Unconditional cash transfers to reduce homelessness?

Although I think we all now recognize the fundamental importance of social determinants of health, I don’t think emergency medicine has found a way to adequately integrate this into our care. That is a shame, considering the huge amount of effort we put into things like tetanus shots or TXA therapy that have tiny (or nonexistent) benefits. The massive benefits and limited harms of social interventions are better than almost any therapy we provide in emergency medicine. People get incredibly upset about a 20 minute delay to antibiotics in sepsis, but then turn around and discharge an unhoused individual to the street with no support. (Yes, I am as guilty as you are.) This study is very interesting because instead of addressing the downstream effects of homelessness with various social net resources, it addresses the root cause: a lack of money. There are many politically driven messages about homelessness, but I think research like this is incredibly important to ensure that our approach to fundamental medical and societal issues is science based. Unfortunately, despite some glowing reporting and some promising results, this study is actually pretty weak.

The paper

Dwyer R, Palepu A, Williams C, Daly-Grafstein D, Zhao J. Unconditional cash transfers reduce homelessness. Proc Natl Acad Sci U S A. 2023 Sep 5;120(36):e2222103120. doi: 10.1073/pnas.2222103120. Epub 2023 Aug 29. PMID: 37643214

The Methods

The primary study is a cluster-randomized trial.

There are 2 additional survey studies embedded.

Participants

Individuals experiencing homelessness from one of 22 shelters in the greater Vancouver area. 

They included adult individuals, aged 19-65, homeless for less than 2 years, with nonsevere levels of substance use, alcohol use, and mental health symptoms.

Intervention / Comparison

There were 4 groups in this study:

• Group 1 received a one time unconditional cash transfer of $7,500 Canadian dollars, plus 1 hour motivational training workshops every 3 months, plus one on one coaching.
• Group 2 received a one time unconditional cash transfer of $7,500 Canadian dollars, plus 1 hour motivational training workshops every 3 months, but no coaching.
• Group 3 received 1 hour motivational training workshops every 3 months, plus one on one coaching, but no money. 
• Group 4 received nothing. 

$7500 represented the average income assistance in the province in the year of the study. These dollars did not impact other services available to the participants, and so represent an addition to currently received assistances.

They performed a cluster randomization, so that an entire shelter either would or would not get the cash transfer, based on the logic that it would limit theft.

Outcome

There was not a predetermined primary outcome, but instead they looked for “better outcomes” across a range of survey results:

•       Fluid Intelligence
•       Executive functioning
•       Satisfaction with life
•       Positive affect
•       Negative affect
•       Subjective well-being
•       Self-efficacy

The Results

229 individuals passed their screening, but they lost contact with 114, so 115 are included in the final sample. (50 given cash and 65 not given cash). 

There were no statistical differences in any of the pre-registered outcomes for cognitive or subjective well being. “The cash transfer did not have overall impacts on employment, cognitive function, subjective well-being, alcohol use severity, education, or food security, yet there were some short-term impacts on these outcomes.”

They ran multiple analyses on the data that they designed after collecting the data. They appropriately label these as exploratory analyses.

Comparing both groups who received cash to those who didn’t, the cash recipients spent 99 fewer days homeless and 55 more days in stable housing. The cash recipients also retained more savings and spent more money on durable goods, rent, food, and transit. There was no difference between the groups on spending on ‘temptation goods’, such as drugs, alcohol, and cigarettes. 

In Vancouver, a night in a shelter costs society an average of $93. Therefore, based on the decreased shelter use, they calculate that their intervention generated savings of $8,277, or a net savings (subtracting the cash transfer) of $777 per person. There are likely other benefits as well, such as opening up shelter beds to other individuals how may have otherwise been sleeping on the street. 

This paper included two other embedded survey studies, both using the online Amazon Mechanical Turk system. The first demonstrates that people generally think (contrary to the above results) that homeless people are more likely to spend cash on temptation goods, like drugs, alcohol, and cigarettes, which is an important finding if you are trying to institute public policy based on these results. The second survey demonstrated that explaining the study results, either in the utilitarian frame emphasizing that cash transfers decreased shelter used and saved money overall, or in the counter-stereotype frame emphasizing the lack of spending on temptation goods, increased public support for a cash transfer program.  

My thoughts

This is a really important topic, and the authors deserve massive credit for organizing and executing an RCT on this topic. Although the results are really interesting, and the topic is important, this study has tremendous limitations. (The authors are pretty good at pointing out the limitations in their write up, and discuss these results as preliminary. News coverage, as expected, did not capture that nuance.) 

Most importantly, their initial analysis was negative, so all of the benefits discussed (both in the paper, where they were discussed fairly, and in the media, where discussion lacked scientific nuance) are based solely on an exploratory analysis that collapsed the 4 study groups into 2 groups after all the data was collected. The lack of a clear primary outcome is problematic, especially when the majority of outcomes measured were not statistically significant. You might anticipate, therefore, that all of the statistical findings could be chance findings, or the result of p-hacking. 

Despite the reporting, this study is not as simple as a one time cash transfer. The cash was always accompanied by at least motivational training workshops, and for half the individuals that was supplemented with one on one coaching. Only half of the control groups received coaching and workshops, which makes simple comparisons based only on the money somewhat inappropriate. 

They talk about enrolling 229 patients but only having 115 in their final sample. It is not clear at  which point these individuals were lost to follow-up, but it might mean that many people were given money who weren’t actually counted in these results. 

The selection of individuals who were homeless for less than 2 years is an important source of selection bias. This likely excludes a number of individuals with more severe mental health or substance abuse issues that would contribute to longer term homelessness. They also explicitly excluded patients with more severe mental health or substance use issues. Therefore, these results cannot be generalized to all unhoused individuals. 

Given the small size of the study, it is not surprising, but the groups don’t look evenly matched. I honestly don’t know enough about the contributors to homelessness to know which factors are most important, but the fact that significantly more people in the cash groups were experiencing homelessness for the first time seems potentially important. Furthermore, the appropriateness of randomizing by shelter depends a lot on whether the shelters are serving similar populations. (You could imagine that shelters in different areas of a city might service very different populations, with very different anticipated trajectories, which would significantly bias these results.) We would need the insights of someone familiar with the Vancouver shelter system to opine on whether this is a significant concern. 

Although they ran some math to choose the $7500, it may have been a poor choice for Vancouver, which is one of the most expensive cities in the world. That sum might not even cover 3 full months of housing/expenses, which would obviously limit the impacts of the intervention. It would be interesting to see similar experiments repeated in cities with lower costs of living. 

This is not an isolated study. They cite prior research (which I have not personally appraised) demonstrating that unconditional cash transfers provide a large variety of benefits, from improved physical health, psychological well-being, and increased education and employment. However, almost all of the previous research was from low and middle income countries.

Bottom line

Despite being widely reported as positive in the media, this was a negative study with some interesting secondary analyses. This is a very important topic for almost every emergency department, and warrants intensive research. (If we run another RCT looking at vitamins in sepsis when we could spend that money on research like this, we should seriously question the morals of the entire medical community.)

References

Dwyer R, Palepu A, Williams C, Daly-Grafstein D, Zhao J. Unconditional cash transfers reduce homelessness. Proc Natl Acad Sci U S A. 2023 Sep 5;120(36):e2222103120. doi: 10.1073/pnas.2222103120. Epub 2023 Aug 29. PMID: 37643214