How to massively expand victimization data

A significant issue with our crime data is that it only includes crimes reported to the police.¹ Most crimes that occur are never reported (especially property crimes). This is a long-term trend, according to a Bureau of Justice Statistics report that compared reported crime to victimization data. In 2020, for example, only about 40% of violent crimes, including 23% of sexual assaults, and 33% of property crimes were reported to the police. We know about the underreporting to police issue because there is a national survey that asks people whether they were victimized. The survey asks fairly detailed questions about the incident, including whether the respondent reported it to the police and, if not, why they didn’t. But this survey is fairly limited due to its small sample size.

The National Crime Victimization Survey, which the Bureau of Justice Statistics runs, surveys about 240,000 people annually.² This is a massive sample in most respects, but consider a few things. Most people never experience a crime in any given year. When examining certain victim groups, such as Black women under 30 who were assaulted, the number of respondents that meets these criteria may be too small to use reliably. This survey is also representative only nationally or regionally, with insufficient data to generate local or even state estimates. The Bureau of Justice Statistics is working on creating subnational estimates for their complete survey. However, as far as I’m aware, this is still a work in progress - and my proposal is still a valuable supplement to the data even after they finish this project. While you technically can get subnational geographic information for respondents, this still has a small sample size at any local geography and requires accessing the restricted data from the BJS rather than using their public data.³ As crime is highly local, with major differences between and even within cities, local estimates are far more critical than national numbers. We want to know, for example, whether reporting went down in City A, as that will impact our understanding of crime there. National reporting trends don’t tell us much about any particular city. So we need a way to massively increase the survey’s sample, at least for a bare minimum number of questions.

There is a simple - and, importantly, politically feasible - solution to this issue. Each year the Census Bureau surveys about 3.5 million households for its American Community Survey, essentially a longer form of the Decennial Census. The questions asked include respondent demographics, income, education, citizenship, language spoken, the field of work, household type, disability status, and household location. In other words, it already asks about nearly every variable that criminologists are interested in when studying causes and correlates of crime. What it does not ask about is crime victimization. It should.

Adding five questions about crime victimization to this survey can significantly expand our understanding of crime in this country. Each respondent should be asked the following:

1. Did you experience a [crime and definition for a set number of crimes] during the past year?

If they say no, then there won’t be any more questions to answer. If they say yes, then for the most serious crime chosen:

2. How many times did this occur?

The following are about the most recent of these incidents that occurred.

3. What month and year did the crime occur?

4. Did you report this incident to the police?

5. If you did not report it, why not? (The National Crime Victimization Survey has a standard set of answers for this question, including believing that the police wouldn’t help and not wanting to get the offender in trouble.)

Of course, more questions could be asked, but these five are key. You could expand these questions by asking about all crimes experienced, not just the most serious or recent ones. And asking about multiple crimes would make it more than five questions. So you could randomize which crime(s) are asked about, but most people won’t experience a crime in the previous year so for these people it adds few questions.

This is, I believe, a proper minimal set of questions to achieve our goal. Using the scale of the American Community Survey scale allows us to accurately measure crime victimization at local levels and among small subgroups of people. This is certainly no panacea; the American Community Survey only gets to local levels for large jurisdictions. But still a vast improvement over the NCVS’s geographic impreciseness. And it will cover a sample almost 15 times as large. These questions should also be included in the Decennial Census, allowing us an extremely granular look at victimization in the country, albeit only once a decade.

To be clear, this proposal is to supplement, not replace, the NCVS. These five questions are a small, shallow fraction of what is asked in the complete NCVS survey. We need both. And ideally, the NCVS will also be expanded to give their longer study to many more people. But surveying more people is very expensive; adding five questions to a survey already taken by millions each year is far more accessible. It would require few additional resources but have a significant return. These data should also be a supplement to official crime data. For example, we can use it alongside FBI crime data to measure the number of crimes. If City A reported 100 robberies and we know from victimization data that people in City A only report 50% of robberies, we’d know that the actual number is 200. And that 50% rate is likely far more accurate to that city - it is a survey, after all, so estimates the actual amount - than what we’d get from national or subnational estimates.

Crime is exceptionally local and affects certain groups more than others. In our current time of elevated murders and gun violence, coupled with lower other crimes, we need more and more local crime data than ever before. This includes more victimization data that are detailed enough to look at all states and specific municipalities. And that allows us to see if victimizations change for small subgroups for certain crimes and at certain places. We can do this; all that’s necessary is adding five short questions to a survey already taken by millions.

1. Another issue is agencies not reporting data to the FBI or releasing it publicly. But that’s worth an entirely different post.

2. It is limited to those 12 and older.

3. This is a much more involved process than just getting IRB approval and requires physically going to a government “Federal Statistical Research Data Center” to access the data.