o-+------,r-----,----,.-------, intercept of zero and a slope of one. These requirements can even be formally tested if the standard errors of the slope and intercept are also supplied. It should be noted that statistical packages will usually give a significance test of the slope compared to zero rather than one. So, by using a slightly more complex analysis, we can start to assess the specific concept of agreement rather than association. However there are still problems with this approach. Standard linear regression analysis assumes that the dependent (y) variable is measured with error, but that the independent (x) variable is not. This leads to two different lines of best fit, depending on the choice of the dependent variable. The magnitude of the difference between the two lines increases as the correlation between the two variables decreases. This is not just a purely statistical problem, but can give two substantially different lines (Fig. 1). This can be eliminated by the use of a method assuming errors in both variables which will give a symmetrical answer, such as Deming regression,
Sally Hollis (Mon,) studied this question.