| Script | Role |
|---|---|
| phase2/code/deliverable05/01_extract_law_citations.py | Scans CE/EA/EIS document pages for explicit ARRA / BIL / IRA citations, with context-based disambiguation of the IRA and BIL acronyms |
| phase2/code/deliverable05/02_build_ce_categories.py | Parses the document-level ce_category metadata into normalized categorical-exclusion codes (DOE 10 CFR 1021, DOI 516 DM 11, EPAct 2005 Section 390) |
| phase2/code/deliverable05/03_analyze_spikes.R | Joins dates, citations, and categories; produces all figures and diagnostic tables |
NEPA Decarbonization Technology Analysis: Deliverable 5
Categorical Exclusion Spikes After Major Infrastructure Legislation
Executive Summary
This deliverable examines whether the use of categorical exclusions (CEs) — the lightest form of NEPA review — spikes after the passage of major infrastructure legislation, and if so, whether those actions are demonstrably tied to the new law. We anchor every CE to the date its determination was issued (from the Deliverable 4 timeline), detect explicit citations to the legislation in the CE documents, and read the specific categorical-exclusion category each action invoked.
The three named laws are the American Recovery and Reinvestment Act (ARRA, Feb 2009), the Bipartisan Infrastructure Law / Infrastructure Investment and Jobs Act (BIL/IIJA, Nov 2021), and the Inflation Reduction Act (IRA, Aug 2022).
Key Findings
There is a large, unmistakable CE spike after ARRA — and it is almost entirely a Department of Energy phenomenon. DOE categorical exclusions jumped from 19 in 2008 to 698 in 2009 and 3,868 in 2010, while the other major CE user, the Bureau of Land Management, stayed flat (74 in 2010). The agency that administered ARRA’s energy money is exactly the agency whose CEs surged.
The spiking actions are demonstrably tied to the law. 59.4% of CEs issued in the ARRA window explicitly cite the Recovery Act in their documents — direct attribution, not a temporal coincidence.
The type of CE used shifted decisively toward the law’s purpose. Among DOE categorical exclusions in the ARRA window, category B5.1, “Actions to conserve energy or water,” rose to 49.6% of all DOE CEs (from 1.3% at baseline) — the energy-efficiency stimulus showing up directly in the categorical-exclusion mix.
BIL and IRA show a real but far more muted pattern. DOE CE activity rose after BIL (1.59× the pre-law monthly baseline), but explicit CE citations to BIL and IRA are rare — those laws are referenced far more often in larger reviews (EISs) than in CEs (IRA is cited in 18.9% of post-IRA EISs vs a negligible share of CEs).
Read raw counts with care. The dataset’s coverage thins before 2009 and is incomplete for 2024–2025, so aggregate counts overstate the ARRA step-change. The DOE-vs-BLM split and the citation evidence isolate the genuine policy signal from this coverage ramp.
Methodology
The question, in three parts
The deliverable asks three distinct questions, each answered by a distinct analysis:
| Question | Analysis |
|---|---|
| Is there a spike in CE use after each law? | CE counts by determination year, with legislative markers — overall, by review type, by department, and DOE-vs-BLM. |
| Are the actions associated with the law? | Detection of explicit law citations in the CE document text. A citation can only post-date the law, so it is coverage-robust attribution evidence. |
| What types of CEs were used? | The specific categorical-exclusion category invoked (e.g. DOE’s B5.1), plus the technology mix of the projects, in the spike window vs a baseline. |
Base population
Every CE is placed in time by its determination (decision) date. Where that is absent we fall back to the initiation date as a same-year proxy — safe for CEs, whose median duration is roughly three weeks, so the two dates almost always fall in the same calendar year. This yields a base of 52,089 of 54,040 CE projects (96.4%) that can be located on the timeline — far broader than the complete-timeline base used for the duration analysis in Deliverable 4, because placing a CE in a year requires only one date, not two.
Data pipeline
A note on the coverage ramp
A raw count of CEs per year mixes three things: genuine policy-driven surges, the growth of NEPATEC’s document coverage over time (the dataset is sparse before 2009), and a drop-off in 2024–2025 caused by recent documents not yet being ingested. We therefore do not lean on aggregate counts to make the causal claim. Two devices isolate the real signal: conditioning on agency (the ARRA surge is DOE-only while BLM — subject to the same dataset — is flat), and citation evidence (a Recovery Act citation cannot appear before the Recovery Act existed).
The Spike
Categorical exclusions rise sharply around ARRA and again, more modestly, in the BIL/IRA period.
The spike is specific to categorical exclusions. Faceting the same series by review type shows that the more demanding Environmental Assessment and Environmental Impact Statement processes do not exhibit the post-ARRA jump — consistent with a surge of fast, low-burden actions rather than a general increase in NEPA activity.
The spike is a DOE phenomenon
The single most important result is that the post-ARRA CE surge belongs almost entirely to the Department of Energy — the agency that administered ARRA’s energy grants, loan guarantees, and weatherization funds. DOE CEs went from 19 (2008) to 698 (2009) to a peak of 3,868 in 2010. Over the same years the Bureau of Land Management — the only other agency that issues CEs at scale, and one drawing on the same dataset — barely moved (40, 74, 104 in 2009–2011). BLM’s own growth comes later (2016 onward) and is unrelated to the legislation. Because both agencies are subject to the identical coverage ramp, this contrast is the cleanest available evidence that the spike is a policy effect, not a data artifact.
| Mean monthly CE volume: spike window vs pre-law baseline | ||||
| Subset | Law | Window (CEs/mo) | Baseline (CEs/mo) | Spike ratio |
|---|---|---|---|---|
| All CE | ARRA | 203.3 | — | — |
| All CE | BIL | 425.7 | 348.4 | 1.22 |
| All CE | IRA | 328.0 | 379.2 | 0.86 |
| DOE | ARRA | 203.0 | — | — |
| DOE | BIL | 260.9 | 164.5 | 1.59 |
| DOE | IRA | 207.5 | 195.8 | 1.06 |
| BLM | ARRA | 6.3 | — | — |
| BLM | BIL | 163.1 | 182.0 | 0.90 |
| BLM | IRA | 119.5 | 181.6 | 0.66 |
| ARRA has no usable pre-law baseline (the dataset is sparse before 2009); its spike is established by the DOE-vs-BLM contrast and citation evidence rather than a window/baseline ratio. | ||||
By energy type and department
Splitting the CE series by energy type (decarbonization, fossil, other) and by lead department provides the all-data and breakdown views requested for every analysis.
Association With the Law
A temporal coincidence is not attribution. The strongest test is whether the spiking documents say the action stems from the law. We detect explicit citations to each statute in the document text (with context disambiguation for the ambiguous “IRA” and “BIL” acronyms).
For ARRA the attribution is overwhelming: 59.4% of CEs issued in the ARRA window cite the Recovery Act by name, versus a negligible rate outside it. The Recovery Act citations themselves peak in 2010 and decay over the following years as stimulus obligations wound down — the citation curve tracks the disbursement curve.
BIL and IRA tell a more nuanced story. Their citation rates rise after passage relative to baseline, confirming a real association, but the absolute rate within CEs stays low — these laws are seldom named in a one-page categorical exclusion. They surface far more often in larger reviews: IRA is cited in 18.9% of post-IRA EISs (vs 8.7% before), and BIL in 17.5% of post-BIL EISs. In other words, the citation evidence for BIL/IRA lives in the EIS record, even though the volume signal (more DOE CEs) is real.
| Law-citation rate by scope and period | ||||
| Law | Scope | Period | % citing | N |
|---|---|---|---|---|
| ARRA | All CE | spike window | 59.4% | 6,913 |
| BIL | All CE | spike window | 0.4% | 10,642 |
| BIL | All CE | baseline | 0.0% | 12,544 |
| IRA | All CE | spike window | 0.1% | 8,200 |
| IRA | All CE | baseline | 0.0% | 13,651 |
| ARRA | All EIS | spike window | 25.6% | 297 |
| BIL | All EIS | spike window | 17.5% | 206 |
| BIL | All EIS | baseline | 3.7% | 431 |
| IRA | All EIS | spike window | 18.9% | 222 |
| IRA | All EIS | baseline | 8.7% | 390 |
What Types of CEs Were Used
NEPATEC records the specific categorical-exclusion category each action invoked. For DOE these are the codes of 10 C.F.R. 1021 (e.g. B5.1, A9, B1.3); for the Interior/BLM they are the 516 DM 11 series; oil-and-gas drilling uses the Energy Policy Act of 2005 Section 390 exclusion. Comparing the mix inside the ARRA window against a stable baseline shows the categorical-exclusion profile shifting directly toward the stimulus’s purpose.
The standout is B5.1, “Actions to conserve energy or water,” which accounts for 49.6% of DOE categorical exclusions in the ARRA window versus only 1.3% at baseline — a roughly forty-fold concentration. Information-gathering and technical-assistance categories (A9, A11, A1) are also elevated, consistent with a wave of grant- and study-related actions. By contrast, the baseline period is dominated by routine-maintenance exclusions (B1.3), which recede during the stimulus. The categorical-exclusion mix, in short, reorganizes around energy efficiency exactly when ARRA’s efficiency money arrives.
| Top DOE categorical-exclusion codes in the ARRA window | |||
| Code | Description | ARRA window % | Baseline % |
|---|---|---|---|
| B5.1 | Actions to conserve energy or water | 49.6% | 1.3% |
| A9 | Information gathering, data analysis & document preparation | 36.4% | 15.2% |
| B3.6 | Small-scale research, development & demonstration projects | 23.4% | 28.7% |
| A11 | Technical advice and planning assistance | 17.9% | 7.7% |
| B2.5 | — | 12.5% | 2.5% |
| A1 | Technical/financial assistance, advice, training & education | 11.2% | 3.7% |
| B3.1 | Site characterization and environmental monitoring | 6.4% | 5.5% |
| B1.3 | Routine maintenance activities | 6.0% | 28.6% |
Technology mix
By project technology, the ARRA-window CEs are dominated by DOE environmental-management and research-and-development categories alongside the expected clean-energy build-out (solar, transmission).
| Technology tag | Share of ARRA-window CEs | N |
|---|---|---|
| Waste Management | 57.0% | 3,934 |
| Research and Development | 41.9% | 2,895 |
| Utilities (electricity, gas, telecommunications) | 34.6% | 2,390 |
| Land Development - Urban | 14.8% | 1,023 |
| Water Resources - Other | 14.7% | 1,013 |
| Manufacturing | 13.8% | 952 |
| Renewable Energy Production - Solar | 13.7% | 947 |
| Electricity Transmission | 13.1% | 906 |
Caveats and Limitations
- Correlation vs. attribution. The volume spike is correlational; the citation analysis is the stronger attribution layer. We present both and lean on citations for the causal claim.
- Pre-2009 coverage is thin. NEPATEC contains few CE documents before 2009, so ARRA’s apparent level shift is partly a coverage ramp. The ARRA claim rests on the DOE-vs-BLM contrast and the 59.4% citation rate, not on raw counts, and we report no ARRA window/baseline ratio.
- BIL and IRA windows overlap. Passed nine months apart, their post-law windows cannot be cleanly separated by date alone; citations are used to attribute where possible.
- 2024–2025 are incomplete. The decline in the most recent years reflects ingestion lag, not a real fall in CE use.
- Dates are extracted, not authoritative. The determination date comes from the Deliverable 4 extraction (≈96.4% of CEs placeable); roughly the remaining few percent carry no usable date and are absent from the time series.
Reproducibility
All figures and tables in this report regenerate from committed code. Run from the repository root in the nepa conda environment:
python phase2/code/deliverable05/02_build_ce_categories.py
python phase2/code/deliverable05/01_extract_law_citations.py --source all
Rscript phase2/code/deliverable05/03_analyze_spikes.ROutputs are written to phase2/output/deliverable05/{figures,diagnostics}/; the intermediate citation and category datasets to phase2/data/analysis/deliverable05/.