The country specific data were collected for this ecological study:
The WHO Global Health Observatory (GHO) data
The WHO Global Health Observatory (GHO) data on estimated prevalence rates of obesity and overweight (percent of population aged 18+ with BMI ≥ 30 and 25 kg/m2 respectively) and on mean BMI of the population aged 18+ by country was obtained for the year 2010 . We did not use the most recent version of three levels of BMI (BMI = 30, BMI = 25 and mean BMI) in 2014, but used the 2010 year data because of other key variables of interest (described below). We included overweight prevalence and mean BMI in our study in case meat availability was a late-stage predictor of obesity.
We also captured the estimated prevalence rate of physical inactivity for each country for the population aged 18+ . The estimated prevalence rate of physical inactivity is defined as percent of defined population attaining less than 150 minutes of moderate-intensity physical activity per week, or less than 75 minutes of vigorous-intensity physical activity per week, or equivalent.
The GHO is an initiative of the WHO to share data on global health, including statistics by country and information about specific diseases and health measures. The GHO specifically assembles prevalence data of the biological risk factors, including obesity, overweight and mean BMI for WHO Member States using standardized protocols (http://www.who.int/gho/ncd/methods/en/).
The FAOSTAT Food Balance Sheet (FBS) data
The FAOSTAT Food Balance Sheet (FBS) data on major food group availability per capita per day of: i) total meat; ii) starch crops (mixed cereals and starchy root); iii) fibers (vegetables and pulses); iv) fats (plant oils and animal fats) and v) fruits . The food items in each food group are indicated in the Supporting Information (Additional file 1: Table S1).
We also extracted the availability of grand total calories and macro-nutrients of fats (animal and plant, in g/capita/day) and proteins (animal, plant and meat, in g/capita/day) from FBS for our study. As animal protein includes meat protein, we subtracted meat protein from the animal protein to obtain the variable, “Animal protein, excluding meat protein” for more precise data analysis. Following the Atwater system , we calculated the energy from carbohydrates using the formula: carbohydrates energy per day = total calories- fat (grand total, in gram/day) × 9 – protein (total, in gram/day) × 4. For carbohydrates availability in g/capita/day, we used the energy in kilocalories (kcal) divided by 4. Because obesity develops after cumulative exposure to dietary risks (i.e. high intake of risk food groups today does not lead to immediate obesity, but a prolonged exposure to high intake of risk food type(s) is required.), we calculated the mean grams per person per day over a 3-year period (2007–2009) in each of these food categories to represent typical long-term exposure to each of these dietary components. The rationale for this decision is that studies have shown that three years is a practical period to develop metabolic syndrome leading to obesity after exposure to dietary risks (i.e. high intake of meat today does not lead to immediate obesity) [25–27]. Using the mean of three years of nutrients and food groups may also reduce the random errors during the data collection and calculation by FAO.
The FAOSTAT database disseminates statistical data collected and maintained by the FAO. FAOSTAT data are provided as a time-series from 1961 in most domains through the Food Balance Sheet (FBS, http://faostat3.fao.org/home/E). The FBS presents a comprehensive picture of the pattern of a country's food supply during a specified reference period. The FBS shows for each food item i.e. each primary commodity availability for human consumption which corresponds to the sources of supply and its utilisation. The total quantity of foodstuffs produced in a country added to the total quantity imported and adjusted to any change in stocks that may have occurred since the beginning of the reference period gives the supply available during that period. On the utilisation side a distinction is made between the quantities exported, fed to livestock + used for seed, losses during storage and transportation, and food supplies available for human consumption. The per capita supply of each such food item available for human consumption is then obtained by dividing the respective quantity by the related data on the population actually partaking in it .
Minimum Dietary Energy Requirements, expressed as kcal per person per day, is the weighted average of the minimum energy requirements of the different gender-age groups in the population with light activity. Grantham et al. reported that when a mixed meal of protein, carbohydrate and fat is consumed, carbohydrates and fats are digested faster and metabolised to satisfy body’s energetic needs while slower digested protein is ultimately and stored as fat . Therefore, we extracted the Minimum Dietary Energy Requirements from the FAO website (http://www.fao.org/) and compared it and with the energy from carbohydrates and fats by country to see if the energy from the proteins is the surplus.
The World Bank data
The World Bank dataset measures progress on aggregate outcomes for member countries for selected indicators. GDP PPP is gross domestic product converted to international dollars using purchasing power parity rates (http://data.worldbank.org/indicator/NY.GDP.PCAP.PP.CD) . GDP PPP is the measure of average income in constant 2010 $US adjusted for purchasing power parity for cross-country comparability.
WHO, FAO and the World Bank are intergovernmental organizations using specialized information relevant to their respective fields. Their professional personnel should have evaluated these data in consideration of their possible use, e.g. for scientific research and decision making, before they were published. Therefore, the data reporting is as free of bias and error as it can be with government statistics. This means that errors are reduced but some inaccuracies related to reporting quality may still be present in the data. Similar data from the same sources were recently used to analyse the relationships between nutrients and obesity [31, 32] and diabetes [33–35] in a number of publications.
We obtained data for 170 countries after we matched the prevalence estimates of obesity and overweight and mean BMI to the year-and country-specific food and other variables. Each country was treated individually as the subject and all their availability for other variables information was analysed. The detailed information of country-level estimates is in the Supporting Information (Additional file 2: Table S2).
For particular analyses, the number of countries included may have differed somewhat because all information on other variables was not uniformly available for all countries due to unavailability from relevant UN agencies. All the data were extracted and saved in Microsoft Excel® for analysis. Data sources and summary statistics are further described in the Supporting Information (Additional file 3: Table S3).
The prevailing dogma of obesity is that obesity is an affluence related medical conditions , which is generally caused by eating too much (too much calories intake)  and moving too little (physically inactive) . Therefore, in this study we used GDP PPP, total calories and prevalence of physical inactivity as the potential confounders and the other variables are divided into two sets, i.e. major food group and macronutrient for data analysis in 5 steps.
Spearman rank correlation analyses was used to evaluate the strength and direction of the associations between food group and macronutrient availability for consumption and prevalence estimates of overweight and obesity and mean BMI.
Partial correlation was used to find the unique variance between each food group and macronutrient and prevalence of obesity and overweight and mean BMI respectively while eliminating the variance from total calories, GDP PPP and physical inactivity. In order to show the independent correlation of meat and meat protein to the three variables defined by BMI (BMI ≥ 30, BMI ≥ 25 and mean BMI) respectively, we controlled for three potential confounders (total calories, GDP PPP and physical inactivity) plus all other food groups and all other macronutrient variables respectively for partial analysis.
Stepwise multiple linear regression modelling was performed to identify and rank predictors (independent variables) of prevalence of obesity, overweight and mean BMI respectively from two sets of data of food groups and macronutrients respectively.
Scatter plots were used to explore the relationship between meat and meat protein (both GDP adjusted) and three variables defined by BMI. Scatter plots were also used to explore the relationship between prevalence of obesity and each food group and macronutrient respectively.
Human diet patterns varying in different food components may be affected by the types of food availability in a particular region, socio-economic status and cultural beliefs. In order to demonstrate that correlation universally exists between meat availability and obesity regardless of these factors, countries were grouped for correlation analyses. The criteria for grouping countries the World Bank income classifications , WHO regions , countries sharing specific characteristics like geography, culture, development role or socio-economic status, like Latin America and the Caribbean (LAC) , Organisation for Economic Co-operation and Development (OECD) , Asia-Pacific Economic Cooperation (APEC) , Southern African Development Community (SADC) , the Arab World , Latin America (LA), and Asia Cooperation Dialogue (ACD) . All the country listings are sourced from their official websites for matching except LA which is self-classified based on region primarily speaking romance languages. Countries included in LA are listed in the Supporting Information (Additional file 4: Table S4).
SPSS v. 22 (SPSS Inc., Chicago Il USA) was used for data analysis and the statistical significance was set at the 0.01 level (two-tailed). Prior to analysis data were log-transformed to bring their distributions close to normal.