1 paper · 1 filter
Giuseppe Mignemi, Antonio Calcagnì, Andrea Spoto +1
In several observational contexts where different raters evaluate a set of items, it is common to assume that all raters draw their scores from the same underlying distribution. Ho…