1

I was reading about the coefficient of variation and it seems that it can only be computed for ratios (non-negative values). I have a list of values which indicate the % increase in a particular quantity. In certain rare cases, there is a decrease, and as a result the % increase is a negative value.

Does this mean that I cannot use the coefficient of variation on my dataset? Are there any other robust inferences I can make to understand the nature of variation in my dataset?

Nick Cox
  • 48,377
  • 8
  • 110
  • 156
  • 2
    Non-negative variables are not all ratios, nor vice versa. But the existence of negative values does usually make using coefficient of variation a bad idea. The real question is what you want to do and why you thought coefficient of variation might help: "understanding the nature of variation" doesn't tie it down as that would be a fair definition of statistics as a whole. It's not clear what you are thinking of as "robust inference" but the coefficient of variation as a ratio of SD/mean is highly non-robust statistically. – Nick Cox Apr 07 '15 at 10:32
  • I'd like to be able to say that "for the first dataset, the variation is higher than the second dataset"(The means for the two sets might vary a lot). –  Apr 07 '15 at 10:37
  • I believe what I may be looking for is "homogeneity of variance". http://stats.stackexchange.com/questions/107756/can-we-use-a-coefficient-of-variation-as-a-statistic-for-testing-homogeneity-of –  Apr 07 '15 at 10:40
  • 1
    There are formal tests for that, but side-by-side graphs usually tell you as much or more. – Nick Cox Apr 07 '15 at 10:49

0 Answers0