Do Deceptive Interrogation Bans Affect Juvenile Case Resolution? Nine-State Results, Forthcoming in Crime & Delinquency

Update (August 2026): This post describes the final, peer-reviewed version of the study, now forthcoming at Crime & Delinquency. It replaces an earlier preliminary post based on a seven-state analysis; the interactive dashboard has been fully rebuilt around the accepted results.

The question

American police may legally use deception when interrogating suspects --- fabricated evidence, false promises of leniency. Because juveniles are especially prone to false confession under such pressure, nine states restricted the practice for minors between 2021 and 2024: Illinois and Oregon (January 2022), Utah (May 2022), Delaware (October 2022), Indiana (July 2023), Colorado (August 2023), Connecticut (October 2023), and Nevada and California (July 2024).

Restrictions on interrogation can be understood not only as safeguards against unreliable admissions but also as potential constraints on police fact-development, raising the hypothesis of a public-safety tradeoff: fewer confessions, fewer cases solved, more prosecutions refused. Our study, “Do Deceptive Interrogation Bans Affect Case Resolution? Evidence from Juvenile Justice Reforms” (Mourtgos & Adams, forthcoming in Crime & Delinquency), tests whether that hypothesized cost appears in the data.

What we did

Using FBI NIBRS incident data covering 510,582 serious juvenile-involved incidents reported by 11,333 agencies from 2021 through 2024, we compare the nine adopting states to 42 never-treated state units (including Washington, DC) on two outcomes: arrest clearances (the case is solved by arrest) and prosecution-declined exceptional clearances (police present the case; the prosecutor refuses). The primary model is a hierarchical Bayesian binomial-logit difference-in-differences on the agency-month panel; each state also gets its own Bayesian synthetic control, and the results are stress-tested with randomization inference, rank-based permutation tests, an adult-offender triple-difference placebo, dose-response models over statute strength, and a battery of sample and window robustness checks.

What we found

We find no aggregate decline in case resolution within the study period.

• Arrest clearances did not decline. The primary estimate is +2.25 percentage points (95% CrI +0.99 to +3.60), with a posterior probability of any decline of 0.5%. Every one of twelve estimation strategies lands null-to-positive; all nine state-specific synthetic-control estimates are positive (only Indiana’s credible interval excludes zero). By the paper’s primary frequentist standard --- randomization inference, because nine treated clusters are too few for reliable clustered standard errors --- the pooled effect is not statistically significant (p = .47): the honest headline is “no decline,” not “bans raised clearances.” We report the positive drift as descriptive, not causal.
• The strictest laws show no larger declines. A dose-response model over a statute-strength composite finds the strongest laws (Oregon, Connecticut) associated with the largest positive annual change in arrest clearances (+2.28 pp/yr, CrI +1.00 to +3.51) --- the opposite of the hypothesized gradient, though the strength cells are thin (one or two states per level).
• Prosecution declinations fell, but not just for juveniles. Declinations drifted down in ban states (-0.26 pp, P(increase) = .13) --- the opposite of the hypothesized increase. Adult cases in the same states show a comparable decline, and the juvenile-minus-adult triple-difference is null (-0.27 pp, CrI -1.11 to +0.49), so the pattern is not attributable to the juvenile bans.
• Offense-specific results are mixed but broadly consistent with the aggregate pattern. Property offenses show positive, relatively precise estimates. Sex crimes show the most credible evidence of a possible reduction (P(decline) = .89; -0.63 pp, with an interval including zero), and the murder estimate is negative but highly imprecise --- patterns consistent with the possibility that cases depending more heavily on statement evidence respond differently, though not confidently distinguishable from sampling noise.

Caveats: Nevada and California contribute only about six post-ban months; pre-trend tests reject equal slopes for both outcomes (for arrest clearance the divergence biases against the positive finding; for prosecution declined it warrants genuine caution); we observe case outcomes, not interrogation practice, so mechanisms remain unmeasured; the analysis is aggregate, so individual cases may still have been affected even where no aggregate change is detectable; and the statutes studied cover juveniles only --- results could differ for adult interrogation restrictions.

Explore the results

The full walkthrough --- the nine statutes and their strength coding, the twelve-method forest plot, state-by-state synthetic controls, dose-response, the adult placebo, subgroup and offense-specific estimates, and the robustness suite --- is in the interactive dashboard:

Methods summary

• Primary estimator: hierarchical Bayesian binomial-logit difference-in-differences with random treatment slopes at the state and agency level (brms/CmdStan), incident-weighted average marginal effects
• State-specific: Bayesian synthetic control models with AR(1) latent structure, one per treated state
• Frequentist convergence: two-way fixed effects with two-way clustering, randomization inference (500 permutations; the primary significance measure), rank-based permutation tests (10,000 permutations), augmented synthetic control (multisynth)
• Placebos and robustness: adult-offender triple-difference, leave-one-state-out, drop-California refit, consistent-reporter restriction, extended 2016-2024 panel, pre-trend diagnostics