11.2 Longitudinal Load Modeling: ACWR, EWMA, Monotony, Strain, and Change Detection

11.2 Longitudinal Load Modeling: ACWR, EWMA, Monotony, Strain, and Change Detection

Quick Answer: Evaluating weekly workload totals in isolation hides how load is distributed across the microcycle and how current stress compares to historical conditioning. Foster's Training Monotony ($\text{Mean Daily Load} / \text{7-Day SD}$) and Training Strain ($\text{Weekly Load} \times \text{Monotony}$) penalize unvaried schedules that lack recovery troughs. For multi-week progression, Uncoupled Rolling Averages avoid the mathematical coupling of traditional $7:28$ ratios, while Exponentially Weighted Moving Averages ($\text{EWMA}$, $\lambda_a = 2/[N+1]$) model non-linear physiological decay. However, because workload ratios suffer from scaling artifacts, practitioners should pair longitudinal load tracking with Statistical Process Control ($\text{SPC}$) and rolling intra-individual $Z\text{-scores}$.

---

Foster's Training Monotony and Training Strain: Quantifying Microcycle Distribution

Developed by Carl Foster (1998), the Session Rating of Perceived Exertion ($\text{sRPE}$) method quantifies daily internal training load in Arbitrary Units ($\text{AU}$) by multiplying the athlete's post-session CR-10 RPE score by session duration in minutes ($\text{sRPE Load} = \text{RPE} \times \text{Minutes}$). When an athlete completes multiple sessions in one day, daily load is the sum of those sessions, and Weekly Training Load is the 7-day total ($\sum_{i=1}^{7} \text{Load}_i$).

However, two microcycles with the exact same Weekly Training Load ($2{,}800\text{ AU}$) can induce vastly different physiological adaptations depending on day-to-day variability:

$$\text{Training Monotony} = \frac{\text{Mean Daily Load (7-Day Average)}}{\text{Standard Deviation of Daily Load (7-Day SD)}}$$ $$\text{Training Strain} = \text{Weekly Total Training Load} \times \text{Training Monotony}$$

Why Identical Moderate Days Spike Monotony and Strain

• The Mathematical Mechanism: Because the 7-day Standard Deviation ($\text{SD}$) sits in the denominator of the Training Monotony equation, prescribing nearly identical moderate workouts every day causes the day-to-day $\text{SD}$ to approach zero. Dividing the mean daily load by a tiny $\text{SD}$ drives Training Monotony above the commonly cited risk threshold of $2.0$ (Monotony is a unitless ratio).
• The Undulating Solution: Alternating high-intensity/high-volume acquisition days (MD-4, MD-3, and Match Day MD) with true off days ($0\text{ AU}$) and low-load regeneration/activation days (MD+1, MD-1) creates a large 7-day $\text{SD}$, keeping Training Monotony low ($< 1.5$) even during high-volume blocks.
• Physiological Consequence of High Strain: When high Weekly Total Load combines with high Monotony ($> 2.0$), Training Strain spikes dramatically. Prolonged high Training Strain blunts autonomic heart rate variability ($\text{lnRMSSD}$), suppresses mucosal immunity (salivary secretory immunoglobulin A, $\text{sIgA}$), increases upper respiratory tract infection ($\text{URTI}$) incidence, and precipitates non-functional overreaching ($\text{NFOR}$) and overuse tissue pathology.

---

Acute:Chronic Workload Ratio (ACWR): Rolling Averages, EWMA, and Methodological Critiques

Inspired by Banister's two-factor impulse-response model—where performance equals Fitness (long-term chronic load) minus Fatigue (short-term acute load)—the Acute:Chronic Workload Ratio ($\text{ACWR}$) compares an athlete's acute workload (typically $7\text{ days}$) against their chronic workload (typically $28\text{ days}$).

1. Coupled vs. Uncoupled Rolling Average ACWR

• Coupled Rolling $\text{ACWR}$: Divides the current 7-day acute load ($W_4$) by the 4-week rolling average that includes the current week in the denominator:

$$\text{Coupled ACWR} = \frac{W_4}{0.25 \times (W_1 + W_2 + W_3 + W_4)}$$

As demonstrated by Lolli et al. and Impellizzeri et al., including $W_4$ in both the numerator and denominator creates mathematical coupling: the numerator and denominator are intrinsically correlated even when generated from random numbers, and the ratio is mathematically capped at a maximum possible value of $4.0$ (which occurs when $W_1 = W_2 = W_3 = 0$).

• Uncoupled Rolling $\text{ACWR}$: Removes the acute week from the chronic denominator, dividing current acute load ($W_4$) strictly by the average of the preceding $3$ (or $4$) independent weeks:

$$\text{Uncoupled ACWR} = \frac{W_4}{\frac{1}{3}(W_1 + W_2 + W_3)}$$

2. Exponentially Weighted Moving Average (EWMA, Williams et al., 2017)

A fundamental physiological flaw of any simple rolling average (coupled or uncoupled) is equal time-weighting: a 28-day rolling average assigns the exact same weight ($1/28 \approx 3.57\%$) to a training session performed $27\text{ days}$ ago as it does to a grueling match played yesterday, and then abruptly drops Day 1's physiological stimulus to zero on Day 29.

To reflect the continuous exponential decay of physiological fitness and fatigue, Williams et al. (2017) introduced the Exponentially Weighted Moving Average ($\text{EWMA}$):

$$\text{EWMA}_{\text{today}} = \text{Load}_{\text{today}} \times \lambda_a + (1 - \lambda_a) \times \text{EWMA}_{\text{yesterday}} \quad \text{where } \lambda_a = \frac{2}{N + 1}$$

• Acute Decay Constant ($N = 7\text{ days}$):

$$\lambda_{\text{acute}} = \frac{2}{7 + 1} = \frac{2}{8} = 0.250$$

Today's workload receives a $25.0\%$ weight, while yesterday's acute $\text{EWMA}$ retains a $75.0\%$ weight ($1 - 0.25 = 0.75$).

• Chronic Decay Constant ($N = 28\text{ days}$):

$$\lambda_{\text{chronic}} = \frac{2}{28 + 1} = \frac{2}{29} \approx 0.069$$

Today's workload receives a $6.9\%$ weight, while yesterday's chronic $\text{EWMA}$ retains a $93.1\%$ weight. Because $\text{EWMA}$ weights recent sessions more heavily, it exhibits significantly higher sensitivity to acute workload spikes than rolling averages.

3. Impellizzeri et al. Methodological Critiques of ACWR

While early observational papers popularized an $\text{ACWR}$ "$0.80\text{–}1.30$ sweet spot" and a "$> 1.50$ danger zone," Impellizzeri et al. (2020, 2021) established why practitioners must never treat $\text{ACWR}$ thresholds as deterministic injury predictors:

1. Spurious Correlation from Coupling: Coupled ratios manufacture artificial statistical associations between the ratio and injury incidence.
2. Ratio Scaling Artifacts (Failure of Isometric Scaling): Ratios assume the numerator scales proportionally with the denominator across the entire range of chronic loads. In reality, an athlete returning from rehab with a very low chronic load ($100\text{ m}$ of sprint distance) who runs $180\text{ m}$ records a terrifying $\text{ACWR} = 1.80$ despite trivial absolute tissue stress, whereas an elite winger with a huge chronic base ($1{,}000\text{ m}$) who spikes to $1{,}350\text{ m}$ ($+350\text{ m}$ absolute spike) records a benign-looking $\text{ACWR} = 1.35$.
3. Discretization Loss of Information: Slicing continuous workload metrics into arbitrary categorical buckets ($< 0.80, 0.80\text{–}1.30, > 1.50$) discards statistical power and ignores baseline tissue capacity (eccentric strength, aerobic fitness, prior injury history).

---

Complete Worked 7-Day Calculation Example: Monotony, Strain, ACWR, and EWMA

Part A: Comparing Two 7-Day Microcycles with Identical Weekly Load ($2{,}800\text{ AU}$)

Part B: Worked Coupled vs. Uncoupled ACWR

Suppose an athlete's weekly loads across 4 consecutive weeks are $W_1 = 2{,}000\text{ AU}$, $W_2 = 2{,}200\text{ AU}$, $W_3 = 2{,}400\text{ AU}$, and current acute week $W_4 = 3{,}400\text{ AU}$:

• Coupled 4-Week Chronic Average: $\frac{2{,}000 + 2{,}200 + 2{,}400 + 3{,}400}{4} = \frac{10{,}000}{4} = 2{,}500\text{ AU} \rightarrow \text{Coupled ACWR} = \frac{3{,}400}{2{,}500} = 1.36$.
• Uncoupled 3-Week Preceding Average: $\frac{2{,}000 + 2{,}200 + 2{,}400}{3} = \frac{6{,}600}{3} = 2{,}200\text{ AU} \rightarrow \text{Uncoupled ACWR} = \frac{3{,}400}{2{,}200} = 1.55$.

Notice how the coupled formula ($1.36$) masked a $55\%$ workload surge above the preceding 3-week baseline ($1.55$) because the acute spike inflated its own denominator!

Part C: Worked 1-Step EWMA Update

Assume yesterday's acute $\text{EWMA}$ ($N=7, \lambda_a = 0.25$) was $400\text{ AU}$ and yesterday's chronic $\text{EWMA}$ ($N=28, \lambda_c = 0.069$) was $380\text{ AU}$. Today the athlete completes Match Day (MD) with $\text{Load}_{\text{today}} = 800\text{ AU}$:

• $\text{EWMA}_{\text{acute, today}} = (800 \times 0.25) + (400 \times 0.75) = 200 + 300 = 500.0\text{ AU}$.
• $\text{EWMA}_{\text{chronic, today}} = (800 \times 0.069) + (380 \times 0.931) = 55.20 + 353.78 = 408.98\text{ AU}$.
• $\text{EWMA ACWR}_{\text{today}} = \frac{500.0}{408.98} \approx 1.22$.

---

Statistical Process Control (SPC) and Individual Time-Series Change Detection

To monitor individual athlete time series without relying on flawed group ratios, sport scientists deploy industrial Statistical Process Control ($\text{SPC}$) and rolling intra-individual standardization:

1. Shewhart Individual Moving-Range ($I\text{-}MR$) Control Charts: Plots an individual athlete's daily metric (e.g., morning $\text{lnRMSSD}$ or $\text{CMJ}$ flight-time:contraction-time ratio) around their healthy rolling baseline mean ($\mu_{\text{indiv}}$) with Upper and Lower Control Limits ($\pm 2\text{ SD}$ warning limits and $\pm 3\text{ SD}$ action limits, estimated from the average moving range $\overline{MR} / d_2$). Practitioners apply Nelson / Western Electric run rules—flagging not only a single extreme point beyond control limits, but also $7\text{–}8$ consecutive points falling on the same side of the mean line (which detects a systematic baseline shift).
2. Cumulative Sum ($\text{CUSUM}$) Charts: Calculates the cumulative sum of deviations from an athlete's target baseline mean ($S_t = \max[0, S_{t-1} + (X_t - \mu_0 - k)]$). Because $\text{CUSUM}$ accumulates small, persistent daily deficits (such as a $-0.6\text{ SD}$ suppression sustained for 6 straight days), it detects insidious non-functional overreaching far earlier than single-day threshold alerts.
3. Rolling Intra-Individual $Z\text{-Scores}$ ($Z_{\text{indiv}} = [X_{\text{today}} - \mu_{28\text{d}}] / \sigma_{28\text{d}}$): Standardizes today's readiness score against the athlete's own 28-day rolling mean and $\text{SD}$, overlaid with Smallest Worthwhile Change ($\text{SWC} = 0.2 \times \text{SD}$) and Typical Error ($\pm \text{TE}$) bands to confirm that a flagged shift exceeds daily measurement noise.