Chapter 02 · Methods

How This Analysis Isolates the Congestion Zone's Effect From Broader Atmospheric Change.

This chapter explains the placebo comparison, the satellite measurements, the neighborhood aggregation, the statistical model, and the limits of what the data can responsibly say.

The placebo comparison

The Analysis Compares Two Pairs, One Year Apart.

Air pollution changes for many reasons — weather, fleet turnover, post-pandemic recovery, seasonal cycles. To estimate whether congestion pricing mattered, the analysis needs a comparison representing what would likely have happened without it.

The "real" comparison measures NO₂ change between Jan 2023 – Apr 2024 and Jan 2025 – Apr 2026, around the January 5, 2025 toll start. The "placebo" comparison repeats the same exercise shifted one year earlier — 2024 vs. 2023 — when no congestion zone existed.

Anything appearing in both comparisons is background atmospheric drift. Anything appearing only in the real comparison is more plausibly the policy. The placebo year shows little comparable signal over Manhattan, strengthening the interpretation that the post-2025 pattern reflects more than ordinary atmospheric variability. (See the placebo map on the air chapter.)

The satellite

What TROPOMI Measures, and What It Doesn't.

This project uses nitrogen dioxide measurements from TROPOMI, a spectrometer aboard the European Space Agency's Sentinel-5P satellite. TROPOMI estimates atmospheric NO₂ by measuring how sunlight reflects through the atmosphere. The result is a "column density" measurement: the total amount of NO₂ present in the air column above each location.

Column NO₂ is not identical to street-level exposure. Because most NO₂ remains concentrated in the lower atmosphere near emission sources, satellite measurements are strongly correlated with traffic-related urban pollution patterns, but they are a regional proxy rather than a direct measure of breathing-level air.

The analysis uses monthly average grids rather than daily measurements to reduce weather noise. The comparison windows are season-matched (Jan 2023 – Apr 2024 against Jan 2025 – Apr 2026), so winter peaks and summer troughs cancel out. TROPOMI's native footprint is roughly 5×3 km — sufficient for identifying broad regional patterns, but not for individual blocks or roadside conditions.

Uncertainty

Confidence Intervals From Bootstrap Resampling.

Confidence intervals throughout the project come from bootstrap resampling — repeatedly rerunning the same comparison on slightly altered versions of the data to estimate how stable results remain across resamples.

Statistical model

Which Neighborhood Traits Predict Bigger Pollution Drops?

A Spatial Lag Model accounts for the fact that nearby neighborhoods influence one another environmentally. Air pollution does not stop at administrative boundaries. The model estimates how strongly different neighborhood traits predict NO₂ change while controlling for the others simultaneously.

Each row shows how strongly a neighborhood trait predicts NO₂ change, holding the others constant. Bars to the right of the dashed reference line at 0 are positive predictors; bars whose interval does not cross 0 are statistically distinguishable. Distance to the zone, the share of households without a car, and the share of transit commuters show the strongest evidence.

The model explains roughly one-third of neighborhood-level variation — substantial enough to interpret the coefficients meaningfully, but limited enough to avoid overstating precision.

A note on the South Bronx

Different Instruments, Different Answers.

While this analysis was being completed, Columbia and South Bronx Unite released ground-monitor findings showing PM2.5 increased at 17 of 19 South Bronx monitoring sites during the congestion zone's first year, especially near major expressways.

Both findings may be true simultaneously.

This project measures NO₂ using a satellite instrument with a native footprint of roughly 5×3 km. The South Bronx study measures PM2.5 using ground-level monitors capable of detecting highly localized roadside changes. A regional satellite signal showing cleaner air over lower Manhattan does not rule out hyperlocal pollution increases along truck diversion corridors.

The fairest interpretation is that congestion pricing's environmental effects were measurable but uneven, and that some communities already burdened by traffic pollution may have absorbed costs this analysis cannot fully resolve.

Things this analysis cannot see

Limits and What They Mean.

• TROPOMI's native satellite footprint is roughly 5×3 km, making broad regional patterns more reliable than block-level neighborhood interpretation.
• This project measures NO₂ only. It does not directly measure PM2.5, ozone, indoor exposure, or cumulative health outcomes.
• Sixteen months is enough to detect a signal, but not enough to determine long-term trajectory.
• High-resolution FHV and rideshare trip records were unavailable at the spatial scale needed to evaluate Uber/Lyft substitution effects.
• Some transportation findings in later chapters partially reflect simultaneous MTA service changes rather than pure congestion-pricing effects alone.

These limitations do not invalidate the findings. They define the scale and scope of what this analysis can responsibly claim.