Target area and sample
Japan mainly consists of eight regions (Hokkaido, Tohoku, Kanto, Chubu, Kinki, Chugoku, Shikoku, and Kyushu). The Ministry of Internal Affairs and Communications of Japan determines city scale based on city population and classifies cities into the following categories: large cities (population of more than 500 thousand), core cities (population of 200–500 thousand or prefectural capital), medium cities (population of 100–200 thousand), small cities (population of 10–100 thousand), and town and village (population of less than 10 thousand) [19].
Our research team comprised 18 researchers from 15 research institutions. We selected and mailed surveys to 78 schools from the eight regions and asked them to complete them. Among them, 76 schools (61 public school, 12 national school, 3 private school) of 78 schools accepted the survey. [20] The survey involved a self-report questionnaire. We asked the schools to give it to 11- to 18-year-old adolescents between 2017 and 2019. In sum, we obtained 21491 questionnaire responses. The purpose, method, benefits, and risks of this study were explained to the principals of the schools. We also explained to the participants that their private information would be protected and that answers to the questionnaires were not related to their school records. Participants provided consent before answering the questionnaire. We excluded 3598 national primary school and secondary school students and 9473 high school students from this study because it was difficult to specify their school district. In Japan, national schools and high schools do not have school districts, and some students may go to school far from their homes. Since we could not specify these participants' addresses, we could not assess their neighborhood environments. We also excluded 318 adolescents due to missing data regarding sex. As a result, 8102 adolescents (4087 males and 4015 females from 48 schools, 11–15 years old) were included in the analysis (Fig. 1).
Individual-level characteristics
We obtained basic information regarding each participant’s sex, birth year, birth month, height, body weight, and organized sports participation through their self-reports. We divided birth month into four groups: Q1 (April to June), Q2 (July to September), Q3 (October to December), and Q4 (January to March). We calculated the body mass index percentile by sex and age from the height and body weight. Those with a body mass index above the 85th percentile were categorized as overweight [21]. We asked participants to report their organized sports participation. Those who participated in least one organized sport (e.g., school sports club activity, neighborhood sports group activity, private sports lesson) were categorized as “Active,” and those who did not participate in any sports activities were categorized as “Inactive.”
We used the International Physical Activity Questionnaire for Japanese Early Adolescents to assess students’ physical activity levels. We also asked students, “How often do you engage in physical activity per week?” and “How long do you engage in physical activity per day on average?”, regarding moderate physical activity (MPA) and vigorous physical activity (VPA) [22]. Based on a previous study, moderate to vigorous physical activity (MVPA) per day was calculated as follows [23].
$$MVPA = \left\{(MPA frequency \times MPA duration) + (VPA frequency \times VPA duration)\right\} / (7 days)$$
Neighborhood-level characteristics
Each board of education in Japan determines the school district based on geographical conditions (e.g., streets and rivers), neighborhood traditions, and residents’ preferences [24]. We defined a school district as a neighborhood unit in this study because the size of the school district corresponded to the daily living area [25, 26]. Public school students in Japan must go to their designated school as per their residential address, and they are instructed not to go out of their school district without their guardians [27].
To apply the results of the national study data collected by the municipalities or by the block (cho-cho-aza), we conducted weighting interpolation with a geographic information system (Fig. 2). First, we overlapped a block-level neighborhood factor and school district polygon data. Then, we computed scores by the ratio of the size of the overlapped area per size of each school district. We calculated the mean of the overlapped area in the school district as the school district level score.
In Japanese culture, asking someone’s academic background or income is frowned upon. A previous study in Japan also reported that only a few participants reported their academic background and income [28]. Thus, we substituted three neighborhood socioeconomic factors: areal deprivation, neighborhood education level, and average annual income. Areal deprivation is an index that reflects the relative size of poor household ratio. We used the Areal Deprivation Index (ADI), a weighted index wherein the following eight variables were associated with poverty from the Population Census in 2010: proportion of elderly single households, elderly couple households, single mother households, rental housing households, sales and service workers, agricultural workers, blue-collar workers, and unemployed persons [29,30,31]. We overlapped block-level ADI and school district data and calculated the mean of each block that was composed in a school district as neighborhood ADI. In addition, we obtained data regarding income and estimated the average annual income from the Housing and Land Survey in 2013 and the Population Census in 2015 [32, 33]. The division of the population according to income consisted of six classes: less than 3 million yen, 3–5 million yen, 5–7 million yen, 7–10 million yen, 10–15 million yen, and more than 15 million yen. We multiplied the class value by the number of households in each income class, summed up the product, and divided it by the number of general households in the school districts. We estimated block-level income data by overlapping the municipality-level income data obtained from the Housing and Land Survey in 2013 and block-level population data from the Population Census in 2015. Then, we overlapped block-level income data and school district data and calculated neighborhood-level income data. Furthermore, we calculated the proportion of people who graduated from university or graduate school from the Population Census in 2015 to yield neighborhood education levels. Finally, we referred to population data from the Population Census in 2015 [32] and calculated the population density of school districts.
Statistical analysis
We estimated lacking data by multiple imputation (the frequency of multiple imputation was five) and calculated the average imputed score for the analysis. In addition, MVPA was normalized by a Box-Cox transformation.
Since this study included both individual-level and neighborhood-level variables, we used multilevel modeling. First, we examined only the birth month (Model 1) to calculate the interclass correlation coefficient. Then, we added each socioeconomic factor to Model 1 (Model 2; Model 2a: Areal deprivation; Model 2b: Average annual income; Model 2c: neighborhood education level). Furthermore, we examined the cross-level interaction between birth month and each socioeconomic factor (Model 3). As for Model 3, birth month was included as a random effect. To prepare for multilevel regression analysis, we used the centering method for all independent variables and covariates.
We conducted multilevel logistic regression analysis to examine whether adolescent organized sports participation was associated with birth month and socioeconomic factors. We estimated a 95% confidence interval (95% CI) using the Wald test. We also ran multilevel linear regression analysis to clarify whether adolescent MVPA was related to birth month and socioeconomic factors. If statistical significance was observed, we conducted a simple effect analysis [34]. To express cross-level interaction, we estimated a single slope of the birth month at mean ± 1 standard deviation [35].
Males are more likely to participate in sports and engage in physical activity more frequently than females [36, 37]. Additionally, males are more likely to demonstrate the relative age effect than females [38]. Therefore, all models considered sex. We adjusted for the following covariates: age, body weight, and population density. We conducted statistical analysis using SPSS 28.0 and EZR (Easy R) [39], and statistical significance was set at p < 0.05.