| Author |
 Topic  |
|
|
Theiden
Italy
3 Posts |
 Posted - 09/25/2019 : 7:39:25 PM
|
Hello,
Could anyone explain me, please, the reason for this:
From a data distribution I've plotted a histogram. Going in the plot details menu, distribution tab I've selected the lognormal curve (black curve in figure). Since I need the parameters of the curve, I've also opened the bin worksheet, plotted the data points and performed the curve fitting (offset y0 fixed or not at 0) but the resulting fitting (red curve in figure) is very different from the previous one (and it seems even worse). Why? some suggestions?
Another thing: Does the xc parameter from the lognormal fit represent the mean or the median?
Thank you very much! |
Edited by - Theiden on 09/25/2019 7:41:40 PM |
|
|
YimingChen
1691 Posts |
Posted - 09/26/2019 : 10:56:31 AM
|
Hi,
If you want to run a distribution fit on your raw data, please go to menu Statistics: Descriptive Statistics: Distribution Fit... where you can choose lognormal distribution. When you have a histogram plot, if you perform a nonlinear curve fit on the binned data, what you end up fitting is the binned data not the raw data, which is not accurate. While the distribution curve turned on in the histogram is accurate since it's equivalent to running distribution fit on raw data.
xc parameter in lognormal fit is neither mean nor median. If you check the formula, xc is like a scaling factor. And estimated ln(xc) value is the mean of ln(x).
Thanks
James |
Edited by - YimingChen on 09/26/2019 10:57:16 AM |
 |
|
|
Theiden
Italy
3 Posts |
Posted - 10/11/2019 : 07:01:52 AM
|
Thank you for your answer, it was very useful.
But at this point I have other two questions:
1) if I have a distribution with two population (two peaks), can't I use the Distribution fit? The only way is to use the multiple peak fit on the bin considering the approximation that you mentioned, right? I suppose that if the bin size is small enough the approximation should be nice, or not?
2) I see that in the Distribution fit report there is the mean and median with their standard deviation (very useful because I have to describe my distribution in the form M+-SD). But is there a way to extract this information also from the xc and w parameters of the curve fit?
Thank you |
|
|
|
YimingChen
1691 Posts |
Posted - 10/11/2019 : 4:03:50 PM
|
Hi Theiden,
1) Well, you can get good estimate of the dist function as long as you have enough samples falling into each bin that can be used to approximate relative density. You will also need to normalize the fitted curve function to 1 in order to be as dist function.
2) Yes, you can get the standard error of parameters after performing curve fitting.
Thanks,
James |
|
|
|
Theiden
Italy
3 Posts |
Posted - 10/15/2019 : 12:25:31 PM
|
Hi James,
So do you confirm that there is noway to perform a distribution with two peaks?
Which formula should I use to convert xc, w, mu and sigma in mean, median and SD?
Thanks
Theiden |
|
|
| |
Topic |
|
|
|
Lognormal: distribution vs fitting