then correlated. This correlation was known as the test-retest reliability coefficient, or the coefficient of stability.

The closer each respondent’s scores were on T1 and T2, the more reliable the test measure. A stability coefficient of 1 would indicate that each respondent’s scores were perfectly correlated; that is, each respondent scored the exact same value on T1 as they did on T2. A correlation coefficient of 0 indicates that the respondents’ scores at T1 were completely unrelated to their scores at T2; therefore, the test was not reliable.

Correlation is an effect size, and the strength of the correlation between items can be described using the interpretation guide for the absolute value of the correlation coefficient (Table 2).

In our study, we used a 1-factor model assessing the Cronbach α score for the 20 items on the translated IAT. A statistical expert calculated the ideal sample size for validation before data collection. According to the widely accepted participant to item ratio approach, each item should include 2 to 20 participants, and a minimum sample size of 100 to 250 participants is required for validation [42]. A sample size of 200 individuals (10 participants per item) was considered adequate because the IAT has 20 items. To assess test-retest validity, 20% (n=40) of the sample was also enlisted.

The first 200 students who identified as internet users were included in validation analysis. This was done by calculating the Cronbach α coefficient. A score of ≥0.7 indicated that the interview schedule in Sinhala had a high level of validity and internal consistency.

For cross-cultural adaptation of psychometric instruments, the original IAT was translated into Sinhala by a nonmedical professional competent in both Sinhala and English. This was subsequently reviewed by a consultant psychiatrist and a specialist in computer technology. Discrepancies were discussed and resolved to create a single preliminary Sinhala version. They identified a few items requiring cultural modification to ensure relevance and comprehensibility: (1) modification to age-appropriate language, (2) adjustment of phrasing for natural Sinhala expression while preserving meaning, and (3) clarification to reflect Sri Lankan social context.

It was then retranslated into English by a different nonmedical professional, and the retranslated version was cross-checked by the review team. Necessary corrections and amendments were made to the translated Sinhala version so that the retranslated English version was as close as possible in meaning to the original interview schedule.

Following the abovementioned validation, the study tool was used to assess the presence and severity of internet addiction in the sample population. Internet addiction was assessed using the Sinhala translated and validated version of the IAT. A questionnaire assessing demographic data and the IAT was administered concurrently to the study population. To ensure the questionnaire was comprehensible across varying literacy levels, the language was simplified, and a research assistant was present during administration to provide clarifications when needed.

All gathered data were handled confidentially, with each participant assigned a unique reference number to ensure anonymity. SPSS software (version 23.0; IBM Corp) was used for data entry and analysis.

The translated IAT was validated using the first 200 entries; test-retest reliability was evaluated using the Pearson correlation coefficient, and internal consistency was evaluated using Cronbach α.

Following validation, the entire dataset was analyzed to assess the level of internet addiction and its nature. Descriptive statistics (means, SDs, frequencies, and percentages) were used to summarize sociodemographic variables and internet use patterns. Chi-square tests were used for categorical variable comparisons (sex [male and female] and academic profile), whereas independent 2-tailed t tests analyzed continuous variables (age differences between groups). One-way ANOVA was applied for multigroup comparisons, and the Pearson correlation coefficient was used to assess relationships between continuous measures.

Prior to statistical analysis, the dataset was screened for missing values. Missing data were analyzed for patterns using the Little MCAR (Missing Completely At Random) test to determine whether data were missing completely at random. For variables with less than 5% missing data, listwise deletion was used.

Ethics approval and other necessary permissions were obtained prior to the study. Ethical clearance for this study was obtained from the ethical clearance committee of the Faculty of Medicine, Ragama, University of Kelaniya (P/240/12/2014). Formal permission to translate and validate the IAT into Sinhala was obtained from its author, Dr Kimberley Young. All participants received an information sheet outlining the study details and what was required of them. Parental notification was conducted through school principals, who were asked to inform parents about the study and communicate any objections to their children’s participation. No parental objections or complaints were received, and students provided written assent after being informed about voluntary participation and confidentiality measures. In addition, written approval was granted by the relevant school principals to conduct the study in their respective schools, and informed written consent was obtained from students aged 18 years and above for participating in the study. No compensation was provided to participants, and participation was voluntary.

The data collected were anonymized and stored in a password-protected electronic database accessible only to the principal investigator and authorized research staff, and all paper-based records were locked in secure storage.

We anticipated that students might not provide fully accurate information regarding their internet use due to fear of reprimand or other concerns. An anonymous self-administered questionnaire was administered under the supervision of a researcher. Subsequently, students were instructed to place the completed questionnaires into a closed collection box to ensure that their responses could not be traced to them or seen by other students. Furthermore, responses on internet use duration were compared with other reported digital behaviors, such as time spent on social media, gaming, and educational browsing. Extreme or inconsistent responses were flagged and excluded during data cleaning. Nonresponse and incomplete questionnaires were excluded from the final analysis.

This study demonstrates that internet addiction is prevalent among school children in the Western Province of Sri Lanka, with a substantial proportion exhibiting mild to moderate levels of problematic use. These findings add to the limited body of evidence from Sri Lanka on adolescent internet addiction and highlight that problematic use is not confined to university or clinical populations. The prevalence pattern observed in this study is broadly consistent with findings from other South Asian countries, including India and Bangladesh, where studies among adolescents have reported a wide range of prevalence estimates depending on methodology, age group, and assessment tools used [23,24,32]. Differences in reported prevalence across studies may reflect variations in internet access, cultural norms, parental supervision, and educational demands.

The predominance of mild and moderate addiction observed in this study mirrors trends reported globally. Meta-analyses and large-scale studies from China, Europe, and other regions suggest that early-stage problematic internet use is considerably more common than severe addiction among adolescents [28,43]. This pattern is important from a public health perspective, as it suggests opportunities for early identification and intervention before more severe impairment develops. The lack of significant associations between internet addiction and basic sociodemographic characteristics in this study is consistent with findings from Hong Kong, Taiwan, and South Korea [29,30,44]. Instead, behavioral factors related to patterns of use appear to be more salient. Longer duration of use and recreational or gaming-focused internet activities have been repeatedly associated with higher addiction scores across different cultural settings [15,25,38]. These findings support the view that how the internet is used may be more important than who is using it.

From a public health and educational perspective, the findings underscore the need for school-based approaches to promote healthy internet use. International evidence suggests that digital literacy education, parental engagement, and clear school policies regarding internet use can reduce harmful patterns and associated psychosocial problems [27,45]. In Sri Lanka, where internet access among school children continues to expand rapidly, integrating such interventions into existing school health and education programs may be both feasible and effective.

While the cross-sectional study design provides a snapshot at one point in time and helps estimate prevalence and associated factors, it does not allow for the establishment of causal relationships. Therefore, future longitudinal studies are needed for more comprehensive assessment.

Colombo and Gampaha districts have the highest internet penetration, the greatest urbanization, and higher average socioeconomic status. So the prevalence, patterns, and correlates of internet addiction likely differ substantially compared with rural areas with limited connectivity, which represents a major limitation. Our findings should not be extrapolated to these contexts without appropriate research.

As the study involved validating the Sinhala translation of the Young IAT, the Tamil-speaking population was not assessed for internet addiction. Therefore, these findings may not be generalizable to the entire Western Province and limit generalizability across ethnic groups. Tamil-speaking communities may have different cultural attitudes toward technology use, different parental supervision patterns, different media consumption preferences, and potentially different prevalence of problematic use. There is potential for selection bias, as the findings may not reflect the experiences of adolescents in rural and remote areas, and the exclusion of Tamil-speaking students may introduce ethnic and linguistic bias.

The prevalence of internet addiction among internet users in the study sample was 12.6% (227/1803), and only 0.9% (16/1803) were identified as having severe internet addiction. This is less than that identified in some studies in other countries, including India [24,32,33]. This may have been due to the students not being forthcoming with information when data were collected in a school setting by research assistants, even though the questionnaire was self-administered and confidential. Students may have underreported problematic use, creating the desirability bias. Therefore, many studies in this area have been conducted using internet-based surveys, which are more anonymous and may yield more accurate information, potentially minimizing desirability bias.

We did not collect detailed socioeconomic status indicators, such as family income, parental occupation, household assets, or parental education level. This represents a significant limitation, as socioeconomic status likely operates through multiple pathways to influence internet addiction risk. We did not screen for depression, anxiety disorders, attention-deficit/hyperactivity disorder, conduct disorders, or other psychiatric conditions that commonly occur with internet addiction. Future research must include comprehensive mental health assessment as well.

Internet addiction is prevalent among school children in the Western Province of Sri Lanka, predominantly at mild to moderate levels. Patterns of internet use, rather than sociodemographic characteristics, are key correlates. Incorporating screening and preventive interventions into school health and education programs may help mitigate the long-term psychological and social consequences of excessive internet use among adolescents.