A Randomized Controlled Trial on the Provision of Financial and Social Capital to Low-Income Households in the United States

In this large-scale, longitudinal randomized controlled trial, we aim to understand the extent to which the negative effects of poverty can be ameliorated through financial and social capital support. Approximately 1,800 low-income households in the Boston, MA area will be randomly assigned to either (a) receive a “no-strings-attached” unconditional cash transfer of $500 per month (totaling $9,000 over 18 months), (b) be encouraged to develop their social capital by creating and engaging with a “peer group” of four to eight participating households, (c) receive both the financial assistance and social capital interventions, or (d) not receive any intervention. We will track the households' financial, psychological, physical health, and family well-being outcomes at regular intervals over the 18 month trial. We predict that the provision of financial and social capital alone will each have a significant and positive effect on outcomes, and that the effect of combining financial and social capital will be larger than the effect of providing either financial or social capital alone. We propose that the positive effects of combining the two forms of capital will occur in part because financial and social capital may address different problems of low-income households, and in part because financial and social capital may interact to mutually enhance one another, amplifying the treatment intensity of each.

2021-06-29

2023-05-31

We will assess outcome measures on four main dimensions: financial well-being, psychological well-being, physical health, and family well-being. For each, we construct an index, described in detail in the attached preregistration manuscript. All survey questions are in the attached appendix.

To be eligible for the trial, households must reside in Cambridge or Boston, MA, and have a total earned household income between 0% and 200% of the federal poverty line. Participants in the Cash-only (hereafter, “C”) treatment arm will receive $9,000 over 18 months in fully unconditional payments, paid out as $500 per month. For a household of four living at the poverty line, these payments will raise participants’ annual earned income by 22%. That is, participants in this arm who earn near the poverty line will receive unconditional payments equal to almost three months of income per year. The Social-only (“S”) arm will be placed into peer groups with three to seven other participants in this treatment arm and will be encouraged to interact with their peer groupmates through monthly meetings. They will also be given access to an online platform that aims to help people develop and strengthen social ties, in part by decreasing barriers to help-seeking and -giving. Finally, they will be given a small unconditional payment of $20 per month. Those in the Cash+Social (“CS”) arm will receive the same financial benefits as the Cash treatment arm and the same social capital benefits as the Social treatment arm. Finally, those in the No Treatment Control (“NTC”) arm will receive only the small unconditional payment ($20 per month) and will not be given any opportunity to participate in any of the social interventions.

We will examine the effects of these treatments on a wide range of outcomes, leveraging panel data and a rich and complementary array of datasets. All participants will be asked to complete quarterly surveys, which will ask about financial, psychological, health, and family well-being outcomes. We will also observe transaction-level financial outcomes from checking and savings accounts, including savings stocks, direct deposits of earned income, and late fees charged. Finally, for participants who are recipients of any welfare program administered by our partner state agency, or who have received benefits in the five years before the trial begins, we will also observe administrative data from that agency. This data will include a comprehensive past and present financial profile, welfare applications and receipts, Supplemental Nutrition Assistance Program benefit usage, and employment outcomes, enabling us to verify and complement many survey measures.

The first and most fundamental aim of this study is to quantify the effects of the treatment arms (C, S, and CS), relative to the NTC arm, on our pre-registered outcomes. Specifically, we seek to measure the effects of our treatments on financial (e.g., subjective financial well-being, number of bills left unpaid), psychological (e.g., sense of agency, depression), health (e.g., sleep quality, nutritional quality), and family well-being (e.g., parenting quality, quality of relations with one’s partner) outcomes. Second, we will aim to identify the underlying mechanisms that may drive our treatment effects. To this end, we will examine the roles of both intrapersonal mechanisms (e.g., cognitive capacity, time preferences, risk preferences), as well as interpersonal mechanisms (e.g., sense of support from others, comfort with asking for help, extent to which others are helping one achieve one’s goals). Finally, our study design gives us the opportunity to test whether the treatments work particularly well for specific subpopulations.

Please see the attached preregistration manuscript for more details.

Not available

Randomization will be done in an office using a computer program.

To accommodate the fact that two treatment arms (S and CS) have peer groups while the other two arms (C and NTC) do not, we will use a two-stage randomization process. In the first stage of randomization, we will randomize the enrolled households into either the “need a peer group” meta-treatment (which will ultimately become the S and CS arms) or the “do not need a peer group” meta-treatment (which will ultimately become the NTC and C arms).

We will then match households in the “need a peer group” meta-treatment into peer groups of four to eight households based on the language(s) they speak and their geographic location. In particular, we will first match households based on their preferred language, grouping them into either English- or Spanish-speaking pools. Next, within each language pool, we will use a set of prespecified rules that consider public transit maps, natural barriers (e.g., rivers), and cultural differences to generate eight distinct “geographic regions.” Finally, within each language pool and geographic region, we will match households in such a way as to minimize the geographic distance between the home addresses of each member of the peer group. In situations in which participants do not provide usable address data for their household, we will use the center of the zip code they provide (which will be required and vetted to be a valid Boston or Cambridge zip code). This matching code will be made publicly available.

Once households have been matched into peer groups, the peer groups will be randomized to the S or CS treatment arm. At the same time, households that were randomized into the “do not need a peer group” meta-treatment will be randomized to the NTC or C treatment arm.

For the No Treatment Control and Cash groups, randomization occurs at the household level. For the Social and Cash+Social groups, randomization occurs at the peer group level.

Yes

We plan to randomize households in approximately these proportions:

No Treatment Control (NTC): 668 households (no clustering).
Cash-only (C): 240 households (no clustering).
Social-only (S): 111 peer groups (of ~6 households each).
Cash+Social (CS): 40 peer groups (of ~6 households each).

We aim for our final sample to be 1,500 households. To account for attrition, we intend on recruiting approximately 1,816 households.

We aim for our final sample to be 216 households in each of the cash arms (C and CS) and 534 households in each of the non-cash arms (NTC and S), for a total of 1,500 households at endline. Furthermore, we assume that 10% of the cash arms and 20% of the non-cash arms will leave over the course of the trial. Therefore, to ensure our final sample is approximately 1,500 households, we plan to recruit approximately 1,816 households at the start of the trial: 240 in each of the cash arms and 668 in each of the non-cash arms.

To calculate the detectable difference given this sample size, we make the following assumptions. We conservatively assume an intraclass correlation of 0.05 and a correlation between baseline measurements and outcomes of 0.10. We further conservatively assume that all peer groups in the S and CS arms will be eight households, the maximum number allowed. To adjust for multiple hypotheses, we use a conservative Bonferroni correction, dividing a standard 0.05 significance level by four—the total number of outcomes we study, described in further detail below. The sample sizes used in these calculations reflect our predicted sample sizes at the end of the eighteenth month, after approximately 17% of the sample has left the trial. Thus, these estimates are a lower bound on our power earlier in the trial, when presumably fewer households will have left the trial. With these assumptions and a target rate of 90% power, we calculate standardized effect sizes across outcomes. Because our outcomes are in standardized effect sizes, this calculation applies to each of our key outcomes (financial, psychological, health, and family well-being). Because of the longitudinal nature of our study, we can leverage the multiple endline measurements to increase our power (or, equivalently, to decrease our minimum detectable effect [MDE] size while holding our power level constant). Specifically, we can treat each quarter after the baseline as an endline measurement, giving us six endline measurements. We assume a correlation of 0.50 between follow-up measures. At 90% statistical power, we find that we are able to detect a MDE between 0.14 and 0.23 for pairwise comparisons between treatment arms. Specifically, we estimate an MDE of 0.15 when comparing NTC to C; 0.14 when comparing NTC to S; 0.18 when comparing NTC to CS; 0.23 when comparing C to CS; 0.17 when comparing S to CS; and 0.17 when comparing C to S.