Nick Tsivanidis, Professor

Closed (1) Measuring slums from space

Applications for fall 2021 are now closed for this project.

My research team has developed a machine learning methodology (convolutional neural network) to measure slums from high resolution satellite imagery in Mumbai, India. In this new project, co-funded by the World Bank, we are looking to extend this methodology to measure slums around the world, how they have changed in the past 20 years, and what factors explain slum growth and conversion in developing country cities.

We are looking for motivated undergraduate research assistants (RA) to help us develop our code library, and learn how to extend it measure slums from satellite imagery in more than one city. Under our guidance, the RA would build additional functionality into our code library, apply it to new cities (starting in India), and then iterate so that the algorithm can eventually perform well in predicting slums around the world. They would also process satellite imagery and label data for the algorithm.

This is an opportunity for the RA to gain experience working with big data, satellite imagery and machine learning methods in a developing country context. This will provide good preparation for private sector jobs or for further graduate study.

Qualifications: The ideal candidate will be a 3rd or 4th year undergraduate with coursework in computer/data science and (ideally) statistics. We are seeking two types of candidates: one who would be comfortable working with the neural network library in python in addition to the pipeline that processes and prepares satellite imagery and label data in ArcGIS and python. Therefore please apply if you are motivated and believe you could work well with the data processing aspect of the work but do not have experience with or knowledge of neural networks. For the data processing work, a knowledge of India would be preferred since some work will involve manually classifying slums or correcting existing label data we have for some cities.

Weekly Hours: to be negotiated

Related website: http:///www.nicktsivanidis.com

Closed (2) Using Cellphone Metadata to Measure Migration and Assimilation of Refugees in Jordan

Applications for fall 2021 are now closed for this project.

Over the past 10 years, the population of Amman has doubled. Most stems from the influx of refugees from Syria who left the destitution of refugee camps for the opportunities in Jordan's capital. This project has two central aims.

First, we are using cellphone metadata (CDR) from one of the country's largest operators merged with a proprietary phone survey to use machine learning techniques to classify refugees from signatures in the CDR. We will then use this data to examine patterns of refugee migration and assimilation into urban life (where they live and work, how social networks integrate over time) and characterize the role that social networks play in this process.

Second, we combine the CDR with administrative data available before and after the refugee influx to examine the impact of the influx on land and labor markets, as well as congestion through traffic and strain on public resources such as waste collection and water provision. We will use economic models to quantify these forces, decompose the welfare effects on refugees and incumbent jordanians, and assess the ability of government policies to benefit from the opportunities of rapid urban influxes while minimizing the challenges they pose.

This is an opportunity for the RA to gain experience working with big data and machine learning methods in a developing country context. This will provide good preparation for private sector jobs or for further graduate study.

Qualifications: The RA will work primarily with the cellphone metadata, but there is scope for involvement across the range of tasks within the project. The ideal candidate will be a 3rd or 4th year undergraduate with coursework in computer/data science and (ideally) statistics. Economics would be a plus but not required. Familiarity with some machine learning and classification methods is desirable. Some coding experience in python is required; R/Stata/ArcGIS desirable but not essential.

Weekly Hours: to be negotiated

Related website: https://www.nicktsivanidis.com

Closed (3) Informal Transit Networks: Evidence from Africa’s Largest Megacity

Applications for fall 2021 are now closed for this project.

African cities will double in population by 2050. A key limit to the opportunities provided by such rapid urban growth will be the rise congestion as roads become clogged. At present, the vast majority of trips in African cities are taken using informal transit. This project seeks to understand their costs and benefits, and ultimately try to understand whether these justify government intervention in terms of reforms and regulation or government provision of bus services.

The project provides an opportunity to work with a variety of frontier datasets. First, we are conducting a census of the transit network in Lagos - Africa’s largest megacity with over 23million residents - using surveyers who ride the system and record GPS traces, bus stops and passenger loads using cellphones. Second, we are arranging access to Home Location Register (HLR) data from Nigeria’s largest cellphone operator which reports the tower location of each handset at 5 second intervals to measure commuting, vehicle volumes and traffic speeds. Third, we will use machine learning methods to predict the transit network from both the HLR as well as high resolution daytime satellite imagery (all informal transit in Lagos, danfo, are yellow).

We are partnering with the Lagos State Government on a randomized evaluation of a new bus reform initiative where the government is introducing a new centrally-managed bus system with 820 high-capacity, modern buses. We will use the HLR and survey data to evaluate the impact on commuters, traffic congestion, and the response of the informal danfo industry. We will then use a dynamic economic model that incorporates the routing decisions of informal buses to quantify informal transits costs and benefits, and the impact of potential government reforms.

This is an opportunity for the RA to gain experience working with big data and machine learning methods in a developing country context. This will provide good preparation for private sector jobs or for further graduate study.

Qualifications: The RA will work primarily with the cellphone and eticketing metadata, but there is scope for involvement across the range of tasks within the project. The ideal candidate will be a 3rd or 4th year undergraduate with coursework in computer/data science and (ideally) statistics. Economics would be a plus but not required. Comfort in developing code in Python is required, doing so within large teams preferred.

Weekly Hours: to be negotiated

Related website: https://www.nicktsivanidis.com