TRANSPORT
DATABOOK

Global transit data, maps, and analysis.

Loading county data…

{{ scatterTitle }}

{{ scatterSub }}

{{ smo.label }}

{{ chip.label }}

Election: {{ syo.label }}

Election: {{ chip.label }}

Include: {{ chip.label }} Size by density Log scale

More Republican More Democratic Local average (LOESS)

Counties plotted: {{ scatterN }} Correlation (r): {{ scatterR }} {{ scatterStrength }}

{{ calloutNote }}

{{ scatterLabelBtn }}

Dot size: fewer workers per sq mi more

⬇ Save Graph⧉ Copy Graph

{{ rankTitle }}

{{ rankSub }}

{{ rmo.label }}

{{ chip.label }}

Colored by: {{ ryo.label }}

Colored by: {{ chip.label }}

Rank: Highest Lowest by {{ chip.label }} 50K+ Workers

Show next 10 More Republican More Democratic

⬇ Save Graph⧉ Copy Graph

{{ trendTitle }}

{{ trendSub }}

{{ tmo.label }}

{{ chip.label }}

Group counties into: {{ chip.label }} Weight by workers

⬇ Save Graph⧉ Copy Graph

{{ bucketTitle }}

{{ bucketSub }}

Election: {{ byo.label }}

Election: {{ chip.label }}

Modes shown: {{ chip.label }}

Average: Per county Weighted by workers

⬇ Save Graph⧉ Copy Graph

{{ shiftTitle }}

{{ shiftSub }}

{{ hmo.label }}

{{ chip.label }}

From: {{ chip.label }} To: {{ chip.label }}

Include: {{ chip.label }} Size by density

Shifted Republican Shifted Democratic Local average (LOESS)

Counties plotted: {{ shiftN }} Correlation (r): {{ shiftR }} {{ shiftStrength }}

{{ shiftCalloutNote }}

{{ shiftLabelBtn }}

Dot size: fewer workers per sq mi (2020–24) more

⬇ Save Graph⧉ Copy Graph

Notes: Commute mode share data from the U.S. Census Bureau American Community Survey, 5-year estimates (2020–2024), sourced via IPUMS NHGIS; data represent workers 16 years and over. Throughout this page, “workers” means resident workers — people who live in a county and are employed, counted by where they live rather than where they work. It is not a count of jobs or of the workforce present in a county: commuters who travel in from elsewhere are counted in their home county, not their destination. Worker density therefore describes how densely a county’s own working residents are settled, not how many people work there, and places that draw large numbers of in-commuters — central business districts especially — will show lower worker counts and densities than their daytime activity suggests. Presidential election returns from the MIT Election Data + Science Lab, County Presidential Election Returns 2000–2024 (Harvard Dataverse, doi:10.7910/DVN/VOQCHQ). Democratic vote share is calculated as the Democratic candidate’s share of all votes cast. The vote margin labels (D+10, R+20, and so on) therefore describe the Democratic share measured against all other votes cast, which are mostly but not exclusively Republican; in counties with significant third-party or independent voting, an “R+” label indicates a non-Democratic majority rather than a strictly Republican one. Each election is paired with the contemporaneous census or ACS vintage rather than with current commute data: the 2000 election uses the 2000 Census, 2004 and 2008 use the 2005–09 ACS, 2012 and 2016 use the 2015–19 ACS, and 2020 and 2024 use the 2020–24 ACS. The one exception is the vote-shift chart, where counties must remain in a single group across the whole series, so groups are defined by 2020–24 commute shares. Worker density is each county’s commuting workers divided by its Census land area (ALAND, which excludes inland water), in workers per square mile. With “size by density” enabled, each dot’s radius reflects that county’s worker density. Because density spans roughly five orders of magnitude, radius is scaled on log density rather than linearly, so the densest counties remain comparable in size to the rest rather than dominating the plot; dot area is therefore an ordinal guide to density, not a proportional readout. On the change-over-time chart, dot size always uses 2020–24 worker density regardless of which endpoint years are compared, so size means the same thing across every comparison. The change-over-time chart compares only the four elections that map to distinct vintages (2000, 2008, 2016, 2024), since consecutive elections sharing an ACS vintage would show no change in commuting. The local average drawn on the scatter plots is a LOESS fit (local linear regression, tricube weighting, 45% span) computed in the plotted coordinate space. Commute and election data are joined on county FIPS code; 3,118 of 3,244 counties match. Alaska is excluded, since it reports results by state legislative district rather than by borough or census area, as are a small number of counties with boundary changes across the two sources. Correlation coefficients describe association only and do not imply causation — counties differ in density, income, and age, all of which shape both commuting and voting.