Getting Started

Welcome to the quickstart guide for OpenPoliceData (OPD)! Here, you should find all you need to learn the basics of OPD.

• New to Python?: Check out the free first python notebook course

• Questions or Comments?: If you questions or comments about anything related to installing or using OPD, please reach out on our discussion board.

Installation

Install OPD with pip from PyPI

pip install openpolicedata

For installation in a Jupyter Notebook, replace pip with %pip.

See here for advanced installation including how to install GeoPandas alongside OPD to enable geospatial analysis of data loaded by OPD.

Import

To use OPD, you must always start by importing it into your Python code:

import openpolicedata as opd

We recommend shortening openpolicedata to opd to make your code more readable.

The Basics

OPD provides access to over 300 police datasets with just 2 simple lines of code:

# Load traffic stops data from Lousiville for the year 2022.
src = opd.Source("Louisville")
tbl = src.load(table_type="TRAFFIC STOPS", year=2022)

The table attribute contains the loaded data as a pandas DataFrame so it can be analyzed with pandas’ simple and powerful capabilities.

# View the 1st 5 rows with pandas' head function
tbl.table.head()

	TYPE_OF_STOP	CITATION_CONTROL_NUMBER	ACTIVITY_RESULTS	OFFICER_GENDER	OFFICER_RACE	OFFICER_AGE_RANGE	ACTIVITY_DATE	ACTIVITY_TIME	ACTIVITY_LOCATION	ACTIVITY_DIVISION	ACTIVITY_BEAT	DRIVER_GENDER	DRIVER_RACE	DRIVER_AGE_RANGE	NUMBER_OF_PASSENGERS	WAS_VEHCILE_SEARCHED	REASON_FOR_SEARCH	ObjectId
0	COMPLAINT/CRIMINAL VIOLATION	DU03293	CITATION ISSUED	M	WHITE	21 - 30	01/02/2022	21:44	M ST ...	4TH DIVISION	BEAT 4	M	WHITE	26 - 30	2	YES	0	1
1	COMPLAINT/CRIMINAL VIOLATION	DV75866	CITATION ISSUED	M	WHITE	51 - 60	07/21/2022	02:00	KEEGAN WAY ...	7TH DIVISION	BEAT 1	M	HISPANIC	16 - 19	1	YES	4	2
2	COMPLAINT/CRIMINAL VIOLATION	DV87754	CITATION ISSUED	M	WHITE	51 - 60	07/21/2022	02:00	KEEGAN WAY ...	7TH DIVISION	BEAT 1	M	HISPANIC	16 - 19	1	NO	0	3
3	COMPLAINT/CRIMINAL VIOLATION	DW19051	CITATION ISSUED	M	WHITE	21 - 30	01/25/2022	11:23	4500 BLOCK SOUTHERN PKWY	4TH DIVISION	BEAT 6	M	WHITE	20 - 25	0	YES	4	4
4	COMPLAINT/CRIMINAL VIOLATION	DX65321	CITATION ISSUED	M	WHITE	31 - 40	01/13/2022	05:30	PRESTON HWY @ OUTER LOOP ...	7TH DIVISION	BEAT 6	M	WHITE	51 - 60	1	YES	3	5

Finding Datasets

OPD provides the datasets module for querying what datasets are available in OPD. To get all available datasets, query the source table with no inputs:

all_datasets = opd.datasets.query()
all_datasets.head()

	State	SourceName	Agency	AgencyFull	TableType	coverage_start	coverage_end	last_coverage_check	Description	source_url	readme	URL	Year	DataType	date_field	dataset_id	agency_field	min_version
0	Arizona	Chandler	Chandler	Chandler Police Department	ARRESTS	2018-01-07	2020-08-26	08/27/2023	Arrest reports completed by a Chandler Police ...	https://data.chandlerpd.com/catalog/arrest-boo...	<NA>	https://data.chandlerpd.com/catalog/arrest-boo...	MULTIPLE	CSV	arrest_date_time	<NA>	<NA>	0.2
1	Arizona	Chandler	Chandler	Chandler Police Department	CALLS FOR SERVICE	2018-01-01	2023-08-26	08/27/2023	This dataset contains details for all of the c...	https://data.chandlerpd.com/catalog/calls-for-...	<NA>	https://data.chandlerpd.com/catalog/calls-for-...	MULTIPLE	CSV	call_received_date_time	<NA>	<NA>	<NA>
2	Arizona	Chandler	Chandler	Chandler Police Department	INCIDENTS	2018-01-01	2023-08-20	08/27/2023	This dataset contains details for all of the g...	https://data.chandlerpd.com/catalog/general-of...	<NA>	https://data.chandlerpd.com/catalog/general-of...	MULTIPLE	CSV	report_event_date	<NA>	<NA>	0.4.1
3	Arizona	Gilbert	Gilbert	Gilbert Police Department	CALLS FOR SERVICE	2006-11-15	2023-07-05	07/06/2023	<NA>	https://data.gilbertaz.gov/maps/2dcb4c20c9a444...	<NA>	https://maps.gilbertaz.gov/arcgis/rest/service...	MULTIPLE	ArcGIS	EventDate	<NA>	<NA>	<NA>
4	Arizona	Gilbert	Gilbert	Gilbert Police Department	EMPLOYEE	NaT	NaT	07/06/2023	A data set of all employees that have previous...	https://data.gilbertaz.gov/datasets/TOG::gilbe...	<NA>	https://services1.arcgis.com/JLuzSHjNrLL4Okwb/...	NONE	ArcGIS	<NA>	<NA>	<NA>	<NA>

The source table provides the information needed to create sources and load data as well as background information. It is a DataFrame that can be filtered with pandas filtering operations. Key information includes:

• State: Optionally used when creating a Source to distinguish ambiguous sources (i.e. same city name in different states)

• SourceName: Original source of the data (typically a shortened name for a police department). Used when creating a Source.

• Agency: Shortened agency / police department name. Typically the same as SourceName. However, it may be MULTIPLE if a datasets contains data for multiple agencies.

• TableType: Type of data (TRAFFIC STOPS, USE OF FORCE, etc.). Used when loading data.

• coverage_start: Start date of data contained in dataset. Combined with coverage_end, this determines the years available for this datasets when loading data. NOTE: Often, agencies store their data in different datasets for different years so one table type may be spread across multiple datasets corresponding to each year of data.

• coverage_end: Most recently checked date for data contained in dataset. Combined with coverage_start, this determines the years available for this datasets when loading data. If the data has been updated by the dataset owner since the date in last_coverage_check, more recent years may be available. NOTE: Often, agencies store their data in different datasets for different years so one table type may be spread across multiple datasets corresponding to each year of data.

• source_url: Homepage for dataset

• readme: Direct URL for data dictionary containing definitions of columns, etc. If empty, the source_url may also contain a data dictionary.

With its optional inputs, query can be used to filter for desired data. Here is a very specific query using all optional inputs:

ds = opd.datasets.query(source_name="Menlo Park", state="California", agency="Menlo Park", table_type="CALLS FOR SERVICE")
ds

	State	SourceName	Agency	AgencyFull	TableType	coverage_start	coverage_end	last_coverage_check	Description	source_url	readme	URL	Year	DataType	date_field	dataset_id	agency_field	min_version
85	California	Menlo Park	Menlo Park	Menlo Park Police Department	CALLS FOR SERVICE	2018-01-01	2018-12-31	07/06/2023	<NA>	https://data.menlopark.gov/datasets/4036c27030...	https://data.menlopark.org/datasets/4036c27030...	https://services7.arcgis.com/uRrQ0O3z2aaiIWYU/...	2018	ArcGIS	<NA>	<NA>	<NA>	<NA>
86	California	Menlo Park	Menlo Park	Menlo Park Police Department	CALLS FOR SERVICE	2019-01-01	2019-12-31	07/06/2023	<NA>	https://data.menlopark.gov/datasets/e88877f5d9...	https://data.menlopark.org/datasets/e88877f5d9...	https://services7.arcgis.com/uRrQ0O3z2aaiIWYU/...	2019	ArcGIS	<NA>	<NA>	<NA>	<NA>
87	California	Menlo Park	Menlo Park	Menlo Park Police Department	CALLS FOR SERVICE	2020-01-01	2020-12-31	07/06/2023	<NA>	https://data.menlopark.gov/datasets/510eb69337...	https://data.menlopark.org/datasets/510eb69337...	https://services7.arcgis.com/uRrQ0O3z2aaiIWYU/...	2020	ArcGIS	<NA>	<NA>	<NA>	<NA>
88	California	Menlo Park	Menlo Park	Menlo Park Police Department	CALLS FOR SERVICE	2021-01-01	2021-12-31	07/06/2023	<NA>	https://data.menlopark.gov/datasets/4c04a71c71...	https://data.menlopark.org/datasets/4c04a71c71...	https://services7.arcgis.com/uRrQ0O3z2aaiIWYU/...	2021	ArcGIS	<NA>	<NA>	<NA>	<NA>

get_table_types finds available table types in OPD. Here, we use optional contains input to only get the table types containing the word “STOPS”:

table_types = opd.datasets.get_table_types(contains="STOPS")
table_types

['PEDESTRIAN STOPS', 'STOPS', 'TRAFFIC STOPS']

Loading Data

The Source class is used to explore datasets and load data. We first need to create a source, which we can use to view all datasets from that source. Let’s create a source of Columbia, South Carolina. We need to specify the state because there are datasets from Columbias from multiple states

src = opd.Source("Columbia", state="South Carolina")
src.datasets

	State	SourceName	Agency	AgencyFull	TableType	coverage_start	coverage_end	last_coverage_check	Description	source_url	readme	URL	Year	DataType	date_field	dataset_id	agency_field	min_version	query
782	South Carolina	Columbia	Columbia	Columbia Police Department	ARRESTS	2016-01-01	2022-12-31	07/07/2023	<NA>	https://coc-colacitygis.opendata.arcgis.com/da...	<NA>	https://services1.arcgis.com/Mnt8FoJcogKtoVBs/...	MULTIPLE	ArcGIS	Arrest_Date	<NA>	<NA>	0.2	NaN
783	South Carolina	Columbia	Columbia	Columbia Police Department	FIELD CONTACTS	2016-01-01	2022-12-31	07/07/2023	Field Interview is a collection of data result...	https://coc-colacitygis.opendata.arcgis.com/da...	<NA>	https://services1.arcgis.com/Mnt8FoJcogKtoVBs/...	MULTIPLE	ArcGIS	TOC	<NA>	<NA>	<NA>	NaN

To get a list of available table types:

src.get_tables_types()

['ARRESTS', 'FIELD CONTACTS']

You can get the number of records for a dataset using get_count. Let’s get the number of records in the year 2022 for the FIELD CONTACTS dataset.

src.get_count("FIELD CONTACTS", 2022)

2382

You can find which years are available for a given table type:

src.get_years(table_type="FIELD CONTACTS")

[2016, 2017, 2018, 2019, 2020, 2021, 2022]

Now, let’s load in some field contacts data for 2022.

tbl = src.load("FIELD CONTACTS", 2022)
tbl

state: South Carolina,
source_name: Columbia,
agency: Columbia,
table_type: FIELD CONTACTS,
year: 2022,
description: Field Interview is a collection of data resulting from citizen contact related to suspicious activity.,
url: https://services1.arcgis.com/Mnt8FoJcogKtoVBs/arcgis/rest/services/FieldInterview/FeatureServer/0,
date_field: TOC,
source_url: https://coc-colacitygis.opendata.arcgis.com/datasets/ColaCityGIS::field-interview-1-1-2016-3-31-2022/about

The loaded data is contained in a pandas DataFrame in the table attribute.

tbl.table.head(2)

	OBJECTID	Case_Num	TOC	Address	City	Zip	State	Age	Race	Sex	Contact_Type	Year	geometry
0	25351	220000108	2022-01-01 21:47:00	12XX Main St		29201		32	W	M	Field Interview	2022.0	POINT (1989801.776 788862.968)
1	25350	220000161	2022-01-02 15:05:00	21XX Main St		29201		29	B	M	Field Interview	2022.0	POINT (1988210.189 793174.093)

Data can be saved locally as CSV files. This allows you to:

• Open the data using the software of your choice

• Re-open the data in OPD from a local copy

tbl.to_csv()
new_src = opd.Source("Columbia", state="South Carolina")
new_tbl = new_src.load_from_csv(2022, table_type="FIELD CONTACTS")
new_tbl.table.head(2)

	OBJECTID	Case_Num	TOC	Address	City	Zip	State	Age	Race	Sex	Contact_Type	Year	geometry
0	25351	220000108.0	2022-01-01 21:47:00	12XX Main St	NaN	29201	NaN	32.0	W	M	Field Interview	2022.0	POINT (1989801.7762467265 788862.9678477645)
1	25350	220000161.0	2022-01-02 15:05:00	21XX Main St	NaN	29201	NaN	29.0	B	M	Field Interview	2022.0	POINT (1988210.189304456 793174.0931758583)

Some datasets contain data for every agency in a state. In this case, you may want to know what agencies are available and optionally, only want agencies containing the word Arlington.

src = opd.Source("Virginia")
agencies = src.get_agencies(table_type="STOPS", partial_name="Arlington")
agencies

["Arlington County Sheriff's Office", 'Arlington County Police Department']

We may also want only load data from a specific agency:

tbl = src.load(table_type="STOPS", year=2022, agency="Arlington County Police Department")

Data Standardization

One of the challenges in analyzing police data is that different agencies will use different column names for the same data and will use different codes and terms for the data in the columns. Particularly, if you are looking at multiple datasets, it is valuable for the data to be standardized so that you know in advance what some key columns will be called and what values will be in those columns.

To provide the user with more consistent column names and data, OpenPoliceData provides powerful tools to automatically standardize column names and data in order. Columns that OpenPoliceData can standardize include:

• Date

• Time

• Gender

• Age

• Race

• Ethnicity

In addition, OpenPoliceData will combine separate date and time columns into a single datetime column and race and ethnicity into a single combined race column.

Let’s examine what columns are in the Phoenix Use of Force dataset:

src = opd.Source("Phoenix")
tbl = src.load(table_type="USE OF FORCE", year=2022, pbar=False)  # pbar=False does not show progress bar
# Only showing 1st 30 due to large number of columns
tbl.table.columns[:30]

Index(['INC_IA_NO', 'INC_IR_NO', 'EMP_BADGE_NO', 'CIT_NUMBER', 'INC_DATE',
       'INC_YEAR', 'INC_HOUR', 'INC_DAY_WEEK', 'INC_LOC_COUNTY',
       'HUNDRED_BLOCK', 'INC_CITY', 'INC_STATE', 'INC_ZIPCODE', 'INC_PRECINCT',
       'CIT_INJURY_YN', 'CIT_GENDER', 'CIT_AGE', 'SUBJ_AGE_GROUP', 'CIT_RACE',
       'CIT_ETHNICITY', 'SIMPLE_SUBJ_RE_GRP', 'SIMPLE_EMPL_RE_GRP', 'EMPL_SEX',
       'CIT_RESIST_AGG_ACTIV_AGGRESSN', 'CIT_RESIST_ACTIVE_AGGRESSN',
       'CIT_RESIST_ACTIVE_RESISTANCE', 'CIT_RESIST_PASSIVE_RESISTANCE',
       'CIT_RESIST_PSYCH_INTIMIDATION', 'CIT_RESIST_VRBL_NONCOMPLIANCE',
       'CIT_RESIST_NONE'],
      dtype='object')

The data has several columns related to subject demographics: - ‘CIT_GENDER’ - ‘CIT_AGE’ - ‘SUBJ_AGE_GROUP’ - ‘CIT_RACE’ - ‘CIT_ETHNICITY’ - ‘SIMPLE_SUBJ_RE_GRP’

These are not common labels used by datasets from other agencies (i.e. they cannot be predicted in advance). Additionally, when looking at them, the column labels are a bit hard to decipher because they are not all clear and are not consistent in their naming conventions. The data uses 2 different short descriptions for the same subject (CIT and SUBJ), and the RE in ‘SIMPLE_SUBJ_RE_GRP’ is for race/ethnicity so there 3 columns related to race and ethnicity.

Similarly, the office demographics data uses the same RE abbreviation, and the user must know that EMPL is short for employee, which means the officer. - ‘SIMPLE_EMPL_RE_GRP’ - ‘EMPL_SEX’

OPD’s data standardization will automatically identify columns and rename them to standard column names (while optionally keeping the original columns as RAW_{original name}). This enables the user to know in advance what the column names will be.

Now let’s examine what’s in the subject race and ethnicity columns:

print(f"The unique values in the race column (CIT_RACE) are {tbl.table['CIT_RACE'].unique()}")
print(f"The unique values in the ethnicity column (CIT_ETHNICITY) are {tbl.table['CIT_ETHNICITY'].unique()}")
print(f"The unique values in the race ethnicity column (SIMPLE_SUBJ_RE_GRP) are {tbl.table['SIMPLE_SUBJ_RE_GRP'].unique()}")

The unique values in the race column (CIT_RACE) are ['White' 'Black' 'American Indian / Alaskan Native'
 'Asian / Pacific Islander' 'Unknown' 'AmIndian']
The unique values in the ethnicity column (CIT_ETHNICITY) are ['Hispanic' 'Non-Hispanic' 'Unknown']
The unique values in the race ethnicity column (SIMPLE_SUBJ_RE_GRP) are ['Hispanic' 'Black or African American' 'White' 'Other']

A few items to note: - Naming conventions are not consistent: Indigenous subjects are labeled ‘American Indian / Alaskan Native’ and ‘AmIndian’. Black subjects are labeled ‘Black’ in the race column and ‘Black or African American’ in the race/ethncity column - ‘Asian / Pacific Islander’ and ‘American Indian / Alaskan Native’ appear in the race column but not the race/ethnicity column, which does not seem correct unless ALL Asian/Pacific island and Indigenous subjects were Hispanic/Latino (since Hispanic/Latino is typically used for Latinos of all races in race/ethnicity columns), which may seem unlikely.

Let’s look at some cases where the race is ‘Asian / Pacific Islander’ or ‘American Indian / Alaskan Native’:

i = tbl.table['CIT_RACE'].isin(['American Indian / Alaskan Native', 'Asian / Pacific Islander'])
tbl.table[['CIT_RACE', 'CIT_ETHNICITY', 'SIMPLE_SUBJ_RE_GRP']][i].head()

	CIT_RACE	CIT_ETHNICITY	SIMPLE_SUBJ_RE_GRP
5228	American Indian / Alaskan Native	Non-Hispanic	Other
5229	American Indian / Alaskan Native	Non-Hispanic	Other
5233	Asian / Pacific Islander	Non-Hispanic	Other
5234	American Indian / Alaskan Native	Hispanic	Hispanic
5236	Asian / Pacific Islander	Non-Hispanic	Other

Subjects labeled as ‘Asian / Pacific Islander’ and ‘American Indian / Alaskan Native’ in the race column are relabeled as ‘OTHER’ in the race/ethnicity column. Thus, although it is often preferred to use a combined race/ethnicity column, the way that ‘SIMPLE_SUBJ_RE_GRP’ has been generated actually removes key information.

OPD’s standardization allows the user to more quickly analyze data by automatically identifying columns, renaming them to standard column names, and standardizing the data in those columns.

tbl.standardize()

Let’s look at what the standardization did using get_transform_map:

std_map = tbl.get_transform_map(minimize=True)
for t in std_map:
    print(f"{t}\n")

orig_column_name: INC_DATE,
new_column_name: DATE,
data_maps: None

orig_column_name: CIT_RACE,
new_column_name: SUBJECT_RACE,
data_maps: {'White': 'WHITE', 'Black': 'BLACK', 'American Indian / Alaskan Native': 'INDIGENOUS', 'Asian / Pacific Islander': 'ASIAN / PACIFIC ISLANDER', 'Unknown': 'UNKNOWN', 'AmIndian': 'INDIGENOUS'}

orig_column_name: SIMPLE_EMPL_RE_GRP,
new_column_name: OFFICER_RACE/ETHNICITY,
data_maps: {'White': 'WHITE', 'Hispanic': 'HISPANIC / LATINO', 'Other': 'OTHER', 'Black or African American': 'BLACK', nan: 'UNSPECIFIED'}

orig_column_name: CIT_ETHNICITY,
new_column_name: SUBJECT_ETHNICITY,
data_maps: {'Non-Hispanic': 'NON-HISPANIC/NON-LATINO', 'Hispanic': 'HISPANIC/LATINO', 'Unknown': 'UNKNOWN'}

orig_column_name: ['SUBJECT_RACE', 'SUBJECT_ETHNICITY'],
new_column_name: SUBJECT_RACE/ETHNICITY,
data_maps: None

orig_column_name: SUBJECT_RACE/ETHNICITY,
new_column_name: SUBJECT_RE_GROUP,
data_maps: None

orig_column_name: OFFICER_RACE/ETHNICITY,
new_column_name: OFFICER_RE_GROUP,
data_maps: None

orig_column_name: CIT_AGE,
new_column_name: SUBJECT_AGE,
data_maps: None

orig_column_name: SUBJ_AGE_GROUP,
new_column_name: SUBJECT_AGE_RANGE,
data_maps: {'30s': '30-39', '20s': '20-29', '40s': '40-49', '<20': '0-20', '50s': '50-59', '60s': '60-69', 'Not Available': 'Not Available', '70s': '70-79', '90s': '90-99', '80s': '80-89'}

orig_column_name: CIT_GENDER,
new_column_name: SUBJECT_GENDER,
data_maps: {'Male': 'MALE', 'Female': 'FEMALE'}

orig_column_name: EMPL_SEX,
new_column_name: OFFICER_GENDER,
data_maps: {'Male': 'MALE', 'Female': 'FEMALE', nan: 'UNSPECIFIED'}

get_transform_map shows changes made by standardization including the following: - Identifying all demographics columns - Identifying the more informative CIT_RACE as the race column instead of SIMPLE_SUBJ_RE_GRP - The identified race (CIT_RACE) and ethnicity columns (CIT_ETHNICITY) were converted to SUBJECT_RACE and SUBJECT_ETHNICITY, respectively and then SUBJECT_RACE and SUBJECT_ETHNICITY are combined into SUBJECT_RACE_ETHNICITY. - The SUBJECT_RACE_ETHNICITY was copied to another column called SUBJECT_RE_GROUP. RE_GROUP columns like SUBJECT_RE_GROUP and OFFICER_RE_GROUP are added for those who want to be able to easily use a RACE_ETHNICITY column if it exists or a RACE column otherwise. The RE_GROUP column will be a copy of the RACE_ETHNICITY column if a RACE_ETHNICITY column has been generated or a RACE column if a RACE column was found but a RACE_ETHNICITY was not generated. - EMPL was identified as indicating officer demographics - The cryptically named SIMPLE_EMPL_RE_GRP was identified as an OFFICER_RACE column - Values of race, gender, age, and age group are standardized to values that will be consistent across all OPD-loaded tables

In data_maps, get_transform_map also includes dictionaries indicating original values (key) and the resulting standardize value (value).

Printing the columns shows that the standardized columns are in the front while the original columns are prepended with RAW and moved to the back of the list.

print(f"The 1st 10 columns after standardization are: {tbl.table.columns[:12]}")
print(f"The last 11 columns after standardization are: {tbl.table.columns[-9:]}")

The 1st 10 columns after standardization are: Index(['DATE', 'SUBJECT_RACE', 'OFFICER_RACE/ETHNICITY', 'SUBJECT_ETHNICITY',
       'SUBJECT_RACE/ETHNICITY', 'SUBJECT_RE_GROUP', 'OFFICER_RE_GROUP',
       'SUBJECT_AGE', 'SUBJECT_AGE_RANGE', 'SUBJECT_GENDER', 'OFFICER_GENDER',
       'INC_IA_NO'],
      dtype='object')
The last 11 columns after standardization are: Index(['FORCE_OTHER', 'RAW_INC_DATE', 'RAW_CIT_GENDER', 'RAW_CIT_AGE',
       'RAW_SUBJ_AGE_GROUP', 'RAW_CIT_RACE', 'RAW_CIT_ETHNICITY',
       'RAW_SIMPLE_EMPL_RE_GRP', 'RAW_EMPL_SEX'],
      dtype='object')

Finally, we can view what values that the new SUBJECT_RE_GROUP column contains:

tbl.table["SUBJECT_RE_GROUP"].unique()
# NOTE: We also provide a columns enum so that the user does not have to remember the standardized column names. This would produce the same thing:
# tbl.table[opd.defs.columns.SUBJECT_RE_GROUP].unique()

array(['HISPANIC/LATINO', 'BLACK', 'WHITE', 'INDIGENOUS',
       'ASIAN / PACIFIC ISLANDER', 'UNKNOWN'], dtype=object)

Other Topics

• Data Standardization Guide (including methods for customizing the standardization process)

• Related Tables