Interactive Analysis

Does a Six-Hour Interrogation Carry a 10% Risk?

Smith et al. (2026) report that the risk of a false confession wrongful conviction reaches 10% at about six hours of interrogation, and 25% at twelve. Both numbers depend on choices the evidence cannot settle.

Based on Mourtgos & Adams, a reply to Smith et al. (2026), forthcoming in the Journal of Criminal Justice

To estimate the risk of a false confession wrongful conviction (FCWC) at a given interrogation length, a model needs two duration distributions: one for interrogations that produced an FCWC, and one for those that did not. The estimate turns on how much more common a given duration is under the first than the second.

Smith et al. build the first from Redlich et al. (2011) — 35 people with serious mental illness who self-reported a false confession, with a mean total questioning time of 3.07 hours across multiple sittings. They build the second by pooling twelve sets of summary statistics from eight studies, weighted by sample size.

Neither distribution was collected to answer this question, and the two are not measuring the same thing. Below is what that costs.

Change One Study, and the Answer Nearly Triples

Smith et al. motivate the duration contrast with Drizin and Leo's (2004) documented false confession cases, which had a reported mean length of 16.3 hours — but their model uses Redlich et al. instead. Below, everything is held fixed except which false confession study supplies that first distribution. Drag the slider to any duration.

Interrogation duration 12.06 hours
15 minutes24 hours
1.53 h 3.07 h 6.28 h 12.06 h 16.23 h
Redlich only Published
26.3%
35 self-reported false confessors with serious mental illness
Both sources
62.1%
Redlich and Drizin & Leo, weighted by sample size
Drizin & Leo only
72.6%
44 documented cases with a reported length; mean 16.3 h
Estimated FCWC Risk by Interrogation Duration
The vertical line follows the slider; the dotted line marks 10%.
These three estimates do not bracket a true value. They show that the model's answer is determined substantially by inputs that neither underlying study was designed to estimate. A quantity that moves from roughly one-quarter to nearly three-quarters on a choice of source is not a calibrated risk boundary.

And the Six-Hour Threshold Moves With It

The published thresholds sit far out in the tails of fitted curves. Only about 0.116% of the fitted comparison distribution lies above 12.06 hours — an output of an assumed curve, not an observed frequency. So the threshold depends heavily on how fast that curve is assumed to fall, and on which studies are allowed into it. Each row below changes one such choice.

Duration at Which Estimated Risk Reaches 10%
Each row holds the FCWC base rate at .019 and the published false confession distribution, varying only the comparison distribution.
The most comparable check is the one that never reaches 10%. Redlich et al. (2011) reported durations for both confession groups, drawn from the same population and answering the same questions. Using that study's own 30 true confessors as the comparison, estimated risk is 3.6% at 6.28 hours and 4.0% at 12.06 hours, never exceeding about 4.2%. The sharp increase in the published model does not come from the difference observed within that study. It appears when that study's false confessors are compared against the narrower distribution pooled from other sources.
Within the published specification, applying Smith et al.'s own bootstrap procedure to both duration distributions and carrying the draws through Bayes' theorem gives a 95% interval of roughly 5.7%–14.6% at 6.28 hours and 7.1%–42.3% at 12.06 hours. At the point where the published estimate reaches 10%, uncertainty in the fitted curves places the risk on either side of that threshold.

What This Does, and Does Not, Show

It does not show that long interrogations are safe

Nothing here establishes that prolonged interrogation is safe, or that duration is unrelated to false confession. Prolonged interrogations may well increase risk. The finding is narrower and about measurement: the available studies do not identify the magnitude of the relationship, and six hours is a product of selected inputs and fitted tails rather than a boundary established by observed interrogations.

The underlying problem is a missing denominator

Finding that prolonged questioning is common among known false confessors does not tell us whether a six-hour interrogation carries a 2%, 10%, or 50% risk — unless we also know how often six-hour interrogations occur without producing an FCWC. Studies assembled from known cases cannot supply that number. Bayes' theorem can update probabilities using evidence; it cannot create the denominator that more than 45 years of research never collected.

Where we agree. Smith et al.'s replication of our original analysis is a real contribution, and it confirms its central result. Both papers agree on what the field should do next: code actual interrogations for tactic presence, duration, and outcome as recordings become more widely available. Both agree that even a 1% FCWC rate would be problematic.

Recording alone, though, will not supply the denominator. Researchers would also need to define the population interrogations are sampled from, separate individual sessions from cumulative questioning, record the time of the first admission, link interrogation characteristics to case outcomes, and distinguish a false confession from one that contributed to a wrongful conviction.

The disagreement is narrower than it may appear. It concerns whether the available evidence establishes the magnitude of the risk. The seriousness of an outcome informs how much risk society should tolerate; it cannot establish how often that outcome occurs.

Reproducibility

Both figures are computed live in your browser from the published duration model — the study inputs and model functions are those in Smith et al.'s deposited code, with the FCWC base rate fixed at .019. The values match the reply's replication package exactly.

Data and code: github.com/smourtgos/fcwc-duration-reply

Citation

Mourtgos, S. M., & Adams, I. T. (forthcoming). Interrogation duration and the estimation of false confession wrongful conviction risk: A reply to Smith and colleagues. Journal of Criminal Justice.

Replying to: Smith, T. B., Catlin, M., May, B., Redlich, A. D., Meissner, C. A., & Kelly, C. E. (2026). Simulating the probability of false confession-wrongful conviction conditional on features of police interrogation. Journal of Criminal Justice, 107, 102742.