Interclass Correlation Coefficient

An alternative correlation statistic is the Intraclass Correlation Coefficient (ICC). This is commonly used to compare observers (for inter-observer correlation/reliability) or sets of measurements for one observer (for intra-observer correlation/reliability). Suppose that an observer makes a series of measurements on MRI scans. He then repeats the measurement after a period of time and we want to quantify the intra-observer reliability. The data set MRI.rda contains 62 measurement pairs. Load it into the Data Viewer to proceed.

A Pearson’s correlation coefficient can be calculated using the Deducer GUI, or the console as described:

cor(MRI$FirstMeasurement,MRI$SecondMeasurement,method=’pearson’)
[1] 0.9962104

For the ICC, the package “irr” 1 should be installed:

library(irr)

And call the icc funtion to calculate the ICC:

icc(MRI,model=”twoway”,type=”agreement”)
 Single Score Intraclass Correlation

   Model: twoway
   Type : agreement

   Subjects = 60
     Raters = 2
   ICC(A,1) = 0.996

 F-Test, H0: r0 = 0 ; H1: r0 > 0
 F(59,54.9) = 525 , p = 1.93e-60

 95%-Confidence Interval for ICC Population Values:
  0.993 < ICC < 0.998

The interclass correlation coefficient is 99.6% and the p-value for the test of no association is highly significant.

1.
Gamer M, Lemon J, Fellows I, singh P. irr: Various Coefficients of Interrater Reliability and Agreement [Internet]. 2012. Available from: http://CRAN.R-project.org/package=irr