I need to compute modes and means and more stuff about some datasets. For this purpose, I decided to work with R. I am using MASS package and function fitdistr. Here is sample code:

a <- c(1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,3,3,3,3,3,3,3,4,4,4,4)
b <- fitdistr(a, 'lognormal')
    meanlog       sdlog   
  0.66214543   0.51994590 
 (0.08788690) (0.06214542)

And here I am stuck, because I dont know how to interpret this results. 0.67 is not mean of a. Somewhere I found that estimates are on log scale. Does it mean that mean and sd of a are exponentially transformed estimates? Or does it mean that this estimates are parameters of underlying normal distribution?

  • $\begingroup$ The parameters of the lognormal distribution are explained in help("dlnorm"). $\endgroup$
    – Roland
    Jan 26 '17 at 12:37
  • $\begingroup$ Yep, so parameters are on log scale - what does it actually mean? $\endgroup$
    – Robert
    Jan 26 '17 at 13:06
  • $\begingroup$ FWIW, there isn't even an astronomically remote chance that these data are from any lognormal distribution. "Meanlog" is shorthand for "mean(log())" while "sdlog" is shorthand for "sd(log())". That makes it easy to interpret the results. $\endgroup$
    – whuber
    Jan 8 '19 at 18:18

I finally found the solution. As someone stated here, meanlog and sdlog are moments of underlying normal distribution.


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.