Apache OpenOffice (AOO) Bugzilla – Issue 87324
regression curve with error bars incorrect
Last modified: 2014-03-02 18:47:14 UTC
Thanks for the new possibility to add custom error bars, this is a fantastic improvement for the use in Sciences. However, the current implementation of the regression curve is incorrect when using error bars. Usually, error bars are interpreted as the "standard deviation" of the y. When making a regression, you should minimize Chi^2 = Sum((y-yi)^2/s^2) where s is the "standard deviation" Currently, the s parameter is not used. Whatever the values in the error column, the fit is the same. I have attached a file with simple data: X Y Error 1 1 0.1 2 1.5 1 3 3.2 O.1 With OOo, slope=1.1 intercept= -0.3 With Gnuplot, slope=1.1 intercept= -0.102985 If you use equally weighted data in Gnuplot (A constant error of 1), you get the same values that in OOo.
Created attachment 52239 [details] Comparison of the same fit between OOo and Gnuplot
Hi Iha, pls have a look. kla
I just checked with Excel2006 and Gnumeric. Both have the same problem than OOo-DEV300. I still think that the gnuplot approach is the correct one, but it may be better for compatibility reasons to either stick with the current implementation or to add an option for the fitting "Weight data using error values"
SBA: This issue has a target set but is still in state of "Unconfirmed". Please re-check with OOo 3.0 or younger if it is (still) valid. Then confirm it or set an appropriate resolution. Thank you.
Yes it is still valid but it is a design issue. The question is to know if you prefer to be "excel-compatible" even when excel is wrong or even buggy or if you prefer to implement the best possible behavior (For curve fitting, gnuplot is definitely better than excel).
Confirmed with AOO410m14(Build:9760) - Rev. 1573062 2014-03-01_04:11:01 - Rev. 1573123 Debian