Skip to contents

Validate triangle structure before building a development

Source: R/triangle.R
validate_triangle.Rd

Check that each (groups, cohort) cohort has a consecutive dev sequence within its observed range. Non-consecutive cohorts produce non-consecutive age-to-age links downstream (e.g., 14 -> 17 instead of 14 -> 15), which breaks summary.Link() key uniqueness and causes cartesian joins in fit_ratio().

This function inspects the raw data without modifying it. Use it before as_triangle() to decide whether to fix the data source, drop offending cohorts, or pass fill_gaps = TRUE to as_triangle().

Two checks are performed:

  1. Cohort dev-sequence gaps – for each (group, cohort), report missing dev values within the observed range.

  2. Row-level calendar consistency – when calendar is supplied, report rows where calendar < cohort. Such rows are logically impossible (claims cannot precede policy issue) and downstream they show up as negative dev_m, polluting cohort dev sequences.

Usage

validate_triangle(
  df,
  groups = NULL,
  cohort,
  calendar = NULL,
  dev = NULL,
  grain = "auto"
)

Arguments

df

A data.frame.

groups

Grouping variable(s).

cohort

A single cohort variable (raw column name).

calendar

Optional calendar period variable for row-level consistency check. When supplied, rows where calendar < cohort are flagged as invalid. Default NULL (skip this check).

dev

A single development-period variable (raw column name). Optional when calendar is supplied – dev is then derived from (cohort, calendar) at the resolved grain (same dispatch as as_triangle()).

grain

Grain string ("M" / "Q" / "H" / "Y") or "auto" (default) – used only when dev is derived from (cohort, calendar). Ignored when dev is supplied.

Value

A data.table of class "TriangleValidation" with one row per cohort containing gaps. Columns:

groups, cohort

Cohort identifier.

n_dev

Number of distinct observed dev values.

n_expected

max(dev) - min(dev) + 1 for that cohort.

missing

List column of missing dev values.

Returns a zero-row data.table when no gaps are found.

Row-level violations (when calendar is supplied and the check finds any) are attached as the "invalid_rows" attribute – a data.table with columns [groups, cohort, calendar, dev (if present), reason]. Use attr(out, "invalid_rows") or rely on print.TriangleValidation which displays both sections.

See also

as_triangle()