Scientific plot digitizer

The bar runs from 0 to 78 percent. The whisker half-width is the SD of 5, so the caps are at 73 and 83. Lower 73 is the bottom cap. It is drawn, not typed as a second mean.

A review often has the figure and not the spreadsheet. The scientific task is narrow: put the printed mean and the printed error into a table that matches the caption. ChartDigitizer’s bar bench is built for that reading. This page is the checklist around it, not a second tool.

Start with a screenshot or an exported bitmap, PNG or JPG, under 10 MB. Crop to one panel. If the PDF has two panels, digitize them separately so one pair of Y anchors cannot be asked to serve two different scales.

Place Y1 and Y2 on the printed yield or response axis, including zero when the bar baseline is zero. Place X1 and X2 on the category or temperature axis. The 300 °C catalyst bar is the worked example: anchors at 200 and 400 °C, 0 and 100 percent.

Read the top of the bar as the mean and the whisker half-width as the error the caption defines. In the preset, 300 °C is mean 78 and SD 5. The lower end is 73 and the upper end is 83. Those four numbers are the whole observation.

The free download will not include the SD. That is intentional. You can see 5.0000 on the bench because the whisker is already on the figure. The column in a spreadsheet is part of the paid matrix, next to Mean and SE.

If the caption says SE and prints N, convert with SD = SE × √N before you compare with a study that printed SD. The reverse is SE = SD / √N. Using the catalyst SD of 5 and a hypothetical N of 25 gives SE = 1.00. Do not invent N from the bar width.

If N is missing, leave SE blank. The paid sheet does the same: an SE cell stays empty when N was not supplied. Filling it with the SD would double-count the error in any later inverse-variance weight.

Different papers use different whiskers: SD, SE, or a confidence interval. The digitizer stores one half-width. Your extraction sheet needs a column for the definition. A 95% CI half-width is about 1.96 SE for a large normal sample, and that factor is not applied automatically.

Time-to-event figures are a different extraction. Kaplan–Meier curves on the log bench can be aligned to a shared month axis. Row 10 of a 100-row linear grid from 0 to 60 months is 5.4545, with treatment 0.9636 and control 0.9345. Censoring ticks are not inferred.

This page does not give clinical advice. A digitized survival of 0.52 at 60 months is a coordinate from a figure, not a prognosis. Pooling, bias tools, and protocol decisions sit outside the coordinate read.

Keep the source citation beside the row: figure number, panel, and the sentence that defines the whisker. The proof PNG, at most 1200 px wide with a PROOF band, is a check that the overlay sat on the bar. It is not the data extract.

When two arms share a plot, use the multi-curve bench only if they are separate polylines. Grouped bars are already separate series: Control 22 ± 2.5, low dose 31 ± 3, high dose 47 ± 4. Each arm keeps its own SD.

Axis mistakes dominate error, not the whisker click. A Y axis read as 0 to 1 instead of 0 to 100 scales every mean by 100. Check one known tick, such as 50% or 100 °C, against the midpoint readout before you trust the rest of the panel.

Images stay in the browser. The paid route receives the numbers you already checked. It does not receive the screenshot. That split matters for a review folder: the PDF stays on your machine, and the table you export is the only thing the server sees.

Open the bar bench for the five-row catalyst check, the formula page for the linear map behind the bar height, and the home digitizer when the figure is a curve rather than a bar. The same four anchors serve all three.

Place the anchors on the graph to CSV converter, or start from the digitizer. Axis algebra is on the formula page.