pling schemes are also used to draw inferences about known populations. The NCVS uses a complex random sampling scheme. In Chapter 7, there is a detailed discussion of case-control schemes that can be especially useful for studying rare events like violence and crime.
Many of the data sets used to study firearms and violence are not random samples from well-defined populations of interest, nor are they exhaustive enumerations of any population. These types of data may provide some information, as described above, but using them to assess the effects of policy can be more complicated.
Accuracy of measurement is an essential criterion for a data source to be useful for understanding firearms and violence. Two key features of accuracy are the validity and reliability of measurement. In general terms, a measure is valid to the degree that it represents the underlying phenomenon of interest, and it is reliable to the degree that it yields the same data over repeated applications. Many of the debates over the relationship between firearms and violence center on questions of validity and reliability. For example, some analysts question the validity of the NCVS for measuring the prevalence of defensive firearms use because, as a survey of crime victims, the NCVS may not fully capture crimes that are averted by the use of firearms. Other researchers question the reliability of one-time sample surveys for measuring rare events, such as defensive use of guns. The chief function of data standardization is to ensure reliability of measurement. The more comprehensive a system, the more likely it will yield valid measurements of the connection between firearms and violence.
Response errors are a vital component of the validity of any data. The validity of data that measure firearms ownership, use, and violence on the basis of respondent self-reports depends on the ability and willingness of persons to disclose highly personal and sometimes incriminating or traumatic information to interviewers. As discussed above, there are reasons to expect response errors in regard to questions about ownership and use, as elicited in the GSS and other gun use surveys. Although there is much speculation on the extent and nature of response errors (see Chapter 5), there is almost no relevant research. Likewise, validity is compromised by nonresponse rates ranging from 20 percent (in the GSS) to over 50 percent in some of the phone surveys used to measure ownership. Without making unsubstantiated assumptions about gun ownership among nonrespondents, the GSS data cannot reveal whether ownership is increasing or decreasing over time.