Clarifying the use of aggregated exposures in multilevel models: self-included vs. self-excluded measures

Contextual fallacy in MLMs with cross-level interaction: A critical review of neighborhood effects on psychiatric resilience

In the multilevel modeling literature, contextual effect is defined as or identified by the effect of the target group-level variable while controlling for the corresponding individual-level variable. This paper extends the notion of "contextual effects" (or "neighborhood" or "school" effects) to an interaction setting, such that the effect of one explanatory variable X_ij on the outcome Y_ij is modeled as a function of a group-level 'moderating' or predisposing variable Z_j* as well as its counterpart at the individual level Z_ij. Researchers frequently use regression models that only contain a cross-level interaction between X_ij and Z_j* to test contextual hypotheses in an interaction setting, but this modeling strategy is unable to discriminate the immediate rival hypothesis that attributes a causal role to the corresponding individual-level variable. This paper points out the prevalence of this type of fallacy through a review of past research on contextual determinants of psychiatric resilience. It is argued that the simple step of adding an appropriate individual-level interaction X_ijZ_ij could help more robustly test substantive hypotheses about how neighborhood context alters the effect of proximal stressors on health outcomes.

Mental Health in Adolescents with a Migration Background in 29 European Countries: The Buffering Role of Social Capital

Previous research is inconclusive as to whether having an immigration background acts as a risk factor for poor mental health in adolescents, and furthermore, what contribution the social context in which adolescents grow up may make. To address these questions, the current study uses an integrative resilience framework to investigate the association between immigration background and adolescent mental health, and the moderating role of social capital at the individual, the school, and the national level. The study uses data gathered from nationally representative samples of adolescents aged 11, 13, and 15 years (N_girls = 63,425 (52.1%); M_age = 13.57, SD = 1.64) from 29 countries participating in the 2017/18 Health Behaviour in School-aged Children (HBSC) study. Data analysis reveals that first- and second-generation immigrants reported higher levels of life dissatisfaction and psychosomatic symptoms than their native peers, and that this association varied across schools and countries. In addition, social capital was found to moderate the association between immigration background and adolescent mental health. Individual-level social support from peers and family and national-level trust protected against poor mental health in adolescents with an immigration background, while the opposite was true for individual-level teacher support. Supportive teacher-student relationships were found to provide more protection against poor mental health for native adolescents than for immigrant adolescents. Our findings indicate the importance of taking an ecological approach to design interventions to reduce the negative effects of having an immigration background on adolescent mental health.

Workplace social capital and refraining from seeking medical care in Japanese employees: a 1-year prospective cohort study

OBJECTIVES

We examined the association of workplace social capital (WSC), including structural and cognitive dimensions, with refraining from seeking medical care (RSMC) among Japanese employees.

DESIGN

One-year prospective cohort study.

SETTING AND PARTICIPANTS

We surveyed 8770 employees (6881 men and 1889 women) aged 18-70 years from 12 firms in Japan using a self-administered questionnaire comprising the WSC scale and the items on potential confounders (ie, age, educational attainment and equivalent annual household income) at baseline (from April 2011 to March 2013).

OUTCOME MEASURES

At a 1-year follow-up, we measured RSMC using a single-item question 'In the past year, have you ever refrained from visiting a hospital, clinic, acupuncturist or dentist despite your sickness (including a slight cold or cavity) or injury?'

RESULTS

The results of Cox regression with robust variance showed that, after adjusting for potential confounders, the low WSC group (ie, the lowest tertile group) had a significantly higher relative risk (RR) of RSMC compared with the high WSC group (ie, the highest tertile group) among both men and women (overall WSC: RR 1.09 (95% CI 1.01 to 1.17) and 1.20 (95% CI 1.06 to 1.37); structural dimension: RR 1.13 (95% CI 1.04 to 1.22) and 1.25 (95% CI 1.07 to 1.45); and cognitive dimension: RR 1.11 (95% CI 1.03 to 1.20) and 1.21 (95% CI 1.06 to 1.38), respectively). Trend analysis using a continuous score of the WSC scale also showed a significant association of low WSC with a higher risk of RSMC among both men and women.

CONCLUSIONS

Our findings suggest that the lack of social capital in the workplace is associated with RSMC among Japanese employees.

Psychometric assessment of a scale to measure bonding workplace social capital

OBJECTIVES

Workplace social capital (WSC) has attracted increasing attention as an organizational and psychosocial factor related to worker health. This study aimed to assess the psychometric properties of a newly developed WSC scale for use in work environments, where bonding social capital is important.

METHODS

We assessed the psychometric properties of a newly developed 6-item scale to measure bonding WSC using two data sources. Participants were 1,650 randomly selected workers who completed an online survey. Exploratory factor analyses were conducted. We examined the item-item and item-total correlations, internal consistency, and associations between scale scores and a previous 8-item measure of WSC. We evaluated test-retest reliability by repeating the survey with 900 of the respondents 2 weeks later. The overall scale reliability was quantified by an intraclass coefficient and the standard error of measurement. We evaluated convergent validity by examining the association with several relevant workplace psychosocial factors using a dataset from workers employed by an electrical components company (n = 2,975).

RESULTS

The scale was unidimensional. The item-item and item-total correlations ranged from 0.52 to 0.78 (p < 0.01) and from 0.79 to 0.89 (p < 0.01), respectively. Internal consistency was good (Cronbach's α coefficient: 0.93). The correlation with the 8-item scale indicated high criterion validity (r = 0.81) and the scale showed high test-retest reliability (r = 0.74, p < 0.01). The intraclass coefficient and standard error of measurement were 0.74 (95% confidence intervals: 0.71-0.77) and 4.04 (95% confidence intervals: 1.86-6.20), respectively. Correlations with relevant workplace psychosocial factors showed convergent validity.

CONCLUSIONS

The results confirmed that the newly developed WSC scale has adequate psychometric properties.

Social Capital and Suicidal Ideation in Community-Dwelling Older Residents: A Multilevel Analysis of 10,094 Subjects in Japan

OBJECTIVE

Social capital, the collective resources of groups including perceptions of trust and reciprocity, is recognized as an important contributor to suicide. We examined the association of individual- and community-level social capital with suicidal ideation after adjusting for social support among older adults living in the community.

METHODS

In August 2010 we sent questionnaires to all residents aged 65 years and older living in 3 rural municipalities (N = 21,232) in Okayama Prefecture, Japan; 13,919 questionnaires were returned (response rate: 65.6%). The final analysis included 10,094 participants. The outcome variable was suicidal ideation. Exposure variables were individual-level mistrust and lack of reciprocity (level 1), and the aggregated responses of these variables from 35 communities in the municipalities (level 2). Covariates included age, sex, educational attainment, marital status, the number of cohabitants, years of residence, self-rated socioeconomic status, disability, social support, and psychological distress. Multilevel logistic regression analysis was performed to obtain odds ratios (ORs) and 95% confidence intervals (CIs).

RESULTS

After adjusting for social support and psychological distress, we found that mistrust and lack of reciprocity were only associated with suicidal ideation at the individual level. Stratified analysis showed that among subjects with psychological distress, mistrust was associated with suicidal ideation at individual (OR: 1.88; 95% CI: 1.42-2.51) and community levels (OR: 1.98; 95% CI: 1.02-3.81).

CONCLUSIONS

Our findings show that individual- and community-level social capital is a possible protective factor for suicidal ideation, particularly for people with psychological distress.

School- and Individual-level Predictors of Weight Status Misperception among Korean Adolescents: A National Online Survey

Background

Growing body of literature has reported that weight status estimation pattern, including accurate-, under-, and overestimation, was associated with weight related behaviors and weight change among adolescents and young adults. However, there have been a few studies investigating the potential role of school contexts in shaping adolescents’ weight status estimation pattern among Korea adolescents.

Objective

The aim of the present study was to investigate the association between weight status misperception patterns and factors at individual-, family-, and school-level, simultaneously, and whether there was significant between schools variation in the distribution of each weight status misperception pattern, underestimation and overestimation respectively, among Korean adolescents aged 12–18 years.

Method

Data from the Eighth Korea Youth Risk Behavior Web-based Survey (KYRBS), 2012, a nationally representative online survey of 72,228 students (boys = 37,229, girls = 34,999) from a total of 797 middle and high schools were used. Sex stratified multilevel random intercept multinomial logistic models where adolescents (level 1) were nested within schools (level 2) were performed.

Results

At the school level, attending a school with higher average BMI (kg/m²) was positively associated with weight status underestimation, and inversely associated with weight status overestimation among boys and girls. Single-sex schooling was positively associated with weight status underestimation among girls. At the family level, higher household income (high/middle versus low) was inversely associated with both weight status under- and overestimation among boys and girls. Higher maternal education (equal to or more than college graduate versus equal to or less than high school graduate) was positively associated with weight status overestimation among boys, and living with both parents (compared to not living with both parents) was inversely associated with weight status underestimation among girls. At the individual level, high academic achievement (compared to low) was positively associated with weight status underestimation among boys and girls.

Conclusions

While further research with prospective designs and objectively measured anthropometric information is needed, school environmental factors such as sex composition and school average BMI, as well as, family contexts such as socioeconomic status need to be considered when developing and implementing obesity prevention programs.

Comparison of Social Trust's effect on suicide ideation between urban and non-urban areas: The Case of Japanese Adults in 2006

An increasing number of studies have addressed the determinants of suicide. Social capital is a key factor in preventing suicide. However, little is known about the experience of suicide ideation using subjective values. From the viewpoint of suicide prevention, it is worth examining how people think of suicide. This paper attempts to examine the effect of social capital on suicide ideation. Furthermore, the paper compares the effect of social capital between urban and non-urban areas. In this paper, urban areas are equivalent to mega-cities with populations over one million. Non-urban areas are cities with populations of less than one million, towns and villages. Individual-level data from the Japanese General Social Surveys (JGSSs) are used. The survey, which was conducted in 2006, provides information about the subjective value of suicide ideation. The survey was answered by 1413 subjects with a mean age of 54.5. Of the subjects, 49% were male. Social trust is used to measure the degree of social capital, and the outcome of interest is suicide ideation within the past 5 years. After controlling for various factors, the major findings are that both individual-level social trust and social trust accumulated in one's residential administrative district reduce the probability that one will consider suicide. After dividing the sample into urban and non-urban residents, particularized trust plays a role in deterring suicide ideation in urban areas, while generalized trust plays a role in deterring suicide ideation in non-urban areas. The effect of each type of trust depends on its scarcity in residential areas.

Contextual influence on condom use in commercial sex venues: A multi-level analysis among female sex workers and gatekeepers in Guangxi, China

This study aims to assess the influence of commercial sex venues on consistent condom use among female sex workers (FSWs) and to examine associations between individual and venue level factors and consistent condom use with clients. Analysis was based on a sample of 637 FSWs and 123 gatekeepers from 51 venues in Guangxi, China. Multi-level logistic regression using Bayesian simulation via Markov Chain Monte Carlo was applied to investigate whether FSWs' individual propensity to use condom with clients was statistically dependent on the venue of working. Multi-level modeling revealed considerable variability across venues in the likelihood of consistent condom use with clients among FSWs. Characteristics at both individual and venue levels helped to explain the observed variation. Certain venue-level factors exerted their influence on condom use over and above the effect of individual-level characteristics. The contextual influence exerted on condom use behaviors among FSWs may imply a potential to harness the path to individual behaviors from a higher and more dominant level, and shed light on the design of more effective sexual risk reduction intervention among venue-based FSWs.

Individual and contextual factors associated with verbal bullying among Brazilian adolescents

Background

Few studies have been carried out in low- middle-income countries assessing contextual characteristics associated with bullying. This study aimed to assess the relative importance of contextual (school and city) and individual-level factors to explain the variance in verbal bullying among a nationally representative sample of Brazilian adolescents.

Methods

59,348 students from 1,453 schools and 26 state capitals and the Federal District participated in the National Survey of School Health among 9th Grade Students (PeNSE, 2009). We performed multilevel logistic regression in a three level model (individual, school and city).

Results

The 30-day prevalence of verbal bullying among these students was 14.2%. We found that 1.8% and 0.3% of the total variance in bullying occurred at school-level and city-level, respectively, and 97.9% at individual-level. At city-level, all factors included failed to demonstrate a significant association with bullying (p <0.05) whereas at school-level, private schools presented more bullying than public schools (OR = 1.17, CI 1.04-1.31). At individual-level, male gender, younger age, not living with both parents, exposed to domestic violence, under or overweight were all associated with bullying.

Conclusions

All socioeconomic indicators assessed contributed little to explain the variance in bullying at individual, school or city-level. Population subgroups at risk identified according to their individual profile could be targeted in future interventions in Brazil.

Community economic status and intimate partner violence against women in bangladesh: compositional or contextual effects?

In this research, we used a multi-level contextual-effects analysis to disentangle the household- and community-level associations between income and intimate partner violence (IPV) against women in Bangladesh. Our analyses of data from 2,668 women interviewed as part of the World Health Organization (WHO) multi-country study on women's health and domestic violence against women showed that household income was negatively associated with women's risk of experiencing IPV. Controlling for residence in a low-income household, living in a low-income community was not associated with women's risk of experiencing IPV. These results support a household-level, not community-level, relationship between income and IPV in Bangladesh.

Group involvement and self-rated health among the Japanese elderly: an examination of bonding and bridging social capital

Background

To date, only a small amount of research on bonding/bridging social capital has separately examined their effects on health though they have been thought to have differential effects on health outcomes. By using a large population-based sample of elderly Japanese people, we sought to investigate the association between bonding and bridging social capital and self-rated health for men and women separately.

Methods

In August 2010, questionnaires were sent to all residents aged ≥65 years in three municipalities in Okayama prefecture (n = 21232), and 13929 questionnaires were returned (response rate: 65.6%). Social capital was measured from survey responses to questions on participation in six different types of groups: a) the elderly club or sports/hobby/culture circle; b) alumni association; c) political campaign club; d) citizen’s group or environmental preservation activity; e) community association; and f) religious organization. Participant perception of group homogeneity (gender, age, and previous occupation) was used to divide social capital into bonding or bridging. Odds ratios (ORs) and 95% confidence intervals (CIs) for poor self-rated health were calculated.

Results

A total of 11146 subjects (4441 men and 6705 women) were available for the analysis. Among men, bonding and bridging social capital were inversely associated with poor self-rated health (high bonding social capital; OR: 0.55, 95% CI: 0.31–0.99; high bridging social capital; OR: 0.62, 95% CI: 0.48–0.81) after adjusting for age, educational attainment, smoking status, frequency of alcohol consumption, overweight, living arrangements, and type-D personality. The beneficial effect among women was more likely limited to bonding social capital (high bonding social capital; OR: 0.34, 95% CI: 0.12–1.00), and the association between bridging social capital and self-rated health was less clear (high bridging social capital; OR: 0.69, 95% CI: 0.44–1.07).

Conclusions

Bonding/bridging social capital could have differential associations with self-rated health among the Japanese elderly depending on the individual’s sex. Considering the lack of consensus on how to measure bonding and bridging social capital, however, we need to carefully assess the generalizability of our findings. Further research is warranted to identify health-relevant dimensions of social capital in different cultural or economic settings.