
How We Measure Crime Creates Biased Surveillance Systems
Crime data looks like a neutral record of reality, but it’s shaped by what gets reported, where police patrol, and how officers act.
以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。
Crime data looks like a neutral record of reality, but it’s shaped by what gets reported, where police patrol, and how officers act.
SUMMARY
- American crime data is used to make political and policing decisions that shape communities.
- This data has predictable gaps and biases shaped by reporting, enforcement, and officer discretion.
- When this data guides surveillance decisions, it can reinforce existing inequalities.
Crime statistics produced by law enforcement agencies are widely cited in news reports, police press releases, and political debates. Police records are often treated as accurate measures of crime, but they aren’t neutral measurements of reality.
Crime data reflects what gets reported, where police patrol, and how officers respond, so crime statistics can reflect patterns of policing as much as patterns of crime.
Note
Crime is also measured by federal agencies and independent researchers, but here we are specifically interested in what data is used to make decisions in
Police crime statistics measure offenses reported to or recorded by police, not every crime that happens.
Departments collect different kinds of data, including 911 calls, incident reports, and arrests. These records can be used to track patterns over time, but they measure different things: a call isn’t necessarily a crime, and an arrest doesn’t always lead to a conviction.
What gets counted depends on who reports an incident, where officers are deployed, and how police document what happened.
It’s difficult to know how many crimes happen because many are never reported or detected. Estimates of underreporting vary by offense, ranging from 17% to 68%[1]; this means that some crimes are undercounted more than others. One estimate puts the number of total unreported crimes at more than 3.4 million per year.[2]
Crimes go unreported for a number of reasons:[3][4]
This missing data isn’t random. It reflects who feels able to report a crime and impacts the percentage of crimes that come to police attention. Crimes that happen in private or involve vulnerable people may be especially underrepresented.
Similar incidents do not always produce the same police response. Officers make decisions in the field that impact who appears in police records and how those encounters are documented.
Officer discretion gives individual officers the ability to choose how they respond to a situation. That includes who to stop or search and whether to arrest someone, issue a warning, or let them go.
Those decisions are shaped by laws, which leave room for interpretation,[5] department policies, and individual judgments. Time of day, location, and assumptions about what is “normal” or “acceptable” in a neighborhood can influence an officer’s decision. This means that the same behavior can lead to different outcomes depending on the officer, the individuals involved, and where the incident happens.
Racial and class disparities in discretionary policing are well documented.
These differences impact who is stopped, searched, and entered into police records. The resulting data can make unequal enforcement look like some groups of people commit more crimes.
Mapping White Collar Crime
White Collar Crime Risk Zones
This satirical site “predicts” financial crime in the same way predictive policing software maps street crime. Why map a $100 convenience store robbery but not millions of dollars stolen in a downtown high rise? What we choose to measure shapes where we expect crime to happen and whom we see as suspicious or criminal.
New laws and enforcement policies change which activities bring people into contact with police. Department leaders and elected officials help decide which laws receive attention. A city might deprioritize minor traffic violations, marijuana possession, or “quality of life” offenses like fare evasion to free up resources for other offenses.
These choices shape not only who is policed, but also what gets recorded.
Police also have discretion in how incidents are classified, documented, and entered into law enforcement databases. This includes decisions about:
Changes in the way incidents are classified or in reporting practices can alter reported crime trends without any underlying change in resident behavior. These administrative changes might not be obvious to the public.
Charging practices offer another example of how enforcement data can change.
Charge-stacking means bringing overlapping or multiple charges arising from the same incident.[11] Prosecutors might file distinct charges for one encounter[12] or charge different degrees of an offense.[13] Relatedly, sentence enhancements can increase penalties based on factors like weapon use or proximity to a school.
These practices can increase prosecutors’ leverage in plea negotiations, which is how most criminal cases are resolved.[14] They can also make charging data look more severe or extensive than the incident would suggest. Charges aren’t the same as recorded offenses: whether multiple charges impact a published crime rate depends on how that statistic is compiled.
When municipal leaders compare trends, they need to know the story behind the data. Are they looking at incidents, arrests, charges, or convictions? Each provides a different picture.
Overpolicing is excessive police enforcement or surveillance in certrain communities.
Higher patrol concentrations account for 57% higher arrest rates in Black neighborhoods.[15]
When officers patrol one neighborhood more heavily than another, they have more opportunities to detect offenses there. Even if the underlying level of crime is similar, the heavily patrolled neighborhood may produce more stops and arrests. One study found that differences in police presence helps to explain a 57% higher arrest rate in Black neighborhoods.[16] The resulting data can reinforce the belief that those neighborhoods are more dangerous.
In this way, policing patterns can produce the very statistics used to justify continued policing.
Modern surveillance systems might rely on decades of incomplete police data. That data carries the effects of decisions made by elected officials, prosecutors, police leaders, and individual officers.
When a system uses those records to determine risk, it can reproduce the assumptions behind them.
Surveillance technology vendors and police departments might use historical records to define what counts as “suspicious” or “high risk.” Place-based predictive policing tools, for example, can use past incident data to identify where police should focus their attention. Other surveillance technologies have different functions and data sources, but can still be deployed according to those same risk assessments.
Police leaders can use maps and those risk scores to decide where to send officers or install surveillance, directing more attention to places already labeled high risk.
When a software model treats higher numbers of recorded incidents or arrests as evidence of greater risk, it can learn patterns created partially by earlier policing decisions. Its rules and training data decide what gets flagged as suspicious. Without additional context, the system can’t reliably separate patterns of crime from patterns of enforcement.
Automated systems can apply these judgments across records or locations, leading to repeated flagging of people or places for attention.
These issues lead to concentrated policing, which we learned earlier produces more records (stops, arrests, and documented incidents).
If officials interpret those records as evidence of higher risk, they might direct more officers and surveillance tools to those same areas. This will generate even more records, while other neighborhoods receive less attention.
Some surveillance tools can monitor the same places continuously. Where surveillance devices are installed matters: activity in monitored areas becomes more visible and more likely to enter official records than similar activity in places without devices.
When historical police records shape future deployment decisions, neighborhoods that were heavily policed in the past might continue to be treated as high risk, even when local conditions change.
This can create a form of lock-in: past enforcement influences today’s risk assessments, which shape tomorrow’s patrol deployments and surveillance. Without regular review, a neighborhood might have little opportunity to escape a “high risk” label based partly on its policing history.
Data-driven surveillance is often presented as objective and evidence-based. But if officials don’t examine how the underlying data was created, these systems can carry past inequalities into future decisions and make those decisions appear fair.
Before using crime data to justify new surveillance, municipalities should examine how the data was produced, which activities they represent, and which communities receive the most police attention.
Brayne, S. (2017). Big Data Surveillance: The Case of Policing. American Sociological Review, 82(5), 977–1008. https://doi.org/doi.org/10.1177/0003122417725865 ↩︎
Sanchez, S. (2019). Ch 2. Dark or Hidden Figure of Crime. In Introduction to the American Criminal Justice System (1st ed.). Open Oregon Educational Resources. https://openoregon.pressbooks.pub/ccj230/chapter/2-1-dark-figure-of-crime-or-unreported-crime/ ↩︎
Chapman, A., Grylls, P. J., Ugwudike, P., Gammack, D., & Ayling, J. (2022). A Data-driven analysis of the interplay between Criminological theory and predictive policing algorithms. Conference on Fairness, Accountability and Transparency. https://doi.org/10.1145/3531146.3533071 ↩︎
Glaser, J. (2024). Disrupting the Effects of Implicit Bias: The Case of Discretion & Policing. Daedalus, 153(1), 151–173. https://doi.org/10.1162/daed_a_02053 ↩︎
Pierson, E., Simoiu, C., Overgoor, J., Corbett-Davies, S., Jenson, D., Shoemaker, A., Ramachandran, V., Barghouty, P., Phillips, C., Shroff, R., & Goel, S. (2020). A large-scale analysis of racial disparities in police stops across the United States. Nature Human Behaviour, 4(7), 736–745. https://doi.org/10.1038/s41562-020-0858-1 ↩︎
NYCLU. (n.d.). Stop-and-Frisk Data. NYCLU. Retrieved March 26, 2026, from https://www.nyclu.org/data/stop-and-frisk-data ↩︎
Lofstrom, M., Hayes, J., Martin, B., & Premkumar, D. (2022, October). Racial Disparities in Traffic Stops. Public Policy Institute of California. https://www.ppic.org/publication/racial-disparities-in-traffic-stops/ ↩︎
Alden, H. O. (2020). Discovering the Victim: The Enduring Problem with “High-Crime Areas.” Stanford Journal of Civil Rights & Civil Liberties, 16(2), 385–420. ↩︎
Feigenberg, B., & Miller, C. (2024). Class Disparities and Discrimination in Traffic Stops and Searches. National Bureau of Economic Research, Inc. ↩︎
Harvard Law Review. (2023, March 10). Stacked: Where Criminal Charge Stacking Happens — And Where it Doesn’t. Harvard Law Review. https://harvardlawreview.org/print/vol-136/stacked-where-criminal-charge-stacking-happens-and-where-it-doesnt/ ↩︎
Bennett, P. (1979). Prosecutorial Overcharging. US Department of Justice. https://www.ojp.gov/ncjrs/virtual-library/abstracts/prosecutorial-overcharging ↩︎
Vera Institute of Justice. (2020, September). In the Shadows. Vera Institute of Justice. https://www.vera.org/publications/in-the-shadows-plea-bargaining ↩︎
Chen, M. K., Christensen, K. L., John, E., Owens, E., & Zhuo, Y. (2025). Smartphone Data Reveal Neighborhood-Level Racial Disparities in Police Presence. Review of Economics and Statistics, 1–9. https://doi.org/10.1162/rest_a_01370 ↩︎
Ashe Dryden (she/they)
Mastodon
An independent researcher and technologist studying policing, surveillance technology, and racial inequality. She examines how these systems intersect with patterns of residential segregation and how stronger governance and oversight could reduce harm. Her background is in public interest technology, computer sc
正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。