Boston crime in times of COVID-19

I am writing this at a time when COVID-19 paralyzes the world. The fatalities, geographic patterns, and economic impact of the disease are subject to fantastic visualizations elsewhere.

However, the lockdown in response to the disease has an impact on almost all aspects of our life and, hence, creates striking patterns in otherwise consistent data.

Here, we’ll take a look at other (equally sad) graphs—the crime statistics in Boston. Did the number of reported offenses change during COVID-19? Did the occurrence of specific offenses vary in comparison to other periods?

The Boston Police Department reports crime statistics on the Open Data Portal of the city of Boston. This rich publicly-available data resource lists, descriptions, dates, and locations of incidences that involved the BPD.

The data set lists more than 400 different types of offenses. Unfortunately, the data is inconsistently labeled. For instance, the corresponding metadata lists duplicate offense IDs with different descriptions.

Furthermore, we find offense codes that appear only for 2020, while others do not appear for 2020 at all. Did these types of offenses not occur in the period, or were not reported? Were offenses unknowingly labeled inconsistently? Did the granularity of offense reporting change without (public) documentation? As these questions remain unanswered, any conclusions from the following analysis remain preliminary.

I cleaned up the labels to the best of my knowledge, yet had to omit offenses with inconsistent labeling that I could not resolve without subject matter expertise. I’d be thrilled to connect with anyone who has insight into data labeling or collection at the Boston Police Department.

Let’s take a first, bird-eye-view of the data. Below is a table with counts per offense type and year, sorted by the highest number of reports over the years. This tabular view makes the zero data (or reporting gaps?) quite apparent.

Note that the most-recent year contains partial data (i.e., up to today). These (and all following) visualizations use an OData data source. That means they directly fetch data from the repository of the city of Boston and will refresh accordingly.

Before drilling down into short-term changes due to the novel coronavirus, I want to understand longer-term seasonal fluctuations. From the graph below, we can conclude that the total number of reported crimes is very stable over the years, hovering around 100,000 incidences per year (see totals in the table above).

Above, we observed a robust seasonal trend over the years. February is consistently the month with the lowest number of reported offenses. For previous years, the numbers of reported crimes continuously pick up in March and peak in August. Strikingly, data for 2020 does not follow this general trend. Let’s take a closer look at the counts in March and April of 2020.

Plotting the total number of reported incidences per day reveals an interesting trend: the number of reported offenses drastically decreases after Governor Charlie Baker declared a state of emergency. When we compare the daily totals to the same day in 2019, the number of reported incidences drops as much as 50% (e.g., March 29th). This can mean that fewer offenses took place, or fewer where reported.

Did all kinds of offenses decrease proportionally? Let’s drill down further and look for changes within particular types of offenses.

Below, we look at the five most frequent types of offenses. Each bar represents the number of incidents for a particular year. More precisely, for each year, we include only the counts that occurred from Jan 1st to yesterday’s date. Restricting the period in this way, allows us to compare the numbers of the ongoing year with the same period in previous years.

As we can see for these five offense type, the numbers did not uniformly decrease. Quite on the contrary, the offense “Verbal dispute” increased by 90% compared to the same period last year. This sad finding might not be surprising given that everyone is currently confined to their home.

Unfortunately, I lack the subject-matter expertise (i.e., policing knowledge) to interpret a 60% drop in the offense category “Sick/Injured/Medical – Person [3006].” Was the police called to the scene after someone was wounded?

In contrast, the category “Sick/Injured/Medical – Police [3018]” has an increase of 550% (at the time of writing: 59 cases in the same period 2019, while there have been 386 cases in 2020). Please comment below, if you have insight into the categorization of incidents.

The graph below plots the number of cases in March and April each year. The data points for 2020 are colored blue when the number of incidences decreases. When the counts in 2020 fell in comparison to March and April in 2019, the data point is pink.

Given that the overall number of incidences fell by 20%, we see that most types of offenses decreased in their numbers (i.e., end with a pink dot). Given the order to shelter in place, it is not surprising that incidences involving motor vehicles decreased by approximately half.

The sharp decrease in reported graffiti vandalism (48 cases in 2019 versus 1 in 2020) might have many reasons. Did fewer people report incidences—as they did not leave their homes, or had other things on their mind? Did sprayers fear infection by the virus SARS-CoV-2? Or did they fear discovery by the police, which indeed investigated twice as many people compared to 2019 (849 vs. 1247 at the time of writing.)

Interestingly, the number of armed bank robbery increased by 80%. Job loss, financial strain, but also the opportunity to meet fewer by-standers, might contribute to that increase (75 vs. 135). It might even seem easier to escape when the majority of people on the streets wear a face mask.

I mentioned the doubled “Verbal Dispute [3301]” category already above. However, that is nothing compared to the 1,200% increase in reported cases of “Simple Assault [801]” (655 cases in March/April 2020 versus 48 reports in the same period 2019).

Above, we uncovered a substantial increase in simple assault cases. I wondered if areas with a lower household income would see a steeper increase in such offenses. COVID-19 related stressors (such as job loss, limited access to health care) might act stronger on these households.

In the below graph, each circle presents data for a police district in Boston. The size of the circle is relative to the absolute number of cases (i.e., a larger circle means more simple assault reports).

The color of the circle indicates the magnitude of the increase. That means a drastic increase compared to 2019 is represented in dark blue, a modest rise in light blue.

The map itself is colored according to the 2018 per capita income (provided by Tableau). Darker shades of gray represent higher income.

There is no striking correlation between income and an increase in simple assault cases. Notably, we observe the most substantial increase in the relatively affluent West Roxbury district. In 2019, the area reported only 1 simple assault incidence, but 23 during the ongoing COVID-19 crisis.

The overall number of reported offenses fell significantly during the lockdown due to COVID-19. Sadly, predictions that people might show increased aggression when confined to their homes did come true. The number of verbal disputes and simple assaults rose dramatically, reflecting the stressful situation. The available data do not point to an increase in such aggressive behavior in areas with a lower average household income.

As growing health and economic concerns increase, Bostonians likely feel little joy that their streets are currently a lot safer.