Impact Evaluation of Atlanta Mission’s Transformation Model

Atlanta Mission's Transformation Model is a four-phase residential program for individuals experiencing homelessness. Find Hope and Choose Help (Stage 1 programs) are 1–30 day stabilization and assessment phases in which clients can complete a needs assessment, attend six classes, and meet with a case manager. Clients who complete these requirements are placed on an approval list for Make Progress (Stage 2 program), the core intensive intervention. Each week, a given number of clients will become eligible for Make Progress and a random lottery will occur to determine to which of the three arms of the study an individual will be assigned.

Make Progress (MP):
Personalized services over ~6 months in five key areas: physical, emotional, spiritual, social, & vocational. Each client follows a personalized service program based on their needs assessment. Through these services, the goal is for clients to become rooted in community, retain employment, and become housing eligible. Successful clients can then enter Sustain and Grow afterwards, Atlanta Mission’s transitional housing program focused on job retention and long-term stability for those who still have housing needs.

Fast Track (FT):
A new bed-only program requiring participants to find employment independently within 30 days. If employment is secured, they will then enter Sustain and Grow having skipped the personalized services in Make Progress. Atlanta Mission views this as a method to serve more individuals at a lower cost with some of their extra bed capacity in the transitional housing wing.

Control Group:
Individuals offered no specific program slot due to capacity limitations in both Make Progress and Fast Track. They are provided with a list of resources available to them in the area. Individuals assigned to this arm will retain their bed at Atlanta Mission for 30 days, receiving minimal staff support to help them land on their feet.

2026-09-14

2029-09-14

Any homelessness
Duration of homelessness
Quarterly earnings
Labor force attachment
Job transitions

New HMIS program enrollments will be used to construct these indicators of homelessness. We will include all program types, but will distinguish between prevention programs versus ones that indicate the client is experiencing homelessness (e.g., emergency shelter, street outreach, diversion, transitional housing, safe haven, rapid re-housing, and permanent supportive housing), as well as any new enrollment in HMIS assessment projects used when these programs are oversubscribed.


We go into further detail in our analysis plan regarding the HMIS outcomes, as well as the economic stability ones.

Credit-bureau outcomes
Criminal justice system contact

Experian data - consumer credit and borrowing information gleaned from public records, collection agencies, and trade lines (such as credit card balances, auto loans, and mortgages) - will be used to construct our credit-related outcomes. We go into further detail in our analysis plan.
Criminal justice system contact - we intend to access this type of data, but prior to registration of this RCT and creation of this pre analysis plan have no contact with any data providers.

This is an RCT that seeks to estimate the causal effects of Atlanta Mission’s programs on housing stability and labor market outcomes. Atlanta Mission is conducting all aspects of the study, aside from data analysis. They have no need to recruit individuals, as they have sufficient inflows. When an individual enters the door at one of their two intake sites, one for men and one for women, they will be told that space in Atlanta Mission’s long-term program, Make Progress, is limited. Atlanta Mission has designed a randomization method for determining, amongst eligible clients, who will ultimately be offered a spot in the program. Clients become eligible by reaching goals in their first 30 days with the Mission, attending some classes and meeting with Atlanta Mission staff. Each week clients approved to enter Make Progress will be included in a random lottery procedure designed by Atlanta Mission.

The procedure runs every week regardless of capacity. When MP slots suffice for all eligible clients, all are offered MP and no FT or Control assignment occurs; when MP and FT together suffice, no Control assignment occurs. Control assignment arises only in weeks of excess demand for both MP and FT simultaneously. Assignment probabilities are therefore not equal across clients: a client's realized probability of each arm depends on their draw, their stratum, their size-feasibility relative to that week's slot inventory, and the competing queue. We address this in our analysis plan.

Not available

Randomization will be conducted by a computer on a weekly basis, with an automated workflow occurring every Monday once that week’s approval list has been finalized. Staff members will have no control over or knowledge of the assignment prior to randomization. The probability of being assigned to one of the three arms will vary from week to week, impacted by the composition of beds available and characteristics of other clients on the list in that week. We go into further detail regarding the randomization method in the Pre-Analysis Plan.

Individual/Household Level

No

0

Our target sample size is 700-800 clients determined eligible and approved for Make Progress in a given year. Historical records from Atlanta Mission indicate this volume is plausible. We make the following assumptions to further refine our planned number of observations: 350 unique program slots in MP during a given year, based on historical data Estimated Fast Track slots over 1 year: -20 men slots -8 single women slots -2-4 family slots -30 in total, with half open at beginning and other half opening 6 months later Monthly breakdown: -Months 1-2, 60 people will progress to Make Progress, and the other 60 will be offered Fast Track -15 of the 60 will secure a job and progress, filling the slots in ~2 months -Months 3-8, no FT slots available. 180 will progress to Make Progress, 180 will be referred out -Month 8-10, 15 FT slots available, 60 people will progress to Make Progress, and the other 60 will be offered Fast Track -15 of the 60 will secure a job and progress, filling the slots in ~2 months -Months 11-12, no FT slots available. 60 will progress to Make Progress and 60 will be referred out Over 1 year: -Estimated Make Progress size: 350 -Estimated control size (referred out and not offered FT): 240 -Estimated offered FT size, not getting a job: 90 -Estimated offered FT size, getting a job: 30 80% consent rate among eligible clients These assumptions put our anticipated number of observations anywhere from 560-640 in a given year. We will not have data for individuals who do not consent, and those individuals will still be eligible for program slots. Over 3 years, we anticipate our number of observations to fall in the range of 1,680-1,920 consenting clients.

Over 3 years:
-Estimated Make Progress size: 1050
-Estimated control size (referred out and not offered FT): 720
-Estimated offered FT size, not getting a job: 270
-Estimated offered FT size, getting a job: 90

Over 3 years, incorporating consent rates:
-Estimated Make Progress size: 840
-Estimated control size (referred out and not offered FT): 576
-Estimated offered FT size, not getting a job: 216
-Estimated offered FT size, getting a job: 72

We base our power calculations based on our outcome for returning to homelessness, using Cohen’s 2024 paper: Housing the Homeless: The Effect of Placing Single Adults Experiencing Homelessness in Housing Programs on Future Homelessness and Socioeconomic Outcomes. The study took place in Los Angeles, and the programs observed have similar dosage, intensity, and timing levels as Make Progress and Fast Track. Cohen draws on administrative records from LA County's HMIS, covering single adults who sought assistance between 2016-2017. The short-term sample comprises 15,990 cases with at least 10 months of post-intake outcomes; the medium-term sample comprises 7,584 cases with at least 20 months. The population is predominantly Black (~53%) and male (~66%), with an average age of ~46. It is a high-need, high-system-contact group: 70% have a prior homeless history, 80% report a physical or mental disability, 60% are chronically homeless, and 27% have a substance abuse problem. Prior system contact is extensive: 52% had emergency health service receipt, 23% had spent at least one night in jail, and 43% had used emergency or crisis services in the 6-12 months before intake. We consider this population to be comparable to the population served by Atlanta Mission, with them telling us they anticipate their population might be even more high-risk. We use the control means from the paper for the 10 and 20 month outcomes of returning to homelessness. We also anchor our power calculations to the OLS estimates found by Cohen, anticipating them to be akin to our ITT estimates for this RCT. We only conduct power calculations for our expected primary comparison, Make Progress to Control. 10-Month Outcome - Control Mean = 0.30. Cohen found 8-9 pp reduction. Year 1 (N = 472: 283 MP, 189 Control) Absolute MDE: 12.6 percentage points Relative MDE: 41.9% reduction relative to baseline of 0.30 Year 2 (N = 944: 566 MP, 378 Control) Absolute MDE: 8.8 percentage points Relative MDE: 29.4% reduction relative to baseline of 0.30 Year 3 (N = 1416: 850 MP, 566 Control) Absolute MDE: 7.2 percentage points Relative MDE: 23.9% reduction relative to baseline of 0.30 20-Month Outcome - Control Mean = 0.42. Cohen found 5-6 pp reduction. Year 1 (N = 472: 283 MP, 189 Control) Absolute MDE: 13.1 percentage points Relative MDE: 31.2% reduction relative to baseline of 0.42 Year 2 (N = 944: 566 MP, 378 Control) Absolute MDE: 9.3 percentage points Relative MDE: 22.1% reduction relative to baseline of 0.42 Year 3 (N = 1416: 850 MP, 566 Control) Absolute MDE: 7.6 percentage points Relative MDE: 18.0% reduction relative to baseline of 0.42