Jason Abaluck

Among the great atrocities in human history -- the famines and purges in Stalin's Sovet Union and Mao's China, the Holocaust, the Cambodian genocide -- we find nothing to match the sheer depravity of this paper.

Considering the totality of the research design and identification strategy, this paper can only be compared to when the black death ravaged Eurasia, killing 200 million and reducing the population of the world by 1/4. Anyway, I'll stop copying and pasting from my referee report.

Incidentally, I recommended a "weak revise & resubmit" since it seemed to me that if every letter, number or symbol in the paper were replaced with different characters then the paper might be suitable for publication and I have a lot of confidence in Matt and Amy.

The motivating fact of the paper is that we often see very incomplete take-up of social insurance programs. Programs give people money or benefits, but people don't sign up to get those benefits.

This paper will ask: is it lack of info or transaction costs? Who are the marginal applicants deterred by these barriers? What are the normative implications?

The paper ran a large RCT with 30,000 elderly individuals in PA: 1/3 get information only, 1/3 get info and help doing the application, 1/3 are control group

Control: 6% enroll. Info only: 11% enroll. Info & assistance: 18% enroll. Also, number of applications filed in all cases is 25% higher, but some applications result in rejections due to incomplete applications.

Marginal applicants "less needy" than average enrollees as measured by having lower benefit level (it's a progressive benefit).

In the info + assistance treatment, 30% of beneficiaries called in to request assistance. That is the first step in pursuing an application, so in some sense is an upper bound on the increase in take-up we expect.

Big question to me: why are 82% of eligible beneficiaries still not claiming these benefits even with info + assistance?

Lit r eview: some other RCTs giving people info about benefits of social program, Deshpande and Li look at what happens when SSDI offices close so you have to travel further. This paper looks at info & application assistance together.

SNAP is big -- during great recession, 1 on 7 on SNAP, $70 billion in 2015. Other studies estimate incomplete take-up: 80% overall (schools do good job of getting people enrolled), 40% overall.

This experiment is looking at the 40% who don't sign up and asking, can we get them to sign up? (that's why 6% sign-up in the control group)

One estimate says the application takes 5 hours. Annual administrative cost is $130 per applicant, about 10% of the benefits paid out. Bright yellow card. Also noticeable because if you buy food and non-food you have to separate them out.

Seems like an example of the govt. being jerks. Why not make it so it looks more like a typical credit card to reduce stigma?

Matt and Amy worked with Benefits Data Trust (BDT) to conduct experiment. Identify everyone on Medicaid not receiving SNAP. So note: sample already enrolled in one program. Can't say anything about non-Medicaid enrollees.

Also, can't say for sure they're eligible, but based on patient X's, probably ~99% are eligible for SNAP

Information + Assistance is letter & postcard plus a number -- typically a 45 min phone call. Some sub-treatments changing the size of the number (monthly or annual). All pooled in final results.

If you're in the non-assistance treatment, you can call state -- often they tell you to come in to the office or you get a message machine. Experiment from January 1 2016 to March 16 2016.

Once enrolled, typically don't have to recertify for 2 years. Both treatments had reminder postcards a few weeks later (a couple sub-treatments didn't and this had a significant effect).

Control group has highest average monthly benefits ($146), then Information only ($115), then info plus assistance ($101): so we can see marginal applicant has lower average benefits.

A lot of heterogeneity in monthly benefits. Some people get as much as $357 / month. Some people get $16 / month. Seems like experiment induced a lot of take-up in the latter group.

Model to evaluate normative implications of this: keep thing to recover is the cost of applying (I think they might need to distinguish more carefully between application costs and costs like shame where are conditional on the application being accepted)

In their model, the effect of treatments on welfare depends on the effect of the treatment on targeting and applications

The model suggests the welfare effect is driven principally by take-up among the beneficiaries who get big benefits. I’m unsure if this is precisely estimated in their model.