Read on Twitter
I’m presenting a new project next week — first of two new projects on mental healthcare. This one is one mental health court and it’s impact on repeat offending using data from a large urban correctional complex. Two years in the making. I’m frankly surprised I got the data.
I got the idea for the project pretty randomly. I wanted to do something with mental healthcare and criminal justice due to the high volume of mental illness in the inmate population, but really didn’t know what to study bc I wasn’t familiar with the programs. 2/n
It was more like I’d heard of programs from people in the programs but only kind of half awake bc when I would hear about the programs, I had a different hat on if that makes sense. I wasn’t sure what to google either. “Mental health and prison” brought up a lot. 3/n
So instead I started cold calling attorneys. I contacted an attorney who worked in indigent defense, traveled to her city, and took her out for tacos. I learned about the mental health court in her community, and decided I would study it. 4/n
So then I got the name of attorneys associated with the mental health court in that county. Tons of traveling and tons of tacos and tons of Indian food. Interviewed attorneys, director of inmate mental health, judges, shadowed defense attorneys at the mental health court 5/n
Interviewed police, learned about many other interesting interventions, approached a PhD student to work on this with me. It was probably 8-9 months of canvasing two urban counties to better understand the people and the agencies. 6/n
One of the large counties worked out and the other one just never worked out. The former is the one I’m presenting and the other I’m just hoping maybe another day I can make happen. The latter county just was such a complex bureaucracy. 7/n
But the first one worked out. I made a relationship with a large correctional complex and pitched a meeting with the Sheriff. There were therapists, officers and attorneys from all over the county. I went through three project proposals. 8/n
The first was a project I would do for the correctional complex that was part of public service in my mind — a model I have come to feel more strongly about in my career. Using ML, my students and I would develop a prediction model for suicide or any self harm risk. 9/n
The second was the mental health courts: using a variation on the judge fixed effects design, I’d estimate the impact of mental health court on earnings, repeat offending and other outcomes. It would require a lot of linking. 10/n
And the third was something I’m drawing a blank on. Anyway, the sheriff and the others were genuinely moved by the presentation. I think they knew I wasn’t trying to “use” or exploit them for data, but cared about the topic, and honestly even then abd the inmates. 11/n
Which was true. Two years in and it’s still true. So they wrote off on it. The correctional complex and me entered into a memorandum of understanding and a long process of learning their data began. I recruited two PhD students from our health policy (“health services research”)
And miraculously, last spring we got the data. Several years worth of inmate booking data for this large urban county. We built a team, one student is writing her dissertation on suicide risk in a correctional complex, and we are teamed up on the project. 13/n
Thing is, I also took a different route simultaneously. I also worked with the courts themselves because they own more of the data. And in the end, I just wasn’t able to secure all of their data. I thought the mental health court project was dead actually. 14/n
But then I realized I’d always needed the correctional complex bc they hired the therapists and the therapists were key to my design. The design which I’ll share another time came to me during interviews with the attorneys. 15/n
I kept asking attorneys “how do inmates get into mental health court?” Only it was more like I was asking a more narrow question than that: explain treatment assignment to me like I’m 5 years old. 16/n
I must have asked that a hundred different ways, bc I was trying to keep my ears alert for any exogenous treatment assignment. The defendants in mental health court already select on mental illness and are a group of “super users” who intensively use public resources. 17/n
So the selection issues are gargantuan, and you can’t merely compare defendants in them to those not in them, especially not without careful understanding of selection and treatment assignment. But the experience proved for me a hunch I’ve had. 18/n
There is almost always quasi random assignment in large areas in 2020. They’re using running variables due to digitization, or they’re using explicit randomization. It’s usually there but it requires all the planets to align to find it. 19/n
So amazingly, the project I set out to do I am mostly doing. I’m not looking at earnings and employment bc I ultimately am working with incomplete data, but I am looking at repeat offending, which is a good start. 20/n
I think it’ll be a useful contribution to the area of mental healthcare and criminal justice, to mental health courts, to even constitutional questions around the financing of indigent defense. And I’m presenting preliminary results next week. 21/n
Today through Sunday, I’m making slides from scratch, trying to remember all the stuff I learned about the mental health-criminal justice And mental health court literature all the way back to 2018-2019. I was relieved I took good notes but still not great notes. 22/n
Going to be thinking about how to improve that part of the work flow bc I am a voracious reader but also tend to have a Swiss cheese brain and forget things. Probably need a better note taking system. 23/n
Anyway, point being — I think but don’t want to overstate this that there’s lots of great projects in large urban communities that can be surfaced if you have a narrow enough idea and socialize extensively within the area. Be courteous, polite, respectful, curious. 24/n
Don’t exploit them, but see if you can help them while doing your own projects. Build relationships. I still have drinks with some of the people I have met, still text new friends. Relationships and alliances are important for some research agendas.25/n
