Skip to main content
Unpublished Paper
Predicting Violence
ExpressO (2011)
  • Shima Baradaran
The last several years have seen a marked rise in state and federal pretrial detention rates. There has been very little scholarly analysis of whether increased detention is reducing crime, and the discussion that has taken place has largely relied on small scale local studies with conflicting results. This article asks whether the United States is making substantially mistaken judgments about who is likely to commit crimes while on pretrial release and whether we are detaining the right people. Relying on the largest dataset of pretrial defendants in the U.S., this article determines what factors, if any, are relevant in predicting “dangerousness” pretrial and what percentage of defendants can be released safely before trial. Prior work in this area disagrees as to whether the current charge or past convictions are relevant as predictors of future crimes, whether flight risk is linked to pretrial violence, and whether judges can accurately predict which defendants are dangerous. This article—for the first time—relies on empirical methods and a nationally representative fifteen-year dataset of over 100,000 defendants to determine what factors are reliable predictors of who will commit pretrial crime. This analysis suggests several important conclusions. First, judges often detain the wrong people. Judges often overhold older defendants, defendants with clean records, and defendants charged with fraud and public order offenses. Second, using our model, judges would be able to release 25% more defendants while decreasing both violent crime and total pretrial crime rates.
Publication Date
February 23, 2011
Citation Information
Shima Baradaran. "Predicting Violence" ExpressO (2011)
Available at: