Summary

A review Boston Police Department (BPD) Field Interrogation and Observation (FIO) data from the period 2011-April 2015 revealed that among the BPD officers who submitted FIO reports in that period, the 20% of the officers who submitted the most FIO reports were more likely to have encounters with black suspects, and the remaining group of officers were more likely to have encounters with suspects of other racial groups.

Introduction

The Boston Police Department uses proactive policing tactics, including the Field Interrogation, Observation, Frisk, and/or Search program (referred to in this report as FIO) to disrupt criminal activities. In the FIO program, someone who the BPD officer believes is involved with some criminal activity, may be observed or even stopped and searched by that officer.

According to the BPD, FIO reports provide officers with a mechanism to document up-to-date information regarding known criminals and their associates for law enforcement purposes and include situations ranging from observation without any interaction to consensual encounters with an individual to a stop and frisk or search.

The FIO reports that were used in this analysis contained information on the date and location of the FIO activity, as well as demographic information about the officers and suspects involved in the FIO encounter.

In a study of an earlier set of FIO data from 2007-2010, Jeffrey Fagan of Columbia Law School and School of Public Health, and Anthony A. Braga of the School of Criminal Justice at Rutgers University in Newark (and a research fellow at the Kennedy School at Harvard) focused on racial disparity when it came to the race of the officers and of the persons who were the subject of FIO reports, and found that there were racial disparities when it came to the treatment of people subject to an FIO encounter, specifically the likelihood of being frisked and searched (An Analysis of Race and Ethnicity Patterns in Boston Police Department Field Interrogation, Observation, Frisk, and/or Search Reports).

As part of the BPD’s effort to maintain transparency about the FIO process, in early January 2016, BPD released a set of data covering the period from January 2011 to April 2015, along with a few summary statistics about the data.

The BPD also stated that their analysis of that data was not yet complete, and implied that their intention with their analysis was to show decreases in the kinds of disparities observed in the earlier study.

Who is this report for?

This report may be useful for the following kinds of groups:

  • Anyone who is interested in the policing practices of the BPD, in the patterns of suspected criminal activity in Boston, and racial makeup of those who were subjected to the BPD’s FIO process.

  • Organizations and individuals who are concerned about possible discriminatory policing practices by the BPD.

  • Law enforcement agencies who are interested in the experiences of the BPD when it comes to implementing proactive policing practices similar to the FIO program.

Methods and Data

After downloading the data and removing variables that were either redundant or were not relevant to the analysis, the data was first in order to create summaries of the data with respect to how the FIO reports were distributed by day of the week, month of the year, and by racial category of those subjected to an FIO encounter, and to identify the populations for BPD officers and the locations within Boston that were responsible for a disproportionate share of FIO reports.

The BPD officers were divided into a pair of categories, the 20% of the officers with the most FIO reports, and all other officers who submitted at least one FIO report, and these two groups were compared with respect to the proportion of FIO reports involving specific racial groups.

This division was directly related to the intent of this analysis - to explore the possible racial disparities in the officers conducting the FIO encounters based on the officers’ performance, rather than their race. In contrast, one of the analytical strategies in the earlier study by Fagan and Braga involved looking at the officers’ race to see if there was a relationship between the race of the officers and the rate at which youths from a particular racial group were stopped.

FIO data preparation

The FIO data was downloaded from a City of Boston site that provided the data in numerous formats. For this analysis, the CSV version of the data was downloaded.

This data was a redacted version of the full FIO data, excluding any personal identifying information about those who were subject to an FIO encounter. Unlike the data used in the earlier study of Fagan and Braga, the data released by the BPD in January 2016 did not contain data that would associate an FIO report with a particular suspect.

Upon request, the BPD also provided a data dictionary that defined 43 of the 44 variables in the database. For the analysis done for this study, the raw FIO was processed and updated in the following ways:

  • Fourteen of the variables were removed because they were either redundant, or if they did not add potentially useful information for this particular study.

  • For consistency, variables with blank entries, or with an entry code indicating missing or unknown data, were changed to NA for Not Available. NA is a code used in R, the statistical analysis program used for this analysis.

  • Six variables were added, five of which were related to a transformation of the date of of the FIO encounter, and the sixth of which was a transformation of a variable describing the racial identity of the subject of the FIO encounter.

Combining racial codes used by BPD

A review of the BPD raw data revealed that seven codes were used for a racial identifier, and five of those were combined in the racial identity variable that was added. Two of codes in the raw data indicated no data entered, or unknown, and they were changed to NA. Three of the codes, for Asian, Middle East/East Indian, and American Indian/Alaska native, were combined into one code (Other) for two reasons. First, this analysis was primarily concerned with racial disparities among FIO reports involving black, white, and Hispanic subjects, and second because less than 2% of FIO reports involved other races.

This transformed FIO data is available at the the following address: http://www.airsafe.com/analyze/fio_airsafe.csv

Summary and overview of the data

This summary will be divided into three sections. The first section will focus on comprehensive statistics covering either the entire body of data, or data that makes comparisons between years. The second type of summary will exclude data from 2015, because the data only covers the period up to April 30, 2015.

Frequency of FIO reports per day

From January 2011 to April 2015, the number of FIO reports submitted per day ranged from 3 to 247 with an average of 94.6. Below are the summary statistics and a chart showing the distribution of daily submissions.

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##    3.00   71.00   93.00   94.59  118.00  247.00

FIO Reports by day of the week

Below are the totals for the number of FIO reports by day of the week for the period 2011-2014, both as numerical values and a bar graph. Because the data from 2015 covered only the first four months of the year, the following graphics are limited to the years 2011-2014.

## 
##   Sun   Mon   Tue   Wed   Thu   Fri   Sat 
## 15651 20323 21681 21071 21195 22308 20458

FIO Reports by month of the year

Below are the totals for the number of FIO reports by month of the year week for the period 2011-2014, both as numerical values and a bar graph. Because the data from 2015 covered only the first four months of the year, the following graphics are limited to the years 2011-2014.

##    Month Records
## 1    Jan  10,745
## 2    Feb  11,237
## 3    Mar  13,099
## 4    Apr  12,330
## 5    May  15,100
## 6    Jun  11,686
## 7    Jul  12,845
## 8    Aug  12,528
## 9    Sep  13,295
## 10   Oct  12,361
## 11   Nov   8,980
## 12   Dec   8,481

FIO encounters by combination of weekday and month

It is also possible to summarize FIO distribution by weekday and month using either a matrix of numbers or a heat map. The matrix of values would consist of 84 cells representing a combination of the month and day of the week. The heat map would reflect the data in the previous table, with the The colors corresponding to a level of intensity with white being on the low end of the scale and dark blue on the upper end.

Below is a matrix showing the number of FIO reports by both day of the week and month of the year for 2011-2014, followed by a heat map of the same information. In the heat map, the darkest cell corresponds to the cell (combination of month and day of the week) with the most FIO reports.

##      
##        Sun  Mon  Tue  Wed  Thu  Fri  Sat
##   Jan 1227 1737 1631 1591 1595 1630 1334
##   Feb 1222 1553 1560 1597 1774 1792 1739
##   Mar 1363 1860 1908 1940 1913 2052 2063
##   Apr 1303 1858 1879 1665 1751 1998 1876
##   May 1637 2024 2225 2253 2527 2436 1998
##   Jun 1474 1654 1814 1739 1613 1767 1625
##   Jul 1552 1939 2258 1890 1755 1738 1713
##   Aug 1307 1475 1804 2004 2005 2083 1850
##   Sep 1499 1829 2057 1996 1997 2049 1868
##   Oct 1158 1954 2011 1906 1802 1841 1689
##   Nov 1043 1250 1380 1297 1116 1481 1413
##   Dec  866 1190 1154 1193 1347 1441 1290

FIO distribution by BPD officers

From 2011 to April 2015, 149,545 FIO reports were submitted by 1,785 BPD officers. There was a very wide range for the number of submissions for this group of officers, and below is a summary that illustrates that range:

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##    1.00    3.00   15.00   83.78   73.00 2904.00

On the low end, 266 officers, or 14.9% of all officers, submitting a single FIO report during this period. A total of 632 officers, representing the bottom 35.4% of the 1,785 officers who submitted FIO reports during the study period, were responsible for 1% of the total FIO reports. These officers filed an average of 2.37 FIO reports during this period.

On the high end, a total of 357 officers, representing the top 20% of the 1,785 officers who submitted FIO reports during the study period, were responsible for 81.3% of the total FIO reports. These officers filed an average of 340.6 FIO reports during this period.

Among the very top performers, 18 officers, representing the top 1.01% of the 1,785 officers who submitted FIO reports during the study period, were responsible for 26,456 , or 17.7% of all FIO reports. These officers filed an average of 1,470 FIO reports during this period.

To illustrate the differences in FIO contributions, the following chart splits shows the contributions of four officer groups:

  • First 1% output - Those with the lowest number of FIO submissions who collectively submitted 1% of the FIO reports,

  • Other BPD - Those with the lowest number of FIO submissions who collectively submitted 20% of the FIO reports,

  • Top Quintile - The highest performing 20% of the officers who FIO submissions who collectively submitted about 80% of the FIO reports, and

  • Top 1% - The top 1% in FIO submissions.

Note that there is overlap in the first pair and second pair of performance measures. The ‘First 1% output’ group is part of the ‘Other BPD’ group, and the ‘Top 1%’ group is part of the ‘Top quintile’ group.

FIO distribution by race

A total of 143,073 FIO reports, or 95.7% of all FIO reports had a racial identity assigned by the BPD. Of those FIO reports with an assigned racial identity, 98.7% of these being either black (61.2%), White (23.9%), or Hispanic (13.7%).

Below is the breakdown of FIO reports by race, excluding FIO reports with no race identifier and with the following racial identifiers grouped into the “All other” category: Asian, Middle East/East Indian, and American Indian/Alaska native.

## 
##    Black    White Hispanic    Other 
##    87540    34142    19587     1804

FIO distribution by age

A review of the ages listed in the FIO reports indicate that a number of these ages may either be in error or a value that is a place holder for a missing value indicator. For example, 3,978 FIO reports give an ages that are either zero or -1, and a total of 32 that are older than 100.

It may be more reasonable to look at a narrower age range, specifically from 12 to 65. There were 144,841 FIO records that fell into that age range, 96.9% of the total records in this study. Below are the top 10 ages and a graph illustrating the distribution of ages.

##    Age Reports
## 1   22    9181
## 2   21    8605
## 3   23    8534
## 4   20    8306
## 5   24    7486
## 6   19    7365
## 7   25    6836
## 8   26    5945
## 9   18    5943
## 10  27    4995

FIO distribution by age and race

The age and race information can be combined to enhance the previous graphic by including information about the distribution of race categories for each year.

Testing for racial disparities

The 2015 Fagan and Braga study of FIO reports from 2007-2010 was able to test for a variety of disparities with their database that was simply not possible in this study. For example, the earlier study was able to determine that 72,619 unique individuals who were associated with the 204,739 FIO encounters that they studied. As a result, they were also able to determine that a 5% of these individuals were associated with more than 40% of the reports. Fagan and Braga also had information on the race of officers and were able to make comparisons between the races of the officers and of the subjects of the FIO encounters.

The data released by the BPD in January 2016 lacked key information that made it impossible to replicate many of the comparisons from the Fagan and Braga study. Specifically the raw data released January 2016 that covered the period from January 2011 to April 2015 did not have the following information:

Due to these limitations, it was not possible to answer the following key questions about the latest BPD FIO data:

Due to the limitations of the data, and the focus on looking for racial disparities between BPD officer groups defined by their performance, the goal of this analysis was to look for differences in the racial makeup of the FIO reports of two groups of BPD officers, the ‘Top quintile’ group (the 20% who were responsible the highest number of FIO reports), and the ‘Other BPD’ group (the 80% of the officers who were responsible for the remaining FIO reports).

Summary of FIO subjects

  • Total FIO reports - 149,545

  • FIO reports with a racial identifier - 143,073 (95.67%)
    • Black - 87,540 (61.2%)
    • White - 34,142 (23.9%)
    • Hispanic - 19,587 (13.7%)
    • Other - 1,804 (1.3%)

Summary of BPD officer performance

  • Total BPD officers - 1,785

  • ‘Top quintile’ group
    • Number of officers - 357 (20%)
    • Number of FIO reports - 121,582 (81.3%)
  • ‘Other BPD’ group
    • Number of officers - 1,428 (80%)
    • Number of FIO reports - 27,963 (18.7%)

FIO racial splits by BPD officer groups

If there are no racial disparities in the two officers groups, then there should be no significant differences between the percentage of the four racial groups (Black, White, Hispanic, and Other) in the FIO report populations of the two officer groups, and the percentages for the population as a whole.

A chi-square test was run on the data that describes the racial breakdown of FIO reports of the ‘Top quintile’ officers compared with the other ‘Other BPD’ officers. Note that only those FIO reports that included a racial category were included in this test.

There were a total of eight categories corresponding to the two officer groups and the four racial groups (Black, White, Hispanic, and Other). If there was no significant differences between the actual and the expected values for each there would be little difference in the racial breakdown of their reports. However, this was not the case, as the chi-square value was too high to accept the null hypothesis that there is no evidence of racial disparities in between the two officer groups with respect to the observed distribution of FIO reports by race.

FIO breakdown by officer group and race
##          Top_quintile Other_BPD
## Black           74165     13375
## White           25176      8966
## Hispanic        15662      3925
## Other            1262       542
Chi-square test results

The chi-square test with 3 degrees of freedom produced a value of 2130.2. Given the number of degrees of freedom, the null hypothesis of there being no significant differences in the racial breakdowns of the FIO reports for these two groups of officers would be rejected at the 0.05 level for chi-square values above 7.815, and in this case the value was well above this level.

## 
##  Pearson's Chi-squared test
## 
## data:  chival
## X-squared = 2130.2, df = 3, p-value < 2.2e-16

Visual depiction of racial disparities

Another way to show this disparity in reporting is visually. Below is a grouped bar chart that shows how far above or below expectations each officer group was with respect the the percentage of their FIO reports in each racial group.

The horizontal bar indicates a likelihood of a BPD officer group having an FIO report involving a particular racial group being equal to that of the entire population of BPD officers. The chart shows that the ‘Top quintile’ officer group were slightly more likely to have black suspects in their FIO reports compared to other racial groups. ‘Other BPD’ officers were somewhat less likely to have black suspects in their FIO reports, but more likely than the ‘Top quintile’ officer group to have FIO reports featuring suspects from other racial groups.

Discussion

The FIO encounter data provided by the Boston Police Department (BPD) represents an important resource for investigating possible biases or disparities in how the FIO process is applied. This study was focused on two distinct BPD officer groups, the top 20% of officers who submitted the most FIO reports, and a group consisting of the remaining 80% of BPD officers.

The patterns of FIO submissions show that these two groups have a statistically significant differences when it comes to the likelihood that a particular racial group is subject of an FIO encounter, with the most productive officers being more likely to investigate black suspects, and the other group of officers being more likely to investigate suspects from other racial groups.

While these results clearly suggest that a racial disparity in FIO reporting exists, it does not address why these disparities exist, or to what extent that those who were subject to FIO encounters may have contributed to these disparities.

The data provided by the BPD provided a unique identifier for each officer, making it possible to determine how many times an officer submitted an FIO report. However, there were no unique identifiers for the subjects of the FIO encounters, so there was no way to associate an FIO report with a particular suspect, and no way to see if a particular subgroup of suspects had a disproportionate contribution to the population of FIO reports.

Also, because the data provided by the BPD that covered the January 2011 to April 2015 period provided less detailed information than the data provided to researchers (Fagan and Braga) who studied FIO data from 2007-2010, it was not possible to show if their earlier findings were true for the most recent data.

Resources

Redacted Boston Police Department FIO data from 2011-April 2015
https://data.cityofboston.gov/Public-Safety/Boston-Police-Department-FIO/xmmk-i78r

FIO data dictionary
http://www.airsafe.com/analyze/fio_data_dictionary.pdf

Boston Police Department overview of FIO data from 2011-April 2015
http://www.airsafe.com/analyze/fio_data-frequency.pdf

Processed FIO data used in this study
http://www.airsafe.com/analyze/fio_airsafe.csv

Fagan and Braga study of FIO data from 2007-2010
http://airsafe.com/analyze/bpd_fio_207_2011_study.pdf

Observed racial disparities in the Boston Police Department FIO program (this report)