Explanatory Economic Factors in Femicides

By Montserrat Ponce-Parra


Gender violence is categorized by the World Health Organization (WHO) as a major public health problem. The United Nations Office on Drugs and Crime (UNODC) estimated that in 2017, 87 thousand women were murdered and that 58% of the homicides were perpetuated by an intimate partner or family member. Homicide is the most extreme expression of violence and the particular killing of women based on their gender has a specific term, femicide. Femicide is defined as “killing of women by men motivated by hate, contempt, pleasure or the assumption of ownership of women” (Caputi, 1992). Femicide is the killing of women because they are women.

Graph 1: Average Number of Femicides in Latin America

As shown in Graph 1, Latin American countries have been experiencing rising rates of femicides over the last 15 years. The United Nations Economic Commission for Latin America and the Caribbean (ECLAC) reports on femicides in the region, and Graph 1 is the result of the average of these reported numbers. According to ECLAC in 2018, an average of 12 women are murdered every day in the region. The high murder rates and prevalence of impunity have resulted in various protests and social movements by women who demand that their respective governments take action in lowering the crime rate, prosecuting the guilty parties, believing accusations of assault, decreasing the disappearances of women, stopping the concealment of femicides and classifying femicides as a specific type of crime in the legal code. Femicides are attributed to an unequal power structure that subordinates women to men. 

Femicides Analysis

This memorandum will focus on femicides in Latin America from 1990 – 2017 and will broaden the previous definition so that it encompasses any killing of a woman, making the dependent variable intentional female homicides per 100,000 people. This change in definition is due to the diverse legal definitions Latin American countries have for the violent act and the lack of data for recognized femicides. Femicide was only recently recognized by the international community in 2013 in the 57th session of UN Commission on the Status of Women so research and data on it is scarce (Saccomano 2015). Some Latin American countries have just recently modified their laws to sanction and classify this crime as a specific offense but there are some in the continent that still do not recognize it in their penal code (ECLAC 2018).

For this study to take place, the rate of intentional homicides of women per 100,000 by country was used as a dependent variable. The economic indicators that will be used as explanatory variables for femicides will be decided based on hypotheses and previous research papers that studied the link between crime and homicide rate. Listed below are the explanatory variables for this model, their symbols, their definitions: 

  • Intentional male homicides (ihMale): the male homicide rate is per 100,000 and it was used to capture the general “danger level” of a country in a given year and so the magnitude of the remaining effects can be interpreted as their contribution to female homicide specifically. 
  • Unemployment rate (Unemployment): the rate of unemployed people that are part of the labor force and are seeking employment. 
  • Proportion of seats held by women in national parliaments (Parliaments): the percentage of parliamentary seats in a single or lower chamber held by women.
  • Rule of Law (Law): measurement ranging from -2.5 to 2.5 that reflects the perception that people have confidence in and abide the rules of society; particularly the quality of contract enforcement, property rights, the police and the courts as well as the likelihood of crime.
  • Infant mortality (InfMortality): the number of infants dying before reaching one year of age per 1,000 live births.
  • Life expectancy (LifeExp): the number of expected years a newborn would live if the patterns of mortality at the time of its birth stayed the same throughout its life. 
  • Education index (Education): the average of mean years of schooling of adults and expected years of schooling of children, both expressed as an index. 
  • GINI coefficient (GINI): measurement from 0 to 100 that reflects the income distribution between individuals or households within a country.
  • GDP per capita (GDP): gross domestic product converted to international dollars using purchasing power parity rate divided by the country’s population. 
  • Urban population (UrbanPop): people living in urban areas as defined by national offices. 
  • Population density (PopDensity): the midyear population of a country divided by the land area of the same country in square kilometers.

The countries that this paper will focus on are Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, Guatemala, Honduras, México, Nicaragua, Panamá, Perú, Paraguay, El Salvador, Uruguay, and Venezuela. Argentina, Cuba and Puerto Rico will not be included in this research because of reasons that will now be stated. However, Argentina is excluded in this study due to the risk of misleading statistics. As reported in The Economist in 2011, Argentina has presented inaccurate data in previous years to avoid making payments on index-linked debt. Argentina’s underreporting was so misleading that the country was reprimanded by the International Monetary Fund (IMF) in 2013. Puerto Rico is also excluded because it is a US territory and as such, its data is merged with that of the US. This means that there is no way of analyzing Puerto Rico as an independent country. And finally, Cuba was excluded from this analysis because of its inaccurate InfMortality. In “Infant Mortality in Cuba: Myth and Reality”, Roberto Gonzalez reported  that Cuba’s actual IMR is substantially higher than that reported by authorities (Gonzales 2015, 19). This proved that underreporting is the reason why Cuba’s data was not used in this analysis.

Statistical Analysis

After collecting the data from the World Bank Database, all the excluded values were removed so that they would not affect the model. The log transformation was applied to both the dependent variable and multiple explanatory variables, thus creating a log-log model to allow for nonlinearities in parameters. The only two explanatory variables that were not transformed were Rule of Law, which takes negative values, and GINI coefficient, which already had an approximately symmetrical distribution. Once all variables were standardized and most of them were logarithmically transformed, the model was fit.

A close up of text on a white surface

Description automatically generated
Figure 1: Model f1

A white sign with black text

Description automatically generated
Figure 2: Model f2

The full model (f1) created included all variables; their coefficients, standard errors, t-values and p-values can be seen in Figure 1. Given the p-value of Unemployment, Law, PopDensity and Education, they contribute very little to the prediction of femicides. InfMortality and LifeExp were highly correlated and because of this their prediction power cannot be assessed individually. The correlation between these variables is not surprising as both of them measure the general health of the country and it is expected that both yield similar trends. As an alternative approach, a second model (f2) was created in order to see how this relationship between variables would affect the model if only one of the variables would be included in the model as a proxy for the country’s health. The coefficients, standard errors, t-values and p-values of f2 are shown in Figure 2. F2 leads to the same conclusion of disregarding Unemployment, Law, PopDensity and Education as relevant predictors because they are not statistically significant. Because of the correlation between InfMortality and LifeExp, any effect InfMortality had in f1 is attributed to LifeExp in f2. Thus, it can only be concluded in using both variables or none in my final model that it cannot be determined which one is the most relevant. An F test based on a joint hypothesis that both parameters are equal to zero was used to decide that both were not significant predictors. For my final model, reduced model (f3), it was decided to eliminate Unemployment, Law, PopDensity, Education, InfMortality and LifeExp given their p-values. Figure 3 shows f3 statistical values. With an adjusted R2 of 0.893, I can conclude that femicides in the selected Latin American countries from 1990 – 2017 can be greatly explained by f3. IhMale is not interpreted as an explanatory variable as it was used to capture the general “hostility level” and so the magnitude of the remaining variables can be interpreted as their contribution to femicides specifically. From f3, we can conclude that a 1% increase in Parliaments, GDP, UrbanPop will result respectively in a 4%, -22%, 13.63% of femicides in the selected Latin American countries. As for GINI, it is expected that a one unit increase in the variables will result respectively in a -13.43% change in femicides. 

A white sign with black text

Description automatically generated
Figure 3: Model f3

Even though these interpretations are correct, given the type of data used in this study and their transformation for statistical analysis as well as how these economic factors interact in the real world, it is better to look at the results from a broader angle. We would achieve this by using the sign and the magnitude of the regression coefficients to identify the socioeconomic factors that are most correlated with female femicides. These are indeed the factors that should drive policy changes in order to reduce femicides; recommendations will be addressed later in the memorandum. A better and more general interpretation of f3 is that GDP has a bigger impact in femicides, followed by GINI and UrbanPop. Parliaments, even though they are statistically significant, they have the smallest effect in explaining femicides in Latin America. In relative terms, that means that  wealth and inequality have the largest correlation with femicides in these observational terms when compared to other factors such as Parliaments, education, health, law, and other factors. Also, future policies should consider that the level of femicides are higher in densely populated areas. Note that these conclusions based on the magnitude and sign of the effects do not change when considering the first model either, as this is a strength since the final conclusion is robust to the model selection process.


There are two major recommendations for Latin American countries based on the research done and the data analysis of this memorandum: 

Firstly, they should have more data and research on femicides. As stated previously, there are multiple legal definitions among the countries for the violent act, and in some cases, they do not recognize femicide as a specific type of violence. This results in a lack of data collection for recognized femicide, if it is not legally defined there can be no data on them and thus no research. All countries, going beyond those in Latin America, should legally recognize femicide as a specific type of gender violence and treat it as a type of discrimination. More specifically within Latin America, once all governments acknowledge femicides in their legal codes, they can finally start gathering data on them. The collection of data should be the responsibility of the governments and should have specific registries to report the number of murders; Argentina and Peru are at the forefront of this by already having femicide registries. Activists can also take part in this data collection solution, like in the case of María Salguero Bañuelos. Salguero is a Mexican activist that has been mapping all of the femicides in Mexico since 2013. Her map presents a big contrast from the under reporting of the Mexican government and the problem this underreporting creates despite the legal recognition of this specific crime. Governments have a responsibility of accurately reporting the number of femicides and it falls on activists to pressure governments to do so.

The second recommendation is based on the concluding data analysis. As seen before, the GDP and GINI coefficient play a major role in the number of femicides in a country. It is expected that investing in human capital such as education, particularly for women, can result in an increment of GDP and reduce the income inequality in a country. The GINI coefficient should be the main target of economic policies as its reduction would be reflected in GDP growth. The main recommendation for the reduction of the GINI coefficient in Latin American countries is increasing the minimum wage and raising marginal income tax rates on the highest earners. The Pigou-Dalton principle states that, ceteris paribus, the transfer of some income from a wealthier person to a poorer person (but not so much that the poorer person is now wealthier than the originally wealthy person), will result in a more equal income distribution and this can be achieved through proper taxation.


The trend of femicides in Latin American countries can be explained not only by social and historical factors but also through economic ones and it falls upon governments to target all of these. Defining what encompasses femicide is the first step towards reducing this type of homicide, not only in Latin America but also across the world. Delineating what constitutes as a femicide and what does not, gives a specific target for governmental policies. Then from an economical aspect, the policies Latin American governments should take to reduce the number of murdered women would need to focus on inequality. A redistribution of wealth through proper taxation can lead to better standards of living for the general population, and thus better quality of life for women. This reduction of inequality can lead to a betterment of human capital and in turn GDP growth in the long run. Also, as seen in the previous model, inequality and GDP greatly influence the number of femicides in a country. The purpose of this memorandum is to highlight the economic factors that can explain these heinous crimes and give recommendations to the Latin American governments in order to reduce them. 


Carcedo Cabañas, Ana. 2000. Femicidio En Costa Rica 1990-1999. San José: Instituto Nacional de las Mujeres.

“ECLAC: At Least 2,795 Women Were Victims of Femicide in 23 Countries of Latin America and the Caribbean in 2017.” United Nations. 2018.

Gonzalez, Roberto M. 2015. “Infant Mortality in Cuba: Myth and Reality.” Cuban Studies 43(1): 19–39. 

Radford, Jill. 1992. The Politics of Woman Killing. New York: Twayne Publishers.

Saccomano, Celeste. 2015. “The Causes of Femicide in Latin America.”

Steil, Justin. 2014. Responding to Rising Inequality. Berkeley: Haas Institute for a Fair and Inclusive Society.

Global Study on Homicide: Gender-Related Killing of Women and Girls. United Nations Office on Drugs and Crime. Vienna. 2018.

Widmer, Mireille and Irene Pavesi. 2016. “A Gendered Analysis of Violent Deaths.” Geneva: Small Arms Survey.

World Health Organization. 2017. Violence against Women. Vienna.

Also published on Medium.