data-raw/05_impute-china-1952-gdpPercap.R

In YTLogos/gapminder: Data from Gapminder

#' ---
#' date: "`r format(Sys.Date())`"
#' output:
#'   html_document:
#'     keep_md: TRUE
#' ---

library(readr)
suppressPackageStartupMessages(library(dplyr))
library(tidyr)
library(ggplot2)

#' There is no data for China in 1952. I have always had an incredibly low-tech 
#' imputation. I put it in its own script at the suggestion of Hilmar Lapp.
#' <https://github.com/jennybc/gapminder/issues/6>

gap_dat_orig <- read_tsv("04_gap-merged.tsv")

#' See? No data for 1952.
(china <- gap_dat_orig %>%
   filter(country == "China"))

#' Why is this problem? Big picture, it's not a problem! But to teach 
#' visualization and data exploration, I wanted my final dataset to be extremely
#' clean and balanced. Ultimately, each country has data for 12 years: 1952, 
#' 1957, 1962, ..., 2007. And I didn't want to lose a large country like China. 
#' So I imputed the data in order to retain it in `gapminder`.
#' 
#' In the past,  I imputed the China data after filtering for the years 1952,
#' 1952, etc. so I must do that here as well.
china <- china %>% 
  filter(year %% 5 == 2)

#' What does the data look like?
china_tidy <- china %>%
  gather(key = "variable", value = "value",
         pop, lifeExp, gdpPercap)
ggplot(china_tidy, aes(x = year, y = value)) +
  facet_wrap(~ variable, scales="free_y") +
  geom_point() + geom_line() +
  scale_x_continuous(breaks = seq(1950, 2011, 15))

#' Begin extremely low, low tech imputation for 1952. I wouldn't necessarily do
#' it this way again, but I'm committed now to replicating what I did late at
#' night long ago.
#' 
#' Linear fit for GDP per capita up to 1982.
china_gdp_fit <- lm(gdpPercap ~ year, china, subset = year <= 1982)
summary(china_gdp_fit)
(china_gdp_1952 <- china_gdp_fit %>%
  predict(data.frame(year = 1952)) %>% 
  round(6))
## historically this has given: 400.4486 

#' Linear fit for population.
china_pop_fit <- lm(pop ~ year, china)
summary(china_pop_fit)
(china_pop_1952 <- china_pop_fit %>%
  predict(data.frame(year = 1952)) %>% 
  as.integer())
## historically this has given: 556263527

#' Pulling a number out of thin air for life expectancy, but no simple linear
#' fit was appropriate.
china_lifeExp_1952 <- 44

#' Append these values to the full data frame.
gap_dat_new <- rbind(gap_dat_orig,
                     data.frame(country = 'China', year = 1952,
                                pop = china_pop_1952, continent = 'Asia',
                                lifeExp = china_lifeExp_1952,
                                gdpPercap = china_gdp_1952))
gap_dat_new <- gap_dat_new %>%
  arrange(country, year)

#' Isolate the China data again for some plots.
china_tidy <- gap_dat_new %>%
  filter(country == "China") %>%
  gather(key = "variable", value = "value",
         pop, lifeExp, gdpPercap)
ggplot(china_tidy, aes(x = year, y = value)) +
  facet_wrap(~ variable, scales="free_y") +
  geom_point() + geom_line() +
  scale_x_continuous(breaks = seq(1950, 2011, 15))

#' Save for now.
write_tsv(gap_dat_new, "05_gap-merged-with-china-1952.tsv")

devtools::session_info()