Apache OpenOffice (AOO) Bugzilla – Issue 72331
Wrong error indicator placement for standard deviation
Last modified: 2013-02-24 21:21:51 UTC
When user wants to display the error indicator on a chart (Statistics tab) with the Standard deviation, the error indicator is centered on each value, it should be centered on the mean value only. See attachment. MS Excel chart is fine, OOo chart is wrong.
Created attachment 41173 [details] Comparison between OOo and Excel chart
@IHA: Maybe an RFE? Same in new chart.
No idea for what the current error bar design could be useful. I believe it should show whether a "measured value" is within the standard deviation (referring to the arithmetic mean) or outside, and that's only possible with the EXCEL-like design as it is shown in "Standard_Deviation.ods". But I am not an expert.
In my opinion, both OOo and Excel perform a poor job here. Of course, in the given example, Excel is slightly better, BUT I would personally prefer the following method: - IF all the data points of ONLY one group of data are displayed (i.e. there is only one mean and one SD) - then allow user to select the confidence interval he wants e.g. CI95% (+/- 2 SD), (or 68% aka +/- 1 SD) - and draw the corresponding horizontal lines [Confidence intervals could be calculated using more robust methods, like bootstrapping sometime in the future.] This would look something like: http://addictedtor.free.fr/graphiques/RGraphGallery.php?graph=65 (well, NOT quite; it would look much more simple, with only 2 horizontal lines and a number of bars).
In statistics, there is no use to display only one set of values. The point is to compare several sets of data to visualize the distribution graphically.
> In statistics, there is no use to display only one set of values. > The point is to compare several sets of data to visualize > the distribution graphically. Usually, we want to create some confidence interval (CI). The 95% CI is the most useful one. So creating 2 horizontal lines that delimit the 95% CI is the better approach. Of course, we could add a parameter to specify the size of the CI. Selecting 68% will generate the (-1 SD; +1 SD) interval that is so accustomed in the literature (the "error bars"), though dedicated statistician rarely use it. The http://addictedtor.free.fr/graphiques/RGraphGallery.php?graph=65 graphics display a very complex confidence interval. Of course, in our example, the lines delimiting the 95% CI would be horizontal (differing from the very complex lines in the previous graph). AND, because there are really lines drawn that look like grid lines, you could more easyly compare the distribution of the individual values. See my attached graphic. That's how I imagine it.
Created attachment 44652 [details] Confidence Interval vs Error Bars
confirmed, changed target
changed target due to limited resources
For information, workarounds available here: http://user.services.openoffice.org/en/forum/viewtopic.php?f=34&t=202&p=22409#p22406
I fixed the placement in CWS chart27. Error indicators for standard deviation are now located at the mean value position. The different length will be addressed with issue 66788.
@Thomas, please verify in CWS chart27. For all old files including the attached one it is necessary to double click the chart once or update it otherwise (e.g. by pressing F9) to see the corrected rendering.
Seen ok in CWS chart27 -> verified
Seen ok in current master -> closed