Empirical reporting-completion summary from a reporting triangle

Empirical reporting-completion summary from a reporting triangle

Usage

reporting_completion_v1(
  triangle,
  max_delay,
  delay_window = NULL,
  period = c("all", "year", "month")
)

Arguments

triangle

A `csfmt_reporting_triangle_v3`.

max_delay

Delay horizon in weeks.

delay_window

Optional: use only settled weeks within roughly this many weeks (drift-aware). `NULL` uses all settled weeks. Ignored for the shape of `period` stratification, which slices time itself.

period

Time stratification of the settled weeks, by the calendar year or month of each week's Thursday. Choose `"all"` (one pooled curve, default), `"year"`, or `"month"` (one row per period). Use `"year"` or `"month"` to see whether completion time is trending up or down.

Value

One row per series, and per period when stratified. The columns are identity columns + `period` + `n_settled`, `mean_delay`, `complete_by_md`, and `pct_delay0`..`pct_delay<max_delay-1>`. Each `pct_delayD` is the pooled % of cases reported by the end of week reference + D – the delay ECDF, no interpolation. `pct_delay0` is the reference week itself, NOT the week after. Every one of these is computed AFTER delays `>= max_delay` are discarded. They describe the cases that arrive within the horizon, not all eventual cases.

complete_by_md is always 1

`complete_by_md` is the last cumulative fraction of a total that was itself summed over the truncated delay axis. So it equals 1 for every series and every period, and `pct_delay<max_delay-1>` equals 100. It does NOT measure whether reporting continues past `max_delay`. To look for a tail, re-run with a larger `max_delay` and compare `mean_delay` and the `pct_delayD` curve.

See also

vignette("pipeline", package = "csalert"), which runs this function on its synthetic triangle.

Other reporting completion functions: reporting_completion_trend_v1()

Examples

w <- cstime::dates_by_isoyearweek$isoyearweek
i <- match("2023-01", w)
set.seed(1)
d <- data.table::data.table(
  isoyearweek_reference = w[i + rep(0:39, each = 3)],
  isoyearweek_reporting = w[i + rep(0:39, each = 3) + rep(0:2, 40)],
  numerator = rpois(120, c(30, 15, 5)),
  indicator_tag = "x", location_code = "nation", age = "total", sex = "total"
)
d <- d[isoyearweek_reporting <= w[i + 39]]
tri <- csfmt_reporting_triangle_v3(
  d,
  id_cols = c("indicator_tag", "location_code", "age", "sex")
)

# one pooled curve: pct_delay0 is the share in during the reference week itself
reporting_completion_v1(tri, max_delay = 3)
#>    indicator_tag location_code    age    sex period n_settled mean_delay
#>           <char>        <char> <char> <char> <char>     <int>      <num>
#> 1:             x        nation  total  total    all        38       0.51
#>    complete_by_md pct_delay0 pct_delay1 pct_delay2
#>             <num>      <num>      <num>      <num>
#> 1:              1       58.8       89.9        100

# sliced by month, to expose drift in how fast reporting arrives
head(reporting_completion_v1(tri, max_delay = 3, period = "month"), 3)
#>    indicator_tag location_code    age    sex  period n_settled mean_delay
#>           <char>        <char> <char> <char>  <char>     <int>      <num>
#> 1:             x        nation  total  total 2023-01         4       0.50
#> 2:             x        nation  total  total 2023-02         4       0.50
#> 3:             x        nation  total  total 2023-03         5       0.53
#>    complete_by_md pct_delay0 pct_delay1 pct_delay2
#>             <num>      <num>      <num>      <num>
#> 1:              1       59.3       91.0        100
#> 2:              1       60.2       90.0        100
#> 3:              1       58.5       88.4        100