The Crime Survey That Asks Victims Instead of Asking Police Departments

Ask a police department how much crime there is, and you will learn how much crime people told the police about. Those are two different numbers, and the gap between them has a name: the dark figure of crime. The federal government has been trying to measure that gap for more than fifty years, by doing something so simple that it is almost embarrassing. It knocks on doors and asks.

The program is the National Crime Victimization Survey, the NCVS, and it is the most useful crime statistic that most politicians have never read. Conservatives should read it more than anyone. We are the ones who say that victims matter and that government exists to protect them, and this survey is the one national instrument built around asking the victims themselves.

Why a second instrument was needed

By the 1960s, American crime had risen and the public was alarmed, and the only national measure was the FBI's Uniform Crime Reporting program, which I describe in Why the FBI's Switch to NIBRS Made National Crime Numbers Harder to Read. It counts offenses known to police. Statisticians had understood for a long time what that means: if a victim does not call, the crime is not in the count.

In 1967 the President's Commission on Law Enforcement and Administration of Justice, appointed by Lyndon Johnson, published its report on the problem of crime. The commission recommended a national crime survey, and the National Opinion Research Center at the University of Chicago ran an early national victimization survey of about 100,000 households. A key early finding was that many crimes never reached the police. That finding set the course for a permanent national survey. The Census Bureau began fielding one in 1973, under what was then called the National Crime Survey. In the early 1990s the survey was redesigned, with new questions added, and renamed the National Crime Victimization Survey. The Bureau of Justice Statistics, part of the Justice Department, publishes it. The Census Bureau does the fieldwork.

So the survey is old enough to have a long record, and it has been run by a federal statistical agency under the same basic logic since 1973. That is an asset. A trend line that spans decades under one federal agency is rare in social data, though the redesign of the early 1990s means a careful reader does not splice the old and new series without checking.

How the survey works

Let me define the pieces, because the design explains both the strengths and the limits.

The sample. The NCVS draws a national sample of households. The Census Bureau's interviewers contact each sampled household and interview every member who is twelve years old or older. That age line matters: the survey says nothing directly about crimes against young children.

The panel. A household stays in the sample for three and a half years. It is interviewed every six months, for a total of seven interviews. That structure lets the survey follow households over time and helps keep the interview periods consistent.

The screening questions. Interviewers do not ask respondents "Were you the victim of a burglary?" and leave it there. They use a battery of cues, asking whether anything was stolen, whether anyone broke in, whether anyone attacked or threatened the respondent, and so on. Cues are important because people do not always label their experiences with legal terms. Someone who was shoved and robbed in a parking lot may not volunteer that they were "the victim of a robbery."

The incident report. When a respondent describes something, the interviewer collects details: what happened, where, when, who the offender was, what the relationship was, whether a weapon was involved, what injuries followed, whether the police were called and, if not, why not.

The weights. The Census Bureau and the BJS weight the sample so that it represents the national population of households and persons, and then produce estimates of the number of victimizations and the rate per 1,000 persons or households.

The survey covers personal crimes such as rape and sexual assault, robbery, aggravated assault, simple assault and personal theft, and household crimes such as burglary, motor vehicle theft and other theft. These are the categories that map most closely onto what the FBI counts, which is why comparisons between the two are possible at all.

What the survey can measure that the FBI cannot

Here is the argument for taking the NCVS seriously, in numbered form.

One: it measures unreported crime. This is the point of the whole exercise. If a respondent says a crime happened and also says it was never reported to the police, the survey has found a crime that the FBI's count could never include. Over the history of the survey, a large share of victimizations has gone unreported, and the share varies a great deal by type of crime. Check the BJS's current publications for the exact percentages, because they move from year to year, and I will not give you a number I cannot source here.

Two: it measures why victims do not report. The survey asks, and the answers are valuable. Some victims believe the police could not or would not do anything. Some consider the matter private or minor. Some fear retaliation. Some report to another official instead, such as a school administrator or an employer. Each answer is a clue to a policy problem. If victims say they do not call because they expect nothing to come of it, that is a verdict on how the system treats them, and it should concern anyone who thinks the first job of government is to protect the innocent.

Three: it measures victim characteristics and circumstances. Age, sex, race of the victim, the victim's relationship to the offender, the time of day, whether the offender was armed: the survey records all of it from the victim's side. That helps in understanding who is hurt and how, and it helps in checking the claims people make about crime and its causes.

Four: it is independent of police practice. If a department changes how it classifies offenses, or tightens or loosens its recording, the FBI figures for that agency can move without any change in the real world. The survey asks the same questions of households whatever their police department does. That makes it a check on police-recorded data, in either direction.

This fourth point is a conservative point. We distrust any government institution that grades its own homework. The NCVS gives the public a second set of books that the local police chief does not keep.

What the survey cannot measure

Now the other half, which any honest citation requires.

It cannot count homicide. The dead cannot be interviewed. For murder, the FBI's data and public health records are the right sources, and they are also the most reliable crime figures we have, for reasons I take up in Seven Things Crime Statistics Cannot Tell You, and Two Things They Can.

It cannot count crimes against businesses. The survey is of households and persons. A shoplifting ring that hits a drugstore chain, a burglary of a warehouse, a fraud against a company: none of those are victimizations of a household, so they are outside the sample. That matters for the argument about retail theft in Organized Retail Theft Is a Business Model, Not a Cry for Help. The household survey is the wrong instrument for that question, and anyone who cites it as proof that retail theft is not a problem has picked the wrong tool.

It cannot count crimes against children under twelve. The survey does not interview them.

It does not reach people outside the household frame. The sample is of households. Persons living in institutional settings, such as correctional or hospital facilities, persons on military bases, and people who are homeless are outside it. These are populations in which victimization may be unusually high.

It cannot reliably produce small-area rates. The sample is designed to give good national estimates and estimates for a few large domains. A national sample cannot tell you the robbery rate in your neighborhood, or even in most cities, because the number of respondents in any small area is tiny. You should not expect the survey to resolve a debate about a specific town.

It depends on memory and candor. Respondents may forget, may place an event in the wrong six-month period, or may not want to describe a painful experience to an interviewer. The survey tries to manage this with a bounded recall period and careful questions, but the limitation remains. This cuts toward undercounting for sensitive offenses such as sexual assault and, in a different way, toward misdating events.

It has sampling error. Any survey estimate has a margin. A year-over-year change that falls within the margin is not evidence of a change. The BJS reports standard errors and tests of significance, and the responsible way to cite the survey is to say whether the change it reports is statistically significant.

That is a long list, and it is a fair list. A tool is useful in proportion to your honesty about what it was built to do.

The steelman against the survey

Skeptics of the NCVS make arguments that deserve a hearing, and here is the best version.

The survey's critics say that it counts experiences, not legal crimes. A respondent decides what counts as an assault or a threat, and the survey's classifiers then fit the description to a category. Two respondents describing similar events may label them differently. A survey that relies on self-report cannot verify that a crime occurred; it only records that someone says so. And a statistic built on a sample can swing on a handful of respondents when an offense is rare.

There is truth in each of these points. The first is a reason to be modest about small categories and sensitive offenses. The second is a reason to prefer police records for offenses the police actually process, and homicide above all. The third is a reason to read confidence intervals before reading headlines.

But none of it supports the dismissal some partisans offer, which runs: "A survey is only what people say, so ignore it." People are the victims. Their account is the evidence. A court of law relies on testimony every day, and the standard is not that testimony be uncorroborated by a government database but that it be credible, specific and tested. The NCVS tests it with a cued interview, a bounded recall period and a consistent design over decades. And note what the police-reported data relies on: a victim deciding to call and an officer deciding to write it down. That is also testimony, filtered through two people instead of one.

How both sides abuse the survey

The best conservative posture is to cite the NCVS with its limits, because both sides misuse it.

The left tends to use it to argue that crime is overstated, pointing to a long-run decline in survey victimization rates. That decline is real, and a fair reader acknowledges it. But the survey's national rate says nothing about a particular city's violence, nothing about homicide, and nothing about business crime, so using it to dismiss a community's alarm is an overreach.

The right sometimes goes the other way, pointing to the unreported share of crime as proof that police data hides a catastrophe. The unreported share is real and important, but it has always been large. A large dark figure is not evidence of a change; it is a standing feature of crime measurement. To claim that crime has risen because the survey shows many unreported crimes is to confuse a level with a trend.

The honest use is narrower and stronger. Use the NCVS to show the scale of what the police never hear about. Use it to show why victims stay silent. Use it, alongside police data, to see whether the two instruments are moving together. And use it to remind policymakers that a failure of reporting is a failure of public confidence, which is a governance problem.

The Founders and the victim

There is a constitutional thread here worth pulling. The Sixth Amendment gives the accused a long list of protections, the right to a speedy and public trial, to confront witnesses, to counsel. The original Constitution says little about the victim, and the omission was not a judgment about the victim's importance but a reflection of the era's model of prosecution, in which the state brought the case in the name of the public. We have since learned, and the movement described in The Victims' Rights Movement Began With a 1982 Presidential Task Force spent decades teaching, that a justice system which forgets the person who was hurt loses its moral footing.

The NCVS is a small part of that correction. It is a government instrument whose whole premise is that the victim's account counts. A republic that takes the dignity of the individual seriously should want to know what individuals suffered, whether or not they filled out a police report, and what the government did about it.

A short guide to citing the NCVS

If you plan to use the survey in an argument, here are five rules.

1. Name the survey and the year. "The National Crime Victimization Survey, from the Bureau of Justice Statistics" is specific. "Studies show" is not.
2. Say what it covers. Households and persons twelve and older, personal and household crimes, national scope.
3. Mention the margin. If the change you cite is within the survey's sampling error, say so, or do not cite the change.
4. Pair it with another source. For violent crime, set it beside police-reported data. For homicide, use the FBI and public health records instead.
5. Do not use it for what it cannot do. No neighborhood claims, no business crime, no homicide.

Follow those, and you will be better informed than most of the people who argue about crime on television.

The case in brief

1. Police statistics count only the crimes that reach police, so a second national instrument was needed, and the Census Bureau began fielding one for the Justice Department in 1973.
2. The NCVS interviews members of a household sample who are twelve or older, repeatedly over three and a half years, using cued questions and detailed incident reports.
3. It can measure unreported crime and the reasons victims give for not reporting, and it does not depend on police recording practices.
4. It cannot count homicide, crimes against businesses or against young children, or reliably estimate rates for small areas, and its estimates carry sampling error.
5. Both parties abuse it, and the honest course is to cite it with its limits and beside police data.

What you can do: look up the Bureau of Justice Statistics page for the NCVS, read its latest summary table, and note how much of the crime it reports was never reported to the police.