Closing my tabs weekly roundup

Tales from a tab hoarder

Welcome to closing out my tabs, a weekly substack where I post things that have been on my browser open on one condition. If I post it, I have to close the tab. I’ll start with a new paper of mine, as it’s open on my tabs.

1. My new JHR with Karen Clay, Vivian Vigliotti and Jonathan Seward (randomized ordering) on mental illness, self harm in jails

This paper is our first publication from a longer research agenda on studying self harm in the jails and prisons. I started this project back in 2018, got the Sheriff to green light it, and the recruited Jonathan and Vivian in the spring. Vivian wrote her dissertation on these data partly studying the mental health courts. Then we began working with Karen and it all came together to study the effect of being classified as severely low functioning (which was a mental illness classification the booking therapists used) on their suicidality (measured as ideation and attempts), recidivism and time in jail (length of stay). We used a leniency design because of the randomization of therapists who processed inmates within 36 hours of booking. You can find the paper here. Here’s some tables and figures

For those that do not the leniency design, this is a picture that has become standard based on papers by Crstyal Yang, Will Dobbie and co-authors attempting to show how there is variation in the therapists average scoring of inmates. The histograms are the therapists own average times they scored someone as having severe low functioning. But as the therapists are randomly assigned to these inmates, these distributions are unlikely to be the product of chance variation. They are rather suggestive of systematic tendencies to rank someone as more or less likely mentally ill.

The inmates only meet with the booking therapist briefly by law and for a short time (as short as 15 minutes) in a sequence of events that we’d call “booking”. They aren’t getting therapy from these therapists and do not work with them again. They receive this score of a 0-3 where 2 and 3 are markers of moderate to severe mental illness which makes them eligible for mental health court after they leave jail and 0-1 is none or mild and keeps them in the same traditional adjudication. We use a binary treatment variable of 0 if mild or no signs of low functioning and 1 if moderate to severe. We then look at the effect that this classification had on outcomes within the jails (i.e., before they exit). Here’s what we found.

We broke it up by misdemeanors and age. We call one group a “transitional age” group as they are less than 24 and the reason is that the onset of severe mental illness symptoms like bipolar and schizophrenia is overlapping with exactly the age when they would often times have first contact with the police. We estimate 2SLS using the residualized leave-one-out mean as the instrument (think of it as the therapist’s average scoring of high or moderate low functioning in all his recent cases except for with this persons, but residualized with baseline things and fixed effects). But as others have noted, this is like a propensity score that reduces a multi dimensional problem of multiple therapist fixed effects into a single scalar, which hides a weak instruments problem. JIVE has bias from too many covariates. We therefore could either use unbiased JIVE by Kolesar (2013), but as that paper is not published and his work is in Matlab, and I personally felt a little insecure about adapting it, we used a published paper by Chernezokov, et al. on IV Lasso which handled multiple covariates and multiple instruments. LASSO has sparsity, which does not make sense in a leniency design, so we put thresholds on how far down to go, but it never made any difference.

I’ll report column 3 for kids. Length of stay increases by an extra week (relative to counterfactual). Anderson-Rubin confidence intervals are reported in brackets under 2SLS coefficients. Suicide attempts rise 4pp. Scaled by length of stay, suicide attempts rise 0.7pp per day in jail. Nothing noticeable on recidivism.

Then we look at adults. Again misdemeanors. Misdemeanors make you eligible for mental health court if you score a high score. Suicide attempts rise 2pp, and adjusted for length of stay, around 0.4pp per day. Some sign of recidivism upon release, but release is confounded by the mental health court so it’s less easy to attribute to the black box of in-jail scoring than it is how the score affects them when they leave.

Next we looked at felonies. Large increases in jail if a youth with a felony charge is tagged as mentally ill. They spend almost 2 months longer in jail relative to counterfactual, and it’s more than double the OLS result. So something is clearly going on with that mental illness tagging that's causing these kids to not be let out. Nothing on suicide attempts. Adults, similarly, length of stay rises 3 weeks relative to counterfactual mean and suicide attempts per day is positive at around 0.1pp per day in jail.

Feel free to leave any comments, suggestions or criticisms and I’ll answer what I can.

2. Heckman and Pearl on causal inference

This is from a 2012 conference held at chicago on causal inference and both men spoke. There’s videos for each of them, and you get some discussion from Heckman in particular. Sadly, I never did finish it and now I must close it.

3. Abadie and on synthetic controls

Here is another review of synth by Alberto and Vives-i-Bastida. It’s full of recommendations for empirical practice.

4. Out of body experiences, reliving things, not memories

I had a home invasion this summer by a kid who broke into my car, then a second time though the second time was worse and violent. He came to the house, ranting about freemasons and had a copy of my Jude the Obscure in his hand (he’d gotten it from the car), and saying I was in his dreams. He then went around back, hitting the house with some junk metal screaming, shattered the window of my 1991 240 Volvo wagon and stole a license plate in the back. They don’t make those anymore; I had to go on a subreddit enthusiast forum for 240s and buy a guy’s entire hatch, but I did and it’s fixed. But watching it all happen and not being able to stop it was surreal.Then the next day, he did a home invasion while I was camping with the girls. It all got caught on video (as I have them everywhere now). Police arrested but this summer he got out and on Sunday and Monday he returned. Came to the door, acting weird again, two days in a row. Clearly taunting me. Kept calling me by my full name, and I think was casing the house again. He never did get the computers or television when he was here; just my Ted Lasso “Believe” portrait.