JustResearch Edition no. 9

Research in Profile (cont'd)


By: Thomas Gabor, Professor, University of Ottawa Kwing Hung, Statistical and Methodological Advisor, Stephen Mihorean, Principal Statistician, and Catherine St-Onge, Research Assistant, Research and Statistics Division, Department of Justice Canada


Historically, homicide has been the subject of considerable criminological research. It has also served as the leading indicator of crime and violence in a number of contexts. Cross-national comparisons often rely on homicide figures, as legislative and definitional differences often undermine the validity of comparisons involving other offences.

For example, sexual assault is defined more broadly in Canada than in countries that have retained "rape" statutes. Rape laws tend to be confined to acts involving attempted or actual sexual penetration; whereas, in the Canadian context, "sexual assault" includes such behaviours as non-consensual sexual touching (Comack, 2000). While definitions of homicide also vary to some degree across nations, there is likely to be a greater consensus, on the international level, about the nature and gravity of this offence than is the case with other offences.

Homicide also tends to be viewed as the best barometer of crime in analyses within countries, as this offence is more likely to be discovered and to be the subject of a serious investigation. In societies not beset by major upheaval, there is the intuitive belief that dead bodies are usually accounted for. Even the blue-ribbon panel on violence organized by the National Academy of Sciences in the United States glossed over potential reliability problems, writing that:

While murders are counted rather accurately, counts of nonfatal violence are incomplete. Gaps and discrepancies occur because victimizations may not be recognized as crimes, because embarrassment or psychological stigma inhibits reporting, because victims are sometimes reluctant to involve authorities, because their consequences may not be thought worth reporting as crimes and because of discretion classifying and counting violent crimes (Reiss and Roth, 1993).

Homicide figures are also viewed as quite accurate as this offence tends to be the subject of a more thorough police investigation. One indication of the priority given homicide investigations is the number of major police departments that have specialized units for the investigation of homicide.

The reliance on homicide as the most serious and the pre-eminent indicator of crime is also illustrated by the attention devoted to this crime by the news media. Studies of the print media have repeatedly shown that this offence receives coverage that is profoundly disproportionate to its volume (Gabor and Weimann, 1987).

Homicide is also more likely, than are other offences, to be the focus of special data collection efforts and reports by public agencies (e.g., Statistics Canada's Homicide Survey and numerous statistical reports on homicide, as well as the Supplemental Homicide Reports in the United States). It is also consistently one of a relatively small number of crimes used in national crime indexes.

The need for reliable measures of homicide is also underscored by the critical policy issues informed by research in this area. As one example, Bowers and Pierce (1975) have argued that conclusions drawn by Isaac Ehrlich (1975), in his seminal study on the deterrent effects of capital punishment, may have been unwarranted due to his reliance on the FBI's Uniform Crime Reports as the measure of homicide. Bowers and Pierce recommended the use of the National Center for Health Statistics' Vital Statistics as the more accurate homicide measure.

Given the profile, importance of and reliance upon homicide figures, there is both a vested interest and implicit belief in the validity of homicide data. Often overlooked, however, is that many countries collect two largely independent sets of statistics on homicide. One set is derived from police services, the other from death certificates (mortality data). Police-based statistics are likely to benefit from the criminal investigative experience of police departments. Mortality figures, on the other hand, are likely to benefit from coroners' investigations and medical opinions regarding the cause of deaths. Criminological studies of homicide have relied overwhelmingly on police statistics and have usually ignored mortality data. The possibility of a significant divergence of the two entails consideration of both sources.

In Canada, Statistics Canada is responsible for compiling both databases. With regard to the Homicide Survey, police departments across Canada provide detailed information each year to Statistics Canada on all homicides occurring within their jurisdiction. The second source, the Mortality Survey, covers all deaths occurring in Canada, as well as of Canadian residents occurring in the United States. The number and causes of deaths are obtained from death registration forms and forwarded to Statistics Canada by the central Vital Statistics Registry of each province and territory.

This research note intends to raise awareness about the existence of two distinct sources of homicide data, compares the homicide figures yielded by these two sources over the last three decades, and advances some preliminary reasons for the differences uncovered. The time frame covered by the analysis was 1970-1997, 1997 being the last year for which figures from the Mortality Survey were available at the time of this study. A significant divergence of the police-based Homicide Survey and the Mortality Database would call into question the reliability of one or, perhaps, both sources in accurately representing the level of homicide.

The only systematic comparisons conducted in Canada to this point involved case by case analyses by Hung (1987) for 1984 and 1985. Hung found that just 70 percent of "legitimate" homicides (i.e., events properly classified as homicides) were identified as such in both the Homicide Survey and Mortality Database.


The line graphs in Figure 1 display the Canadian homicide counts yielded by both the Homicide Survey and the Mortality Database, while the bar graph illustrates the absolute difference between the two sources during the study period.

We can observe that the figures drawn from the Homicide Survey (HS) exceed those yielded by the Mortality Database (MD) every year from 1970 to 1997. Furthermore, the HS trend line shows stabilization in the figures following a growth period

in the early 1970s, whereas the MD suggests a growth period in the early 1970s followed by a slight decline from the latter part of the 1970s on. The figure also indicates that the gap between the two databases has grown over time.

Figure 1: A comparison of homicide counts in Canada (1970-1997)

Figure 1, vertical line and bar graphs representing the Canadian mortality rates provided by the Homicide Survey (HS) and the Mortality Database (MD) and the difference between the two data sources, 1970-1997.

Sources: Statistics Canada: Causes of Death (Catalogue 84-208) Statistics Canada: Homicide Survey.

The homicide figures are correlated strongly and positively across the two databases. The Pearson correlation coefficient is .85 when the absolute homicide figures are considered and .89 when the homicide rates are considered. This finding indicates that the trends revealed by the two databases are similar; however, it does not reflect the magnitude of the differences between the two sources of homicide data.

The results indicate that the HS yielded an average of 89 more homicides per year than the MD (627 as compared to 538), a 14.2% difference. The average annual homicide rate in Canada, according to the HS, is .34 (per 100,000) higher than that indicated by the MD.

In absolute terms, the HS revealed a range in the homicide counts during the study period that was considerably in excess of that shown by the MD. The databases differed by about 10% in terms of the bottom of that range-467 and 421 homicides yielded by the HS and MD, respectively, in 1970. The difference, however, was more substantial at the top of the range as HS showed a peak number of 755 homicides in 1991, while the MD showed a peak of just 622 homicides during the same year, a difference of 133 homicides or about 18%. When rates are considered, the HS showed a greater peak by .42 incidents per 100,000. Further investigation is required to explain why it is that the two databases had a similar minimum figure, but were so dissimilar in terms of the peak years. Might it be that during these peak years there were more incidents of a certain type or in jurisdictions in which medical examiners were less likely to classify incidents as homicides?


This study raises some concerns with regard to the reliable measurement of homicide. While the figures yielded by the two Canadian data sources-the Homicide Survey and Mortality Database-did not differ dramatically, the differences between them were sufficient to merit further investigation. Without exception, the police-based Homicide Survey consistently yielded higher homicide counts, with the differences averaging almost 15% during the study period. The differences were more pronounced during some of the peak years for homicide. In addition, these differences appear to be increasing over time.

The two databases not only yielded different volumes of homicide, they arrived at somewhat different conclusions regarding homicide trends. When the raw figures were consulted, the Homicide Survey suggested a stabilization and the Mortality Database a decline in homicide between 1977 and 1997.

Future research will need to determine the reasons for the discrepancies between the two databases. Are they an inevitable result of the reliance on two different sets of professionals-law enforcement versus health-or are they avoidable through more rigorous record-keeping practices?

Hung's (1987) study of case files indicated that the majority of discrepancies were due to differences of opinion between police and coroners regarding the cause of death. Data quality issues and different data collection procedures accounted for a smaller number of discrepancies.

The question also arises as to whether the differences between the databases are systematically related to the nature of the homicide (e.g., family versus acquaintance incidents) or the circumstances surrounding the incident. It is possible that the growing divergence of the two databases over time may be related, in part, to a decline in the proportion of homicides among family members. The motives and circumstances of these incidents are likely to be less ambiguous than those involving acquaintances.

American investigations have found that differences between the two data sources are attributable to a number of factors. While mortality statistics include civilian justifiable homicides, the Supplemental Homicide Reports exclude them (Wiersema, Loftin, and McDowall, 2000). Also, some American law enforcement agencies do not participate in the UCR program and, hence, the reporting of homicides to the FBI is incomplete. A third reason for the difference between the sources relates to the timing of the recording of an incident. Police agencies either record the date of the incident or the date of their discovery of it, while mortality statistics record the date of the victim's death. This discrepancy may result in the reporting, by police and health professionals, of the same incident in two different reporting periods.

The use of different research methods in the American and Canadian studies done to date prevents us from even speculating about the different situations prevailing in the two countries. A cross-border study, involving both a case file analysis and interviews with police and medical examiners, would appear necessary to sort out the cross-national and inter-source differences. The reliance placed by professionals, the media, and public on homicide figures as a key barometer of crime, as well as the use of these figures in a large body of criminological research, underscores the importance of such research.


[2] This research paper has been reproduced and edited specifically for JustResearch, with permission from the authors.