Index numbers from extracted trends

Turns a trend on the transformed scale into an index equal to scale on average over a base period, separately for each key.

Usage

trend_index(
  tr,
  key_cols,
  time_col,
  base_period,
  value_col = "trend",
  transform = c("log", "logit", "none"),
  scale = 100
)

Arguments

tr

A data.frame or data.table of trends (e.g. from extract_trend()).

key_cols

Character vector of key columns.

time_col

Name of the time column (class Date or IDate).

base_period

Either two dates (start and end, inclusive) or the name of a logical column of tr marking the base rows.

value_col

Name of the trend column (default "trend").

transform

Scale of value_col: "log", "logit" or "none".

scale

Value of the index in the base period (default 100).

Value

A data.table with columns key_cols, time_col, value_col and indice, ordered by key and time.

Details

• log: indice = scale * exp(trend - mean(trend in base)). The base is geometric: the mean of log(indice) over the base period is exactly log(scale), so the geometric mean of the index in the base is scale. A geometric base keeps indices multiplicative, which makes the decomposition total x share in compose_skill_index() exact.

• logit: the trend is back-transformed to a share p = plogis(trend) and indice = scale * p / mean(p in base): the arithmetic mean of the index in the base is scale.

• none: indice = scale * trend / mean(trend in base); keys whose base mean is not positive get NA.

Keys without non-missing trend values in the base period get NA and a warning.

Examples

weeks <- data.table::as.IDate("2024-01-01") + 7L * (0:103)
tr <- data.table::data.table(
  source = "a", week = weeks, trend = 5 + 0.002 * seq_along(weeks)
)
idx <- trend_index(tr, "source", "week",
  base_period = c("2024-01-01", "2024-12-31")
)
in_base <- idx$week <= as.Date("2024-12-31")
exp(mean(log(idx$indice[in_base])))
#> [1] 100