Working with time series across prefectures • estatr

e-Stat’s strength is long, consistent time series across all 47 prefectures. This article shows a typical workflow: pull a series, filter by time and area, and plot it.

library(estatr)
library(dplyr)
library(ggplot2)

Filtering server-side

Filter before the data leaves e-Stat by passing cd* codes to get_estat(). This is far cheaper than pulling everything and filtering in R:

# One category, a time window, all areas
d <- get_estat(
  "0003217721",
  cdCat03 = "1",            # e.g. a single sex category
  cdTimeFrom = "2015000000",
  cdTimeTo = "2023000000"
)

cdTime, cdArea, and cdCat01..cdCat15 all accept vectors, which are comma-joined into e-Stat’s code-list form for you:

# Just Tokyo and Osaka (whole-prefecture area codes)
get_estat("0003217721", cdArea = c("13000", "27000"))

The bundled prefectures table has the codes:

prefectures |>
  filter(name_en %in% c("Tokyo", "Osaka")) |>
  select(name_en, area_code)

Parsing the time axis

e-Stat time codes are numeric (time_code) with a human label (time). For plotting, derive a proper date/period from whichever is convenient. Quarterly labour-force codes, for example, encode the year in the first four digits:

d <- d |>
  mutate(year = as.integer(substr(time_code, 1, 4)))

Plotting

Because get_estat() returns a tidy tibble with a numeric value, it drops straight into ggplot2:

d |>
  filter(!is.na(value)) |>
  ggplot(aes(year, value, colour = cat01)) +
  geom_line() +
  labs(y = unique(d$unit), colour = NULL)

Handling suppressed and annotated values

Non-numeric markers (suppressed cells, footnotes) are not silently coerced to NA — they are preserved in the annotation column, with the legend attached as an attribute. Inspect them before dropping:

d |> filter(!is.na(annotation)) |> count(annotation)
attr(d, "notes")