Questioning Canadian criminal incidence rates: a re-analysis of the 2004 Canadian Victimization Survey.

• Author: Nazaretian, Zavin

Beginning in 1988, Statistics Canada implemented a victimization survey as part of the Canadian General Social Survey (GSS). This effort followed the success of the National Crime Victimization Survey (NVCS) in the United States in 1976. Victimization surveys are designed to complement police data from the Uniform Crime Report (UCR) and provide estimates of crime incidence determined from victims' accounts. Further, victimization surveys allow criminologists to investigate unreported crimes, the victims of crime, and the contextual setting of criminal incidents. In addition to the United States and Canada, the United Nations and United Kingdom each conduct regular victimization surveys to provide researchers and policy makers with the empirical data needed to understand crime and to protect their citizens.

Like all surveys, victimization surveys can suffer from inaccurate recall by survey respondents. Unfortunately, most official analysts that estimate criminal incidence rates rely on a practice known as capping to deal with inaccurate responses (Genn 1988; Farrell and Pease 2007). Capping describes the process whereby data analysts attempt to correct for response bias by limiting the number of individual victimization incidents that any individual victim can represent in official rates. For instance, in the Canadian Victimization Survey (CVS), although each respondent is queried about the details of up to 20 crimes, official statistics are based on a cap of only 3 incidents per respondent. That is, a respondent who reports being victimized 20 times only accounts for 3 incidents in the official tabulation of national incidence rates (Statistics Canada 2005).

The practice of capping is particularly troubling, given that abundant empirical evidence indicates that individuals who are victims of a crime are likely to be victimized again (Farrell and Pease 2001; Farrell, Tseloni, and Pease 2005). The prevalence of repeat victimization suggests that individuals reporting a larger number of victimizations may be relaying genuine accounts; capping crime reports, therefore, may lead to a substantial under-estimation of criminal incidence (Farrell, Sousa, and Weisel 2002). The current study addresses the degree to which capping affects estimates of criminal incidence in Canada. Estimates of criminal incidence using different capping levels from the 2004 CVS indicate that capping results in a significant under-estimation of the number of crimes that take place in Canada. Further, the patterns of repeat victimization seen in the CVS suggest that many reports may be genuine accounts of victimization incidents. Given the uncertainty regarding the effects of capping, this article concludes that models of response accuracy should be developed, and that a range of criminal incidence rates should be presented in official reports.

Background

Capping in victimization surveys

Any attempt to estimate crime incidence from survey data must weigh the risk of not counting actual victimizations versus that of including incidents that did not occur within the survey time window. When collecting data based on retrospective victimization surveys, researchers generally attempt to correct for the contradictory effects of memory decay and telescoping. Memory decay occurs because respondents tend to forget events that happened in the more distant past. Telescoping, in contrast, refers to the inclusion of criminal incidents that did not occur inside the survey time window. In general, researchers believe that memory decay leads to an under-reporting of criminal incidents, whereas telescoping leads to an over-estimation of criminal incidents (Gottfredson and Hindelang 1977; Schneider and Sumi 1981; Skogan 1981; Wolfer 1999; Farrell et al 2002).

Early victimization researchers questioned the effect of respondent recall on reported victimization rates (Biderman 1980; Reiss 1980; Skogan 1981). In each case, the primary question being raised is: Do attempts at preserving the integrity and accuracy of victimization surveys skew the gross rate of crime incidents reported? Given that early victimization surveys suggested that there is far more crime than police-based estimates indicate, analysts often assumed that victimization surveys over-estimate criminal incidence. Yet, several studies have indicated that, although response accuracy problems bias estimates of the number of crimes in any specific time period within the survey time window (e.g., one month in a year-long survey window), these biases are minimal over the entire time window. Biases are mitigated because the processes of telescoping and memory decay offset each other and lead to a generally accurate estimate of the total number of criminal incidents over the entire survey time window (Gottfredson and Hindelang 1977; Schneider and Sumi 1981).

Currently, nearly every large victimization survey relies on a methodological convention referred to as capping to account for response bias (Skogan 1981; Tseloni and Pease 2005; Farrell and Pease 2007). Capping occurs when analysts simply limit the number of victimization incidents of each crime type that any one person can represent in the total crime incidents calculation. For example, in the CVS, an individual who reports 20 assaults represents only 3 incidents in the calculation of the assault incidence rate (Statistics Canada 2005: 15). Although different victimization surveys use different methods of capping, nearly every large victimization survey employs some type of cap on victimization incidents, typically around 3 or 5 incidents per person (Genn 1988; Farrell and Pease 2007). Capping is based on the general assumption that respondents over-report criminal victimizations; yet this practice is not based on any specific model of respondent recall. The decision to trust respondents for some crime reports and not others has been called "schizoid at best and hypocritical at worst" (Farrell and Pease 2007: 12). By arbitrarily determining which responses can be trusted, victimization surveyors risk under-estimating the criminal incidence rate and misreporting crime patterns in Canada.

Effects of capping on incidence rates

To what extent does capping attenuate criminal incidence rates? Research using the NCVS and British Crime Survey (BCS) indicates that capping leads to a substantial reduction in criminal incidence rates. Planty and Strom (2007) calculated the rate of victimization in the United States by incorporating crimes that were counted as series-incidents. Series-incidents are crimes that are repeated against the same victim and are recorded as one victimization in the NCVS. Planty and Strom (2007) show that accounting for series-incidents increases estimates of crime incidence anywhere from 62% (in 2000) to 174% (in 1996). In terms of the number of incidents, analysts estimated that, in 2004, the total number of violent crime incidents in the United States was 27,375,186, compared to the official number of 11,365,000. Similarly, Farrell and Pease (2007) examined the BCS, which uses a cap of five incidents per respondent. Farrell and Pease (2007) show that removing the cap of five leads to a 52% increase in personal victimization. This evidence points to the importance of questioning the practice of capping because this practice may lead to a drastically inaccurate depiction of crime.

Capping criminal incidents has the effect of essentially under-representing repeat victimization. However, research suggests that repeat victims are prevalent and that a small portion of victims represent a disproportionate percentage of criminal victimization (Farrell and Sousa 1997). For instance, research on repeat victimization in Canada indicates that 33% of violent crime victims and 25% of those suffering from household victimization were victimized more than once (Perreault, Sauve, and Bums 2004). In addition to describing the prevalence of repeat victimization in Canada, this research also illustrates the relationship between crime type and repeat victimization; namely, that victims of more serious crime types are most likely to be repeat victims (Farrell and Sousa 1997; Tseloni and Pease 2005). The relationship between crime type and repeat victimization indicates that, if capping eliminates genuine crime reports, the resulting rise in incidence rates will not be evenly distributed among crime types...