economy_ai_at_risk_population_county_year
County-year (2024 baseline). The AI at-risk population layer — the measured denominator the displacement shock sweep acts on. Reports how much of each county's knowledge workforce (SOC major groups 11/13/15/17/19/23/25/27) sits in AI-exposed occupations, the earned income it represents, and where it sits in the county wage distribution. Two at-risk measures: at_risk_headcount (knowledge workers in occupations at/above the national employment-weighted top-tercile AEI automation threshold) and labor_at_risk_continuous (employment × automation share, graded). AIOE columns are a robustness cross-check. The d/r-bar displacement sweep is parameterized in the analysis layer, not in this table.
overview
3,170 counties for 2024 (single baseline year). automation_coverage_knowledge reports the share of each county's knowledge employment carrying an AEI automation score; the AEI scores only the ~3.4k O*NET tasks seen in Claude.ai traffic, so occupations with thin task coverage carry noisier automation shares.
current vintage — 2024
history — No — single 2024 baseline vintage
source & licensing
fields
| name | type | definition |
|---|---|---|
| aioe_high_share_of_knowledge | float64 · share (0–1) | knowledge_employment_aioe_high divided by knowledge_employment; robustness share of knowledge workers in above-average-AIOE occupations. |
| at_risk_earned_income | float64 · dollars (nominal, annual) | Aggregate annual wages of the at-risk knowledge employment (headcount × OEWS occupation mean wage). The earned income exposed to displacement. |
| at_risk_headcount | float64 · workers | Knowledge-sector employment in occupations whose AEI automation share is at or above automation_threshold. The at-risk population as a headcount.An exposure denominator, not a job-loss forecast. |
| at_risk_household_buffer | float64 · share (0–1) | Employment-weighted household-income pass-through buffer of the at-risk knowledge mass — the earner-to-household bridge applied. Each occupation's wage percentile is rank-matched to its SOI agi_stub income band, whose filing-status mix gives a pass-through buffer (single and head-of-household returns pass fully; joint returns cushion). A household-income change is roughly the wage change times this factor; lower means more two-earner cushioning.Null where no SOI income band matched (a few small counties). |
| at_risk_mean_wage_percentile | float64 · percentile (0–1) | Employment-weighted mean wage percentile of the at-risk knowledge mass within the county wage distribution. Positions the at-risk population.Expected upper-middle (~0.78), not the bottom — the bridge to the distributional analysis. |
| at_risk_share_of_all | float64 · share (0–1) | at_risk_headcount divided by total_employment; the at-risk knowledge population as a share of all county employment. |
| at_risk_share_of_knowledge | float64 · share (0–1) | at_risk_headcount divided by knowledge_employment; share of the county knowledge workforce in at-risk occupations. |
| automation_coverage_knowledge | float64 · share (0–1) | Share of knowledge-sector employment carrying an AEI automation score (coverage flag). Lower values mean more knowledge employment is unscored and the at-risk measures less complete.AEI scores only the ~3.4k O*NET tasks in Claude.ai traffic out of ~18k. |
| automation_threshold | float64 · share (0–1) | National employment-weighted top-tercile cutoff of the Anthropic Economic Index automation share; knowledge-sector occupations at or above this value are counted as at-risk. Identical across all counties.A stated threshold, not data — emitted so the at-risk cut is auditable. |
| country_id | string | ISO alpha-2 country code (always 'US' for domestic tables). |
| county_idkey | string | 5-character FIPS code identifying the county.Part of primary key. Joins dim.counties on county_id. |
| knowledge_employment | float64 · workers | County employment in knowledge-sector occupations (SOC major groups 11, 13, 15, 17, 19, 23, 25, 27: management, business/finance, computer/math, architecture/engineering, science, legal, education, arts/media).The knowledge-sector denominator; a stated definition, not an assumption. |
| knowledge_employment_aioe_high | float64 · workers | Knowledge-sector employment in above-average AIOE (Felten-Raj-Seamans, AIOE > 0) occupations.Robustness cross-check against the AEI automation cut. |
| labor_at_risk_continuous | float64 · workers (exposure-weighted) | Employment-weighted sum of the AEI automation share over knowledge-sector occupations (sum of employment × automation share). The graded partial-risk measure.Occupations without an automation score contribute zero — a downward bias; see automation_coverage_knowledge. |
| state_id | string | 2-character FIPS code identifying the state.Joins dim.states on state_id. |
| total_employment | float64 · workers | Total county employment summed across occupations, from the QCEW county industry mix expanded by OEWS national NAICS×SOC staffing shares (shift-share estimate). Higher values indicate a larger county workforce. |
| yearkey | int64 | Reference year of the at-risk baseline (2024).Part of primary key. Single 2024 vintage. |
joins
how to use this table
County NAICS employment (QCEW) is expanded into occupation employment via OEWS national NAICS×SOC staffing shares (shift-share), priced at the OEWS SOC mean annual wage. Per-occupation AEI automation share is rolled up from O*NET task collaboration modes (directive + feedback_loop) to 6-digit SOC. Knowledge-sector occupations (SOC major groups 11/13/15/17/19/23/25/27) at or above the national employment-weighted top-tercile automation share (automation_threshold) are counted as at-risk (headcount); the continuous measure sums employment × automation share. Each occupation's mid-percentile in the county wage distribution gives the positioning. The d/r-bar displacement sweep is applied in the analysis layer, not here.
Reading at_risk_headcount as a job-loss forecast (it is an exposure denominator, not a displacement count — the d/r-bar sweep lives in the analysis layer); single-year change detection (single 2024 vintage); treating the automation share as realized automation rather than a graded exposure measure.
AEI scores only the ~3.4k O*NET tasks in Claude.ai traffic out of ~18k, so per-SOC automation shares are computed over ~20% of an occupation's tasks on average (automation_coverage_knowledge flags this); labor_at_risk_continuous treats unscored knowledge employment as automation zero, a downward bias. QCEW statewide aggregate-area pseudo-counties (000, 996-999) and CT/territory FIPS without a dim.geographies match are filtered upstream in int_county_occupation_employment.