Table 1

Mean performance of methods used to estimate the effect of training load on injury risk (n simulations=1900).

RelationshipMethodExternal RMSE*Internal RMSEAICCoverage (%)AWCoverage MCSE
Absolute training load
ConstantRolling average4.850.1135471422.9234.75.174780.90
EWMA4.770.1135481423.4236.35.171790.91
REDI5.530.1135571424.1020.33.401140.74
DLNM1.440.1124341317.1534.82.056000.95
DecayRolling average5.380.1135901421.8030.25.169300.87
EWMA5.170.1135871421.8531.85.125540.88
REDI6.210.1136051423.8018.73.421540.71
DLNM1.550.1122451295.3032.42.079770.93
Exponential decayRolling average2.130.1135991424.6585.05.546950.58
EWMA1.880.1135881423.8685.15.371410.61
REDI1.970.1136031425.0074.23.692080.64
DLNM0.760.1133681407.0881.62.026330.65
Relative training load (%Δ)†
ConstantACWR0.1136431426.16
Week-to-week %Δ0.1136461426.40
DLNM %Δ0.1136271389.28
DecayACWR0.1136151424.73
Week-to-week %Δ0.1136171425.12
DLNM %Δ0.1135531383.52
Exponential decayACWR0.1135651423.33
Week-to-week %Δ0.1135661423.27
DLNM %Δ0.1137001401.39
  • *Monte Carlo SE for RMSE was <0.001 for all simulations. The scale of the RMSE depends on the scale of the coefficients, and it is therefore only interpretable by comparing values in the same analysis – the values cannot be interpreted in isolation.

  • †Due to differences in scale between methods and simulation for relative training load, external RMSE, coverage, and AW could not be calculated in a comparable manner.

  • ACWR, acute:chronic workload ratio; AIC, Akaike’s information criterion; AW, average width of 95% CIs; Coverage, coverage of 95% CIs; DLNM, distributed lag non-linear mode; EWMA, exponentially weighted moving average; MCSE, Monte Carlo Standard Error; REDI, robust exponential decreasing index; RMSE, root-mean-squared error.