Question 3: What are the advantages and disadvantages of the intraclass correlation coefficient in the study of the reliability of measurements?
The intraclass correlation coefficient has two advantages over the ordinary product moment correlation coefficient. The order in which measurements were made does not matter, which it should not if the measurements are true replicates and there is no change in the quantity being measured. Also, the ICC can be calculated for any number of repeated measurements, whereas the product moment correlation coefficient can only be calculated for pairs of measurements.
The intraclass correlation coefficient has an advantage over statistics based on the within-subject standard deviation. It is independent of the units in which measurements were made. This means that it can be used to compare the measurement errors of measurements made in different ways or measured in arbitrary units, for example anxiety scales.
The intraclass correlation coefficient has a disadvantage compared to statistics based on the within-subject standard deviation. It depends on the variability of the true values of the quantity being measured. A sample of fetuses which were had high variability in femur length, such as one with a wide range of gestational ages or one chosen to have a lot of early and a lot of late gestational ages, would produce a larger ICC than one with little variability in femur length, e.g. a sample with a narrow range of gestational ages, or repeated pairs of measurements on the same fetus. If the fetuses all had the same femur length the ICC would be zero. Hence ICC is specific to the population sampled and can be interpreted only if the sample can be assumed to be representative of that population, equivalent to a random sample.
Back to question.
This page maintained by Martin Bland.
Last updated: 23 February, 2007.