3
$\begingroup$

I have data that turn out to be describable by a lognormal distribution. However, I feel like the peak value (mode) of the lognormal distribution would be more representative of the data points. Is there an analytical expression for this, and what kind of dispersion measure would be most suitable to report along with it?

$\endgroup$
0
1
$\begingroup$

Mode of lognormal distribution has closed-form solution:

$$\exp(\mu-\sigma^2)$$

As about dispersion, assuming that you are interested in variability around the mode, if I were you I'd simply use bootstrap to compute intervals over it. You could also use mean absolute deviation around the mode. However if you are interested in general measure of variability, why not simply use standard deviation?

$\endgroup$
5
  • $\begingroup$ Bootstrapping in order to describe the dispersion of a dataset sounds like overkill. $\endgroup$
    – whuber
    Oct 3 '16 at 21:12
  • $\begingroup$ @whuber agree, I was thinking about confidence intervals around the mode. I don't have better ideas at the moment - I would be happy to hear about better ones. $\endgroup$
    – Tim
    Oct 3 '16 at 21:20
  • $\begingroup$ It's natural to describe the logarithms of the data in such cases: after all, we have been told the logs will be approximately normally distributed, so we can be confident that their mean and standard deviation will give an accurate picture of them. Furthermore, there are simple and well-known "analytical expressions" for these statistics! $\endgroup$
    – whuber
    Oct 3 '16 at 21:21
  • $\begingroup$ @whuber right, but as I understand the question it's about variability around the mode. $\endgroup$
    – Tim
    Oct 3 '16 at 21:28
  • $\begingroup$ Ah, interesting interpretation! You might want to ask the OP whether that was the intention. $\endgroup$
    – whuber
    Oct 3 '16 at 21:30

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.