Rolling geometric-mean wait time between consecutive first-time authors, by source and popularity stratum

Aggregates author_interval_tbl()'s event-level intervals into a (popularity stratum x month) grid: each interval is binned by the calendar month of its event_time, and reported as a window-month trailing geometric mean of interval_days. A geometric (not arithmetic) mean is used because wait times between authors are heavily right-skewed - in a sparse stratum-month cell, a single repo that went quiet for years would otherwise dominate an arithmetic mean of just a handful of intervals. A handful of near-simultaneous arrivals (interval_days at or near 0) are floored at one minute before logging, since log(0) = -Inf would otherwise wreck that whole cell's geometric mean rather than just pulling it down.

Usage

author_interval_trend_tbl(
  issue_authors_tbl,
  repo_tbl,
  source_name,
  n_strata = 4L,
  window = 12L,
  date_start = as.Date("2015-01-01"),
  date_end = NULL
)

Arguments

issue_authors_tbl: As returned by fetch_issue_authors().
repo_tbl: As returned by build_repo_tbl().
source_name: One of repo_tbl$source ("pypi", "npm", "joss", "ropensci").
n_strata: Number of popularity strata.
window: Trailing aggregation window, in months. Default 12: each reported month's n_metric/n_repo_months sum that month and the preceding 11.
date_start, date_end: Date bounds on the analysis window; date_end defaults to the start of the current month.

Value

A tibble with one row per (popularity stratum, month): popularity_stratum, month, n_events (trailing sum of author arrivals contributing an interval that month), geo_mean_days (the window-month trailing geometric mean of interval_days, NA where n_events is 0). Carries window and source_name as attributes.

Examples

if (FALSE) { # \dontrun{
it <- author_interval_trend_tbl (issue_authors_tbl, repo_tbl, "pypi")
} # }