Annotated Bibliography

Anastasia Mavrakis - Data 150

Problem Statment

The harms that I am looking into is the role that poverty plays in the daily lives of those living in Rwanda, and how it consequently leads to food insecurity, as both of those issues are closely intertwined in this nation. Poverty is a growing issue within the country, and has lead to an unfortunate high mortality rate, particularly among children. The complex challenge in analyzing this topic is that Rwanda is an incredibly densely populated nation with not enough resources, and is also largely rural agriculturally. Basing an entire economy and means of livelihood upon the growth of crops has proved detrimental, particularly as fluctuations in climate continue to be unpredictable. Furthermore, Rwanda is one of many African countries that has had limited public access spatial data that would provide useful in allocating necessary resources to combat poverty, thus, making the achievement of any sustainable development goals all the more difficult.

Annotations

1. Akinyemi, F., & Bigirimana, F. (2012). A spatial analysis of poverty in KIGALI, Rwanda using indicators of household living standard. KIST Journal of Science and Technology. doi:10.4314/rj.v26i1.1

Rwanda is a densely-populated nation with even more refugees pouring in on a yearly basis, and it has seen a consistent pattern of poverty throughout the years. In this article, Akinyemi aims to determine the leading causes of this high number of individuals living in poverty as a way to form a concrete plan on how to tackle this problem in the most efficient way possible through Economic Development and Poverty Reduction Strategies (EDPRS). The answer lies in a Geographic Information System (GIS) based image of poverty. Cartographic visualizations of indicators of poverty such as education level and nutritional status can be combined with geographic and environmental information in order to design programs target poverty at its source. This particular study determines poverty levels using the household living standard indicators from the National Institute of Statistics of Rwanda as gathered by the Integrated Living Condition Survey (EICV). ArcGIS was used for integrating all spatial data into a vector format. The dimensions of household living standards that were analyzed were household health, education, and accessibility to services. For household health the parameters were how frequently health care facilities were being used and the number of people lacking medical insurance. For education, illiteracy rates as well as enrollment in primary and secondary school were analyzed, and lastly for accessibility the number of people living in unauthorized housing was mapped out in combination with the distribution of safe and accessible water. In looking at these parameters and assessing their frequency, aid may be distributed more effectively, thus contributing to the freedom that Sen speaks about in defining human development. The dimension of human development being addressed is poverty assessment and analysis, aligning with SDG #1 which is to end poverty in all forms everywhere. The geospatial datasets used are GIS, which then uses ArcGIS 9.3 software package to map the data for this study. The question that Akinyemi is seeking to answer is which aspect of household living needs to be addressed first in order to decrease poverty patterns.

2. Jean, N., Burke, M., Xie, M., Davis, W. M., Lobell, D. B., & Ermon, S. (2016). Combining satellite imagery and machine learning to predict poverty. Science. doi:10.1126/science.aaf7894

A huge hindrance that is currently plaguing the developing world is a lack of reliable and accurate economic data. This has led to a lack of development in research and policy as well as a lack of effective intervention. Based on World Bank Data, from 2000-2010 nearly ⅔ of African countries conducted less than 2 surveys on measures of poverty. Furthermore, most of this data is not publicly accessible. The same is true of a majority of nationwide health statistics. The authors proposed solution is utilizing high-resolution satellite imagery in order to correlate landscape features with economic data. Recently, this tactic has failed due to an inability to assess the discrepancies between countries living at the poverty line vs below it at such a large scale, even with extensive manual analysis. What Jean considers, however, is to overcome this challenge by implementing “transfer leaning”. This would mean that a proxy for poverty would be utilized in order to teach a deep learning model, which would then be able to estimate the mean household wealth on the lowest level for which longitude and latitude data is available for the public domain. This, in combination with the World Bank’s Living Standards Measurement Study, would create a standard by which national poverty statistics could be determined in developing countries. This transfer learning model would begin with a convolutional neural network (CNN) model that has already been trained on ImageNet. Then, this CNN can be used on a new task and can perform nonlinear mapping from each input image to a precise vector representation. The results were that this transfer learning model was actually very precisely able to predict average household consumption and wealth across multiple African countries. Using this transfer learning technique would allow aid efforts to be fine-tuned and implemented where they are needed most, hopefully allowing for an eventual freedom from poverty, in accordance to Sen’s perspective on human development. Poverty assessment and analysis is at the forefront of this article, making its priority to reduce poverty as outlined in SDG #1. The data used is though a CNN model, which then utilizes transfer learning to become a feature extractor. Overall, the question that Jean hopes to answer is whether these satellite images when combined with transfer learning can be used in order to accurately predict poverty measures effectively on a lower scale.

3. Muhire, I., Ahmed, F., Abutaleb, K., & Kabera, G. (2015). Impacts of projected changes and variability in climatic data on major food crops yields in Rwanda. International Journal of Plant Production.

This article looked at the future trends in climate as well as precipitation changes in order to determine how this will affect crops being grown in Rwanda. This study is particularly important due to the fact that on a yearly basis the unpredictability of these trends can account for up to 30% of variation in the yields of the most widely-grown crops worldwide. In developing countries such as Rwanda, which on average have much smaller crop inputs and whose agriculture relies solely on the rain, alterations in climate can account for a 80% variability in overall crop yield. Changes in temperature as well as overall land productivity are among the reasons that climate patterns play such a key role. Furthermore, it is essential to note that Rwanda already has a low water supply, which could only increase the rate at which the country becomes increasingly arid. The authors began their study utilizing data gathered from the Rwandan Meteorological Center, which they used to project the daily data for 2000-2050 and from there implemented a stochastic weather generator in order to calculate the average temperature and precipitation records. Then, agricultural records of yields per crop from 2000-2010 were obtained in order to create a model that could predict crop yields from 2011-2050 based on the climate data. This model could also determine future climate suitability of the growth of major food crops in Rwanda. The results were that future precipitation data is likely to vary more than the temperatures of Rwanda, which are not likely to change very much at all, and will likely only affect crop growth in high altitude areas of Rwanda. Rwanda is a very agriculturally-based society that obtains most of its food intake through these crops, and thus predictions of future crop yields can allow for shifts in agricultural methods to promote productivity. This relates to Sen’s definition of development as freedom as it would allow people in Rwanda to maximize agricultural efforts and aid in making sure people are able to obtain as many resources as possible, potentially giving them the freedom of escaping food insecurity. The dimension of human development that is being evaluated in this article is environmental impact, but particularly how it pertains to crops and future food security. Again, this article would go hand-in-hand with SDG #2 as it centralizes around promoting sustainable agriculture. The datasets utilized are created by LARS-WG, a weather generator and the method used to analyze this data and create a model was through multiple regression analysis. The purpose the authors have in publishing this article and the question they are trying to answer is how the future of Rwanda’s crops look based on climate trends.

4. N., D., K.V., R., J., M., & Sharma, A. (2018). The study of dietary diversity score in children between 6 months to 23 months: A hospital based study. International Journal of Contemporary Pediatrics. http://dx.doi.org/10.18203/2349-3291.ijcp20181541

According to the World Health Organization, one of the most important and accurate predictors child malnutrition lies in one’s dietary diversity score, which is defined as the number of food groups consumed over a period of time. This number can be used as a sufficient predictor for nutritional adequacy, and therefore it is a good indicator of a child’s growth and overall weight. In nations that are impoverished and have less resources, such as Rwanda, it is common to see that a majority of children do not have a sufficient dietary diversity score and in most cases, do not even meet the minimum dietary diversity of eating 4 different food groups. In this article, the authors conducted a study at a hospital in an impoverished part of the world, where they analyzed what children between the ages of 6-23 months were consuming based on a 24 hour recall method, where the children were interviewed to determine what they had eaten the previous day. The data was analyzed using statistical software over the course of three days. The results were grouped by age, and revealed that the median dietary diversity score for children between the ages of 6-12 months was a 2 out of the possible 7 and for children between 13-23 months old, this score rose to 4, the minimum dietary diversity score. The conclusion was then made that there is in fact a positive correlation between age and dietary diversity score. This is pertinent due to the fact that the biggest predictor of a child’s diversity score is annual household funds. Thus, it is likely that developing nations will have decreased child nutritional status. This article relates to Sen’s definition of development in that it emphasizes the fact that mean income is not the only predictor of growth, and freedom, in this case freedom to have multiple nutritional options, is key. The authors are addressing the poverty and food insecurity aspects of human development, with the most pertinent SDG being #2, that is, to end hunger and improve nutrition. The data that is pertinent to the study is the demographic profile of those involved as well as the anthropometric measurements of the subjects, and the methods used were the Mann Whitney U test and the Kruskal Wallis test, which are both tests of significance to identify the mean difference between groups of qualitative data. Overall, the question that the authors were trying to answer was whether or not the age of children living in poverty affects their dietary diversity score when all other confounding variables are removed.

5. Njuguna, C., & McSharry, P. (2017). Constructing spatiotemporal poverty indices from big data. Journal of Business Research. doi:https://doi.org/10.1016/j.jbusres.2016.08.005

Njuguna begins the article by addressing the fact that in order for poverty to be eradicated completely, as is outlined in the agenda of the UN General Assembly, we must first define and measure what constitutes poverty itself. He divides poverty into monetary-based issues, such as income, as well as non monetary-based issues, like the multidimensional poverty index, and argues that while census data and surveys are common and reliable sources of data, they can only be implemented periodically due to their high cost. In Rwanda in particular, a population census is only taken every 10 years. Njuguna points to the growth of big data as the answer and an alternative to obtain data. Big data, also known as business intelligence and analytics, has reached its third version and stores records in relational database management systems (RDBMSs), as well as implements features such as statistical analysis and data mining. One widely promising category of big data is call detail records (CDR), which are extremely labor and cost efficient. The study proposed is to use features that are representative of socioeconomic status as proxies, which are then statistically analyzed and aggregated at the sector level to match the spatial resolution of the true value, or, the MPI. Features that do not correlate to the MPI are then eliminated. Those that do correlate to the MPI are analyzed and either eliminated via the Least Absolute Shrinkage and Selection Operator (LASSO) or ridge regression prior to the final model. In accordance to Sen’s definition of human development as freedom, one of the greatest freedoms is economic freedom, which could be granted through humanitarian aid should poverty data become more reliable and accessible through the methods outlined above. In the article, Njuguna emphasizes the importance of SDG #1, which is the sustainable development goal that he is focusing on, the dimension of human development being poverty assessment and analysis. The datasets that are used by the author include the multi-dimensional poverty index as well as call detail records. He also touches on the idea of using night light satellite data through visible infrared imaging radiometry suite (VIIRS), as it is freely available and has good spatial resolution. The method used is extracting mobile ownership and call volume from CDRs. The question that Njuguna seeks to answer is whether or not CDRs can replace census data about poverty and be a more cost-efficient way of doing so.