To provide the most accurate nationwide perspective of the extent of the gang problem, it was necessary to estimate:

• The number of jurisdictions reporting gangs.

• The number of gangs.

• The number of gang members.

• The number of homicides.

To determine the estimated number of jurisdictions reporting gangs in small cities and rural counties, the percentage of agencies reporting gangs was multiplied by the total number of small cities and rural counties included in the group from which the sample was derived. The same method was used for large cities and suburban counties in order to incorporate nonrespondents.

Estimating the number of gangs and gang members for small cities and rural counties was a slightly more complex task. For these samples, the following steps were completed:

1. Calculate the mean for agencies responding either "yes" or "no" to the question regarding the number of gangs and gang members.

2. Multiply the total number of jurisdictions from which the sample was derived by the percentage of agencies responding either "yes" or "no" to the question regarding the number of gangs and gang members.

3. Multiply the product of step 2 by the mean calculated in step 1.

To produce the most accurate nationwide estimate, it was necessary to extrapolate for nonrespondents in large cities and suburban counties. If this had not been done, the result would have been a systematic underestimation of the extent of gang activity in these areas. In addition, any change in the proportion of agencies responding for large cities and suburban counties in future surveys would likely have resulted in a commensurate change in the number of gangs and gang members; this change could lead to a false conclusion that gang activity has increased or decreased.

To estimate the number of gangs and gang members for large cities and suburban counties, the average or mean number of gangs and gang members per jurisdiction was calculated. These estimates were controlled for population by stratification of respondent agencies into population groups of 50,000 and by calculation of a mean for each population group. To acquire the most accurate mean, the survey designers established a minimum number of agencies in each population group from which a mean could be derived. The minimum number was set at 40 to allow for the inclusion of a large number and wide range of agencies in each population group. Unfortunately, not all population groups included 40 or more agencies. Therefore, the population groups were expanded equally on the high and low ends until 40 or more agencies were included. Once this expansion was completed, a mean was calculated and that mean was matched with each nonresponding agency within the corresponding population group (see Appendix D).

As with gangs and gang members, estimating the number of homicides required extrapolation for both random samples and for nonrespondents in large cities and suburban counties. The number of agencies responding to the homicide questions and reporting homicides in the random samples was comparatively low. Consequently, extrapolations for the random samples would not have been reliable. Responses for large cities and suburban counties were analyzed because the entire universe of each group was included in the sample. Therefore, an estimated number of homicides was reported only for large cities and suburban counties. A nationwide estimate could not be calculated.

The estimated number of homicides that likely occurred in large cities and suburban counties during 1996 was obtained by multiplying the average number of homicides per jurisdiction by the estimated number of jurisdictions reporting gangs in 1996.