1
$\begingroup$

Which of the following two data sets is more dispersed? Assume that both data sets are normally distributed.

Data Set A: $\mu_A = 1$, $\sigma_A = 1$.

Data Set B: $\mu_B = 10$, $\sigma_B = 1$.

Answer 1: Data Set A because $CV_A = \dfrac{\sigma_A}{\mu_A} = \dfrac{1}{1} = 1 > 0.1 = \dfrac{1}{10} = \dfrac{\sigma_B}{\mu_B} = CV_B$.

Answer 2: They are equally dispersed since the graph of Data Set B is exactly the same graph as Data Set A shifted $\mu_B-\mu_A=10-1=9$ units to the right.

How do we reconcile these two equally convincing answers?

$\endgroup$
  • 1
    $\begingroup$ Is this a question from a course or textbook? If so, please add the [self-study] tag & read its wiki. $\endgroup$ – S. Kolassa - Reinstate Monica Jul 23 at 14:29
  • 2
    $\begingroup$ It is a good example of using a ambiguous word (dispersed) to smear the very clear and good defined statistical concepts (mean and standard deviation). $\endgroup$ – user158565 Jul 23 at 14:35
  • $\begingroup$ @StephanKolassa, not really. I'm just reading about these topics and this question popped to mind. $\endgroup$ – Henry Razon Jul 23 at 14:39
  • 1
    $\begingroup$ They're not equally convincing... $\endgroup$ – Glen_b Jul 24 at 5:59
3
$\begingroup$

Standard deviation and coefficient of variation are both measures of dispersion of a distribution, but which one is more useful will depend on context. SD is widely applicable, but there are situations where you definitely should not use CV.

Some measurements are taken on an interval scale, meaning that there's no true, non-arbitrary zero point on the scale - you can really only compare differences in measurements. Temperature, for example, could be on an interval scale, since 0 degrees F and 0 degrees C are both equally valid, and the zero point has no special meaning (although measuring temperature in Kelvin has a non-arbitrary zero and therefore isn't an interval measure). For anything measured on an interval scale, coefficient of variation is rather meaningless, since you could shift the scale to any numerical value arbitrarily and get an entirely new CV - there's no reason why you should prefer a CV measured in Fahrenheit rather than Celsius, suggesting that it's not a useful measure. CV should only be used for ratio scales for things like mass or length that have a non-arbitrary zero point.

If the data is on a ratio scale, CV and SD are both acceptable, but must be interpreted differently. In your example, both A and B have identical SD, indicating their variation is the same in an absolute sense. The difference in means results in different CVs, however, so dataset A has greater dispersion in a relative sense, in that the distribution varies more widely as a percentage of the mean.

$\endgroup$
  • $\begingroup$ Nuclear Wang, thank you for the enlightening answer! My understanding: Even if the data sets are on a ratio scale, there is still no clear-cut answer as to which is more dispersed. Relative sense: Data Set A is more dispersed. Absolute sense: Both data sets are equally dispersed. Did I understand it correctly? $\endgroup$ – Henry Razon Jul 23 at 15:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.