The COVID-19 threat and the Shelter-in-Place (SIP) order has changed many things about life in Oakland. One recurring question has been: what about crime? In many ways, things seem more quiet and we might expect crime has gone down, but we also hear stories of isolated events.

Data collected by the OakCrime.org project can give one perspective on this question. This project by OpenOakland.org captures data sources published by Oakland Police Department. The collection has gone on for years, and several interfaces for analysis are provided. The current experiment looks at the data during the period since March 17, 2020 and asks how it compares to historical data.

The most obvious comparison is with the days just before March 17. Figure 1 shows the number of crimes reported each day. Indeed, the level of crime has gone generally down since that date, and quite dramatically. In fact, the major drop in crime anticipated the March 17 date by almost a week. The average number of crimes reported was 179 per day before March 11 and 118 after, a drop of almost a third.

Figure 1: Crime frequency

A second reasonable comparison is to the same early spring period In 2019. A plot of daily crime frequencies for the two years as shown in Figure 2. The 2019 data do not show any similar dramatic drop in early March.

Figure 2: Crime frequency – compared to 2019

Beyond looking at just the total frequency of crime, it is also possible to consider how different types of crime have changed before and after the SIP order. Figure 3 shows the variation in types of crime as they change before and after this date.

Figure 3: Change in crime categories before/after SIP

The statistic reported is the difference in the daily average rate of occurrence of each crime type before and after March 17, with negative numbers suggesting they have dropped since SIP, and positive numbers that they have increased. Increases (red) and decreases (green) that are statistically significant^§ have been highlighted. (Note that two crime categories, VANDALISM and LARCENY_BURGLARY_AUTO, have dropped so dramatically that they would correspond to bars going far below the rest of the graph.)

The number of HOMICIDE and HOMICIDE_UNEXPLAINED ^# have both gone up, but only the change in HOMICIDE_UNEXPLAINED is significant. There have been reports nationally regarding increasing incidence of domestic violence, but the situation in Oakland is mixed: DOM-VIOL and DOM-VIOL_BATTERY have gone down, while DOM-VIOL_BATTERY-SPOUSE is up. None of these changes are statistically significant, however.

As with all statistics, the devil is in the details and it is important to ask exactly how the categories are derived from OPD data. The OakCC crime classification system, including the open source code on which it is based, is described in detail elsewhere. In brief, OakCC classification depends on three attributes provided in OPD records. In some cases, the California Penal Code (PC) is provided. When it is, this attribute is the most reliable indicator of the type of crime; unfortunately it is provided for only about 10% of reported incidents. The remaining 90% are classified based on OPD’s CRIME_TYPE and DESCRIPTION attributes.

Table 1 demonstrates for a single category, WEAPONS, some of the variability of OPD-reported incidents captured in that category. There were 33 different combinations of CRIME_TYPE and DESCRIPTION included by OakCC in the WEAPONS category; Table 2 lists only the most frequent. Of this set, 10 crimes have depended on the PC code, 19 on a combination of CRIME_TYPE and DESCRIPTION, and 60 on DESCRIPTION alone.

Table 1: Range of incidents in WEAPONS

These details help to make clear that the question of changing crime patterns in Oakland before vs. after SIP shares the same fundamental issue of transparency as all citizen analysis of OPD crime data: Are we seeing evidence of changes in real crime activity, or in OPD’s reporting of it? For example, it is easy to imagine that OPD officers and record keepers, in the churn of huge changes in their jobs and the city they patrol, that a field like CRIME_TYPE just happens to have been left blank more often than during calmer times. Or, are more weapons being used since SIP?

A great, simple step towards transparency would be to include California Penal Code (or Vehicle Code, Health and Safety Code, etc.) as well as the UCR code (Uniform Crime Reporting, supplied to the FBI) as part of primary OPD data publishing via data.oaklandnet.com. This would reduce the impact of the vagaries of crime reporting via the textual CRIME_TYPE and DESCRIPTION attributes, and make the connection between crime in Oakland and OPD’s policing more clear.