Timeline for Normal QQ-plot of logarithm of data does not match log-normal QQ-plot of data itself
Current License: CC BY-SA 3.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 25, 2017 at 19:37 | history | bumped | CommunityBot | This question has answers that may be good or bad; the system has marked it active so that they can be reviewed. | |
| Oct 8, 2016 at 20:56 | comment | added | eos | I edited the question to provide a direct scipy version. I fit the data to a lognormal distribution, get the parameters, and make a probability plot accordingly. 1) why do the statsmodels and scipy plots look so different? 2) why does the scipy plot still look like it's not a very good fit? | |
| Oct 8, 2016 at 20:54 | history | edited | eos | CC BY-SA 3.0 |
added scipy plot
|
| Oct 8, 2016 at 20:44 | comment | added | Matthew Gunn | Clearly from second figure, something f'ed up is going on when fitting the log-normal distribution. | |
| Oct 8, 2016 at 20:37 | comment | added | Matthew Gunn | Do you know how it fits the lognormal distribution? Does it estimate parameters with maximum likelihood? Or what is it doing? Eg. the maximum likelihood estimator of $\mu$ for lognormal distribution is $\frac{1}{n} \sum_i \log x_i$. That's almost certainly not what it's doing though. FYI a general issue with log-normal distribution is that if you compute raw moments (eg. $\frac{1}{n} \sum_i x_i$ etc...) to estimate parameters $\mu$ and $\sigma$ you tend to perform horribly because all your estimation is driven by a few observations in far right tail. | |
| Oct 8, 2016 at 20:01 | history | edited | eos | CC BY-SA 3.0 |
added descriptive stats of the values
|
| Oct 8, 2016 at 11:45 | answer | added | blackeneth | timeline score: 4 | |
| Oct 8, 2016 at 2:41 | history | asked | eos | CC BY-SA 3.0 |