[R] missclassification in estimating proportions

Frank Harrell f.harrell at Vanderbilt.Edu
Wed Oct 2 13:34:54 CEST 2013

This is dichotomania and is an inappropriate use of continuous 
variables.  Use an information-preserving approach such as rank 
correlation.  Also, this is not an R question.

Assume that I want to compare two methods, say skinfold measurement and 
BMI, in how they classify subjects into four categories. In a 2x2 table 
I would simply calculate sensitivity and specificity and NPV an PPV, 
using the standard formulas, but can the same be directly applied in the 
4x4 table? Or are other methods preferable?

Thanks in advance


Frank E Harrell Jr Professor and Chairman      School of Medicine
                    Department of Biostatistics Vanderbilt University

More information about the R-help mailing list