First, what was the study trying to find out? The researchers wanted to see how mental health difficulties (called psychopathology) and gaming disorder (GD) symptoms were connected across three points in time. They also wanted to know whether changes in one were linked with later differences in the other.
How the researchers analyzed the data
- They checked the data. From December 2024 to March 2025, they checked whether each measure followed a pattern that their statistical tests could use. When a measure did not, they changed its scale using a method called a log transformation.
- They checked how mental health measures fit together. A method called confirmatory factor analysis tested whether different mental health measures could be combined to represent one overall mental health factor.
- They tested links over time. The researchers used a statistical model called a cross-lagged panel model (CLPM). In simple terms, it checks whether a person’s earlier score on one measure is connected with their later score on another. They first examined mental health and GD, while accounting for sex. Then they added personal characteristics, including impulsivity, difficult life events, family conflict, and bullying.
- They compared two types of model. They also tested a random-intercept CLPM, which separates people’s usual, long-term differences from changes within each person. The researchers chose the regular CLPM as their main model because their main question was about whether adolescents who differed in mental health also tended to differ in GD over time.
Which model fit the data better?
The regular CLPM had fit scores of CFI = 0.95, RMSEA = 0.04, and SRMR = 0.05. The random-intercept version had scores of CFI = 0.92, RMSEA = 0.06, and SRMR = 0.06. These scores describe how well a model matches the data; generally, lower RMSEA and SRMR scores and higher CFI scores indicate a better fit. The formal comparison did not find a meaningful difference between the models (P ≥ .99). However, the regular CLPM’s scores looked slightly better, so the researchers used it as their main model.
How they handled missing information and checked the results
- Missing data: They used a method called full-information maximum likelihood, which uses the available information rather than simply dropping everyone with a missing answer.
- Different research sites: As an extra check, they used models that accounted for participants being grouped within 21 study locations. This helped test whether the results might be affected by differences between locations.
- Other checks: They tested whether the mental health measures worked consistently across time and also ran separate models for each mental health measure.
The analyses were run in the R computer program using the lavaan package. The researchers used a significance cutoff of P < .05. This is a commonly used rule for judging whether a result is statistically significant; it does not, by itself, show how important the result is.
How large were the connections?
The researchers used these guideposts for the model’s effect sizes, shown by the Greek letter beta (β):
- β = 0.03: small
- β = 0.07: medium
- β = 0.12: large
In short: The researchers used several checks and statistical models to study how mental health and gaming-disorder symptoms were related across time. They also accounted for important personal factors, missing answers, and differences between research locations.