Andrew Gelman, Jennifer Hill, Aki Vehtari 2021-04-20
Tidyverse version by Bill Behrman.
Outcomes after civil war in countries with and without United Nations peacekeeping. See Chapter 1 in Regression and Other Stories.
# Packages
library(tidyverse)
library(haven)
library(lubridate)
# Parameters
# Outcomes after civil wars with and without U.N. peacekeeping
file_peace <- here::here("Peacekeeping/data/pk&pkept_old.dta")
# Common code
file_common <- here::here("_common.R")
#===============================================================================
# Run common code
source(file_common)Estimating the effects of United Nations peacekeeping, using pre-treatment variables to adjust for differences between treatment and control groups
Data
peace <-
file_peace %>%
read_dta() %>%
select(id:pkop, pcw, morewar, hazard1) %>%
mutate(across(where(is.labelled), as_factor))
glimpse(peace)#> Rows: 206
#> Columns: 14
#> $ id <dbl> 10, 15, 20, 30, 40, 45, 60, 70, 70, 70, 75, 90, 95, 100, 100,…
#> $ ccode <dbl> 700, 700, 700, 615, 615, 615, 540, 540, 540, 540, 540, 160, 3…
#> $ cname <chr> "Afghanistan-Mujahideen", "Afghanistan-Taliban", "Afghanistan…
#> $ cfdate <date> 1992-04-25, 1993-03-07, NA, 1963-01-15, 1997-10-15, NA, 1991…
#> $ faildate <date> 1992-08-10, 1993-04-15, NA, NA, NA, NA, 1992-10-11, 1998-12-…
#> $ firstob <dbl> 1, 1, 1, 1, 1, 1, 1, 1, NA, NA, 1, 1, 1, 1, NA, 1, NA, 1, 1, …
#> $ t0 <date> 1992-04-25, 1993-03-07, NA, 1963-01-15, 1997-10-15, NA, 1991…
#> $ t1 <date> 1992-08-10, 1993-04-15, NA, 2004-12-31, 2004-12-31, NA, 1992…
#> $ outcome <fct> Victory-Rebels, Settlement, NA, Victory-Rebels, Truce, NA, Se…
#> $ pk_dum <dbl> 0, 0, NA, 0, 0, NA, 1, 1, 1, 1, NA, 0, 0, 0, 0, 0, 0, 0, 0, 0…
#> $ pkop <dbl> 0, 0, NA, 0, 0, NA, 2, 2, 4, 3, NA, 0, 0, 0, 1, 0, 0, 0, 0, 0…
#> $ pcw <dbl> 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 0, 0, 1…
#> $ morewar <dbl> 1, 1, NA, 0, 0, NA, 1, 1, 1, 1, NA, 0, 1, 0, 0, 0, 0, 0, 1, 1…
#> $ hazard1 <dbl> 3.11e-03, 3.00e-03, NA, 1.85e-05, 2.95e-04, NA, 7.24e-02, 1.2…
Variable labels.
peace %>%
keep(~ "label" %in% names(attributes(.))) %>%
map_chr(~ attr(., "label"))#> ccode outcome
#> "COW country id number" "Outcome of the War"
#> pk_dum pkop
#> "peacekeeping dummy" "peacekeeping operation, mission type"
#> pcw hazard1
#> "post cold war termination" "predicted hazard"
Conflict ID.
peace %>%
count(id, name = "nrows") %>%
count(nrows)#> # A tibble: 5 x 2
#> nrows n
#> * <int> <int>
#> 1 1 148
#> 2 2 14
#> 3 3 5
#> 4 4 2
#> 5 7 1
148 of the conflict IDs have only one row, but 22 have multiple rows. One has seven rows. Let’s look at an ID with multiple rows.
peace %>%
group_by(id) %>%
filter(n() == 7) %>%
select(id, cname, cfdate, faildate, t0, t1, pk_dum, pkop)#> # A tibble: 7 x 8
#> # Groups: id [1]
#> id cname cfdate faildate t0 t1 pk_dum pkop
#> <dbl> <chr> <date> <date> <date> <date> <dbl> <dbl>
#> 1 590 Haiti 1994-09-18 NA 1994-09-18 1995-03-31 1 5
#> 2 590 Haiti NA NA 1995-04-01 1996-06-30 1 5
#> 3 590 Haiti NA NA 1996-07-01 1997-07-31 1 3
#> 4 590 Haiti NA NA 1997-08-01 2000-03-14 1 2
#> 5 590 Haiti NA NA 2000-03-15 2004-02-28 0 0
#> 6 590 Haiti NA NA 2004-02-29 2004-05-31 1 5
#> 7 590 Haiti NA NA 2004-06-01 2004-12-31 1 5
Haiti had a ceasefire that started on 1994-09-18. The seven rows
correspond to seven different time periods. There were peacekeepers for
six of these periods (pk_dum), though the peacekeeping operation and
mission (pkop) varied.
We’ll add a new variable peacekeepers to indicate whether there were
peacekeepers at any time. And we’ll now collapse the data to one row per
conflict ID, using the first row.
peace <-
peace %>%
group_by(id) %>%
summarize(
peacekeepers = any(pk_dum == 1),
across(everything(), first)
)We are only interested in conflicts where we know the peacekeeper status.
peace %>%
filter(is.na(peacekeepers)) %>%
select(id, cname, peacekeepers, cfdate, faildate, outcome, pk_dum, pkop)#> # A tibble: 13 x 8
#> id cname peacekeepers cfdate faildate outcome pk_dum pkop
#> <dbl> <chr> <lgl> <date> <date> <fct> <dbl> <dbl>
#> 1 20 Afghanistan-Ta… NA NA NA <NA> NA NA
#> 2 45 Algeria-GIA/GS… NA NA NA <NA> NA NA
#> 3 75 Angola NA NA NA <NA> NA NA
#> 4 200 Burundi NA NA NA <NA> NA NA
#> 5 255 Chad NA NA NA <NA> NA NA
#> 6 300 Colombia NA NA NA <NA> NA NA
#> 7 375 CongoD.R./Zaire NA NA NA <NA> NA NA
#> 8 630 India-Kashmir NA NA NA <NA> NA NA
#> 9 685 Indonesia-Aceh NA NA NA <NA> NA NA
#> 10 855 Liberia NA NA NA <NA> NA NA
#> # … with 3 more rows
There are 13 conflicts with no peacekeeper data. We’ll remove them.
peace <-
peace %>%
drop_na(peacekeepers)
peace %>%
count(peacekeepers)#> # A tibble: 2 x 2
#> peacekeepers n
#> * <lgl> <int>
#> 1 FALSE 114
#> 2 TRUE 43
157 conflicts remain, of which 41 had peacekeepers.
We will say that a country remained at peace if the ceasefire hadn’t
failed by the end of the data collection, in other words if faildate
is NA. The NA status of faildate is also encoded in morewar.
peace %>%
count(is.na(faildate), morewar)#> # A tibble: 2 x 3
#> `is.na(faildate)` morewar n
#> <lgl> <dbl> <int>
#> 1 FALSE 1 85
#> 2 TRUE 0 72
The ceasefires held 72 times and failed 85 times.
Let’s see the relationship between having peacekeepers and a return to conflict.
peace %>%
count(peacekeepers, peace = is.na(faildate))#> # A tibble: 4 x 3
#> peacekeepers peace n
#> <lgl> <lgl> <int>
#> 1 FALSE FALSE 65
#> 2 FALSE TRUE 49
#> 3 TRUE FALSE 20
#> 4 TRUE TRUE 23
peace %>%
group_by(peacekeepers) %>%
summarize(peace_prop = mean(is.na(faildate)))#> # A tibble: 2 x 2
#> peacekeepers peace_prop
#> * <lgl> <dbl>
#> 1 FALSE 0.430
#> 2 TRUE 0.535
In about 51% of conflicts with peacekeepers, the ceasefires held, compared to 44% of conflicts without peacekeepers.
If conflict did return, let’s see the relationship between having peacekeepers and the length of time for the return.
peace %>%
filter(!is.na(faildate)) %>%
mutate(return_years = interval(cfdate, faildate) / dyears(1)) %>%
group_by(peacekeepers) %>%
summarize(across(return_years, list(mean = mean, median = median)))#> # A tibble: 2 x 3
#> peacekeepers return_years_mean return_years_median
#> * <lgl> <dbl> <dbl>
#> 1 FALSE 4.17 2.00
#> 2 TRUE 2.61 1.25
In the conflicts where ceasefires did not hold, the ceasefires ended sooner when there were peacekeepers.
If conflict did not return, let’s see the relationship between having peacekeepers and the duration of peace. We’ll use the end of data collection, 2004-12-31, as the end date.
peace %>%
filter(is.na(faildate)) %>%
mutate(peace_years = interval(cfdate, ymd("2004-12-31")) / dyears(1)) %>%
group_by(peacekeepers) %>%
summarize(across(peace_years, list(mean = mean, median = median)))#> # A tibble: 2 x 3
#> peacekeepers peace_years_mean peace_years_median
#> * <lgl> <dbl> <dbl>
#> 1 FALSE 27.6 22.3
#> 2 TRUE 12.4 10.5
In the conflicts where the ceasefires held, the duration of peace was shorter when there were peacekeepers.
Years until return of war.
ceasefire_failed <-
peace %>%
filter(!is.na(faildate)) %>%
mutate(return_years = interval(cfdate, faildate) / dyears(1))
return_years_means <-
ceasefire_failed %>%
group_by(peacekeepers) %>%
summarize(return_years_mean = mean(return_years))
peacekeepers_labels <-
c(
"FALSE" = "Without peacekeeping",
"TRUE" = "With peacekeeping"
)
ceasefire_failed %>%
ggplot(aes(return_years)) +
geom_histogram(binwidth = 0.5, boundary = 0) +
geom_vline(aes(xintercept = return_years_mean), data = return_years_means) +
scale_x_continuous(breaks = scales::breaks_width(2)) +
facet_grid(
rows = vars(peacekeepers),
labeller = labeller(peacekeepers = peacekeepers_labels)
) +
labs(
title = "Years until return of war",
x = "Years",
y = "Count"
)
Years until return of war.
ceasefire_failed %>%
drop_na(hazard1) %>%
ggplot(aes(hazard1, return_years, color = peacekeepers)) +
geom_point() +
scale_x_log10(
breaks = c(1.4e-5, 4.5e-2),
minor_breaks = NULL,
labels = c("Better conditions", "Worse conditions")
) +
scale_y_continuous(breaks = scales::breaks_width(5)) +
scale_color_discrete(
breaks = c(FALSE, TRUE),
labels = c("Without peacekeeping", "Without peacekeeping")
) +
theme(
axis.ticks.x = element_blank(),
legend.position = "bottom",
panel.grid.major.x = element_blank()
) +
labs(
title = "Years until return of war",
x = "Pre-treatment measure of conditions in the country",
y = "Years",
color = NULL
)
The plot uses hazard1 as a measure of the conditions in the country at
the time of the ceasefire. Below a threshold of this measure, it was
possible for ceasefires to last 10 years or longer. Over this threshold,
all of the countries returned to war in 5 years or less.