Australian-births.R

   1 # Analyse births in Australia
   2 library(tidyverse)
   3 library(ggthemes)
   4 library(forecast)
   5 AU_births <- read_csv("data/BIRTHS_MONTH_OCCURRENCE_17102017230358770.csv")
   6 # Exclude 2015 as data seems incomplete
   7 AU_births <- filter(AU_births, !(TIME == 2015))
   8 # Remove subtotals in rows
   9 AU_births <- filter(AU_births, MONTH_OCCURENCE != "TOT")
  10 # Make it a number as it was meant to be
  11 AU_births$MONTH_OCCURENCE <- as.numeric(AU_births$MONTH_OCCURENCE)
  12 # Add median by month to df
  13 AU_births_medians <- AU_births %>%
  14   group_by(Region, MONTH_OCCURENCE) %>%
  15   summarise(median = median(Value))
  16 AU_births <- merge(AU_births, AU_births_medians)
  17
  18 # Create a seasonal subseries plot
  19 ggplot(filter(AU_births, Region == "Australia"), aes(x= Time, y = Value)) +
  20   geom_line() +
  21   geom_line(aes(y = median), color = "red") +
  22   facet_grid( ~ MONTH_OCCURENCE) +
  23   labs(x = paste("\nFrom", min(AU_births$Time), "to",max(AU_births$Time)),
  24        caption = "Data: Australian Bureau of Statistics",
  25        y = "Births per month",
  26        title = "Births in Australia",
  27        subtitle = "Bottoms out in summer months; boomed from 2005/2007") +
  28   theme_economist() +
  29   theme(axis.text.x = element_blank())
  30
  31 # Let's check the seasonality
  32 # Uses some base R graphics subtleties to right align the caption...
  33 AU_births_ts <- as.ts(arrange(filter(AU_births, Region == "Australia"), Time, MONTH_OCCURENCE)$Value)
  34 Acf(AU_births_ts, lag.max = 48, main = "")
  35     title(main = "ACF plot of births in Australia by month")
  36     title(sub = "Data: Australian Bureau of Statistics", adj = 1)
  37
  38 # The acf peaks at multiples of 12, showing a yearly seasonality
  39
  40 # Seasonality doesn't appear to increase over time
  41 #plot(AU_births_ts)
  42 #trend_AU_births <- ma(AU_births_ts, order = 12, centre = T)
  43 #plot(trend_AU_births)
  44 #detrend_AU_births <- AU_births_ts - trend_AU_births
  45
  46 AU_births_decomposed <- decompose(ts(AU_births_ts, frequency = 12))
  47 years <- rep(c(min(AU_births$TIME):max(AU_births$TIME)), each = 12)
  48
  49 # For ggplot, we prefer a long dataframe over a list of timeseries
  50 AU_births_decomposed_df <- as.tibble(cbind(observation = c(1:length(years)),
  51                                  years,
  52                                  observed = as.vector(AU_births_ts),
  53                                  seasonal = as.vector(AU_births_decomposed$seasonal),
  54                                  random = as.vector(AU_births_decomposed$random),
  55                                  trend = as.vector(AU_births_decomposed_trend)))
  56 # Add a character vector only after the above cbind, to avoid everything becoming a chr or a factor
  57 AU_births_decomposed_df <- mutate(AU_births_decomposed_df,
  58                                   seastrend = seasonal + trend,
  59                                   month = rep(c("01","02","03","04","05","06","07","08","09","10","11","12"), times = 19))
  60 AU_births_decomposed_df <- gather(AU_births_decomposed_df, key = component, value = value, -observation, -years, - month)
  61
  62 # Plot the various components of the observed births underneath each other
  63 ggplot(filter(AU_births_decomposed_df, component != "seastrend"), aes(x = as.factor(observation), y = as.numeric(value), group  = 1)) +
  64   geom_line() +
  65   facet_grid(component ~ ., scales = "free_y") +
  66   # Hackish way of showing only the year
  67   scale_x_discrete(breaks = filter(AU_births_decomposed_df, month == "01")$observation,
  68                    labels = filter(AU_births_decomposed_df, month == "01")$years) +
  69   theme_fivethirtyeight() +
  70   theme(axis.text.x = element_text(angle = 45)) +
  71   labs(title = "Births in Australia",
  72        subtitle = "Seasonally decomposed",
  73        caption = "Data: Australian Bureau of Statistics",
  74        x = "",
  75        y = "")
  76
  77 # Let's look at everything in one plot instead
  78 ggplot(AU_births_decomposed_df, aes(x = yearmonth, y = as.numeric(value), group  = 1)) +
  79   geom_line(data = dplyr::filter(AU_births_decomposed_df, component =="trend"), color = "orange", size = 2) +
  80   geom_line(data = dplyr::filter(AU_births_decomposed_df, component =="observed"), color = "grey50") +
  81   geom_line(data = dplyr::filter(AU_births_decomposed_df, component =="seastrend"), color = "black") +
  82   scale_x_discrete(breaks = filter(AU_births_decomposed_df, month == "01")$yearmonth,
  83                    labels = filter(AU_births_decomposed_df, month == "01")$years) +
  84   theme_economist_white() +
  85   theme(axis.text.x = element_text(angle = 45)) +
  86   labs(title = "Births in Australia",
  87        subtitle = "Seasonally decomposed",
  88        caption = "Data: Australian Bureau of Statistics",
  89        x = "",
  90        y = "")
  91
  92 # TODO Horizon plot