Correctional Officers vs. Inmates: Who Prevents Self-Harm in Prison?
Suicide is the leading cause of death in U.S. jails and the second-leading cause in prisons. Traditional prevention relies on correctional officers, but chronic staffing shortages—25–30 percent vacancy rates in Texas—have made surveillance-based approaches increasingly untenable. Using administrative records from the Texas Department of Criminal Justice covering 11 million inmate-months from 2017 to 2024, we evaluate two approaches to reducing suicide and self-harm. First, using panel estimators, we find that a one standard deviation increase in correctional officer staffing reduced completed suicides by 50 percent but had no effect on self-harm or suicide attempts—correctional officers intervene in acute crises but do not prevent the upstream behaviors that precede them. Second, we evaluate a peer support program in 16 prisons that pairs high-risk inmates with trained inmate volunteers. A logit model of baseline self-harm risk produced 3,614 unique fitted values which, for operational simplicity, were coarsened into 39 discrete scores; inmates above a prison-specific threshold were assigned peer support. This coarsening created quasi-randomization: inmates with identical underlying risk received different treatments depending on how rounding shifted them relative to the threshold. We exploit this variation alongside a threshold-progression difference-in-differences estimator that aggregates effects across facilities with varying entry thresholds. Peer support reduces self-harm by 30–60 percent, with effects appearing precisely where quasi-randomization is strongest and absent where assignment is nearly deterministic. Expanding staffing would cost $1.6–3.1 billion annually; peer support achieves comparable reductions at far lower cost, though it raises ethical questions about using inmate labor for institutional functions.
LOCATION:CGIS Knafel Building, Room K354 STATUS:CONFIRMED DTSTART:20260401T160000Z DTEND:20260401T173000Z END:VEVENT END:VCALENDAR