Pre-registration
Biden's IRA/CHIPS industrial policy will show partial success on capacity-building metrics and mixed results on job creation, consistent with the conditional view that industrial policy works where targeting is technically competent and governance is strong.
Falsification criterion — what would disprove this
This hypothesis is considered falsified if:
PRIMARY (dispositive, evaluated 2030): the conditional claim is SUPPORTED iff ALL THREE of: (a) US semiconductor manufacturing value added gains >30% over a 2021 baseline (capacity-success), (b) US sector employment gains <15% over the same baseline (jobs-mixed), AND (c) cumulative private capex commitments triggered by CHIPS exceed $200B (leverage). REFUTED iff capacity gain <10% AND capex <$100B by 2030 (no industrial-base shift). PARTIAL otherwise. METHOD_VALID gate: the primary outcome series — oecd:STAN_INDUSTRY ISIC C26, bls:CES3133, ilostat semiconductor employment, and a CHIPS capex tracker — must be available, AND the post-treatment window must be at least 4 years of observed data. If either fails, verdict is INCONCLUSIVE (data gap or window too short), not refuted.
formal test & threshold
test: chips_ira_2030_multi_metric_pattern_check threshold: PRIMARY: capacity_VA_change_vs_2021 > 0.30 AND sector_employment_change_vs_2021 < 0.15 AND cumulative_private_capex_USD_bn > 200. METHOD_VALID: all 4 spec-named series available AND post_treatment_years >= 4.
Method
- Template
multi_metric_checklist- Sample
- 5 countries · 2018 – 2030
- Evidence type
- associational
Pre-registered multi-metric pattern check by 2030: (a) US semi-fab capacity gain >30% over 2021 baseline = capacity-success metric; (b) aggregate sector employment gain <15% = jobs-mixed metric; (c) cumulative private capex commitments >$200B = leverage metric. Hypothesis SUPPORTED if pattern matches all three conditional metrics. Caveat: 2030 horizon means primary spec runs incomplete until then — flagged TODO swarm-6e. Cyclical-demand confound (post-COVID semiconductor cycle) addressed by cross-country comparison to KOR/TWN/JPN/DEU peers.
Data
| Variable | Source | Transform |
|---|---|---|
log_semiconductor_manufacturing_value_added outcome | oecd:STAN_INDUSTRYtier 2 | log |
semiconductor_employment_thousands outcome | bls:CES3133tier 1 | log |
semiconductor_employment_ilostat outcome | ilostat:semiconductor_employmenttier 2 | log |
capex_announced_chips_act outcome | constructed:hand-coded CHIPS Act award announcements + private capex commitments (DoC database). Total $ billions cumulative.tier 5 | level |
chips_act_post_2022_indicator treatment | constructed:indicator = 1 for USA from 2022-08 (CHIPS and Science Act enactment); 0 for comparator pool.tier 5 | indicator |
ira_post_2022_indicator treatment | constructed:indicator = 1 for USA from 2022-08 (Inflation Reduction Act); separate from CHIPS for the energy/clean-tech subsidy chantier 5 | indicator |
log_gdp_per_capita control | world_bank_wdi:NY.GDP.PCAP.KDtier 2 | log |
trade_openness control | world_bank_wdi:NE.TRD.GNFS.ZStier 2 | level |
● ready · ● pending · ● reconstruct-needed
Detailed result card
CHIPS Act / IRA semiconductor industrial policy effectiveness
Verdict: inconclusive — Stacked: (4/4) spec-named semi-specific series unavailable on disk (oecd:STAN_INDUSTRY ISIC C26, bls:CES3133, ilostat:semiconductor employment, constructed CHIPS capex), AND post-treatment window is 0 year(s) vs spec horizon 2030 (8-year gap). CHIPS/IRA fab build-out lead-times of 3-5 years mean leading-edge US fabs (TSMC AZ, Intel OH, Samsung TX) are not expected to reach steady-state output until 2026-2028; the spec's 2030 horizon is a deliberate acknowledgement of this. Re-run when semi-specific fetchers land AND post-2022 vintage extends to 2027+.
Summary
Spec proposes a 2030 multi-metric pattern check on US CHIPS Act + IRA industrial policy: capacity-VA gain >30%, employment gain <15%, private capex >$200B. Two stacked obstacles prevent dispositive evaluation today:
- Data gap — 4/4 spec-named semiconductor-specific series are not available in current vintages. The proxy on disk (broad industry VA) is too coarse, and for the treated unit (USA) the proxy itself is sparse.
- Window too short — Treatment year is 2022; latest available data extends to 2022 (0 post-treatment years vs 4-year minimum). CHIPS Act fab build-out lead-times are 3-5 years before commissioning; the spec's deliberate 2030 horizon reflects that lead-time. Even with perfect data, primary tests are not informative until 2027-2028 at the earliest.
Method
v1 promotion (2026-04-24). The replication encodes:
- A series-availability audit against the four spec-named series,
- A post-treatment window-length check (≥4 years required),
- A best-available proxy probe (world_bank_wdi:NV.IND.TOTL.KD) showing the treated unit (USA) has near-zero coverage in the post-2018 window even at the broad industry level.
Either obstacle alone would justify inconclusive; both together make the case unambiguous. The script is structured to re-run cleanly once the missing series land — primary thresholds (30% capacity / 15% employment / $200B capex) are pinned as constants.
Data
Required (missing — fetcher backlog):
oecd:STAN_INDUSTRY_C26— OECD STAN ISIC C26 (computer, electronic, optical products) value addedbls:CES3133— BLS CES semiconductor & other electronic component mfg employmentilostat:semiconductor_employment— ILOSTAT cross-country sectoral employment, NACE C26constructed:chips_act_capex_commitments— Hand-coded CHIPS Act award announcements + private capex commitments
Available (used for window probe only):
- (none)
Reproduction
.venv/bin/python3 engine/runs/industrial_policy_semiconductor_chips_act_effectiveness/replication.py
Notes
Origin is auto-generated coverage-gap stub seeded from developmentalist + pragmatist framings of CHIPS/IRA producing partial capacity gains and mixed jobs effects. Human review required.