TIGER_algorithmic_audit/institutional_risk_original.R at main · aclu-national/TIGER_algorithmic_audit · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
library(tidyverse)

sample_size = 1000000

df <- data.frame(
  age_first_arrest = sample(8:76, sample_size, replace = TRUE),
  mental_health_level = sample(1:6, sample_size, replace = TRUE),
  gender = sample(0:1, sample_size, replace = TRUE),
  marijuana_conviction = sample(0:1, sample_size, replace = TRUE),
  history_revocations = sample(0:2, sample_size, replace = TRUE),
  age_release = sample(2920:33580, sample_size, replace = TRUE), # Sketchy
  employment_history = sample(0:1, sample_size, replace = TRUE),
  history_drugs_alcohol = sample(0:2, sample_size, replace = TRUE),
  felony_convictions = sample(0:2, sample_size, replace = TRUE),
  crime_type_other = sample(0:1, sample_size, replace = TRUE),
  crime_type_property = sample(0:1, sample_size, replace = TRUE),
  crime_type_sex = sample(0:1, sample_size, replace = TRUE),
  crime_type_sexual_violent = sample(0:1, sample_size, replace = TRUE),
  crime_type_violent = sample(0:1, sample_size, replace = TRUE),
  offender_class = sample(1:22, sample_size, replace = TRUE),
  time_served = sample(0:48, sample_size, replace = TRUE),
  p07_job_placement = sample(0:1, sample_size, replace = TRUE),
  prior_recidivisms = sample(0:5, sample_size, replace = TRUE)
)

institutional_risk_model1 <- function(X) {
  logit <- function(X) {
    -0.5304268 +
    0.017552*X$age_first_arrest +
    -0.0415209*X$mental_health_level +
    0.288077*X$gender +
    -0.1017762*X$marijuana_conviction +
    1.070666*X$history_revocations +
    -0.0005054*X$age_release +
    0.0000000108*X$age_release^2 +
    0.1199185*X$employment_history +
    0.0921725*X$history_drugs_alcohol +
    0.2977796*X$felony_convictions +
    0.8565472*X$crime_type_other +
    0.6621011*X$crime_type_property +
    1.616689*X$crime_type_sex +
    2.049594*X$crime_type_sexual_violent +
    1.421824*X$crime_type_violent +
    0.3183547*X$offender_class +
    0.0425726*X$time_served +
    0.3052627*X$p07_job_placement +
    1.135524*X$prior_recidivisms +
    -0.0962842*X$prior_recidivisms^2 +
    0.000000795*X$age_release*X$prior_recidivisms
  }
  risk <- exp(logit(X)) / (1 + exp(logit(X)))
  return(risk)
}


df$pred <- institutional_risk_model1(df)

df %>%
  ggplot(aes(x = pred)) +
  geom_histogram()

mean(df$pred)
median(df$pred)
sd(df$pred)