The current study was a cross-sectional study of elementary school children in the 5th year in state-run schools in Saitama, Japan. The research protocol was reviewed and approved by the Research Ethics Committee at Saitama Prefectural University (No. 19078).
Procedures
This study was conducted collaboratively between Saitama Prefectural University and the Saitama educational board. The measures described below were distributed to the children between April and December 2019. Each school district chose the periods for administration. Prior to distribution, school district officers visited the schools, described the aim of this survey to school teachers, and obtained agreement to conduct the survey. School district officers explained the aim and asked children to participate in a homeroom session while the school teachers facilitated students’ group activities that were not relevant to the subjects or teaching. Data collection was completed by teachers in each classroom.
Participants
The educational board manages 802 state-run elementary schools (99% of total elementary schools) in Saitama. Those state-run elementary schools are grouped into 72 school districts matched to municipalities. We targeted 8 school districts (selecting 2 school districts from the southern, northern, western, and eastern areas). Chairpersons of the school districts provided permission to implement the survey.
Children begin elementary education at six years of age in Japan. Thus, elementary school children in the 5th year were either 10 or 11 years old during the data collection period. There were 8045 eligible children who were in the 5th year for this survey (Supplemental material Table). The district with small school numbers employed no school counselors. Unlike European countries with inclusive policies, children who require special support in school because of physical, psychological, and intellectual issues attend segregated special schools [4]. No school providing special education was included in this survey.
Hypothesized model
We hypothesized that a positive perception towards attending school was influenced by positive relationships with friends and school teachers, positive perceptions of current circumstances, positive subjective health, and having someone to share experiences and thoughts with (Fig. 1). To examine the model, an original questionnaire was developed through a collaboration between members of the Saitama educational board and faculty whose specialty was education. Several meetings took place between the members of the development team to finalize the questionnaire. Items were developed based on the literature, national guidelines and the expertise of the development team. It was strongly felt by the development team, given their experience working in and around schools, that a large number of questions would be a burden to the participants; thus, a shorter tool was desirable.
Data collection
There were 14 items in the original questionnaire. The first 13 items targeted intrapsychic and interpersonal determinants of positive perceptions towards attending school. The 14th item listed potential persons who were accessible for the participants to share their experiences.
To capture how children felt about attending school, the item wording was “I am looking forward to going to my school.” The responses were 4 = strongly agree, 3 = agree, 2 = disagree, and 1 = strongly disagree.
To capture positive relationships with friends and school teachers, the following five items were positively worded: “I play with friends a lot,” “I get along well with friends,” “Other kids like me,” “School teachers approve of my efforts” and “I am happy when talking to school teachers.” For the five items, the responses were 3 = very much, 2 = a little bit, and 1 = not at all.
To capture the positive perceptions of current circumstances, the following four items were positively worded: “I am able to prepare for classes by my own,” “I am proud of myself,” “I am helpful to others” and “I believe my efforts will be rewarded.” For the item about being able to prepare for classes independently, the responses were 4 = very well, 3 = fairly, 2 = poorly, and 1 = not at all. For other items, the responses were 3 = very much, 2 = a little bit, and 1 = not at all.
To capture subjective health, the following three items were employed: “My health status,” “I get tired easily” and “I have anxieties and/or worries.” For the general health status, the responses were 4 = good, 3 = fair, 2 = not very good, and 1 = not good at all. For the item for tiredness, the responses were 3 = not tired at all, 2 = sometimes tired, and 1 = frequently tired. For the item about anxieties and/or worries, the responses were 3 = no anxiety/worry at all, 2 = a little bit of anxiety/worry, and 1 = a lot of anxiety/worry.
To capture the persons with whom the children shared their experiences and thoughts as the 14th item, the following eight responses were selected: (a) my mother, (b) my father, (c) my siblings, (d) other family members/relatives, (e) school teachers, (f) a school counselor, (g) instructors of extracurricular activities, and (h) friends. The item also asked the participants the frequency of accessing those persons. The responses were 3 = frequently, 2 = sometimes, 1 = rarely, and 0 = there is no such person.
Data analysis
The statistical data analysis included descriptive statistics, Cramer’s V, Spearman’s rank correlation coefficient, and a structural equation model (SEM). The analysis was conducted with HAD 17.0 [29], SPSS v.26 for Japanese (IBM, Japan), and Mplus version 7.3 [30].
Before conducting the SEM, we observed the data characteristics. Because the data were collected from 8 school districts, we explored individual-level variance and group-level (school district-level, in this study) variance [31]. Group homogeneity was identified with intraclass correlation coefficient (ICC). A high ICC results in a biased error variance in conventional regression models, overestimating the relationship between variables. An ICC of 0.25 and higher indicates that much of the variation in the dependent variables is due to the features of groups rather than the characteristics of individuals [32]. In our study, we employed a more conservative value of 0.05 [33] because we collected data from a large sample.
After confirming the ICC values, we observed sample descriptive data with individual-level variance. We compared boys and girls according to the survey items. For the comparisons, we used Cramer’s V, which indicates how strongly two categorical variables are associated, with 1 indicating a strong association and 0 indicating no association. Values of 0.1, .03, and 0.5 are considered to be small, medium, and large effect sizes, respectively [34]. We also used Spearman’s rank correlation coefficient between the variables. Coefficients were interpreted as limited (0.00 to 0.25), fair 0(.25 to 0.50), moderate (0.50 to 0.75) and excellent (0.75 to 1.0) [35].
SEM with ordinal data was conducted using the modified weighted least squares method (WLSMV). We examined the hypothesized model using all data obtained by the original questionnaire and then, modified the model. The model fit indices were comparative fit index (CFI), Tucker Lewis index (TLI), and root mean square error of approximation (RMSEA). For CFI and TLI, a value higher than 0.9 is the best model fit. For RMSEA, a value of 0.05 and smaller is a close fit, a value of 0.08 and smaller is a reasonable fit, and a value of 0.1 and higher is a poor fit [36]. The RMSEA value was supplemented with a 90% confidence interval (90% CI).