Data Sources and Methodology

Background on the Development of the Conceptual Framework

The Food Systems Dashboard framework is used to define and describe food systems by summarizing the drivers, components, and outcomes of food systems. The framework depicts four components of food systems – food supply chains, food environments, individual factors, and consumer behavior – which are all interrelated. The framework also shows how these components impact diets and, ultimately, nutrition and health outcomes. Additionally, it depicts the drivers that push and pull food systems - climate change, globalization and trade, income growth and distribution, population growth and migration, politics and leadership, and socio-cultural context.

This framework was adapted from the conceptual framework developed by the High-Level Panel of Experts on Food Security and Nutrition (HLPE), the science-policy interface of the United Nations Committee on World Food Security (CFS) in their Nutrition and Food Systems report. The HLPE framework aimed to illustrate how food systems influence diets and nutrition, emphasizing the importance of diets in linking food systems with nutrition and health outcomes. The HLPE framework identified three foundational components of food systems - food supply chains, food environments, and consumer behavior. It stressed the importance of food environments in consumer food choices. Additionally, the HLPE framework depicted how food systems impact environmental, social, and economic sustainability. The HLPE framework is now being used to inform the United Nations Voluntary Guidelines on Food Systems for Nutrition.

In adapting the HLPE framework for the Dashboard, we added a fourth component of food systems, individual factors, to capture the economic, cognitive, aspirational, and situational factors that define people and their behavior. We also expanded upon the food environments subcomponents to include additional elements of food product properties as well as food vendor properties. Additionally, among the drivers, we narrowed these down as well as adding income growth and distribution. Our framework aims to describe food systems, the complicated interactions between food systems components, and the impacts on diets, nutrition, and health.

Indicators and Their Sources

The Dashboard includes 166 indictors that are mapped across the framework. We identified indicators with high-quality data for countries at all income levels, when available, for each of the food systems components and subcomponents in the framework. When indicators were not available, we included that in the Dashboard to advocate for more data collection in these areas, as understanding each component is critical for a complete understanding of food systems. The 166 indicators are from over 20 different sources, including both those that are publicly available such as the Food and Agriculture Organization (FAO), Global Burden of Disease (GBD), United Nations Children's Fund (UNICEF), World Health Organization (WHO), World Bank, Climate Watch, and others as well as those that are not publicly available, such as Euromonitor International.

More information about each indicator can be found in the metadata, which can be accessed by clicking on the “i” next to the indicator name in both Compare and Analyze and Country Profiles. The metadata includes the indicator definition, its relevance to food systems and why it was included in the Dashboard, for calculated indicators, how this calculation was made, how missing values are treated in the Dashboard, the indicator source with a link to this source, and a link to Additional Information when applicable. This metadata Aims to provide a complete understanding of the indicators and their importance.

For more details on data indicators, the gaps, and future areas, please read the paper published in Nature Food’s May 2020 issue.

Methodology for Developing the Food Systems Typology

The dashboard typology was developed to identify common patterns among country food systems. By grouping food system into types, policymakers, practitioners, and researchers (all potential users of the dashboard) may explore these broad patterns without needing to conduct an in-depth assessment of each individual country’s data. Together with their stylized descriptions, these food system types may enhance understanding of the transformations country food systems are undergoing and enable generation of hypotheses related to their effects on diets and nutrition. However, it is important to note that this country-level typology cannot capture the additional heterogeneity in food systems that exists sub-nationally, across geography and different food commodities. Food systems are dynamic, with multiple types often existing in parallel, and transforming over time.

Like the design of the dashboard itself, the development of the typology was guided by our conceptual framework, aiming to characterize food systems according to their core elements, including: food supply chains, food environments, consumers, and external food system drivers.

The following steps were undertaken:

1. Literature review

A preliminary literature review was conducted to assess how previous typologies related to food systems had been developed. Searches were conducted during the summer of 2018 in PubMed, Web of Science, and Scopus databases. The literature review sought to explore the different components of food systems that typologies aimed to characterize, the variables used to define the typologies, and the methods used to develop typologies.

The search yielded 61 relevant articles. Illustrative examples of the typologies are included in the table below. Typologies describing the entire food system rather than an individual component were most relevant for the dashboard. These typologies tended to utilize theoretical, nonempirical methods of categorizing food systems, or alternatively, a relatively small number of indicators for which thresholds or cutoffs could be applied.

Food Supply ChainsFood EnvironmentConsumersFood System
Farming households:
  • Alvarez et al (2018)
  • Andersen et al (2007)
  • Lopez-Ridaura et al (2018)
Farming and agricultural production systems:
  • Gunia et al (2010)
  • Le Noe et al (2017)
  • Madry et al (2013)
  • Niles et al (2017)Supply and value chains:
  • Carbone (2018)
  • Gómez and Ricketts (2013)
  • Nsamzimshuti et al (2018)
  • Smith (2008)
  • Tudisca et al (2014)
Food outlets:
  • Tyrrel et al (2016)
Retail or neighborhood food environments:
  • Hutchinson et al (2012)
  • McInerney et al (2016)
  • Mezuk et al (2016)
  • Timperio et al (2018)
  • Zhang, van der Lans, Dagevos (2012)
Food, physical activity, and social environments:
  • Myers, Denstel, and Broyles (2016)
Note: Urban Food SystemsCountry food systems:
  • Ericksen (2007)
  • IFPRI (2015)
  • HLPE (2017)
  • McCullough, Pingali, and Stamoulis (2008)
  • UNEP (2016)
Agri-food system:
  • Baer-Nawrocka and Sadowski (2019)
  • Pingali, Ricketts, and Sahn (2015)
Urban food systems:
  • Tefft et al (2008)

2. Selection of methods for constructing the typology

The dashboard typology most closely resembles those presented in IFPRI (2015) and HLPE (2017) in its methods, which both made use of a core set of indicators to group countries into categories (food system types). This method was preferred over non-empirical methods, which may be more subjective and more difficult to operationalize. IFRI (2015) separated countries into types using quantiles, while HLPE (2017) separated countries based on data values that were above or below the median values. More robust, data-driven methods are available for building typologies, including cluster analysis and latent variable modeling. These approaches were tested by the dashboard team but were found to be limited by sample size and skewed distributions with extreme outliers. The dashboard typology utilizes quintiles to categorize countries, similar to IFPRI (2015), as described below.

3. Prioritization of indicators

Prioritization of the specific indicators to include in the typology were based on the following criteria: 1) the group of indicators chosen should reflect the different components of the food system, as put forth in the conceptual framework; 2) the literature should support the indicators’ association with food system patterns and transitions; and 3) indicators should have high coverage among countries included in the dashboard.

Based on these criteria the following four typology indicators were used:

  • Agriculture value added per worker (constant 2010 USD)[1]
  • Share of dietary energy from cereals, roots, and tubers[2]
  • Number of supermarkets per 100,000 population[3]
  • Percent urban population of total population[4]

IFPRI (2015) and HLPE (2017) also used agriculture value added per worker, share of dietary energy from staples, and urbanization in their typologies – the only new indicator added for the dashboard typology was supermarkets per 100,000 population. Supermarkets influence how supply chains are organized and how consumers shop for food, therefore, this indicator was considered a critical gap to fill.

4. Ranking on individual typology indicators

151 countries were included in the typology, those for which there was no missing data across the four typology indicators. For each of these indicators, countries were ranked from highest to lowest, under the hypothesis that higher values were associated with more “modern” food systems, and lower values more “traditional” food systems. The ranking was inverted in the case of the share of dietary energy from cereals, roots, and tubers, which is theorized to be lower in more modern food systems.

5. Score and categorization of five food system types

A score for each country was assigned based on the sum of its ranks on each of the four indicators. For example, if a country ranked 10th on agriculture value added, 15th on share of dietary energy from cereals roots and tubers, 17th on number of supermarkets per 100,000 population, and 8th on urbanization, it’s score was 50. Once scores were calculated for each country, all countries were sorted from lowest score to highest score. The typology was created by separating the distribution of scores into quintiles, with the lowest quintile representing the most modern food system type and the highest quintile representing the most traditional food system type. Our team pre-selected five food system types as the most appropriate number that would capture heterogeneity in country food systems, while remaining significantly distinct and easy to communicate. For example, typologies based on three food systems will not pick up important differences within regions, while more than five typologies may include too much overlap between types.

The median and interquartile ranges (25th percentile to 75th percentile) are include in the table below:[5]

Agricultural value added per worker, constant 2010 USDShare of dietary energy from staplesNumber of supermarkets per 100,000 populationPercent of total population living in urban areas
Rural and traditional814 (522 – 1,218)0.67 (0.58 – 0.71)0.34 (0.24 – 0.45)0.34 (0.24 – 0.37)
Informal and expanding2,428 (1,559 – 3,344)0.58 (0.51 – 0.62)0.65 (0.51 – 1.83)0.52 (0.43 – 0.59)
Emerging and diversifying5,511 (3,907 – 6,955)0.46 (0.41 – 0.51)4.00 (2.02 – 5.85)0.57 (0.50 – 0.68)
Modernizing and formalizing14,382 (10,519 – 20,331)0.39 (0.35 – 0.43)7.15 (3.92 – 14.25)0.76 (0.66 – 0.84)
Industrialized and consolidated53,180 (27,842 – 80,456)0.29 (0.27 – 0.31)13.97 (10.73 – 22.70)0.83 (0.77 – 0.91)

6. Validation

Aims to provide a complete understanding of the indicators and their importance.

For the first hypothesis, we examined the distribution of other food system indicators—percentage of the rural population with a financial account, nutrition functional diversity of agriculture production, retail value of ultra-processed food sales (logarithm), and income elasticity for food, beverages, and tobacco—by food system type. These distributions were displayed in box plots (see examples below). While substantial overlap tended to exist among neighboring food system types, for the most part, the median values followed the patterns that we would expect. (Note: 1 – rural and traditional; 2 – informal and expanding; 3 – emerging and diversifying; 4 – modernizing and formalizing; and 5 – industrialized and consolidated).

For the second hypothesis, we conducted the same exercise for prevalence of stunting, prevalence of anemia among women of reproductive age, prevalence of obesity among adult women, and NCD death rate per 100,000 population. These distributions also revealed trends across food system types that well aligned with diet and nutrition outcomes associated with the nutrition transition.

7. Stylized descriptions

The dashboard team also developed a set of narrative descriptions to accompany the food system typology. These descriptions are meant to aid users to visualize what the food system types look like and also to emphasize the dynamic elements within each type.

8. Limitations

In addition to the issues previously raised related to the inability of our typology to capture the full complexity of food systems at the subnational level, there are a couple limitations associated with the method used to construct the typology. First, quintiles assume an equal number of countries per typology, which may not actually be the case. Second, we used a rank-based score rather than a score based on actual data values, meaning that the typology is driven more by the rank order of countries rather than the actual indicator distributions. An equal interval method of categorizing countries would avoid these issues, however, due to skewness, results in a typology that is difficult to communicate. For example, two of three countries with extreme outlier values could be assigned their own unique food system type. Our chosen method seeks to balance the desire for an empirical approach with the desire for a typology that can be easily communicated.

References

  1. World Bank (2018). World Bank DataBank.
  2. FAO (2012). FAOSTAT, Suite of Food Security Indicators.
  3. Euromonitor International (2018). Passport database.
  4. National Population Division (2018). World Bank DataBank.
  5. Note that the dashboard will use population weighted averages, therefore typology means included in the dashboard will differ from this table, which presents median values.

How the 42 Actions Were Identified

Overview of our Methods

The starting point for identifying the actions was a review of major international evidence-based expert reports on food systems which include detailed recommendations on how to orient food systems towards healthier diets. A set of inclusion criteria was rigorously applied to select the reports, leading to eight being included from the original 45. A list of all the recommended actions was recorded in detail. A pathway-to-impact was formulated to assess if each action could plausibly have impact on the availability, affordability, appeal/acceptability of food. Actions were reworded where necessary so they clearly communicate the specific way they can lead to healthier diets. This led to a list of 42 specific actions with the potential to improve diets if fully implemented.

Detailed Methods

Step 1. Identify evidence-based expert reports

The first step was to identify evidence-based expert reports with clearly articulated recommendations for reorienting food systems towards healthier diets. An initial list of potential reports was prepared drawing on reports already known to the research team and an extensive review of additional compilations and searches on relevant websites. The result was a list of 45 potentially relevant reports, all published since 2013.

Each report was reviewed against four criteria for inclusion: (1) deals explicitly with how food systems can be reoriented towards healthier diets in at least part of the report (2) makes detailed recommendations of policies and actions on how to reorient food systems towards healthier diets (3) provides an evidence informed review of the topic with references to research studies; and (4) indication of peer-review.

Recommendations had to be action-oriented and specific – broader perspectives that were not accompanied by specific actions were not included (e.g. “Align policies with health outcomes” or “Improve the science-policy interface”). Reports which referenced nutrition, but only included generic recommendations on food security (e.g. producing more food or increasing rural incomes) rather than specifically improving diets were excluded. Reports focused on food and the environment were largely excluded on the basis that recommendations were primarily related to either (1) changing consumer behaviours to improve environmental sustainability (e.g. shifting demand for meat to combat climate change) and/or (2) changing agricultural production methods to improve the environment (e.g. different farming methods). From the original list of 45 reports, eight met all of the criteria.

Step 2. Extract actions

The reports were reviewed in detail to identify recommended actions. Actions were included that aimed to increase the availability, affordability, appeal, nutritional quality or safety of nutritious foods and/or decrease the availability, affordability or appeal of foods, snacks and beverages high in energy, sugars, fat and/or salt. Recommendations with no clear pathway towards availability, affordability, etc. were not included. For example, a handful of reports made recommendations on decreasing the amount of antibiotics used in animals to combat anti-microbial resistance. While this is important for improving the overall health of the population, there was no clear pathway presented for how this would specifically improve the availability, affordability, etc. of nutritious foods. Similarly, actions that focused only on changing production methods with no clear pathway to impact were not included. Recommended actions were entered into a spreadsheet using near-verbatim language to how it was presented in the report.

Step 3. Combine and consolidate similar actions

Many reports recommended similar actions, but with different details on effective implementation. For example, one report recommended donating city land for urban farming, another report recommended providing training programmes and funding for women to farm urban land while another recommended building markets exclusively for urban-grown food. These various details were combined into a more comprehensive action on delivering urban agriculture through funding, training and provision of inputs. In order to ensure that recommendations were combined in such a way that did not lose the original intent or detail from the report, each one was taken through a decision tree to determine if the original report intended the specific recommendation to be part of a more comprehensive approach. If it was, it was combined into a more comprehensive action; if not it was included as its own action. The effect of this process was to produce a list of actions that were both specific and broad enough to plausibly have impact. The final outcome of this process was the list of 42 Actions.

Step 4. Further clarify and refine actions according to their pathway to impact

Each of the 42 Actions were then sense-checked against a pathway to impact to ensure their impact on availability, affordability, etc. was clear. A pathway to impact was written for each action detailing how the action was expected to impact (1) the supply chain, (2) food environments, (3) consumer behaviour and (4) consumption. The wording of the actions was then further clarified and refined to ensure it clearly communicated the specific way in which the action could lead to healthier diets.