This data analysis is part of my senior thesis, presented to the Department of Mathematics at Bates College, in partial fulfillment of the requirements for the Degree of Bachelor of Arts. Presented below is my abstract:

A mindset shift in the 1970s to more “tough-on-crime” politics led to harsher sentencing policies across the United States. As a result, in 1976, Maine became the first state out of 16 to abolish parole, which is the conditional release of an inmate before they complete their prison sentence. In the last 5 years, some states began to doubt their decisions on the effectiveness of abolishing parole. In 2021, a bill to reestablish parole in Maine was introduced and as a result of this bill a 13 member commission to study restoring parole in Maine was created. This project is inspired by the ongoing conversation about parole in Maine, and the fact that inmates in Maine have very few opportunities for early release and rehabilitation. In this work I conduct a data analysis to model the potential impact of reinstating parole in Maine. There is a particular focus on changes in inmate sentence lengths as well as the economic impact if Maine were to implement the parole systems of Wyoming and Vermont.

Data Setup

#Installing and loading necessary packages

# install.packages("skimr", "haven", "tidyverse", "dplyr", "psych", "DT", "foreign", "gridExtra", "knitr", "kableExtra")
library(haven)
library(tidyverse)
library(dplyr)
library(psych)
library(DT)
library(foreign)
library(skimr)
library(gridExtra)
library(kableExtra)

#loading dataset
load("C:/Users/smangane/OneDrive - bates.edu/Bates Parole Thesis/BatesParoleThesis/DS0001/38492-0001-Data.rda") 

#filtering for Maine
 ME_term_records <- da38492.0001 %>%
  filter(STATE == "(23) Maine")

Basic Summary Statistics/EDA - (ME Term Records)

The skim() function in R gives us a broad overview of our dataframe. The first and second dataframes that they print are the factor and numeric variables in our dataframe. It shows the amount of missing, and unique (or distinct) values as well as the highest values for each factored variable. Based on this, I order some of the factored variables. For our numeric variables, we can see missing values as well as the mean, median, and distribution of values.

#skimming the data set and renaming columns
skim(ME_term_records) %>%  
  kbl(caption = "Dataset Overview", 
      col.names = c("Variable Type", "Variable (Column Name)", "Missing Rows", "Completion Rate", "Ordered?", "Unique Rows", "Most Common Values", "Mean", "Standard Deviation", "Minimum", "25th Quartile", "Median", "75th Quartile", "Maximum", "Histogram")) %>%
  kable_minimal() %>% 
  scroll_box(height = "500px")#creating table

Dataset Overview
Variable Type	Variable (Column Name)	Missing Rows	Completion Rate	Ordered?	Unique Rows	Most Common Values	Mean	Standard Deviation	Minimum	25th Quartile	Median	75th Quartile	Maximum	Histogram
factor	ABT_INMATE_ID	0	1.0000000	FALSE	9694	A23: 7, A23: 6, A23: 6, A23: 5	NA	NA	NA	NA	NA	NA	NA	NA
factor	SEX	0	1.0000000	FALSE	2	(1): 11093, (2): 1516	NA	NA	NA	NA	NA	NA	NA	NA
factor	ADMTYPE	69	0.9945277	FALSE	3	(1): 12077, (2): 397, (3): 66	NA	NA	NA	NA	NA	NA	NA	NA
factor	OFFGENERAL	135	0.9892934	FALSE	5	(1): 4825, (2): 3610, (3): 2668, (4): 1280	NA	NA	NA	NA	NA	NA	NA	NA
factor	EDUCATION	0	1.0000000	FALSE	1	(9): 12609	NA	NA	NA	NA	NA	NA	NA	NA
factor	SENTLGTH	164	0.9869934	FALSE	7	(2): 4320, (0): 3098, (1): 2999, (3): 1231	NA	NA	NA	NA	NA	NA	NA	NA
factor	OFFDETAIL	135	0.9892934	FALSE	13	(12: 2668, (05: 2197, (07: 1664, (08: 1345	NA	NA	NA	NA	NA	NA	NA	NA
factor	RACE	243	0.9807281	FALSE	3	(1): 10619, (2): 1154, (4): 593, (3): 0	NA	NA	NA	NA	NA	NA	NA	NA
factor	AGEADMIT	0	1.0000000	FALSE	5	(2): 5297, (3): 3147, (1): 1964, (4): 1602	NA	NA	NA	NA	NA	NA	NA	NA
factor	AGERELEASE	1726	0.8631136	FALSE	5	(2): 4590, (3): 2982, (4): 1536, (1): 1113	NA	NA	NA	NA	NA	NA	NA	NA
factor	TIMESRVD	1726	0.8631136	FALSE	5	(0): 4914, (1): 3332, (2): 2077, (3): 417	NA	NA	NA	NA	NA	NA	NA	NA
factor	RELTYPE	2670	0.7882465	FALSE	3	(1): 5043, (2): 4770, (3): 126	NA	NA	NA	NA	NA	NA	NA	NA
factor	STATE	0	1.0000000	FALSE	1	(23: 12609, (01: 0, (02: 0, (04: 0	NA	NA	NA	NA	NA	NA	NA	NA
numeric	ADMITYR	0	1.0000000	NA	NA	NA	2014.650	4.165646	1970	2013	2015	2018	2020	▁▁▁▁▇
numeric	RELEASEYR	1726	0.8631136	NA	NA	NA	2016.167	2.538238	2012	2014	2016	2018	2020	▆▇▃▇▇
numeric	MAND_PRISREL_YEAR	12609	0.0000000	NA	NA	NA	NaN	NA	NA	NA	NA	NA	NA
numeric	PROJ_PRISREL_YEAR	12609	0.0000000	NA	NA	NA	NaN	NA	NA	NA	NA	NA	NA
numeric	PARELIG_YEAR	12609	0.0000000	NA	NA	NA	NaN	NA	NA	NA	NA	NA	NA

#factoring the ages and offenses renaming columns 
ME_term_records <- ME_term_records  %>% 
    mutate_at(vars(AGEADMIT, AGERELEASE),
              factor,
              levels = c("(1) 18-24 years", "(2) 25-34 years",  "(3) 35-44 years",  "(4) 45-54 years", "(5) 55+ years"),
              ordered = TRUE) %>% 
  mutate_at(vars(OFFGENERAL),
            factor,
            levels = c("(1) Violent", "(2) Property",   "(3) Drugs",    "(4) Public order", "(5) Other/unspecified"),
            ordered = TRUE)

Frequency of Admittance in Maine (1991-2020) - looking at reoffenders

To see each individual case, I grouped the data frame by ID number, since a person could be released more than once and as a result have more than one record (only available in DS0001). There are 9,694 distinct observations of prisoners in the ME_term_records. Here I create a histogram of the distribution of admittance in Maine, to get an estimate of reoffenders in Maine. About 41% of all inmates are repeat offenders.

#creating a variable that states inmates as repeat offenders or not
me_analysis <- ME_term_records %>%
  count(ABT_INMATE_ID, sort = TRUE) %>%
  mutate(Repeat_Offender = case_when(
    n > 1 ~ "Yes",
    TRUE ~ "No"
  )) %>% 
  full_join(ME_term_records, by = "ABT_INMATE_ID") %>% 
  rename("In_Datasetx" = "n")

#creating a histogram
 me_analysis %>%
  ggplot(aes(In_Datasetx)) + 
  geom_bar() +
     labs(title = "Frequency of Admittance in Maine (1991-2020)", x ="Number of Times an Inmate was Admitted", y = "Frquency of Inmates Admitted")

Frequency of Parole Violator Admissions in Maine

Below I create a histogram of the distribution of parole violator admittance (those who were admitted because they violated parole) in Maine. Of those 9,694 observations, 238 have been admitted because they violated their parole. Although it can happen more than once, most inmates have returned to prison only once for violating parole.

#code for percentages of distribution of admissions type
# me_analysis %>% 
#   distinct(ABT_INMATE_ID, .keep_all = TRUE) %>% 
#   count(ADMTYPE) %>% 
#   mutate(Percent = paste((n/9694)*100,seperate = "%"))

#creating graph
 me_analysis %>%
   filter(ADMTYPE == "(2) Parole return/revocation") %>%
   count(ABT_INMATE_ID, sort = TRUE) %>%
   ggplot(aes(n)) + 
  geom_bar() +
     labs(title = "Frequency of Parole Violator Admissions in Maine (1991-2020)", x ="Number of Times an Inmate was Admitted for Violating Parole", y = "Frquency of Inmates Admitted")

Distribution of Time Served in Maine

These are the percentages of the distributions of time served. More than 50% of the data set are serving less than 2 years in prison.

#Create table of percentages of time served    
me_analysis %>%
  distinct(ABT_INMATE_ID, .keep_all = TRUE) %>%
  count(TIMESRVD) %>% 
  mutate(Percent = paste(round((n/9694)*100,digits = 2),seperate = "%"))  %>% 
  rename("Time Served" = "TIMESRVD",
         "Number of Inmates (Observations)" = "n",
         "% of Inmates in Dataset" = "Percent") %>% 
  kbl(caption = "Percentages of Time Served") %>%
  kable_minimal()

Percentages of Time Served
Time Served	Number of Inmates (Observations)	% of Inmates in Dataset
< 1 year	3735	38.53 %
1-1.9 years	2506	25.85 %
2-4.9 years	1596	16.46 %
5-9.9 years	383	3.95 %
>=10 years	138	1.42 %
NA	1336	13.78 %

#creating an exact time served  variable
me_analysis <- me_analysis %>%
  mutate(exact_time_served = RELEASEYR - ADMITYR)

Average Sentence Length By Crime

This is the average sentence length for each offense and their percentage of the data set population.

#table of percentages of dataset population by offense
ro <- me_analysis %>% 
  distinct(ABT_INMATE_ID, .keep_all = TRUE) %>%
  count(OFFDETAIL) %>% 
  mutate(Perc_of_prisonpop = paste(round((n/9694)*100,digits = 2),seperate = "%")) %>%
  select("OFFDETAIL", "Perc_of_prisonpop")

#combing dataframe above with table of average sentence lengths 
avg_senlgth <- me_analysis %>%
  distinct(ABT_INMATE_ID, .keep_all = TRUE) %>%
  group_by(OFFDETAIL) %>%
  summarize(Avg_sentlgth = mean(exact_time_served, na.rm = TRUE)) %>% 
  full_join(ro, by = "OFFDETAIL") %>% 
  mutate(Avg_sentlgth = round(Avg_sentlgth, digits = 1))  %>% 
  rename("Offense Categorization" = "OFFDETAIL",
         "Avg. Sent. Length" = "Avg_sentlgth",
         "% of Prison Pop." = "Perc_of_prisonpop")

#creating table
avg_senlgth %>% 
  kbl(caption = "Average Sentence Length for Each Offense") %>%
  kable_minimal()

Average Sentence Length for Each Offense
Offense Categorization	Avg. Sent. Length	% of Prison Pop.
Murder (including non-negligent manslaughter)	14.8	3.08 %
Negligent manslaughter	5.2	1.72 %
Rape/sexual assault	3.5	9.2 %
Robbery	2.9	6.56 %
Aggravated or simple assault	1.5	16.93 %
Other violent offenses	2.0	1.33 %
Burglary	1.7	12.13 %
Larceny	1.1	9.93 %
Fraud	1.3	2.49 %
Other property offenses	1.9	2.24 %
Drugs (includes possession, distribution, trafficking, other)	1.4	22.33 %
Public order	1.2	10.14 %
Other/unspecified	3.9	0.79 %
NA	0.5	1.11 %

Initial Analysis

How Certain Are Prison’s Sentence Length in Maine?

I calculate what percentage of inmates were released before, within, or after your mandatory prison release year (determined by the maximum sentence length)? Despite not knowing the release year of about 15% of inmates (due to missing release years or maximum sentence lengths), only 0.86% aren’t released earlier or at the release date they’re supposed to (maximum sentence length). Thus it seems like prison sentence length are pretty certain in Maine.

#creating 4 variables: earliest and latest mand. prison release, released early variables (in character and numeric form)
me_analysis <- me_analysis %>% 
  mutate(Earliest_MAND_PRISREL_YEAR = case_when( #max sent.length + admit yr
     SENTLGTH == "(0) < 1 year" ~ ADMITYR,
     SENTLGTH == "(1) 1-1.9 years" ~ ADMITYR + 1, 
     SENTLGTH == "(2) 2-4.9 years" ~ ADMITYR + 2, 
     SENTLGTH == "(3) 5-9.9 years" ~ ADMITYR + 5, 
     SENTLGTH == "(4) 10-24.9 years" ~ ADMITYR + 10,
    SENTLGTH == "(5) >=25 years" ~ ADMITYR + 25,
   ), Latest_MAND_PRISREL_YEAR = case_when(
     SENTLGTH == "(0) < 1 year" ~ ADMITYR + 1,
     SENTLGTH == "(1) 1-1.9 years" ~ ADMITYR + 2, 
     SENTLGTH == "(2) 2-4.9 years" ~ ADMITYR + 5,
     SENTLGTH == "(3) 5-9.9 years" ~ ADMITYR + 10, 
     SENTLGTH == "(4) 10-24.9 years" ~ ADMITYR + 25,
    SENTLGTH == "(5) >=25 years" ~ NA_real_,
   )
   ) %>% 
  mutate(Released_early_dbl = case_when(
    RELEASEYR < Earliest_MAND_PRISREL_YEAR ~ paste(Earliest_MAND_PRISREL_YEAR - RELEASEYR, " year(s) earlier"),
    RELEASEYR == Earliest_MAND_PRISREL_YEAR ~ "0",
    RELEASEYR > Earliest_MAND_PRISREL_YEAR & RELEASEYR < Latest_MAND_PRISREL_YEAR ~ paste(Latest_MAND_PRISREL_YEAR - RELEASEYR, " year(s) within mand prison release"),
    RELEASEYR > Earliest_MAND_PRISREL_YEAR & RELEASEYR == Latest_MAND_PRISREL_YEAR ~ "0",
    RELEASEYR > Earliest_MAND_PRISREL_YEAR & RELEASEYR > Latest_MAND_PRISREL_YEAR ~ paste(RELEASEYR - Latest_MAND_PRISREL_YEAR, " years later")
  )) %>%
    mutate(Release_early_chr = case_when(
      RELEASEYR < Earliest_MAND_PRISREL_YEAR ~ "Yes",
      RELEASEYR == Earliest_MAND_PRISREL_YEAR ~ "Same Time",
      RELEASEYR > Earliest_MAND_PRISREL_YEAR & RELEASEYR < Latest_MAND_PRISREL_YEAR ~ "Same Time",
    RELEASEYR > Earliest_MAND_PRISREL_YEAR & RELEASEYR == Latest_MAND_PRISREL_YEAR ~ "Same Time",
    RELEASEYR > Earliest_MAND_PRISREL_YEAR & RELEASEYR > Latest_MAND_PRISREL_YEAR ~ "No"))
      

#table to see when people are released
me_analysis %>%
  count(Release_early_chr) %>% 
  mutate(Percent = paste(round((n/12609)*100, digits = 2),seperate = "%")) %>% 
  rename("Released Before Mandatory Release Year" = "Release_early_chr",
         "Number of Inmates (Observations)" = "n",
         "% of Inmates in Dataset" = "Percent") %>% 
  kbl(caption = "Percentages of When Inmates Are Released") %>%
  kable_minimal()

Percentages of When Inmates Are Released
Released Before Mandatory Release Year	Number of Inmates (Observations)	% of Inmates in Dataset
No	108	0.86 %
Same Time	8270	65.59 %
Yes	2349	18.63 %
NA	1882	14.93 %

Truth in Sentencing?

Below I calculate the percentage of inmates serving at least 85% of their sentence. This isn’t possible for prisoners who serve less than 1 year and more than 25 years, since the units are in years and I can’t determine what their maximum sentence length is. Nonetheless for ht einmates we can calculate it for, a majority of inmates in Maine serve at least 85% of their prison sentence lengths, which lines up with truth-in-sentencing laws passed.

#creating variable which determines the 85% mark of a prisoner's sentence
me_analysis <- me_analysis %>% 
  mutate(etisyear = case_when(
     SENTLGTH == "(1) 1-1.9 years" ~ 1 * .85, 
     SENTLGTH == "(2) 2-4.9 years" ~ 2 * .85, 
     SENTLGTH == "(3) 5-9.9 years" ~ 5 * .85, 
     SENTLGTH == "(4) 10-24.9 years" ~ 10 * .85,
    SENTLGTH == "(5) >=25 years" ~ 25 * .85,
   ),ltisyear = case_when(
    SENTLGTH == "(0) < 1 year" ~ 1 *.85,
     SENTLGTH == "(1) 1-1.9 years" ~ 2 * .85, 
     SENTLGTH == "(2) 2-4.9 years" ~ 5 * .85, 
     SENTLGTH == "(3) 5-9.9 years" ~ 10 * .85, 
     SENTLGTH == "(4) 10-24.9 years" ~ 25 * .85
   ), servedperc = case_when( #variables above compared to their exact time served
     SENTLGTH == "(0) < 1 year" ~ "Served less than a year",
     SENTLGTH == "(5) >=25 years" ~ "Served more than 25 years",
     exact_time_served >= etisyear & exact_time_served <= ltisyear ~ "85% or more",
     exact_time_served >= ltisyear ~ "85% or more",
     exact_time_served < etisyear ~ "Less than 85%"
   ))

#creating table of percentages of inmates of above
me_analysis %>% 
  count(servedperc) %>% 
  mutate(Percent = paste(round((n/12609)*100, digits = 2),seperate = "%")) %>% 
  rename("Percentage Served" = "servedperc",
         "Number of Inmates (Observations)" = "n",
         "% of Inmates in Dataset" = "Percent") %>% 
  kbl(caption = "Percentages of Sentence Served") %>% 
  kable_minimal()

Percentages of Sentence Served
Percentage Served	Number of Inmates (Observations)	% of Inmates in Dataset
85% or more	5413	42.93 %
Less than 85%	2283	18.11 %
Served less than a year	3098	24.57 %
Served more than 25 years	269	2.13 %
NA	1546	12.26 %

#for those serving less than 85%, shows what perc they're actually serving
# me_analysis %>%
#   filter(servedperc == "Less than 85%") %>% 
#   mutate(servedpercent =  case_when(
#      SENTLGTH == "(1) 1-1.9 years" ~ exact_time_served/1, 
#      SENTLGTH == "(2) 2-4.9 years" ~ exact_time_served/2, 
#      SENTLGTH == "(3) 5-9.9 years" ~ exact_time_served/5, 
#      SENTLGTH == "(4) 10-24.9 years" ~ exact_time_served/10,
#     SENTLGTH == "(5) >=25 years" ~ exact_time_served/25,
#    )) %>% 
#   count(servedpercent)

Missing Values

Some of the NA categories may be due missing RELEASEYR values. As we can see in the graph below, this may be due to the fact that most inmates are still in custody and haven’t been released yet.

#created a minimum sentence length, and then the earliest projected release yr and the latest projected release yr

me_analysis3 <- me_analysis %>%
  filter(is.na(RELEASEYR)) %>%
  mutate(exact_maxsentlgth = case_when(
           SENTLGTH == "(0) < 1 year" ~ 1,
           SENTLGTH == "(1) 1-1.9 years" ~ 1.9,
           SENTLGTH == "(2) 2-4.9 years" ~ 4.9,
           SENTLGTH == "(3) 5-9.9 years" ~ 9.9,
           SENTLGTH == "(4) 10-24.9 years" ~ 24.9),
         min_sentlgth = case_when(
           SENTLGTH == "(1) 1-1.9 years" ~ 1, #365 * 1
           SENTLGTH == "(2) 2-4.9 years" ~ 2, #365 * 2
           SENTLGTH == "(3) 5-9.9 years" ~ 5, #365 * 5
           SENTLGTH == "(4) 10-24.9 years" ~ 10, #365 * 10
           SENTLGTH == "(5) >=25 years" ~ 25,  #365*25, murder to be exact
         ),
         min_sentlgth = case_when(
           SENTLGTH == "(6) Life, LWOP, Life plus additional years, Death" ~ "NA - Life",
           TRUE ~ as.character(min_sentlgth)
         ),
         eproj_releaseyr = case_when(
           min_sentlgth == "NA - Life" ~ NA_real_,
           TRUE ~ ADMITYR + as.numeric(min_sentlgth)
         ),
         lproj_releaseyr = case_when(
           min_sentlgth == "NA - Life" ~ NA_real_,
           TRUE ~ floor(ADMITYR + as.numeric(exact_maxsentlgth))
         )) %>% 
  select("ADMITYR", "SENTLGTH", "eproj_releaseyr", "lproj_releaseyr") 

me_analysis3 %>%
  ggplot(aes(x = lproj_releaseyr)) +
  geom_bar() +
  geom_vline(xintercept = 2020, color = "blue") + labs(title = "Distribution of Projected Release Years", subtitle = "(For inmates with a `RELEASEYR` of `NA`)", x ="Projected Release Year", y = "Count / Frequency")

Only 4.78% of inmates in the dataset have a projected release year before or in 2020. 78.65% of inmates in this dataset have a projected release year after 2020 and the projected release year for 16.57% of inmates in the dataset are unknown.

me_analysis3 %>%
  count(lproj_releaseyr) %>% 
  mutate(Percent = paste(round((n/1726)*100,digits = 2),seperate = "%")) %>% 
  rename("Latest Projected Release Year" = "lproj_releaseyr",
         "Number of Inmates (Observations)" = "n") %>% 
  kbl(caption = "Percentages of Projected Release Years") %>% 
  kable_minimal()

Percentages of Projected Release Years
Latest Projected Release Year	Number of Inmates (Observations)	Percent
1991	1	0.06 %
1992	1	0.06 %
1998	1	0.06 %
2001	1	0.06 %
2003	1	0.06 %
2005	1	0.06 %
2007	2	0.12 %
2008	2	0.12 %
2009	1	0.06 %
2012	2	0.12 %
2013	3	0.17 %
2014	2	0.12 %
2016	5	0.29 %
2017	7	0.41 %
2018	7	0.41 %
2019	6	0.35 %
2020	40	2.32 %
2021	250	14.48 %
2022	117	6.78 %
2023	260	15.06 %
2024	235	13.62 %
2025	28	1.62 %
2026	60	3.48 %
2027	74	4.29 %
2028	59	3.42 %
2029	61	3.53 %
2030	1	0.06 %
2031	1	0.06 %
2032	3	0.17 %
2033	9	0.52 %
2034	12	0.7 %
2035	14	0.81 %
2036	15	0.87 %
2037	19	1.1 %
2038	14	0.81 %
2039	25	1.45 %
2040	13	0.75 %
2041	27	1.56 %
2042	34	1.97 %
2043	19	1.1 %
2044	7	0.41 %
NA	286	16.57 %

What’s the maximum percentage of a sentence that can be reduced through good time credits?

Prisoners can be released at their maximum sentence lengths or earlier due to “good time” credit deductions (earned through good conduct and program participation). Prisoners convicted of murder, and any crime listed under sexual assaults (not limited to these two!) are eligible for only five days of credit per month for good conduct and program participation. Besides that prisoners can earn up to 4 days for good conduct and an extra 5 days per month for participating in programs (that you assigned to!). Here, I calculate the maximum percent of a sentence that can be reduced if prisoners earn all their good time credits as well as the averages of their sentence lengths (excluding good time credits) for each crime.

Note that there can be way more observations because we had to use the NCRP’s broad categories which we don’t know the criteria for (keep in mind other crimes are only limited to 5 days of good time credit, but they don’t fit in the dataset’s categories). However, for prisoners convicted of sexual assault or murder (about 10.9% of our data set), the release date with all good time credits earned is about ~83.5% of the maximum sentence given by a judge. For all other prisoners (89.1% of our data set), the release date with all good time credits earned is about ~70% of the maximum sentence given by a judge.

#create dataframe for murder/rape offenses and calculate their good time credits and time served (subtracting those credits)

#Formula: (12 \* time served) \* (5 days per month) = good time credits available (in days)

df1 <- me_analysis %>%
  filter(OFFDETAIL == "(01) Murder (including non-negligent manslaughter)" | OFFDETAIL == "(03) Rape/sexual assault") %>%
  mutate(Good_time_credits = (12 * exact_time_served) * 5,
         gtc_exact_time_served = (( (365 * exact_time_served) - Good_time_credits ) / (365)))
         
#create dataframe that calculates average sent length minus good time credits by offense
avg_senlgthv <- df1 %>%
  distinct(ABT_INMATE_ID, .keep_all = TRUE) %>%
  group_by(OFFDETAIL) %>%
  summarize(avg_sentlgth_w_gtc = round(mean(gtc_exact_time_served, na.rm = TRUE), digits = 1)) 


#create dataframe for all other offenses and calculate their good time credits and time served (subtracting those credits)

#Formula: (12 * time served) * (9 days per month) = good time credits available (in days)

df2 <- me_analysis %>%
  filter(OFFDETAIL != "(01) Murder (including non-negligent manslaughter)" & OFFDETAIL != "(03) Rape/sexual assault" | is.na(OFFDETAIL)) %>%
  mutate(Good_time_credits = (12 * exact_time_served) * 9, 
         gtc_exact_time_served = (( (365 * exact_time_served) - Good_time_credits ) / (365)))  

#create dataframe that calculates average sent length minus good time credits by offense
avg_senlgthnv <- df2 %>%
  distinct(ABT_INMATE_ID, .keep_all = TRUE) %>%
  group_by(OFFDETAIL) %>%
  summarize(avg_sentlgth_w_gtc = round(mean(gtc_exact_time_served, na.rm = TRUE), digits = 1)) 


#combine the two dataframes above and calculate thier sentence reductions
gtc_table <- bind_rows(avg_senlgthv, avg_senlgthnv) %>% 
  full_join(avg_senlgth, by = c("OFFDETAIL" = "Offense Categorization")) %>% 
  mutate(perc_reduction = paste(round(((`Avg. Sent. Length` - avg_sentlgth_w_gtc)/`Avg. Sent. Length`), digits = 2) * 100, seperate = "%")) %>% 
  rename("Offense Categorization" = "OFFDETAIL",
         "Avg. Sentence Length w/ Good Time Credits" = "avg_sentlgth_w_gtc",
         "Avg. Sentence Length" = "Avg. Sent. Length",
         "Sentence Reduction %" = "perc_reduction")
 
#creating table
kbl(gtc_table, caption = "Sentence Length Comparison with Good Time Credits") %>%
  kable_minimal()

Sentence Length Comparison with Good Time Credits
Offense Categorization	Avg. Sentence Length w/ Good Time Credits	Avg. Sentence Length	% of Prison Pop.	Sentence Reduction %
Murder (including non-negligent manslaughter)	12.4	14.8	3.08 %	16 %
Rape/sexual assault	2.9	3.5	9.2 %	17 %
Negligent manslaughter	3.6	5.2	1.72 %	31 %
Robbery	2.0	2.9	6.56 %	31 %
Aggravated or simple assault	1.1	1.5	16.93 %	27 %
Other violent offenses	1.6	2.0	1.33 %	20 %
Burglary	1.2	1.7	12.13 %	29 %
Larceny	0.8	1.1	9.93 %	27 %
Fraud	0.9	1.3	2.49 %	31 %
Other property offenses	1.3	1.9	2.24 %	32 %
Drugs (includes possession, distribution, trafficking, other)	1.0	1.4	22.33 %	29 %
Public order	0.8	1.2	10.14 %	33 %
Other/unspecified	2.7	3.9	0.79 %	31 %
NA	0.4	0.5	1.11 %	20 %

Wyoming and Vermont Parole Eligibility:

In Wyoming and Vermont, an inmate must have served his or her minimum term, less any good time (15 days a month in Wyoming) and special good time earned (only in Wyoming, but won’t consider for below) to be eligible. LWOP or life sentences aren’t eligible for parole.

Wyoming

Based on Wyoming’s parole model when we compare the time served with and without parole (given they’re released on the parole eligibility date), there sentence lengths are dramatically reduced, compared to the time served with good time credits. Although Wyoming’s good time credits are very generous, they don’t have a huge impact because it’s based on the time served until parole, which is much shorter. Compared to good time credits, being released on parole can serve as a better incentive for prisoners If those with the longest sentences no longer pose a threat to public safety and are released on parole, they can save Maine the most.

me_analysis1 <- me_analysis %>% 
  filter(SENTLGTH != "(6) Life, LWOP, Life plus additional years, Death" | is.na(SENTLGTH)) %>% #filter out LWOP sentences
  mutate(PARELIG_YEAR = case_when( #parole eligibility year (minimum sentence defined based on smallest value in range of time served)
           TIMESRVD == "(0) < 1 year" ~ ADMITYR, #minimum term = 0.5
           TIMESRVD == "(1) 1-1.9 years" ~ floor(1 + ADMITYR), #365 * 1
           TIMESRVD == "(2) 2-4.9 years" ~ floor(2 + ADMITYR), #365 * 2
           TIMESRVD == "(3) 5-9.9 years" ~ floor(5 + ADMITYR), #365 * 5
           TIMESRVD == "(4) >=10 years" ~ floor(10 + ADMITYR)  #365*10
           ),
         time_until_parole_wy = case_when( #sentence length until parole eligibility year 
           PARELIG_YEAR == ADMITYR ~ 0.5,
           TRUE ~ PARELIG_YEAR - ADMITYR),
         Good_time_credits_wy = (12 * time_until_parole_wy) * 15,
         gtc_exact_time_served_wy = (( (365 * time_until_parole_wy) - Good_time_credits_wy ) / (365)), #time served w/ good time credits
         PARELIG_YEAR = PARELIG_YEAR - (Good_time_credits_wy/365)) #adjusting for any good time credit they earned 

#find the avg. time until parole for each offense and combine with avg. sentence length df for comparison
 avg_time_until_parole_wy <- me_analysis1 %>%
  distinct(ABT_INMATE_ID, .keep_all = TRUE) %>%
  group_by(OFFDETAIL) %>%
  mutate(time_until_parole_gtc = PARELIG_YEAR - ADMITYR) %>%  #sentence leng. until parole w/ gtc
  summarize(avg_time_until_parole = round(mean(time_until_parole_gtc, na.rm = TRUE), digits = 1)) %>% 
  full_join(avg_senlgth, by = c("OFFDETAIL" = "Offense Categorization")) 


wyparole_table <- avg_time_until_parole_wy %>%  
  mutate(incarceration_savings_pp = round((`Avg. Sent. Length` - avg_time_until_parole) * (78732.15/365), digits = 2), #calculated the savings if inmates were released on parole eligibility year (avg. cost of incarceration per prisoner per day * sentence length not served)
         avg_perc_reduction = paste(round(((`Avg. Sent. Length` - avg_time_until_parole)/`Avg. Sent. Length`), digits = 2) * 100, seperate = "%")) %>% #calculated the avg. sent. reduction perc. for each offense
  rename("Offense Categorization" = "OFFDETAIL",
         "Avg. Sent. Length w/ Parole" = "avg_time_until_parole",
         "Average Sentence Reduction %" = "avg_perc_reduction",
         "Incarceration Savings Per Prisoner" = "incarceration_savings_pp")
 
#creating table
kbl(wyparole_table, caption = "Sentence Length Comparison with Parole in Wyoming") %>%
  kable_minimal()

Sentence Length Comparison with Parole in Wyoming
Offense Categorization	Avg. Sent. Length w/ Parole	Avg. Sent. Length	% of Prison Pop.	Incarceration Savings Per Prisoner	Average Sentence Reduction %
Murder (including non-negligent manslaughter)	3.3	14.8	3.08 %	2480.60	78 %
Negligent manslaughter	1.7	5.2	1.72 %	754.97	67 %
Rape/sexual assault	1.1	3.5	9.2 %	517.69	69 %
Robbery	0.9	2.9	6.56 %	431.41	69 %
Aggravated or simple assault	0.4	1.5	16.93 %	237.27	73 %
Other violent offenses	0.5	2.0	1.33 %	323.56	75 %
Burglary	0.4	1.7	12.13 %	280.42	76 %
Larceny	0.1	1.1	9.93 %	215.70	91 %
Fraud	0.2	1.3	2.49 %	237.27	85 %
Other property offenses	0.5	1.9	2.24 %	301.99	74 %
Drugs (includes possession, distribution, trafficking, other)	0.3	1.4	22.33 %	237.27	79 %
Public order	0.1	1.2	10.14 %	237.27	92 %
Other/unspecified	0.5	3.9	0.79 %	733.40	87 %
NA	-0.1	0.5	1.11 %	129.42	120 %

Vermont

In Vermont, if an inmate doesn’t have a minimum sentence, then they’re eligible after a year. They also wrote their good time bill in 2020 and it’s effective 01/01/2021. Since this data is up until 2020, good time wasn’t taken into consideration for the parole eligibility date calculation.

As a result of the slightly different parole criteria (being eligible for parole after a year), sentence reductions aren’t as drastic for those with shorter sentences, which encompasses most of the prisoners in Maine from 1991-2020.

#same comments as Wyoming, but for vermont (excluding good time credits)
me_analysis2 <- me_analysis %>% 
  filter(SENTLGTH != "(6) Life, LWOP, Life plus additional years, Death" | is.na(SENTLGTH)) %>%
  mutate(PARELIG_YEAR_VT = case_when(
           TIMESRVD == "(1) 1-1.9 years" | TIMESRVD == "(0) < 1 year" ~ floor(1 + ADMITYR), #365 * 1
           TIMESRVD == "(2) 2-4.9 years" ~ floor(2 + ADMITYR), #365 * 2
           TIMESRVD == "(3) 5-9.9 years" ~ floor(5 + ADMITYR), #365 * 5
           TIMESRVD == "(4) >=10 years" ~ floor(10 + ADMITYR)  #365*10
           ),
         time_until_parole_vt = case_when(
           PARELIG_YEAR_VT == ADMITYR ~ 0.2,
           TRUE ~ PARELIG_YEAR_VT - ADMITYR))
         # sent_reduction_perc_vt = case_when(
         #    PARELIG_YEAR_VT == ADMITYR ~ 1.0, #0% reduction
         #    ADMITYR == RELEASEYR ~ 2.0, #INF! 
         #    RELEASEYR == PARELIG_YEAR_VT ~ 3.0, #100% reduction
         #    TRUE ~  round(((exact_time_served - time_until_parole_vt)/exact_time_served), digits = 2) * 100)) 

 avg_time_until_parole_vt <- me_analysis2 %>%
  distinct(ABT_INMATE_ID, .keep_all = TRUE) %>%
  group_by(OFFDETAIL) %>%  
  summarize(avg_time_until_parole = round(mean(time_until_parole_vt, na.rm = TRUE), digits = 1)) %>% 
  full_join(avg_senlgth, by = c("OFFDETAIL" = "Offense Categorization")) 


 
vtparole_table <- avg_time_until_parole_vt %>% 
  # group_by(OFFDETAIL) %>% 
  # summarise(avg_perc_reduction = round(mean(sent_reduction_perc_vt, na.rm = TRUE), digits = 2)) %>% 
  # full_join(avg_time_until_parole_vt, by = "OFFDETAIL") %>% 
  mutate(incarceration_savings_pp = round((`Avg. Sent. Length` - avg_time_until_parole) * (78732.15/365), digits = 2),
         avg_perc_reduction = paste(round(((`Avg. Sent. Length` - avg_time_until_parole)/`Avg. Sent. Length`), digits = 2) * 100, seperate = "%")) %>% #calculated the avg. sent. reduction perc. for each offense
  rename("Offense Categorization" = "OFFDETAIL",
         "Avg. Sent. Length w/ Parole" = "avg_time_until_parole",
         "Average Sentence Reduction %" = "avg_perc_reduction",
         "Incarceration Savings Per Prisoner" = "incarceration_savings_pp")
 
 

#creating table
kbl(vtparole_table, caption = "Sentence Length Comparison with Parole in Vermont") %>%
  kable_minimal()

Sentence Length Comparison with Parole in Vermont
Offense Categorization	Avg. Sent. Length w/ Parole	Avg. Sent. Length	% of Prison Pop.	Incarceration Savings Per Prisoner	Average Sentence Reduction %
Murder (including non-negligent manslaughter)	6.7	14.8	3.08 %	1747.21	55 %
Negligent manslaughter	3.6	5.2	1.72 %	345.13	31 %
Rape/sexual assault	2.5	3.5	9.2 %	215.70	29 %
Robbery	2.1	2.9	6.56 %	172.56	28 %
Aggravated or simple assault	1.3	1.5	16.93 %	43.14	13 %
Other violent offenses	1.8	2.0	1.33 %	43.14	10 %
Burglary	1.4	1.7	12.13 %	64.71	18 %
Larceny	1.1	1.1	9.93 %	0.00	0 %
Fraud	1.2	1.3	2.49 %	21.57	8 %
Other property offenses	1.5	1.9	2.24 %	86.28	21 %
Drugs (includes possession, distribution, trafficking, other)	1.3	1.4	22.33 %	21.57	7 %
Public order	1.2	1.2	10.14 %	0.00	0 %
Other/unspecified	2.0	3.9	0.79 %	409.84	49 %
NA	1.0	0.5	1.11 %	-107.85	-100 %

If you don’t want to keep scrolling up ad down, below is a joint table of the sentence lengths and their reductions with Maine’s good time credit system, and Wyoming’s and Vermont’s parole system

#creating joint table of the 3 tables above for thesis poster

gtc_table %>% 
  select(1,2,5) %>% 
  rename("(ME) Avg. Sent. Length w/ Good Time Credits" = "Avg. Sentence Length w/ Good Time Credits",
         "(ME) Sent. Reduction % w/ Good Time Credits" = "Sentence Reduction %") %>% 
  full_join(wyparole_table, by = "Offense Categorization") %>%
  # select(1,2,3,6) %>% 
   rename("(WY) Avg. Sent. Reduction %" = "Average Sentence Reduction %",
         "(WY) Avg. Sent. Length w/ Parole" = "Avg. Sent. Length w/ Parole",
         "(WY) Incarceration Savings Per Prisoner" = "Incarceration Savings Per Prisoner") %>% 
  full_join(vtparole_table, by = "Offense Categorization") %>% 
  select(1:9,12:13) %>%
  rename("(VT) Avg. Sent. Reduction %" = "Average Sentence Reduction %",
         "(VT) Avg. Sent. Length w/ Parole" = "Avg. Sent. Length w/ Parole",
         "(VT) Incarceration Savings Per Prisoner" = "Incarceration Savings Per Prisoner",
         "Avg. Sent. Length" = "Avg. Sent. Length.x",
         "% of Prison Pop." = "% of Prison Pop..x") %>% 
  select(1,6,5,2,4,9,3,8,11,7,10) %>% 
  kbl(caption = "Sentence Length Comparisons") %>% 
  kable_minimal()

Sentence Length Comparisons
Offense Categorization	% of Prison Pop.	Avg. Sent. Length	(ME) Avg. Sent. Length w/ Good Time Credits	(WY) Avg. Sent. Length w/ Parole	(VT) Avg. Sent. Length w/ Parole	(ME) Sent. Reduction % w/ Good Time Credits	(WY) Avg. Sent. Reduction %	(VT) Avg. Sent. Reduction %	(WY) Incarceration Savings Per Prisoner	(VT) Incarceration Savings Per Prisoner
Murder (including non-negligent manslaughter)	3.08 %	14.8	12.4	3.3	6.7	16 %	78 %	55 %	2480.60	1747.21
Rape/sexual assault	9.2 %	3.5	2.9	1.1	2.5	17 %	69 %	29 %	517.69	215.70
Negligent manslaughter	1.72 %	5.2	3.6	1.7	3.6	31 %	67 %	31 %	754.97	345.13
Robbery	6.56 %	2.9	2.0	0.9	2.1	31 %	69 %	28 %	431.41	172.56
Aggravated or simple assault	16.93 %	1.5	1.1	0.4	1.3	27 %	73 %	13 %	237.27	43.14
Other violent offenses	1.33 %	2.0	1.6	0.5	1.8	20 %	75 %	10 %	323.56	43.14
Burglary	12.13 %	1.7	1.2	0.4	1.4	29 %	76 %	18 %	280.42	64.71
Larceny	9.93 %	1.1	0.8	0.1	1.1	27 %	91 %	0 %	215.70	0.00
Fraud	2.49 %	1.3	0.9	0.2	1.2	31 %	85 %	8 %	237.27	21.57
Other property offenses	2.24 %	1.9	1.3	0.5	1.5	32 %	74 %	21 %	301.99	86.28
Drugs (includes possession, distribution, trafficking, other)	22.33 %	1.4	1.0	0.3	1.3	29 %	79 %	7 %	237.27	21.57
Public order	10.14 %	1.2	0.8	0.1	1.2	33 %	92 %	0 %	237.27	0.00
Other/unspecified	0.79 %	3.9	2.7	0.5	2.0	31 %	87 %	49 %	733.40	409.84
NA	1.11 %	0.5	0.4	-0.1	1.0	20 %	120 %	-100 %	129.42	-107.85

Parole Release Predictions

Since we don’t have access to LSI Risk scores, we can’t predict who would be released on parole. Instead, I use the grant releases from the previous years to predict how many people would be released on parole and the savings for Maine after 2015 (we have total expenditure data for 2016 and above). I use the grant releases to probation in Maine (since parole release data before they abolished it is unavailable), and releases to parole in Wyoming and Vermont, all of which are then compared to each other. Presented below is the formula I use to calculate net savings in Maine and the summary statistics used for the formula.

Net incarceration savings $=$ (projected parole non-returnees $\times$ cost per prisoner in Maine) $-$ (projected parole releases $\times$ cost of parole per prisoner)

Model 1: Based on Maine Probation Releases

Probation Grant Rates in Maine

Average Cost of Parole: $229.1 Average Probation Grant Rate: 50.87% Average Parole Success Rate: 87.38%

If Maine implements parole and inmates are released to parole at the same rate they’re released to probation, then Maine will have an 25% average increase in their total budget.

parolerelease_me <- me_analysis %>% 
  filter(SENTLGTH != "(6) Life, LWOP, Life plus additional years, Death" | is.na(SENTLGTH)) %>%  #filter out LWOP
  filter(RELEASEYR >= 2016) %>% #2016 and above 
  count(RELEASEYR) %>% #to get the amount of prisoners per release yr after 2016
  mutate(parole_entries_me = case_when( #projected parole entries based on probation grant each yr
    RELEASEYR == 2016 ~ floor(n * .52728),
    RELEASEYR == 2017 ~ floor(n * .514),
    RELEASEYR == 2018 ~ floor(n * .5063),
    RELEASEYR == 2019 ~ floor(n * .5003),
    RELEASEYR == 2020 ~ floor(n * .4956)
  ), 
  parole_costs = parole_entries_me *250, #total parole costs each yr (parolees * cost of parole per prisoner)
  parole_nonreturnees_me = case_when( #projected parole nonreturnees based on probation success each yr
    RELEASEYR == 2016 ~ floor(parole_entries_me * .857),
    RELEASEYR == 2017 ~ floor(parole_entries_me * .822),
    RELEASEYR == 2018 ~ floor(parole_entries_me * .89),
    RELEASEYR == 2019 ~ floor(parole_entries_me * .897),
    RELEASEYR == 2020 ~ floor(parole_entries_me * .903)
  ), 
  reduced_incarceration_costs = case_when( #total reduced incarceration costs each yr (parole nonreturnees * cost of incarcaration per prisoner)
    RELEASEYR == 2016 ~ floor(parole_nonreturnees_me * 83874.5),
    RELEASEYR == 2017 ~ floor(parole_nonreturnees_me * 78732.15),
    RELEASEYR == 2018 ~ floor(parole_nonreturnees_me * 79306.1),
    RELEASEYR == 2019 ~ floor(parole_nonreturnees_me * 85417.9),
    RELEASEYR == 2020 ~ floor(parole_nonreturnees_me * 106548)
  ),
  net_incarceration_savings = reduced_incarceration_costs - parole_costs, #benefit - costs = net benefit
  Maine_prison_budget_proj = case_when( #maine's projected budget if they implement parole
    RELEASEYR == 2016 ~ floor(185865999 + net_incarceration_savings),
    RELEASEYR == 2017 ~ floor(186596104 + net_incarceration_savings),
    RELEASEYR == 2018 ~ floor(193824055 + net_incarceration_savings),
    RELEASEYR == 2019 ~ floor(194411181 + net_incarceration_savings),
    RELEASEYR == 2020 ~ floor(202761410 + net_incarceration_savings)
  ),
  Maine_prison_budget_perc = case_when( #maine's projected budget percent change if they implement parole
    RELEASEYR == 2016 ~ floor(((Maine_prison_budget_proj - 185865999)/185865999) * 100),
    RELEASEYR == 2017 ~ floor(((Maine_prison_budget_proj - 186596104)/186596104) * 100),
    RELEASEYR == 2018 ~ floor(((Maine_prison_budget_proj - 193824055)/193824055) * 100),
    RELEASEYR == 2019 ~ floor(((Maine_prison_budget_proj - 194411181)/194411181) * 100),
    RELEASEYR == 2020 ~ floor(((Maine_prison_budget_proj - 202761410)/202761410) * 100)
    ))
  
parolerelease_me_tidy <- parolerelease_me %>% #from wide to tidy format for graphing purposes
  pivot_longer(cols = n:Maine_prison_budget_proj, names_to = "Statistic", values_to = "Value")

parolerelease_me_tidy %>% 
  filter(Statistic == "net_incarceration_savings" | Statistic == "reduced_incarceration_costs"| Statistic == "parole_costs") %>% 
  ggplot(aes(x = RELEASEYR, y = Value, fill = Statistic)) +
   geom_col(position = position_dodge()) + ylim(0, 55500000) + #goes past 4e+07 
  geom_text(aes(label = Value), vjust = -0.5, size = 2, position = position_dodge(0.9)) +
  scale_fill_discrete(labels=c('Incarceration Savings (Net)', 'Parole Costs', 'Incarceration Costs (Reduced)')) +
  labs(title = "Net Incarceration Savings (2016-2020)", subtitle = "(If Parolees are Released Based on Probation Grant Rates in Maine)", x ="Release Year", y = "Costs (in Dollars)")

Model 2: Based on Wyoming Parole Releases

Parole Grant Rates in Wyoming

Average Cost of Parole: $1,918.45 (excludes 2019) Average Probation Grant Rate: 67.02% Average Parole Success Rate: 54.18%

If Maine implements parole and inmates are released to parole at the same rate they’re released to parole in Wyoming, then Maine will have a 19.5% increase in their total budget.

parolerelease_wy <- me_analysis %>% 
  filter(SENTLGTH != "(6) Life, LWOP, Life plus additional years, Death" | is.na(SENTLGTH)) %>%  #filter out LWOP
  filter(RELEASEYR >= 2016) %>% #2016 and above 
  count(RELEASEYR) %>% #to get the amount of prisoners per release yr after 2016
  mutate(parole_entries_wy = case_when( #projected parole entries based on parole grant each yr
    RELEASEYR == 2016 ~ floor(n * .6637),
    RELEASEYR == 2017 ~ floor(n * .6801),
    RELEASEYR == 2018 ~ floor(n * .6582),
    RELEASEYR == 2019 ~ floor(n * .6712),
    RELEASEYR == 2020 ~ floor(n * .6773)
  ), 
  parole_costs = case_when( #total parole costs each yr (parolees * cost of parole per prisoner), excludes 2019
    RELEASEYR == 2016 ~ floor(parole_entries_wy * 2179.87),
    RELEASEYR == 2017 ~ floor(parole_entries_wy * 1906.66),
    RELEASEYR == 2018 ~ floor(parole_entries_wy * 1766.26),
    RELEASEYR == 2020 ~ floor(parole_entries_wy * 1821)
  ), 
  parole_nonreturnees_wy = case_when( #projected parole nonreturnees based on parole success each yr
    RELEASEYR == 2016 ~ floor(parole_entries_wy * .603),
    RELEASEYR == 2017 ~ floor(parole_entries_wy * .574),
    RELEASEYR == 2018 ~ floor(parole_entries_wy * .603),
    RELEASEYR == 2019 ~ floor(parole_entries_wy * .488),
    RELEASEYR == 2020 ~ floor(parole_entries_wy * .441)
  ), 
  reduced_incarceration_costs = case_when( #total reduced incarceration costs each yr (parole nonreturnees * cost of incarcaration per prisoner)
    RELEASEYR == 2016 ~ floor(parole_nonreturnees_wy * 83874.5),
    RELEASEYR == 2017 ~ floor(parole_nonreturnees_wy * 78732.15),
    RELEASEYR == 2018 ~ floor(parole_nonreturnees_wy * 79306.1),
    RELEASEYR == 2019 ~ floor(parole_nonreturnees_wy * 85417.9),
    RELEASEYR == 2020 ~ floor(parole_nonreturnees_wy * 106548)
  ), 
  net_incarceration_savings = reduced_incarceration_costs - parole_costs, #benefit - costs = net benefit
  Maine_prison_budget_proj = case_when( #maine's projected budget if they implement parole
    RELEASEYR == 2016 ~ floor(185865999 + net_incarceration_savings),
    RELEASEYR == 2017 ~ floor(186596104 + net_incarceration_savings),
    RELEASEYR == 2018 ~ floor(193824055 + net_incarceration_savings),
    RELEASEYR == 2019 ~ floor(194411181 + net_incarceration_savings),
    RELEASEYR == 2020 ~ floor(202761410 + net_incarceration_savings)
  ),
  Maine_prison_budget_perc = case_when( #maine's projected budget percent change if they implement parole
    RELEASEYR == 2016 ~ floor(((Maine_prison_budget_proj - 185865999)/185865999) * 100),
    RELEASEYR == 2017 ~ floor(((Maine_prison_budget_proj - 186596104)/186596104) * 100),
    RELEASEYR == 2018 ~ floor(((Maine_prison_budget_proj - 193824055)/193824055) * 100),
    RELEASEYR == 2019 ~ floor(((Maine_prison_budget_proj - 194411181)/194411181) * 100),
    RELEASEYR == 2020 ~ floor(((Maine_prison_budget_proj - 202761410)/202761410) * 100)
    ))

parolerelease_wy_tidy <- parolerelease_wy %>% #from wide to tidy format for graphing purposes
  pivot_longer(cols = n:Maine_prison_budget_proj, names_to = "Statistic", values_to = "Value")

parolerelease_wy_tidy %>%
  filter(Statistic == "net_incarceration_savings" | Statistic == "reduced_incarceration_costs"| Statistic == "parole_costs") %>% 
  ggplot(aes(x = RELEASEYR, y = Value, fill = Statistic)) +
   geom_col(position = position_dodge()) + ylim(0, 55500000) + #goes past 4e+07 
  geom_text(aes(label = Value), vjust = -0.5, size = 2, position = position_dodge(0.9)) +
  scale_fill_discrete(labels=c('Net Incarceration Savings','Parole Costs', 'Reduced Incarceration Costs')) +
  labs(title = "Net Incarceration Savings (2016-2020)", subtitle = "(If Parolees are Released Based on Parole Grant Rates in Wyoming)", x ="Release Year", y = "Costs (in Dollars)")

Model 3: Based on Vermont Parole Releases

Parole Grant Rates in Vermont

Average Cost of Parole: $401.2 Average Probation Grant Rate: 24.62% Average Parole Success Rate: 56.9% (excludes 2019 and 2020)

If Maine implements parole and inmates are released to parole at the same rate they’re released to parole in Vermont, then Maine will have a 7% increase in their total budget.

#recidivism rate: (2016) 43.8% (2017) 42.9% (2018) 42.6%

parolerelease_vt <- me_analysis %>% 
  filter(SENTLGTH != "(6) Life, LWOP, Life plus additional years, Death" | is.na(SENTLGTH)) %>%
  #filter out LWOP
  filter(RELEASEYR >= 2016) %>% #2016 and above 
  count(RELEASEYR) %>% #to get the amount of prisoners per release yr after 2016
  mutate(parole_entries_vt = case_when( #projected parole entries based on parole grant each yr
    RELEASEYR == 2016 ~ floor(n * .3462),
    RELEASEYR == 2017 ~ floor(n * .2228),
    RELEASEYR == 2018 ~ floor(n * .1616),
    RELEASEYR == 2019 ~ floor(n * .1619),
    RELEASEYR == 2020 ~ floor(n * .3384)
  ),  
  parole_nonreturnees_vt = case_when( #projected parole nonreturnees based on parole success each yr, don't have recidivism rates after 2018
    RELEASEYR == 2016 ~ floor(parole_entries_vt * .562),
    RELEASEYR == 2017 ~ floor(parole_entries_vt * .571),
    RELEASEYR == 2018 ~ floor(parole_entries_vt * .574)
  ),
  reduced_incarceration_costs = case_when( #total reduced incarceration costs each yr (parole nonreturnees * cost of incarcaration per prisoner)
    RELEASEYR == 2016 ~ floor(parole_nonreturnees_vt * 83874.5),
    RELEASEYR == 2017 ~ floor(parole_nonreturnees_vt * 78732.15),
    RELEASEYR == 2018 ~ floor(parole_nonreturnees_vt * 79306.1),
    RELEASEYR == 2019 ~ floor(parole_nonreturnees_vt * 85417.9),
    RELEASEYR == 2020 ~ floor(parole_nonreturnees_vt * 106548)
  ), 
  parole_costs = case_when( #total parole costs each yr (parolees * cost of parole per prisoner)
    RELEASEYR == 2016 ~ floor(parole_entries_vt * 327.134),
    RELEASEYR == 2017 ~ floor(parole_entries_vt * 419.861),
    RELEASEYR == 2018 ~ floor(parole_entries_vt * 390.898),
    RELEASEYR == 2019 ~ floor(parole_entries_vt * 437.468),
    RELEASEYR == 2020 ~ floor(parole_entries_vt * 430.615)
  ), 
  net_incarceration_savings = reduced_incarceration_costs - parole_costs, #benefit - costs = net benefit 
  Maine_prison_budget_proj = case_when( #maine's projected budget if they implement parole
    RELEASEYR == 2016 ~ floor(185865999 + net_incarceration_savings),
    RELEASEYR == 2017 ~ floor(186596104 + net_incarceration_savings),
    RELEASEYR == 2018 ~ floor(193824055 + net_incarceration_savings),
    RELEASEYR == 2019 ~ floor(194411181 + net_incarceration_savings),
    RELEASEYR == 2020 ~ floor(202761410 + net_incarceration_savings)
  ),
  Maine_prison_budget_perc = case_when( #maine's projected budget percent change if they implement parole
    RELEASEYR == 2016 ~ floor(((Maine_prison_budget_proj - 185865999)/185865999) * 100),
    RELEASEYR == 2017 ~ floor(((Maine_prison_budget_proj - 186596104)/186596104) * 100),
    RELEASEYR == 2018 ~ floor(((Maine_prison_budget_proj - 193824055)/193824055) * 100),
    RELEASEYR == 2019 ~ floor(((Maine_prison_budget_proj - 194411181)/194411181) * 100),
    RELEASEYR == 2020 ~ floor(((Maine_prison_budget_proj - 202761410)/202761410) * 100)
    ))

parolerelease_vt_tidy <- parolerelease_vt %>% #from wide to tidy format for graphing purposes
  pivot_longer(cols = n:Maine_prison_budget_proj, names_to = "Statistic", values_to = "Value")

parolerelease_vt_tidy %>%
  filter(Statistic == "net_incarceration_savings" | Statistic == "reduced_incarceration_costs"| Statistic == "parole_costs") %>% 
  ggplot(aes(x = RELEASEYR, y = Value, fill = Statistic)) +
   geom_col(position = position_dodge()) + ylim(0, 55500000) +  
  geom_text(aes(label = Value), vjust = -0.5, size = 2, position = position_dodge(0.9)) +
  scale_fill_discrete(labels=c('Incarceration Savings (Net)', 'Incarceration Costs (Reduced)', 'Parole Costs')) +
  labs(title = "Net Incarceration Savings (2016-2020)", subtitle = "(If Parolees are Released Based on Parole Grant Rates in Vermont)", x ="Release Year", y = "Costs (in Dollars)")

Parole in Maine? Modeling Parole in Maine and It’s Economic Impacts

Sokona Mangane

April 2023