Abstract
Objective. We assessed the effectiveness of South Africa’s Firearm Control Act (FCA), passed in 2000, on firearm homicide rates compared with rates of nonfirearm homicide across 5 South African cities from 2001 to 2005.
Methods. We conducted a retrospective population-based study of 37 067 firearm and nonfirearm homicide cases. Generalized linear models helped estimate and compare time trends of firearm and nonfirearm homicides, adjusting for age, sex, race, day of week, city, year of death, and population size.
Results. There was a statistically significant decreasing trend regarding firearm homicides from 2001, with an adjusted year-on-year homicide rate ratio of 0.864 (95% confidence interval [CI] = 0.848, 0.880), representing a decrease of 13.6% per annum. The year-on-year decrease in nonfirearm homicide rates was also significant, but considerably lower at 0.976 (95% CI = 0.954, 0.997). Results suggest that 4585 (95% CI = 4427, 4723) lives were saved across 5 cities from 2001 to 2005 because of the FCA.
Conclusions. Strength, timing and consistent decline suggest stricter gun control mediated by the FCA accounted for a significant decrease in homicide overall, and firearm homicide in particular, during the study period.
With the revisiting of the gun control debate both in South Africa following the high-profile shooting incident involving celebrity paralympian, Oscar Pistorius, and in the United States after recent the killings at Newtown, Connecticut, it is instructive to assess the impact of stricter gun control applied in South Africa through the Firearms Control Act (FCA) of 2000. The FCA was implemented in response to the prominent role of firearms in violent crime, identified by the National Crime Prevention Strategy adopted in 1996, active lobbying by an alliance of nongovernment organizations and opinion leaders, and a groundswell of popular support.1 It provided a complete overhaul of the existing firearms-control regime and was unambiguous in its intent to reduce the number of firearms in the country, particularly those in civilian hands. It also included provisions for restrictions and prohibition on the ownership of particular types of firearm and background checks to establish an individual’s physical and mental capacity to use a firearm responsibly.
In 2000, South Africa had one of the highest homicide rates in the world.2 The National Injury Mortality Surveillance System (NIMSS), a sentinel system comprising mainly urban mortuaries, indicates that until 2003 firearms were the leading external cause of homicide for all age groups from the age of 5 years,3–7 and in 2003, gunshot injuries accounted for 53% of male and 41% of female homicides.6 The South African Police Service has reported a consistent decline in homicide from 1994 to 2012. The NIMSS data, although not comparable year-on-year because of erratic coverage after 2005, suggest that this decrease might be ascribed to fewer firearm deaths—the latest report for 2008 revealed that sharp force injuries (stabbings) accounted for 41% of homicides compared with just 30% for gunshots.8
The stricter licensing conditions under the Firearms Control Act No. 60 of 2000 were part of a broader strategy to reduce the number of guns in circulation, and there was a firearm-access gradient between the FCA being passed in 2000 and its full implementation in 2004. The period was characterized by firearm amnesties and hand-ins, in which legal and illegal guns were recovered by the authorities, more rigorous application of existing licensing conditions, and the destruction of surplus and illegal weapons. The South African Defense Force destroyed an estimated 270 000 weapons between January 1998 and May 2001, and the police destroyed 30 000 in 2001.9 It was also notable that there was a 75% increase in the number of firearms recovered by the police in 2003 compared with the previous year (i.e., from 20 000 weapons confiscated and recovered in 2002 to 35 000 in 2003).10 This coincides with the period in which the decline in firearm homicide became evident. After full implementation of the FCA in 2004, this effort was redoubled and a further 100 000 firearms were collected in the first 6 months of 2005.11
The aim of the current study was to investigate a decrease in homicide rates over 2001 to 2005 and to assess whether this decrease was specific to firearm homicide or other factors associated with homicide risk. We hypothesized there would be a significant decrease in firearm homicide specifically attributed to the increasingly stricter gun control coinciding with the phased implementation of the FCA that was fully implemented by 2004. The corresponding nonfirearm homicide rate was expected to remain stable.
We conducted a retrospective population-based study which included all 37 067 homicide cases presenting to mortuaries over a 5-year period from 2001 to 2005 across 5 South African cities: Cape Town, Durban, Johannesburg, Port Elizabeth, and Pretoria (Table 1).
We acquired data from the NIMSS from 2001 to 2005, the period during which the NIMSS provided full coverage of the 5 cities. The NIMSS collates information gathered from death registers, autopsy reports, and ancillary police and laboratory documentation from forensic pathology laboratories. The availability of data corresponded to the provisions of the Inquests Act, which required that all deaths in South Africa associated with causes other than natural causes (i.e., injury deaths) be subject to postmortem examination by a district surgeon, forensic pathologist or medical practitioner.12
We obtained population estimates for these cities by age, sex and race as denoted by the South African Municipal Demarcation Board from the 2001 South African Census.13 Estimated population growth rates between 2001 and 2004 were derived from the Actuarial Society of South Africa AIDS and Demographic for 200314 for 4 provinces—KwaZulu-Natal, Gauteng, the Eastern Cape, and the Western Cape—and applied to the cities in those provinces.
Based on a review of risk factors for violence15 and an assessment of the suitability and completeness of available data, mortality counts for firearm homicide and nonfirearm homicide were categorized by age, sex, race, city, alcohol-relatedness, year, and the day of week.
Age of the deceased was grouped into the following intervals 0 to 4 years, 5 to 14 years, 15 to 29 years, 30 to 44 years, 45 to 59 years, and 60 years or older. This accommodated 5-year digit preference, presumably because of medical specialists estimating the ages of unidentified cases. Missing ages (10%) were imputed according to the observed city-race-sex-year-day of week specific distribution. Sensitivity analysis indicated that the imputation of missing ages did not have a differential effect.
The race of the deceased was recorded according to the apartheid population group classifications, which includes the terms African, Colored, Asian, and White. Race in South Africa was a social construct that served a particular political purpose. Race and socioeconomic status may well be conflated, but race conceivably also masks a range of invidious psychological and social effects on individuals, families, and communities that account for the differences in the risk for violence. In the absence of alternative measures of socioeconomic status, race provides a gross proxy measure to reflect the differential manner in which apartheid affected, and continues to affect, the lives and health of South Africans.
Day of week was derived from recorded date of death. Missing dates of death (1.5%) were imputed as the midpoint of the death dates of the preceding and subsequent postmortems at the same facility. Blood was analyzed for alcohol concentration, using standard gas chromatography, at the State Chemical Laboratory.
We calculated crude homicide mortality rates per 100 000 population for levels of each covariate. Generalized linear models with Poisson family, log link, and robust standard errors were constructed with firearm and nonfirearm homicide counts for age, sex, race, city, day of week and year of death strata. Covariate-specific population size provided a measure of person-time. We omitted alcohol because of the high fraction of incomplete data (40%).
Year of death was fitted as a continuous variable. Sensitivity analysis indicated that this provided a model of comparable fit to a model with year of death as a categorical variable. After assessing the time trend in homicides overall, we included an interaction term for year of death and type of homicide, to assess differences in the firearm and nonfirearm homicide time trends. Statistical analysis was conducted using Stata version 12.16 Statistical significance was set at 5%.
We have shown crude homicide mortality rates per 100 000 population and 95% confidence intervals (CIs) for firearm and nonfirearm homicide for selected covariates in Table 2. Year-on-year comparison indicated that there was a consistent decline in firearm homicide rates. Nonfirearm homicides indicated a more moderate downward trend. Firearm homicide rates were significantly higher among youths and young adults aged 15 to 44 years and among males. Analysis by race showed that firearms accounted for the majority of homicides in each race group, except among Coloreds. Firearm homicide rates were significantly higher in Durban and Johannesburg than in other cities. The alcohol-related nonfirearm homicide mortality rate was significantly higher than that for firearm homicides, which was also consistent with the lower proportion of firearm homicides on weekends. A fitted generalized linear model with an overall 2001-2005 time trend (i.e., not distinguishing between the time trends of firearm and nonfirearm homicides) indicated that all covariates were significantly associated with homicide rates (Table 3).
TABLE 1— Homicide Cases for 5 Cities: National Injury Mortality Surveillance System, South Africa, 2001–2005
| Cases/Year, No. | ||||||
| City | 2001 | 2002 | 2003 | 2004 | 2005 | Total |
| Cape Town | 2478 | 2454 | 2195 | 1823 | 2046 | 10 996 |
| Durban | 2096 | 2175 | 2080 | 1894 | 1873 | 10 118 |
| Johannesburg | 2275 | 2284 | 1936 | 1550 | 1440 | 9485 |
| Port Elizabeth | 738 | 763 | 694 | 663 | 754 | 3612 |
| Pretoria | 640 | 620 | 537 | 595 | 464 | 2856 |
| All homicides | 8227 | 8296 | 7442 | 6525 | 6577 | 37 067 |
TABLE 2— Crude Firearm and Nonfirearm Homicide Rates for 5 Cities per 100 000 for Selected Covariates: National Injury Mortality Surveillance System, South Africa, 2001–2005
| Covariate | Nonfirearm Homicide (n = 17 876), Rate (95% CI) | Firearm Homicide (n = 19 191), Rate (95% CI) |
| Year | ||
| 2001 | 29.9 (29.4, 30.4) | 37.5 (36.9, 38.1) |
| 2002 | 29.3 (28.9, 29.8) | 37.5 (36.9, 38.1) |
| 2003 | 28.0 (27.6, 28.4) | 31.1 (30.6, 31.6) |
| 2004 | 26.5 (26.1, 27.0) | 24.7 (24.3, 25.0) |
| 2005 | 28.5 (28.1, 29.0) | 22.5 (22.2, 22.9) |
| Age, y | ||
| 0–4 | 4.3 (4.1, 4.4) | 1.4 (1.4, 1.5) |
| 5–14 | 1.7 (1.6, 1.7) | 1.6 (1.5, 1.6) |
| 15–29 | 42.4 (42.0, 42.9) | 49.2 (48.7, 49.7) |
| 30–44 | 45.4 (44.8, 46.0) | 47.9 (47.2, 48.5) |
| 45–59 | 22.8 (22.3, 23.3) | 22.7 (22.2, 23.3) |
| ≥ 60 | 13.3 (12.7, 13.9) | 9.1 (8.7, 9.5) |
| Sex | ||
| Female | 7.8 (7.8, 7.9) | 6.0 (5.9, 6.0) |
| Male | 50.3 (50.0, 50.6) | 56.6 (56.2, 56.9) |
| Race | ||
| African | 35.0 (34.8, 35.2) | 40.9 (40.7, 41.1) |
| Asian | 8.0 (7.6, 8.3) | 11.8 (11.3, 12.4) |
| Colored | 32.8 (32.1, 33.4) | 20.4 (20.0, 20.8) |
| White | 7.0 (6.8, 7.1) | 8.1 (7.9, 8.3) |
| City | ||
| Cape Town | 42.4 (41.9, 43.0) | 30.8 (30.4, 31.2) |
| Durban | 27.6 (27.3, 27.9) | 37.0 (36.5, 37.4) |
| Johannesburg | 18.8 (18.6, 19.0) | 37.6 (37.2, 38.1) |
| Port Elizabeth | 55.1 (52.9, 57.2) | 16.2 (15.6, 16.8) |
| Pretoria | 12.0 (11.7, 12.2) | 15.9 (15.6, 16.2) |
| Day of week | ||
| Sunday | 6.5 (6.5, 6.5) | 5.5 (5.5, 5.6) |
| Monday | 3.3 (3.2, 3.3) | 4.0 (4.0, 4.0) |
| Tuesday | 2.5 (2.5, 2.5) | 3.4 (3.4, 3.5) |
| Wednesday | 2.4 (2.3, 2.4) | 3.4 (3.4, 3.4) |
| Thursday | 2.3 (2.3, 2.3) | 3.3 (3.3, 3.3) |
| Friday | 3.7 (3.7, 3.7) | 4.5 (4.5, 4.5) |
| Saturday | 7.8 (7.8, 7.8) | 6.4 (6.4, 6.4) |
| Alcohol | ||
| 0.00 g/100ml | 5.5 (5.5, 5.5) | 11.7 (11.6, 11.7) |
| > 0.00 g/100ml | 11.3 (11.2, 11.3) | 7.6 (7.6, 7.6) |
| Unknown | 11.7 (11.7, 11.7) | 11.3 (11.3, 11.3) |
Note. CI = confidence interval.
TABLE 3— Generalized Linear Model Homicide Rate Ratio Estimates for Selected Covariates: National Injury Mortality Surveillance System, South Africa, 2001–2005
| Covariate | Homicide, RR (95% CI) |
| Year | 0.916 (0.903, 0.930) |
| Age, y | |
| 0–4 (Ref) | 1.000 |
| 5–14 | 0.604 (0.500, 0.729) |
| 15–29 | 16.892 (14.519, 19.653) |
| 30–44 | 18.718 (16.109, 21.749) |
| 45–59 | 10.608 (9.103, 12.362) |
| ≥ 60 | 6.902 (5.862, 8.126) |
| Sex | |
| Female (Ref) | 1.000 |
| Male | 7.701 (7.390, 8.026) |
| Race | |
| African (Ref) | 1.000 |
| Asian | 0.234 (0.216, 0.254) |
| Colored | 0.520 (0.490, 0.552) |
| White | 0.193 (0.181, 0.206) |
| Day of week | |
| Sunday (Ref) | 1.000 |
| Monday | 0.600 (0.559, 0.644) |
| Tuesday | 0.488 (0.453, 0.525) |
| Wednesday | 0.473 (0.440, 0.509) |
| Thursday | 0.462 (0.430, 0.497) |
| Friday | 0.677 (0.631, 0.727) |
| Saturday | 1.182 (1.103, 1.266) |
| City | |
| Cape Town (Ref) | 1.000 |
| Durban | 0.721 (0.678, 0.767) |
| Johannesburg | 0.525 (0.492, 0.561) |
| Port Elizabeth | 0.802 (0.741, 0.868) |
| Pretoria | 0.280 (0.260, 0.301) |
| Type of homicide | |
| Nonfirearm (Ref) | 1.000 |
| Firearm | 1.077 (1.034, 1.122) |
Note. CI = confidence interval; RR = rate ratio.
A second generalized linear model fitted with an interaction term for year of death and type of homicide indicated that, adjusting for other covariates, the decline in homicide rates from 2001 was far more pronounced for firearm homicide than for nonfirearm homicide, with a year-on-year estimated firearm homicide rate ratio of 0.864 (95% CI = 0.848, 0.880), representing a decrease of 13.6% per annum. This decrease was also evident in sensitivity analyses with year of death as a categorical variable. The decrease in nonfirearm homicide rates was more modest and statistically significantly less than that of firearm homicides, with a nonfirearm homicide rate ratio of 0.976 (95% CI = 0.954, 0.997), a 2.4% decrease per annum.
The significant decline in firearm homicide from 2001 has altered the ratio of firearm to nonfirearm homicide rates. At the beginning of the study period, firearm homicide rates were significantly higher than nonfirearm rates. By 2005 the estimated firearm homicide rates were lower than nonfirearm homicide rates, but not significantly so (Figure 1).
FIGURE 1— Estimated firearm and nonfirearm homicide rates for males aged 15–29 years in Johannesburg on Saturdays: National Injury Mortality Surveillance System, South Africa, 2001–2005.
Note. Whiskers indicate pointwise 95% confidence intervals.
Assuming that without the imposition of stricter gun control the firearm homicide rate would have followed the same trajectory as the nonfirearm homicide rate (i.e., a small decrease of 2.4% per annum), the modeled homicide rates suggest that, instead, 4585 (95% CI = 4427, 4723) lives were saved across the 5 cities from 2001 to 2005. For each year, the number of lives saved was estimated at 521 (95% CI = 493, 545) in 2002, 976 (95% CI = 940, 1007) in 2003, 1372 (95% CI = 1328, 1410) in 2004, and 1716 (95% CI = 1666, 1761) in 2005.
The analysis suggests that after adjusting for the effects of other covariates, the significant decline in homicide across 5 South African cities from 2001 to 2005 was attributable primarily to the decline in firearm homicide. The introduction of the South African FCA is a seemingly obvious explanation for this decline, but inferring causality in the case of upstream policy interventions is often more complicated than in the case of programmatic interventions that are more easily controlled by investigators.
Surprisingly, a report by the United Nations Office on Drugs and Crime that draws on data submitted by the South African Police Service suggests that the decrease in firearm homicides was not substantially greater than the decrease in nonfirearm homicides between 1994 and 2007.17 The method of analysis is not described in any detail and the data that were utilized were not included among the annual crime statistics released by the police. Considering that the data in our urban study show a considerable decrease only in firearm homicide, it seems that for the police analysis for South Africa as a whole to be valid, there would have had to exist a corresponding increase in rural firearm homicides coinciding with the study period.
There is a paucity of studies describing firearm homicide trends from rural areas and the 2 sources we identified both contradict this hypothesis. In Mpumalanga, a predominantly rural province, firearm homicide accounted for 20% of all homicides in 200118 compared with only 10% in 2009.19 Firearm homicides during farm attacks have also decreased since the implementation of the FCA.20 A more likely explanation may be a differential reporting bias of firearm homicides in police statistics. That is, if firearm deaths have become less prone to underreporting than nonfirearm deaths, the cumulative effect would be an increasing proportion of firearm deaths reflected in police murder statistics. Without access to the source data, it is impossible to assuredly note whether the data were valid or whether the method of analysis was appropriate.
Our analysis, which has shown that the decrease in the homicide rate in the 5 cities is likely attributable to the implementation of the FCA, is based on more reliable mortuary data. Several studies have demonstrated the comprehensiveness of fatalities recorded in mortuaries through the medicolegal process with regard to occupational injuries,21 female homicides,22 and railway injuries.23 Although it is conceivable that a few deaths associated with the late effects of injury may be certified as being because of natural causes, and therefore bypass the mortuaries, this is unlikely because medical practitioners would be cognizant of the medicolegal implications of these cases not being referred to the relevant authorities.23
The probable mechanism by which the FCA effected a decline was the reduction in the gun pool among licensed gun owners and, consequently, criminals, who would in many cases have acquired stolen guns from the pool of registered weapons. Although the Central Firearm Registry maintained by the South African Police Service does not report on licensing of gun owners or registered firearms, there are indications that the FCA has restricted access to both these groups. Civilian firearm ownership has decreased by 14% from more than 3.5 million registered firearms in 199824 to approximately 3 million in 2011.20 A report in the Small Arms Survey 2008 Yearbook indicated that during the period of the current study the number of firearms reported stolen from April 2003 to March 2004 was 14% lower than for the corresponding March 1998 to April 1999 baseline, and 35% lower in March 2005 to April 2006.
Certainly the strength of the estimated association, a 13.6% decrease year-on-year in firearm homicide rates from 2001 to 2005, suggests a dramatically significant decrease in firearm homicide that was explained neither by other covariates included in the model, nor a decline in homicide more generally. Reasons for the more modest decrease in nonfirearm homicide remain unclear. Police attribute decreases in violent crimes to greater cooperation between police and community policing forums, increased police visibility, and higher arrest rates for robbers.25 There is also the possibility that the decrease may reflect the gradual stabilization in South African society, both politically following the end of apartheid, and socially with the slowing rate of urbanization. Urbanization is an important catalyst for increased levels of violence, and South African cities grew at a lower rate between 1996 and 2005 than in preceding decades.26
Although comprehensive mortuary data from the 5 cities were not available before and after the study period, there are other sources that indicate that firearm homicide was not declining prior to the inception of the FCA. In Cape Town, where mortuary-based injury surveillance was first piloted, there was an increasing trend in firearm homicide prior to the FCA. In 1994 firearm homicide accounted for just 28% of homicide cases in Cape Town27 (compared with 42% between 2001 and 2005), and there was evidence of an increasing trend before 2001.28 Prior to the FCA, police homicide statistics indicated that there had been a slight increase in the number and rate of firearm homicides from 1994 to 1998.24 The first NIMSS data for 1999 recorded 51% of homicides as being firearm-related.29 This increased to 55% in the 2000 annual report,3 which was similar to the 56% of homicides recorded in the 5 cities in 2001, the first year of this study.
Analysis of a longer time trend post-FCA was similarly not possible because full coverage of the 5 cities was only maintained during the 5-year period of the current study. Nevertheless, after 2005, partial data from injury mortality studies in 3 of the cities confirm that the decrease in firearm homicide had been sustained. In 2010, firearms accounted for 32% of homicides in Cape Town,30,31 compared with 42% during the period of this study (Table 2) and 36% of homicides in Gauteng Province,32 which included Johannesburg and Pretoria, which recorded 67% and 57% of homicides being firearm-related from 2001 to 2005.
Evaluation of natural experiments typically relies on before-after study designs assessing the effect within a single society and evidence is enhanced through replication across settings. Our findings are consistent with international data showing that the variation in homicide rates between countries is closely associated with the rate of firearm homicides.33 Despite considerable variation across settings, there is also emerging evidence that introducing legislation and regulation can reduce firearm-related suicides and homicides, as well as public demand for firearms.34
In the United States, the effectiveness of legislation has only been shown in relation to unintentional deaths35 and suicides among young people36 following the introduction of child access prevention laws. No reduction in firearm homicide was found in 32 states following the implementation of the Brady Handgun Violence Prevention Act in 1994 compared with 18 control states.37 However it should be noted that the Brady Act legislation applies to the purchase of new firearms rather than those already in circulation, and a considerable time lag might be expected before any notable decrease could be observed. Interventions that demonstrate the most notable population-level effects are those that have been applied in settings with particularly high firearm fatality rates,34 such as in Colombia38 and Brazil,39 or that included wide-ranging restrictive measures, such as in New Zealand40 and Austria,41 and measures to reduce firearms already in circulation, as in Australia.42,43 Recent events have increased the feasibility of considering similar upstream policy and regulatory interventions in the United States.
In the absence of identified confounding factors, the strength, timing, and consistency of the decline in firearm homicides in South Africa suggest that stricter gun control through the FCA accounted for a significant decrease in homicide overall, and firearm homicide in particular, during the study period. The findings are consistent with currently available research that suggests that stricter gun control decreases homicide rates, particularly in countries such as South Africa and the United States where homicide rates are already relatively high.
Acknowledgments
Data collection was funded by the Medical Research Council of South Africa and provincial Departments of Health.
We are grateful to the forensic pathologists, the academic staff in forensic medicine departments and the mortuary personnel for supporting this study. We acknowledge William Msemburi of the Medical Research Council’s Burden of Disease Research Unit for calculating population estimates for the five cities.
Note. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the article.
Human Participant Protection
The University of Cape Town’s Health Sciences Faculty Research Ethics Committee approved the study and permission to utilise the data were granted by the National Department of Health.
