Volume 33, Issue 1: Paper 1
Determinants of Food Insecurity in Africa: A Panel and Multi-level Analysis
Jonas Baune, Minnesota State University Moorhead
In the 1960’s President Lyndon Johnson called for the United States to “lead the world in a war against hunger.” Yet, in the 2020s, billions of people still suffer from food insecurity. Food insecurity offers a more nuanced view of hunger and starvation. Food insecurity includes people who suffer from starvation and insufficient caloric intake, but also includes people whose diets contain enough calories but lack essential minerals and vitamins. A lack of understanding prompts food insecurity to be viewed through a binary lens, with individuals and populations being viewed as either food secure or food insecure.
Often, food insecurity is viewed as a monolithic phenomenon, with considerable existing literature labeling someone as either food secure or food insecure. While these approaches become adopted due to data availability, they do not provide accurate insight into addressing food insecurity. For instance, someone may consume their required calories for the day; however, they may be deficient in essential vitamins or other health categories, resulting in detrimental health effects. To fill the research vacuum on depth of hunger, this research examines the causes of food insecurity at three distinct levels (World Bank) across 36 nations in Africa. The first and most basic level of food insecurity describes the percentage of a nation’s population that cannot afford a calorie sufficient diet, referred to as severe food insecurity. The second level considers the percentage of people who cannot afford a diet that meets essential nutritional requirements, such as calories, protein, and essential vitamins and minerals. This level is referred to as moderate food insecurity. The third and final level measures the percentage of people who cannot afford a diet that meets dietary standards given by governments and public health agencies, referred to as mild food insecurity. If different factors contribute to different levels of food insecurity, then policy recommendations can be targeted and enacted strategically. To complement this research, a panel regression explores the time aspect of food insecurity, noting variables that can affect food insecurity over time.
Based on relevant research and economic theory, this paper incorporates multiple cross-sectional OLS regressions and a panel regression to highlight potential factors of food insecurity. The cross-sectional regressions measure food insecurity as a dependent variable at three different levels (severe, moderate, and mild) to respond to the two research questions: 1) Are the factors that affect food insecurity the same at every level of food insecurity? and 2) Are these factors statistically significant over time?
This paper extends the general understanding of this topic, particularly regarding how different levels of food insecurity manifest themselves. Two important findings emerged from this research. Universal factors, which influence all three levels of food insecurity, can be identified. In addition, variables that affect alternative levels of food insecurity also exist.
II. Literature Review
Upon taking a focused look at the literature, several global and local variables vital to this study became evident. Global variables affect all nations, regardless of circumstance. In contrast, local variables affect some nations; however, it depends on the individual nation’s unique characteristics. To distinguish between global and local variables and gather actionable insights for future food security policies, researchers utilize a spatial non-parametric analysis technique called Geographically Weighted Regression (Sassi 2015). The local variable, rainfall, exhibits high importance in arid countries and milder importance in temperate climates. Statistically significant global variables related to the percentage of the population undernourished include GDP per capita, agricultural labor productivity, and arable land as a share of total land. In addition, the study finds evidence for regional factors affecting undernourishment as well, even recommending that African countries transition from individualized protectionist policies to more regionally designed policies instead. In this context, a foundation for relating global and local variables to different levels of food insecurity is established. Related research, described in the remainder of this section, exposes additional variables for inclusion in this analysis.
A study using stochastic simulation (Popp et al. 2019) compares the countries of Nigeria and Uganda with the goal of predicting if these countries could meet increased food demand in the future. Researchers utilize a system dynamics approach to predict food security, supplemented by scenario analysis to refine the results. Some relevant variables that the study examined include demand for food, fertility ratio, number of workplaces in agriculture and food industry, and agro-food export. The study also suggests a number of ways that may increase food security, including providing a social safety net for the most vulnerable population, increasing value added content, advancing transportation and storage capabilities, fostering political stability, managing water resources, and increasing arable land.
Multiple studies (Hesselberg and Yaro 2006; Agidew and Singh 2018) identify rainfall, or lack thereof, as a contributing factor to food insecurity. The first focuses on food insecurity determinants in a small region of Ghana (Hesselberg and Yaro 2006). By observing a few sectors in Northern Ghana, the study identifies a negative relationship between rainfall and food insecurity. Local residents also listed rainfall as an important factor affecting food insecurity, given that the study area experiences fluctuating and low rainfall. Additionally, household size and food insecurity display a positive correlation, which contrasts with other research on the topic. A study focusing on rural Ethiopia’s South Wollo Zone (Agidew and Singh 2018) recognizes the challenging dry climate with variable rainfall. The associated drought risk contributes to food insecurity in the region. The study features a case study approach, incorporating household survey data, to measure food insecurity. The case of the Telayayen sub-watershed provides insight regarding how food insecurity manifests itself in rural areas. Residents from these drier regions recognize the influence of rainfall on food insecurity.
HIV proves to be a surprising variable for inclusion in this analysis. Utilizing regression analysis, Benzekri et al. (2015) discover a positive correlation between HIV and food insecurity (Benzekri et al. 2015). Interestingly, the same study finds no link between food insecurity and household size, although the study did attribute that result to a limited sample size. The study’s authors also recommend including perspectives outside Africa to get a broader understanding of food insecurity.
Though not on the continent of Africa, Laos suffers from food insecurity in a manner similar to many African countries. In a study that used an ordinary least squares (OLS) regression to analyze variables affecting food security status, an indexed measurement that incorporates what and how much people eat emerges (Phami et al. 2020). The study finds negative relationships for household size, price of food, drought, and shock (combining unforeseen factors, at the household level, like disease, pests, and price increases). Interestingly, a positive relationship exists between the number of laborers in a household and the food security status of that household. This indicates that as children grow and begin to contribute more at the household level, they become a net benefit to a family’s food security status. In addition to this finding, a positive correlation exists between household monthly income, the number of laborers in a household, and gender of household head (with male-headed households being more likely to be food secure).
Literature with a focus on food insecurity centers around a national or sectorial perspective. The sheer volume of variables to be measured poses a challenge for such broad study. Education, rainfall, GDP per capita, and food aid summarize only a few of the many variables that have been found to be statistically significant in regard to food insecurity. Reviewing the literature highlights the need to consider a large number of variables in order to study this issue effectively. For example, usage of the percentage of the population that is undernourished as a dependent variable applies not only to previous research (Sassi 2015), but this study as well. In addition, measurements such as cereal production per capita, percent of GDP from agriculture, arable land as a share of total land, political stability, access to an improved water source, and access to improved sanitation can all serve as useful metrics. Hesselberg and Yaro (2006) highlight soil quality and inflation as additional factors noted by Ghanaians, further emphasizing the multifaceted nature of food insecurity. In the study of Ethiopia’s South Wollo Zone, a link exists between education and food security, albeit measured through literacy rate as opposed to education level (Agidew and Singh 2018). This study provides evidence that family size has a negative effect on food security. Additionally, the number of agricultural laborers, food aid received, and farming experience all display statistical significance in regard to food security. Soil loss, deforestation, pest incidence, rapid population growth, poverty, conflict, and rural-urban migration can all be considered as possible food security factors in a given country.
In conjunction with the information that food insecurity occurs across a spectrum, the studies included in the literature review offer important insights regarding how food insecurity manifests itself. The study in Northern Ghana highlights that during harvest time, most people can be classified as food secure. However, as time passes and the harvest season ends, an increasing number of families fall into food insecurity, according to Hesselberg and Yaro (2006). Residents could rarely be food secure exclusively from farming; non-farm activities supplement these residents’ food security. Many households, especially rural ones, receive limited support from the government. In Ethiopia, only 21.9% of households report access to a specific food-related policy implemented by the Ethiopian government (Agidew and Singh 2018).
While reviewing the literature on food insecurity helps to establish a multitude of factors, it is necessary to obtain an even deeper understanding of these underlying factors. To this end, this study next turns to economic theory to provide insights into individual and macro-level factors, such as GDP per capita and the fertility rate.
III. Conceptual Framework
As prior research has shown, household size, for which fertility rate functions as a proxy, exhibits a positive relationship with food insecurity. As a nation’s household size increases, that nation can be expected to have a lower percentage of its citizens afford the diet. Children, especially infants, strain a household’s financial resources since they can be expensive to care for and typically do not contribute to household productivity. This is the primary reason an increase in fertility rate is correlated with an increase in food insecurity.
There are a few additional explanations that may illustrate this relationship with more clarity. One explanation is that households may place an inflated value on retirement quality since children, who have negative economic impacts until they are old enough to work, can provide “insurance for a better life in old age” (Hesselberg and Yaro 2006). This view can lead households to sacrifice food insecurity for an opportunity of retirement at an elevated standard of living. In the framework of this study, this behavior may manifest itself as an increase in food insecurity, with families substituting for cheaper, less nutritious diets across all levels. In addition, a family may decide to send their children to school. This decision decreases current food security even further since children will not contribute as much of their labor toward household income and the household undertakes the additional expense of education. However, as educated children age into adulthood, their labor will be worth significantly more and will be able to increase the food security of their family. Family prestige may likewise play a role in household decision making. Hesselberg and Yaro noted that households often provide food to elderly family members while other (younger) members of the household are forced to fast. Unmet food needs of elderly family members impact the prestige and social standing of the household. In this unique tradeoff, a household may make decisions detrimental to their food security status in order to protect their social status.
Across all households, supply and demand highlight some of the factors that may affect food insecurity. An increase in income correlates with a higher percentage of people being able to afford a particular diet. This could be attributed to the income effect. As GDP per capita increases in a nation, the number of people who cannot afford a calorie-sufficient diet declines as individuals expand their caloric consumption. For this reason, nations with higher GDP per capita include a higher percentage of people able to afford diets at a given level.
The substitution effect can explain household reactions to inflation. If a country experiences high inflation, some consumers will substitute with a cheaper, less fulfilling diet. Due to this relationship, inflation will be positively correlated with the percentage of people who cannot afford a diet at a given level. However, this may be less likely to be true if wage inflation keeps pace with price inflation. Given the connection between inflation and food insecurity over time, this study uses a 10-year average for inflation.
The labor force participation rate also affects the ability to afford a diet. This variable relates to GDP per capita in that the more people a country has working in the labor force, the higher GDP will be for the nation. Higher GDP increases the likelihood that the income effect will increase consumers’ consumption habits. Therefore, as the labor force participation rate rises, it is expected that the percentage of a country that cannot afford a diet should decrease.
It would be a glaring omission to not include the price of a given diet as something that would affect a person’s ability to afford a diet. As the price of a diet increases, some consumers will find the diet unaffordable. Consequently, nations with higher prices for the same level of diet will exhibit a larger percentage of people who cannot afford said diet. This means that nations with relatively inexpensive food prices will see a smaller percentage of their population unable to afford a certain diet.
In conclusion, households may make decisions that do not maximize short-term food security in the pursuit of other long-term household goals, causing a positive relationship between food insecurity and fertility rate. Other macro-level factors such as the fertility rate, inflation rate, and the price of a diet also affect food insecurity, with positive increases in these metrics expected to cause an increase in food insecurity. Conversely, increases in labor force participation and GDP per capita correlate with reduced food insecurity.
IV. Regression Model & Data
Food insecurity remains a complex and multifaceted topic, requiring an understanding of multiple variables. Pertinent literature and economic theory identified potential variables that could impact the dependent variable. In order to use these variables to answer the two research questions (Are the factors that affect food insecurity the same at every level of food insecurity and are these factors statistically significant over time?), this study incorporated alternative regression models.
Twelve independent variables exist in these regressions, with three rotating price variables subject to each level of food insecurity. Due to lack of data for one or multiple variables, the model excludes some African countries. While relevant data was included, some data availability challenges were encountered. Namely, observed data for every African country for every year does not exist. This analysis incorporates 2017 data due to the broad availability of data in this year across variables and countries. These data offer an added advantage because they predate the effects of the Covid-19 pandemic, which falls outside the scope of this study.
The regressions incorporate three dependent variables, acting as proxies for food insecurity, including the percentage of people who cannot afford a calorie-sufficient diet, the percentage of people who cannot afford an adequate diet, and the percentage of people who cannot afford a healthy diet. The three dependent variables compare factors affecting the three different levels (mild, moderate, and severe) of food insecurity. Appendix A describes the regression variables in terms of source and measurement.
In addition to the multi-level cross-sectional regression, this study also employs a fixed effects panel regression that utilizes the same independent variables as the cross-sectional regression. The standards for measurement for each variable denoted in Appendix B remain the same for the panel regression, excluding the following notes. For the access to basic water variable if the collection of water takes over thirty minutes round trip, a person is characterized as lacking access to basic drinking water services. Consequently, this metric does not necessarily mean that everyone who does not have access to basic drinking water services cannot get improved water, only that it will take a significant amount of time to do so. Conversely, a person recorded as having access to basic water services may have access to more developed water services than just basic water services. The literacy rate variable could not be found for every country in 2017. In such cases, the model utilized the nearest available year. Since literacy rate does not fluctuate in a noticeable manner in a short period, this does not impact data validity. This regression analysis omits the average rainfall variable as all observations for this variable had the same country‑specific value across different years. In addition, price data can only be accessed for 2017. Finally, the panel regression uses a yearly observation for inflation rather than a ten-year average to avoid including the same year-specific inflation value multiple times for a single country. The panel regression uses data from 2001-2018 for 39 countries. In the panel regression, the prevalence of undernourishment, sourced from the World Bank, functions as a replacement for the three dependent variables used in the OLS regressions. Prevalence of undernourishment measures the percentage of the population whose food consumption does not meet required energy levels for an active and healthy life. The availability of prevalence of undernourishment data for years 2001-2018 (compared to data by food security levels for 2017 only) supports this adaptation.
One reason to expect that an increase in per worker added value in agriculture will cause an increase in food insecurity across all variables relates to the fact subsistence farming often gets excluded from value added metrics. Consequently, workers appear to join the labor force when they worked as subsistence farmers previously. Therefore, subsistence farmers can cause per agricultural worker value added to decrease by transitioning to a less efficient, but measured, agricultural job. To gain a more complete understanding of the study countries, Appendix C includes the descriptive statistics for each variable by mean, median, and standard deviation.
Among the descriptive statistics, the difference for the price of food across the three different levels of food insecurity varies significantly. The mean value ranges from an average of only $0.87 per day for a calorie sufficient diet to $3.66 per day for a healthy diet. If we expand this to a year’s worth of consumption, a calorie sufficient diet requires a little more than $317 per person per year, while a healthy diet requires almost $1336 per person per year. Given this gap in excess of $1000 per person per year, a healthy diet requires a family of five over $5000 more than a calorie sufficient diet. Since GDP per capita has a mean of approximately $2350, the price of a healthy diet per capita represents over half of the mean GDP per capita. Namely, diet represents a significant expenditure among the citizens of the countries studied.
The descriptive statistics associated with several other variables further frame the context of this analysis. Access to basic water services averaged only 70.69% in the study countries. Access to basic sanitation services averaged even less, with a mean of only 44.82% in the study countries. The mean of per worker value added in agriculture, forestry, and fishing surpasses the mean of GDP per capita. Although per worker value added functions as a proxy for the marginal product of labor for agriculture, the absence of subsistence farming data can help to explain the difference between these GDPs per capita. If these data existed, it is anticipated that the amount added by an individual worker would decrease. This is expected because subsistence farming achieves lower efficiency than other forms of farming. Finally, the study countries exhibited a ten-year inflation median of 6.13%. The U.S. usually targets inflation at around 2 percent, meaning a majority of African countries suffer over three times the target inflation rate of the U.S.
V. Results
All four of the regressions are detailed below. Equations 1, 2, and 3 denote the cross-sectional OLS regressions corresponding to severe, moderate, and mild food insecurity, respectively. The dependent variable in Equation 1 measures the percentage of people that cannot afford a calorie sufficient diet, implying severe food insecurity. The dependent variable in Equation 2 measures the percentage of people that cannot afford an adequate diet and experience moderate food insecurity. The experiences of those with mild food insecurity are reflected with the Equation 3 dependent variable which measures the percentage of people that cannot afford a healthy diet. Regression equation 4 describes the panel regression.
Equation 1: CANNOT AFFORD A CALORIE SUFFICIENT DIET (%) = β0 + (β1) α + (β2) γ + (β3) δ + (β4) ζ + (β5) η + (β6) θ + (β7) ι + (β8) κ + (β9) λ + (β10) μ + (β11) ν + (β12) π + ε
Equation 2: CANNOT AFFORD AN ADEQUATE DIET (%) = β0 + (β1) α + (β2) γ + (β3) δ + (β4) ζ + (β5) η + (β6) θ + (β7) ι + (β8) κ + (β9) λ + (β10) μ + (β11) ν + (β12) ρ + ε
Equation 3: CANNOT AFFORD A HEALTHY DIET (%) = β0 + (β1) α + (β2) γ + (β3) δ + (β4) ζ + (β5) η + (β6) θ + (β7) ι + (β8) κ + (β9) λ + (β10) μ + (β11) ν + (β12) τ + ε
Equation 4: PREVALENCE OF UNDERNOURISHMENT (%) = β0 + (β1) α + (β2) γ + (β3) δ + (β4) ζ + (β5) η + (β6) θ + (β7) ι + (β8) κ + (β9) λ + (β10) μ + (β11) ν + ε
Note: Initial regressions for each equation include every independent variable. Variables that displayed weak or no statistical significance were discarded in favor of final regressions with the remaining variables.
Table 1: Variable Symbols
Table 1 displays the variable names associated with each symbol in equations 1-3.
| Variable | Symbol | ||
|---|---|---|---|
| Fertility Rate | α | ||
| Per Worker Value Added in Agriculture, Forestry, and Fishing | γ | ||
| Arable Land % | δ | ||
| Access to Basic Water Services % | ζ | ||
| Access to Basic Sanitation Services % | η | ||
| GDP per Capita | θ | ||
| Avg. Annual Precipitation (mm) | ι | ||
| Avg. Inflation (CPI) | κ | ||
| HIV % | λ | ||
| Literacy Rate | μ | ||
| Labor Force Participation Rate | ν | ||
| Price of a Calorie Sufficient Diet | π | ||
| Price of an Adequate Diet | ρ | ||
| Price of a Healthy Diet | τ |
Table 2: Initial Regression Coefficients
Table 2 shows the initial regression results (5 percent level), with three dependent variables and three linear regression equations all utilizing the same independent variables. Each equation in the table below utilizes observations from 36 countries.
| Variable | Calorie Sufficient Diet | Adequate Diet | Healthy Diet |
|---|---|---|---|
| Fertility Rate | -2.967 | 6.913** | 8.829** |
| Per Worker Value Added in Agriculture, Forestry, and Fishing | 0.0004 | 0.0003 | -0.0003 |
| Arable Land % | -0.122 | 0.138 | 0.060 |
| Access to Basic Water Services % | -0.313 | -0.168 | -0.064 |
| Access to Basic Sanitation Services % | 0.051 | -0.050 | 0.010 |
| GDP per Capita | -0.002 | -0.004** | -0.004** |
| Avg. Annual Precipitation (mm) | 0.004 | 0.009 | 0.004 |
| Avg. Inflation (CPI) | -0.540 | 0.103 | 0.844** |
| HIV % | 0.111 | 1.080** | 0.900** |
| Literacy Rate | -0.121 | -0.193 | -0.107 |
| Labor Force Participation Rate | 0.585 | 0.567** | -0.297 |
| Price of a Calorie Sufficient Diet | 50.086** | ||
| Price of an Adequate Diet | 15.222** | ||
| Price of a Healthy Diet | 5.946** | ||
| Observations | 36 | 36 | 36 |
| Adj. R2 | 0.704 | 0.739 | 0.776 |
Statistically Significant Variables are shown in bold. Blank entries denote variables omitted to avoid confounding variables. *, **, and *** denote p values of p<.1, p<.05, and p<.01 respectively.
Table 3 presents the final linear regression results. In addition, testing the model for multicollinearity, heteroskedasticity, and an F-test for overall regression integrity further substantiated the validity of this model. Multicollinearity does not appear to be a factor, since all VIFs scored below a value of five. Serial correlation tests like the Durbin-Watson test did not get administered as this is a cross-sectional study, and therefore unlikely to suffer from serial correlation. Administering the White test for heteroskedasticity further assessed the model’s reliability. Final regression coefficients comprise Table 3.
Table 3: Final Regression Coefficients
Table 3 presents the final linear regression results. In addition, testing the model for multicollinearity, heteroskedasticity, and an F-test for overall regression integrity further substantiated the validity of this model. Multicollinearity does not appear to be a factor, since all VIFs scored below a value of five. Serial correlation tests like the Durbin-Watson test did not get administered as this is a cross-sectional study, and therefore unlikely to suffer from serial correlation. Administering the White test for heteroskedasticity further assessed the model’s reliability. Final regression coefficients comprise Table 3.
| Variable | Calorie Sufficient Diet | Adequate Diet | Healthy Diet |
|---|---|---|---|
| Fertility Rate | 9.311** | 10.052** | |
| Per Worker Value Added in Agriculture, Forestry, and Fishing | |||
| Arable Land % | |||
| Access to Basic Water Services % | |||
| Access to Basic Sanitation Services % | |||
| GDP per Capita | -0.002** | -0.005** | -0.004** |
| Avg. Annual Precipitation (mm) | 0.009** | ||
| Avg. Inflation (CPI) | 0.782** | ||
| HIV % | 1.014** | 0.776** | |
| Literacy Rate | |||
| Labor Force Participation Rate | 0.676** | 0.633** | 0.442** |
| Price of a Calorie Sufficient Diet | 49.250** | ||
| Price of an Adequate Diet | 14.577** | ||
| Price of a Healthy Diet | 5.918** | ||
| Observations | 36 | 36 | 36 |
| Adj. R2 | 0.710 | 0.765 | 0.799 |
Blank entries denote unmeasured or discarded variables. *, **, and *** denote p values of p<.1, p<.05, and p<.01 respectively.
Several variables (arable land percentage, the percentage of people with access to basic water services, the percentage of people with basic sanitation services, the literacy rate, and per worker value added in agriculture, forestry, and fishing) displayed no statistical significance despite references to their significance in previous literature. While these variables may affect food insecurity as local variables, no clear evidence exists of their statistical significance as global variables. Global variables include variables displaying statistical significance at all three levels (severe, moderate, and mild) of food insecurity.
These analyses suggest that different levels of food insecurity have unique factors affecting them. Fertility rate exhibited a positive coefficient for the price of both an adequate and healthy diet. However, fertility rate did not display statistical significance at the calorie sufficient price level. Average annual precipitation represents an additional explanatory variable that does not display statistical significance at every level. Surprisingly, rainfall exhibited positive correlation, with an increase in rainfall correlating to an increase in the percentage of people who cannot afford an adequate diet. Flooding works as a possible explanation, as too much rainfall would both reduce and destroy crop yields. Inflation also provided support for the hypothesis as it only carries statistical significance at the healthy diet level. As anticipated, inflation had a positive coefficient, meaning an increase in inflation increases the percentage of people who cannot afford a healthy diet. Finally, the prevalence of HIV, as a percentage of the population, has a positive coefficient, confirming the hypothesis that HIV increases with access to food. HIV displays statistical significance at the adequate and healthy diet level. For the adequate diet level, a one percent increase in the HIV rate correlates to a 1.01 percent increase in the percentage of people who cannot afford an adequate diet. HIV had a slightly less severe effect regarding the healthy diet level, with a one percent increase in the HIV rate correlating to a 0.78 percent increase in the percentage of people who cannot afford a healthy diet.
The analysis identifies three statistically significant global variables: GDP per capita, labor force participation, and the price of food at a certain level of diet. GDP per capita displays a negative relationship with the percentage of people who cannot afford a diet, at all levels, supporting the alternative hypothesis. GDP per capita is characterized by a beta value of -.00151 at the calorie sufficient level. This implies that a $1000 increase in GDP per capita correlates to a reduction of 1.51 percent in the percentage of people who cannot afford a calorie sufficient diet. The most severe changes in GDP per capita happen at the adequate diet level, with a $1000 increase in GDP per capita correlating to a 4.54 percent reduction in the percentage of people who cannot afford an adequate diet. Two more global variables, labor force participation rate and price to avoid food insecurity (at any of the three levels) exist as well. An increase in the labor force participation rate results in an increase in the percentage of people who cannot afford a diet at a given level. A large t-statistic related to this result implies a failure to reject the null hypothesis.
Food insecurity can exacerbate via price of food. Namely, the price of a food diet at every level had a positive correlation with the dependent variable. Additional price sensitivity exists at lower levels of food insecurity; small changes in price cause a larger swing in the percentage of people who cannot afford a food diet. At the price sufficient diet level, a 10-cent increase in the price of said diet correlates to a 4.93 percent increase in the amount of people who cannot afford a price sufficient diet. This shows that the most severe level of food insecurity coincidentally happens to be the most susceptible to changes in price, even small ones. The price of an adequate diet, while still sensitive to price, does not match the price sufficient level. A $0.10 increase in the price of an adequate diet correlates to a 1.46 percent increase in the percentage of people who cannot afford an adequate diet. Finally, the healthy diet level displays the least sensitivity to price, with a $0.10 increase in price correlating to a .59 percent increase in the percentage of people who cannot afford a healthy diet.
The panel regression relies upon the Newey-West estimator to address potential autocorrelation. The results for the panel regression can be seen in the Newey-West Corrected Regression in Table 4. This regression utilizes observations from 39 countries, in contrast to the 36 countries for the multi-level regressions, as panel data was available for more countries.
Table 4: Newey-West Corrected Regression
| Variable | Prevalence of Undernourishment |
|---|---|
| Fertility Rate | |
| Per Worker Value Added in Agriculture, Forestry, and Fishing | -0.001** |
| Arable Land % | -0.139** |
| Access to Basic Water Services % | -0.328** |
| Access to Basic Sanitation Services % | |
| GDP per Capita | 0.001** |
| Avg. Annual Precipitation (mm) | |
| Avg. Inflation (CPI) | |
| HIV % | |
| Literacy Rate | |
| Labor Force Participation Rate | 0.245** |
| Observations | 39 |
Blank entries denote unmeasured or discarded variables. *, **, and *** denote p values of p<.1, p<.05, and p<.01 respectively.
The panel regression provides interesting results, with per worker value added negatively correlated with the prevalence of undernourishment. This result aligns with expected results from the literature. Access to basic water services and arable land both displayed a negative correlation as well. Surprisingly, as GDP per capita increases, so too does the prevalence of undernourishment. This result stands in opposition to the expected sign, although this may be due to sample size (n=36). Finally, the labor force participation rate also yielded an unexpected sign, being positively correlated with undernourishment. Although the labor force participation rate did not follow the expected sign, all four regressions supported a positive correlation with food insecurity. This further indicates that the effects of subsistence farmers entering the workforce are not as strong as expected.
VI. Conclusion
This paper aimed to address two questions as its primary objectives: Do the factors that affect food insecurity have the same effect at every level of food insecurity? And are these factors statistically significant over time? To address the first question, the model measured the severity of food insecurity across three levels: severe, moderate, and mild food insecurity. An OLS cross-sectional model suggests that food insecurity manifests itself as a multi-level issue that has different effects at each of the three measured levels of food insecurity. Global variables universally affect the model, with GDP per capita and labor force participation rate having effects at all three levels of food insecurity. Furthermore, some local variables affect only parts of the food insecurity spectrum. Notably, fertility rate and inflation both have effects upon the percentage of people suffering from mild food insecurity, but those effects cannot be felt at the severe level of food insecurity. In addition, some variables had a more notable effect over time. A panel regression model helped to answer the second question. The model suggests that these factors do affect the model over time, but in ways that differ from the cross-sectional model. Multiple factors have a larger effect over time. Access to water and arable land display statistical significance over time. Additionally, variables that exhibit statistical significance in the cross-sectional model, like the inflation and fertility rate, did not exhibit statistical significance in the panel model.
In future research related to food insecurity, access to more complete data will help to ease sample size issues. In the present analysis, data availability issues may have contributed to higher variance within the regressions. This study has another limitation in that it does not wholly account for the presence of local variables, variables that affect countries with only a specific set of characteristics. One possible local variable could be average rainfall, as it may only affect countries that get below a certain level of rainfall per year. A future study that utilizes the multi-level approach used in this study and combines it with Geographic Weighted Regression would be able to remedy this issue by localizing variables even further to isolate the factors and needs of individual areas.
On a policy-level, recommendations should always include strategies to positively affect global variables, while tailoring the directives toward relevant local variables. On a general level, policies aimed at increasing GDP per capita or reducing the price of food for consumers appear to be universally effective in reducing food insecurity and should therefore be included in any national or regional plan addressing food insecurity. On a more specific level, policy should depend on relevant local variables. For example, areas with low rainfall should consider designing policy around increasing irrigation. In addition, the level of food insecurity a nation wants to combat most should be considered. If nations want to focus on getting their citizens a more complete diet, an added emphasis should be put on reducing inflation, incentivizing households with fewer children, and addressing HIV. By designing policy with awareness of both global and local variables, nations will be able to start with a policy foundation and build their policy around the unique characteristics of their nation.
VII. Appendices
Appendix A
Variable Overview
| Variable | Source | Measurement |
|---|---|---|
| Fertility Rate | World Bank | Number of children that would be born to a woman if she lived to the summation of her childbearing years and births children at an age-specific rate for the current year (2017) |
| Per Worker Value Added in Agriculture, Forestry, and Fishing | World Bank | Net output of three industries, agriculture, forestry, and fishing, and divides that by the number of workers, measured in 2015 dollars |
| Arable Land % | World Bank | Arable land divided by the total amount of land |
| Access to Basic Water Services % | World Bank | Percent of the population with access to water from an improved water source, whether that be piped water, protected well water, or packaged water |
| Access to Basic Sanitation Services % | WHO JMP | Percent of the population with access to either basic or safely managed sanitation |
| GDP per Capita | World Bank | Gross domestic product divided by the population in the middle of the year, in current (2021) dollars |
| Avg. Annual Precipitation (mm) | World Bank | Long-term average, in terms of depth, of annual rainfall in a country (mm) |
| Avg. Inflation (CPI) | World Bank | Average change in CPI inflation per year, over a ten-year period, from 2008 to 2017 |
| HIV % | World Bank | Percentage of people, ages 15-49 in a country who have contracted HIV |
| Literacy Rate | CIA World Factbook | Percentage of a population, aged 15 years and older, who can both read and write (2017 data not available for all countries) |
| Labor Force Participation Rate | World Bank | Proportion of the population aged 15 and older who supply their own labor for the production of goods or services in a given year |
| Price of a Calorie Sufficient Diet | Our World in Data | Lowest price bundle of foods that meets calorie requirements using available staple foods in their country
(2017 international dollars per day) |
| Price of an Adequate Diet | Our World in Data | Lowest price bundle of foods that would be within the bounds for both calorie intake and all essential nutrients
(2017 International dollars per day) |
| Price of a Healthy Diet | Our World in Data | Lowest price bundle of foods that meets dietary guidelines from governments and public health agencies
(2017 International dollars per day) |
Appendix B
Explanatory Variables and Expected Coefficients’ Signs
| Variable | Calorie Sufficient Diet | Adequate Diet | Healthy Diet |
|---|---|---|---|
| Fertility Rate | – | + | + |
| Per Worker Value Added in Agriculture, Forestry, and Fishing | + | + | + |
| Arable Land % | – | – | – |
| Access to Basic Water Services % | – | – | – |
| Access to Basic Sanitation Services % | – | – | – |
| GDP per Capita | – | – | – |
| Average Annual Precipitation (mm) | – | – | – |
| Average Inflation (CPI) | + | + | + |
| HIV % | + | + | + |
| Literacy Rate | – | – | – |
| Labor Force Participation Rate | – | – | – |
| Price of a Calorie Sufficient Diet | + | ||
| Price of an Adequate Diet | + | ||
| Price of a Healthy Diet | + |
Blank entries denote variables omitted to avoid confounding variables
Appendix C
Descriptive Statistics
| Variable | Mean | Median | Std. Dev. |
|---|---|---|---|
| Fertility Rate | 4.16 | 4.37 | 1.06 |
| Per Worker Value Added in Agriculture, Forestry, and Fishing | 3254.82 | 1938.55 | 3550 |
| Arable Land % | 15.05 | 10.86 | 13.9 |
| Access to Basic Water Services % | 70.69 | 70.85 | 15.78 |
| Access to Basic Sanitation Services % | 44.82 | 41.89 | 25.35 |
| GDP per Capita | 2350.10 | 1393.40 | 2402.83 |
| Avg. Annual Precipitation (mm) | 999.84 | 929 | 729.81 |
| Avg. Inflation (CPI) | 7.18 | 6.13 | 5.15 |
| HIV % | 4.79 | 1.75 | 7.29 |
| Literacy Rate | 67.48 | 72.2 | 19.54 |
| Labor Force Participation Rate | 64.065 | 63.35 | 12.56 |
| Price of a Calorie Sufficient Diet | 0.87 | 0.85 | 0.31 |
| Price of an Adequate Diet | 2.24 | 2.25 | 0.45 |
| Price of a Healthy Diet | 3.66 | 3.53 | 0.71 |
VIII. References
Agidew, Alem-meta Assefa, and K. N. Singh. “Determinants of Food Insecurity in the Rural Farm Households in South Wollo Zone of Ethiopia: The Case of the Teleyayen Sub-Watershed.” Agricultural and Food Economics 6, no. 1 (2018). https://doi.org/10.1186/s40100-018-0106-4.
Benzekri, Noelle A., Jacques Sambou, Binetou Diaw, El Hadji Sall, Fatima Sall, Alassane Niang, Selly Ba, et al. “High Prevalence of Severe Food Insecurity and Malnutrition among HIV-Infected Adults in Senegal, West Africa.” PLOS ONE 10, no. 11 (2015). https://doi.org/10.1371/journal.pone.0141819.
Craig Gundersen and James P. Ziliak, Coleman-Jensen A, Ziliak JP, Gundersen C, Carmichael SL, Eicher-Miller HA, Skalicky A, et al. “Food Insecurity and Health Outcomes: Health Affairs Journal.” Health Affairs, November 1, 2015. https://www.healthaffairs.org/doi/10.1377/hlthaff.2015.0645.
Hesselberg, Jan, and Joseph A. Yaro. “An Assessment of the Extent and Causes of Food Insecurity in Northern Ghana Using a Livelihood Vulnerability Framework.” GeoJournal 67, no. 1 (2006): 41–55. https://doi.org/10.1007/s10708-006-9007-2.
“JMP Data.” JMP. Accessed November 15, 2021. https://washdata.org/data/household#!/ago.
Phami, Phouvong, Jianhua He, Dianfeng Liu, Su Ding, Patrik Silva, Chun Li, and Zhijiao Qin. “Exploring the Determinants of Food Security in the Areas of the Nam Theun2 Hydropower Project in Khammuan, Laos.” Sustainability 12, no. 2 (2020): 520. https://doi.org/10.3390/su12020520.
Popp, József, Judit Oláh, Anna Kiss, and Zoltán Lakner. “Food Security Perspectives in Sub-Saharan Africa.” Amfiteatru Economic 21, no. 51 (May 2019): 361–76.
Sassi, Maria. “A Spatial, Non‐parametric Analysis of the Determinants of Food Insecurity in Sub-Saharan Africa.” African Development Review 27, no. 2 (2015): 92–105. https://doi.org/10.1111/1467-8268.12126.
The World Bank Group. “Open Data,” n.d. Accessed October 10, 2021.
The World Factbook. “Literacy.” Central Intelligence Agency. Accessed November 14, 2021. https://www.cia.gov/the-world-factbook/field/literacy/.
IX. Acknowledgements
I would like to express my gratitude to my professors Dr. Tonya Hansen and Dr. Flores-Ibarra for their support and input during my research. Their help has been crucial in improving the quality of my paper.