Do Violent Crime Rates in Chicago Justify Operation Legend?
A Machine Learning Approach to Modeling Crime
* For the full analysis, see Chitown Crime: Facts Versus the Feds
On July 22nd, the President announced he would send ‘hundreds of federal troops’ into cities like Chicago to combat urban crime.
Why?
Rates of violent crime across the United States have decreased dramatically over the past quarter-century.
Chicago has been singled out as a media poster-city for violence. It’s often touted as the murder capital of the US, due to its annual homicide counts.
As America’s third-largest city, it ranked 6th (2018) and 14th (2017) in terms of overall violent crime and homicide rates, respectively, compared to the rest of the United States on a per capita basis. Chicago fares better than other cities with at least 250,000 residents. To fit its characterization — and the federal intervention’s “rationale” — Chicago should rank much higher.
Anchorage and Memphis took the top two spots, respectively, in per capita overall violent crime. Neighboring St. Louis and Detroit ranked first and third place in terms of homicides.
In case you’re wondering: Portland does not even begin to near the top of the list in terms of either homicide rates (78th in 2017) or overall violent crime (128th in 2018).
The portrayal of Chicago as a center of urban crime has historical origins — events with context for the racial tensions, the pandemic, and the characterization of 2020.
Statistics show the present crime rate is nothing like the past. There is no justification or cause in the data for a dehumanizing federal response — a political strategy thinly veiled as a required imposition of law and order.
Publicly available data were sourced from the Chicago Data Portal (CDP) and include all reported crimes from the beginning of 2001 to present for the Chicago Metropolitan Statistical Area. Crimes were classified as violent (including assault, homicide, sexual assault, robbery, kidnapping, arson, etc.) or non-violent. A total of 7,158,040 crimes were reported; 29.6% fell under the violent category and the rest were classified as non-violent. Season was categorized as Winter (Dec-Feb), Spring (Mar-May), Summer (Jun-Aug), and Fall (Sep-Nov).
The research question is simple: “Are current violent crime rates in Chicago any different from those of previous months and years?”
Despite doom-and-gloom news stories, the overall number of reported crimes had been trending downwards in recent years (Plot 1).
Additionally, violent crimes have maintained a steady trend since 2015, neither decreasing nor increasing.
This trend falls in line with research from the Pew Center, which finds that the public thinks that violent crime rates are rising, when rates have actually fallen drastically over the past quarter century.
If the data appear seasonal, it’s due to strong periodicity: peaks in crime occur in the summer, and valleys in the winter. Focusing on just violent crimes over the years by season (Plot 2), it’s clear that the summer months have a higher mean number of crimes (indicated by the middle line in each box) than any other month. This seasonal effect is statistically significant at the p<0.001 level (F = 142,641).
It’s peak summer in the Windy City as of this writing, and protests abound — how do current violent crime rates compare to what we would expect to see, given historical data and context?
A random forest regression was constructed to model the number of violent crimes by month and year. A forest of 500 trees was grown, which explained 96.14% of the variance in crimes. The model was trained and tested on data prior to April of 2020; crimes were then forecasted for the period of April — December 2020. The plot below displays the actual, fitted, and forecasted number of violent crimes by month.
The fitted values map quite closely to actual historic crime numbers, with little error. The expected (forecasted) values for April — June 2020 are greater than actual rates of violent crime. Given that our model demonstrates a very good fit (96.14% of variance in violent crimes is explained), we can be quite confident that violent crime is statistically, significantly lower than expected. In terms of probability, the number of observed violent crimes in April had a 0.10% chance of happening due to randomness (p < 0.001); May a 4.95% chance (p < 0.05), and June 33.3% chance (p = Not Significant).
In terms of variable importance, year matters more than month. A lot of information would be lost if year were not included in the model (as measured by increasing node purity). Extracting the partial dependence of each predictor yields its individual effect on the predicted change in number of crimes; year also has a stronger fit compared to month on this measure (R²). Plot 4, below, illustrates the relationship between each predictor, the predicted change in number of violent crimes, and the relationship between year and month. Each plot is colored by year, with more recent years indicated by blue/purple points, and earlier years by red points. Crime is significantly lower after 2010; fall and winter months experience significantly lower numbers of reported violent crimes.
An interesting interaction effect between year and month emerges in the rightmost plot: the summertime boost in crime is less dramatic for more recent years (blue points lower on the y-axis during summer months); and while cooler months bring lower crime rates for all years, that effect has been lessening in recent years (blue points indicate higher predictions in winter).
This interaction effect is evidenced in Plot 1 as well; the seasonal spread of crime has been lessening as the overall rate of crime decreases over time. While violent crime is a fact of life in Chicago (as it is in every major U.S. city), there is no evidence for increasing rates of overall violent crime.
In fact, comparing the forecast from April-June 2020 (left) to actual data reveals 1,145 fewer violent crimes per month on average than expected.
But what about truly violent crimes, like homicides? Hasn’t the news reported that Chicago is already racking up more homicides in 2020 than in previous years? The data reported by the Tribune appear to be incomplete, representing a fraction of actual homicides collected and publicly available via the CDP. By CDP numbers, Chicago hasn’t exceeded any previous years in terms of homicides for which we sourced data (since 2001). Chicago currently counts 409 homicides as of 24 July 2020; in 2016 the homicide rate peaked at just under 800.
The number of monthly homicides over time was fit using an ARIMA(2,1,1) model; the model was built using data from January 2001 until June 2020. Data were then fit to historic data and forecasted from July to December 2020.
The model yields a decent fit (Plot 6, left); slight to moderate variations in homicides are easily fitted while large departures are more difficult to capture. Notice the upticks in cases during Presidential election years; Chicago has experienced an uptick in homicides every election year (except for 2004), and this year is no exception. With 339 homicides as of 30 June 2020, and another 401 forecasted through 31 December, 2020’s total count is expected to reach (and likely surpass) 740, 90% Prediction Interval (605, 877).
On the Ground: Chicago’s Fight.
Following President Trump’s 22 July announcement that he would send federal troops into Chicago, Mayor Lori Lightfoot responded via Twitter, comparing his move to “tyranny … a dictatorship.”
Indeed, the data do not support the decision to send a surge of federal troops to the Chicago MSA. Overall violent crime rates are down, even in the face of unprecedented historical protests. Racial tensions and protests are also nothing new; protests against inequality, discrimination, and police brutality have continued throughout the years, from the 1960s to the present day. Civilians are exercising their civic rights, peacefully.
Although Chicago may see more homicides this year over last; there is no comparison to the rates of the 1990’s (when no troops were dispatched to The Blues Brothers destination), especially when couched in the context of a downward trend in overall violent crime.
Homicide rates for both 2016 and 2020 (forecasted) pale in comparison to Chicago’s previous homicide records set in 1992 and 1994, when the homicide rate was 33 people per 100,000 residents compared to 28 per 100,000 in 2016 and a forecasted 27.5 per 100,000 residents in 2020.
This year’s uptick is readily interpretable in the context of economic depression, growing discontent with the status quo of racial inequality, and fears amid a global pandemic, all accompanied by an impending Presidential election. Just as families suffered economically during the Spanish flu, families are suffering now due to the loss of employment due to Covid-19. Businesses suffered from decreased spending in 1919; businesses today are struggling with the compound problems of decreased spending and a reduced workforce. While there is undoubtedly a myriad of other factors affecting current crime rates, the data simply don’t support a significant, outlying upswing in violent crime in Chicago.
No one denies that Chicago has a serious crime issue connected to gun violence. It’s not a new story). Few understand the root socio-economic causes. A cycle continues — of failed interventions, continued deaths and trauma, particularly for the city’s historically marginalized, Black southside.
Properly eliminating the causes of violent crime requires America to take a hard look at the reasons for its inequality. To re-evaluate market values and the true utility of an economic system in an age of growing disparities and environmental externalities. To think critically about the systemic motivations for violent crime and implement meaningful, widespread reforms. To respect the facts.
