periods.R

library(tidyverse)
library(slider)

final <- tibble::tribble(
           ~ccode, ~year, ~publicdept, ~lifeexpec, ~gdp_pc, ~cpia, ~xm_gdp, ~polity,
            "AFG", 2006L,          23,       58.8,   0.264,  3.33,    103L,    -66L,
            "AFG", 2007L,        20.1,       59.4,    0.36,  3.17,    102L,    -66L,
            "AFG", 2008L,          23,       58.8,   0.264,  3.33,    103L,    -66L,
            "AFG", 2009L,        20.1,       59.4,    0.36,  3.17,    102L,    -66L,
            "AFG", 2010L,          28,         60,    0.36,  3.33,    103L,    -66L,
            "AFG", 2011L,          24,         68,    0.36,  3.17,    102L,    -66L,
            "AFB", 2006L,          19,         66,   0.264,  3.17,    102L,    -66L,
            "AFB", 2007L,          20,         67,    0.36,  4.17,    103L,    -66L,
            "AFB", 2008L,          21,         68,   0.456,  5.17,    104L,    -66L,
            "AFB", 2009L,          22,         69,   0.552,  6.17,    105L,    -66L,
            "AFB", 2010L,          23,         70,   0.648,  7.17,    106L,    -66L,
            "AFB", 2011L,          24,         71,   0.744,  8.17,    107L,    -66L
           )

# calculate mean per ccode with the new summarise across introduced in dyplr 1.0.0
final %>% 
  group_by(ccode) %>%
  summarise(
    across(where(is.numeric), mean)
  )

# we will use the slider package to create periods every three years
final %>% 
  group_by(ccode) %>%
  mutate(period = paste0(
    slide_dbl(year, min, .step = 3L, .after = 2),
    "-",
    slide_dbl(year, max, .step = 3L, .after = 2)
    )) 

# this gives us nasty NA-NA interfalls every 2nd and 3rd step
# we get rid of this with a nasty hack and fill the values from top
# not really elegant but it works
final %>% 
  group_by(ccode) %>%
  mutate(period = paste0(
    slide_dbl(year, min, .step = 3L, .after = 2),
    "-",
    slide_dbl(year, max, .step = 3L, .after = 2)
  )) %>%
  mutate(period = ifelse(str_detect(period, "NA"), NA_character_, period)) # hack

# now combine everything
final %>% 
  group_by(ccode) %>%
  mutate(period = paste0(
    slide_dbl(year, min, .step = 3L, .after = 2),
    "-",
    slide_dbl(year, max, .step = 3L, .after = 2)
  )) %>%
  mutate(period = ifelse(str_detect(period, "NA"), NA_character_, period)) %>%
  fill(period) %>% # fill the values from top
  select(-year) %>%
  group_by(ccode, period) %>%
  summarise(
    across(where(is.numeric), mean)
  )