Home › Air Quality › Methods

Air Quality: Methods and Validation

Updated August 13, 2026

Every air quality figure on this site is recomputed from EPA's raw monitor records rather than copied from a summary table, because the question we are asking, what the same air looks like under two different rulebooks, cannot be answered from published summaries. That puts the burden on us to show the calculation is right. This page shows it, including where it is wrong and what it cannot support.

Two definitions, both frozen

The Air Quality Index converts a pollutant concentration into a 0–500 score. The conversion table has changed, so we compute every day twice:

• The 2010 rulebook. 40 CFR part 58, Appendix G, Table 2, as printed in the 7-1-2011 edition of the Code of Federal Regulations. This is the table in force when EPA published its 2011 trends report.
• Today's rulebook. The current Appendix G, reflecting the 2013 and 2024 fine particle revisions and the 2015 ozone revision.

Both are applied to the same monitors, the same metro boundaries and the same treatment of wildfire days. Only the conversion table differs, which is what makes the two series comparable to each other.

How the calculation works

Following Appendix G section 12, for each area and day: take the highest concentration among all monitors in the area for each pollutant; truncate it and convert it to a sub-index by interpolating between the bracketing breakpoints; the area's AQI for that day is the highest sub-index across pollutants. Source data is EPA's AirData daily files for ozone, fine particles (both reference and non-reference methods), coarse particles, carbon monoxide, sulfur dioxide and nitrogen dioxide, 2000–2025, 182 files in total.

Ozone design values, used on the attainment pages, follow 40 CFR part 50 Appendix P: the three-year average of the annual fourth-highest daily maximum 8-hour concentration, truncated to three decimals, with a county taking the highest value among its monitors.

Validation against EPA's own numbers

Before any of it was used, the calculation was checked against what EPA publishes:

Where we use EPA's numbers instead of ours

On the attainment page we use EPA's published design values, not our own. Legal attainment excludes wildfire and dust days that states have applied to have discounted, and EPA's public daily files do not carry the discounted values: only 202 monitor-days nationally in 2022. Our calculation cannot reproduce that exclusion, and at the margin where attainment is judged, that is exactly what matters. Our values remain the right tool for consistent-method trends, where the same rule is applied to every year.

What this cannot show

• Whether the standards are set at the right level. That rests on health evidence we have not independently evaluated.
• Whether observed improvements were caused by any particular rule. The comparison world, in which the rule was never written, cannot be observed.
• Why ozone stopped responding to emission cuts in the East. Several explanations fit and this data cannot separate them.
• Wildfire smoke from wildfire weather. Both come from the same hot dry summers.
• Any short-term link between ozone and respiratory illness. We built that analysis and it failed its own validity check. See the corrections below.

Corrections

Investigative work raises the cost of being wrong, so every error we have found in our own analysis is listed here rather than quietly fixed.

1. “Ozone progress has no trend since 2014.” Eyeballed and too strong. The fitted trend is a 17-fold deceleration, not a stop. Corrected before publication.
2. “The East ran out of emission reductions.” Our own explanation for the eastern ozone plateau, inferred from the shape of the ozone data. Testing it against EPA's emissions inventory falsified it: emissions kept falling at nearly the same rate. The published version reflects the corrected finding.
3. A short-term ozone and respiratory illness estimate. Our model returned a clean, statistically significant result, and so did a control test using influenza, which cannot respond to ozone within days. That means the model was measuring something other than ozone. The estimate is not published, in either direction.
4. A benefits calculation that came out backwards. Comparing EPA's modelled ozone projections against observed monitor readings produced a negative result. The two are not the same measurement: EPA's modelled values sit about 11 ppb below EPA's own published design values, because the projections work in ratios rather than absolute levels. Redone correctly, the sign reversed.
5. Two calculation bugs caught by the validation above: a rounding rule that differed from EPA's at exact half-values, and an omitted class of fine-particle monitor that carried most of the unhealthy days in some western cities.

Sources

• EPA AirData daily monitor files and annual summaries.
• EPA air quality design value reports.
• EPA Green Book nonattainment designations.
• EPA Air Pollutant Emissions Trends.
• 40 CFR part 58 Appendix G (Air Quality Index) and part 50 Appendix P (ozone design values).