We have recorded indoor temperatures for each unit April 1 through Sept. 30 of 2022, 2023, 2024 and 2025. According to the Texas Department of Criminal Justice, indoor temperatures are taken at 3 p.m. every day from April through September at all units that do not have air conditioning. The data was acquired by Lauren McGaughy at KUT through a public information request to the Texas Department of Criminal Justice. All temperatures are recorded in Fahrenheit.
In this notebook, we will explore the following questions:
# Loading our libraries
library(tidyverse)
library(janitor)
library(ggthemes)
library(DT)
library(ggwaffle)
# suppress group warning
options(dplyr.summarise.inform = FALSE)
# Reading in the cleaned and combined indoor logs.
indoor_logs_clean <- read_rds("data-processed/01-indoor-temps-all-cleaned.rds")
# Adding variables to help analysis by date.
indoor_logs <- indoor_logs_clean |> mutate(
flo_mo = floor_date(date, unit = "month"),
year = year(date)
) Checking where we might have missing data … Here we show the year without 183 recorded days in red so you can see them.
indoor_logs |>
group_by(unit, year) |>
summarize(days_recorded = n()) |>
pivot_wider(names_from = year, values_from = days_recorded) |>
datatable() |>
formatStyle(
columns = c("2022", "2023", "2024", "2025"),
backgroundColor = styleInterval(
c(182.5, 183.5),
c("#f8d7da", NA, "#f8d7da")
)
)Data takeaway: For each year, we have 183 total days of indoor temperature recordings. For each year of data, there are 68 TDCJ units that recorded indoor temperatures. However, the units recorded differ by year, with 74 total units appearing across all four years of data. Sixty-three units appear in all four years of data. The 11 units that differ by year are as follows:
The top 10 hottest days indoors for each year.
indoor_logs |>
group_by(year) |>
slice_max(temperature, n = 10) |>
select(year, unit, date, temperature)Summarizing how many times each unit appears in the top 10 hottest temperature recordings each year.
indoor_logs |>
group_by(year) |>
slice_max(temperature, n = 10) |>
select(year, unit, date, temperature) |>
group_by(unit, year) |>
summarize(n = n()) |>
arrange(desc(n))Note - several years include more than 10 recordings for the top 10 because of days/units where the temperature recorded was the same.
Data takeaways:
days100 <- indoor_logs |>
group_by(year) |>
select(year, unit, date, temperature) |>
filter(temperature >= 100) |>
arrange(desc(temperature))
days100 |>
datatable()Data takeaway: It reached 100 degrees indoors 211 times across all units between 2022 and 2025, 108 of which occurred in 2023. Twice, it reached 105 or above. The hottest temperature across all years was recorded at the Mountain View unit on June 19, 2023, which reached 106.8 degrees. The Mountain View unit was renamed to Patrick O’Daniel (appears as O’Daniel in the data) in January 2024 and houses female prisoners in Gatesville, Texas, about 45 miles west of Waco. It houses up to 644 prisoners at full capacity.
It reached 100 degrees more than 100 times in 2003.
days100 |>
filter(year == 2023)This shows how many days per month a unit reached a specific indoor temperature.
count_temps_unit_month <- indoor_logs|>
group_by(unit, flo_mo) |>
summarize(
total_days = n(),
cnt_85 = sum(temperature >= 85, na.rm = T),
cnt_90 = sum(temperature >= 90, na.rm = T),
cnt_95 = sum(temperature >= 95, na.rm = T),
cnt_100 = sum(temperature >= 100, na.rm = T),
cnt_105 = sum(temperature >= 105, na.rm = T),
.groups = "drop"
)
count_temps_unit_month |> datatable()Number of days unit reached a certain temperature within a year (which is really a 6-month period). Also the percentage of recorded days it was reached.
count_temps_unit_year <- indoor_logs |>
group_by(unit, year) |>
summarize(
total_days = n(),
cnt_n85 = sum(temperature < 85, na.rm = T),
cnt_85 = sum(temperature >= 85, na.rm = T),
cnt_90 = sum(temperature >= 90, na.rm = T),
cnt_95 = sum(temperature >= 95, na.rm = T),
cnt_100 = sum(temperature >= 100, na.rm = T),
.groups = "drop"
) |> mutate(
pct_85 = cnt_85/total_days,
pct_90 = cnt_90/total_days,
pct_95 = cnt_95/total_days,
pct_100 = cnt_100/total_days
) |>
mutate(across(starts_with("pct"), \(x) round(x, 2))) # rounds percentage
count_temps_unit_year |>
datatable()How many “prison days” it reached a certain temperature each year. “Prison days” include all days recorded across all units each year.
count_temps_year <- indoor_logs|>
group_by(year) |>
summarize(
total_days = n(),
cnt_85 = sum(temperature >= 85, na.rm = T),
cnt_90 = sum(temperature >= 90, na.rm = T),
cnt_95 = sum(temperature >= 95, na.rm = T),
cnt_100 = sum(temperature >= 100, na.rm = T)
) |> mutate(
pct_85 = cnt_85/total_days,
pct_90 = cnt_90/total_days,
pct_95 = cnt_95/total_days,
pct_100 = cnt_100/total_days
) |>
mutate(across(starts_with("pct"), \(x) round(x, 2))) # rounds percentage
count_temps_year Data takeaways:
Starting with the 85 threshold data above, we find total units, then subtract a count units that didn’t reach 85.
count_temps_unit_year |>
group_by(year) |>
summarise(
total_units = n(),
units_met_85 = total_units - sum(cnt_85 == 0),
)Let’s find that unit in 2022 that didn’t reach 85 degrees. These are the lowest of the “highest temperature for each unit” where we can see Sayle didn’t reach 85 degrees twice. We only have 2022 data for the Sayle unit.
indoor_logs |>
filter(year == 2022) |>
group_by(unit) |>
slice_max(temperature) |> # find highest temp each unit
arrange(temperature) |> # find lowest of those
select(unit, date, temperature) |>
filter(temperature <= 90) # cutting here to show the next units for comparisonTakeaway: Every unit hit 85 degrees at least once every year with one exception in 2022, where the Sayle unit recorded 84.6 degrees as its highest temperature on two separate days. The Sayle unit is located a little over 100 miles west of Forth Worth and was not included in the data for 2023, 2024 or 2025.
This counts the number of units that reached 85 degrees for half of the days measured. We also calculate for 75% of the days measured.
count_temps_unit_year |>
group_by(year) |>
select(year, total_days, cnt_85, pct_85) |> arrange(desc(cnt_85)) |>
summarize(
total_units = n(),
cnt_half_85 = sum(pct_85 >= .50),
cnt_34ths_85 = sum(pct_85 >= .75)
)Data takeaways:
This shows units each year with the most days to reach 85.
at85_unit_year <- count_temps_unit_year |>
group_by(year) |>
slice_max(pct_85, n = 3) |> # n argument adjust number per year
select(unit, year, total_days, cnt_85, pct_85)
at85_unit_yearWe’re taking the results above and focusing on Garza West for clarity.
at85_unit_year |>
filter(unit == "Garza West")Data takeaway:
Here we find the fewest number of days it reached the 85-degree threshold within each year. We remove the Sayle unit that didn’t ever reach 85 in 2022.
count_temps_unit_year |>
filter(cnt_85 > 0) |> # removes Sayle in 2022
group_by(year) |>
slice_min(cnt_85, n = 1) |>
select(unit, year, total_days, cnt_85, pct_85)Data takeaways:
count_temps_unit_year |>
group_by(year) |>
summarise(
total_units = n(),
units_met_90 = total_units - sum(cnt_90 == 0),
)Data takeaway: Almost every unit also reached 90 degrees inside at least once each year, all but three units in 2022, one in 2023 and six in 2025.
This counts the number of units that reached 90 degrees for half of the days measured. We also calculate for 75% of the days measured.
count_temps_unit_year |>
group_by(year) |>
select(year, total_days, cnt_90, pct_90) |> arrange(desc(cnt_90)) |>
summarize(
total_units = n(),
cnt_half_90 = sum(pct_90 >= .50),
cnt_34th_90 = sum(pct_90 >= .75)
)Let’s find those units that reach 90 for 75% of the time.
count_temps_unit_year |>
# filter(year == 2023) |>
select(unit, year, total_days, cnt_90, pct_90) |>
filter(pct_90 >= .75) |>
arrange(year)Data takeaways:
This shows units with the most days to reach 95 within each year.
units_at_90 <- count_temps_unit_year |>
group_by(year) |>
slice_max(pct_90, n = 3) |> # n argument adjust number per year
select(unit, year, total_days, cnt_90, pct_90)
units_at_90units_at_90 |> filter(unit == "Garza West")Data takeaway: In the Garza West unit, the temperature reached at least 90 degrees indoors on at least 65% of days recorded each year from 2022 to 2025. In 2023 and 2024, 73% and 75% of days reached 90 degrees, respectively.
count_temps_unit_year |>
group_by(year) |>
summarise(
total_units = n(),
units_met_95 = total_units - sum(cnt_95 == 0),
)Takeaway: In each year from 2022 to 2024, at least 50 of 68 units reached 95 degrees inside at least once between April 1 and Sept. 30. In 2025, 26 units reached 95 degrees at least once.
Here we calculate the number of units that reach 95 degrees at least a quarter of recorded days each year. This is a lower percentage than other thresholds.
count_temps_unit_year |>
group_by(year) |>
summarize(
total_units = n(),
cnt_quarter_95 = sum(pct_95 >= .25)
)Data takeaway: In 2023, there were 13 units where it reached 95 degrees inside a quarter of the time, or one in every four days between April and September. This was true for six units in 2022, three units in 2024, and one in 2025.
This shows units with the most days to reach 95 within each year.
units_at_95 <- count_temps_unit_year |>
group_by(year) |>
slice_max(pct_95, n = 1) |> # n argument adjust number per year
select(unit, year, total_days, cnt_95, pct_95)
units_at_95Data takeaway: At the Garza West unit in 2023, it reached 95 degrees indoors more than 100 times, or individual days. In other words, nearly three out of every five days between April 1 and Sept. 30. reached 95 degrees indoors.
count_temps_unit_year |>
group_by(year) |>
summarise(
total_units = n(),
units_met_100 = total_units - sum(cnt_100 == 0),
)count_temps_unit_year |>
group_by(unit, year) |>
arrange(desc(cnt_100))indoor_logs |>
filter(unit == "Garza West", temperature == 100)Data takeaways: Between 2022 and 2024, at least a dozen units each year reached 100 degrees inside at least once. In 2023, 18 units reached 100 degrees indoors at least once between April and September. In 2025, only one unit reached 100 degrees - Garza West on August 29.
This attempts to show how many units reached 100 degrees at least once in a given year.
Since we’ll want to make this chart for several years, I’ll make it into a function to reuse.
prison_waffle <- function(.data, yr) {
.data |>
select(unit, year, cnt_100) |>
mutate(r_100 = if_else(cnt_100 > 0, "At least 100°", "High below 100°")) |>
filter(year == yr) |>
waffle_iron(aes_d(group = r_100)) |>
ggplot(aes(x, y, fill = group)) +
geom_waffle() +
theme_waffle() +
labs(
title = paste("Number of units to reach 100 in", yr),
x = "", y = "", fill = "High temp"
)
ggsave(paste0("figures/waffle_plot_", yr, ".png"))
}
count_temps_unit_year |> prison_waffle(2022)
count_temps_unit_year |> prison_waffle(2023)
count_temps_unit_year |> prison_waffle(2024)
count_temps_unit_year |> prison_waffle(2025)
How many times each year a unit hit 100 degrees inside.
unit_days_100 <- count_temps_unit_year |>
filter(cnt_100 > 0) |>
select(unit, year, total_days, cnt_100, pct_100) |>
arrange(year, cnt_100 |> desc())
unit_days_100Let’s take that and peek at just Garza West
unit_days_100 |> filter(unit == "Garza West")Takeaway: Garza West reached 100 degrees inside on 46 days between April 1 and Sept. 30 in 2023. In other words, 25%, or one in four days, reached 100 degrees.
Here we’ll attempt to plot how many days a unit reached a certain temperature within a given year. We need to recalculate our thresholds.
Here we bin the number of days at certain temperatures, which is different than we we’ve been doing previously (which was total days at/above a temp). We also have to account for missing data, so we create cnt_na to add any missing days to the data.
temp_thrsh_unit_year <- indoor_logs |>
group_by(unit, year) |>
summarize(
cnt_n85 = sum(temperature < 85, na.rm = T),
cnt_85 = sum(temperature >= 85 & temperature < 90, na.rm = T),
cnt_90 = sum(temperature >= 90 & temperature < 95, na.rm = T),
cnt_95 = sum(temperature >= 95 & temperature < 100, na.rm = T),
cnt_100 = sum(temperature >= 100, na.rm = T),
.groups = "drop"
) |>
mutate(cnt_na = 183 - (cnt_n85 + cnt_85 + cnt_90 + cnt_95 + cnt_100)) |>
relocate(cnt_na, .before = cnt_n85)
temp_thrsh_unit_yearTo plot this we have to reshape the data, then plot as a bar chart. We save the chart before displaying it so we can control the size since there are so many units to show.
The plotting has been turned into a function to keep it dRy.
First we pivot the data and turn the thresholds into a factor.
thresh_data <- temp_thrsh_unit_year |>
pivot_longer(
cols = starts_with("cnt"),
names_to = "threshold",
values_to = "days"
) |>
mutate(threshold = fct_inorder(threshold))
thresh_data |> glimpse()Rows: 1,632
Columns: 4
$ unit <chr> "Allred", "Allred", "Allred", "Allred", "Allred", "Allred", …
$ year <dbl> 2022, 2022, 2022, 2022, 2022, 2022, 2023, 2023, 2023, 2023, …
$ threshold <fct> cnt_na, cnt_n85, cnt_85, cnt_90, cnt_95, cnt_100, cnt_na, cn…
$ days <dbl> 0, 60, 43, 48, 31, 1, 0, 82, 30, 33, 33, 5, 0, 77, 45, 51, 9…# plot the data into an object and then save that object into the figures folder.
plot_thresh_data <- function(yr) {
thresh_plot <- thresh_data |>
filter(year == yr) |>
ggplot(aes(x = days, y = fct_rev(unit), fill = threshold)) +
geom_bar(stat = "identity", position = position_stack(reverse = TRUE)) +
scale_fill_brewer(
palette='OrRd',
labels = c(
cnt_na = "Missing",
cnt_n85 = "Under 85°",
cnt_85 = "85° to 89°",
cnt_90 = "90° to 94°",
cnt_95 = "95° to 99°",
cnt_100 = "100° or more"
)) +
# theme(legend.position = "bottom") +
labs(
title = paste("Tracking indoor heat in prisons,", yr),
x = "Number of days at each temperature", y = "",
fill = "Indoor high"
)
ggsave(paste0("figures/thresh_plot_", yr, ".png"), width = 6, height = 12)
}
plot_thresh_data(2022)
plot_thresh_data(2023)
plot_thresh_data(2024)
plot_thresh_data(2025)
We are using the 2023 version of this threshold chart for a story on KUT.org, so we’ll work out the language, etc.
thresh_pot_2023_print <- thresh_data |>
filter(year == 2023) |>
ggplot(aes(x = days, y = fct_rev(unit), fill = threshold)) +
geom_bar(stat = "identity", position = position_stack(reverse = TRUE)) +
scale_fill_brewer(
palette='OrRd',
labels = c(
cnt_na = "Missing",
cnt_n85 = "Under 85°",
cnt_85 = "85° to 89°",
cnt_90 = "90° to 94°",
cnt_95 = "95° to 99°",
cnt_100 = "100° or more"
)) +
# theme(legend.position = "bottom") +
labs(
title = str_wrap("More often than not, temperatures reached 85 degrees inside unair-conditioned prisons in the summer of 2023", width = 55),
subtitle = str_wrap("The Texas Department of Criminal Justice records temperatures at 3 p.m. every day from April 1 through Sept. 30 at all prison units that do not have air conditioning. We plotted how many of those days in 2023 fell into different temperatures ranges for each prison.", width = 68),
x = "Number of days at each temperature", y = "",
fill = "Indoor high
temperature",
caption = "The Manuel A. Segovia Unit did not have data for all days.
By: Media Innovation Group; Source: TDCJ"
)
ggsave("figures/thresh_pot_2023_print.png", width = 6, height = 12)
Here we reformat this data to build this chart in datawrapper.
thresh_2023_dw <- thresh_data |>
filter(year == 2023) |> # remove for all years
pivot_wider(names_from = threshold, values_from = days) |>
mutate(
cnt_na = 183 - (cnt_n85 + cnt_85 + cnt_90 + cnt_95 + cnt_100)
) |>
select(
Unit = unit,
# Year = year,
`Missing` = cnt_na,
`Under 85°` = cnt_n85,
`85° to 89°` = cnt_85,
`90° to 94°` = cnt_90,
`95° to 99°` = cnt_95,
`100° or more` = cnt_100
)
thresh_2023_dwthresh_2023_dw |> write_csv("data-processed/thresh_2023_dw.csv")Determining the indoor temperatures on the days and unit where prisoners died.
indoor_logs |>
select(unit, date, temperature) |>
filter(unit == "Murray", date == "2023-06-30")Data takeaway: On June 30, 2023, Elizabeth Hagerty died in her cell at the Murray unit, located in Gatesville, Texas, about 100 miles north of Austin. According to the autopsy report, she died at 12:48 a.m. The temperature in her cell was 95.7 degrees. The recorded indoor temperature for the Murray unit that day, reportedly taken at 3 p.m., was 90 degrees. Hagerty was 37 years old.
TDCJ hand-recorded outdoor temperature log for 12:30 a.m. on June 30, 2023 at the Murray unit: 83 degrees with a heat index of 87.
NWS log for Gatesville (station ID KGOP) at 12:15 a.m.: 82 degrees with a heat index of 85.6.
indoor_logs |>
select(unit, date, temperature) |>
filter(unit == "Hughes", date == "2023-08-05")Data takeaway: On Aug. 5, 2023, John Castillo died indoors at the Hughes unit, located in Gatesville, Texas, about 100 miles north of Austin. According to the autopsy report, he died at 11:33 p.m. His core body temperature was 107.5 degrees. The recorded indoor temperature for the Hughes unit that day, reportedly taken at 3 p.m., was 94.4 degrees. Castillo was 32 years old.
TDCJ hand-recorded outdoor temperature log for 11:30 p.m. on Aug. 5, 2023 at the Hughes unit: 88 degrees with a heat index of 89.
NWS log for Gatesville (station ID KGOP) at 11:15 p.m.: 86 degrees with a heat index of 86.8.
indoor_logs |>
select(unit, date, temperature) |>
filter(unit == "Coffield", date == "2023-08-21")Data takeaway: On Aug. 21, 2023, Patrick Womack died in his cell at the Coffield unit, located in Tennessee Colony, Texas, about 100 miles northwest of Waco. According to the autopsy report, he died at 1:53 p.m. His core body temperature was 106.9 degrees. The recorded indoor temperature for the Coffield unit that day, reportedly taken at 3 p.m., was 96.6 degrees. Womack was 50 years old.
TDCJ hand-recorded outdoor temperature log for 1:30 p.m. on Aug. 21, 2023 at the Coffield unit: 104 degrees with a heat index of 113.
NWS log for Palestine (station ID KPSN) at 2:15 p.m.: 104 degrees with a heat index of 105.3. Note: the Palestine Municipal Airport station is located 15 miles away from the Coffield unit.
indoor_logs |>
select(unit, date, temperature) |>
filter(unit == "Goree", date == "2024-06-07")Data takeaway: On June 7, 2024, Oscar R. Cortez died outdoors at the Goree unit, located in Hunstville, Texas, about 65 north of Houston. According to the Death in Custody report, he died at 5:20 p.m. The recorded indoor temperature for the Goree unit that day, reportedly taken at 3 p.m., was 86.6 degrees. Cortez was 44 years old.
NWS log for Hunstville (station ID KUTS) at 4:53 p.m.: 91 degrees with a heat index of 93.