Are the interconnection withdrawal dates published for the non-ISO West representative of the region?
Berkeley Lab's Queued Up reports a median duration from interconnection request to withdrawal for the West. That figure rests on the 12.1% of western withdrawals that carry a recorded withdrawal date. Of the 31 western transmission providers with withdrawals in the file, six record any withdrawal date and five record one for more than a quarter of their withdrawals; the rest — including BPA, Idaho Power, APS, NorthWestern, NV Energy and WAPA — record none at all. PacifiCorp, whose 1,627 withdrawals are 2.7 times those of the next-largest western provider, records a date for 6.0% of them — which is why the count reads six or five depending on where the line is drawn.
This repository tests whether the reporting subset resembles the region it is used to represent, and estimates how far the published median is displaced as a result.
The reporting subset is compositionally unlike the region. Technology mix, queue cohort and interconnection service type all differ from the regional population at p < 10⁻¹⁰ after Holm correction, on both defensible definitions of "the reporting subset." Project capacity does not differ. The dated records over-represent stand-alone solar (44.8% against 31.5%) and wind (30.7% against 23.1%), and under-represent solar-plus-storage and stand-alone storage.
But that barely moves the median. Standardising to the region's actual technology-by-cohort composition shifts the median from 1.83 to 1.77 years — a shift of −0.06 years, 95% CI [−0.23, +0.00]. Under a truncation-matched window the shift is +0.19 years. The sign flips, so the direction of the bias is not identified. A validation exercise inside the well-covered ISO regions explains why: imposing the West's compositional distortion on a population whose durations are fully observed shifts the median by about eight days. Technology, cohort and capacity together explain 4.6% of the variance in withdrawal duration.
The binding caveat is coverage, not composition. A logistic regression over the 7,991 western requests with complete predictors — which needs no dates — finds solar the most withdrawal-prone technology, larger projects more likely to be withdrawn, and 26% of the latent variation in withdrawal risk sitting between providers rather than between projects. Whether the five reporting providers withdraw on a different schedule from the twenty-six that publish nothing cannot be determined from this data file, and that, not project mix, is the open question.
The full write-up is in PAPER.md.
data/LBNL_Ix_Queue_Data_File_thru2025.xlsx — the Queued Up 2026 Edition data
file, covering interconnection requests through the end of 2025. Sheet
03. Complete Queue Data holds the project-level records.
Source: https://emp.lbl.gov/queues
The Queued Up data file is licensed CC BY 4.0. Attribution: Lawrence Berkeley National Laboratory and GridTracker.
Requires R (tested on 4.6.1) with readxl, dplyr, tidyr, ggplot2 and
lme4. Open queue-analysis.Rproj in RStudio and run the scripts in order.
| Script | Produces |
|---|---|
00-setup.R |
shared definitions; sourced by everything below |
01-explore.R |
the original exploratory session, kept as a record |
02-figures.R |
figures/01–03 |
03-differences.R |
the roster and the five comparisons |
04-standardise.R |
the ISO calibration, validation and bias estimate |
05-logistic.R |
the withdrawal model |
06-figures.R |
figures/04–05 |
04-standardise.R runs 2,000 bootstrap replicates for each of eight estimates
and takes several minutes. It and 05-logistic.R set a seed (20260906), so
results are exactly reproducible. Result tables are written to results/.
See CITATION.cff.
Analysis code and write-up: CC BY 4.0, matching the licence of the underlying data file.