Data sources
This cross-sectional study was conducted using the 2016 Ethiopian Demographic and Health Survey (EDHS) dataset, a nationally representative survey. It was implemented by the Central Statistical Agency (CSA). Data collection took place from January 18, 2016, to June 27, 2016 (22). The study is conducted in compliance to the STROBE cross sectional reporting guidelines (33)
Population and sampling procedures
The 2016 EDHS data was collected using a sampling frame from the Ethiopian population and housing census which was conducted in 2007 by the Ethiopian CSA. The survey was designed to represent the country as a whole, for urban and rural areas separately. All of the nine administrative regions and the two city administrations in Ethiopia were included in the survey. A two-stage stratified cluster sampling was used. Each region and one city administration were stratified into urban and rural, except Addis Ababa, which is entirely urban. In a total, 21 sampling strata were created. A total of 645 enumeration areas (202 in urban areas and 443 in rural areas) were selected with probability proportional to size of enumeration areas and with independent selection in each sampling stratum. In the second stage, a fixed number of 28 households per cluster were selected with an equal probability systematic selection from newly created household listing (22).
Data collection and anthropometric measurements
EDHS survey data was collected using structured interviewer administered questionnaire. All men aged 15-59 years who were either permanent residents of selected households or visitors who stayed in the household the night before the survey were eligible to be interviewed for the male version of the questionnaire. The interview was carried out using tablet computers to record responses. Height and weight of all men aged 15-59 were measured. Weight was measured using light weight SECA mother-infant scales with digital screen designed and manufactured under guidance of UNICEF. Height measurements were carried out using a Shorr measuring board. A detailed explanation of data collection procedures can be found elsewhere (22).
Outcome variable
Overweight and obesity is the outcome variable of this study which was derived from body mass index (BMI) data. BMI was calculated as weight in kilogram divided by square of height in meters. According to the World Health Organization classification, BMI of <18.5 kg/m2, 18.5-24.99 kg/m2, 25-29.99 kg/m2, and ≥30 kg/m2 was categorized as underweight, normal, overweight, and obese respectively (2). Individuals with either overweight or obesity were combined into one category and coded as “1” and others were coded as “0”.
Independent variables
The independent variables were classified as either individual level or community level factors. Individual level variables included variables such as age (categorized into 15-29 years, 30-44 years, 45-59 years), marital status (categorized as single, married/living with partner, and widowed/divorced/separated), educational level (labeled as no education, primary education, secondary education, and higher than secondary education), current employment status (categorized as employed or unemployed), frequency of watching television (categorized as not at all, less than once a week, at least once a week), listening to the radio (categorized as not at all, less than once a week, at least once a week), and reading magazine (categorized as not at all, less than once a week, at least once a week).
Community level variables were administrative region and level of “rich” households in the community level. Ethiopia was divided into nine regions and two city administrations. These regions were re-categorized based on the settings associated with the prevalence of overweight and obesity (34). Accordingly, region was defined as “Metropolis” (containing Addis Ababa, Harari and Diredawa), Tigray, Amhara, Oromia, and SNNPR and “Other” (Afar, Benshangul-Gumuz, Gambela and Somali). Level of “rich” households in the community was defined as the proportion of men in richer and richest households within each cluster. The household wealth index variable was used, which indicates a household's cumulative living standard and includes items like owning a bicycle, television, or refrigerator. In the EDHS data, the majority of urban dwellers are very rich. This variable was categorized as high rich or low rich using a median split. Then the proportion of high rich households for each cluster was calculated.
Data analysis procedures
Data analysis was conducted using STATA version 14. To calculate the overweight or obesity status of men, we merged anthropometric variables from household members recode (PR file) to the men’s recode (MR file) using the cluster, household and line numbers (35). After merging both datasets, a total of 2259 weighted sample of men aged 15-59 years living in urban areas were included in the final analysis. Sampling weight was used during data analysis to adjust for non-proportional allocation of sample and possible differences in response rates across regions included in the survey. No missing data was present. Due to the hierarchical nature of the EDHS data and presence of intra-class correlation (ICC) multilevel logistic regression was used instead of ordinary logistic regression. Both bi-variable and multi-variable multilevel logistic regression was performed to assess the independent effect of the individual and community level variables on overweight and obesity. Independent variables with p-values less than 0.05 in bi-variable analysis were entered into the multi-variable logistic regression. Fixed effects were reported using adjusted odds ratio (AOR) with 95% confidence intervals (CI). Random effect parameters were measured by ICC, proportional change in variance (PCV) and median odds ratio (MOR), which measure the variability between clusters in the multilevel models. ICC explains the cluster variability, while MOR can quantify unexplained cluster variability (heterogeneity). MOR translates cluster variance into OR scale. In the multilevel model, PCV can measure the total variation due to factors at the community and individual level (36). Model comparison was conducted for the null model (model without explanatory variables), model 1 (Model adjusted for individual level factors), model 2 (Model adjusted for community level factors), and model 3 (Final model adjusted for both individual and community level factors). Multi-collinearity among independent variables was checked using variance inflation factors (VIF).
Ethical considerations
The Ethiopian Demographic and Health Survey was conducted after the approval of the Ethiopian National Research Ethics Review Committee. Permission to use the 2016 EDHS data for further analysis was sought from http://www.dhsprogram.com and no ethics committee approval was necessary. The data was analyzed and reported in aggregate; household and individual identifiers were not reported in the dataset.