Disability Insurance (DI) application costs are the time, effort, travel, and administrative friction imposed on potential applicants by the process of filing for disability benefits. These costs are not intrinsic to the disability standard itself but are artifacts of program delivery infrastructure — office locations, staffing, wait times, online access, and form complexity. Take-up refers to the fraction of eligible individuals who actually apply and receive benefits. Application costs reduce take-up; the policy question is whether this reduction is desirable (by screening out low-severity applicants) or undesirable (by deterring high-severity, approved-quality applicants).
The classical public economics argument for using hassles and application costs as a targeting device — due to Nichols and Zeckhauser (1982) — holds that applying for a means-tested or disability program is more costly for individuals with higher opportunity costs (higher wages, better labor market attachment), who are also more likely to be low-severity applicants who do not genuinely need the benefit. Under this logic, requiring an in-person visit, a stack of medical records, or a long wait functions as a self-selection mechanism: it screens out the least deserving while allowing the most deserving (those with no better alternative) to persist through the process.
The implication would be that reducing application costs — through online filing, more field offices, shorter wait times — loosens the screen and worsens targeting by enabling low-severity applicants to apply with less friction.
Deshpande and Li (2019) test the Nichols-Zeckhauser prediction directly, using 118 Social Security Administration (SSA) field office closings between 2000 and 2014 as quasi-random variation in application costs. The key finding: office closings reduce applications by 10% but reduce recipients by 16%. Because recipients fall more than applications, the composition of applicants worsens — those deterred by the added friction are disproportionately likely to have been approved had they applied. The closings screen out approved-quality applicants, not denied-quality ones.
This directly falsifies the Nichols-Zeckhauser prediction for the DI/Supplemental Security Income (SSI) context:
| Nichols-Zeckhauser Prediction | Deshpande-Li Finding |
|---|---|
| Higher costs deter low-severity / low-need applicants | Higher costs deter medium-severity, approved-quality applicants |
| Targeting improves: fewer undeserving recipients | Targeting worsens: fewer deserving recipients relative to total applications |
| Hassles function as a beneficial self-selection device | Hassles impose dead-weight losses without targeting benefit |
The mechanism of the Deshpande-Li effect is not primarily travel distance. Their channel decomposition:
| Channel | Share of decline |
|---|---|
| Congestion at neighboring offices (processing delays, wait times) | 54% |
| Office switching friction (unfamiliarity with new office, relationship disruption) | 42% |
| Increased travel distance | 4% |
The dominant channel — congestion — is indirect and counter-intuitive: closing an office does not directly raise costs for applicants in the closed office's zip code by requiring them to drive further. It raises costs by routing their applications through already-busy neighboring offices, generating queuing delays, appointment backlogs, and processing slowdowns that deter potential applicants before they even begin.
This implies that the relevant application cost for policy is not the geographic accessibility of the nearest office but the throughput capacity of the local office network. Marginal closings that push cases into congested networks impose larger deterrence effects than closings in low-density areas with excess capacity at nearby offices.
Application costs do not deter uniformly across the potential applicant pool:
The heterogeneity pattern is consistent with a model where application costs screen on navigational capability and bureaucratic experience rather than on disability severity or work capacity.
Administrative simplification is not a program loosening. Reducing application costs (better online filing, more office hours, more caseworkers) corrects an unintentional screen that falls on the wrong population — it does not open the program to low-severity applicants.
Office placement matters for equity. If closings disproportionately deter lower-education applicants (who have more difficulty navigating administrative alternatives), office placement decisions have distributional consequences independent of benefit eligibility rules.
Congestion is the key variable to manage. SSA should monitor average processing capacity at offices when making closure or staffing decisions; a closing that pushes cases into a congested neighboring office has much larger deterrence effects than one with surplus nearby capacity.
Optimal closure decisions. Deshpande and Li show that SSA's actual closings had net social costs more than double the lowest-cost counterfactual closings. Applying a cost-minimization algorithm — targeting offices with fewer potential applicants and adequate nearby capacity — could achieve the same savings at much lower welfare cost.
This paper reframes the debate about program access. Prior to Deshpande and Li, the standard framing was: "How strict should the disability standard be?" Their contribution reframes it as: "What fraction of eligible individuals successfully navigate the application process, and who is lost in that gap?" The answer has implications not just for field office policy but for how to interpret program statistics — measured application and receipt rates are endogenous to administrative capacity, not just to disability prevalence.
A complementary finding from the SSI children's literature extends the take-up picture from individuals to households. Deshpande (2016), studying families of children removed from SSI via medical review, finds that 65% of SSI children have a parent or sibling who ever applies for disability; 15% have a family member apply within 60 days of the child's own application. When a child is removed from SSI, family member disability applications fall by 50–100% — but family member disability receipt does not fall, because the deterred applications come from marginal applicants who would not have been approved anyway.
This pattern exactly mirrors Deshpande-Li (2019): application costs or loss of familiarity with the program deter the marginal, likely-to-be-denied applicant — not the inframarginal, clearly eligible one. Combined, these two papers suggest the mechanism is general: any friction or negative shock to program familiarity reduces applications more at the margin (low-probability cases) than at the inframarginal end (high-probability cases), leaving the recipient pool unchanged or slightly worsened in composition.
The clustering finding also reframes the decision to apply as a household decision, not purely individual. Household income shocks — job loss, learning about program availability — trigger simultaneous multi-member applications. Application rate trends therefore reflect household economic conditions as well as individual health.
Asymmetric entry vs. exit responses. The Deshpande (2016) examiner-leniency instrumental variables (IV) estimates the entry-margin earnings elasticity for parents (); the medical-review exit-margin IV produces an elasticity . Households respond much more strongly to losing benefits than to gaining them — consistent with loss aversion and habit formation. This asymmetry implies that application costs on the way in to a program are less deterring (people still apply when potential gains are large) than benefit removal on the way out is mobilizing (people substitute aggressively when a familiar income stream is cut). The policy implication: reducing application barriers will raise take-up more at the margin than simply comparing entry and exit elasticities would predict.
The contemporary null result on Administrative Law Judge (ALJ) award rates deterring applications (Deshpande et al. 2025) must be read against the historical expansion of the ALJ stage as a successful appeals pathway. Autor and Duggan (2006) document the scale of this expansion through 2002:
| Measure | ca. 1977–1979 | ca. 2000–2002 |
|---|---|---|
| Attorney representation of claimants | 37% | 70% |
| ALJ awards as share of all DI awards | 12% | 27% |
| SSA ALJ appeals win rate |
The SSA lacks its own legal representation at ALJ hearings — claimants are represented by contingency-fee attorneys while SSA relies on case records alone. This asymmetric advocacy explains why SSA loses approximately 75% of contested ALJ appeals. The result is that the ALJ stage functionally reverses a large share of Disability Determination Services (DDS) initial denials: by 2002, more than one in four DI awards was granted at the ALJ level rather than by DDS examiners.
This historical pattern is the background against which the Deshpande et al. (2025) ALJ reform must be evaluated: the ALJ stage expanded substantially over the prior three decades, and the 2017–2019 reform sought to constrain it. The null result on application deterrence (applicants did not reduce filing in response to lower ALJ award rates) is consistent with applicants being aware of DDS-stage prospects but inattentive to ALJ-stage changes that are temporally remote from the initial filing decision.
A distinct but related question is whether potential applicants respond to changes in award probabilities rather than administrative friction. If applicants rationally weigh expected award rates in their application decision, then policies that reduce award rates should deter applications — a form of cost-via-discouragement rather than cost-via-friction.
Deshpande, Kellogg, Mogstad, and Tseng (2025) test this hypothesis using the 2017–2019 ALJ reform, which sharply reduced ALJ award rates through stricter documentation requirements and adjudicator quality review pressure. The finding: applications did not respond to the ALJ award rate decline. The elasticity of DI applications to ALJ award rates is approximately zero.
This null result has two interpretations:
Either way, the implication is that tightening adjudication at the appeal stage does not function as an application deterrent. Policies aimed at reducing DI applications must operate at the initial stage or upstream (labor market conditions, field office access), not through ALJ standards.
This complements the Deshpande-Li (2019) finding: both types of application barriers — physical access costs (field office closures) and procedural stringency at the appeal stage (ALJ reform) — leave the entry margin largely unaffected for the population that is already committed to applying. The action happens earlier: at the decision to file at all, driven by labor market conditions and initial-stage expectations.
The discussion of application costs has focused on who is deterred. Deshpande, Gross, and Su (2021) add a complementary dimension: when do applicants file, and what does timing reveal about their financial state?
Using the universe of SSA disability applicants (2000–2014) linked to nationwide bankruptcy, foreclosure, and eviction records, they document that disability applicants' rates of adverse financial events rise steadily in the months before application, peak at the application date, and fall afterward — relative to both the general population and the applicants' own lifetime profile. This "Ashenfelter's peak" pattern in financial distress implies:
The welfare implication is direct: shortening the statutory waiting period would avert additional financial distress at peak need, though it must be weighed against administrative costs and potential compositional changes in the applicant pool.
Deshpande and Lockwood (2022) clarify why the Substantial Gainful Activity (SGA) earnings limit is a beneficial screen for DI targeting, in contrast to the harmful screen created by field office closings.
| Feature | SGA Earnings Limit | Field Office Closings |
|---|---|---|
| Operates on | Work capacity: screens out workers with viable earnings alternatives | Administrative friction: screens out applicants unable to navigate bureaucracy |
| Who is deterred | Higher-earning individuals with meaningful outside options | Medium-severity, high-need applicants deterred by congestion |
| Effect on targeting | Improves: applicant pool is more financially vulnerable | Worsens: applicant pool includes more low-value, denied-quality cases |
| Welfare effect | Positive (SGA generates 68% of U.S. Disability Program (USDP)'s insurance value alone) | Negative (net social cost approximately $1.2 billion, Deshpande and Li 2019) |
The distinction is what each screen selects on. SGA screens on earnings capacity — the dimension most directly predictive of financial need from disability. Distance and congestion screen on administrative navigation ability — a dimension correlated with education and resources but not with financial need conditional on health impairment.
A simulation replacing all medical screening with only the SGA earnings limit (Earnings-Test-DI) produces a $5,900 surplus — recovering 68% of actual USDP's $8,700 surplus. The SGA limit is therefore the load-bearing element of DI's insurance value, not merely a secondary formality. Policy reforms that substantially weaken or remove the SGA limit risk admitting a higher-earning population for whom the insurance value is low, while reforms that preserve SGA but simplify medical screening can maintain most of DI's value at lower administrative cost. See Nonhealth Risk and DI Insurance Value.
A distinct take-up phenomenon concerns racial and ethnic differences in application propensity — the rate at which groups apply for DI relative to their underlying disability prevalence. Mitchell and Thompson (2025), using 2015 American Community Survey (ACS) data linked to SSA administrative records, document striking disproportionalities:
These disproportionalities have a direct implication for understanding raw allowance rate gaps. When a group applies at higher rates relative to its underlying disability burden, the margin of the applicant pool is lower-quality — it includes more individuals who are near or below the medical eligibility threshold. Under purely race-neutral adjudication, this produces a lower allowance rate for that group. The raw NH Black allowance rate deficit (−9.5 percentage points [pp] for males) is almost entirely explained by this compositional channel rather than by disparate treatment in adjudication.
The proposed mechanism: progressive benefit replacement rates. DI's benefit formula is progressive — it replaces a higher fraction of pre-disability earnings for low-wage workers. Because Black workers are disproportionately concentrated in low-wage occupations, DI's actuarial value is relatively higher for them, making application more attractive even at lower disability severity. The combination of higher relative benefit replacement and lower wages may explain a substantial share of the Black over-application pattern.
The access disparity this reveals. The application propensity paradox matters regardless of what happens at adjudication. Even if DI adjudication were perfectly race-neutral — which Mitchell and Thompson largely find for DDS-stage decisions — the high application rate relative to disability prevalence for Black applicants is an upstream disparity. It suggests that some combination of financial need, benefit attractiveness, information networks, and behavioral responses to labor market conditions is driving Black applicants disproportionately into the DI pipeline. Reducing this disproportionality — or ensuring adequate support for those who apply and are denied — requires addressing conditions upstream of the determination process itself.
Compare with the Deshpande-Li (2019) finding: field office closings deter approved-quality applicants, which is a barrier into the pipeline. The Mitchell-Thompson application propensity finding identifies the opposite pressure for Black applicants: over-entry into the pipeline relative to disability severity. Both are access problems, but of opposite signs and requiring different remedies.
Two mechanisms connecting eligibility stringency to take-up emerge from Staubli's (2011) Austrian natural experiment that extend the Deshpande-Li framework.
Self-screening (Parsons 1991): Stricter criteria reduce not only awards but also applications. When Austria raised the relaxed-eligibility age from 55 → 57, DI enrollment fell through both channels: fewer applications (workers who inferred they no longer qualified near age 55 did not bother filing) and fewer approvals among those who did apply. This is the Parsons (1991) prediction: under stricter rules, the marginal would-be applicant screens herself out before incurring application costs. The Deshpande-Li (2019) result is the mirror image: administrative friction screens out applicants who do qualify, not those who do not. Together, the two papers bracket the application margin — stringency and friction are distinct screens that select on different dimensions (expected eligibility vs. bureaucratic navigation capacity).
Postponement rather than deterrence: Staubli's transition analysis reveals that DI enrollment at ages 55–56 drops ~10 pp immediately post-reform but fully recovers at ages 58–59. The reform did not permanently deter workers from the DI pipeline — it delayed their entry to the new threshold age of 57. This postponement pattern has a direct parallel to the "denial roundabout" in the U.S. literature (French and Song 2014: of ALJ-denied applicants eventually receive benefits within 10 years). Both patterns suggest that reforms targeting eligibility standards at a particular age or adjudication stage shift enrollment in time or across programs rather than achieving permanent exclusion.
The policy implication reinforces Deshpande-Li: application barriers and eligibility stringency create dead-weight processing costs without achieving the targeting improvements assumed by the Nichols-Zeckhauser model. In Austria's case, most of the cost fell on workers who ultimately received DI anyway — they simply waited longer and transited through Unemployment Insurance (UI) and sickness insurance in the interim.
Distinct from the pre-filing friction documented by Deshpande and Li (2019), Benítez-Silva et al. (1999) quantify a post-denial cost: the 10–15 month delay imposed on applicants who appeal an initial DDS rejection. Using Health and Retirement Study (HRS) data, they find that while the appeal option is highly valuable on average — raising the effective award rate from 45.9% to 72.5% — the delay it entails deters a substantial share of truly disabled applicants from exercising it.
The delay deterrence finding: 32% of rejected applicants do not appeal. Among this non-appealing group, 55% report a health limitation that prevents work (HLIMPW=1) — meaning the deterred population includes a majority who self-identify as genuinely disabled. By contrast, first-stage awardees wait a median of only 4 months; appeal awardees wait a median of 13–15 months. The delay imposes substantial opportunity costs — foregone income support, medical expense accumulation, human capital depreciation — that are particularly severe for individuals with no other income source.
This result adds a distinct type of application-stage cost to the typology: appeal delay cost operates on denied applicants (not just potential filers) and selects on the ability to bear financial uncertainty during a multi-year adjudication gap, rather than on navigational capacity or geography.
Complementarity with Deshpande-Li: Both findings indicate that application barriers in the DI system fall disproportionately on the wrong population. Deshpande and Li show that administrative friction at the filing stage deters medium-severity, approved-quality applicants. Benítez-Silva et al. show that delay costs at the appeal stage deter genuinely disabled rejectees who could win on appeal. Together, they identify a two-stage attrition problem in the DI pipeline.
The application cost literature — led by Deshpande and Li (2019) — focuses on frictions that arise before or at the point of filing: travel distance, office congestion, bureaucratic navigation complexity. A distinct cost operates after filing, during the waiting period: the administrative processing time imposed by SSA's adjudication queue.
Autor, Maestas, Mullen, and Strand (2015) show that each additional month of Social Security Disability Insurance (SSDI) processing time reduces applicants' employment by 0.44–0.52 pp and annual earnings by $133 three years post-decision, with effects persisting 6+ years. The mechanism — confirmed by a falsification test using the 5-month statutory waiting period — is human capital depreciation through labor force non-participation: skills atrophy, employer relationships weaken, and job search capital erodes during the waiting period. This is distinct from the application access cost channel; it operates on all applicants regardless of their proximity to an SSA field office.
A typology of application-stage costs:
| Cost Type | Stage | Channel | Evidence |
|---|---|---|---|
| Geographic access friction | Pre-filing | Travel distance, congestion at neighboring offices | Deshpande and Li (2019) |
| Administrative complexity | Pre-filing | Bureaucratic navigation capacity, form burden | Deshpande and Li (2019); household clustering |
| Processing time delay | Post-filing | Human capital depreciation from labor force participation (LFP) exit | Autor et al. (2015) |
| Appeal roundabout delay | Post-denial | Extended non-participation during re-application | French and Song (2014) |
The key distinction: geographic and complexity costs operate before an applicant enters the queue, and their main effect is deterrence (fewer applications). Processing time costs operate inside the queue and fall on all applicants regardless of their ultimate award decision — denied applicants bear them just as allowed applicants do. The Autor, Maestas, Mullen, and Strand (AMMMS) (2015) estimate of 2.4 pp LFP reduction for denied applicants (5.8% of their mean LFP) and 3.1 pp for allowed applicants (36.5% of their mean LFP after DI receipt) quantifies the processing time cost directly.
Policy implication: Reducing DI processing times is welfare-improving independent of the benefit receipt channel. Faster adjudication preserves employment capacity for both denied applicants (who can return to the labor market more readily) and allowed applicants (who enter DI with less skill atrophy). The gains are largest for allowed applicants whose LFP is already severely curtailed by the benefit receipt effect — the delay imposes an additional 3 pp LFP reduction on top of the 48 pp receipt effect. See Causal Effects of DI Receipt.
The application costs literature — Deshpande-Li (2019), Staubli (2011), Deshpande et al. (2025) — can be unified under the type I/type II error framework formalized by Low and Pistaferri (2020).
The Nichols-Zeckhauser (1982) targeting hypothesis is, at bottom, a claim that application hassles reduce type I errors without increasing type II errors — that friction screens on work capacity rather than on bureaucratic navigation ability. Deshpande-Li (2019) directly refutes this: field office closings increase type I errors (medium-severity, approved-quality applicants deterred) without reducing type II errors (low-severity applicants are the least deterred group).
Five policy dimensions and their error effects. Low and Pistaferri (2020) identify five axes along which DI design decisions affect the type I/II error balance:
| Dimension | Policy choice | Effect on type I | Effect on type II |
|---|---|---|---|
| Medical test stringency | Stricter standards | Increases | Decreases |
| Application process / labor market attachment | US 20-quarter requirement | Increases (coverage gap for irregular workers) | Modest decrease |
| Eligibility structure | Partial vs. all-or-nothing | Decreases (graded coverage) | Decreases (partial benefits lower stakes) |
| Benefit generosity and progressivity | Higher replacement rates | No direct effect | Increases (higher incentive to apply) |
| Reassessment/Continuing Disability Review (CDR) | More frequent review | No direct effect at entry | Decreases dynamically |
The US's distinctive combination — strict medical standard, 20-quarter labor market attachment, no partial disability, five-month waiting period — is calibrated heavily toward reducing type II errors at the cost of large type I errors. The application cost evidence (Deshpande-Li) and the coverage gap evidence (Low-Pistaferri) both point in the same direction: the current system under-covers relative to its coverage goal.
International comparison. The Netherlands model — partial disability benefits tied to the degree of residual work capacity, mandatory rehabilitation, and employer co-responsibility for the first two years of disability — substantially reduces both error types relative to the US binary standard. The cost is higher administrative complexity and reliance on employer cooperation. See Low and Pistaferri 2020 — Disability Insurance Theoretical Trade-Offs and Empirical Evidence.
The application costs literature focuses on friction that prevents eligible applicants from entering the DI pipeline. A complementary access problem runs in the opposite direction: proposals to divert potential applicants toward employment before they file implicitly assume that applicants are workforce-attached shortly before application. Contreary, Honeycutt, Stegman Bailey, and Mastrianni (2017) show this assumption fails for roughly half the applicant pool.
Using Survey of Income and Program Participation (SIPP)–SSA linked data for DI applicants ages 25–55, they find that of applicants had intermittent or no employment in the 24 months before application (Type 2), compared to who were consistently employed or had recently ceased employment (Type 1). Employer-focused early-intervention programs — which target workers before health conditions cause application — can only reach Type 1 applicants. Type 2 applicants, who are more likely to be female, Black, and enrolled in Medicaid or Supplemental Nutrition Assistance Program (SNAP), are not in the workforce and must be reached through the public programs they already participate in.
This creates a two-track intervention requirement:
The finding complements Deshpande-Li (2019) on field office closings: both papers identify distinct populations who are structurally disconnected from a particular intervention pathway — one from the administrative pipeline (deterred by office closings), the other from the workforce-based diversion pipeline (Type 2 applicants). Together, they suggest the DI access problem has two dimensions: some eligible individuals cannot navigate the entry process; others could not have been diverted at the point of application even with ideal employer-based programs.