This started as a final Project for Econ 554: Urban Economics at Duke University. It is an extension of Jan Brucckner and Stuart Rosenthal's 2008 paper "Gentrification and Neighborhood Housing Cycles: Will America's Future Downtowns Be Rich?". This paper makes two main contributions: first it uses modern data (2000-2019) to test some of Bruckner and Rosenthal's predictions. Second, it tries some additional measures of city centrality and regression forms.
This project is released under the MIT License, but if you want to use any portion of it for any scholarly work, I would ask that you cite it as an unpublished paper written by Logan Cooper in 2023. If you have any questions, feel free to reach out to me.
Most of the data needed for reproduction is packaged in the /data directory. The one exception is the ~8GB tract distance file, which can be downloaded from the NBER here. Note that the file used is the 50-mile data for 2010.
The dependencies for this project are saved in ENV.yml. You can install them with conda env create -n econ554 --file ENV.yml.
/data: This directory contains all of the data for this project./assets/tables: This directory is for tables generated either in the notebook or bytable-maker.py./models: This directory contains the summary printouts of all of the models defined inreg.py.census-req.py: This file has some code to populate/datawith Census data. All of that data is committed, so there shouldn't be any reason to run it.ENV.yml: Conda environment definition.LICENSE: Licensing information.notebook.ipynb: The main file for this project. This notebook includes data cleaning, data wrangling, model evaluation, and visualization.reg.py: Python code for running the regressions. Separate from the notebook so modelling could be done in the cloud.secrets.json.template: If you want to usecensus-req.pyyou'll need to rename this tosecrets.jsonand fill it in with a Census API code.table-maker.py: Takes statsmodels' model output and turns it into LaTeX tables used in my paper.wilson_cbd_req.py: Python script to get Census Tract FIPS code from latitude-longitude coordinates.
Other than the NBER datafile, everything necessary to run this code should be present in this repository. However, note that notebook.ipynb does not contain any code to run the models. The models are extremely memory-intensive to run, and were run on Duke's Economics Computing Cluster. There are 60 models in all, and they range in size from 7,546 data points and 12 features (including fixed effects) to 77,055 data points with 6,862 features, including squared and cubed ones. The largest models took more than 8GB of RAM to run. If you want to reproduce those, you will need a beefy machine.