Statistical Analysis

We estimated survival curves of the cumulative percentage of respondents exposed to firearm violence by race, sex, and cohort by a given age using the nonparametric Turnbull maximum-likelihood estimator (NPMLE), a generalization of the Kaplan-Meier estimator applicable to interval-censored data.^16,17 Accounting for interval censoring is necessary for the seen shot outcome because age at time of exposure is uncertain for respondents exposed in the years between wave 3 and 1 year prior to wave 5. This is less consequential for the been shot outcome, since wave 5 respondents were asked for the age when first shot, resulting in smaller censoring intervals.

We estimated associations between sociodemographic characteristics and exposure to firearm violence using multivariable semiparametric Turnbull (SPT) proportional hazards models.^18,19 The SPT model is analogous to the Cox proportional hazards model but estimates the baseline hazard using the NPMLE to account for interval censoring. A Cox model with interval midpoints as exposure times yields similar estimates (eFigure 1 in Supplement 1). No formal test of the proportional hazards’ assumption exists for SPT models, but accelerated failure time models that relax the proportional odds assumption produce substantively equivalent estimates (eFigure 2 and eAppendix 2 in Supplement 1).

We used a negative binomial regression model to estimate the association between our covariates and counts of past-year shootings occurring within 250 m of a respondent’s residence. These models include combined survey design and attrition weights to permit inferences to the population of Chicago children. We performed statistical analyses in R software version 4.2.2 using the interval, icenReg, and MASS packages (R Project for Statistical Computing), with a 2-tailed significance level of P < .05.^17,19,20 Because our analyses are descriptive and not directed at formal hypothesis testing, the purpose of calculating CIs is to draw inferences from our sample to the population. Data analyses were conducted from May 2022 to March 2023.