Survey coverage

The National Crime Victimization Survey (NCVS) is an annual data collection conducted by the U.S. Census Bureau for the Bureau of Justice Statistics (BJS). Each year, data are obtained from a nationally representative sample of about 134,690 households and 224,520 persons, on the frequency, characteristics, and consequences of criminal victimization in the United States. The NCVS collects information on nonfatal personal crimes (i.e., rape or sexual assault, robbery, aggravated and simple assault, and personal larceny) and household property crimes (i.e., burglary, motor vehicle theft, and other theft) both reported and not reported to police. In addition to providing annual level and change estimates on criminal victimization, the NCVS is the primary source of information on the nature of criminal victimization incidents.

Survey respondents provide information about themselves (e.g., age, sex, race and Hispanic origin, marital status, education level, and income) and whether they experienced a victimization. The NCVS collects information for each victimization incident about the offender (e.g., age, race and Hispanic origin, sex, and victim-offender relationship), characteristics of the crime (e.g., time and place of occurrence, use of weapons, nature of injury, and economic consequences), whether the crime was reported to police, reasons the crime was or was not reported, and victim experiences with the criminal justice system

The NCVS is administered to persons age 12 or older from a nationally representative sample of households in the United States. The NCVS defines a household as a group of persons who all reside at a sampled address. Persons are considered household members when the sampled address is their usual place of residence at the time of the interview and when they have no usual place of residence elsewhere. Once selected, households remain in the sample for 3 years, and eligible persons in these households are interviewed every 6 months either in person or over the phone for a total of seven interviews.

Typically, first interviews are conducted in person with subsequent interviews conducted either in person or by phone. New households rotate into the sample on an ongoing basis to replace outgoing households that have been in the sample for the 3-year period. The sample includes persons living in group quarters (e.g., dormitories, rooming houses, and religious group dwellings) and excludes persons living in military barracks and institutional settings (e.g., correctional or hospital facilities) and persons who are homeless.

Nonresponse and weighting adjustments

In 2016, about 134,690 households and 224,520 persons age 12 or older were interviewed for the NCVS. Each household was interviewed twice during the year. The response rate was 78% for households and 84% for eligible persons. Victimizations that occurred outside of the United States were excluded from these reports. In 2016, less than 1% of the unweighted victimizations occurred outside of the United States and were excluded from the analyses.

Estimates in NCVS reports generally use data from the 1993 to 2016 NCVS data files, weighted to produce annual estimates of victimization for persons age 12 or older living in U.S. households. Because the NCVS relies on a sample rather than a census of the entire U.S. population, weights are designed to inflate sample point estimates to known population totals and to compensate for survey nonresponse and other aspects of the sample design.

The NCVS data files include both person and household weights. Person weights provide an estimate of the population represented by each person in the sample. Household weights provide an estimate of the U.S. household population represented by each household in the sample. After proper adjustment, both household and person weights are also typically used to form the denominator in calculations of crime rates.

Victimization weights used in this analysis account for the number of persons present during an incident and for high-frequency repeat victimizations (i.e., series victimizations). Series victimizations are similar in type but occur with such frequency that a victim is unable to recall each individual event or describe each event in detail. Survey procedures allow NCVS interviewers to identify and classify these similar victimizations as series victimizations and to collect detailed information on only the most recent incident in the series.

The weight counts series incidents as the actual number of incidents reported by the victim, up to a maximum of 10 incidents. Including series victimizations in national rates results in large increases in the level of violent victimization; however, trends in violent crime are generally similar, regardless of whether series victimizations are included. In 2016, series incidents accounted for fewer than 2% of all victimizations and fewer than 4% of all violent victimizations. Weighting series incidents as the number of incidents up to a maximum of 10 incidents produces more reliable estimates of crime levels, while the cap at 10 minimizes the effect of extreme outliers on rates. Additional information on the series enumeration is detailed in the report Methods for Counting High-Frequency Repeat Victimizations in the National Crime Victimization Survey (NCJ 237308, BJS web, April 2012).

Standard error computations

When national estimates are derived from a sample, as with the NCVS, caution must be used when comparing one estimate to another estimate or when comparing estimates over time. Although one estimate may be larger than another, estimates based on a sample have some degree of sampling error. The sampling error of an estimate depends on several factors, including the amount of variation in the responses and the size of the sample. When the sampling error around an estimate is taken into account, estimates that appear different may not be statistically different.

One measure of the sampling error associated with an estimate is the standard error. The standard error can vary from one estimate to the next. Generally, an estimate with a small standard error provides a more reliable approximation of the true value than an estimate with a large standard error. Estimates with relatively large standard errors are associated with less precision and reliability and should be interpreted with caution.

To generate standard errors around numbers and estimates from the NCVS, the Census Bureau produced generalized variance function (GVF) parameters for BJS. The GVFs take into account aspects of the NCVS complex sample design and represent the curve fitted to a selection of individual standard errors based on the Balanced Repeated Replication (BRR) technique. The GVF parameters were used to generate standard errors for each point estimate (e.g., counts, percentages, and rates) in these reports.

BJS conducted statistical tests to determine whether differences in estimated numbers, percentages, and rates in these reports were statistically significant once sampling error was taken into account. Using statistical analysis programs developed specifically for the NCVS, all comparisons in the text were tested for significance. The primary test procedure was the Student's t-statistic, which tests the difference between two sample estimates. Unless otherwise noted, the findings described in these reports as higher, lower, or different passed a test at the 0.05 level of statistical significance (95% confidence level). Findings that passed a test at the 0.10 level of significance are noted as such in the text (i.e., 90% confidence level). Caution is required when comparing estimates not explicitly discussed in these reports.

Data users can use the estimates and the standard errors of the estimates provided in these reports to generate a confidence interval around the estimate as a measure of the margin of error. The following example illustrates how standard errors can be used to generate confidence intervals:

According to the NCVS, in 2016 the violent victimization rate among persons age 12 or older was 21.1 per 1,000 persons (see table 2 in Criminal Victimization, 2016, NCJ 251150, December 2017). Using the GVFs, BJS determined that the estimated victimization rate has a standard error of 0.95 (see appendix table 5 in Criminal Victimization, 2016, NCJ 251150, December 2017). A confidence interval around the estimate is generated by multiplying the standard error by ± 1.96 (the t-score of a normal, two-tailed distribution that excludes 2.5% at either end of the distribution). Therefore, the 95% confidence interval around the 21.1 estimate from 2016 is 21.1 ± (0.95 × 1.96) or (19.24 to 22.96). In other words, if BJS used the same sampling method to select different samples and computed an interval estimate for each sample, it would expect the true population parameter (rate of violent victimization) to fall within the interval estimates 95% of the time.

For these reports, BJS also calculated a coefficient of variation (CV) for all estimates, representing the ratio of the standard error to the estimate. CVs provide a measure of reliability and a means for comparing the precision of estimates across measures with differing levels or metrics.

Methodological changes to the NCVS in 2006

Methodological changes implemented in 2006, including the decennial sample redesign that also occurred in 2016, may have affected the crime estimates for that year to such an extent that they are not comparable to estimates from other years. Evaluation of 2007 through 2015 data from the NCVS conducted by BJS and the Census Bureau found a high degree of confidence that estimates for 2007 through 2015 are consistent with and comparable to estimates for 2005 and previous years.

• Criminal Victimization, 2006 (NCJ 219413, December 2007)
• Criminal Victimization, 2007 (NCJ 224390, December 2008)
• Criminal Victimization, 2008 (NCJ 227777, September 2009)
• Criminal Victimization, 2009 (NCJ 231327, October 2010)
• Criminal Victimization, 2010 (NCJ 235508, September 2011)
• Criminal Victimization, 2011 (NCJ 239437, October 2012)
• Criminal Victimization, 2012 (NCJ 243389, October 2013)
• Criminal Victimization, 2013 (NCJ 247648, September 2014)
• Criminal Victimization, 2014 (NCJ 248973, August 2015)
• Criminal Victimization, 2015 (NCJ 250180, October 2016)

 Methodological changes to the NCVS in 2016

The ability to compare 2016 estimates of crime to 2015 or other years is limited due to a redesign of the NCVS sample. In 2016, BJS introduced new areas to the NCVS sample to reflect population changes based on the 2010 Decennial Census and to produce state- and local-level victimization estimates. Therefore, 2016 victimization estimates are not comparable with NCVS data from prior years. BJS will continue to monitor victimization estimates in future years to determine whether the impact of the sample redesign is unique to the 2016 estimates. Following the NCVS 2006 decennial redesign, BJS found that data since 2007 were comparable