Biology, images, analysis, design...
Use/Abuse Principles How To Related
"It has long been an axiom of mine that the little things are infinitely the most important" (Sherlock Holmes)

Search this site



Measurement of agreement

In the More Information page we have looked at measures of the level of association between two variables. These included the risk and odds ratios for nominal variables, and correlation and regression for measurement variables. Sometimes we may wish to go further and assess whether two sets of observations are not just associated, but are actually the same - or at least very similar. In other words we need measures of agreement, rather than measures of association.

For example we may have two observers, a farmer and a researcher, each of whom assess whether or not lice are present on cattle on a number of farms. We want to know how closely the two observers agree in their assessments. Alternatively we may have two methods, rectal measurement and axillary measurement, to measure the temperature of young children. We want to quantify the level of agreement between the two methods.

You will often find the measures of association we have given in the More Information page, especially the correlation coefficient, wrongly used in this context. One can see why it is wrong with a simple example. If rectal measurement always gave a temperature reading 0.4 degrees higher than axillary measurement, the correlation coefficient between the two measures would equal 1 indicating perfect association. Yet the level of agreement is poor because they always give a different reading!

We will look at appropriate measures of agreement between variables in the next unit.