Smoking, alcohol abuse, unhealthy eating habits and low level of physical activity are great risk factors for a large number of serious conditions and diseases [2].
All municipal upper schools (161 schools) in 21 Swedish municipalities (out of 290 municipalities in Sweden) were asked to participate in the study. These municipalities were strategically selected to ensure a geographical and socio-demographical spread among the participants in the study. The selected municipalities included both urban (city with a population of 50,000 or more or a municipality near a city of this size where a high percentage commute to the city) and non-urban-populated municipalities (less than 50,000 inhabitants and not a municipality near a city of at least 50,000 inhabitants where a high percentage commute to the city) as defined by the Swedish Association of Local Authorities and Regions [55]. Nineteen schools in 14 municipalities agreed to participate. Ten schools (in nine municipalities) were chosen, representing equal distribution of schools and pupils in urban and non-urban-populated municipalities, which were geographically spread out (805 registered pupils, 412 in non-urban-populated and 393 in urban-populated municipalities). The final sample in the study consisted of 492 pupils (61%) (in school grades 8 and 9 (ages 15–16 years) from compulsory school). Non-responses were owed to one school failing to return their questionnaires (123 questionnaires), teachers failing to distribute the questionnaires (151 questionnaires) and pupils who were absent on the particular day that the questionnaire was distributed (39 pupils). Out of the 492 pupils, 240 adolescents lived in urban-populated municipalities (49%) and 252 adolescents in non-urban-populated municipalities (51%). This sample is representative of the Swedish population with regard to geographical spread, education level of parents and number of pupils with passing certificates upon graduation from grade 9 [55,56,57].
Procedure
An 82-item questionnaire was handed out to pupil participants who completed it using paper and pencil (in approximately 30 min) in the classroom during class hours. The questions were divided into subsections with different subheadings. Twenty-six of the questions were used in this cross-sectional study (see Additional file 1). Twenty-five of these questions had been used in previous studies and one was developed for this study (‘self-perceived economic situation’). To ensure the validity of the questions in the questionnaire, a pilot study was conducted in a school class (grade 9). This school did not provide data for the current study. A test–retest was performed on the same school class. A few questions and answering alternatives were then reformulated. A second pilot study was performed in four school classes to test the validity of the edited questions. The test–retest was performed to ensure reliability of the questions in the questionnaire that were developed by the research team. The test–retest analysis was performed in SPSS Statistics (version 17.0) using cross-tabulation and Spearman correlation and the reliability was found to be adequate (Correlation coefficients ranging from 0.6 to 0.9).
Ethics, consent and permissions
The first page of the questionnaire provided information about the study’s aim, confidentiality, informed consent (which was given by answering the questionnaire), voluntary participation and the option to withdraw from the study at any time. Questionnaires were returned in June 2009.
The Swedish law of ethical regulations and guidelines for humanistic and social science research [58] were followed in this study. The study was performed according to the Declaration of Helsinki and the ethical standards of the ethics committee at the Faculty of Medicine at Uppsala University, Sweden. Following ethical standards and this law (Law, 2003:460), the ethics confirmed that the study was exempt from requiring ethical approval.
Study variables
Variables
Health-related behavioural variables
Meal frequency was selected to measure eating habits. Three items assessed meal frequency (‘How often do you eat the following meal during a regular week? Breakfast? Cooked lunch? Cooked food in the evening?’ (Cronbach’s Alpha 0.671) (Additional file 1)). One question for each health-related behaviour was used for the other health-related behaviours: ‘How often do you exercise in your spare time for more than 30 minutes so that you get out of breath or sweat?’ ‘Do you smoke?’ and ‘Do you drink alcohol so that you become drunk?’
Psychosocial condition variables
How often should you see a new guy?
He told Bravo TV, "It's important to stick to twice a week only so that you have plenty of time away from your new date to give your deepest...
Subjective well-being was measured by two commonly used variables [59, 60] (‘How do you feel?’ and ‘How satisfied are you with your life?’) (Cronbach’s Alpha 0.800). Social relationships were measured using the interpersonal distrust subscale (7 questions) from the Eating Disorder Inventory-Children (EDI-C) [61], Swedish version [62,63,64]. The variable ‘social relationships’ (referring to the quality of relationships, for example, trust and social support) consisted of the total score for the subscale. Items measuring social relationships included, for example, ‘I trust others’ and ‘I have close friends’. Self-esteem was measured with the ineffectiveness subscale (10 questions) from the EDI-C [61], Swedish version [62,63,64]. The variable ‘self-esteem’ consisted of the total score on the subscale and included questions that measured factors such as feelings of autonomy and control (for example, ‘I feel that I can attain the things I try to’ and ‘I feel that I can control things in my life’). As recommended for these validated subscales [61], mean values were used to replace isolated missing responses on the interpersonal distrust subscale and ineffectiveness subscale. These subscales have good internal consistency [61] (Cronbach’s Alpha was 0.760 for the interpersonal distrust subscale and 0.881 for the ineffectiveness subscale).
Socio-demographic variables
Demographic variables included age and gender (Additional file 1). Adolescents’ self-perceived economic situation was also assessed (using the question ‘How often do you feel that you have less money than your peers?’); this is a good measure of socio-economic status in relation to health-related behaviours [65].
First-order latent variables
Health-related behaviours, psychosocial condition variables and socio-demographic variables were all analysed as first-order latent variables (constructs consisting of one or more items from the questionnaire). We first determined whether items represented latent phenomena [66]; for example, whether the items ‘How do you feel?’ and ‘How satisfied are you with your life?’ represented subjective well-being. Variables for social relationships and self-esteem were constituted by the total scores on their respective subscales ‘interpersonal distrust’ and‘ineffectiveness’ (indicating quality of social relationships and level of self-esteem) in the EDI-C [61]. These variables were then included in the analysis as first-order latent variables. To use only one indicator for a latent variable is common practise in SEM analysis when this indicator alone measures what the authors intend to measure [66].
Second-order latent variable
It was hypothesised that vulnerability represented a latent phenomenon underlying the tendencies to smoke, consume alcohol, eat irregularly and refrain from physical activity (Fig. 1). A second-order latent variable [67,68,69], which is a construct that represents several first-order latent variables [68, 69] was used to test whether there was a shared underlying psychosocial vulnerability for this set of unhealthy behaviours (i.e. smoking, alcohol consumption, irregular meal frequency and low level of physical activity level).
Statistical analyses
The statistical programme LISREL (version 8.80) was used for measurement modelling analysis, correlation analysis and structural equation modelling (SEM) [67]. Maximum likelihood was used to deal with missing values. There were no differences in demographics, in terms of participants who were and were not missing data.
To confirm or reject whether certain indicator variables represented latent variables [66], and to examine the validity of these variables, measurement modelling was performed by confirmatory factor analysis measuring the degree to which each item significantly loaded (as indicated by path coefficients) onto its designated first-order latent variable [67]. Items measuring regular meal frequency of ‘breakfast’, ‘cooked lunch’ and ‘cooked food in the evening’ were tested to determine whether they loaded onto a first-order latent variable for ‘meal frequency’. Similarly, the items ‘How do you feel?’ and ‘How satisfied are you with your life?’ were tested to see if they were reliable measures of subjective well-being.
The definition of the unit of measurement, which has to be obtained when second-order latent variables are measured in SEM, was specified by fixing the unstandardised direct effect of the second-order latent variable and the first-order latent variable ‘smoking’ to 1.00 [67, 68]. ‘Smoking’ was chosen because it had the strongest loading with the second-order latent variable in the correlation analysis (Additional file 2) [69]. This scaling controls the remaining path coefficients between the second-order latent variable and its first-order (indicator) latent variables.
Who cheats more often male or female?
men In general, men are more likely than women to cheat: 20% of men and 13% of women reported that they've had sex with someone other than their...
The hypothesised model (Fig. 2) was tested by SEM analysis. The total scores of the social relationships scale and the self-esteem scale measured good social relationships and high self-esteem. For this reason, as well as to make the presentation of the results more legible, the direction of variables in the SEM model were analysed as presented in Fig. 2 (i.e., high well-being, good social relationships, high self-esteem, high socio-economic status, regular meal frequency and high level of physical activity were tested in the analysis with the expectation of negative associations with the vulnerability variable).
SEM analysis was chosen because it offers certain advantages. For example, when compared with regression analysis, SEM allows the researcher to hypothesise and test the strength of relationships between first- and second-order latent variables; i.e. associations with underlying factors may be estimated [68, 69]. SEM also allows complex model analyses, including mediating associations [68, 69]. Previous studies that are similar to the present one have used second-order latent variable modelling to measure the underlying structure of unhealthy behaviours [21, 70,71,72]. SEM analysis determined whether the variables outlined in the hypothesised model (Fig. 1) (which was based on theoretical and empirical evidence) were associated with each other. The strength of the hypothesised associations in the model was indicated by path coefficients [67]. Total associations (direct and indirect associations) between all variables in the SEM model were tested. The total path coefficients determined the associations between the individual health-related behaviours and each psychosocial condition variable, and indicated whether the psychosocial conditions and the socio-demographic groups reflected an underlying vulnerability to the set of unhealthy behaviours included in the study. As this is a cross-sectional study, however, the path associations are not causal.
Path coefficients of associations between variables in the measurement model analysis (see Results section) and the SEM analysis (see Additional file 3) were confirmed when they were statistically significant at 95% CI. The polychoric correlation associations were confirmed when they had significant p-values (Additional file 2). The measurement model, correlation analysis and SEM analysis were all evaluated by fit measures. These measures indicated how closely these analyses fitted the data. Good fit is indicated by a low root-mean-square error of approximation (RMSEA, acceptable fit below 0.08), non-significant χ2, high Goodness of Fit Index (GFI > 0.90), high Goodness of Fit Index Adjusted for df (AGFI >0.90) and low standardised root-mean-square residual (SRMR, acceptable fit below 0.08) [73].
Infatuation Rules
Photo: Alena Shekhovtcova
Promotion
Promotion
Promotion
Promotion
Promotion