# typeset large uncertainties using siunitx

I have some entries in a table where the uncertainties are very large compared to the value due to outliers in the data:

\documentclass{article}

\usepackage{siunitx}
\usepackage{booktabs}

\begin{document}

\resizebox{1\textwidth}{!}
{%
\large
\sisetup{detect-weight=true,detect-inline-weight=math}
\begin{tabular}{S[table-format=1.2(2)]S[table-format=1.2(2)]}
\toprule
8.39(12) & 13.11(25)\\
4.64(18) &  0.76(10000000)\\
\bottomrule
\end{tabular}
}

\end{document}


Is there a standard way for typesetting this neatly? I couldn't find a way to get siunitx to typeset just the uncertainties using scientific notation, e.g. "0.76(e7)", is this possible? I have also been unsuccessful when trying to denote these outliers using a text symbol "-", e.g. "0.76({-})". Any ideas?

-
unrelated to your question but avoid distorting your tables with \resizebox{1\textwidth}{!} it means you get inconsistent font sizes in every table, depending on the table data which is very jarring for the reader. –  David Carlisle Jun 4 '13 at 21:12
The problem is that I have lots of data and tables are arranged side-by-side. If I don't do the resize-huge font hack, the font becomes too small. –  Patrick Jun 4 '13 at 21:15
choose a font size that works say \large or \huge and stick to it for all such tables so they use the same font. there can't be any need to scale the thing (if you do do that you are missing a % after \end{tabular}) –  David Carlisle Jun 4 '13 at 21:17
If the uncertainty is seven orders of magnitude greater than the measured value, shouldn't the value be left out entirely? Reporting a value with two significant decimal places but an uncertainty of 10 million does not make much sense. Maybe you should also check your method for calculating your uncertainty: Surely your knowledge of the true value is not limited to "It's somewhere between 10 million and negative 10 million"? –  Jake Jun 4 '13 at 21:18
@Jake: Actually the numbers are quantiles and the errors are the 95% confidence of the Marritz-Jarrett standard errors. I'm not a statistician, so any suggestions are welcome! Really I would just like a way to say the numbers aren't trustworthy without having to put in the large errors... –  Patrick Jun 4 '13 at 21:25