Scientific Plot Maker

Use Scientific Plot for individual observations when you want group means with SD, SEM, a t confidence interval or range. Use Bar and Column Chart Maker when you already have category values to draw. Scientific Plot does not treat a supplied mean or SD as an individual-data summary table.

After confirming a valid figure, choose R code and download the .R file. Run it locally in base R (4.0 or later). It contains the accepted observations, group order and source cells, and defines cleanfigure_result with the numerical summaries. It does not recreate the artwork, install packages or write files. Methods text describes the current mode, summary, exclusions and display; generating either output does not execute R.

A bar of means hides how many values there are and how they are spread. With small samples (n ≤ 9) journals such as PLOS Biology discourage bar graphs and ask for the individual values, so readers can see outliers, gaps and group sizes.

Each violin is a smoothed outline of where a group’s values lie: wider where values are common, narrow where they are rare. CleanFigure draws a Gaussian kernel density, computed as the exact kernel sum at 512 points between the group’s lowest and highest value, so the outline never extends past the data. A narrow type-7 box and the individual values are drawn inside by default. Violin uses a linear axis and supports up to 3,500 accepted observations.

By default each group uses R’s bw.nrd0 rule (Silverman’s rule of thumb), the same default as R density() and ggplot2 geom_violin. Adjust multiplies it (0.25 to 4): smaller values show more detail and noise, larger values a smoother outline. You can also type a bandwidth in data units. The summary table lists every bandwidth, and the sentence under the figure and the Methods text state the rule, so you can report it.

Different software uses different defaults. GraphPad Prism offers smoothing levels without a documented numeric bandwidth, and seaborn uses Scott’s rule and extends tails past the data. CleanFigure follows R: bw.nrd0, outlines trimmed to the data range, and equal area by default, as ggplot2 does. R density() uses a fast binned approximation, so its values typically differ from the exact sum by less than 0.1% of the peak density.

With fewer than 10 values an outline mostly reflects the bandwidth rather than the data, so CleanFigure always shows the individual values in such groups and notes it under the figure. A group with one value shows just that point; a group whose values are all equal shows its points without an outline. For small samples a Dot plot is often clearer.

Equal area (the default, as ggplot2) gives every violin the same area, so shapes are comparable. Area proportional to n scales each violin by its sample size, which shows which groups have more data. Equal maximum width makes every violin equally wide, which helps compare shapes when one group is much more concentrated than the others.

The box spans type-7 Q1 to Q3, with a line at the median. Whiskers end at observed values within 1.5 times the IQR of Q1 and Q3. Outliers always remain in the data and axis range. Show all observations or only outliers; singleton observations are always shown. Box uses a linear axis and supports up to 3,500 accepted observations. These quartiles need not equal base R boxplot.stats hinges.

Each bar starts at zero and shows its group’s mean. Choose SD, SEM, a 95% t confidence interval or the observed minimum–maximum range; individual observations are shown by default. Group labels show sample sizes. A singleton has a mean but no estimable error interval. Bar mode uses a linear axis and supports up to 3,500 accepted observations; Dot remains available for larger tables.

Each column with numbers becomes a group (or, with a Group column and a Value column, each group name does). The card “How we read your table” shows the groups, their number of values and missing cells, the header rows, the decimal mark and every cell that was left out, with its row. You can change any of these, and you confirm the reading before you export.

They are missing values: counted per group and listed, never turned into 0. Text in a number column, such as “<0.1” or “12 mg”, is left out and listed with its row; it is never converted into a number.

Yes. “12,5” is read as 12.5, and semicolon CSV files from European Excel work. When every value looks like “1,250”, which could be 1.25 or 1250, the tool asks which one you mean.

They answer different questions. SD (n − 1) shows how spread out the values are. SEM (SD/√n) and the 95% confidence interval (t distribution, as R’s t.test) show how precisely the mean is estimated, and shrink as n grows. Say which one you show in the legend; the sentence under the figure does this for you. The numbers match R to nine significant digits; quartiles use R’s default (type 7, as Excel QUARTILE.INC), which can differ slightly from GraphPad Prism.

Group names are written to the summary CSV and the copied table exactly as you typed them, so a name starting with “-”, “+”, “=” or “@” (such as “-LPS”) may show as a formula error (#NAME?) in Excel. Renaming the group fixes both; a CSV file can also be imported with Excel’s Data › From Text/CSV, which keeps the name as text.

No. Pasted tables and opened Excel or CSV files are read in your browser, and the figure is drawn there too. Nothing is sent to a server.

No. This tool draws your data; it does not choose or run statistical tests.