| ass.measures {MethComp} | R Documentation |
Computes correlation, mean squared difference, concordance correlation
coefficient and the association coefficient.
middle and ends are useful utilities for illustrating the
shortcomings of the association measures, see the example.
ass.measures(x, y) middle(w, rm = 1/3) ends(w, rm = 1/3)
x |
vector of measurements by one method. |
y |
vector of meásurements by another method. |
w |
numerical vector. |
rm |
fraction of data to remove. |
These measures are all flawed since they are based on the correlation in various guises. They fail to address the relevant problem of AGREEMENT. It is recommended NOT to use them. The example gives an example, illustrating what happens when increasingly large chunks of data in the middle are removed.
ass.measures return a vector with 4 elements.
middle and ends return a logical vector pointing to the middle
or the ends of the w after removing a fraction of rm from data.
Bendix Carstensen, Steno Diabetes Center, http://www.biostat.ku.dk/~bxc
Shortly...
cbind( zz <- 1:15, middle(zz), ends(zz) ) data( sbp ) bp <- subset( sbp, repl==1 & meth!="J" ) bp$meth <- factor( bp$meth ) tab.repl( bp ) plot.meth( bp ) bw <- to.wide( bp ) with( bw, ass.measures( R, S ) ) # See how it gets better with less and less data: rbind( with( subset( bw, middle( R+S ) ) , ass.measures( R, S ) ), with( bw , ass.measures( R, S ) ), with( subset( bw, ends( R+S ) ), ass.measures( R, S ) ), with( subset( bw, ends( R+S, 0.4 ) ), ass.measures( R, S ) ), with( subset( bw, ends( R+S, 0.6 ) ), ass.measures( R, S ) ), with( subset( bw, ends( R+S, 0.8 ) ), ass.measures( R, S ) ) )