What people are saying
Nobody has said anything yet.
Brand New Thoughts/Justice & Data
Dirty Data: How Wrongful Convictions Can Distort What We Think We Know About Crime
A wrongful conviction is first a human catastrophe. It can also become a bad data point that follows a person through records, research and public arguments long after the verdict.
By ElleJanelle|August 7, 2026|4 min read
What Happens After the Wrong Person Becomes the Official Answer?
A conviction does more than send a person to prison. It creates an official record.
That record can be counted, searched, cited, summarized and fed into other systems. It can appear in criminal histories, court databases, academic datasets, news stories, risk assessments and political arguments about crime.
That is the problem ElleJanelle started worrying through on Brand New Thoughts while discussing the case of Pierre Rushing. Rushing was convicted of first-degree murder in Alameda County in 2011 and sentenced to 50 years to life. In July 2026, the key eyewitness from his trial testified at an evidentiary hearing that Rushing was not the shooter and said he had committed perjury at the original trial. KQED reported on the recantation and hearing.
The live raised a deceptively simple question: if the wrong person becomes the official solution to a crime, where does that wrong answer travel next?
Bad justice can become bad data.
Brand New Thoughts
The answer requires one important correction first.
A Conviction Is Not the Same Thing as a Crime Statistic
It is tempting to assume that every wrongful conviction automatically inflates a police department's clearance rate. That is not quite how the FBI's Uniform Crime Reporting system works.
The FBI defines a cleared offense primarily as one cleared by arrest or by exceptional means, not by a later conviction. The FBI's clearance guidance makes that distinction explicit.
That means a wrongful conviction does not necessarily create a new offense or directly change the original count of crimes reported to police. The crime itself still happened. And the clearance may have been recorded earlier in the process.
That correction actually makes the larger question more interesting.
The data problem is not simply "wrong conviction equals fake crime rate." It is that criminal justice data is made of many different records created at different stages, and an error at one stage can still contaminate the systems that depend on it.
Where one wrong record travels
Consider what a conviction can become downstream.
A conviction can become:
- a criminal-history record;
- an input into sentencing decisions;
- part of a dataset used to study recidivism;
- a feature in risk-assessment systems;
- a case counted when researchers examine conviction patterns;
- evidence in a news story about who commits particular crimes;
- a rhetorical example in an argument about race, policing or punishment.
Researchers studying criminal justice algorithms have already warned that the inputs themselves can be unstable. One study of a pretrial risk tool found that booking charges that never resulted in convictions could still change recommended supervision levels for some defendants. The broader lesson is straightforward: a model can only be as meaningful as the records it has been given.
The National Registry of Exonerations exists precisely because official convictions are not infallible. The Registry describes itself as a living archive collecting information on known exonerations in the United States. Its cases document mistaken identifications, false accusations, official misconduct, false confessions and unreliable forensic evidence. Explore the Registry.
Four questions to ask of any crime statistic
Numbers gain authority because they look settled.
"Convicted" looks more definitive than "accused." A database field looks cleaner than a contested human story. Once the system assigns a value, the value can move farther than the argument that produced it.
That creates a data-literacy problem.
When someone cites a criminal justice statistic, we should ask:
What exactly is being counted? Arrests? Reported offenses? Clearances? Charges? Convictions? Incarcerated people?
At what stage was the information recorded?
Can that record later be corrected?
If it is corrected, do all downstream datasets update too?
Those questions matter because criminal justice statistics are frequently used as if they describe criminal behavior directly. Often they describe the behavior of both the public and institutions: who gets reported, stopped, arrested, charged, prosecuted, convicted and entered into the database.
Wrongful Convictions Reveal the Difference Between Reality and Records
This is not an argument that crime data is useless.
It is an argument that a record of institutional decisions is not identical to reality itself.
The distinction is especially important when the consequences are severe. Eyewitness identification, for example, has been a major factor in wrongful convictions. The National Institute of Justice has long documented the risks of mistaken eyewitness identification and the reforms developed to reduce them. NIJ's overview is here.
If a justice system can sometimes record the wrong person as guilty, then responsible analysis has to preserve room for error.
That is what "dirty data" means in this context. Not that every number is false. Not that crime cannot be measured. But that institutional data has a provenance, and provenance matters.
The Data Point Is a Person
The easiest thing to lose when talking about datasets is scale.
A single wrong row in a spreadsheet sounds small.
A single wrongful conviction can mean decades of a human life.
That is why the data question and the justice question belong together. The same error that harms one person can also become evidence used to describe a neighborhood, a demographic group, a criminal justice policy or the effectiveness of an institution.
Before building an argument from the numbers, we should know what the numbers actually record.
Watch the episode
What should count as clean criminal justice data?
Add to the DiscussionWhich criminal justice statistics do you trust most, and what would you want to know about how they were produced before using them in an argument?