11

When taking in data through a file (such as table[x index=0, y index=1, y error index=2]{plots/mydata.table};) is it possible to automatically calculate the error bars (instead of putting them in manually)? Error calculation is a rather straightforward manner, and I would think that a package as complete as pgfplots would include it as a feature.

I'm open to a number of options. Obviously, a pure LaTeX (i.e. pgfplots) solution is best. If there is a way to automate the run of a script upon compilation that alters the data files before their use, that would be okay too (and this is probably the easiest, although I don't know how to effect the automation).

MWE

\documentclass{article}
\usepackage{pgfplots}
\usepackage{tikz}

\pgfplotsset{compat=1.7}

\begin{document}
\begin{tikzpicture}
  \begin{axis}[grid=major]
    \addplot+[smooth,
    error bars/.cd,
    y dir=both,
    y explicit]
    table[x index=0, y index=1, y error index=2]
    {plots/mydata.table}; % simple space-delimited data file
  \end{axis}
\end{tikzpicture}
\end{document}

Sample mydata.table

# Test input file
Sample Measure1 Measure2 Measure3 Measure4 ...
5 180 190 200 210
15 420 410 400 390
25 650 640 630 640
35 1100 1200 1150 1020

I have a script that will produce

5 194.4 4.36898157469
10 195.6 1.4310835056
15 207.4 2.23785611691
20 250.4 1.4587666023

given the input file.

7
  • 1
    What's the rule for error computation? It's fairly straightforward to add a column to the input table.
    – percusse
    Mar 18, 2013 at 16:40
  • stddev(<dataset>)/sqrt(len(<dataset>)) (I'm trying to convert a biologist friend of mine, and automatic error bar calculation would do her in. However, I'm no biologist, and I underestimated the challenge (and the very problem itself).) Mar 18, 2013 at 16:55
  • 3
    @SeanAllred: Are you sure that's the error calculation your friend wants? If there's a trend in the dataset (like it's the case in your dummy data), she'd probably want the standard error of the regression, or maybe even some confidence interval?
    – Jake
    Mar 18, 2013 at 17:14
  • @Jake, that's exactly what I meant. My apologies for not getting back to you sooner; internet here is unbearably unreliable. The question will be updated shortly. Mar 18, 2013 at 17:32
  • Can the bashful package be applied here, as far as the script is concerned? Mar 18, 2013 at 17:41

1 Answer 1

12

PGFPlots comes with the PGFPlotstable package, which can process tabulated data. It doesn't include functions to calculate summary statistics like the mean, standard deviation or standard error for data columns, but these can be added quite easily.

After the necessary code has been included in the document, you can tell PGFPlots to make the standard error of the data in columns 2 to 5 available in a column called stderror by putting the following lines somewhere before your graph:

\pgfplotstableset{
    summary statistics/end index=5,
    create on use/stderror/.style={create col/standard error}
}

By default, the code assumes that the data columns start at index 1 (so the second column in the table) and end at column 4, but this can be changed using the keys summary statistics/start index and summary statistics/end index.

Then you can plot the mean values of each row with the error bars representing the standard error of columns 2 to 5 using

  \begin{axis}[grid=major]
    \addplot+[
        smooth,
        error bars/.cd,
            y dir=both,
            y explicit
    ]
    table[
            x=Sample,
            y=mean,
            y error=stderror
    ]
    {data.txt};
  \end{axis}

Here's an example using the data you provided (with the values slightly altered for a more dramatic effect):

\documentclass{article}
\usepackage{pgfplots, pgfplotstable}
\usepackage{filecontents}


\begin{filecontents*}{data.txt}
Sample Measure1 Measure2 Measure3 Measure4
5 80 190 200 210
15 520 410 430 350
25 650 640 630 900
35 1100 1200 1150 1020
\end{filecontents*}

\pgfplotsset{compat=1.7}


%% Code chunk for statistics starts here...
\newcommand{\calcrowmean}{
    \def\rowmean{0}
    \pgfmathparse{\pgfkeysvalueof{/pgfplots/table/summary statistics/end index}-\pgfkeysvalueof{/pgfplots/table/summary statistics/start index}+1}
    \edef\numberofcols{\pgfmathresult}
            % ... loop over all columns, summing up the elements
    \pgfplotsforeachungrouped \col in {\pgfkeysvalueof{/pgfplots/table/summary statistics/start index},...,\pgfkeysvalueof{/pgfplots/table/summary statistics/end index}}{
        \pgfmathparse{\rowmean+\thisrowno{\col}/\numberofcols}
        \edef\rowmean{\pgfmathresult}
    }
}
\newcommand{\calcstddev}{
    \def\rowstddev{0}
    \calcrowmean
    \pgfplotsforeachungrouped \col in {\pgfkeysvalueof{/pgfplots/table/summary statistics/start index},...,\pgfkeysvalueof{/pgfplots/table/summary statistics/end index}}{
        \pgfmathparse{\rowstddev+(\thisrowno{\col}-\rowmean)^2/(\numberofcols-1)}
        \edef\rowstddev{\pgfmathresult}
    }
    \pgfmathparse{sqrt(\rowstddev)}
}
\newcommand{\calcstderror}{
    \calcrowmean
    \calcstddev
    \pgfmathparse{sqrt(\rowstddev)/sqrt(\numberofcols)}
}

\pgfplotstableset{
    summary statistics/start index/.initial=1,
    summary statistics/end index/.initial=4,
    create col/mean/.style={
        /pgfplots/table/create col/assign/.code={% In each row ... 
            \calcrowmean
            \pgfkeyslet{/pgfplots/table/create col/next content}\rowmean
        }
    },
    create col/standard deviation/.style={
        /pgfplots/table/create col/assign/.code={% In each row ... 
            \calcstddev
            \pgfkeyslet{/pgfplots/table/create col/next content}\pgfmathresult
        }
    },
    create col/standard error/.style={
        create col/assign/.code={% In each row ... 
            \calcstderror
            \pgfkeyslet{/pgfplots/table/create col/next content}\pgfmathresult
        }
    }
}
%%...code chunk for statistics ends here

\begin{document}
\pgfplotstableset{
    create on use/mean/.style={create col/mean},
    create on use/stddev/.style={create col/standard deviation},
    create on use/stderror/.style={create col/standard error}
}
\pgfkeys{/pgf/fpu=true} % Only needed for \pgfplotstabletypeset
\pgfplotstabletypeset[columns={Sample, Measure1, Measure2, Measure3, Measure4, mean, stddev, stderror}]{data.txt}
\pgfkeys{/pgf/fpu=false}

\begin{tikzpicture}
  \begin{axis}[grid=major]
    \addplot+[
            smooth,
            error bars/.cd,
        y dir=both,
        y explicit
    ]
    table[
            x=Sample,
            y=mean,
            y error=stderror
    ]
    {data.txt};
  \end{axis}
\end{tikzpicture}
\end{document}
7
  • Can the calculations of standard deviation be applied to data sets in the same plot or will one over ride the other?
    – dustin
    May 13, 2014 at 2:09
  • @dustin: By "different data sets", do you mean different tables or different groups of data within the same table?
    – Jake
    May 13, 2014 at 19:00
  • I mean say I have 2 csv files. I want to put them on the same plot and have standard dev bars present on each data set plot.
    – dustin
    May 13, 2014 at 19:05
  • @dustin: That should work out of the box (just repeat the \addplot command with the second csv)
    – Jake
    May 13, 2014 at 19:08
  • This piece of code wont be overwritten for one and hit both csv files no problem: \pgfplotstableset{ create on use/mean/.style={create col/mean}, create on use/stddev/.style={create col/standard deviation}, create on use/stderror/.style={create col/standard error} }
    – dustin
    May 13, 2014 at 19:17

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .