1. Importing and cleaning data:

Details of imports, variable correction, cleaning and editing are ommitted from the markdown file, but can be found in the github repo.

2. Merging data

Merging report
Merge Freq
0 98251
1 3
3 3168
Note:
0: calls only, 1: call-arrest only, 2: call-charge only, 3: call-arrest-charge

Comment : We are starting with a dataset of 99,699 calls. 1448 of the calls are matched to 1768 arrests. 1447 calls are matched to 4560 charges. The table above shows the breakdown of the matching results. Note that out of the arrests, only 3 don’t lead to a charge. All 3 cases concern black arrestees!

3. Some statistics


3.1. What is the racial make-up of complainants?

Complainant race
Complainant race Percentage Instance
American Indian/Native American 0.12 95
Asian 1.70 1400
Black 28.41 23349
Latin/Hispanic 34.21 28111
Middle Eastern 1.18 973
Native Hawaiian or Pacific Islander 0.07 59
TEST 0.00 3
Unknown 0.38 310
White 33.93 27881

Comment: For reference, Dallas African-American residents represent 25% of the total population


3.2. Who calls the police on whom?

Complainant and arrestee race
Complainant race Freq
American Indian arrestee
American Indian/Native American 0
Asian 0
Black 0
Latin/Hispanic 6
Middle Eastern 0
Native Hawaiian or Pacific Islander 0
White 0
Asian arrestee
American Indian/Native American 0
Asian 0
Black 0
Latin/Hispanic 1
Middle Eastern 0
Native Hawaiian or Pacific Islander 0
White 8
Black arrestee
American Indian/Native American 0
Asian 13
Black 371
Latin/Hispanic 223
Middle Eastern 2
Native Hawaiian or Pacific Islander 0
White 196
Hispanic arrestee
American Indian/Native American 0
Asian 7
Black 58
Latin/Hispanic 479
Middle Eastern 12
Native Hawaiian or Pacific Islander 0
White 200
Middle Eastern arrestee
American Indian/Native American 0
Asian 0
Black 2
Latin/Hispanic 63
Middle Eastern 0
Native Hawaiian or Pacific Islander 0
White 0
White arrestee
American Indian/Native American 0
Asian 1
Black 26
Latin/Hispanic 97
Middle Eastern 12
Native Hawaiian or Pacific Islander 0
White 197
Note:
Excluding ‘unknown race’ and ‘hawaiian and pacific islander’ arrestees

3.3. How does response time differ across racial groups?

Mean time to dispatch by complainant race
Complainant race Time to dispatch
American Indian/Native American 00:59:21
Asian 00:49:13
Black 00:53:35
Latin/Hispanic 00:54:51
Middle Eastern 00:48:09
TEST 01:02:17
Unknown 00:46:36
White 00:49:31
NA 00:42:20

3.4 When do calls happen?

Night vs. day calls
Night call Freq
No 83021
Yes 18401
Note:
Night-time is defined as period between 8pm and 5:30 am. An arbitrary split that probably needs to be adjusted for seasons.

3.5 Is outcome of the call different depending on time of day?

Call outcomes and time of day
Outcome Freq
Daytime calls
0 81007
1 3
3 2011
Nighttime calls
0 17244
1 0
3 1157

3.6 What is the split of call-time for arrestees of different races?

Call times, arrestees by race
Time Freq
American Indian arrestee
Daytime 2
Nighttime 1
Asian arrestee
Daytime 4
Nighttime 4
Black arrestee
Daytime 387
Nighttime 265
Hispanic arrestee
Daytime 340
Nighttime 194
Mid-Eastern arrestee
Daytime 3
Nighttime 3
White arrestee
Daytime 137
Nighttime 136

3.7 Where are people calling from?

In progress pending geo data on beat to tract concordance

3.8 Time of incident by arrest

#### When looking at this dimension by race, the shapes remain the same.

3.9 Multiple charges

3.10 Multiple calls

Note that no single charge is linked to multiple calls.

3.11 “Hit rate” by complainant race

Ideally here we would want to know the race of the accused at the call, but that is data we do not have. A margin we can explore is the heterogeneity in the number of charges.

3.12 Pending questions

Please share any pending questions via email or slack so that I can explore them further.

3.13 Identifying court information

Using the calls-arrest-charges dataset, I scraped the Dallas county courts for court appearance and sentencing data, and matched it to our calls data using warrant numbers. Not all warrants numbers were found in the courts data.

Initial review of sentencing data does not show large cross-race discrepancies. First stabs at using the sentencing data are summarized below.

Outcome of Issued Warrants by Racial Group
White
Non-white
No sentence Cleared Convicted No sentence Cleared Convicted
89 128 254 413 542 1213
(19%) (27%) (54%) (19%) (25%) (56%)
Note:
Based on scraped data from the Dallas county court. Total number of unique call-warrant-arrestee is 2639
Outcome of Court Sentencing by Racial Group
White
Non-white
Fine only < 1 year 1-5 years > 5years Fine only < 1 year 1-5 years > 5years
45 154 43 12 258 699 166 90
(10%) (33%) (9%) (3%) (12%) (32%) (8%) (4%)
Note:
Percentages are out of the total warrants issued to each race group