Skip to the content.

Presidential Readability

@unwelcomedata · data from public sources

Have presidential speeches gotten simpler over time? Yes — markedly. This project measures the reading-grade level of U.S. presidential oratory across the spoken/broadcast era (1913–present) and finds a clear long-term decline: the State of the Union has fallen from a mid-teens/college reading level to around 8th grade today.

Reading grade is measured with Flesch–Kincaid (and three companion formulas as a cross-check). A higher number means more complex text — roughly, the U.S. school grade you’d need to read it comfortably.


Presidential readability — infographic


The State of the Union got simpler

State-of-the-Union reading grade level, 1913–2026

Average Flesch–Kincaid grade level of each State-of-the-Union address, spoken/broadcast era (1913 on). The series starts in 1913 on purpose — see How it was measured below.

Four formulas agree

State-of-the-Union readability by decade, four formulas

Flesch–Kincaid, SMOG, Gunning Fog, and Coleman–Liau, averaged by decade. They all trend the same way — the decline isn’t an artifact of one formula’s quirks.

Inaugurals too

Inaugural address reading grade level, 1913–2025

Inaugural addresses — a different, always-spoken speech type — show the same downward drift (noisier, since it’s one speech per term).


How it was measured

Honest limits:

The data

Reproduce it

The pipeline is a single standalone command (raw source → published export):

pip install -r requirements.txt
python scripts/reproduce.py        # Miller Center corpus -> clean -> readability -> export
python scripts/validate_charts.py  # re-checks the chart facts; exits 0 if all good

scripts/reproduce.py runs the same src/ logic the analysis uses, in order (ingest → clean → prepare). Against the same archive snapshot it writes a byte-identical export; the Miller Center updates its archive occasionally, so a fresh download can differ slightly (see Corrections in SOURCES.md).

Sources & license

Full attribution, methodology, and series-break notes: SOURCES.md. Source corpus is public domain (U.S. government works); this project’s code and derived dataset are shared for public use.


AI-Assisted Development This project was built with the assistance of Kiro, an AI-powered development environment. All data sourcing decisions, methodology choices, and published findings are the responsibility of the author. AI was used for code generation, data pipeline construction, and research assistance — not for analysis conclusions or editorial judgment.