Figure 1 illustrates a variety of possible scenarios when repeated measures are plotted against each other. A straight and narrow line indicates the correlation is close to 1 or -1, whilst a cloud of data points demonstrates no relation (i.e. ICC closer to 0).
3. Within-Participant Variation: Absolute Reliability
Absolute reliability looks into the random error between repeated measures (e1 and e2) for one participant. Since it is expressed in the original unit (e.g. kg or cm), or in a proportion of it, it is useful to gauge the precision of a measure [2, 3]. It is, therefore, most relevant to assess the reliability of measures that evaluate change (e.g. tracking strength) [2, 3].
Next, we will briefly present the most common methods to investigate the within-participant variation:
Standard Error of Measurement (SEM)
The SEM is an estimate of the absolute value of the typical deviation between the observed scores and the true score, which is assumed to be the mean of all measured values .
Coefficient of Variation (CV)
The CV is the ratio of the SEM to the mean; it expresses the spread of values around the mean as a percentage of it (e.g. the CV of a fatigue protocol is 10% of the total work performed).
Limits of Agreement (LOA)
The 95% LoA is a range where we can expect someone’s test-retest values to fall 95% of the time . It is based on testing and can be calculated as follows:
The 95 % LoA for specific running task = ±7%
1st test score = 10 min (600s)
Because of this, and providing no change has occurred, we can expect the retest time to be in the range of plus or minus 7% of the 1st test score (10 min or 600s).
[Mean -7 % (Mean) to Mean +7 % (Mean)].
[600 – (7*(600/100) = 558s to (7*(600/100)) = 642s)]
As a result, we can expect (95% of the time) that the retest time will be between 9 minutes 18 seconds and 10 minutes 42 seconds. Practically speaking, this means that no change can be found to have taken place if retest values are within the above-mentioned range.