Crime as a Social Cost of Poverty
□nd Inequality: A Review Focusing
on Developing Countries
F. Bourguignon
the homicide series is likely to be the most reliable4. Moreover, it may be
expected to be somewhat correlated with the actual, as opposed to observed
robbery rate, Thiswould be the ease, for instance, if a more or less constant
proportion of robberies were leading to the death of a victim. In effect,
the correlation between both sets of series in the UN data base is rather
high. The rank correlation computed on all series after pooling all countries
together is 0.40.
Figures la and Ib report the evolution of five-year average crime rates for
the great regions of the world, as it may be very roughly estimated with
the UN data base5. These figures refer to the median of five-year country
averages in each region. It must be kept in mind that: (a) the sample of
countries may change from one period to another, and (b) the number of
years on which averages are computed differs across countries and sub-
periods. However, as the time variation of crime rates in contiguous years
is not very important, the major potential source of bias is due to (a).
Computations made on samples comprising too few countries have been
eliminated -this being the reason the Middle-East and North Africa region
does not appear in Figure 1-. Some observations arc still, however, of
little significance. This is, in particular, the case for the 1985-89 drop in
African crime rates. Following FLL it may be thought that the median is
more reliable than the mean because it is insensitive to possibly
inconsistent extreme values. But in fact, the overall picture is not that
different whether one uses the median or the mean.
Before examining these figures, it may be useful having in mind some
orders of magnitude for further reference. Because data arc more reliable
there, it seems natural to take high-income countries as a basis for
comparison. Among them, the United States stands at the upper extreme
with a frequency of robbery ranging from 170 to 260 for 100,000
inhabitants (Thti) during the 1970-1994 period and a homicide rate ranging
■шашммктшквим
4 To test this reliability, it is also possible to compare these series with data from the World Health
Organization on death causes, one of which is homicide, ʌn informal calculation based on mean
homicide rates from 1970-1984 or sub-periods led to a rank correlation between the two sources
equal to 0.60. This is not too bad but far from satisfactory for both sides of the comparison.
3 Figurela also appears in FLL.
66