class: center, middle, inverse, title-slide .title[ # Predicting and Preventing Homelessness in Franklin County: A Summary of Early Progress and a Few Calls to Action ] .subtitle[ ## Presented on Behalf of Smart Columbus at the Columbus Data & Analytics Wednesdays Meetup ] .author[ ### Ty Henkaline ] .date[ ### April 16, 2024 ] --- ## Not so fast! ### Some notes before you go any further ... - Any pink text you see in these slides is clickable for more information - These slides are basically just a higher-level summary of a much more comprehensive white paper, for which both an executive summary and the full white paper are available: - [Executive Summary of White Paper](https://drive.google.com/file/d/1NXzdPXLYFVfLTBCGC8WZVKMoTNU0NMqH/view?usp=sharing) - [Full White Paper](https://drive.google.com/file/d/1tiNBxX_8uSiEYxX6HsOsfOARI4lPQunF/view?usp=sharing) - Yet another option for learning more about this work is to use some LLM-driven tool. For instance, here is how to chat with the white paper using Claude (the free version is fine): - [Download](https://drive.google.com/file/d/1fPlkL-kBTj3Vp9q1mK90eotRk-lSYgWB/view?usp=sharing) the [full white paper](https://drive.google.com/file/d/1tiNBxX_8uSiEYxX6HsOsfOARI4lPQunF/view?usp=sharing) - [Open a tab for Claude](https://claude.ai/) in your browser - Copy/paste [this text](https://drive.google.com/file/d/1SNz-jZkHuCkwux2uCFTBEWgaIJ8M6MMJ/view?usp=sharing) as your Claude prompt - [Attach](https://drive.google.com/file/d/1puhOVZWr-M35wI8CD9QJHgTSp7DnKV7O/view?usp=sharing) the [white paper](https://drive.google.com/file/d/1WIDv9kZS39qnkzYjnmEVrm10Xe1yqirQ/view?usp=sharing) to your prompt - Press Enter / Return --- ## Acknowledgements ### Key Contributing Organizations #### Funders - [Franklin County](https://www.franklincountyohio.gov/) - [City of Columbus](https://www.columbus.gov/) #### Contributors - [Smart Columbus](https://smartcolumbus.com/) - [RISE Together Innovation Institute](https://www.rtiico.org/) - [Mid-Ohio Food Collective](https://www.mofc.org/) - [Community Shelter Board](https://www.csb.org/) #### Collaborators - [California Policy Lab](https://www.capolicylab.org/) --- ## Acknowledgements ### Key Contributing People #### Smart Columbus - [Jordan Davis](https://www.linkedin.com/in/jordan-davis-96116310) - [Diane Dagefoerde](https://www.linkedin.com/in/dagefoerde/) - [Ty Henkaline](https://www.linkedin.com/in/tyhenkaline/) - [Hailey Allison](https://www.linkedin.com/in/haileyallison/) - [Tosin Guobadia](https://www.linkedin.com/in/tosinguobadia/) - [Ayaz Hyder](https://www.linkedin.com/in/ayaz-hyder-bb2a2630/) #### RISE Together Innovation Institute - [Danielle Sydnor](https://www.linkedin.com/in/danielle-l-sydnor-402823b/) --- ## Acknowledgements ### Key Contributing People #### Mid-Ohio Food Collective - [Matt Habash](https://www.linkedin.com/in/matt-habash-8866578/) - [Nick Davis](https://www.linkedin.com/in/nickdavisinnovation/) - [David Pickering](https://www.linkedin.com/in/david-pickering/) - [Jason Gleim](https://www.linkedin.com/in/jgleim/) - [Matt Thompson](https://www.linkedin.com/in/johnthompsoncsm/) #### Community Shelter Board - [Shannon Isom](https://www.linkedin.com/in/riveted/) - [Lianna Barbu](https://www.linkedin.com/in/lianna/) #### California Policy Lab - [Brian Blackwell](https://www.linkedin.com/in/brian-blackwell-b5137576/) --- ## Key Collaborative Underpinnings --- ## Key Collaborative Underpinnings - **[Rise Together Blueprint](https://risetogether.franklincountyohio.gov/)** - Comprehensive roadmap with 13 strategic goals - **[Rise Together Innovation Institute](https://www.rtiico.org/)** - Hub for driving collaboration and systemic change - **[Smart Columbus](https://smartcolumbus.com/)** - Human-centered, technology-driven solutions for residents - **[Central Ohio Stable Housing Network](https://www.ahaco.org/)** - 28 agencies with dedicated Housing Resource Specialists --- ## The CIE Initiative --- ## The CIE Initiative  --- ## The CIE Initiative  --- ## The CIE Initiative  --- ## The CIE Initiative  --- ## The LA Initiative --- ## The LA Initiative  --- ## The LA Initiative  --- ## The LA Initiative  --- ## The LA Initiative - **Proof points** - Identifies residents up to **48 times** more likely to experience homelessness - **99%** of identified at-risk individuals not currently targeted by other prevention programs - Enables **proactive** caseworker outreach with targeted assistance --- ## The Big Question --- ## The Big Question <span style="font-size: 24pt;">Should we be doing this in Franklin County?</span> --- ## The State of Homelessness in Franklin County --- ## The State of Homelessness in Franklin County - **Staggering and disturbing growth** - Predicted 68% surge in homelessness from 2024 to 2029 - 47% surge in chronic homelessness since 2023 - Only 26 affordable housing units available per 100 extremely low-income households --- ## The State of Homelessness in Franklin County - **Antiquated approach** - Reactive rather than proactive interventions - Focus on crisis management rather than prevention --- ## The Research Questions --- ## The Research Questions **1. Can data be sourced from service providers?** - Will organizations share sensitive data? - Can we overcome technical, legal, and governance barriers? --- ## The Research Questions **2. Can data be linked across service providers at the resident level?** - Is it possible to create unified resident profiles? - Can we overcome data quality challenges? --- ## The Research Questions **3. Do patterns exist in cross-agency data that may help predict homelessness?** - Are there early warning signs in service usage? - Can we identify key risk factors? --- ## The Research Questions **4. Can these patterns be used to predict future homelessness?** - How accurate can our predictions be? - Can we identify high-risk residents for early intervention? --- ## The Data Sharing Partners --- ## The Data Sharing Partners - **Mid-Ohio Food Collective (MOFC)** - Food assistance services - **Community Shelter Board (CSB)** - Housing and homelessness services --- ## Question 1: Can data be sourced from service providers? --- ## Question 1: Can data be sourced from service providers? - **Answer: Yes!** - We received over 30.8 million service records spanning nearly 12 years --- ## Question 1: Can data be sourced from service providers? - **What we requested:** - Identifying information (name, DOB) - Basic demographics - Service details (date, type, description) --- ## Question 1: Can data be sourced from service providers? - **What we received:** - MOFC: 30,582,456 records for 1,283,715 residents - CSB: 264,836 records for 63,124 residents --- ## Question 2: Can data be linked across providers? --- ## Question 2: Can data be linked across providers? - **Answer: Yes!** - We successfully linked records at the resident level --- ## Question 2: Can data be linked across providers? - **Method:** Linking by first name, last name, and date of birth - Less than 1% duplicate or mismatched records - Conservative approach given data quality considerations --- ## Question 2: Can data be linked across providers? - **Key finding:** Most CSB clients also used MOFC services - 67% of shelter users also accessed food assistance - Significant overlap validates our approach --- ## Question 3: Do patterns exist that may help predict homelessness? --- ## Question 3: Do patterns exist that may help predict homelessness? - **Answer: Yes!** - We identified clear warning signs, including a characteristic spike in service usage --- ## Question 3: Do patterns exist that may help predict homelessness?  --- ## Question 3: Do patterns exist that may help predict homelessness? - **Key finding:** Service usage increases sharply before homelessness - Similar pattern observed in **both** Franklin County and LA County - **Critical** intervention window visible in the months before crisis --- ## Question 4: Can these patterns predict homelessness? --- ## Question 4: Can these patterns predict homelessness? - **Answer: Yes!** - Even with limited data, we can identify residents at nearly 10x greater risk --- ## Question 4: Can these patterns predict homelessness?  --- ## Question 4: Can these patterns predict homelessness? - **Our prototype model:** - **Simple** additive logistic regression with seven basic features - AUC of 0.63 (**mildly effective** already) - Highest-risk group nearly **10 times more likely** to experience homelessness --- ## Early Success: Data Sourcing & Linking --- ## Early Success: Data Sourcing & Linking - **Established technical, legal, and governance frameworks** - Comprehensive data use agreements - Secure data transfer protocols - Secure data storage and analysis environment --- ## Early Success: Data Sourcing & Linking - **Ready to scale with additional partners** - Proven value proposition - Streamlined onboarding process --- ## Early Success: Pattern Identification --- ## Early Success: Pattern Identification - **Validated key findings from LA** - Service usage spike before homelessness crisis - Clear patterns despite limited data sources --- ## Early Success: Pattern Identification - **Additional insights** - Divergent service trends during COVID pandemic - Distinct service usage patterns among different user groups --- ## Early Success: Prediction --- ## Early Success: Prediction - **Statistical significance achieved with limited data** - With just two data partners - Simple model shows promising results --- ## Early Success: Prediction - **Enormous potential for improvement** - LA model uses data from 12 county departments - Each additional relevant partner can significantly increase accuracy --- ## Next Step: More Data --- ## Next Step: More Data - **Key potential partners include:** - County agencies - Community-based organizations - Social services agencies - Housing assistance programs - Workforce development agencies --- ## Next Step: Better Modeling --- ## Next Step: Better Modeling - **Key modeling enhancements include:** - Advanced algorithms beyond logistic regression - Expanded feature engineering - Time-series analysis techniques - Ensemble methods combining multiple models --- ## Next Step: Plan Intervention Testing --- ## Next Step: Plan Intervention Testing - **Key planning steps include:** - Establish partnership - Obtain funding - Design intervention --- ## Calls to Action --- ## Calls to Action: More Data! - **For potential partners:** - Join a proven initiative - Receive valuable insights about your service population - Contribute to critical community impact --- ## Calls to Action: Better Modeling! - **For potential collaborators:** - Join a proven initiative - Gain valuable experience - Contribute to critical community impact --- ## Calls to Action: Plan Intervention Testing! - **For potential partners:** - Join a proven initiative - Get additional funding - Contribute to critical community impact --- ## How to Act! Send me a DM on LinkedIn and we can take it from there! Scan this QR code to [connect with me on LinkedIn](https://www.linkedin.com/in/tyhenkaline/):  --- ## Thank You! ### For more details - Our CIE Initiative Plan - [Summary](https://drive.google.com/file/d/1_VErrkR20CAqwPrPuyYCDokBapihd-hw/view?usp=sharing) - [Full](https://drive.google.com/file/d/1VGNAYGL8oBsllEkxjGByHHB6COlhw8Ba/view?usp=sharing) - Our CIE's Homelessness Prediction / Prevention White Paper - [Summary](https://drive.google.com/file/d/1NXzdPXLYFVfLTBCGC8WZVKMoTNU0NMqH/view?usp=sharing) - [Full](https://drive.google.com/file/d/1tiNBxX_8uSiEYxX6HsOsfOARI4lPQunF/view?usp=sharing) - LA's Homelessness Prediction / Prevention Work - [Website](https://capolicylab.org/topics/homelessness/#:~:text=Hosted%20Homelessness%20Research-,Predicting%20and%20Preventing%20Homelessness,-CPL%20has%20multiple)