What is the rational behind not permitting negative values when computing a ratio?
In my initial response I was simply looking at the SPSS menu for the Ratios procedure and did not see anything that was being reported that would require the two variables composing the ratio to be positive. I have now examined the SPSS Algorithms page that Rich Ulrich refers to and no longer need to guess. None of the calculations shown there make the assumption that the ratios (R) are positive. Rich stated that I believe that when he said “these data” he was actually referring to the SPSS algorithm for a distribution-free confidence interval for the median, not to any data values. It is true that the algorithm uses “binomial” and “incomplete beta” formulas, but the algorithm can be applied to any distribution, not just to a positive-valued one. Those formulas enter naturally when thinking about the possible values of a sample median. In particular the SPSS calculations do not assume that the denominator of the ratio R has a binomial distribution, nor that it is positive-integer value
