Local Indicators of Spatial Association in Homicides in Baltimore, 2019

Using Some Open Data and R Programming to Look for Spatial Clusters

Last year was yet another record year for homicides in Baltimore with 348 reported homicides for the year. While this number was the second-highest in terms of total homicides, it was probably the highest in terms of rate per 100,000 residents since Baltimore has been seeing a decrease in population. (The previous record was in 2015 and 2017.) The epidemic of violence and homicides that began in 2015 continues today.

When we epidemiologists analyze data, we need to be mindful of the spatial associations we see. The first law of geography states that “everything is related to everything else, but near things are more related than distant things.” As a result, clusters of cases of a disease or condition — or an outcome such as homicides — may be more a factor of similar things happening in close proximity to each other than an actual cluster.

To study this phenomenon and account for it in our analyses, we use different techniques to assess spatial autocorrelation and spatial dependence. In this blog post, I will guide you through doing a LISA (Local Indicators of Spatial Association) analysis of homicides in Baltimore City in 2019 using R programming.