Adolescence is defined as the transitional phase between childhood and adulthood, which takes place between about 8 and 25 years of age. This phase is characterized by dramatic physical changes and associated mental turmoil, and is the stage during which individuals overcome a range of psychological conflicts and form their own self-image.

In this study, we considered the hypothesized relationships between factors in terms of Bandura’s social cognitive theory, which integrates a range of factors that affect health behavior and behavioral changes [14] , and may explain smoking behavior among adolescents.

We utilized the main concepts of social cognitive theory and devised a hypothesis of the relationships between the environment, individual factors, and early smoking initiation based on the interactions between these factors (Figure 1).

Based on the hypothetical framework of the study, we extracted factors related to smoking by minors from previous studies on this subject that have identified such factors. We searched for original research papers published during the previous 10 years, between January 2005 and June 2016, that included the Japanese key words kitsuen (“smoking”) and yōin (“factors”) in the Ichushi Web Version 5 (hereafter “Ichushi”) Japanese language medical database, and the English key words “initiation”, “smoking”, “factor,” and “children” in the PubMed and CINAHL databases. These searches returned 578 results in Ichushi, 63 in PubMed, and 23 in CINAHL. We extracted factors related to smoking by minors from the titles and abstracts of original research papers. In most cases, the participants of these studies were minors, but in others, they were adults who had started smoking while minors. Three of the overseas studies were excluded because they were written in languages other than English, and four because they were duplicates. A total of 64 papers were finally identified, 27 Japanese studies and 37 from overseas, from which, 228 items were extracted as factors in smoking by minors (Table 1). The items extracted were analyzed by qualitative recursive analysis on the basis of the similarity of their semantic content, and then combined into 110 items. These were then classified deductively into the following 10 categories corresponding to the main concepts of Bandura’s social cognitive theory: physical environment, social environment, situation, behavioral capability, observational learning, reinforcements, expectations, expectancies, self- efficacy, and emotional coping responses. Next, proposed items for the questionnaire were selected. Their content and means of expression were then investigated under the supervision of an investigator experienced in the preparation

of questionnaires using qualitative procedures. The items were then amended to deal with duplicate meanings and issues with their content or means of expression, leading to the final selection of 105 items.

A preliminary survey was carried out on 74 university students aged ≥ 20 years recruited by the snowball sampling method during November and December 2016 to investigate face and content validity. The responses were measured on a 5-point Likert scale by asking the respondents to rate how closely each item was connected to their having started smoking (from “1. No connection” to “5. Very closely connected”). To confirm the appropriateness and consistency of the content of questions, the correlation coefficients between questionnaire items were evaluated by using Item-Total (I-T) correlation analysis, and items with a correlation coefficient of <0.3 were eliminated. Pairs of questions among the proposed 105 items with a correlation coefficient of ≥0.7 were shown to have similar content, and either one of the pair was eliminated or the two were changed to use common expressions. Respondents were asked whether any questions were difficult to answer or whether any of the expressions were difficult to understand, and the expressions used were amended at the same time. This resulted in a revised version with 72 items.

Responses were received from 1010 of the 1597 study participants (63.2% response rate). The participants were 235 individuals who had tried smoking and 775 never-smokers. Responses with no missing data were received from 215 of those who had tried smoking (91.5% valid response rate), and this group was included in the final analysis. The mean age of the participants, who included 170 men (79.1%) and 45 women (20.9%), was 22.3 ± 2.7 years.

Mean age at smoking initiation was 16.9 ± 3.0 years, minimum age was 8 years, and maximum age was 23 years. There were 119 current smokers (55.3%), and the mean number of cigarettes smoked per day was 9.7 ± 7.2. In addition, 123 participants (57.2%) had family members who smoked, and 129 (60%) had undergone smoking prevention education (Table 2).

Ten items were eliminated because their I-T correlation was ≤0.3. Descriptive statistics were calculated, and five items for which either the ceiling or the floor effect applied were eliminated. Six pairs of questions with correlation coefficients of >0.7 were judged to have similar content, and one of each pair was eliminated.

Factor analysis (maximum-likelihood estimation and promax rotation) was performed for the 54 items remaining after 18 items had been eliminated by item analysis. Items for which the factor loading was <0.4 were eliminated, and ultimately, five factors comprising 18 items were extracted, with a cumulative contribution ratio of 59.3%. The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was 0.86, and Bartlett’s test of sphericity was significant at p < 0.01 (Table 3).

The Factor 1, which included items such as a personality susceptible to anxiety or stress and having negative feelings toward oneself, was named self-disgust. Factor 2, which included items such as having been disciplined by violence and living in poverty, was named growing up in an inappropriate home environment. Factor 3, which included items such as a lack of interest in smoking on the part of parents or local communities, was named reduction of norms consciousness for adolescents’ smoking. Factor 4, which included items such as smoking as stress relief and distraction, was named perception of smoking as stress relief.

Factor extraction method: Maximum likelihood method. Rotation method: Promax rotation with Kaiser normalization. KMO = 0.86, Bartlet’s test of sphericity was significant p < 0.01.

Factor 5, which included items such as lack of knowledge about smoking and being unaware of the difficulty of giving up smoking, was named inadequate knowledge of the dangers of smoking (Table 3).

Cronbach’s α coefficient was calculated to confirm the reliability of the internal consistency of each of these factors and the questionnaire items as a whole. Cronbach’s α for all 18 items was 0.89, and the values for the five factors were within the range 0.73 - 0.87 (Table 3).

Construct validity was investigated by comparing the five factors adopted as a result of the exploratory factor analysis with the elements of social cognitive theory used in the hypothetical framework, that is, environment and individual factors. Factors 1, 3, and 4 included items related to individual factors, and factors 2 and 3 included items related to environment.

Based on the proposed hypothesis, we developed a causal model with five latent variables and early smoking initiation as the dependent variable. Subsequently, we searched for causal models with a high degree of explanation. At that time, we attempted to correct from the paths with smaller degrees of explanation using the Wald test. We successfully identified a causal model consisting of the five latent variables (Figure 2).

The fitness indices of this model were GFI = 0.86, AGFI = 0.82, CFI = 0.90, and RMSEA = 0.07. Although these could not be described as sufficiently high, they did demonstrate a high level of fit, and the model was judged to be valid.

The only factor in the selected model that showed a significant direct correlation with early smoking initiation was inadequate knowledge of the dangers of smoking.

Growing up in an inappropriate home environment increased self-disgust and

reduction of norms consciousness for adolescents’ smoking, while self-disgust increased perception of smoking as stress relief, and reduction of norms consciousness for adolescents’ smoking increased perception of smoking as stress relief and inadequate knowledge of the dangers of smoking.

All of these factors acted indirectly to lower the age of smoking initiation via inadequate knowledge of the dangers of smoking. In particular, reduction of norms consciousness for adolescents’ smoking had a greater effect than other factors on indirectly lowering the age of smoking initiation via inadequate knowledge of the dangers of smoking.

In this study, we used a specially produced questionnaire to investigate factors related to smoking initiation during adolescence. In terms of the reliability of the questionnaire items, the I-T correlation coefficient was maintained, while the factor structure extracted by exploratory factor analysis met the standards of the cumulative contribution ratio, the KMO measure of sampling adequacy, and Bartlett’s test of sphericity. Cronbach’s α for each factor was also high, within the range 0.73 - 0.87, indicating that the internal consistency of the factor structure was maintained. These results suggested that the items in the specially produced questionnaire were valid for investigating “factors related to smoking initiation in adolescence”.

Our results contribute to an understanding regarding whether, and if so, how, factors related to smoking initiation in adolescence are related to early smoking initiation.

The factors extracted on the basis of factor analysis―self-disgust [11] [15] [16] [17] , growing up in an inappropriate home environment [18] - [24] , reduction of norms consciousness for adolescents’ smoking [21] [22] [24] [25] [26] [27] , perception of smoking as stress relief [17] [20] [24] [28] [29] , and inadequate knowledge of the dangers of smoking [17] [30] ―have all been identified as factors affecting smoking in various previous studies.