Analysis

Attendance, and where it broke

2021-10-24 to 2026-10-18 — 261 weeks, the widest recent run complete enough to decompose.

Seasonality stripped, the trend runs from 155 to 176 a week.

1 break in that window

• 2022-11-13: 162 -> 178 (+16), over 55 weeks before and 206 after.

261 weeks, of which 7 (2.7%) were filled in by linear interpolation and are drawn differently. Those weeks are not records; nobody counted them.

Was 2020 a step, a dip, or an acceleration?

Unanswerable from the archive as it stands: the weeks the change would be visible in have not been scanned. The levels either side are real and are given below, but a step, a dip and an acceleration all fit them.

Before 2020-03-15: 206 a week over 39 recorded week(s), trending -54.3 a year. After: 164 over 70, trending +14.7 a year.

Caveat: the pre-break side spans 1.4 year(s) of the 3 asked for, so its slope is not carried forward: a trend fitted to that and extrapolated is not a prediction

Caveat: 0 recorded week(s) in the 26 weeks from 2020-03-15, so the shape of the change was never written down. 2019 and 2020 are entirely unscanned; see `pewcount scan-status`.

2019 and 2020 hold no records at all, so the 26 weeks after the break -- where a dip separates from a step -- were never written down. Scanning either year would answer this; it is the single most valuable thing a session could unlock.

Do the children stay?

Aligned to Move Up Sunday, not to January. 4 boundaries on record. No Move Up Sunday recorded for 2025.

K-1 → 2-3

• Of the K-1 class in 2024, 89% as many children were in 2-3 2 years later (44-133% at 95% confidence), from 5 recorded Sunday(s) -- too few to read much into. One of the two cohort-years has no recorded closing Move Up Sunday and was closed 52 weeks after its opening one. thin inferred boundary

2-3 → 4-5

• Of the 2-3 class in 2024, 108% as many children were in 4-5 2 years later (52-164% at 95% confidence), from 5 recorded Sunday(s) -- too few to read much into. One of the two cohort-years has no recorded closing Move Up Sunday and was closed 52 weeks after its opening one. thin inferred boundary

Modern age-band cards only. A legacy room count cannot be resolved to a grade and is never substituted — see docs/questions/room-grade-crosswalk.md.

Giving, in comparable dollars

real, 2026 dollars · CPI covers 1913–2026 · 53 of 53 years of giving can be deflated

Giving per attender, 1970-12-27 to 2026-10-18: $36.82 → $38.80 in 2026 dollars. In the dollars of the day it reads $4.31 → $38.80 — the gap between those two pairs is the inflation.

Next year, for budget season

Weekly giving over the next 52 week(s): about $6,392.52, plausibly $368.95 to $16,129.60 at 95% confidence. The budget of $6,314.43 is above the floor of that range: the central estimate clears it, but a shortfall is within the range the archive's own errors allow.

Over the next 52 week(s): about 174 a week, 9,061 in total, and plausibly between 5,782 and 12,340 at 95% confidence. Drift over the last year: -2.5.

learnt from the last 5 year(s) only; the archive reaches back 56 year(s), and errors from a congregation twice this size are not evidence about next year

empirical quantiles were not monotonic in horizon; the band has been widened to be non-decreasing

only 0 whole-horizon backtest error(s); the total's interval is the sum of the weekly ones, which is wider than the total's own spread warrants

45 week(s) had a lower bound below zero, which this series cannot be; those bounds are floored at zero