The SNAP Participation Gap During the COVID Era

The cited COVID Money Tracker covers trillions of dollars in legislative, administrative, and Federal Reserve responses across public health, state and local governments, businesses, credit markets, and households. Household programs included unemployment supplements, economic impact payments, and nutrition assistance, but they were only one part of the total response. ^[1]

Santa Clara County joined five other Bay Area counties and the City of Berkeley in coordinated stay-at-home orders announced on March 16, 2020. ^[2] The article uses the county as a case for examining SNAP participation estimates during the pandemic era. The data do not evaluate whether federal relief worked.

The article lists a 0.09 percentage-point decrease between the ACS vintages labeled 2019 and 2020. A lower SNAP participation estimate does not by itself indicate improved food security.

Examining the 408 census tracts individually, we learn a little more:

Vulnerability Quintile	Difference Between Published Vintages	Estimate Direction
Q1 (Lowest)	-0.29 pp	Lower estimate
Q2	-0.14 pp	Lower estimate
Q3	-0.04 pp	Similar estimate
Q4	-0.63 pp	Lower estimate
Q5 (Highest), "high-vulnerability tracts"	+1.01 pp from the listed rates	Higher estimate

I'll refer to Q5 as "high-vulnerability tracts" throughout this post; these are the 82 census tracts (20% of the county) with the highest concentrations of poverty, limited education, and other risk factors.

The article reports different changes in SNAP participation across vulnerability groups.

It reports a gap changing from 9.49 percentage points in 2019 to 14.16 in 2023, summarized as a 49% increase. The values are not publicly reproduced.

The series does not identify whether relief, need, eligibility, outreach, or take-up produced the pattern.

This post preserves an article-reported SNAP participation comparison for reconciliation. It does not measure food insecurity or establish a causal COVID effect.

---

What We Already Know

The COVID-19 pandemic created unprecedented economic disruption. National food insecurity initially spiked, but substantial federal intervention (enhanced unemployment benefits, stimulus payments, and expanded SNAP) appeared to stabilize the situation. National food insecurity rates remained relatively flat at 10.5% through 2020. ^[4]

Feeding America's linked pandemic page presents modeled county projections rather than observed county outcomes. ^[5] This article does not use those projections as evidence of measured increases or decreases.

What's missing from the literature:

The literature lacks granular temporal analysis showing:

• Which specific neighborhoods experienced increases versus decreases
• Whether vulnerability groups show different article-reported SNAP participation patterns
• Whether the gap has closed in the years since

Most COVID food security research relies on survey data, which provides valuable national and state-level insights but typically lacks the geographic precision needed for neighborhood-level analysis. ^[3] ACS five-year tract estimates add geographic detail, but each vintage pools 60 months and adjacent vintages overlap. They support comparisons among successive period estimates, not annual neighborhood trajectories or acute-shock timing.

Here's what we can learn now:

The article describes a tract-level SNAP participation analysis for Santa Clara County:

• Analysis file described by the article: 408 tracts × 5 successive ACS five-year vintages labeled 2019-2023; the 2017-2018 vintages appear only in the earlier-period description
• ACS 5-year SNAP participation rates (smoothed estimates)
• Difference-in-differences specification comparing high- versus low-vulnerability tracts; its published coefficients are withdrawn pending a saved output
• Earlier-vintage description (2017-2019), which cannot validate parallel trends with overlapping ACS period estimates

The article-reported pattern is not publicly reproduced. It cannot establish worsening food insecurity, recovery, or a causal pandemic effect.

---

Methodology

Study Area: Santa Clara County, California (408 census tracts, 1.9M residents)

Data Sources:

• Census ACS 5-Year Estimates (2017-2023): SNAP participation rates
• Article-described 2019 vulnerability classification: The exact components and formula conflict with the current index page and remain under reconciliation
• Census TIGER/Line shapefiles (2020): Geographic boundaries

Analytical Approach:

1. File construction: the article reports 408 tracts × 5 successive ACS five-year vintages from 2019-2023 = 2,040 tract-vintage rows; a public output has not reproduced the file
2. Quintile classification: Tracts grouped by 2019 baseline vulnerability (Q1=lowest, Q5=highest)
3. Difference-in-differences specification: Compare Q5 with Q1-Q4 across vintages labeled before and after 2020; the reported coefficients are withdrawn pending a saved output
4. Article-defined change groups: tracts categorized by increases, small changes, or decreases in SNAP participation

Key Definitions:

• SNAP participation rate: Percentage of households receiving SNAP benefits (ACS table B22003)
• COVID-era labels: 2020-2023. Each ACS five-year vintage pools 60 months, and adjacent vintages share most of their underlying years.
• Treatment group: Q5 tracts (top 20% vulnerability, 82 tracts)
• Control group: Q1-Q4 tracts (bottom 80%, 326 tracts)

Primary Limitation: ACS five-year estimates smooth acute change and substantially overlap across adjacent vintages. Census Bureau comparison guidance advises against comparing overlapping five-year datasets. The displayed differences are preserved only as article-reported claims under reconciliation; they cannot recover the size or timing of a 2020 shock and should not be interpreted as annual effects.

Public materials: No matching replication code or saved output is currently posted.

---

Finding 1: The Aggregate Estimate Masks Group Differences

The article reports a lower countywide SNAP estimate and a higher estimate for its highest-vulnerability group. Neither direction establishes a change in food security.

Countywide Statistics for Vintages Labeled 2019 and 2020

• SNAP participation: 4.93% → 4.84%
• Change: -0.09 percentage points (-1.8%)
• Statistical test: t = -0.348, p = 0.73 (not statistically significant)
• Interpretation: No meaningful change at aggregate level

The article reports a small county-wide change in SNAP participation. That is not a direct food-security metric.

By Vulnerability Quintile

The aggregate estimate masks differences between the published group estimates:

By Vulnerability Quintile

Quintile	Mean Difference	Median Difference	% of Tracts with a Higher Estimate
Q1 (Lowest)	-0.29 pp	-0.01 pp	46.3%
Q2	-0.14 pp	-0.11 pp	39.2%
Q3	-0.04 pp	0.00 pp	48.8%
Q4	-0.63 pp	-0.49 pp	22.5%
Q5 (Highest)	+1.01 pp from the listed rates	+0.22 pp	62.5%

Only the highest-vulnerability group has a positive mean difference in the displayed table. It also has the largest displayed share of tracts with a higher estimate. These are comparisons between overlapping period estimates, not annual changes in household hardship.

What High-Vulnerability Tracts Look Like

These tracts are concentrated in:

• East San Jose (particularly Alum Rock area)
• North San Jose industrial zones
• Parts of Gilroy in the south county

They share characteristics:

• 23% average poverty rate (versus 8.5% county-wide)
• 67% Hispanic population (versus 26% county-wide)
• 31% without high school diploma (versus 12% county-wide)
• Higher concentrations of service-sector employment
• More renters, more crowded housing, less remote work capability

The article reports different SNAP participation changes by tract group; it does not establish changes in household hardship.

Why This Matters for Policy

A county-wide SNAP participation rate cannot by itself show whether policy worked. Group comparisons can motivate follow-up tests of need, eligibility, outreach, and take-up.

Here, "high-vulnerability" is an article-assigned tract label. The components named in this article do not match those on the current Neighborhood Vulnerability Index page, and no public formula or assignment file reconciles them. The label should not be treated as a current EVI classification.

The article reports 112,000 residents in its high-vulnerability tract group. The SNAP series cannot determine which of the following mechanisms, if any, explain the reported change:

• Changes in program reach
• Changes in need or eligibility
• Changes in outreach or take-up

The article offers income and housing examples as hypotheses. Its tract-level SNAP participation estimates do not identify relief adequacy or distributional effects.

---

Finding 2: Article-Reported Gaps Across Successive Vintages

The article reports different SNAP participation estimates across overlapping vintages; causal timing is unresolved.

Published Estimates for Vintages Labeled 2019 and 2023

Published Estimates for Vintages Labeled 2019 and 2023

Quintile	2019 Rate	2023 Rate	Change	% Change
Q1 (Lowest)	1.52%	1.40%	-0.12 pp	-7.9%
Q2	2.57%	2.83%	+0.26 pp	+10.1%
Q3	3.45%	4.49%	+1.04 pp	+30.1%
Q4	5.30%	7.63%	+2.33 pp	+44.0%
Q5 (Highest)	11.01%	15.56%	+4.55 pp	+41.3%

The Q5-Q1 gap:

• 2019: 9.49 percentage points (Q5 rate 7.24× higher than Q1)
• 2023: 14.16 percentage points (Q5 rate 11.11× higher than Q1)
• Difference between the two published gaps: +4.67 percentage points, reported as 49.2% of the 2019 gap

Differences Between Successive ACS Five-Year Vintages

The article lists the following differences between overlapping countywide period estimates:

Differences Between Successive ACS Five-Year Vintages

Period	Change	Description
2019 → 2020	-1.8%	Difference between published period estimates
2020 → 2021	+3.7%	Difference between published period estimates
2021 → 2022	+11.8%	Difference between published period estimates
2022 → 2023	+13.9%	Difference between published period estimates

The article reports larger differences between successive ACS vintages. Because adjacent estimates share most of their underlying years, those values are not annual growth rates. They can reflect need, eligibility, outreach, or take-up and do not measure economic recovery.

Possible explanations:

1. Catch-up enrollment: Households eligible since 2020-2021 finally navigating enrollment processes
2. Policy expansions: SNAP eligibility expansions and outreach implemented during pandemic
3. Delayed economic impacts: Job losses and income reductions in vulnerable populations manifesting with lag
4. Reduced stigma: Enhanced visibility of SNAP during pandemic normalized participation

High-Vulnerability Group Across Successive Vintages

The article reports the following SNAP participation estimates by ACS vintage label:

• 2019: 11.01%
• 2020: 12.02% (+1.01 percentage points from the listed 2019 rate)
• 2021: 12.73%
• 2022: 14.02%
• 2023: 15.56% (+4.55 points above the estimate labeled 2019)

The article lists a higher estimate in each successive overlapping vintage. The series does not establish annual growth or continuous deterioration in food security.

---

Finding 3: Regression Values Withdrawn

No regression coefficient is reported from the current public record. The previously published difference-in-differences and event-study values do not reconcile with the displayed group means, and no saved model output is public.

Specification Described by the Article

The article describes a comparison between its Q5 group and Q1-Q4 group using vintages labeled 2019 and 2020, followed by an event-study specification across later labels. That design description remains a proposed analysis, not a reproduced result.

Adjacent ACS five-year estimates share most of their underlying years. Even a reproduced coefficient from this specification would compare overlapping period estimates rather than isolate an acute March 2020 treatment.

The regression table is withdrawn until a saved output identifies the estimation sample, weighting rule, formula, standard-error method, and coefficient path.

What This Means

The descriptive tables report different SNAP participation estimates across tract groups. They do not establish a structural shift, worsening hardship, or failed recovery.

---

Mechanisms a Reproduced Pattern Would Need to Separate

The article proposes labor-market exposure, housing costs, eligibility, outreach, and take-up as possible explanations for differences in SNAP participation estimates. Its numerical support for those mechanisms is not publicly reproduced, and the SNAP series cannot distinguish among them.

A matched rerun would first need to reproduce the tract groups and period estimates. Follow-up data could then test whether changes in need, eligibility, outreach, take-up, employment, or housing costs account for the pattern.

---

Spatial Classification Under Reconciliation

The earlier category table listed 160 tracts with a higher estimate, 107 with a similar estimate, 67 with a lower estimate, and 112 with insufficient data. Those counts total 446, which does not match the page's stated 408-tract study population. The exact classification and percentages are therefore withdrawn until a matched output defines the tract universe and missing-data rule. A higher SNAP participation estimate would still describe program use, not deterioration in food security.

Geographic Concentration

The article reports tracts with higher estimates in:

• East San Jose: Alum Rock, East Foothills, Evergreen
• North San Jose: Industrial and light commercial zones
• South County: Gilroy and unincorporated areas

The article reports tracts with lower estimates in:

• Tech corridor: Cupertino, Sunnyvale, Mountain View
• West Valley: Los Gatos, Saratoga, Campbell (wealthier areas)

If the article-reported geography survives reconciliation, it could motivate testing place-based outreach. The current analysis does not establish pandemic-driven food insecurity or an efficient intervention target.

---

Summing Up

The "Too Early to Say" Questions

Causal mechanism:

• Job loss versus housing burden versus enrollment effects?
• Which factor explains most of the divergence?
• Would require household-level panel data to separate

Successive-vintage pattern:

• Does the group difference remain after reconstructing non-overlapping periods?
• How much of the pattern reflects changes in need, eligibility, outreach, or take-up?
• Later vintages still require the same period-estimate caution

Generalizability:

• Is Santa Clara unique, or does this pattern hold in other affluent counties?
• Do tech-hub economies show similar differential impacts?

Claims Preserved for Reconciliation

What the article reports:

• The listed high-vulnerability SNAP estimates are 11.01% and 12.02% in the vintages labeled 2019 and 2020; the direct difference is +1.01 percentage points and remains under reconciliation
• The published regression values are withdrawn because they do not reconcile with the displayed group means
• The event-study specification is described but has no saved output and cannot establish parallel trends from overlapping ACS vintages
• The article lists a larger group gap across successive overlapping vintages labeled 2019 through 2023

Robustness claims to reproduce:

• Holds across alternative specifications (tract fixed effects, year fixed effects)
• Robust to excluding 2020 (potential measurement noise)
• Consistent across demographic subgroups within high-vulnerability tracts

Reported scale under review:

• Article-reported group: 82 tracts and 112,000 residents
• Article-reported gaps of 9.49 and 14.16 percentage points in vintages labeled 2019 and 2023
• A +4.55 percentage-point difference between the displayed high-vulnerability estimates labeled 2019 and 2023

---

Data described in the article: Census ACS 5-Year Estimates 2017-2023 and Census TIGER/Line Shapefiles 2020. Analysis conducted October 2025. No matching replication package or saved output is currently public.

---

References

[1] Committee for a Responsible Federal Budget. "COVID Money Tracker." 2023. https://www.covidmoneytracker.org/ ↩

[2] San Francisco Department of Public Health. Annual Report, Fiscal Year 2019-2020. The report documents coordinated March 16 orders by San Francisco, five other Bay Area counties, and the City of Berkeley. https://media.api.sf.gov/documents/Full_Report_FY1920-_FINAL-7.14.21.pdf ↩

[3] Ziliak, J.P. "Food Hardship during the COVID-19 Pandemic and Great Recession." Applied Economic Perspectives and Policy, 43(1), 132-152, 2021. https://doi.org/10.1002/aepp.13099 ↩

[4] Coleman-Jensen, A., Rabbitt, M.P., Gregory, C.A., & Singh, A. "Household Food Security in the United States in 2020." USDA Economic Research Service, Report No. 298, September 2021. https://www.ers.usda.gov/publications/pub-details/?pubid=102075 ↩

[5] Feeding America. "The Impact of the Coronavirus on Food Insecurity in 2020 & 2021." Archived county-level projection page, March 2021. https://www.feedingamerica.org/research/coronavirus-hunger-research ↩

Frequently Asked Questions

What does the article report about SNAP participation in high-vulnerability neighborhoods?

The listed high-vulnerability rates move from 11.01% to 12.02%, a direct difference of +1.01 percentage points, while the article reports a -0.09 point county change. These values are not publicly reproduced, and SNAP participation is not a direct measure of food insecurity.

How much does the article report the SNAP participation gap changed?

The article reports a change from 9.49 percentage points in 2019 to 14.16 in 2023. That 49% calculation is under reconciliation and does not identify a causal COVID effect.

What can the article's difference-in-differences model establish?

The previously reported regression values are withdrawn because they do not reconcile with the displayed group means and no saved model output is public. Overlapping ACS five-year vintages also prevent the model from isolating an acute COVID effect.

How should the successive ACS vintages be interpreted?

Each ACS five-year estimate pools 60 months of data and substantially overlaps adjacent vintages. Differences between vintages can reflect need, eligibility, outreach, or take-up, but they are not annual growth rates and do not identify an acute pandemic change.