Data Limitations - PhilaDAO Data Dashboard
Limitations
This website currently only catalogs cases that were handled in adult court, not juvenile court. The website does include cases against juveniles that were handled and completed in adult court.
Nearly all of the data used to construct this website are subject to human error. Police officers enter reports on arrests and court clerks enter data on cases. Mistakes made during data entry can appear in our data. Although these errors exist on individual cases, we have no reason to think that errors are common, and their impact is minimized when large amounts of data are available.
Our data systems are far from perfect, and we do not always have analyzable data on everything that we would like to be able to report on. Because of that, there are reports that would be very useful for transparency and accountability that we cannot make. As an example, as of yet, we do not electronically record the amount of bail our office requests at bail hearings, so we cannot report on how our bail requests match or differ from bail assigned by a bail magistrate. We are working to improve our systems so that we can better record and report on everything that happens in the criminal justice system.
Unless otherwise noted
The method used to categorize an event in the criminal legal system is different for an incident, an arrest, and a case within the court system.
Incidents receive an intial classification by the Philadelpiha Police Department at the time of reporting, which may or may not reflect the reality of the event. When an arrest is made, the police recommend a set of charges that they believe the arrestee should be charged with. Ultimately, the District Attorney’s Office determines whether to charge an arrestee and what to charge that person with. The Dashboard uses incident classification to categorize incidents, police recommended charges to categorize arrests, and the actual charges brought in court to categorize the case from the time of charging and beyond.
The advantage of this system is that it allows users of the Dashboard to view incidents, arrests, and cases as they actually moved through the system. If an incident is originaly classified as an Aggravated Assault and the police recommended charges for Aggravated assault, the dashboard will reflect that even if the DAO ultimately disagreed and charged the case as a Simple Assault. Similarly, the dashboard will count that case as a simple assault for the purpos of court-related activity, like bail, case outcomes, and sentencing, rather than keep the original police category of Aggravated Assault. This system also allows for a more granular look at the system, breaking incidents categorized as, for example, Illegal Firearm Possession into its component offenses of Firearm Possession by a Prohibited Person and Firearm Possession Without a License.
The disadvantage of this system is that it limits the ability to track a set of incidents through the entire system. For example, an incident that was initially called a “Robbery/Gun” on the Incidents page could become an “Aggravated Assault” on the Arrests page, which could then become an “Attempted Homicide” on the Charges page. That means that it isn’t possible to use the dashboard to know, for example, what percentage of Robbery/Gun incidents lead to an arrest, and what the case outcomes for the subsequent cases are.
For the first two years of the Dashboard’s existence, we did the opposite: we used the original police classification of each incident to classify that case as it moved through the system. While there were advantages to this system, it distorted outcomes in the court system in a way that we believe was inaccurate.
We feel that it is important to show the most accurate picture of the criminal legal system that can be understood at each stage of the process. We also feel it is the responsibility of the District Attorney’s Office to show the decisions we have made. We cannot go ‘back in time’ and reclassify incidents and arrests after a charging decision has been made without distorting history and presenting a false view of these stages. Given the constraints of our data, and for these reasons, we feel that categorizing each event at each stage allows for the most complete representation of the system available. For more information about this change, please read this data story.
Over time, offenses may be recategorized if underlying facts change. For example, if a person is shot but not killed, the offense may be categorized as an aggravated assault with a gun. If the victim later dies, the offense may be recategorized as homicide. When offenses are recategorized in this way, the data will update to reflect the new categorization. As a user of the site, you may notice historic numbers changing over time.
Information about Fatal and Non-Fatal Shootings is based on the Open Data Philly Shooting Victims dataset which only includes shooting victims since 2015. That means that all information about shooting victims (arrests, cases charged, outcomes, etc…) only reflects shooting incidents since 1/1/2015. As an example, the Case Outcomes report shows very few cases resolved in 2015 and a very high rate of non-convictions. The reason for this is that very few shooting incidents that occured in 2015 were also resolved in 2015; those that were resolved that quickly were often quickly dismissed.
We have made daily-summary CSV files available to the public so that the public can discover new insights into the criminal justice system and our work. Because these files do not include case-level information, but rather are summaries of all of the cases/arrests on a day, it is not always possible to recreate the information contained in the reports using the CSV files.
There are numerous features that we would like to make available in the future, such as the ability to view data based on demographics of the accused and victims, like race and gender. We aim to continually improve this website to add more data, features, and interactivity in the future.
In addition, the District Attorney’s Office (DAO) is presently working with criminal justice partners in Philadelphia to understand the limitations of current race and gender data, and to analyze what is available.
The reports on this website are generated automatically. Sometimes errors will appear before we find them. Please let us know about any errors so that we can fix them quickly.
The data used to create this dashboard includes all arrests that took place since 1/1/2012 and all cases that had hearings since 1/1/2012. As a general matter, this should not affect any of our statistics. Cases that began with an arrest prior to 2012 will be classified as “other offenses” since the arrest classification was not available to our data.
Cases that are overturned after appeal are often overturned years after the arrest. Because we can only classify arrests from 2012 forward, it is impossible to classify the offense type of most overturned cases. Rather than misclassify these cases as “Other” we decided to not include this categorization. These cases still appear in our data, generally categorized as “Withdrawn” cases.
Generally the Philadelphia Police is the arresting agency, but our data includes other police agencies such as Southeastern Pennsylvania Transportation Authority (SEPTA) police.
The DAO’s declination rate (the rate at which we decline to charge arrests) cannot be calculated from the data on this website. This is because declination is more complicated than comparing arrests to cases charged. For example, it is common for the DAO to open multiple cases based on a single arrest. The DAO can also decline and ask the police to gather more evidence before charging an individual case. If the person is later rearrested and charged, the prior declination will be lost.
When calculating future years of incarceration, we assume that person a sentenced to a life sentence or a death sentence will live to their normal life expectancy using the 2015 actuarial table from the Social Security Administration. For example, a 25-year-old male sentenced to life in prison would be calculated to have a sentence of 52.3 years.
We made a decision to only include people who have completed their diversion program in our data, rather than people who started but never completed. This means that the District Attorney’s Office diverts an even larger number of cases each year, but individuals may not complete their assigned program.
The District Attorney’s Office does not have data on crime incidents in the City other than what is available to the public through the City of Philadelphia’s open data program. We use OpenDataPhilly’s Parts I & II Crime Incidents dataset to study crime events, many of which do not lead to arrests. This leads to a number of limitations:
- This database only records crimes classified as Part I or Part II crimes by the FBI, and does not include all violations. This means that not all crimes that happen within the City are shown in our Incident report.
- The incidents recorded in this dataset are given a generalized of the FBI Uniform Crime Reporting (UCR) code. We cannot determine the specific nature offenses from the data available. For example, all Thefts are recorded as “Theft” in the Incident Report, while we can separate “Theft” from “Retail Theft” in all other reports. Another example of this is that all drug violations are recorded as simply Drug Offenses, and we cannot separate them into possession and sales as we do in all other reports.
- This dataset is intended to allow members of the public, community groups, researchers, and other departments to access up-to-date and actionable data on crime incidents in Philadelphia. It is not intended to be an authoritative crime reporting database. The statistics generated from this database may not reflect the official reporting by the Philadelphia Police Department (PPD).
- As a result of the point above, it is possible to derive three different numbers for Homicides from Police sources. We have summarized the OpenDataPhilly Incidents and derived the total number of incidents that have been classified in that data as Homicides. This is often different than the number reported on the front page of the Police Crime Maps & Stats page. Another number for homicides can be found in the Police’s weekly Crime Stats Reports; this number is also often different than those found in the Incident Data and Crime Maps & Stats page.
We have decided to include the Incident Report to help those who view the dashboard get as complete a view of the criminal justice system as possible, starting from the incident to the future years of supervision or incarceration generated.
We currently only report the amount of bail initial assigned to a case, not whether it was paid or later modified.
On a modest number of older cases, we know the amount of bail that was set, but we don’t know whether that bail was set as Cash bail (requiring the person to pay in order to be released) or SOB (allowing the person to be released without paying any money). In these situations we assume that the person is required to pay Cash bail.
Due to the quality of the data available, our ability to locate incidents and associated arrest and case information throughout the City is currently limited to what we can connect from public data sources. This affects what data we can deliver for download. Data by Police District is the most complete, but data independent from Police Districts, such as Census Tracts, may be incomplete because we’re missing precise location information that allows us to map an incident to a geography. We estimate that approximately 8% of all arrest and case information is classified as Unknown. Census Tract information is limited to weekly due to file size limitations.
The data by ZIP code available for download is actually by ZIP Code Tabulation Area, a geography defined by the United States Census Bureau using the majority ZIP Code for addresses in Census blocks. In densely populated cities like Philadelphia, ZCTAs and postal ZIP codes are functionally identical. For more information, visit the U.S. Census Bureau’s page on ZIP Code Tabulation Areas.
We use ZCTAs in the downloadable sets for two reasons:
- ZCTAs allow for easy and clean connection to U.S. Census data.
- The location data of the incident tends to be more reliable than the ZIP Code in the Court information. Currently, Court data contains a large number of missing, incomplete, wrong, or otherwise bad ZIP Codes. This is likely due to the data being human-entered in the field, where one might not know what ZIP code they are currently in.