# Does a fast retraction bite as hard as a slow one? — results

Computed by `analyse.py` from `~/data/retraction-citations-fast/`. The rule was sealed as `rule.retraction-citations-fast` at 2026-09-20T19:52Z, before any of this population was fetched and before this file existed.

## The sample

- retracted papers resolved: **296** (the rule's floor for a verdict is 150)
- control papers (same journal, same publication year): **862**
- **retracted papers with an EMPTY pre-window, dropped by the rule: 84 of 296 (28.4 %)**
- control papers likewise dropped: **263 of 862 (30.5 %)**

The rule names that drop as the bias this population carries and asks for it prominently: a paper pulled two or three years after publication has one or two years in which to be cited before the notice lands, so the papers that vanish here are the least-cited ones — plausibly the ones a retraction would have silenced most completely. The direction of that bias is toward RR = 1, against the result I expected.

- retracted papers by pre-window length: **1 year: 111**, **2 years: 101**

## The controls on the detector

**Null control A, composition-preserving** — arm labels shuffled inside each stratum, 200 draws: mean RR **0.9997** (range 0.736 to 1.292), band 0.90 to 1.11: **PASS**.

**Null control B, composition-breaking** — labels shuffled across the whole sample and the strata rebuilt, 200 draws: mean RR **0.9975** (range 0.716 to 1.284), band 0.90 to 1.11: **PASS**.

**Positive control** — half the control arm thinned by 50 %, over 35 strata: RR **0.4344** (95 % CI 0.387 to 0.499), band 0.42 to 0.58: **PASS**.

**Floor.** The positive control's interval is 0.112 wide, so this sample cannot distinguish an effect smaller than about 6 % from none at all.

## The estimate

Strata used: **36** of 45, holding **81.6 %** of the retracted pre-window citation mass; the rest sat in strata with no control arm and was dropped.

**RR = 0.3170** (95 % CI 0.277 to 0.364).

### Primary verdict

**KILLED.** A fast retraction at least halves the citation rate, as a slow one does.

### Secondary verdict, declaredly the weaker of the two

RR_slow is held FIXED at the published **0.4345** and is not re-estimated, so the uncertainty below is the fast study's alone. D = RR_fast / RR_slow = **0.7296**.

**DIFFERENT.** The fast study's interval (0.277 to 0.364) excludes the slow study's point estimate.

## The plain counts, which carry no verdict

- fast-retracted papers collecting at least one citation in the three years after the retraction year: **209 of 296 (70.6 %)**
- median post-window citations, retracted: **2**; controls: **5**
- total citations to these 296 fast-retracted papers in the three years after their retraction: **1601**
- of all citations to the retracted papers, OpenCitations flags **126** as author self-citations (3.5 %).

## Per stratum

| publication year | pre-window | pre-window citations | retracted | controls | post/pre retracted | post/pre control |
|---|---|---|---|---|---|---|
| 2015 | 2 years | 1-2 | 2 | 16 | 0.500 | 5.182 |
| 2015 | 2 years | 11-20 | 3 | 8 | 0.692 | 1.581 |
| 2015 | 2 years | 3-5 | 2 | 10 | 0.429 | 2.600 |
| 2015 | 2 years | 6-10 | 5 | 8 | 1.024 | 1.862 |
| 2016 | 1 year | 1-2 | 10 | 19 | 1.643 | 11.292 |
| 2016 | 1 year | 3-5 | 4 | 13 | 1.692 | 6.574 |
| 2016 | 1 year | 6-10 | 2 | 6 | 1.941 | 2.389 |
| 2016 | 2 years | 1-2 | 5 | 18 | 3.167 | 2.630 |
| 2016 | 2 years | 11-20 | 3 | 4 | 0.511 | 2.083 |
| 2016 | 2 years | 3-5 | 2 | 10 | 0.250 | 2.000 |
| 2016 | 2 years | 6-10 | 4 | 10 | 1.138 | 1.816 |
| 2017 | 1 year | 1-2 | 7 | 17 | 1.556 | 4.565 |
| 2017 | 1 year | 11-20 | 1 | 3 | 0.471 | 3.143 |
| 2017 | 1 year | 3-5 | 8 | 15 | 1.276 | 4.491 |
| 2017 | 1 year | 6-10 | 2 | 6 | 2.133 | 3.889 |
| 2017 | 2 years | 1-2 | 5 | 17 | 0.143 | 4.222 |
| 2017 | 2 years | 11-20 | 5 | 8 | 0.846 | 2.035 |
| 2017 | 2 years | 21-50 | 2 | 3 | 1.062 | 4.212 |
| 2017 | 2 years | 3-5 | 3 | 13 | 0.750 | 2.418 |
| 2017 | 2 years | 6-10 | 5 | 12 | 0.854 | 2.765 |
| 2018 | 1 year | 1-2 | 11 | 46 | 3.714 | 5.576 |
| 2018 | 1 year | 11-20 | 1 | 3 | 0.850 | 5.341 |
| 2018 | 1 year | 21-50 | 1 | 1 | 0.441 | 3.926 |
| 2018 | 1 year | 3-5 | 7 | 23 | 1.222 | 4.878 |
| 2018 | 1 year | 6-10 | 6 | 10 | 1.356 | 5.000 |
| 2018 | 2 years | 1-2 | 10 | 32 | 0.667 | 2.814 |
| 2018 | 2 years | 11-20 | 11 | 20 | 0.625 | 1.662 |
| 2018 | 2 years | 21-50 | 6 | 10 | 0.510 | 1.487 |
| 2018 | 2 years | 3-5 | 11 | 42 | 0.674 | 1.655 |
| 2018 | 2 years | 51-200 | 1 | 2 | 0.213 | 2.967 |
| 2018 | 2 years | 6-10 | 15 | 40 | 0.788 | 1.404 |
| 2019 | 1 year | 1-2 | 16 | 65 | 2.200 | 5.000 |
| 2019 | 1 year | 11-20 | 5 | 5 | 0.757 | 2.734 |
| 2019 | 1 year | 21-50 | 1 | 2 | 0.455 | 3.954 |
| 2019 | 1 year | 3-5 | 18 | 53 | 1.276 | 3.510 |
| 2019 | 1 year | 6-10 | 7 | 22 | 0.860 | 3.124 |