The sample I use for this research is all countries in Africa with a population over one million within the time frame of 1990-2010 in order to determine why food price shocks lead to protests in some countries while at the same time, other countries produce no protests. There are 937 observations. All of the countries used in the sample have a full twenty years of observations. For the sample, I will be using the number of recorded events of political unrest for each country. There are more than 6,000 recorded events overall during the twenty year time period. Within this study I will use the country and the year (country-year) as a unit of analysis.
To measure the dependent variable, the number of events of political unrest (protest, riots, and demonstrations), I use a variable measuring the number of events per year for each country as designed in the Social Conflict in Africa Database (SCAD) (www.scaddata.org). The database covers the time period 1990 through 2010 and lists over 6,000 protests, riots, strikes, coups, communal violence, and other types of social unrest compiled by Associated Press (AP) and Agence France Presse (AFP) news wires (Hendrix, Salehyan, Linebarger, Stull, and Williams 2010). Using SCAD, I constructed a count variable for my dependent variable of the number of political unrest events per country per year. I tried using the number of political unrest events as a dependent variable but it was difficult to distinguish what was related to food, water, and subsistence, and what was not.
In my data set, each country is organized by year and the number of SCAD events for each country. The distribution of my dependent variable ranges from zero (zero recorded counts of political unrest) to ninety-five (South Africa in 1995 had 95 instances of political unrest). The dependent variable for recorded events is represented in the model as events.
To measure the independent variable, Food Price Index (foodpriceindex), I used data regarding global food market prices compiled from the Food and Agriculture Organization (FAO). The FAO Food Price Index is a list of the monthly food price indices averaged over the course of the year. It consists of the average of five different commodities weighted by the average export shares for 2002-2004 (FAO). The five commodity price indices used are: meat, dairy, cereals, oils and fats, and sugar. The data was collected from the FAO website (http://www.fao.org/worldfoodsituation/wfs-home/foodpricesindex/en/) and the full index is available for download at its website. In addition to providing a variable for food price index, I created a variable to represent the change from one year to the next. The variable (fp_change) represents the amount of change in the food price index from the food price index of the previous year. Furthermore, I constructed a variable for the percent change in the food price index from the previous year as a percentage (fp_pctchange).
The variable for urban population (urban_pct), represents the percent of total population that reside in urban areas. The data for the urban population variable was obtained through the World Bank World Development Indicators. “Urban population refers to the people living in urban areas as defined by national statistical offices. It is calculated using World Bank population estimates and urban ratios from the United Nations World Urbanization Prospects” (http://data.worldbank.com).
In order to test my theory and hypotheses, I constructed a variable representing net food importing countries. The variable for food importers (nfic) was created by taking FAO database numbers for cereal exports and subtracting them from cereal imports in order to determine if the country was a net importer of cereals (wheat, rice paddy, barley, maize, rye, oats, millet, sorghum, buckwheat, quinoa, fonio, triticale, canary seed, and mixed grain) (FAO 1994). I then created a dichotomous variable by assigning a “0” if the number was less than zero, meaning that it was not a net food importer and “1” if the number was greater than zero, meaning that it was a net food importer.
I also created an interaction variable representing the effect of the food price index and whether or not the country was a food importer (fp_nfic). It is expected that when interacted with the food price index, net food importing countries will have a greater impact on events of political unrest. I created this variable by multiplying the variable for net food importing countries (nfic) with the variable for food price index (foodpriceindex).
Finally, I included several control variables within the model. First, the type of government may be a decisive factor in the number of protests, riots, and demonstrations that take place within a country. If a country is fully democratic, we expect that certain institutions within the government would be devised to provide a buffer or safety net during food price shocks. Furthermore, if citizens have the right to vote and those in power have displayed that it will allow for the change in leadership, the citizens of the country would have the means to alter the government if those in power fail to provide a safety net. To empirically measure the type of government of the country, I used the Polity 2 variable from the database compiled by the Polity IV Project: Political Regime Characteristics and Transitions. Polity 2 is an adjusted variable of the original Polity variable, modified for use in time series analysis. The original Polity variable measured the type of government by subtracting the level of autocracy from the level of democracy to achieve the final variable. In this dataset, the variable, Polity 2, ranges from -10 (full autocracy) to a +10 (full democracy) (Marshall and Jaggers 2000).
Second, a variable representing the Gross Domestic Product of a country (log_GDP) was also added to the model to test if GDP logged over time would have any effect on the number of events of political unrest. It is expected that higher levels of logged GDP will have a negative impact on higher levels of political unrest. With a higher GDP a country would have more flexibility when it comes to a shock in market prices. This variable is measured in U.S. dollars and comes from the Penn World Tables. The Penn World Tables are arranged in time series format and the entries are denominated in a common set of prices in a common currency (U.S. Dollars) to allow for easy comparison (Heston, Summers, and Aten 2009).
Finally, I included variables for my dependent variable lagged (events_lag) and a variable for total population (total_pop). The variable for events lagged was used with the expectation that an event in one year will increase the likelihood of future events in the next year. Total population was used as a control for the total population of each country included in the dataset.