A Brief Conceptual Tutorial on Multilevel Analysis in Social Epidemiology: Investigating Contextual Phenomena in Different Groups of People

Queen Mary, University of London, Londinium, England, United Kingdom
Journal of Epidemiology & Community Health (Impact Factor: 3.5). 10/2005; 59(9):729-36. DOI: 10.1136/jech.2004.023929
Source: PubMed


(1) To provide a didactic and conceptual (rather than mathematical) link between multilevel regression analysis (MLRA) and social epidemiological concepts. (2) To develop an epidemiological vision of MLRA focused on measures of health variation and clustering of individual health status within areas, which is useful to operationalise the notion of "contextual phenomenon". The paper shows how to investigate (1) whether there is clustering within neighbourhoods, (2) to which extent neighbourhood level differences are explained by the individual composition of the neighbourhoods, (3) whether the contextual phenomenon differs in magnitude for different groups of people, and whether neighbourhood context modifies individual level associations, and (4) whether variations in health status are dependent on individual level characteristics. Design and
Simulated data are used on systolic blood pressure (SBP), age, body mass index (BMI), and antihypertensive medication (AHM) ascribed to 25 000 subjects in 39 neighbourhoods of an imaginary city. Rather than assessing neighbourhood variables, the paper concentrated on SBP variance between individuals and neighbourhoods as a function of individual BMI.
The variance partition coefficient (VPC) showed that clustering of SBP within neighbourhoods was greater for people with a higher BMI. The composition of the neighbourhoods with respect to age, AHM use, and BMI explained about one fourth of the neighbourhood differences in SBP. Neighbourhood context modified the individual level association between BMI and SBP. Individual level differences in SBP within neighbourhoods were larger for people with a higher BMI.
Statistical measures of multilevel variations can effectively quantify contextual effects in different groups of people, which is a relevant issue for understanding health inequalities.

Download full-text


Available from: Juan Merlo
  • Source
    • "Several models were built, following the approach of Merlo et al. [42] First, an empty model was built. This investigated the random effect of school on the outcome of interest, and explains how much of the variation in MVPA is explained by school alone. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Little is known about patterns of physical activity engaged in by youth after they immigrate to a new country. This study aims to investigate relationships between immigrant generation and ethnicity with physical activity, and to determine if the relationship between immigrant generation and physical activity was modified by ethnicity. The data sources were Cycle 6 (2009-2010) of the Canadian Health Behaviour in School-Aged Children Study and the 2006 Canada Census of Population. Participants (weighted n = 23,124) were young people from grades 6-10 in 436 schools. Students were asked where they were born, how long ago they moved to Canada, their ethnicity, and how many days a week they accumulated at least 60 minutes of moderate-to-vigorous physical activity (MVPA). Youth born outside of Canada were less likely to be active than peers born in Canada; 11% vs 15% reported 7 days/week of at least 60 minutes of MVPA (p = .001). MVPA increased with time since immigration. Compared to Canadian-born youth, youth who immigrated within the last 1-2 years were less likely to get sufficient MVPA on 4-6 days/week (odds ratio: 0.66, 95% confidence interval: 0.53-0.82) and 7 days/week (0.62; 0.43-0.89). East and South-East Asian youth were less active, regardless of time since immigration: 4-6 days/week (0.67; 0.58-0.79) and 7 days/week (0.37; 0.29-0.48). Time since immigration and ethnicity were associated with MVPA among Canadian youth. Mechanisms by which these differences occur need to be uncovered in order to identify barriers to physical activity participation among youth.
    Full-text · Article · Feb 2014 · PLoS ONE
  • Source
    • "Two measures of variation between municipalities were used. The variance partition coefficient (VPC) represents the proportion of the variance (VA) that is due to the difference between municipalities and is calculated as follows: VPC = VA/(VA + 3.29) [33]. The median odds ratio (MOR) is defined as “the median value of the odds ratio between the area at highest risk and the area at lowest risk when randomly picking out two areas” [33]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Despite an increased use of contraceptive methods by women, unintended pregnancies represent one of the most evident violations of women's sexual and reproductive rights around the world. This study aims to measure the association between individual and community exposure to different forms of violence against women (physical/sexual violence by the partner, sexual abuse by any person, or controlling behavior by the partner) and unintended pregnancies. Data from the 2006/2007 Nicaraguan Demographic and Health Survey were used. For the current study, 5347 women who reported a live birth in the five years prior to the survey and who were married or cohabitating at the time of the data collection were selected. Women's exposure to controlling behaviors by their partners was measured using six questions from the WHO Multi-Country Study on Women's Health and Domestic Violence against Women.Area-level variables were constructed by aggregating the individual level exposures to violence into an exposure measurement of the municipality as a whole (n = 142); which is the basic political division in Nicaragua. Multilevel logistic regression was used to analyze the data. In total, 37.1% of the pregnancies were reported as unintended. After adjusting for all variables included in the model, individual exposure to controlling behavior by a partner (AOR = 1.28, 95% CrI = 1.13-1.44), ever exposure to sexual abuse (AOR = 1.31, 95% CrI = 1.03-1.62), and ever exposure to physical/sexual intimate partner violence (AOR = 1.44, 95% CrI = 1.24-1.66) were significantly associated with unintended pregnancies. Women who lived in municipalities in the highest tertile of controlling behavior by a partner had 1.25 times higher odds of reporting an unintended pregnancy than women living in municipalities in the lowest tertile (AOR = 1.25, 95% CrI = 1.03-1.48). Nicaraguan women often experience unintended pregnancies, and the occurrence of unintended pregnancies is significantly associated with exposure to different forms of violence against women at both the individual and the municipality level. National policies aiming to facilitate women's ability to exercise their reproductive rights must include actions aimed at reducing women's exposures to violence against women.
    Full-text · Article · Feb 2014 · BMC Women's Health
  • Source
    • "Taking into account that the variances at individual and area level were not directly comparable in the two-level logistic regression, variance partition coefficient (VPC) was calculated by means of latent variable method as the proportion of variation that is attributable to the county-level sources. Besides, median odd ratios (MOR) and proportional change in variance (PCV) were also obtained to estimate the area level and individual level variation respectively [14,15]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Childhood undernutrition adversely impacts child health and is one of China's largest health burdens. However, there is limited information on the current rate of childhood undernutrition in rural Western China. The purpose of this study was to investigate the prevalence of childhood undernutrition and explore its association with socio-economic characteristics in Western China. A total of 13,532 children of 0 ~ 36 months of age were recruited as subjects from 45 counties and 10 provinces in Western China with a 3-stage probability proportion to size sampling. The composite index of anthropometric failure (CIAF) was used to assess the childhood undernutrition. The association between socio-economic characteristics and childhood undernutrition was analyzed using a two-level logistic regression. Based on CIAF, the prevalence of undernutrition among children under three years of age in rural Western China in 2005 was 21.7%. The two-level logistic analysis presented a large difference in undernutrition among the 10 provinces with the highest odds ratio in Guizhou (OR: 2.15, 95%CI: 1.50, 3.08). Older children had a higher prevalence of undernutrition. As compared to girls, boys were more likely to be undernourished (OR 1.27, 95% CI: 1.16, 1.39). The likelihood of undernutrition was lower in subjects of Han ethnicity as opposed to subjects of minority ethnicities (OR 0.77, 95%CI: 0.65, 0.90). In addition, the education levels of the mother as well as wealth index were both negatively associated with childhood undernutrition. Childhood undernutrition still remains a large health challenge in rural Western China. This study has important policy implications for the Chinese government to improve childhood undernutrition in the surveyed areas.
    Full-text · Article · Feb 2014 · BMC Public Health
Show more