I-213 heatmaps

library(tidyverse); library(lubridate); library(here); library(skimr); library(yaml); library(gplots)
inputfile <- here::here("write/input/uw-chr-i213-public.csv.gz")

i213 <- read_delim(
    inputfile, delim = "|", 
    col_types = cols(
        .default                     = col_character(),
        source                       = col_factor(),
        sex                          = col_factor(),
        cmplxn                       = col_character(),
        country_of_citizenship       = col_factor(),
        year                         = col_double(),
        month                        = col_double(),
        day                          = col_double(),
        hour                         = col_double(),
        minute                       = col_double(),
        age                          = col_double(),
        accompanied_juvenile_flag    = col_double(),
        unaccompanied_juvenile_flag  = col_double(),
        custody_redetermination_flag = col_double()
))

priority_counties <- read_yaml('../../shared/hand/priority_counties.yaml')

i213 <- i213 %>%
    filter(!is.na(source))

i213 <- i213 %>%
    mutate_at(vars(starts_with('mentions_')), as.logical)

Fields starting with “mentions_” are result of simple str_detect() function for keywords in I-213 “Narrative” fields. Search terms are as follows:

mentions_airport: '[Aa]irport'
mentions_anonymous_tip: '[Aa]nonymous tip|concerned citizen'
mentions_border_patrol: '[Bb]order [Pp]atrol'
mentions_bus: '([Bb]us |[Bb]us\.|[Bb]us,)'
mentions_corrections: '[Cc]orrections'
mentions_courthouse: '([Cc]ourthouse|[Cc]ourt [Hh]ouse)'
mentions_database: '[Dd]atabase'
mentions_detainer: '[Dd]etainer|[I1]-247|[I1]-2[0Oo][0Oo]'
mentions_family_unit: '[Ff]amily [Uu]nit'
mentions_greyhound: '([Gg]reyhound|[Gg]rey [Hh]ound)'
mentions_hsi: 'HSI|[Hh]omeland [Ss]ecurity [Ii]nvestigation'
mentions_jail: '[Jj]ail'
mentions_juvenile: '[Jj]uvenile'
mentions_police: '[Pp]olice'
mentions_prison: '[Pp]rison'
mentions_probation_office: '[Pp]robation [Oo]ffice'
mentions_secure_comm: '[Ss]ecure [Cc]ommunities|Immigration Alien Response|(IAR)'
mentions_sheriff: '[Ss]heriff'
mentions_state_patrol: '[Ss]tate [Pp]atrol'
mentions_surveillance: 'surveillance'
mentions_task_force: '[Tt]ask [Ff]orce'
mentions_traffic: 'traffic'

In the below we compare these with method_location_apprehension values in order to help gather context clues for decoding “Method of Location/Apprehension” codes.

skimr::skim(i213, 
  starts_with("mentions_")
  )

Data summary
Name	i213
Number of rows	4052
Number of columns	62
_______________________
Column type frequency:
logical	23
________________________
Group variables	None

Variable type: logical

skim_variable	complete_rate	mean	count
mentions_airport	1	0.10	FAL: 3662, TRU: 390
mentions_anonymous_tip	1	0.01	FAL: 4021, TRU: 31
mentions_border_patrol	1	0.36	FAL: 2591, TRU: 1461
mentions_bus	1	0.03	FAL: 3922, TRU: 130
mentions_corrections	1	0.13	FAL: 3540, TRU: 512
mentions_courthouse	1	0.03	FAL: 3948, TRU: 104
mentions_database	1	0.30	FAL: 2852, TRU: 1200
mentions_detainer	1	0.32	FAL: 2757, TRU: 1295
mentions_family_unit	1	0.01	FAL: 4008, TRU: 44
mentions_greyhound	1	0.01	FAL: 4014, TRU: 38
mentions_hsi	1	0.04	FAL: 3880, TRU: 172
mentions_jail	1	0.47	FAL: 2156, TRU: 1896
mentions_juvenile	1	0.04	FAL: 3910, TRU: 142
mentions_license_plate	1	0.04	FAL: 3872, TRU: 180
mentions_police	1	0.18	FAL: 3304, TRU: 748
mentions_prison	1	0.19	FAL: 3268, TRU: 784
mentions_probation_office	1	0.00	FAL: 4033, TRU: 19
mentions_secure_comm	1	0.05	FAL: 3868, TRU: 184
mentions_sheriff	1	0.11	FAL: 3614, TRU: 438
mentions_state_patrol	1	0.02	FAL: 3978, TRU: 74
mentions_surveillance	1	0.11	FAL: 3606, TRU: 446
mentions_task_force	1	0.01	FAL: 4002, TRU: 50
mentions_traffic	1	0.05	FAL: 3866, TRU: 186

i213 <- i213 %>% 
  mutate(method_loc_app_clean = case_when(
         str_detect(method_location_apprehension, "PB") ~ "PB",
         str_detect(method_location_apprehension, "CFD") ~ "CFD",
         str_detect(method_location_apprehension, "CST") ~ "CST",
         str_detect(method_location_apprehension, "CLC") ~ "CLC",
         str_detect(method_location_apprehension, "NCA") ~ "NCA",
         str_detect(method_location_apprehension, "LEA") ~ "LEA",
         str_detect(method_location_apprehension, "OA") ~ "OA",
         str_detect(method_location_apprehension, "OTF") ~ "OTF",
         str_detect(method_location_apprehension, "TCB") ~ "TCB",
         str_detect(method_location_apprehension, "ISP") ~ "ISP",
         str_detect(method_location_apprehension, "L") ~ "L",
         str_detect(method_location_apprehension, "O|0") ~ "O",
         is.na(method_location_apprehension) ~ NA_character_,
         TRUE ~ "OTHER"))

First we examine correlations of keyword mentions in I-213 narratives. Some strong positive correlations are trivial: e.g. “Greyhound” and “bus”; “prison” and “corrections”. Others may be interesting: i.e. positive correlation bewteen “database” and “jail” but negative correlation between “database” and “prison”.

data <- i213 %>%
  select(starts_with('mentions')) 

corrplot::corrplot(cor(data))

Comparing simple standardization of method_location_apprehension with mentions_* columns. Note similarity of “O”, “LEA”, “OTF” categories per clustering. Keywords largely seem appropriate in relation to proposed “Method of Location/Apprehension” values as discussed in https://uwchr.github.io/i-213-analysis/.

mentions_method <- i213 %>%
  select(method_loc_app_clean, starts_with('mentions')) %>% 
  group_by(method_loc_app_clean) %>% 
  summarize_all(sum)

m <- mentions_method %>% 
  select(starts_with('mentions')) %>%
  t %>% 
  as.matrix()

colnames(m) <- as.character(mentions_method$method_loc_app_clean)

heatmap.2(m,
          scale = 'row',
          density.info="none",  # turns off density plot inside color legend
          trace="none",         # turns off trace lines inside the heat map
          main="method_location_apprehension\nin search string mentions_*",
          margins =c(12,12) )

Comparison after grouping “LEA/OTF/O”. Note “CLC” similartiy to NA values for method_location_apprehension, which makes sense given “CLC” is most common value after simple standardization.

i213 <- i213 %>% 
  mutate(method_loc_app_clean = case_when(
         str_detect(method_location_apprehension, "PB") ~ "PB",
         str_detect(method_location_apprehension, "CFD") ~ "CFD",
         str_detect(method_location_apprehension, "CST") ~ "CST",
         str_detect(method_location_apprehension, "CLC") ~ "CLC",
         str_detect(method_location_apprehension, "NCA") ~ "NCA",
         str_detect(method_location_apprehension, "LEA") ~ "LEA/OTF/O",
         str_detect(method_location_apprehension, "OA") ~ "OA",
         str_detect(method_location_apprehension, "OTF") ~ "LEA/OTF/O",
         str_detect(method_location_apprehension, "TCB") ~ "TCB",
         str_detect(method_location_apprehension, "ISP") ~ "ISP",
         str_detect(method_location_apprehension, "L") ~ "L",
         str_detect(method_location_apprehension, "O|0") ~ "LEA/OTF/O",
         is.na(method_location_apprehension) ~ NA_character_,
         TRUE ~ "OTHER"))

mentions_method <- i213 %>%
  select(method_loc_app_clean, starts_with('mentions')) %>% 
  group_by(method_loc_app_clean) %>% 
  summarize_all(sum)

m <- mentions_method %>% 
  select(starts_with('mentions')) %>%
  t %>% 
  as.matrix()

colnames(m) <- as.character(mentions_method$method_loc_app_clean)

heatmap.2(m,
          scale = 'row',
          density.info="none",  # turns off density plot inside color legend
          trace="none",         # turns off trace lines inside the heat map
          main="method_location_apprehension\nin search string mentions_*",
          margins =c(12,12) )

heatmap.2(m,
          scale = 'column',
          density.info="none",  # turns off density plot inside color legend
          trace="none",         # turns off trace lines inside the heat map
          main="search string mentions_* in\nmethod_location_apprehension",
          margins =c(12,12) )

I-213 heatmaps

Phil Neff

16 April, 2021