---
title: "Half of a Trained Skill's Gains Are Gone in 6.5 Months. Here's How to Set Re-Assessment Intervals."
description: "Accuracy-based skills decay at about 0.08 standard deviations a month of nonuse. A step-by-step method for deriving a re-assessment interval from the decay rate and the cost of being wrong, with a ten-point checklist."
canonical: https://www.assessall.com/blog/half-of-a-trained-skills-gains-are-gone-in-6-5-months-heres-how-to-set-re
published: 2026-11-08
updated: 2026-11-08
author: "AssessAll Editorial"
source: AssessAll
---

# Half of a Trained Skill's Gains Are Gone in 6.5 Months. Here's How to Set Re-Assessment Intervals.

Accuracy-based skills decay at about 0.08 standard deviations a month of nonuse. A step-by-step method for deriving a re-assessment interval from the decay rate and the cost of being wrong, with a ten-point checklist.

A re-assessment interval is the planned gap between certifying someone on a skill and testing that skill again. It should be derived from the skill's decay rate, not from the training calendar. For accuracy-dependent procedural skills, meta-analytic estimates put the loss at roughly 0.08 standard deviations per month of nonuse — about half the original gain gone in six and a half months.

Most organisations do not set that interval — they inherit it. Annual refreshers exist because the financial year is annual; two-year recertification cycles exist because an accreditation body said two years decades ago. Neither number came from how fast the skill degrades, and that gap is where capability quietly disappears from a workforce still showing as certified in the LMS. Here is how to set it deliberately.

## What the decay evidence actually says

Two meta-analyses carry most of the weight. The older is Arthur, Bennett, Stanush and McNelly's 1998 review in *Human Performance*, which pooled 189 independent data points from 53 articles. Skill loss ran from an effect size of −0.01 immediately after training to **−1.4 after more than 365 days of nonuse** ([Arthur et al., 1998](https://doi.org/10.1207/s15327043hup1101_3)). A year of disuse costs more than the gap most selection tests are asked to detect between a strong and a weak candidate.

The newer one is more useful operationally because it models time continuously. Tatel's 2025 review in *Psychological Bulletin* drew on 1,344 effect sizes from 457 reports and estimated decay as a monthly slope: standardised mean differences grew stronger by **0.08 per month for accuracy-based performance measures** and 0.06 per month for speed-based and mixed measures. In planning terms, half of initial skill-acquisition gains were lost after approximately **6.5 months for accuracy-based measures, 13 months for speed-based and 11 months for mixed** ([Tatel, 2025, *Psychological Bulletin* 151(6), 696–736](https://pubmed.ncbi.nlm.nih.gov/40455501/)).

Two scope limits matter. Tatel's review covers procedural and motor skill retention, so it travels well to operational tasks and poorly to declarative knowledge. And Arthur and colleagues found decay is not uniform: physical, natural and speed-based tasks were **less** susceptible than cognitive, artificial and accuracy-based ones.

## Why the calendar is the wrong input

Resuscitation training is the clearest published case of an interval outliving its evidence. The American Heart Association's 2025 resuscitation education guidance notes that traditional courses are built around retraining "every 1 to 2 years" — while also stating that "CPR skills that are acquired immediately after BLS training often show decay by as early as 3 months" ([AHA, *Circulation* 2025;152(Suppl 2):S719–S750](https://www.ahajournals.org/doi/10.1161/CIR.0000000000001374)).

A three-month decay signal inside a twenty-four-month cycle means that for most of the cycle, the certificate records something that was once true. The same guidance lists "optimal training frequency for specific provider groups" as an open knowledge gap — the honest position: the evidence that long intervals are too long is strong; the evidence for what replaces them is thin.

## The four inputs that set the interval

**Which curve the skill sits on.** Accuracy-dependent work — error-free procedure execution, quality checks, compliance judgement — decays fastest. Speed-dependent work decays more slowly. Classify before you pick a number.

**How often the skill is actually used.** Tatel identifies intermittent performance opportunities as a possible moderator of decay rates — the input most organisations already hold and never use. A procedure performed weekly is rehearsed continuously; the same procedure performed twice a year is not. Pull usage frequency from operational data, not the job description.

**The quality of original acquisition.** Arthur and colleagues concluded that "the single most important determinant of both skill and knowledge retention appears to be the amount or degree of overlearning." A skill trained to bare passing already sits close to your threshold on day one, so it crosses below sooner. Thin training forces short intervals.

**What an undetected loss costs.** This sets your tolerance band. If half a standard deviation of drift is an inconvenience, you can afford a long interval. If it is a safety event or a regulatory breach, you cannot.

## Doing the arithmetic

The method is crude on purpose: an interval derived from a published slope beats a precise-looking one derived from the fiscal calendar.

1. **Fix the decision threshold.** How far below their certified score can someone drift before you would want to know? Express it in standard deviations — 0.25, 0.5, 1.0.
2. **Pick the monthly slope.** 0.08 SD per month for accuracy-based skills, 0.06 for speed-based or mixed, as a starting estimate.
3. **Divide.** A 0.25 SD tolerance on an accuracy-based skill gives roughly three months; 0.5 SD about six; 1.0 SD about a year. Annual cycles are defensible where a full standard deviation of drift is survivable, and indefensible where it is not.
4. **Adjust for on-the-job rehearsal.** Lengthen where the skill is used frequently. Do not shorten below the point where you are measuring test familiarity rather than skill.
5. **Replace the estimate with your own data.** After two cycles you hold paired scores at a known interval. Regress score change on months elapsed and you have your own slope for that skill, which beats any published average.

## Boosters, not recertification events

What follows a re-assessment matters as much as its timing. The AHA's 2025 guidance carries a Class 1 (Level of Evidence B-R) recommendation — "It is recommended to implement booster sessions when utilizing a massed learning approach for resuscitation training" — plus a Class 2a recommendation that spaced learning is reasonable in place of massed. In the underlying studies, the proportion achieving excellent CPR at 12 months was highest among those receiving monthly boosters.

The same guidance records the catch honestly: increased booster frequency reduced attendance rates, with the highest attrition in the most frequent condition. A three-month interval half the workforce skips measures less than a six-month interval everyone completes.

## The re-assessment is itself an intervention

A second reason to shorten intervals has nothing to do with surveillance: retrieval practice *is* learning. In Roediger and Karpicke's experiment, students who repeatedly studied a passage recalled 83% of idea units after five minutes, against 71% for students who repeatedly took tests on it. One week later the ranking inverted — 61% for the tested group against 40% for the studied group. Across that week the study-only group forgot 52% of what it knew; the tested group forgot 14% ([Roediger & Karpicke, 2006, *Psychological Science* 17(3), 249–255](https://doi.org/10.1111/j.1467-9280.2006.01693.x)). Their conclusion: "Testing is a powerful means of improving learning, not just assessing it."

A quarterly ten-minute scenario check is a cheaper way to find decay than an annual recertification — and, on this evidence, a better way to prevent it.

## What a short interval does not fix

- **Non-comparable occasions.** Score change is interpretable only if the two assessments are equivalent in difficulty. Re-using an identical form invites memorisation; writing a fresh one invites drift. Equated forms from a maintained item bank are the requirement, and the *Standards for Educational and Psychological Testing* treat comparability across forms and occasions as a validity issue, not an administrative one ([AERA/APA/NCME Standards](https://www.apa.org/science/programs/testing/standards)).
- **Measurement error mistaken for decay.** A drop smaller than the assessment's standard error of measurement is not evidence of anything. Set the tolerance band outside that error, or you will retrain noise.

Short-cycle re-assessment is what pay-as-you-go pricing exists for: at AssessAll's ₹30 / US$0.50 per assessment with no seat licence, a quarterly check costs a fraction of an annual classroom recertification, and AI-graded scenarios let it be a judgement task rather than a recall quiz. Results carry into a Skill Passport, so a person's decay curve becomes visible rather than inferred.

## The checklist

1. Classify the skill as accuracy-based, speed-based or mixed.
2. Pull actual on-the-job usage frequency.
3. State the tolerance band in SD units, justified by the cost of an undetected miss.
4. Confirm the band is wider than the standard error of measurement.
5. Divide tolerance by the monthly slope for a starting interval.
6. Check that interval against attendance reality.
7. Build equated forms from an item bank so occasions are comparable.
8. Pair each re-assessment with a short booster, not a full recertification.
9. After two cycles, fit your own slope and retire the published estimate.
10. Record the rationale, so the next person inherits a reason, not a number.

**The takeaway:** a certification date tells you when someone was competent, not whether they still are, and the gap between those two things grows at a rate that has been measured. Set the interval from the decay rate and the cost of being wrong — then measure your own slope and stop borrowing anyone else's.

---

Source: https://www.assessall.com/blog/half-of-a-trained-skills-gains-are-gone-in-6-5-months-heres-how-to-set-re