Wednesday, 19 Mar 2008
Wednesday, 19 Mar 2008
Did you ever want to take the values for error bars in a chart from cells in a spreadsheet? Then you are maybe one of those who voted for Issue 366. As you can see by the issue number this is quite an old one. And it has over 130 votes. So, if you are interested in this feature, there are good news: it is implemented, integrated into the DEV300 code-base and thus will make it into OpenOffice.org 3.0!
It was done in the CWS chart20, which was integrated into DEV300.m2 (as chart20_DEV300). So, if you know how to do a build yourself, you can do this for this master workspace or a newer one. I suppose, there will soon be DEV300 developer snapshot builds available on the OpenOffice.org Download Website.
Here is an example for a chart that has error bars at its data points which use their values from spreadsheet data. In this example both, the positive and negative values share the same range of cells, so they have the same extent in both directions. Of course these ranges can be different as well.
In addition to this long awaited feature, the automatically calculated
error-bars now also support standard error (also known as standard
deviation of the mean). So now, there are the error types:
Technically, it would be no problem to have separate parameters for positive and negative values for percentage values and error margins as well, but the ODF file format currently does not allow this.
While implementing this new feature, the dialog for error bar properties had to be adapted. In that course, it was also improved in general. You might have noticed that in OpenOffice.org 2.3 there is one dialog/tab page for statistic properties that contains error bars as well as regression curves (now called trend lines). This dialog was split up into two dialogs for the trendline equation feature. The trendline part was already improved there. Now, the second part was done by cleaning up the error bar dialog. This is how it looks now:

tags: chart openoffice.org
3.0 will be a nice release. I also wrote error bars (it in less detail) just this morning describing new features in 3.0:
http://www.oooninja.com/2008/03/openofficeorg-30-new-features.html
Posted by Andrew Ziem on March 19, 2008 at 03:07 PM CET #
This is an awesome improvement. Once you finalize error bars you should look into including the ability to do a Chi-squared fit. gnuplot is an open source program with a chi-squared routine you could probably integrate into OpenOffice.org. This fitting routine does provides uncertainty in the fit parameters and weights the points by their uncertainty. This would be make it easier for scientists to do more of their data analysis in calc
Posted by Martin Smith-Martinez on March 19, 2008 at 06:31 PM CET #
This is great news! What about errors in x though? Furthermore simultaneous errors in x and y?... is this possible? :D
Posted by Dan Bishop on March 24, 2008 at 11:13 AM CET #
Error bars for x-values are still not possible. (See Issue 42946, http://graphics.openoffice.org/issues/show_bug.cgi?id=42946) Sorry to say.
The main issue here is that the error bar properties (like what type you want to use, or the values you want to use) are not stored at the error-bar element in the file format. Instead these values are stored at the data series. Therefore, it is not easy to extend for x-error bars, as the existing properties have no "y" in their names. So, this is a compatibility problem. The most elegant solution would be to store the properties at the error-bar element, which is incompatible, and therefore not feasible, unless we introduce redundancy.
I hope we can address this issue for OOo 3.1
Posted by Bjoern Milcke on March 25, 2008 at 05:49 PM CET #
This is a significant development, making Chart a useful tool for scientific figure production. I am especially pleased to see that the extent of the upper and lower error ranges can be different, usually the case in data with a skewed distribution. Well Done!
Colin
Posted by Colin Robert Beasley on April 01, 2008 at 03:07 PM CEST #