{ "early_release_partial_reporter_set": true, "generated_on": "2026-08-31", "headline": { "audited_serving_utility": { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 2.330365774648688, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 0.17663575222681632, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 18.70238632158988, "mapping": "audited_serving_utility", "never_treated_units": 1161, "overall_att_pct": 3.3194113343851015, "overall_ci_high_pct": 6.91283910449397, "overall_ci_low_pct": -0.2740164357237673, "overall_p_value": 0.07021202246820163, "overall_se_pp": 1.8333815153616677, "post_treatment_rows": 240, "sample_rows": 17532, "threshold_mw": 5, "treated_units": 35, "utilities": 1108, "utility_state_units": 1196, "year2_att_pct": 1.2881964146687102, "year2_ci_high_pct": 4.306199614078785, "year2_ci_low_pct": -1.7298067847413636, "year2_se_pp": 1.5397975507194255 }, "interpretation": "Replacing the county-wide treatment broadcast with documented serving utilities changes the point estimate and sharply reduces the treated sample in this public-data replication attempt. The confidence intervals overlap and neither overall estimate rejects zero, so this is treatment-definition sensitivity rather than evidence of a statistically different effect.", "paper_county_broadcast": { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 1.7220180144353316, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 0.1263332716956098, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 20.122274769637908, "mapping": "paper_county_broadcast", "never_treated_units": 959, "overall_att_pct": 1.236327204681246, "overall_ci_high_pct": 3.327093822883383, "overall_ci_low_pct": -0.8544394135208905, "overall_p_value": 0.2464556370769312, "overall_se_pp": 1.0667176623480288, "post_treatment_rows": 1019, "sample_rows": 15966, "threshold_mw": 5, "treated_units": 125, "utilities": 1006, "utility_state_units": 1084, "year2_att_pct": 1.3178185485910667, "year2_ci_high_pct": 2.8451138036154524, "year2_ci_low_pct": -0.2094767064333189, "year2_se_pp": 0.7792322729716253 } }, "matched_design": [ "EIA-861 annual bundled-service residential revenue per MWh", "utility-state panel with short-form utilities excluded", "first county-year reaching 5 MW as absorbing treatment", "Callaway-Sant'Anna doubly robust inverse-probability-tilting estimator", "never-treated comparison group and event time -1 normalized to zero" ], "not_an_exact_replication_because": [ "The paper's Aterio treatment panel is proprietary and not included with the paper.", "The paper's S&P rate-case panel is not public; its public Form 1 mechanisms are independently reconstructed here." ], "outcome_vintage": "early_release_2010_2025", "outcome_years": [ 2010, 2025 ], "outputs": [ "mpqw_specification_comparison-2025er.csv", "mpqw_threshold_comparison-2025er.csv", "mpqw_event_study-2025er.csv", "mpqw_analysis_panel-2025er.csv", "mpqw_table2_outcome_comparison-2025er.csv", "mpqw_paper_estimate_catalog.csv", "mpqw_paper_estimate_catalog.json", "mpqw_design_audit.csv", "mpqw_methods.md" ], "paper": { "authors": "Meeks, Pless, Qi, and Wang", "paper_residential_overall_att_pct": 2.1, "paper_residential_se_pp": 0.8, "ssrn_abstract_id": "6967558", "table2_outcomes": [ { "estimate_pct": 2.7, "outcome": "all", "se_pp": 0.9 }, { "estimate_pct": 2.1, "outcome": "residential", "se_pp": 0.8 }, { "estimate_pct": 2.8, "outcome": "commercial", "se_pp": 0.9 }, { "estimate_pct": 4.2, "outcome": "industrial", "se_pp": 1.6 } ], "title": "Who Pays for Growth? Evidence from Data Centers and the Grid" }, "paper_issues": [ { "extension_test": "Re-estimate with the audited campus-to-serving-utility crosswalk and compare with the county-broadcast rule.", "issue": "Serving-utility assignment", "paper_choice": "Treat every utility whose territory touches any part of an opening county.", "status": "tested", "why_it_matters": "A county can contain several utilities; non-serving utilities can be misclassified as treated." }, { "extension_test": "Report 2/5/20/50/150 MW thresholds and the signed repeated-MW model.", "issue": "Treatment intensity", "paper_choice": "Convert first county-year with at least 5 MW into a permanent binary treatment.", "status": "tested", "why_it_matters": "A 5 MW and a 500 MW opening receive the same treatment; later additions are discarded." }, { "extension_test": "Retain every supported annual MW increment with distributed-lag responses.", "issue": "Repeat additions", "paper_choice": "Keep only the first qualifying county-year.", "status": "tested", "why_it_matters": "Multi-phase campuses and later facilities in the same utility are omitted from treatment dose." }, { "extension_test": "Enter documented retirement MW as negative flows and publish the lifecycle ledger.", "issue": "Retirements", "paper_choice": "Permanent treatment after first entry.", "status": "tested", "why_it_matters": "Documented exits cannot reduce exposure." }, { "extension_test": "Publish joint pretrend tests, future-flow placebos, and clean-window/event-timing checks.", "issue": "No anticipation", "paper_choice": "Use activation year as treatment while the paper's own mechanism estimates begin moving before activation.", "status": "tested", "why_it_matters": "Pre-activation investment and planning can violate the estimator's no-anticipation condition and make activation a late treatment date, even though the paper appropriately discusses those early mechanism movements." }, { "extension_test": "Keep this limitation explicit and emphasize the all-utility bundled outcome for direct comparison.", "issue": "Deregulated-state outcome", "paper_choice": "Use bundled/default-service EIA price where retail choice exists.", "status": "partially_testable", "why_it_matters": "Bundled customers are a selected and shrinking subset, so the heterogeneity estimand can change with switching." }, { "extension_test": "Publish mapped serving-utility results and ownership-matched checks where support permits.", "issue": "Ownership heterogeneity controls", "paper_choice": "Compare treated ownership groups with all never-treated utilities.", "status": "extension_documented", "why_it_matters": "Control composition can differ from the treated ownership class." }, { "extension_test": "Publish estimates, intervals, and treated-unit counts at every threshold.", "issue": "Threshold sensitivity", "paper_choice": "State that results are similar and grow with threshold without publishing the full numerical path.", "status": "tested", "why_it_matters": "Readers cannot audit sample collapse, monotonicity, or uncertainty." }, { "extension_test": "Report an influence-function Wald test of the aggregated event-time leads and the project's wild-cluster joint-lead tests.", "issue": "Formal pretrend inference", "paper_choice": "Show event-study leads and describe little visual evidence of differential pretrends, without reporting an omnibus lead test.", "status": "tested", "why_it_matters": "Individual lead intervals are not a formal joint test of the displayed pre-treatment event-time path." }, { "extension_test": "Keep Tables 5 and 6 cataloged, but label their S&P outcome panel as unreleased rather than implying that all mechanism inputs are private.", "issue": "Rate-case mechanism reproducibility", "paper_choice": "Use the proprietary S&P Regulatory Research Associates rate-case panel.", "status": "not_publicly_replicable", "why_it_matters": "The requested and authorized rate-base, revenue, ROE, case-type, and case-duration outcomes are not released with the paper." }, { "extension_test": "Reconstruct Tables A6 and A9 from version-pinned public Form 1 data and a public FERC-to-EIA utility crosswalk.", "issue": "FERC investment mechanism reproducibility", "paper_choice": "Use S&P's compilation of public FERC Form 1 filings.", "status": "public_reconstruction_added", "why_it_matters": "The underlying filings are public even though the paper's cleaned S&P extract is not released." }, { "extension_test": "Flag the apparent label reversal for correction by the authors.", "issue": "Appendix table labels", "paper_choice": "Table A11 labels 'Yes' with 939 utilities and 'No' with 34.", "status": "paper_error", "why_it_matters": "The labels are reversed relative to Table A10's 3.8% tariff prevalence and the paper's own text." } ], "paper_issues_count": 12, "replication_label": "specification-matched treatment-substitution replication attempt", "schema_version": "1.0.0", "table2_outcome_comparison": [ { "estimate_pct": 2.7, "mapping": "proprietary_Aterio_plus_county_broadcast", "outcome": "all", "sample_end_year": 2024, "sample_rows": 14733, "sample_start_year": 2010, "se_pp": 0.9, "source": "MPQW paper Table 2", "utilities": 973, "utility_state_units": 1048 }, { "estimate_pct": 0.4185849500772415, "mapping": "paper_county_broadcast", "outcome": "all", "sample_end_year": 2025, "sample_rows": 15966, "sample_start_year": 2010, "se_pp": 1.2084064433674873, "source": "Independent treatment-substitution replication", "utilities": 1006, "utility_state_units": 1084 }, { "estimate_pct": 3.1733482489312848, "mapping": "audited_serving_utility", "outcome": "all", "sample_end_year": 2025, "sample_rows": 17532, "sample_start_year": 2010, "se_pp": 1.9773241588694725, "source": "Independent treatment-substitution replication", "utilities": 1108, "utility_state_units": 1196 }, { "estimate_pct": 2.1, "mapping": "proprietary_Aterio_plus_county_broadcast", "outcome": "residential", "sample_end_year": 2024, "sample_rows": 14733, "sample_start_year": 2010, "se_pp": 0.8, "source": "MPQW paper Table 2", "utilities": 973, "utility_state_units": 1048 }, { "estimate_pct": 1.236327204681246, "mapping": "paper_county_broadcast", "outcome": "residential", "sample_end_year": 2025, "sample_rows": 15966, "sample_start_year": 2010, "se_pp": 1.0667176623480288, "source": "Independent treatment-substitution replication", "utilities": 1006, "utility_state_units": 1084 }, { "estimate_pct": 3.3194113343851015, "mapping": "audited_serving_utility", "outcome": "residential", "sample_end_year": 2025, "sample_rows": 17532, "sample_start_year": 2010, "se_pp": 1.8333815153616677, "source": "Independent treatment-substitution replication", "utilities": 1108, "utility_state_units": 1196 }, { "estimate_pct": 2.8, "mapping": "proprietary_Aterio_plus_county_broadcast", "outcome": "commercial", "sample_end_year": 2024, "sample_rows": 14733, "sample_start_year": 2010, "se_pp": 0.9, "source": "MPQW paper Table 2", "utilities": 973, "utility_state_units": 1048 }, { "estimate_pct": -0.03523518016361047, "mapping": "paper_county_broadcast", "outcome": "commercial", "sample_end_year": 2025, "sample_rows": 15966, "sample_start_year": 2010, "se_pp": 1.3692063726844839, "source": "Independent treatment-substitution replication", "utilities": 1006, "utility_state_units": 1084 }, { "estimate_pct": 4.0851479045646055, "mapping": "audited_serving_utility", "outcome": "commercial", "sample_end_year": 2025, "sample_rows": 17532, "sample_start_year": 2010, "se_pp": 1.899143549860749, "source": "Independent treatment-substitution replication", "utilities": 1108, "utility_state_units": 1196 }, { "estimate_pct": 4.2, "mapping": "proprietary_Aterio_plus_county_broadcast", "outcome": "industrial", "sample_end_year": 2024, "sample_rows": 14733, "sample_start_year": 2010, "se_pp": 1.6, "source": "MPQW paper Table 2", "utilities": 973, "utility_state_units": 1048 }, { "estimate_pct": 1.9992189991389397, "mapping": "paper_county_broadcast", "outcome": "industrial", "sample_end_year": 2025, "sample_rows": 15966, "sample_start_year": 2010, "se_pp": 1.6566193642066258, "source": "Independent treatment-substitution replication", "utilities": 1006, "utility_state_units": 1084 }, { "estimate_pct": 5.693414414399651, "mapping": "audited_serving_utility", "outcome": "industrial", "sample_end_year": 2025, "sample_rows": 17532, "sample_start_year": 2010, "se_pp": 3.5547111760881616, "source": "Independent treatment-substitution replication", "utilities": 1108, "utility_state_units": 1196 } ], "threshold_results": [ { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 2.0148771752245156, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 0.08319605424497087, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 21.78246267474678, "mapping": "paper_county_broadcast", "never_treated_units": 945, "overall_att_pct": 0.7328636951375328, "overall_ci_high_pct": 2.848078240668108, "overall_ci_low_pct": -1.3823508503930424, "overall_p_value": 0.4970832953847556, "overall_se_pp": 1.0791910946584569, "post_treatment_rows": 944, "sample_rows": 15661, "threshold_mw": 2, "treated_units": 119, "utilities": 989, "utility_state_units": 1064, "year2_att_pct": 0.9793676175641638, "year2_ci_high_pct": 2.553166139593962, "year2_ci_low_pct": -0.5944309044656346, "year2_se_pp": 0.80295842960704 }, { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 1.7220180144353316, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 0.1263332716956098, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 20.122274769637908, "mapping": "paper_county_broadcast", "never_treated_units": 959, "overall_att_pct": 1.236327204681246, "overall_ci_high_pct": 3.327093822883383, "overall_ci_low_pct": -0.8544394135208905, "overall_p_value": 0.2464556370769312, "overall_se_pp": 1.0667176623480288, "post_treatment_rows": 1019, "sample_rows": 15966, "threshold_mw": 5, "treated_units": 125, "utilities": 1006, "utility_state_units": 1084, "year2_att_pct": 1.3178185485910667, "year2_ci_high_pct": 2.8451138036154524, "year2_ci_low_pct": -0.2094767064333189, "year2_se_pp": 0.7792322729716253 }, { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 1.6773972521084162, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 0.5138663749595201, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 13.161268021887043, "mapping": "paper_county_broadcast", "never_treated_units": 1024, "overall_att_pct": -0.7689870256448634, "overall_ci_high_pct": 1.7703320231274393, "overall_ci_low_pct": -3.3083060744171666, "overall_p_value": 0.5528126707487431, "overall_se_pp": 1.295570943251175, "post_treatment_rows": 822, "sample_rows": 16860, "threshold_mw": 20, "treated_units": 120, "utilities": 1062, "utility_state_units": 1144, "year2_att_pct": -0.2355250246337837, "year2_ci_high_pct": 1.6568247186222191, "year2_ci_low_pct": -2.1278747678897862, "year2_se_pp": 0.9654845628857156 }, { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 1.9076852484273443, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 0.046320586206464294, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 23.961080408610446, "mapping": "paper_county_broadcast", "never_treated_units": 1062, "overall_att_pct": -0.7465924512523114, "overall_ci_high_pct": 1.025286645383722, "overall_ci_low_pct": -2.518471547888345, "overall_p_value": 0.4088844311950487, "overall_se_pp": 0.9040199472632824, "post_treatment_rows": 630, "sample_rows": 17246, "threshold_mw": 50, "treated_units": 110, "utilities": 1088, "utility_state_units": 1172, "year2_att_pct": 0.3598385609720989, "year2_ci_high_pct": 2.1563254477032485, "year2_ci_low_pct": -1.436648325759051, "year2_se_pp": 0.9165749422097703 }, { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 1.8619891609112074, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 0.0028947954628628708, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 32.98561593724581, "mapping": "paper_county_broadcast", "never_treated_units": 1115, "overall_att_pct": 2.3255985835336213, "overall_ci_high_pct": 4.430052314264069, "overall_ci_low_pct": 0.22114485280317375, "overall_p_value": 0.03031385799406562, "overall_se_pp": 1.0737008830257386, "post_treatment_rows": 386, "sample_rows": 17647, "threshold_mw": 150, "treated_units": 87, "utilities": 1115, "utility_state_units": 1202, "year2_att_pct": 2.4476025455154824, "year2_ci_high_pct": 5.345091620093627, "year2_ci_low_pct": -0.44988652906266313, "year2_se_pp": 1.4783107523357886 }, { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 1.8964419282811373, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 0.01713851224413753, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 27.387166979498975, "mapping": "audited_serving_utility", "never_treated_units": 1156, "overall_att_pct": 2.7813209880167213, "overall_ci_high_pct": 6.158000200093417, "overall_ci_low_pct": -0.5953582240599737, "overall_p_value": 0.10643567070236437, "overall_se_pp": 1.7227955163656605, "post_treatment_rows": 247, "sample_rows": 17464, "threshold_mw": 2, "treated_units": 35, "utilities": 1105, "utility_state_units": 1191, "year2_att_pct": 2.301426314908376, "year2_ci_high_pct": 5.519184916316895, "year2_ci_low_pct": -0.9163322865001433, "year2_se_pp": 1.6417135721472038 }, { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 2.330365774648688, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 0.17663575222681632, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 18.70238632158988, "mapping": "audited_serving_utility", "never_treated_units": 1161, "overall_att_pct": 3.3194113343851015, "overall_ci_high_pct": 6.91283910449397, "overall_ci_low_pct": -0.2740164357237673, "overall_p_value": 0.07021202246820163, "overall_se_pp": 1.8333815153616677, "post_treatment_rows": 240, "sample_rows": 17532, "threshold_mw": 5, "treated_units": 35, "utilities": 1108, "utility_state_units": 1196, "year2_att_pct": 1.2881964146687102, "year2_ci_high_pct": 4.306199614078785, "year2_ci_low_pct": -1.7298067847413636, "year2_se_pp": 1.5397975507194255 }, { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 1.9414226570029118, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 0.16485179039134237, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 19.002551054592107, "mapping": "audited_serving_utility", "never_treated_units": 1173, "overall_att_pct": 1.2742723549383117, "overall_ci_high_pct": 4.485120912791107, "overall_ci_low_pct": -1.9365762029144837, "overall_p_value": 0.4366546445047462, "overall_se_pp": 1.638188039720814, "post_treatment_rows": 208, "sample_rows": 17704, "threshold_mw": 20, "treated_units": 34, "utilities": 1118, "utility_state_units": 1207, "year2_att_pct": 0.865524353738139, "year2_ci_high_pct": 4.13837737159978, "year2_ci_low_pct": -2.4073286641235017, "year2_se_pp": 1.6698229682967551 }, { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 1.9514207281283846, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 4.5105732364671995e-05, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 44.73084700073008, "mapping": "audited_serving_utility", "never_treated_units": 1186, "overall_att_pct": -1.1894168417539701, "overall_ci_high_pct": 3.205684943977839, "overall_ci_low_pct": -5.58451862748578, "overall_p_value": 0.5958196392755384, "overall_se_pp": 2.2423988702713316, "post_treatment_rows": 152, "sample_rows": 17766, "threshold_mw": 50, "treated_units": 25, "utilities": 1122, "utility_state_units": 1211, "year2_att_pct": 0.26294545359286625, "year2_ci_high_pct": 3.579675469235555, "year2_ci_low_pct": -3.053784562049822, "year2_se_pp": 1.6922091916544326 }, { "joint_pretrend_effective_rank": 14, "joint_pretrend_estimable": true, "joint_pretrend_max_abs_t": 3.0434397966931366, "joint_pretrend_method": "joint Wald test of aggregated event-time leads (t<-1); influence-function covariance; df=effective rank", "joint_pretrend_p_value": 4.7908047546761063e-23, "joint_pretrend_restrictions": 14, "joint_pretrend_wald": 140.85752322004302, "mapping": "audited_serving_utility", "never_treated_units": 1202, "overall_att_pct": 3.9873130357328606, "overall_ci_high_pct": 10.24150423377435, "overall_ci_low_pct": -2.266878162308629, "overall_p_value": 0.21145176734234772, "overall_se_pp": 3.1909138765517806, "post_treatment_rows": 43, "sample_rows": 17782, "threshold_mw": 150, "treated_units": 10, "utilities": 1123, "utility_state_units": 1212, "year2_att_pct": 10.832223092504897, "year2_ci_high_pct": 21.753537558386952, "year2_ci_low_pct": -0.08909137337715922, "year2_se_pp": 5.572099217286763 } ] }