Data and Code for: Using Neural Networks to Predict Micro-Spatial Economic Growth

Principal Investigator(s): View help for Principal Investigator(s) Arman Khachiyan, UC San Diego; Anthony Thomas, UC San Diego; Huye Zhou, UC San Diego; Gordon Hanson, Harvard University; Alex Cloninger, UC San Diego; Tajana Rosing, UC San Diego; Amit Khandelwal, Columbia University

Version: View help for Version V2

Name	File Type	Size	Last Modified
Data and Code for: Using Neural Networks to Predict Micro-Spatial Economic Growth		0

Project Citation:

Khachiyan, Arman, Thomas, Anthony, Zhou, Huye, Hanson, Gordon, Cloninger, Alex, Rosing, Tajana, and Khandelwal, Amit. Data and Code for: Using Neural Networks to Predict Micro-Spatial Economic Growth. Nashville, TN: American Economic Association [publisher], 2025. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 2025-01-27. https://doi.org/10.3886/E158002V2

Project Description

Summary: We apply deep learning to daytime satellite imagery to predict changes in income and population at high spatial resolution in US data. For grid cells with lateral dimensions of 1.2sq-km and 2.4sq-km (where the average US county has dimension of 55.6km), our model predictions achieve R-sq values of 0.85 to 0.91 in levels, which far exceed the accuracy of existing models, and 0.32 to 0.46 in decadal changes, which have no counterpart in the literature and are 3-4 times larger than for commonly used nighttime lights. Our network has wide application for analyzing localized shocks.

Scope of Project

Subject Terms: spatial economics; income measurement; remote sensing

JEL Classification:
R10 General Regional Economics (includes Regional Data)

Geographic Coverage: Contiguous United States

Universe: Analysis is conducted on square images of neighborhoods in the contiguous United States.

Data Type(s): administrative records data; census/enumeration data; geographic information system (GIS) data; images: photographs, drawings, graphical representations

Collection Notes: Please see our README.md file and our manuscript for details on how each dataset was collected and applied in our study.

Methodology

Response Rate: n/a

Sampling: All neighborhoods above a population density threshold were included in our analysis.

Data Source: Landsat 7 imagery via Google Earth Engine
Census Estimates via IPUMS NHGIS

Collection Mode(s): face-to-face interview; mail questionnaire; remote sensing

Scales: No scales were used.

Weights: Weights are not used in our analysis.

Unit(s) of Observation: The unit of observation is a neighborhood (image) in a year.

Geographic Unit: The geographic unit is a neighborhood (image).

Related Publications

Download this file

Published Versions

V2 [2025-01-27]

V1 [2022-11-21]

Export Metadata

Dublin Core

DDI 2.5

Report a Problem

Found a serious problem with the data, such as disclosure risk or copyrighted content? Let us know.

This material is distributed exactly as it arrived from the data depositor. ICPSR has not checked or processed this material. Users should consult the investigator(s) if further information is desired.