How Place and Poverty IntersectGeographic Barriers and Low SNAP Take-upMarianne Bitler, UC Davis & NBERJason Cook, University of UtahSonya R. Porter, US Census Bureau1 / 51

Disclaimer

Any opinions and conclusions expressed herein are those of the authors and do not reflect the views of the U.S. Census Bureau. The statistical summaries reported in this paper have been cleared by the Census Bureau's Disclosure Review Board release authorization number CBDRB-FY21-CES014- 049. All results have been reviewed to ensure that no confidential information is disclosed.

2 / 51

Motivation

US somewhat unique within advanced economies in design of safety net. System very federalized with various levels of government running programs.
People apply to many programs and rules not harmonized and application is not automatic as in much of the developed world $\Rightarrow$ low take-up is more of issue in US than other places (Currie, 2006; Currie and Gahvari, 2008).

3 / 51

Motivation

US somewhat unique within advanced economies in design of safety net. System very federalized with various levels of government running programs.
People apply to many programs and rules not harmonized and application is not automatic as in much of the developed world $\Rightarrow$ low take-up is more of issue in US than other places (Currie, 2006; Currie and Gahvari, 2008).
Social program take-up: Hot topic in public econ, yet understudied.
- Emphasis on understanding role of barriers to accessing safety net.
Barriers impact both take-up (how many people enroll) and targeting (what types of people enroll).

3 / 51

Competing Models to Explain Low Take-Up

Neoclassical model

People make decisions balancing costs and benefits in a utility maximizing framework.
Incomplete take-up is a function of barriers:
- Incomplete information: Program existence, eligibility rules.
- Stigma: Either concerns about whether you should participate or about how others will judge you.
- Transaction costs: E.g., travel costs, difficultly with documentation, time costs.

4 / 51

Competing Models to Explain Low Take-Up

Neoclassical model

People make decisions balancing costs and benefits in a utility maximizing framework.
Incomplete take-up is a function of barriers:
- Incomplete information: Program existence, eligibility rules.
- Stigma: Either concerns about whether you should participate or about how others will judge you.
- Transaction costs: E.g., travel costs, difficultly with documentation, time costs.

Under this framework, barriers could be efficient if improve targeting, i.e., deter those without need or target those in need (Ackerlof, 1978; Nichols & Zeckhauser, 1982; Besley & Coate, 1992; Kleven & Kopczuk, 2011).

4 / 51

Competing Models to Explain Low Take-Up

Behavioral science model

Limited bandwidth or stress impairs people. Suggests incomplete take-up might be harmful or inefficient (Bertrand, Mullainathan, & Shafir, 2004).

Administrative Burden

Government actors may intentionally create "administrative burden" such as learning costs, compliance costs, and psychological costs to limit use, possibly due to limited capacity (Herd and Moynihan, 2019).

Under these models, barriers could be inefficient if worsen targeting, e.g., if the poor face limited bandwidth or scarcity, more needy individuals could be deterred by hassle costs.^*

^* Though take up and targeting alone are not enough to assess normative implications (Finkelstein & Notowidigdo, 2019).

5 / 51

Literature5 / 51

Finkelstein and Notowidigdo (2019)

Model for social welfare incorporates social costs and perceptions of costs of application (looking at targeting is not enough).
Fit parameters with RCT on Medicaid participants not on SNAP.
Did information and application help affect take-up and targeting?

6 / 51

Finkelstein and Notowidigdo (2019)

Model for social welfare incorporates social costs and perceptions of costs of application (looking at targeting is not enough).
Fit parameters with RCT on Medicaid participants not on SNAP.
Did information and application help affect take-up and targeting?

Assistance $>$ information $>$ status quo; Compliers are better off.
Worse targeting for all interventions.

6 / 51

Finkelstein and Notowidigdo (2019)

Model for social welfare incorporates social costs and perceptions of costs of application (looking at targeting is not enough).
Fit parameters with RCT on Medicaid participants not on SNAP.
Did information and application help affect take-up and targeting?

Assistance $>$ information $>$ status quo; Compliers are better off.
Worse targeting for all interventions.
Quantify costs of assessing eligibility?
- Understanding expense of eligibility assessment would help determine size of wedge between social and private welfare.
- Efforts to cut cost of eligibility would help shrink wedge.
- SNAP average administrative costs low relative to benefit amount.

6 / 51

Literature

Remaining literature can be conceptualized as studying how different types of barriers impact take-up and targeting.

7 / 51

Barriers Impact Take-Up

Information Barriers

SNAP-eligible people often don't realize eligibility (Bartlett et al., 2004); providing information increases take-up (Daponte, Sanders, & Taylor, 1999; Finkelstein and Notowidigdo, 2019).
Informational interventions matter in some other contexts as well: EITC (Bhargava and Manoli, 2015) and SSDI (Armour, 2018), but not others: FAFSA (Bettinger et al., 2012).

8 / 51

Barriers Impact Take-Up

Transaction Costs in SNAP

Reducing transaction costs in SNAP increases take-up via: application help (Schanzenbach, 2001; Finkelstein and Notowidigdo, 2019), certification periods (Kabbani and Wilde, 2003), certification reporting requirements (Gray, 2018; Hanratty, 2006; Unrath, 2021).
Less time for SNAP recertification interviews leads to more churning (Homonoff and Somerville, 2021)
Switching to automated SNAP application process decreases take-up (Wu, 2021)

9 / 51

Barriers Impact Take-Up

Transaction Costs in Other Programs

Learning costs with WIC, relative to SNAP, influenced by what stores carry (Barnes, 2021).
WIC participation during COVID affected by pick up and enrollment rules (in person/not) (Barnes and Petry, 2021; Whaley and Anderson, 2021; Vasan, Kenyon, and Roberto, 2021).
Access to program offices matters: SSA offices (Deshpande and Li, 2019) and WIC program offices/vendors (Rossin-Slater, 2013; Ambrozek, 2021).

10 / 51

Barriers Impact Targeting

Information Barriers

Complexity worsens targeting of low-income cases for EITC (Bhargava and Manoli, 2015).
Information mailers induce less-needy households to enroll in SNAP (Finkelstein and Notowidigdo, 2019).

11 / 51

Barriers Impact Targeting

Transaction Costs

Closing SSA offices had mixed impacts on who is deterred (Deshpande and Li, 2019).
SNAP application assistance reduces targeting across all dimensions (Finkelstein and Notowidigdo, 2019).
Automated application system reduces take-up, but it improves targeting efficiency for new recipients and worsens among recertifiers (Wu, 2021).
Recertification rules reduce take-up and lower retention, but improve targeting (Unrath, 2021).

12 / 51

Barriers Impact Targeting

Transaction Costs

Closing SSA offices had mixed impacts on who is deterred (Deshpande and Li, 2019).
SNAP application assistance reduces targeting across all dimensions (Finkelstein and Notowidigdo, 2019).
Automated application system reduces take-up, but it improves targeting efficiency for new recipients and worsens among recertifiers (Wu, 2021).
Recertification rules reduce take-up and lower retention, but improve targeting (Unrath, 2021).

To our knowledge, no documented SNAP interventions increase take-up and improve targeting.

12 / 51

Our Contribution

First assessment of how access to in-person assistance via opening/closing SNAP offices and SNAP-authorized stores (coming soon) impacts participation and targeting.
Provide evidence that reducing transaction costs via access to SNAP offices increases participation and improves targeting.
Our setting includes multiple relevant actors: SNAP-authorized stores and program offices.
- Literature has looked at one of these in isolation with limited focus on private actors (Handbury & Moshary, 2021; Beatty, Bitler, and van Der Werf, 2021; Meckel, 2020).

13 / 51

Our Contribution

Link Census data on residential location with administrative SNAP data and information on SNAP offices and SNAP-authorized retailers.
Not possible to study this question with existing surveys due to limited sample sizes and data quality issues.

14 / 51

Our Contribution

Link Census data on residential location with administrative SNAP data and information on SNAP offices and SNAP-authorized retailers.
Not possible to study this question with existing surveys due to limited sample sizes and data quality issues.
Study population within states and don’t have to focus samples with limited generalizability (previous RCT literature).
Rich administrative data on other programs and mobility.

14 / 51

Our Contribution

Link Census data on residential location with administrative SNAP data and information on SNAP offices and SNAP-authorized retailers.
Not possible to study this question with existing surveys due to limited sample sizes and data quality issues.
Study population within states and don’t have to focus samples with limited generalizability (previous RCT literature).
Rich administrative data on other programs and mobility.
Samples which permit us to explore staggered adoption designs.
We establish a first stage for work to come on outcomes.

14 / 51

Institutional Background14 / 51

SNAP

Backbone of US safety net.
Only US safety net program available to nearly all low-income households.
Means tested (income and asset tests) and includes work requirements for some households (i.e., ABAWDs).
Certification periods typically 6-12 months (seniors 24+ months; ABAWD 3 months).
- In-person or phone interview required along with income verification.

15 / 51

SNAP

Application Process

Some heterogeneity across states.
Many states have online application portals and hotlines.
- Most people still submit applications in person.
Most states have switched from face-to-face to phone interview.
Provide household information and records of income/assets.

Source: Arizona Department of Economic Security Website.

16 / 51

SNAP

Role of SNAP Offices

Provide in-person assistance navigating application process.
Application prevalence: $>$ online $>$ fax $>$ email $>$ phone.
- SNAP offices report that typically 80% of applications are in person.
Typically provide resources to connect SNAP applicants to other assistance programs (e.g., HUD, TANF, Medicaid/Medicare, LEAP, WIC).
Some offices allow direct applications to other programs (typically TANF, but also sometimes HUD and Medicaid).
Some offices even help find jobs, daycare, and housing.

17 / 51

Data17 / 51

Data

SNAP Administrative Data
Master Address File
MAFARF
HUD Administrative Data
ACS
HHS TANF Administrative Data
Collected 243 SNAP office closings and 336 openings.

18 / 51

Measuring Access to SNAP Offices18 / 51

Measuring Access to SNAP Offices

We count number of SNAP recipients in administrative records residing within given distance to each SNAP office in each year.
In the case of overlap, we assign case to the closest office (i.e., "No Overlap").
- Show robustness to counting cases multiple times if overlap
Perform similar exercise for any person with a Census PIK to use as denominator (awaiting disclosure).

19 / 51

Measuring Access to SNAP Offices

We count number of SNAP recipients in administrative records residing within given distance to each SNAP office in each year.
In the case of overlap, we assign case to the closest office (i.e., "No Overlap").
- Show robustness to counting cases multiple times if overlap
Perform similar exercise for any person with a Census PIK to use as denominator (awaiting disclosure).
Next, we illustrate method for counting cases within a ring of an office using data on SNAP offices and SNAP-authorized stores.
- Use SNAP-authorized stores for this example to avoid Census disclosure issues

19 / 51

Data Source: SNAP-Authorized Stores - USDA's Store Tracking and Redemption System (STARS); SNAP Offices - Collected by authors.

Here, SNAP-authorized stores (dots) in 1 mile of SNAP Offices (diamonds).

20 / 51

Data Source: SNAP-Authorized Stores - USDA's Store Tracking and Redemption System (STARS); SNAP Offices - Collected by authors.

Here, SNAP-authorized stores (dots) in 1 mile of SNAP Offices (diamonds).
We assign dots to closest SNAP Office (signified by matching color).
We then count number of dots assigned to each SNAP office each year.

21 / 51

Data Source: SNAP-Authorized Stores - USDA's Store Tracking and Redemption System (STARS); SNAP Offices - Collected by authors.

In context of paper, the dots are SNAP admin cases.
Compute counts for offices both before they open and after they close.
Compute counts for various case types (e.g., no gross income)

22 / 51

SNAP Offices

Overlap

Within a mile, 12% of cases overlap in rural counties.
Within a mile, 28% of cases overlap in urban counties.

Distances

Most analyses focus on SNAP cases within 1 mile of the SNAP Office.
- Roughly 25th percentile of distance distribution for both rural and urban counties.
Also explore distances of $\color{#b92e34 }{(1,10]}$ miles for and $\color{#07506F }{(1,4]}$ for urban.
- 75th percentile of distance distribution.

23 / 51

Share Affected by Openings/Closings23 / 51

"Any Office": All offices, including those that neither open nor close.
Meaningful share of recipients live near SNAP offices.
Of the 27,540,000 rural clients we observe: ^{*^{695,000 live < 1 mile of opening office (781,000 for closing office).}}
Of the 24,590,000 urban clients we observe:
- 452,000 live < 1 mile of opening office (843,000 for closing office).

^* Recall we only observe a subset of states in administrative data.

24 / 51

Client Characteristics by SNAP Office Type

Rural: Monthly income ~ $200 lower near opening/closing offices.
Urban: Monthly income ~ $100 lower near closing offices.
All: Monthly income higher for recipients far away from offices.

25 / 51

Client Characteristics by SNAP Office Type

Urban Counties: Lower share of Black recipients near opening offices.
Both: Recipients far away from offices are less likely to be Black/Hispanic and more likely to be White.

26 / 51

Client Characteristics by SNAP Office Type

Characteristics relatively balanced across office type.

27 / 51

Empirical Design27 / 51

Empirical Design

Preferred Specification - Two-way Fixed Effects

$y_{it} = \sum_{\tau, \tau\neq 1}\beta_\tau 1(t-E_i = \tau) + \gamma_i + \theta_t + \epsilon_{it}$

$i\ -$ SNAP office
$t\ -$ calendar year
$E_i\ -$ year of opening/closing

28 / 51

Empirical Design

Preferred Specification - Two-way Fixed Effects

$y_{it} = \sum_{\tau, \tau\neq 1}\beta_\tau 1(t-E_i = \tau) + \gamma_i + \theta_t + \epsilon_{it}$

$i\ -$ SNAP office
$t\ -$ calendar year
$E_i\ -$ year of opening/closing
Panel design hinges on exogenous timing of openings/closings.
- Unobserved determinants of SNAP participation not differentially trending across office types.

28 / 51

Empirical Design

$y_{it} = \sum_{\tau, \tau\neq 1}\beta_\tau 1(t-E_i = \tau)+ \gamma_i + \theta_t + \epsilon_{it}$

Run on a panel balanced over main event times $\tau \in [-3,3]$ .
Coefficients estimated for all event times, but only report $\tau \in [-3,3]$ .
Cluster standard errors by SNAP office (location where SNAP office will be/is/used to be).

29 / 51

Empirical Design

$y_{it} = \sum_{\tau, \tau\neq 1}\beta_\tau 1(t-E_i = \tau)+ \gamma_i + \theta_t + \epsilon_{it}$

Run on a panel balanced over main event times $\tau \in [-3,3]$ .
Coefficients estimated for all event times, but only report $\tau \in [-3,3]$ .
Cluster standard errors by SNAP office (location where SNAP office will be/is/used to be).
Sample includes all SNAP offices and event time is only calculated for treated offices (i.e., opening or closing).
Test robustness to heterogeneous treatment effects (de Chaisemartin and D'Haultfœuille, 2020).

29 / 51

Empirical DesignExogeneity of our VariationSNAP recipients far away from offices have  higher income and more likely to be White.⇒ Analysis only focuses on recipients in close vicinity of offices.
Office changes unlikely to target neighborhood characteristics in such close vicinity.

Characteristics of SNAP recipients near offices relatively similar across office types (i.e., opening/closing/any).Caveat: Urban offices tend to open in whiter neighborhoods and close in lower-income neighborhoods.

30 / 51

Empirical DesignExogeneity of our VariationSNAP recipients far away from offices have  higher income and more likely to be White.⇒ Analysis only focuses on recipients in close vicinity of offices.
Office changes unlikely to target neighborhood characteristics in such close vicinity.

Characteristics of SNAP recipients near offices relatively similar across office types (i.e., opening/closing/any).Caveat: Urban offices tend to open in whiter neighborhoods and close in lower-income neighborhoods.

Importantly, we leverage timing of changes.
Will use Dun & Bradstreet firms to check whether openings and closings affected by aggregate retail trends.
30 / 51

Results30 / 51

Mean Distance to Office (Miles)

Goal: Measure how travel distances are impacted by SNAP office openings/closings.
Use the Census Master Address file (MAFX); a static file of all known residential locations in US.
Measure average travel distance from every MAFX address within 1 of SNAP Office during years leading up to and following opening/closing.

31 / 51

Mean Distance to Office (Miles)

	Closing Rural	Closing Urban	Opening Rural	Opening Urban
Distance	0-1	0-1	0-1	0-1
Model	TWFE	TWFE	TWFE	TWFE
Avg. Estimate	5.63*** (1.16)	1.97*** (0.70)	-5.35*** (1.03)	-1.30*** (0.26)
Baseline Y	.56	.63	5.3	1.6
Event Study

Data Source: SNAP Administrative Data - various states and years; Census Master Address File (MAFX).

32 / 51

Mean Distance to Office (Miles)

Rural Counties: Open/closings change average distance by 5 miles.
Urban Counties: Open/closings change average distance by 1-2 miles.
Distances change enough to move from office being walkable to requiring transit to access.

33 / 51

Mean Distance to Office (Miles)

Rural Counties: Open/closings change average distance by 5 miles.
Urban Counties: Open/closings change average distance by 1-2 miles.
Distances change enough to move from office being walkable to requiring transit to access.

Next, we explore impact of SNAP office closings and openings on total counts of new SNAP clients living within a mile radius.

33 / 51

# of New SNAP Clients - Short Distance

Office Closings

	Closing Rural	Closing Urban
Distance	0-1	0-1
Model	TWFE	TWFE
Avg. Estimate	-3.17 (12.6)	-88.0 (62.5)
Baseline Y	224	1,020
Event Study

Data Source: SNAP Administrative Data - various states and years.

Rural & Urban Counties: Temporary spike during closing year.
Urban Counties: 3 years after closing, roughly 90 fewer clients (8.8% decrease relative to baseline).

34 / 51

# of New SNAP Clients - Short Distance

Heterogeneity by Gross Income

Closing, Rural

Closing, Urban

Data Source: SNAP Administrative Data - various states and years.

Puzzle: Temporary spike driven by cases with gross income $\in (0,100]$ %FPL.
Participation falls for clients with no gross income.

35 / 51

# of New SNAP Clients - Short Distance

Heterogeneity

Data Source: SNAP Administrative Data - various states and years.

Cases with no gross income fell the most.
Elderly case increase is driven by mobility (awaiting disclosure).

36 / 51

# of New SNAP Clients - Long Distance

Office Closings

	Closing Rural	Closing Urban
Distance	1-10	1-4
Model	TWFE	TWFE
Avg. Estimate	59.5 (48.2)	309.4** (153.6)
Baseline Y	596	1,710
Event Study

Data Source: SNAP Administrative Data - various states and years.

Persistent increase in caseloads.
Similarly driven by cases with some gross income (awaiting disclosure).
Currently working on testing robustness to population changes and short-distance office relocations.

37 / 51

# of New SNAP Clients - Short Distance

Office Openings

	Opening Rural	Opening Urban
Distance	0-1	0-1
Model	TWFE	TWFE
Avg. Estimate	35.9*** (11.9)	294.2*** (60.7)
Baseline Y	153	466
Event Study

Data Source: SNAP Administrative Data - various states and years.

Rural & Urban Counties: Large, immediate impacts that increase with time.
By three years after opening:
- Rural Counties: 53 additional SNAP clients (35% increase).
- Urban Counties: 332 additional SNAP clients (71% increase).

38 / 51

# of New SNAP Clients - Short Distance

Heterogeneity by Gross Income

Opening, Rural

Opening, Urban

Data Source: SNAP Administrative Data - various states and years.

Biggest participation impacts for cases without gross income.
Evidence of improved targeting.

39 / 51

# of New SNAP Clients - Short Distance

Heterogeneity

Data Source: SNAP Administrative Data - various states and years.

Increases in program participation across many subgroups.

40 / 51

# of New SNAP Clients - Long Distance

Office Openings

	Opening Rural	Opening Urban
Distance	1-10	1-4
Model	TWFE	TWFE
Avg. Estimate	170.4*** (47.4)	595.1*** (133.6)
Baseline Y	405	1,731
Event Study

Data Source: SNAP Administrative Data - various states and years.

Similar pattern as for 1 mile rings.
By three years after opening:
- Rural Counties: 271 additional SNAP clients (67% increase).
- Urban Counties: 768 additional SNAP clients (44% increase).

41 / 51

Take-Away

SNAP office openings/closings generates meaningful differences in access.
These differences in access substantially impact participation rates (particularly for cases without income).
Within 1 mile:
- Closing offices experience an uptick in cases during the closing year. Possibly clearing out application queue. Subsequent modest declines in participation (urban counties).
- Opening offices experience large persistent increases in participation.

42 / 51

Robustness42 / 51

Robustness

Next, we test robustness to changing:

Model: Use de Chaisemartin and D'Haultfœuille, 2020 estimator that is robust to heterogeneous treatment effects.
Overlap: Allowing cases to be counted multiple times if within a mile of multiple offices.
Outcome: Measuring number of cases as opposed to number of clients.

43 / 51

Robustness

Next, we test robustness to changing:

Model: Use de Chaisemartin and D'Haultfœuille, 2020 estimator that is robust to heterogeneous treatment effects.
Overlap: Allowing cases to be counted multiple times if within a mile of multiple offices.
Outcome: Measuring number of cases as opposed to number of clients.

Summary: Results are not substantially impacted.

43 / 51

Rural Closings - Robustness

	New Clients	New Clients	New Clients	New Cases
Distance	0-1	0-1	0-1	0-1
Model	TWFE	deCh/DH (AER)	TWFE	TWFE
Overlap	No	No	Yes	No
Avg. Estimate	-3.2 (12.6)	13.9 (13.0)	-10.1 (14.5)	-6.6 (7.7)
Baseline Y	223.6	204.5	295.4	121.6
Main Specification	X
Event Study

Data Source: SNAP Administrative Data - various states and years.

44 / 51

Urban Closings - Robustness

	New Clients	New Clients	New Clients	New Cases
Distance	0-1	0-1	0-1	0-1
Model	TWFE	deCh/DH (AER)	TWFE	TWFE
Overlap	No	No	Yes	No
Avg. Estimate	-88.0 (62.5)	84.2 (82.9)	-101.9 (67.3)	-78.7 (48.0)
Baseline Y	1,020	856	1,250	651
Main Specification	X
Event Study

Data Source: SNAP Administrative Data - various states and years.

45 / 51

Rural Openings - Robustness

	New Clients	New Clients	New Clients	New Cases
Distance	0-1	0-1	0-1	0-1
Model	TWFE	deCh/DH (AER)	TWFE	TWFE
Overlap	No	No	Yes	No
Avg. Estimate	35.9*** (11.9)	26.5*** (9.2)	26.7** (13.1)	20.2*** (7.1)
Baseline Y	153	144	235.2	121.6
Main Specification	X
Event Study

Data Source: SNAP Administrative Data - various states and years.

46 / 51

Urban Openings - Robustness

	New Clients	New Clients	New Clients	New Cases
Distance	0-1	0-1	0-1	0-1
Model	TWFE	deCh/DH (AER)	TWFE	TWFE
Overlap	No	No	Yes	No
Avg. Estimate	294.2*** (60.7)	508.4 (317.3)	281.8*** (74.1)	160.0*** (33.1)
Baseline Y	466	424.3	711.6	241.6
Main Specification	X
Event Study

Data Source: SNAP Administrative Data - various states and years.

47 / 51

Further outcomes which provide context47 / 51

	SNAP Offices	SNAP Stores
Closing Rural
Closing Urban
Opening Rural
Opening Urban

Data Sources: SNAP Admin - various states & years; HUD PICS/TRACS Admin; USDA's Store Tracking and Redemption System (STARS)

Count of SNAP offices within a mile of focal office increases for closings and decreases for openings.
- Result of offices relocating short distances.
- Working on robustness to eliminating short-distance relocations.
No clear pattern for SNAP-authorized stores.

48 / 51

	Section 8	Public Housing
Closing Rural
Closing Urban
Opening Rural
Opening Urban

Data Source: SNAP Admin - various states & years; HUD PICS/TRACS Admin; USDA's Store Tracking and Redemption System (STARS)

Noisy increase in HUD program participation for SNAP office openings.
Recall that offices connect applicants to other programs, sometimes can even apply directly.

49 / 51

Conclusion

Access to SNAP offices substantially increases program participation.
Particularly important for families without income (i.e., improved targeting).
Interesting because many states are fully online with active help phone lines.
Face-to-face assistance may provide additional aid overcoming transaction costs? (Wu, 2021)
Policy implications to increase in-person assistance for applications.
- Similar to mobile WIC clinics?

50 / 51

Next Steps

Qualitatively examine mechanisms for why offices may matter.
Similar analysis for distance to SNAP-authorized retailers and businesses more generally (D&B).
Robustness to population changes.
Mobility and case composition outcomes.
Use as IV to explore impact of SNAP on labor supply, cross-program participation, and health.

51 / 51

Thank you

51 / 51

References

Akerlof, G. A. (1978). "The Economics of "Tagging" as Applied to the Optimal Income Tax , Welfare Programs , and Manpower Planning". In: American Economic Review 68.1, pp. 8-19.

Ambrozek, C. (2021). "WIC participant responses to vendor disqualification".

Armour, P. (2018). "The role of information in disability insurance application: An analysis of the social security statement phase-in". In: American Economic Journal: Economic Policy 10.3, pp. 1-41. ISSN: 1945774X. DOI: 10.1257/pol.20160605.

Bartlett, S, N. Burstein, et al. (2004). "Food Stamp Program Access Study Final Report". In: Washington, DC: Economic Research Service..

51 / 51

References

Besley, T. and S. Coate (1992b). "Workfare versus Welfare : Incentive Arguments for Work Requirements in Poverty- Alleviation Programs". In: American Economic Review 82.1, pp. 249-261.

Bettinger, E. P, B. T. Long, et al. (2012). "The Role of Application Assistance and Information in College Decisions: Results from the H&R Block Fafsa Experiment". In: The Quarterly Journal of Economics 127.3, pp. 1205-1242.

Bhargava, S. and D. Manoli (2015). "Psychological frictions and the incomplete take-up of social benefits: Evidence from an IRS field experiment". In: American Economic Review. Vol. 105. 11. American Economic Association, pp. 3489-3529. DOI: 10.1257/aer.20121493.

Chaisemartin, C. de and X. D'Haultfœuille (2020). "Two-Way Fixed Effects Estimators with Heterogeneous Treatment Effects". In: American Economic Review 110.9, pp. 2964-2996. ISSN: 19447981. DOI: 10.1257/aer.20181169. eprint: 1803.08807.

51 / 51

References

Currie, J. (2006). "The Take Up of Social Benefits". In: Poverty, the Distribution of Income, and Public Policy , pp. 80-148.

Daponte, B. O, S. Sanders, et al. (1999). "Why do low-income households not use food stamps? Evidence from an experiment". In: Journal of Human Resources 34.3, pp. 612-628. ISSN: 0022166X. DOI: 10.2307/146382.

Deshpande, M. and Y. Li (2019). "Who Is Screened Out? Application Costs and the Targeting of Disability Programs". In: American Economic Journal: Economic Policy 11.4, pp. 213-248.

Finkelstein, A. and M. J. Notowidigdo (2019). "Take-Up and Targeting: Experimental Evidence from SNAP". In: The Quarterly Journal of Economics 134.3, pp. 1505-1556. ISSN: 0033-5533. DOI: 10.1093/qje/qjz013.

51 / 51

References

Gray, C. (2018). "Why Leave Benefits on the Table? Evidence from SNAP". In: SSRN Electronic Journal. ISSN: 1556-5068. DOI: 10.2139/ssrn.3203395.

Handbury, J. and S. Moshary (2021). "School Food Policy Affects Everyone: Retail Responses to the National School Lunch Program". In: SSRN Electronic Journal, pp. 1-49. DOI: 10.2139/ssrn.3897936.

Hanratty, M. J. (2006). "Has the food stamp program become more accessible? Impacts of recent changes in reporting requirements and asset eligibility limits". In: Journal of Policy Analysis and Management 25.3, pp. 603-621. ISSN: 02768739. DOI: 10.1002/pam.20193.

Herd, P. and D. Moynihan (2019). Administrative burden: Policymaking by other means. Russel Sage Foundation.

51 / 51

References

Kabbani, N. S. and P. E. Wilde (2003b). "Short recertification periods in the U.S. Food Stamp Program". In: Journal of Human Resources 38.SUPPLEMENT, pp. 1112-1138. ISSN: 0022166X. DOI: 10.2307/3558983.

Kleven, H. J. and W. Kopczuk (2011). "Transfer program complexity and the take-up of social benefits". In: American Economic Journal: Economic Policy 3.1, pp. 54-90. ISSN: 19457731. DOI: 10.1257/pol.3.1.54.

Meckel, K. (2020). "Is the Cure Worse than the Disease? Unintended Effects of Payment Reform in a Quantity-based Transfer Program". In: American Economic Review 110.6, pp. 1821-1865. ISSN: 0898-2937. DOI: 10.3386/w26725.

Moffitt, R. (1983). "An Economic Model of Welfare Stigma". In: American Economic Review 73.5, pp. 1023-1035.

51 / 51

References

Rossin-Slater, M. (2013). "WIC in your neighborhood: New evidence on the impacts of geographic access to clinics". In: Journal of Public Economics 102, pp. 51-69.

51 / 51

How Place and Poverty IntersectGeographic Barriers and Low SNAP Take-upMarianne Bitler, UC Davis & NBERJason Cook, University of UtahSonya R. Porter, US Census Bureau1 / 51

Disclaimer

Any opinions and conclusions expressed herein are those of the authors and do not reflect the views of the U.S. Census Bureau. The statistical summaries reported in this paper have been cleared by the Census Bureau's Disclosure Review Board release authorization number CBDRB-FY21-CES014- 049. All results have been reviewed to ensure that no confidential information is disclosed.

2 / 51

Motivation

US somewhat unique within advanced economies in design of safety net. System very federalized with various levels of government running programs.
People apply to many programs and rules not harmonized and application is not automatic as in much of the developed world $\Rightarrow$ low take-up is more of issue in US than other places (Currie, 2006; Currie and Gahvari, 2008).

3 / 51

Motivation

US somewhat unique within advanced economies in design of safety net. System very federalized with various levels of government running programs.
People apply to many programs and rules not harmonized and application is not automatic as in much of the developed world $\Rightarrow$ low take-up is more of issue in US than other places (Currie, 2006; Currie and Gahvari, 2008).
Social program take-up: Hot topic in public econ, yet understudied.
- Emphasis on understanding role of barriers to accessing safety net.
Barriers impact both take-up (how many people enroll) and targeting (what types of people enroll).

3 / 51

Competing Models to Explain Low Take-Up

Neoclassical model

People make decisions balancing costs and benefits in a utility maximizing framework.
Incomplete take-up is a function of barriers:
- Incomplete information: Program existence, eligibility rules.
- Stigma: Either concerns about whether you should participate or about how others will judge you.
- Transaction costs: E.g., travel costs, difficultly with documentation, time costs.

4 / 51

Competing Models to Explain Low Take-Up

Neoclassical model

People make decisions balancing costs and benefits in a utility maximizing framework.
Incomplete take-up is a function of barriers:
- Incomplete information: Program existence, eligibility rules.
- Stigma: Either concerns about whether you should participate or about how others will judge you.
- Transaction costs: E.g., travel costs, difficultly with documentation, time costs.

4 / 51

Competing Models to Explain Low Take-Up

Behavioral science model

Limited bandwidth or stress impairs people. Suggests incomplete take-up might be harmful or inefficient (Bertrand, Mullainathan, & Shafir, 2004).

Administrative Burden

Government actors may intentionally create "administrative burden" such as learning costs, compliance costs, and psychological costs to limit use, possibly due to limited capacity (Herd and Moynihan, 2019).

Under these models, barriers could be inefficient if worsen targeting, e.g., if the poor face limited bandwidth or scarcity, more needy individuals could be deterred by hassle costs.^*

^* Though take up and targeting alone are not enough to assess normative implications (Finkelstein & Notowidigdo, 2019).

5 / 51

Literature5 / 51

Finkelstein and Notowidigdo (2019)

Model for social welfare incorporates social costs and perceptions of costs of application (looking at targeting is not enough).
Fit parameters with RCT on Medicaid participants not on SNAP.
Did information and application help affect take-up and targeting?

6 / 51

Finkelstein and Notowidigdo (2019)

Model for social welfare incorporates social costs and perceptions of costs of application (looking at targeting is not enough).
Fit parameters with RCT on Medicaid participants not on SNAP.
Did information and application help affect take-up and targeting?

Assistance $>$ information $>$ status quo; Compliers are better off.
Worse targeting for all interventions.

6 / 51

Finkelstein and Notowidigdo (2019)

Model for social welfare incorporates social costs and perceptions of costs of application (looking at targeting is not enough).
Fit parameters with RCT on Medicaid participants not on SNAP.
Did information and application help affect take-up and targeting?

Assistance $>$ information $>$ status quo; Compliers are better off.
Worse targeting for all interventions.
Quantify costs of assessing eligibility?
- Understanding expense of eligibility assessment would help determine size of wedge between social and private welfare.
- Efforts to cut cost of eligibility would help shrink wedge.
- SNAP average administrative costs low relative to benefit amount.

6 / 51

Literature

Remaining literature can be conceptualized as studying how different types of barriers impact take-up and targeting.

7 / 51

Barriers Impact Take-Up

Information Barriers

SNAP-eligible people often don't realize eligibility (Bartlett et al., 2004); providing information increases take-up (Daponte, Sanders, & Taylor, 1999; Finkelstein and Notowidigdo, 2019).
Informational interventions matter in some other contexts as well: EITC (Bhargava and Manoli, 2015) and SSDI (Armour, 2018), but not others: FAFSA (Bettinger et al., 2012).

8 / 51

Barriers Impact Take-Up

Transaction Costs in SNAP

Reducing transaction costs in SNAP increases take-up via: application help (Schanzenbach, 2001; Finkelstein and Notowidigdo, 2019), certification periods (Kabbani and Wilde, 2003), certification reporting requirements (Gray, 2018; Hanratty, 2006; Unrath, 2021).
Less time for SNAP recertification interviews leads to more churning (Homonoff and Somerville, 2021)
Switching to automated SNAP application process decreases take-up (Wu, 2021)

9 / 51

Barriers Impact Take-Up

Transaction Costs in Other Programs

Learning costs with WIC, relative to SNAP, influenced by what stores carry (Barnes, 2021).
WIC participation during COVID affected by pick up and enrollment rules (in person/not) (Barnes and Petry, 2021; Whaley and Anderson, 2021; Vasan, Kenyon, and Roberto, 2021).
Access to program offices matters: SSA offices (Deshpande and Li, 2019) and WIC program offices/vendors (Rossin-Slater, 2013; Ambrozek, 2021).

10 / 51

Barriers Impact Targeting

Information Barriers

Complexity worsens targeting of low-income cases for EITC (Bhargava and Manoli, 2015).
Information mailers induce less-needy households to enroll in SNAP (Finkelstein and Notowidigdo, 2019).

11 / 51

Barriers Impact Targeting

Transaction Costs

Closing SSA offices had mixed impacts on who is deterred (Deshpande and Li, 2019).
SNAP application assistance reduces targeting across all dimensions (Finkelstein and Notowidigdo, 2019).
Automated application system reduces take-up, but it improves targeting efficiency for new recipients and worsens among recertifiers (Wu, 2021).
Recertification rules reduce take-up and lower retention, but improve targeting (Unrath, 2021).

12 / 51

Barriers Impact Targeting

Transaction Costs

Closing SSA offices had mixed impacts on who is deterred (Deshpande and Li, 2019).
SNAP application assistance reduces targeting across all dimensions (Finkelstein and Notowidigdo, 2019).
Automated application system reduces take-up, but it improves targeting efficiency for new recipients and worsens among recertifiers (Wu, 2021).
Recertification rules reduce take-up and lower retention, but improve targeting (Unrath, 2021).

To our knowledge, no documented SNAP interventions increase take-up and improve targeting.

12 / 51

Our Contribution

First assessment of how access to in-person assistance via opening/closing SNAP offices and SNAP-authorized stores (coming soon) impacts participation and targeting.
Provide evidence that reducing transaction costs via access to SNAP offices increases participation and improves targeting.
Our setting includes multiple relevant actors: SNAP-authorized stores and program offices.
- Literature has looked at one of these in isolation with limited focus on private actors (Handbury & Moshary, 2021; Beatty, Bitler, and van Der Werf, 2021; Meckel, 2020).

13 / 51

Our Contribution

Link Census data on residential location with administrative SNAP data and information on SNAP offices and SNAP-authorized retailers.
Not possible to study this question with existing surveys due to limited sample sizes and data quality issues.

14 / 51

Our Contribution

Link Census data on residential location with administrative SNAP data and information on SNAP offices and SNAP-authorized retailers.
Not possible to study this question with existing surveys due to limited sample sizes and data quality issues.
Study population within states and don’t have to focus samples with limited generalizability (previous RCT literature).
Rich administrative data on other programs and mobility.

14 / 51

Our Contribution

Link Census data on residential location with administrative SNAP data and information on SNAP offices and SNAP-authorized retailers.
Not possible to study this question with existing surveys due to limited sample sizes and data quality issues.
Study population within states and don’t have to focus samples with limited generalizability (previous RCT literature).
Rich administrative data on other programs and mobility.
Samples which permit us to explore staggered adoption designs.
We establish a first stage for work to come on outcomes.

14 / 51

Institutional Background14 / 51

SNAP

Backbone of US safety net.
Only US safety net program available to nearly all low-income households.
Means tested (income and asset tests) and includes work requirements for some households (i.e., ABAWDs).
Certification periods typically 6-12 months (seniors 24+ months; ABAWD 3 months).
- In-person or phone interview required along with income verification.

15 / 51

SNAP

Application Process

Some heterogeneity across states.
Many states have online application portals and hotlines.
- Most people still submit applications in person.
Most states have switched from face-to-face to phone interview.
Provide household information and records of income/assets.

Source: Arizona Department of Economic Security Website.

16 / 51

SNAP

Role of SNAP Offices

Provide in-person assistance navigating application process.
Application prevalence: $>$ online $>$ fax $>$ email $>$ phone.
- SNAP offices report that typically 80% of applications are in person.
Typically provide resources to connect SNAP applicants to other assistance programs (e.g., HUD, TANF, Medicaid/Medicare, LEAP, WIC).
Some offices allow direct applications to other programs (typically TANF, but also sometimes HUD and Medicaid).
Some offices even help find jobs, daycare, and housing.

17 / 51

Data17 / 51

Data

SNAP Administrative Data
Master Address File
MAFARF
HUD Administrative Data
ACS
HHS TANF Administrative Data
Collected 243 SNAP office closings and 336 openings.

18 / 51

Measuring Access to SNAP Offices18 / 51

Measuring Access to SNAP Offices

We count number of SNAP recipients in administrative records residing within given distance to each SNAP office in each year.
In the case of overlap, we assign case to the closest office (i.e., "No Overlap").
- Show robustness to counting cases multiple times if overlap
Perform similar exercise for any person with a Census PIK to use as denominator (awaiting disclosure).

19 / 51

Measuring Access to SNAP Offices

We count number of SNAP recipients in administrative records residing within given distance to each SNAP office in each year.
In the case of overlap, we assign case to the closest office (i.e., "No Overlap").
- Show robustness to counting cases multiple times if overlap
Perform similar exercise for any person with a Census PIK to use as denominator (awaiting disclosure).
Next, we illustrate method for counting cases within a ring of an office using data on SNAP offices and SNAP-authorized stores.
- Use SNAP-authorized stores for this example to avoid Census disclosure issues

19 / 51

Data Source: SNAP-Authorized Stores - USDA's Store Tracking and Redemption System (STARS); SNAP Offices - Collected by authors.

Here, SNAP-authorized stores (dots) in 1 mile of SNAP Offices (diamonds).

20 / 51

Data Source: SNAP-Authorized Stores - USDA's Store Tracking and Redemption System (STARS); SNAP Offices - Collected by authors.

Here, SNAP-authorized stores (dots) in 1 mile of SNAP Offices (diamonds).
We assign dots to closest SNAP Office (signified by matching color).
We then count number of dots assigned to each SNAP office each year.

21 / 51

Data Source: SNAP-Authorized Stores - USDA's Store Tracking and Redemption System (STARS); SNAP Offices - Collected by authors.

In context of paper, the dots are SNAP admin cases.
Compute counts for offices both before they open and after they close.
Compute counts for various case types (e.g., no gross income)

22 / 51

SNAP Offices

Overlap

Within a mile, 12% of cases overlap in rural counties.
Within a mile, 28% of cases overlap in urban counties.

Distances

Most analyses focus on SNAP cases within 1 mile of the SNAP Office.
- Roughly 25th percentile of distance distribution for both rural and urban counties.
Also explore distances of $\color{#b92e34 }{(1,10]}$ miles for and $\color{#07506F }{(1,4]}$ for urban.
- 75th percentile of distance distribution.

23 / 51

Share Affected by Openings/Closings23 / 51

"Any Office": All offices, including those that neither open nor close.
Meaningful share of recipients live near SNAP offices.
Of the 27,540,000 rural clients we observe: ^{*^{695,000 live < 1 mile of opening office (781,000 for closing office).}}
Of the 24,590,000 urban clients we observe:
- 452,000 live < 1 mile of opening office (843,000 for closing office).

^* Recall we only observe a subset of states in administrative data.

24 / 51

Client Characteristics by SNAP Office Type

Rural: Monthly income ~ $200 lower near opening/closing offices.
Urban: Monthly income ~ $100 lower near closing offices.
All: Monthly income higher for recipients far away from offices.

25 / 51

Client Characteristics by SNAP Office Type

Urban Counties: Lower share of Black recipients near opening offices.
Both: Recipients far away from offices are less likely to be Black/Hispanic and more likely to be White.

26 / 51

Client Characteristics by SNAP Office Type

Characteristics relatively balanced across office type.

27 / 51

Empirical Design27 / 51

Empirical Design

Preferred Specification - Two-way Fixed Effects

$y_{it} = \sum_{\tau, \tau\neq 1}\beta_\tau 1(t-E_i = \tau) + \gamma_i + \theta_t + \epsilon_{it}$

$i\ -$ SNAP office
$t\ -$ calendar year
$E_i\ -$ year of opening/closing

28 / 51

Empirical Design

Preferred Specification - Two-way Fixed Effects

$y_{it} = \sum_{\tau, \tau\neq 1}\beta_\tau 1(t-E_i = \tau) + \gamma_i + \theta_t + \epsilon_{it}$

$i\ -$ SNAP office
$t\ -$ calendar year
$E_i\ -$ year of opening/closing
Panel design hinges on exogenous timing of openings/closings.
- Unobserved determinants of SNAP participation not differentially trending across office types.

28 / 51

Empirical Design

$y_{it} = \sum_{\tau, \tau\neq 1}\beta_\tau 1(t-E_i = \tau)+ \gamma_i + \theta_t + \epsilon_{it}$

Run on a panel balanced over main event times $\tau \in [-3,3]$ .
Coefficients estimated for all event times, but only report $\tau \in [-3,3]$ .
Cluster standard errors by SNAP office (location where SNAP office will be/is/used to be).

29 / 51

Empirical Design

$y_{it} = \sum_{\tau, \tau\neq 1}\beta_\tau 1(t-E_i = \tau)+ \gamma_i + \theta_t + \epsilon_{it}$

Run on a panel balanced over main event times $\tau \in [-3,3]$ .
Coefficients estimated for all event times, but only report $\tau \in [-3,3]$ .
Cluster standard errors by SNAP office (location where SNAP office will be/is/used to be).
Sample includes all SNAP offices and event time is only calculated for treated offices (i.e., opening or closing).
Test robustness to heterogeneous treatment effects (de Chaisemartin and D'Haultfœuille, 2020).

29 / 51

Empirical DesignExogeneity of our VariationSNAP recipients far away from offices have  higher income and more likely to be White.⇒ Analysis only focuses on recipients in close vicinity of offices.
Office changes unlikely to target neighborhood characteristics in such close vicinity.

Characteristics of SNAP recipients near offices relatively similar across office types (i.e., opening/closing/any).Caveat: Urban offices tend to open in whiter neighborhoods and close in lower-income neighborhoods.

30 / 51

Empirical DesignExogeneity of our VariationSNAP recipients far away from offices have  higher income and more likely to be White.⇒ Analysis only focuses on recipients in close vicinity of offices.
Office changes unlikely to target neighborhood characteristics in such close vicinity.

Characteristics of SNAP recipients near offices relatively similar across office types (i.e., opening/closing/any).Caveat: Urban offices tend to open in whiter neighborhoods and close in lower-income neighborhoods.

Importantly, we leverage timing of changes.
Will use Dun & Bradstreet firms to check whether openings and closings affected by aggregate retail trends.
30 / 51

Results30 / 51

Mean Distance to Office (Miles)

Goal: Measure how travel distances are impacted by SNAP office openings/closings.
Use the Census Master Address file (MAFX); a static file of all known residential locations in US.
Measure average travel distance from every MAFX address within 1 of SNAP Office during years leading up to and following opening/closing.

31 / 51

Mean Distance to Office (Miles)

	Closing Rural	Closing Urban	Opening Rural	Opening Urban
Distance	0-1	0-1	0-1	0-1
Model	TWFE	TWFE	TWFE	TWFE
Avg. Estimate	5.63*** (1.16)	1.97*** (0.70)	-5.35*** (1.03)	-1.30*** (0.26)
Baseline Y	.56	.63	5.3	1.6
Event Study

Data Source: SNAP Administrative Data - various states and years; Census Master Address File (MAFX).

32 / 51

Mean Distance to Office (Miles)

Rural Counties: Open/closings change average distance by 5 miles.
Urban Counties: Open/closings change average distance by 1-2 miles.
Distances change enough to move from office being walkable to requiring transit to access.

33 / 51

Mean Distance to Office (Miles)

Rural Counties: Open/closings change average distance by 5 miles.
Urban Counties: Open/closings change average distance by 1-2 miles.
Distances change enough to move from office being walkable to requiring transit to access.

Next, we explore impact of SNAP office closings and openings on total counts of new SNAP clients living within a mile radius.

33 / 51

# of New SNAP Clients - Short Distance

Office Closings

	Closing Rural	Closing Urban
Distance	0-1	0-1
Model	TWFE	TWFE
Avg. Estimate	-3.17 (12.6)	-88.0 (62.5)
Baseline Y	224	1,020
Event Study

Data Source: SNAP Administrative Data - various states and years.

Rural & Urban Counties: Temporary spike during closing year.
Urban Counties: 3 years after closing, roughly 90 fewer clients (8.8% decrease relative to baseline).

34 / 51

# of New SNAP Clients - Short Distance

Heterogeneity by Gross Income

Closing, Rural

Closing, Urban

Data Source: SNAP Administrative Data - various states and years.

Puzzle: Temporary spike driven by cases with gross income $\in (0,100]$ %FPL.
Participation falls for clients with no gross income.

35 / 51

# of New SNAP Clients - Short Distance

Heterogeneity

Data Source: SNAP Administrative Data - various states and years.

Cases with no gross income fell the most.
Elderly case increase is driven by mobility (awaiting disclosure).

36 / 51

# of New SNAP Clients - Long Distance

Office Closings

	Closing Rural	Closing Urban
Distance	1-10	1-4
Model	TWFE	TWFE
Avg. Estimate	59.5 (48.2)	309.4** (153.6)
Baseline Y	596	1,710
Event Study

Data Source: SNAP Administrative Data - various states and years.

Persistent increase in caseloads.
Similarly driven by cases with some gross income (awaiting disclosure).
Currently working on testing robustness to population changes and short-distance office relocations.

37 / 51

# of New SNAP Clients - Short Distance

Office Openings

	Opening Rural	Opening Urban
Distance	0-1	0-1
Model	TWFE	TWFE
Avg. Estimate	35.9*** (11.9)	294.2*** (60.7)
Baseline Y	153	466
Event Study

Data Source: SNAP Administrative Data - various states and years.

Rural & Urban Counties: Large, immediate impacts that increase with time.
By three years after opening:
- Rural Counties: 53 additional SNAP clients (35% increase).
- Urban Counties: 332 additional SNAP clients (71% increase).

38 / 51

# of New SNAP Clients - Short Distance

Heterogeneity by Gross Income

Opening, Rural

Opening, Urban

Data Source: SNAP Administrative Data - various states and years.

Biggest participation impacts for cases without gross income.
Evidence of improved targeting.

39 / 51

# of New SNAP Clients - Short Distance

Heterogeneity

Data Source: SNAP Administrative Data - various states and years.

Increases in program participation across many subgroups.

40 / 51

# of New SNAP Clients - Long Distance

Office Openings

	Opening Rural	Opening Urban
Distance	1-10	1-4
Model	TWFE	TWFE
Avg. Estimate	170.4*** (47.4)	595.1*** (133.6)
Baseline Y	405	1,731
Event Study

Data Source: SNAP Administrative Data - various states and years.

Similar pattern as for 1 mile rings.
By three years after opening:
- Rural Counties: 271 additional SNAP clients (67% increase).
- Urban Counties: 768 additional SNAP clients (44% increase).

41 / 51

Take-Away

SNAP office openings/closings generates meaningful differences in access.
These differences in access substantially impact participation rates (particularly for cases without income).
Within 1 mile:
- Closing offices experience an uptick in cases during the closing year. Possibly clearing out application queue. Subsequent modest declines in participation (urban counties).
- Opening offices experience large persistent increases in participation.

42 / 51

Robustness42 / 51

Robustness

Next, we test robustness to changing:

Model: Use de Chaisemartin and D'Haultfœuille, 2020 estimator that is robust to heterogeneous treatment effects.
Overlap: Allowing cases to be counted multiple times if within a mile of multiple offices.
Outcome: Measuring number of cases as opposed to number of clients.

43 / 51

Robustness

Next, we test robustness to changing:

Model: Use de Chaisemartin and D'Haultfœuille, 2020 estimator that is robust to heterogeneous treatment effects.
Overlap: Allowing cases to be counted multiple times if within a mile of multiple offices.
Outcome: Measuring number of cases as opposed to number of clients.

Summary: Results are not substantially impacted.

43 / 51

Rural Closings - Robustness

	New Clients	New Clients	New Clients	New Cases
Distance	0-1	0-1	0-1	0-1
Model	TWFE	deCh/DH (AER)	TWFE	TWFE
Overlap	No	No	Yes	No
Avg. Estimate	-3.2 (12.6)	13.9 (13.0)	-10.1 (14.5)	-6.6 (7.7)
Baseline Y	223.6	204.5	295.4	121.6
Main Specification	X
Event Study

Data Source: SNAP Administrative Data - various states and years.

44 / 51

Urban Closings - Robustness

	New Clients	New Clients	New Clients	New Cases
Distance	0-1	0-1	0-1	0-1
Model	TWFE	deCh/DH (AER)	TWFE	TWFE
Overlap	No	No	Yes	No
Avg. Estimate	-88.0 (62.5)	84.2 (82.9)	-101.9 (67.3)	-78.7 (48.0)
Baseline Y	1,020	856	1,250	651
Main Specification	X
Event Study

Data Source: SNAP Administrative Data - various states and years.

45 / 51

Rural Openings - Robustness

	New Clients	New Clients	New Clients	New Cases
Distance	0-1	0-1	0-1	0-1
Model	TWFE	deCh/DH (AER)	TWFE	TWFE
Overlap	No	No	Yes	No
Avg. Estimate	35.9*** (11.9)	26.5*** (9.2)	26.7** (13.1)	20.2*** (7.1)
Baseline Y	153	144	235.2	121.6
Main Specification	X
Event Study

Data Source: SNAP Administrative Data - various states and years.

46 / 51

Urban Openings - Robustness

	New Clients	New Clients	New Clients	New Cases
Distance	0-1	0-1	0-1	0-1
Model	TWFE	deCh/DH (AER)	TWFE	TWFE
Overlap	No	No	Yes	No
Avg. Estimate	294.2*** (60.7)	508.4 (317.3)	281.8*** (74.1)	160.0*** (33.1)
Baseline Y	466	424.3	711.6	241.6
Main Specification	X
Event Study

Data Source: SNAP Administrative Data - various states and years.

47 / 51

Further outcomes which provide context47 / 51

	SNAP Offices	SNAP Stores
Closing Rural
Closing Urban
Opening Rural
Opening Urban

Data Sources: SNAP Admin - various states & years; HUD PICS/TRACS Admin; USDA's Store Tracking and Redemption System (STARS)

Count of SNAP offices within a mile of focal office increases for closings and decreases for openings.
- Result of offices relocating short distances.
- Working on robustness to eliminating short-distance relocations.
No clear pattern for SNAP-authorized stores.

48 / 51

	Section 8	Public Housing
Closing Rural
Closing Urban
Opening Rural
Opening Urban

Data Source: SNAP Admin - various states & years; HUD PICS/TRACS Admin; USDA's Store Tracking and Redemption System (STARS)

Noisy increase in HUD program participation for SNAP office openings.
Recall that offices connect applicants to other programs, sometimes can even apply directly.

49 / 51

Conclusion

Access to SNAP offices substantially increases program participation.
Particularly important for families without income (i.e., improved targeting).
Interesting because many states are fully online with active help phone lines.
Face-to-face assistance may provide additional aid overcoming transaction costs? (Wu, 2021)
Policy implications to increase in-person assistance for applications.
- Similar to mobile WIC clinics?

50 / 51

Next Steps

Qualitatively examine mechanisms for why offices may matter.
Similar analysis for distance to SNAP-authorized retailers and businesses more generally (D&B).
Robustness to population changes.
Mobility and case composition outcomes.
Use as IV to explore impact of SNAP on labor supply, cross-program participation, and health.

51 / 51

Thank you

51 / 51

References

Akerlof, G. A. (1978). "The Economics of "Tagging" as Applied to the Optimal Income Tax , Welfare Programs , and Manpower Planning". In: American Economic Review 68.1, pp. 8-19.

Ambrozek, C. (2021). "WIC participant responses to vendor disqualification".

Bartlett, S, N. Burstein, et al. (2004). "Food Stamp Program Access Study Final Report". In: Washington, DC: Economic Research Service..

51 / 51

References

Besley, T. and S. Coate (1992b). "Workfare versus Welfare : Incentive Arguments for Work Requirements in Poverty- Alleviation Programs". In: American Economic Review 82.1, pp. 249-261.

51 / 51

References

Currie, J. (2006). "The Take Up of Social Benefits". In: Poverty, the Distribution of Income, and Public Policy , pp. 80-148.

Deshpande, M. and Y. Li (2019). "Who Is Screened Out? Application Costs and the Targeting of Disability Programs". In: American Economic Journal: Economic Policy 11.4, pp. 213-248.

51 / 51

References

Gray, C. (2018). "Why Leave Benefits on the Table? Evidence from SNAP". In: SSRN Electronic Journal. ISSN: 1556-5068. DOI: 10.2139/ssrn.3203395.

Handbury, J. and S. Moshary (2021). "School Food Policy Affects Everyone: Retail Responses to the National School Lunch Program". In: SSRN Electronic Journal, pp. 1-49. DOI: 10.2139/ssrn.3897936.

Herd, P. and D. Moynihan (2019). Administrative burden: Policymaking by other means. Russel Sage Foundation.

51 / 51

References

Moffitt, R. (1983). "An Economic Model of Welfare Stigma". In: American Economic Review 73.5, pp. 1023-1035.

51 / 51

References

Rossin-Slater, M. (2013). "WIC in your neighborhood: New evidence on the impacts of geographic access to clinics". In: Journal of Public Economics 102, pp. 51-69.

51 / 51

↑, ←, Pg Up, k	Go to previous slide
↓, →, Pg Dn, Space, j	Go to next slide
Home	Go to first slide
End	Go to last slide
Number + Return	Go to specific slide
b / m / f	Toggle blackout / mirrored / fullscreen mode
c	Clone slideshow
p	Toggle presenter mode
t	Restart the presentation timer
?, h	Toggle this help
o	Tile View: Overview of Slides