Hi! I?m trying to calculate de C index (concordance probability) through the somers2 function (library Hmisc). I?m interesting on including the sampling effort as a weight factor for the evaluation of model predictions with real data. I?ve some questions about that: first of all I?m not really sure if I can include sampling effort as a weight factor. Since the weight factor should be a numeric vector of observation (usually frequencies), I would expect that sampling effort could be a surrogate of the frequency count of the number of subjects (i.e. frequency of observation). However, when I use sampling effort as a weight factor, I get C index larger than one. I guess/know this is statistically wrong. Then, if these values were frequency of observation; what is working incorrectly? What should be the characteristics of the weight vector? Or what could be exactly included as weight factor? Thank you very much!
sara.vallecillo at ctfc.es wrote:> Hi! > I?m trying to calculate de C index (concordance probability) through the > somers2 function (library Hmisc). I?m interesting on including the > sampling effort as a weight factor for the evaluation of model predictions > with real data. I?ve some questions about that: first of all I?m not > really sure if I can include sampling effort as a weight factor. Since the > weight factor should be a numeric vector of observation (usually > frequencies), I would expect that sampling effort could be a surrogate of > the frequency count of the number of subjects (i.e. frequency of > observation). However, when I use sampling effort as a weight factor, I > get C index larger than one. I guess/know this is statistically wrong. > Then, if these values were frequency of observation; what is working > incorrectly? What should be the characteristics of the weight vector? Or > what could be exactly included as weight factor? > Thank you very much!Send me the smallest artificial example you can construct and I'll work on it. Frank -- Frank E Harrell Jr Professor and Chair School of Medicine Department of Biostatistics Vanderbilt University