Deshpande and Dizon-Ross 2022 — The (Lack of) Anticipatory Effects of the Social Safety Net on Human Capital Investment

Summary

A randomized controlled trial (RCT) with ~6,000 parents of Supplemental Security Income (SSI)-recipient children ages 14–17 tests whether providing accurate information about the age-18 SSI redetermination — in which ~40% of SSI children lose benefits under adult eligibility rules — causes parents to increase investment in their children's human capital. The treatment delivers an informational video that raises parental beliefs about SSI removal by +20 percentage points (pp) (F-statistic 94). Despite this strong first stage, the treatment effect on human capital investment (enrollment in education, job training, or work) is −0.2 pp against a control mean of 28% — statistically zero and ruling out effects larger than +1.5 pp. The null result holds in all 34 pre-specified subgroups and directly contradicts both expert elicitations (+14 pp predicted) and calibrated structural model predictions (+11%).

Key Claims

• RCT design: N ≈ 6,000 parents of SSI children ages 14–17; treatment = 15-minute video detailing age-18 redetermination process and removal probability; control = no video; all parents recruited from Social Security Administration (SSA) field offices and community organizations.
• Strong first stage: Treatment raises parental beliefs about SSI removal by +20 pp (from 20% to 40%, vs. actual predicted removal rate ~70%); F-statistic = 94; effect holds across all subgroups.
• Null main result: Treatment effect on human capital investment = −0.2 pp (standard error [SE]: 0.9 pp); control mean = 28%; 95% confidence interval (CI) rules out effects larger than +1.5 pp; p = 0.85.
• Expert prediction failure: Survey of 46 disability and social insurance economists predicted average +14 pp increase; structural model of Heathcote, Storesletten, and Violante predicts +11%; both rejected by the null finding.
• Null holds across all subgroups: 34 pre-specified subgroup analyses (by child's disability type, parent's income, parent's education, child's age, SSI duration, parental employment status) — no subgroup shows a statistically significant positive effect.
• Mechanism 1 — Parental labor supply substitution: 49% of parents in the treatment group report they would work more themselves if their child loses SSI — substituting own labor supply for child human capital investment. When parents learn SSI will be removed, they plan to fill the income gap themselves rather than build the child's earnings capacity.
• Mechanism 2 — Wealth effect depresses college plans: Among parents already aware of SSI removal risk, learning confirms the threat; college enrollment plans fall −5.75 pp in treatment (conditional on prior awareness), consistent with an income shock reducing aspirations.
• Mechanism 3 — Investment constraints: 89% of parents report they are already investing in their child's human capital at the maximum level they can afford given current resources — leaving no margin to increase in response to the information shock.
• Mechanism 4 — Non-financial goals: Qualitative responses indicate many parents prioritize child health and happiness over economic preparation; SSI removal may not increase economic-investment urgency if non-economic goals dominate.
• Age-18 removal facts: ~40% of SSI children are removed under adult eligibility criteria overall; for children with behavioral/attention-deficit-hyperactivity-disorder (ADHD) diagnoses, removal rate is ~70%; parental prior beliefs about removal average only 20% — a severe information gap relative to actual rates.
• Control mean heterogeneity: Human capital investment rates vary widely by child's condition: ~38% for physical/sensory impairments; ~22% for mental/intellectual disabilities — the latter group likely faces greater actual employment barriers.

Concepts Introduced or Extended

• Dynamic Discouragement
• SSI Children's Program
• DI Application Costs and Take-Up

Entities Mentioned

• Manasi Deshpande
• Rebecca Dizon-Ross
• Social Security Administration

Quotes

"We find no evidence that anticipating future loss of SSI benefits causes parents to invest in their children's human capital."

"Our main finding is a precise null: the information treatment has no effect on human capital investment, which rules out effects as small as 1.5 percentage points."

"The most common reason parents report for not increasing investment is that they are already investing at the maximum level they can — 89% of parents in the control group report this."

"Information alone may not change behavior if structural constraints bind: parents who cannot afford to invest more cannot do so even when they learn the stakes have risen."

My Take

This paper matters primarily for what it rules out. The "dynamic discouragement" hypothesis — that safety net generosity depresses parental investment in children's human capital — has been the theoretical foundation for concerns that programs like SSI trap families in dependence across generations. Deshpande and Dizon-Ross design a nearly ideal test of this mechanism: an RCT with a well-powered first stage, exact-matching to SSA records, and 34 pre-specified subgroup analyses. The null result is robust and precisely estimated.

Three implications for this wiki's focus:

1. The null result is not an absence of discouragement — it is evidence of constraints. The 89% already-at-maximum finding suggests that even if parents wanted to invest more in response to SSI removal risk, they cannot. This reframes the dynamic discouragement hypothesis: the mechanism may exist in theory but is inoperative in the low-income SSI population because liquidity and time constraints bind first. This has direct implications for DI Application Costs and Take-Up — barriers to investment parallel barriers to application.

2. The substitution-onto-own-labor-supply finding is underappreciated. If 49% of parents plan to work more themselves when SSI is removed rather than invest in the child's human capital, then SSI removal does not produce the human capital investment the program's critics hoped for. Instead it produces parental labor supply increases — which are separately documented in Deshpande 2016 — The Effect of Disability Payments on Household Earnings and Income Evidence from the SSI Childrens Program (the elasticity ≥ 1 finding). The two papers together suggest a consistent behavioral response: parents treat themselves as the residual labor supplier, not the child.

3. Expert elicitation failure is methodologically significant. A survey of 46 disability economists predicted +14 pp — 70× the actual treatment effect. This is not a small forecasting error; it reflects how poorly even domain experts calibrate behavioral responses in complex social programs. The structural Heathcote et al. model was also off by an order of magnitude. This is a case study in why reduced-form RCT evidence cannot be replaced by expert judgment or theoretical calibrations, especially in populations with severe resource constraints.

Limitations: The study measures short-run human capital investment as self-reported enrollment in education/training/work. It cannot measure long-run outcomes (child earnings at age 20+). The sample is drawn from SSA field offices and community organizations — likely more informed and engaged than the typical SSI family, potentially understating information gaps in the general population. The treatment informs about removal probability but does not provide resources to invest — no vouchers, no income support — so the constraint mechanism is directly relevant.