Tables and Graphs for Monitoring Crime Trends

Tables and graphs for monitoring temporal crime trends: Translating theory into practical crime analysis advice
 
Data based approaches to crime prevention such as CompStat and hot spots policing have become the norm for modern police agencies. Data are often presented in tables that show percent changes in categories of crime for a specific time period. These types of data are not sufficient to determine major crime trends over time, but police officers often use them to infer more general crime trends. 
 
The goal of monitoring crime trends over time is to be able to determine is current crime numbers are outliers compared to historical numbers and trends. However, using percent change to do so is problematic. First, the nature of percent change makes it very difficult to determine how large a percent change is an outlier. This is because the variance is only based on the ratio of the pre- and post-numbers. The percent change variance differs depending on how often crimes occur, and when crimes occur infrequently the variance will be higher. 
 
For example: if there were five robberies in May, and then there were eight in June, we have an increase of three robberies. The percent change is 60% (increase). However, if there were four robberies in May, and then eight robberies in June, the percent change is 100%, even though there was only a one crime change in baseline. This is misleading and does not accurately reflect the actual state of crime; it is what is known as a false positive. When police departments take actions resulting from “false positive” data, it results in wasted time and resources. 
 
The false positive bias exists when using percent change data because percent change is not symmetric. An increase from four to five crimes is an increase of 25%, but a decrease from five to four crimes is a decrease of 20%. Percent changes are more likely to mislead people into thinking that crime is increasing rather than decreasing, leading to wasted time and resources. 
 
An alternative to percent change might be a Poisson z-score. If crime is a Poisson distributed random variable, a standardized statistic can be calculated. This normalizes the distribution to have an average of 0 and a variance of close to 1, instead of being based on pre- and post-values like percent changes are. Unlike percent change, this standardization allows us to know how large a value is needed to mark a change as being significant, and that value does not change with the crime numbers. This also makes increases and decreases symmetric, so there is no bias toward false positives. This statistic can be shown on a simple scale illustrating when to flag a change as being significantly different from past values. 
 
In addition to using a standardized statistic instead of percent change to determine changes in crime, tables and graphics used to display crime data can be improved. When using color in tables, one must be careful to avoid random colors with no reason or scheme behind why certain cells are colored the way that they are. In addition, those with colorblindness must also be considered. Color also decreases text readability. When using color in a table, it may be best to stick with greyscale with only the values that you want to stand out colored. 
 
Numbers and rounding can also impact a table’s readability. It is best to keep numbers in a table rounded to the nearest tenth, as further decimals are often not needed and can decrease readability. Further, it is important to think about what the table is displaying and what data should be emphasized. Sometimes it might be best to display numbers sorted from smallest to largest, or vice versa, depending on what is meant to be emphasized in that table. Or, the table could be arranged according to crime type groupings, such as violent and non-violent crimes. The goal of creating data tables should be to easily identify the data that are intended to be the most important, such as significant changes in crime or in certain crime types. 
 
Finally, using seasonal charts and time series graphs can assist in being able to spot outliers in crime data. Time series charts graphing the Poisson z-scores, observed counts of crime, moving averages of crime, and the currently weekly crime count can help to quickly identify whether there are outliers in the data. Seasonal graphs of certain crimes, such as burglary, with crime counts per month and yearly lines can also help to easily identify seasonal trends, if there are any. 
 
Calculating changes in crime data, as well as displaying that data in an informative and intuitive way is important for evidence-based policing practices. Having accurate data that can be easily interpreted allows police to maximize efficiency and effectiveness in their responses to changes in crime, and avoids wasting resources and time. 
 
Takeaway: 
 
A standardized statistic, such as a Poisson z-score, can improve the accuracy and interpretability of crime statistics as opposed to using percent changes 
The way that tables and charts are displayed impacts the readability of the data and they should be designed so that the most important data are emphasized 
Graphing seasonal and monthly/weekly crime trends can be an effective way to spot outliers in the data 
 
Reference: Wheeler, A. (2016). Tables and graphs for monitoring temporal crime trends: Translating theory into practical crime analysis advice. International Journal of Police Science & Management, 18(3), 159–172. https://doi.org/10.1177/1461355716642781