Impact of online gambling on youth problem behaviors during COVID-19 periods in South Korea

Abstract

This study investigates the impact of online gambling on problem behaviors among South Korean adolescents across three phases of the COVID-19 pandemic: pre-pandemic (2018), early pandemic (2020), and late pandemic (2022). We applied a doubly robust estimation approach that combines propensity score matching with regression analysis using nationally representative survey data from the Korea Problem Gambling Agency. Our findings indicate that online gambling significantly intensified adolescents’ problem behaviors in all periods, with a more pronounced effect observed during the late pandemic phase in 2022 compared to the early phase in 2020. Sensitivity analysis further demonstrated that the estimated effects were substantially robust to unobserved confounding, particularly in 2018 and 2022. We conclude with a discussion of adolescents’ heightened vulnerability to online gambling-related problem behaviors and the corresponding need for targeted interventions and policy responses.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-49414-1.

Introduction

South Korea maintains a highly restrictive gambling regime in which most forms of gambling are prohibited for citizens. Legal gambling is limited to adults and a small set of state-controlled options: the national lottery, sports betting (Toto/Proto), and selected venue-based activities including horse racing, traditional Korean bullfighting, motorboat racing, cycle racing and casino gambling (only one casino admits Korean nationals)^1,2.

Notwithstanding these prohibitions, online gambling among adolescents has become a significant public health concern in South Korea, with prevalence rising from 8.2% in 2018 to 18.3% in 2022^3,4. Concurrently, gambling-related problems among youth have increased, including the use of illegal loans and involvement in online fraud to finance gambling. In response, the government has implemented regulatory measures and public awareness campaigns to reduce gambling-related harm^5,6. Nevertheless, adolescents continue to access illegal platforms with ease, facilitated by the anonymity and convenience of smartphones and by the extraterritorial operation of many sites beyond the reach of domestic regulation^7–9. Existing policies have not matched the scale or complexity of adolescent online gambling, underscoring the need for research that clarifies adolescents’ access pathways and risk factors and that can inform more effective, targeted prevention and regulatory strategies.

Yet, despite the heightened social alarm and policy efforts, it remains unclear whether online gambling truly contributes to problem behaviors among adolescents, particularly when compared to traditional offline gambling. The empirical evidence on this issue is mixed. Some studies report a strong association between online gambling and problematic behaviors, often attributing this to factors such as increased accessibility, anonymity, and immediate reinforcement^6,7. In contrast, other research has found no significant differences between online and offline gambling^10,11, and a few studies have even identified inverse relationships, with offline gambling users showing more problem behaviors¹². Moreover, most of these studies focus on adult populations, limiting the applicability of their findings to adolescents, who are developmentally more susceptible to risk-taking behaviors and impaired impulse control¹³.

The COVID-19 pandemic has further complicated this issue. Social distancing measures and the increased reliance on digital technologies during the pandemic have likely heightened adolescents’ exposure to online gambling platforms. While several studies have examined gambling behavior during the pandemic, most have focused primarily on adult populations^14–17. Recent research has begun to address adolescent gambling during this period, with studies documenting changes in gambling patterns^18,19, examining predictors of problem gambling²⁰, and exploring pandemic-related risk factors²¹. However, little is known about how adolescent gambling behaviors differed across distinct phases of the public health crisis. This gap is particularly salient in South Korea, where adolescent online gambling rates rose markedly during this period, yet empirical research remains substantially limited. Furthermore, most existing studies have used correlational designs, which are inadequate for establishing causal relationships between online gambling and adolescent problem behaviors.

This study addresses these gaps by examining the causal impact of online gambling on adolescent problem behaviors across three phases of the COVID-19 pandemic: pre-pandemic (2018), early pandemic (2020), and late pandemic (2022). Using nationally representative data from South Korean adolescents, the study employs a doubly robust estimation approach that integrates propensity score matching with regression analysis. By controlling for potential confounders and conducting sensitivity analysis, this study aims to provide robust evidence for the impact of online gambling on adolescents’ problem behaviors throughout the pandemic periods in South Korea.

The remainder of this article is organized as follows. The next section presents the main results of the analyses across the three pandemic phases. The following section discusses the findings in relation to existing literature and outlines their implications for policy and future research. The final section describes the data sources, variables, and analytical strategy, including propensity score matching and doubly robust estimation.

Results

Pre-processing with matching

Following the matching procedure, covariate balance between the online and offline gambling groups was assessed. Most standardized mean differences (SMDs) were below 0.1, indicating adequate balance. Unmatched control group samples were subsequently excluded from further analysis. Across the three survey years (2018, 2020, 2022), the matching process substantially improved covariate balance between online and offline gambling groups. SMDs declined from a pre-matching range of 0.065–0.620 to 0.014–0.044 post-matching. A comparable improvement was observed for the Kolmogorov–Smirnov (K–S) statistics, which fell from 0.032–0.298 to 0.007–0.022. Detailed diagnostics of the balance assessment are provided in Table 1 and Supplementary Table S1 as well as Supplementary Figs. S1 and S2. Upon reconstructing the groups, we conducted descriptive statistical analyses of dependent variables. Comprehensive descriptive statistics for each year’s online and offline gambling groups are presented in Supplementary Table S2.

Table 1.

Covariate balance (SMDs) before and after propensity score matching.

Variables	Matching	Pre-pandemic (2018)	Early pandemic (2020)	Late pandemic (2022)
Gender (Male)	Before	0.527	0.377	0.404
	After	0.010	0.014	0.046
Age	Before	0.518	0.462	0.280
	After	0.006	0.023	0.024
Middle school students	Before	0.605	0.474	0.270
	After	0.000	0.029	0.014
High school students	Before	0.155	0.275	0.108
	After	0.009	0.052	0.069
Experience winning money	Before	0.158	0.482	0.066
	After	0.005	0.062	0.013
Who gambles around you	Before	0.816	0.437	0.607
	After	0.000	0.039	0.069
Online gacha item purchase experience	Before	0.343	0.424	0.400
	After	0.019	0.026	0.000
Experience with preventative program	Before	0.027	0.063	0.105
	After	0.015	0.026	0.000
N	Before	4152	1593	1977
	After	872	312	288

Effect of online gambling across COVID-19 periods

Table 2 presents the results of examining the effect of online gambling across the three phases. Although these were not the primary analyses, we additionally conducted, for robustness of findings, simple linear regression (SLR), multiple linear regression (MLR), and propensity score matching (PSM). Despite some variation, all results were consistent with the doubly robust estimation (DR) findings. The doubly robust estimation indicated that online gambling significantly increased the severity of adolescent problem behaviors in the pre-pandemic period (b = 2.10, 95% Confidence Interval (CI) [1.62, 2.58]), exerted a smaller yet significant effect during the early pandemic period (b = 1.30, 95% CI [0.47, 2.13]), and rose again in the late pandemic period (b = 2.44, 95% CI [1.66, 3.25]). These effects were consistent across all analytical models. The results are illustrated in Fig. 1, which highlight variations across COVID-19 phases.

Table 2.

Effect of online gambling on adolescent problem severity across COVID-19 periods.

Independent variables	Models	Pre-pandemic (2018)	Early pandemic (2020)	Late pandemic (2022)
Online Gambling	SLR	3.00*** [2.75, 3.26]	2.23*** [1.82, 2.63]	2.56*** [2.16, 2.97]
	MLR	2.12*** [1.87, 2.38]	1.71*** [1.31, 2.11]	2.11*** [1.72, 2.51]
	PSM	2.11*** [1.59, 2.63]	1.31** [0.45, 2.17]	2.44*** [1.64, 3.25]
	DR	2.10*** [1.62, 2.58]	1.30** [0.47, 2.13]	2.44*** [1.66, 3.25]
N	Before matching	4152	1593	1977
	After matching	872	312	288

95% confidence intervals are presented in bracket. SLR simple linear regression model, MLR multiple linear regression model, PSM propensity score matching, DR doubly robust estimation. *p < 0.05. **p < 0.01. ***p < 0.001.

Fig. 1.

Effect of online gambling on adolescent problem gambling severity across COVID-19 periods (doubly robust estimation). Note The points show the estimated effect for each year, while the vertical lines represent the corresponding 95% confidence intervals.

Following the effect analysis, a linear hypothesis test was conducted to assess whether the estimated effects differed significantly across periods. The results indicated a statistically significant difference between 2020 and 2022 (95% CI [0.03, 2.20]). In contrast, the differences between 2018 and 2022 (95% CI [− 0.56, 1.20]) and between 2018 and 2020 (95% CI [− 1.73, 0.13]) were not statistically significant (Table 3).

Table 3.

Comparison of the effect of online gambling on adolescent problem gambling across COVID-19 periods.

Periods	Estimates	Standard error	p	95% Confidence Interval
				Lower	Upper
2022 vs. 2020	1.116	0.554	0.044	0.030	2.20
2022 vs. 2018	0.317	0.450	0.480	− 0.564,	1.20
2020 vs. 2018	− 0.799	0.475	0.092	− 1.73,	0.13

Sensitivity analysis for the effect of online gambling

The findings consistently revealed that online gambling increased problem behavior across all analyzed models. To ensure the robustness of these findings, we conducted a sensitivity analysis to address omitted variable bias (OVB), based on Cinelli and Hazlett’s²² methods.

As shown in Table 4, the results indicate that the estimated effect remained robust pre-pandemic in 2018, with a partial R² of 7.79%. To invalidate this finding, a confounder would need a robustness value (RV) of 25.14% to nullify the point estimate, or 20.01% to render it statistically non-significant. In contrast, the effect was relatively more vulnerable to hidden bias during the early COVID-19 pandemic phase in 2020 (partial R² = 3.06%), requiring an RV of 16.62% to nullify the estimate and 6.23% to render it non-significant. By 2022, robustness improved substantially and surpassed pre-pandemic levels (partial R² = 12.12%), with corresponding RVs of 30.88% and 22.31%, respectively.

Table 4.

Sensitivity analysis for the effect of online gambling.

Category	Pre-pandemic (2018)	Early pandemic (2020)	Late pandemic (2022)
Partial R2 of the treatment with the outcome (%)	7.79%	3.06%	12.12%
Robustness Value for the point estimate (%)	25.14%	16.62%	30.88%
Robustness Value for the t-value (%)	20.01%	6.23%	22.31%

The robustness value (RV) indicates the minimum confounder strength needed to alter the conclusion. RV for the point estimate is minimum confounder association with both exposure and outcome required to reduce the estimated effect to zero. RV for the t-statistic is minimum confounder association required to make the effect statistically non-significant.

To further evaluate these findings, we benchmarked the RV against our model’s strongest observed covariate: exposure to gamblers within the adolescent’s social environment. Contour plots for all survey years (Fig. 2) demonstrate that the effect of online gambling remained robust even when confronted with a hypothetical confounder three times stronger than this benchmark. In contrast, during the 2020 pandemic period, the estimated effect approached the threshold under the assumption of a confounder three times as strong as our benchmark, indicating increased vulnerability to hidden bias.

Fig. 2.

Contour plots of sensitivity analysis for the estimated effect of online gambling. Note The plots assess robustness to an unobserved confounder. The x- and y-axes show the partial R² of a confounder with the treatment and outcome, respectively. (a) Contours show the adjusted effect estimate. The red dashed line indicates the confounder strength required to reduce the estimate to zero. (b) Contours show the adjusted t-value. The red dashed line indicates the confounder strength required to render the effect statistically non-significant. The partial R² of the benchmark covariate is plotted at 1, 2, and 3 times its value for comparison.

Discussion

This study examined the impact of online gambling on the severity of problem behaviors among South Korean adolescents during the COVID-19 periods. The impact of online gambling on increasing the severity of problem behaviors among adolescents has consistently demonstrated statistical significance across all periods in 2018, 2020, and 2022. These findings suggest that online gambling may serve as a risk factor for exacerbating problem behaviors among adolescents, compared to offline gambling^4,6,23,24.

Our finding that online gambling exacerbates problem behaviors aligns with research by Effertz et al.²⁵ but contrasts with Philander and MacKay¹², who reported an inverse relationship between online gambling and problem behaviors. This divergence can likely be explained by significant differences in study context, population, and period. First, our study is situated in South Korea during the COVID-19 pandemic, a unique environment characterized by a global surge in adolescent online activity²⁶ and a distinct regulatory landscape where illegal online platforms are prevalent^1,4. In contrast, Philander and MacKay’s research was conducted pre-pandemic in the United Kingdom, a nation with a more liberalized and established legal gambling market¹². Second, our focus on adolescents is critical; this population is developmentally more susceptible to impaired impulse control and the immediate reinforcement common in online gambling platforms compared to the general adult population in their study^11,13.

In particular, we found the impact of online gambling increased during the late COVID-19 period (2022) compared to the early phase in 2020. This increase may be associated with heightened engagement in illegal online gambling in South Korea during this time⁴, as several features of this environment contribute to increased risk behaviors among adolescents. With access to smartphones, adolescents can easily sign up for and use online gambling platforms⁴. The anonymity of the internet further enables illegal gambling websites to grant access to underage users²⁷. Many of these websites operate in overseas jurisdictions where online gambling is legal, which poses practical limitations on Korean law enforcement’s ability to regulate them^1,5,28. These factors facilitate adolescents’ unrestricted access to gambling activities, thereby heightening their vulnerability to problem behaviors.

The sensitivity analysis indicated that the findings were robust to unobserved confounding in 2018 and 2022, but more vulnerable to such bias in 2020. This increased susceptibility may reflect disruptions to adolescents’ daily routines during the early stages of the COVID-19 pandemic, particularly due to quarantine measures. During this period, smartphone use rose sharply, which has been linked to more frequent engagement in online gambling^15,26,29,30. In South Korea, adolescents primarily used smartphones as their main device for online gambling⁴.

These results provide broader support for the hypothesis that online gambling contributes to heightened problem behaviors among adolescents. Importantly, rigorous statistical methods were employed to strengthen the causality between these variables. Furthermore, the impact of online gambling remained consistent across the COVID-19 periods, even under social distancing restrictions, suggesting that it is not a temporary phenomenon but a persistent issue requiring sustained attention. The present study thus offers additional empirical evidence that online gambling can lead to harmful consequences for adolescents, including financial distress, compulsive involvement, and adverse psychological outcomes. Accordingly, it is essential to recognize the severity of this issue and to implement prevention and intervention strategies tailored to this vulnerable population.

While this research provided evidence for a causal relationship between online gambling and problem behaviors among adolescents using doubly robust estimation, several limitations should be acknowledged. First, due to the quasi-experimental nature of our observational design, potential unmeasured confounders may still exist, warranting caution in causal interpretation. Second, our classification of gambling types was dichotomous (online versus offline), which may not fully capture the complexity of modern gambling patterns where adolescents often engage in both modalities. This binary categorization also obscures important variations in risk profiles among different gambling forms within each modality. Third, although we analyzed data across three time points, the samples were cross-sectional rather than longitudinal, limiting our ability to track individual trajectories and developmental patterns. Fourth, our reliance on self-reported measures may introduce reporting biases, particularly for socially undesirable behaviors like gambling. Furthermore, the variation in survey administration methods, specifically the inclusion of an online format during the 2020 wave due to COVID-19, may have introduced method variance. However, this hybrid approach might have also compromised data quality, as the lack of direct supervision during online administration could increase the likelihood of careless or inattentive responding compared to traditional offline settings. Fifth, the high percentage of missing data is a significant limitation. Although our sensitivity analysis using imputed data yielded consistent results, suggesting the robustness of our findings, the potential for selection bias remains, particularly for the 2022 sample which had the largest proportion of missing data. This increasing pattern of missing might reflect changing response patterns during the pandemic, potentially introducing systematic biases.

Future research should address these limitations in several ways. First, studies should examine key mediating factors such as emotional regulation, impulsivity, and social influence dynamics^31,32. This approach may clarify the mechanisms through which online gambling contributes to adverse outcomes in adolescents. Second, future research should employ more nuanced classifications that account for both the modality (online/offline) and the specific type of gambling activity, as these interact to influence risk. For instance, online sports betting might carry different risk profiles than online casino games, just as offline sports betting differs from offline lottery participation. Recent research has demonstrated that specific gambling activities carry different levels of inherent risk, with activities like sports betting and casino games typically associated with higher problem gambling rates compared to lottery or scratch cards³³. Third, longitudinal and mixed-methods designs that follow adolescents through different phases of gambling involvement would provide invaluable insights into the temporal sequence, causal pathways, and contextual nuances that quantitative cross-sectional data alone cannot capture^18–21. Fourth, incorporating objective measures (such as digital tracking data or clinical assessments) alongside self-reports would strengthen measurement validity. Finally, as an intellectual exercise to move beyond the limitations of our quasi-experimental data, it is worth considering how this research question could be operationalized through a controlled experiment. Due to strict ethical constraints, it is impossible to randomly assign adolescents to an active online gambling condition to observe subsequent problem behaviors. However, a large-scale randomized controlled trial (RCT) could be designed from a prevention standpoint. For instance, adolescents could be randomly assigned to either a highly targeted, interactive gambling prevention intervention or a waitlist control group. By subsequently tracking both their online gambling engagement and general problem behaviors over time, researchers could infer causality: if the intervention significantly reduces online gambling exposure and concurrently decreases problem behaviors compared to the control group, it would provide robust experimental evidence of the causal pathway. Alternatively, a small-scale, lab-based controlled experiment could utilize simulated, risk-free betting games to measure immediate changes in impulsivity and risk-taking behaviors among adolescents. This approach would isolate the short-term behavioral impacts of gambling mechanics without exposing them to real-world financial or psychosocial harm.

Methods

Data

This study utilized secondary data from a national biennial survey, originally conducted by the Korea Problem Gambling Agency, to assess the prevalence and patterns of gambling-related issues among adolescents across all regions of South Korea. The original survey employed a stratified sampling approach, selecting ten schools from each of the 17 provinces and major cities nationwide. The survey was primarily administered in an offline format. For in-school students, data collection occurred during regular class periods under the supervision of trained research assistants. For out-of-school youth, the survey was administered at youth community centers and vocational training facilities with strict confidentiality protocols. However, due to the outbreak of the COVID-19 pandemic, the 2020 survey wave was administered using a hybrid approach, providing both online and offline formats to accommodate social distancing measures and temporary facility closures.

For the present study, data from the 2018, 2020, and 2022 surveys were analyzed to investigate the impact of online gambling on the severity of problem behaviors across different phases of the COVID-19 pandemic. Although the original survey was conducted at multiple time points to investigate adolescent gambling behaviors, the data for each year were cross-sectional. Consequently, the sample from each survey year was treated as independent in our analysis.

To specifically estimate the impact of online gambling compared to offline gambling, the analysis was restricted to participants who reported having gambled. The 2018 dataset comprised 5101 students (middle to high school) and 518 out-of-school adolescents aged 13 to 18, with data collected between August and October. The 2020 dataset included 2,809 students (middle to high school) and 124 out-of-school adolescents aged 12 to 18, collected during the same months as the 2018 survey. The 2022 dataset consisted of 4665 in-school students (middle to high school) and 148 out-of-school adolescents aged 13 to 18, with data collected between September and November.

The study was approved by the Institutional Review Board of Seoul National University (IRB No. E2311/001-004). All research was performed in accordance with the Declaration of Helsinki and relevant guidelines. Informed consent was obtained from all individual participants included in the study.

Measures

Outcome variable

The outcome variable in this study is the severity of gambling-related problem behaviors, measured using nine items adapted from the Gambling Problem Severity Subscale (GPSS;³⁴). To clarify, these problem behaviors are specifically related to gambling rather than general delinquency or conduct problems. Participants self-reported the frequency of specific gambling-related problem behaviors on a four-point Likert-type scale ranging from 0 (never) to 3 (often). Sample items include "How often have you skipped or dropped out of group activities or extracurricular practices to engage in gambling or betting instead?", "How often have you planned your gambling or betting activities?", and “How often have you felt bad about the amount you bet, or about what happens when you bet money?” The total score, calculated by summing all nine items, indicates the severity of gambling-related problem behaviors. A higher overall score reflects greater severity of problems.

Independent variable

The independent variable was the primary mode of gambling (online vs. offline). Participants were categorized into one of two groups based on their most frequent gambling activity over the preceding three months. The online gambling group (coded as 1) comprised individuals who primarily engaged in internet-based gambling (see Table 5, responses 8–13). In contrast, the offline gambling group (coded as 0) consisted of participants who predominantly engaged in land-based gambling activities (responses 1–7).

Table 5.

Survey question for primary gambling type in the past three months.

Question

Measurement

Which type of gambling/betting have you played most frequently in the last three months? If you played one type of gambling/betting, please select that one; if you played more than one, please select the one you played most often

1 = crane games, 2 = betting on sporting events, 3 = betting on other games (such as board games, bingo, etc.), 4 = card games, 5 = buying lottery tickets, 6 = buying lottery tickets on sporting events, 7 = horse racing, track cycling, motorboat racing, bull fights, etc. 8 = online cards games (using a PC or smartphone), 9 = other online betting games (such as mini-games), 10 = buying online lottery tickets, 11 = betting on Toto/Proto on Betman, 12 = betting on other sports betting systems on websites other than Betman, 13 = other online casino games (such as blackjack, baccarat, etc.)

Crane game is an arcade game in which players use a joystick and buttons to control a crane-like claw to grab a prize from a machine. Toto/Proto is the name for legal gambling on sports events in South Korea. Betman is the website where people can play Toto/Proto online.

Control variables

Informed by prior research, this study accounts for key covariates known to shape adolescent gambling behavior and the severity of associated problems. The control variables included gender, age, school grade level, prior experiences of monetary winnings from gambling, exposure to gamblers within the adolescent’s social environment, and participation in gambling prevention programs. All variables, excluding age, were dichotomously coded (Table 6). These covariates were integrated into both the propensity score matching and multiple regression model to more precisely isolate the effect of online gambling on the severity of problem behaviors.

Table 6.

Control variables and measurement.

Variables	Measurement
Gender	1 = male, 0 = female
Age	continuous
Middle school students	1 = yes, 0 = no (reference group: out-of-school youth)
High school students	1 = yes, 0 = no (reference group: out-of-school youth)
Experiences of monetary winnings from gambling	1 = yes, 0 = no
Exposure to gamblers within the adolescent’s social environment	1 = yes, 0 = no
Online gacha item purchase experience	1 = yes, 0 = no
Participation in gambling prevention programs	1 = yes, 0 = no

A gacha item is a random item in online games that players can purchase using in-game currency or real money.

Statistical analyses

We evaluated the effect of online gambling engagement by applying multiple analytical models, with our main analysis utilizing a doubly robust estimation approach that combines propensity score matching (PSM) and regression analysis. All statistical procedures were performed using R statistical software. Prior to analysis, cases with missing data were removed from each annual dataset via listwise deletion. Missing data rates were 26.1% for 2018, 33.4% for 2020, and 58.9% for 2022. The datasets, after listwise deletion, were used for all subsequent matching and regression analysis. Additionally, to account for potential bias from missing data, we conducted a parallel analysis using imputed data generated with the ‘missForest’ package, a random forest-based method adept at handling mixed datasets with both continuous and categorical variables³⁵. Results from this analysis, presented in the Supplementary Table S3, confirmed the robustness of our primary findings.

Propensity scores were estimated using logistic regression. Subsequently, 1:1 nearest neighbor matching without replacement was conducted using the ‘MatchIt’ package³⁶. This technique pairs individuals from the treatment and control groups based on the most similar propensity scores, a method widely used for its simplicity and interpretability³⁷. Consistent with this matching procedure, the Average Treatment Effect on the Treated (ATT) was estimated, representing the effect of online gambling among adolescents who reported engagement in online gambling^38,39.

To evaluate balance achieved by the matching procedure, we calculated standardized mean differences (SMDs) for covariates within each annual dataset. SMDs provide a standardized measure of difference between treatment and control group means, with values approaching zero indicating strong balance^40,41. For comparing online gambling effects across years, we assessed balance on the combined dataset after applying weights derived from the within-year matching processes⁴². This assessment utilized SMDs, and Kolmogorov–Smirnov (K–S) statistics. Unlike its application in hypothesis testing, the K–S statistic here serves as a descriptive measure of distributional distance, where smaller values indicate greater overlap between distributions and thus superior covariate balance^41,43. The covariates used in the matching process were reintroduced in the subsequent regression analysis to account for residual confounding, consistent with the doubly robust estimation approach⁴⁴. Furthermore, we conducted sensitivity analysis for each survey year to evaluate the robustness of our findings in the presence of potential unobserved confounders²².

Finally, given the substantial post-matching reductions in sample size (e.g., in 2022, from 1977 to 288), we addressed concerns about generalizability by repeating the analysis with the full, unmatched sample. We compared estimates of the effect of online gambling on problem behaviors from the propensity score matched sample with those derived from the full dataset. The direction and statistical significance of the key coefficients were consistent across analyses, indicating that excluding unmatched cases did not distort the primary findings. Thus, the observed impact of online gambling is robust and not attributable to sample selection. Additional details of this analysis are provided in Supplementary Table S4.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

J.P. wrote the main manuscript text. J.P., D.K., and Y.K. contributed to the conceptualization. J.P. was responsible for the methodology. D.K. acquired the funding and data. D.K. and Y.K. supervised the project. All authors reviewed and edited the manuscript.

Funding

This research was supported by the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea (NRF-2020S1A3A2A02103411).

Data availability

The data that support the findings of this study are available from the Korea Problem Gambling Agency but restrictions apply to the availability of these data, which were used under license for the current study and are not publicly available. Data are however available from the corresponding author, upon reasonable request and with permission of the Korea Problem Gambling Agency.

Declarations

Competing interests

The authors have no relevant financial or non-financial interests to disclose.