Transit Data Quality Under Reconciliation

The article's initial statewide analysis classified 49% of California census tracts as mobility deserts under its grocery-proximity and stop-proximity thresholds.¹ That estimate was withheld as a verified result when the initial analysis was conducted. It was disclosed retrospectively on this page as the baseline in the article-reported 49%-to-12% comparison.

A later article run reported 12%. That figure is now under reconciliation rather than presented as a verified correction.²

The article's explanation attributes the fourfold difference to incomplete transit data. It compares General Transit Feed Specification (GTFS) feeds from 8 major transit agencies, reported as 24,421 stops, with a Cal-ITP extract previously described here as 64,060 unique stops. The pinned public artifact uses different raw and unique labels, so that comparison is not yet verified.

This experience offers a case study in how data completeness affects research conclusions. If we're going to make claims about food access, we need to understand how sensitive those claims are to data quality. Here's what we learned.

---

The Original Finding

Using GTFS data from 8 agencies (VTA, AC Transit, Muni SF, BART, LA Metro, SacRT, OCTA, San Diego MTS), we classified every residential census tract in California:

The Original Finding

Classification	Tracts	Percentage
Full Access	2,800	31.0%
Traditional Food Desert	1,810	20.0%
Mobility Desert	4,429	49.0%

The unreproduced original table classified 4,429 tracts as mobility deserts. The current record does not establish the affected households or their travel barriers.

The classification logic is documented, but the processed inputs and outputs are not committed in a form that verifies these calculations. The suspected problem is upstream transit coverage; its exact scale remains to be established.

---

What Raised the Red Flag

Two observations suggested the 49% figure needed scrutiny:

1. Partial LA Metro coverage

The article reports that its LA Metro input covered only part of the service it intended to include. No committed feed inventory currently shows which routes or stops were present, so the missing share and resulting tract errors are under reconciliation.

2. Missing agencies entirely

The article names the following potential agency omissions to verify:

• No Golden Gate Transit (serving Marin and San Francisco counties)
• No SamTrans (San Mateo County)
• No Foothill Transit (San Gabriel Valley)
• Dozens of smaller municipal systems

Each named omission is an input-inventory item to verify. No saved feed crosswalk currently derives the affected tracts.

---

The Article-Reported Validation Record

To investigate the coverage gap, the article describes acquiring a statewide GTFS aggregation from Cal-ITP.³

The Validation Process

Data Source	Transit Stops	Agencies
Original 8-agency sample	24,421	8
Cal-ITP count reported as raw in the article	143,203, under reconciliation	200+
Cal-ITP count reported as deduplicated in the article	64,060, under reconciliation	200+

The article reported 64,060 unique transit stops after filtering and deduplication. The pinned artifact instead labels 64,060 as raw and reports roughly 24,000 unique records. Until that pipeline is reconciled, neither the deduplicated count nor the 2.6-times comparison should be treated as verified.

---

Article-Reported Comparison Under Reconciliation

The article reports the following comparison from two runs. A committed processed output does not currently reproduce it:

Article-reported results under reconciliation

Classification	Original	Corrected	Change
Full Access	31.0%	60.2%	+29.2 pp
Traditional Food Desert	20.0%	27.8%	+7.8 pp
Mobility Desert	49.0%	12.0%	-37.0 pp

The article table shows the mobility desert rate changing from 49% to 12%. Because the stop populations conflict across public materials, the table records an article claim rather than a verified rerun.

---

Where the Errors Concentrated

The article proposes the following candidate mechanisms and locations. No reconciled output currently verifies where misclassification concentrated:

Municipal transit systems: Glendale, Pasadena, Long Beach, and Santa Monica are article-reported examples to check against the original feed inventory.

Multi-agency corridors: The article identifies the East Bay I-680 corridor as a candidate case where a partial agency set could omit service. The magnitude is not verified.

Regional authorities: The article names Monterey, Santa Barbara, and San Luis Obispo as counties to check for omitted agencies. The public record does not verify which tracts were misclassified or what service they had.

---

The Broader Lesson: Data Completeness Matters

The public record supports a coverage warning, but it does not yet isolate the discrepancy to one cause. The classification logic, spatial calculations, raw stop inventory, deduplication rule, and saved output all need to be reconciled together.

Several practices could have prevented this:

1. Know Your Coverage

Before analysis, check what proportion of the relevant infrastructure your data represents. For transit:

• How many agencies operate in the study area?
• What percentage of stops does your data capture?
• Are there known gaps in common data sources?

2. Use Aggregated Data Sources When Available

The Cal-ITP GTFS-Ingest Pipeline Dataset aggregates California feeds for statewide analysis. Its primary dataset page cautions that feeds may not have passed General Transit Feed Specification validation, so researchers still need a dated feed inventory and validation record. The Federal Transit Administration's National Transit Database provides agency inventories. Transitland aggregates GTFS feeds worldwide.⁴

3. Validate Against External Benchmarks

If your findings seem surprising, check against known quantities:

• Total transit stops should match published fleet/infrastructure numbers
• Regional breakdowns should align with agency service areas
• Extreme outliers warrant investigation

4. Report Data Limitations

Even with best efforts, some incompleteness remains. Explicitly stating what data sources were used, what coverage they represent, and what gaps might exist allows readers to assess findings appropriately.

---

What the Article-Reported 12% Would Mean

If the 12% mobility desert rate survives reconciliation, the article's interpretation would identify:

• An article-reported set of 1,086 tracts for follow-up validation
• A comparison between the project's transit threshold and the federal distance metric
• Candidate locations for testing transit and retail interventions

A verified 12% rate would describe a smaller classified population than 49%. Neither rate by itself establishes a household travel barrier.

The policy interpretation is therefore provisional. Targeting specific transit gaps is warranted only if the reconciled inputs reproduce the lower rate and its geographic distribution.

---

Validation Checklist for Transit Analysis

Before finalizing transit accessibility research:

• ☐ List all transit agencies operating in study area
• ☐ Verify GTFS data includes each agency
• ☐ Check stop counts against published agency statistics
• ☐ Use aggregated data sources (Cal-ITP, Transitland) when available
• ☐ Flag and investigate areas with suspiciously low stop density
• ☐ Document data sources, coverage, and known gaps
• ☐ Test sensitivity: how do results change with different data sources?

---

Data and Methods

Original analysis:

• GTFS feeds from 8 agencies: VTA, AC Transit, Muni SF, BART, LA Metro, SacRT, OCTA, San Diego MTS
• 24,421 transit stops
• Downloaded individually from agency websites

Article-reported later analysis, under reconciliation:

• Cal-ITP GTFS-Ingest Pipeline Dataset; the source page says feeds may not have passed GTFS validation
• The article labels 143,203 as raw and 64,060 as unique after deduplication
• The pinned artifact instead labels 64,060 as raw and reports roughly 24,000 unique records

Classification unchanged:

• Mobility desert = grocery within 1 mile, transit stop > 0.5 miles OR < 2 stops within 0.5 miles
• Same thresholds, same spatial calculations, same methodology

---

Notes

1. The initial November 2025 analysis used individually downloaded GTFS feeds from 8 California transit agencies. The 49% estimate was not released as a verified result at that time; this page later disclosed it retrospectively as the baseline in the comparison. ↩
2. The later November 2025 analysis used Cal-ITP GTFS data. Its stop population and resulting tract classifications remain under reconciliation. ↩
3. California Integrated Travel Project (Cal-ITP). Cal-ITP GTFS-Ingest Pipeline Dataset. The source describes statewide analytical tables and cautions that feeds may not have passed GTFS validation. ↩
4. Transitland aggregates GTFS feeds. The Federal Transit Administration's National Transit Database provides official agency and operating-data products. Neither source establishes which feeds entered this article's historical extract without a dated crosswalk. ↩

---

Tags: #FoodSecurity #DataQuality #TransitData #GTFS #CalITP #Validation #Methods #Research

---

Next in this series: Mobility Deserts: Transit Proximity Under Review.

Frequently Asked Questions

How did incomplete transit data affect mobility desert estimates?

The article reports a change from 49% with an 8-agency input to 12% with a Cal-ITP input. Conflicting raw and unique stop counts mean the comparison is under reconciliation.

What was wrong with the original transit data?

The article reports that its original input covered only part of the intended LA Metro service and omitted named agencies. No committed feed inventory establishes the missing share or resulting tract errors.

How many transit agencies actually operate in California?

The number relevant to this article is unknown because no frozen California operator roster or dated crosswalk connects operators to the feeds in the analyzed extract. The National Transit Database provides an agency inventory, but it does not establish the article's historical inclusion count without that crosswalk.

What data source should be used for California transit accessibility research?

The Cal-ITP GTFS-Ingest Pipeline Dataset provides statewide analytical tables. Its primary page cautions that feeds may not have passed GTFS validation, so researchers still need to document feed coverage, coordinate filtering, validation, and deduplication before calling an extract complete.

Where did the misclassification errors concentrate?

The article identifies suburban municipal systems, multi-agency corridors, and rural regional authorities as candidate mechanisms. That geographic concentration is not derived by a committed output and remains article-reported rather than independently verified.