commuter-network

Python code for constructing commuter networks from Census data.

These scripts all produce commuter networks as GraphML objects. There are various relevant node and edge attributes; not all are present in each graph:

Node:

• name: the node names correspond to FIPS codes for a particular level of granularity
• state, county: straightforward descriptives
• town: in states where available, the name of the relevant minor civil divisions
• block, tract: Census administrative units
• Population: The total population of the region in the 2016 5-year ACS. The population data is downloaded using the collect_population_data.py script.
• <18, 18-24, ..., 65+: Distribution of ages in the region, also drawn from the 2016 5-year ACS.
• latitude, longitude: coordinates for the relevant spatial unit, drawn from 2019 Census gazetteer files

Edge:

• weight: the amount of commuting traffic between the edges
• margin: the uncertainty around this magnitude, where available (ACS only)

construct_county_network.py creates a weighted edgelist of commuting flows between counties, drawn from Table 1 of the 2011-2015 5-Year ACS Commuting Flows dataset.

construct_town_network.py creates a weighted edgelist of commuting flows between minor civil divisions (where available; counties where not available), drawn from Table 3 of the 2011-2015 5-Year ACS Commuting Flows dataset.

construct_tract_network.py creates a weighted edgelist of commuting flows between Census tracts, drawn from LODES. Before running it, download the LODES data using the collect_lodes_data.sh script, and then aggregate from block level to tract level using the aggregate_lodes_tract_level.py script.

construct_block_level.py creates a weighted edgelist of commuting flows between Census tracts, drawn from LODES. Before running it, download the LODES data using the collect_lodes_data.sh script.

A note on national versus subset networks

All four scripts are used in the same way.

python construct_block_network.py -s ma,me,vt,ct,ri,nh -o new_england -m 1

The -s flag takes a comma-separated list of USPS state codes; if it is absent, it runs the script across all 50 states plus DC. The output data files are in data/derived; the -o flag adds an additional subdirectory (e.g., data/derived/new_england). The -m flag specifies a minimum weight necessary for an edge to be included, effectively thresholding the network.