Homicide rates

This subsection focuses on the changes in homicide rates in the city of Chicago, with descriptions of what findings were produced by the tests and discussion on how to interpret these results. The first image displays the clustering of change in homicide rates between the years 2000 and 2008 (Figure 5). There are five different colors indicated in the map image, with the legend provided by the GeoDa software in the upper left corner.

The majority of the map is colored light gray, which means there is no statistically significant clustering of any kind when compared to an artificial null hypothesis produced by running Monte Carlo permutations on the data. In these images the number of permutations has been set to 999. The weights matrix used in the analysis is a queen contiguity, meaning that all contiguous neighbors on all sides of an area are weighted in the local Moran’s I analysis.

The four other colors in the map indicate that some type of clustering is present in the area. Areas colored red indicate places where high values of homicide rates are surrounded by similar high values. Areas colored blue are areas where there is clustering of low values surrounded by similar low values. The remaining cases contain clustering of dissimilar values, where light purple areas are low values of homicide rates surrounded by high values, and light red areas are high values of homicide rates surrounded by low values.

In the differential local Moran’s I, the values are interpreted somewhat differently. Whereas in a local Moran’s I statistic low and high values are interpreted in a straight forward manner, where a high value corresponds to a high value in the data, in a differential local Moran’s I the values are determined by examining the changes between the starting point and the ending point. A low value in this analysis means a decrease in the values between the starting point and the ending point, namely that the starting point value is higher than the ending point value. Conversely, a high value means an increase in the value between the two points in time, with the ending point value being larger than the starting point value.

Figure 5

As the map shows, there are four areas where there is clustering of high values surrounded by high values, six areas where there is clustering of low values

surrounded by low values, four areas of lows surrounded by highs, and two of highs surrounded by lows.

What is interesting to note is where the clustering of values occurs. The clustering of decrease in homicide rates is located in and around the area where the Robert Taylor Homes and a large number of other housing projects were demolished during that time period, namely the community areas of Grand Boulevard [38] and

Douglas [35]. The immediate surrounding community areas of Armour Square [34], Oakland [36], and Fuller Park [37] also show clustering of decreases in homicide rates following the demolition of public housing.

However, there are also areas where homicide rates have gone up between the years 2000 and 2008. This is of particular interest as there was an overall trend of decreasing homicide rates in the city following the dwindling crack cocaine boom, starting around the year 2000, with the actual numbers of homicides decreasing from 582 in 2000 (CPD annual report 1999/2000) to 510 in 2008 (CPD annual report 2008). In spite of this 12% decrease in homicide, there are areas where homicide rates have gone up from 2000 to 2008. Of course, it is impossible to claim direct causality based on the indications displayed in the results of the analysis, but they are

interesting anomalies nonetheless.

There are two community areas on the edges of the concentration of

decreasing homicide rates where homicide rates have increased, namely Near South Side [33] and Bridgeport [60]. These are immediately adjacent to the areas where massive demolitions of public housing occurred, which is interesting to note even if concrete verification of causality or connection cannot be made with the data that is available. There are also some areas in the southern parts of the city where homicide rates have gone up.

The following image is a significance map for the differential local Moran’s I statistic between the years 2000 and 2008 in terms of changes in homicide rates (Figure 6). As was described earlier, the statistical significance is calculated based on a pseudo null hypothesis that is constructed by running Monte Carlo permutations on the existing data, and for the purposes of these analyses the number of permutations has been set at 999. Naturally, the significance map corresponds to the earlier cluster map, where only statistically significant clustering is displayed.

In the same way as the cluster map for the differential local Moran’s I, the significance levels are encoded on the map with the use of colors. In this particular map, the majority of the community areas are colored gray, which translates to no statistical significance at the set levels. The areas with the standard p-level of 0.05 are colored a light green. Areas with a p-level of 0.01 are indicated by a medium green color, and highly significant areas with a p-level of 0.001 are colored a dark green.

As the map shows, there are 8 areas that are statistically significant at p-level 0.05, 4 areas that are significant at p-level 0.01, and also 4 areas that are significant at

p-level 0.001. The areas with the lowest p-level of 0.001, or the highest statistical significance are, from top left to bottom right, Bridgeport [60], Armour Square [34], Chatham [44], and Avalon Park [45]. Of these, Bridgeport [60], Chatham [44], and Avalon Park [45] saw increases in homicide rates between 2000 and 2008, and Armour Square [34] experienced a decrease during this period.

Figure 6

The community areas of most interest in terms of public housing, Douglas (35) and Grand Boulevard [38] were statistically significant at p-levels 0.05 and 0.01 respectively. The strength of the significance is especially noteworthy for Grand Boulevard [38], where the notorious Robert Taylor Homes high-rise housing projects had been demolished during this period. In the surrounding community areas with decreases in homicide rates, there were two highly significant ones. Armour Square [34] had a decrease in homicides that was significant at p-level 0.001 and Fuller Park had a decrease significant at p-level 0.01.

Again, making definitive claims of causality with the available data is not possible, but it is worth noting the strongest statistical significances in and around the area where the concentration of high-rise housing projects was. Of these, Fuller Park [37] and Grand Boulevard [38] were significant at a p-level of 0.01 and both areas

experienced decreases in homicide rates between 2000 and 2008. Another two areas, Armour Square [34] and Bridgeport [60], had extremely high statistical significance at p-level 0.001, with the first one experiencing a decrease and the latter one

experiencing an increase in homicide rates.