CSV to Graph

Which shape answers which question

All eight shapes redraw from rows that are already in memory, so switching between them costs nothing and the file is never read twice.

• Line. Did this go up. The stroke carries the direction and the axis carries the size. Use it whenever the horizontal axis is time and you care about the shape of the movement more than the height of any one point.
• Area. The same stroke with a low-opacity fill beneath it. The fill reads as accumulated volume, so pick it when the quantity under the curve is meaningful, such as total revenue booked in a quarter. Because the fill encodes magnitude, an area graph really does want a zero baseline in a way a plain line does not.
• Scatter. Does this move with that. Set Measure to one numeric column and the Y measure dropdown appears for the second. Every row becomes a point, no aggregation happens, and the Pearson coefficient lands in the caption. Add a Dimension and the points take a color per category, up to eight groups.
• Bar and Column. Ranking rather than trend. Bar lays the categories down the left where long labels are readable without rotation; Column stands them up, which suits short labels and short time sequences.
• Donut. Parts of a whole, with the share in the legend. Worth resisting past about six slices.
• Histogram. One measure, binned. This is the shape that tells you whether the average you have been quoting sits anywhere near a typical record.
• Boxplot. Spread per group, with whiskers at 1.5 times the interquartile range and every point beyond them drawn as an outlier dot.

The remaining controls are the same six you would expect. Measure and Dimension pick the columns. Aggregate offers sum, average, count rows, count distinct, minimum, maximum and median. Show and Sort trim and order a categorical axis. Time grain overrides the bucketing.

Worked example: a year of orders

Seven columns, one date, three categories and three numbers, 180 rows across 2024. The first few:

date,region,product,channel,units,unit_price,revenue
2024-01-02,West,Doohickey,Direct,58,211.64,12275.12
2024-01-04,West,Sprocket,Partner,13,40.73,529.49
2024-01-06,North,Sprocket,Retail,31,147.97,4587.07
2024-01-08,East,Gizmo,Direct,7,172.02,1204.15

Dropped here with nothing clicked, you get a line graph titled revenue over time, a kicker reading TREND in the top right corner, and a caption reading Sum of revenue · monthly totals · 12 points. The vertical axis is marked $0, $40k, $80k, because revenue was read as money rather than as a bare number, and the horizontal axis carries three labels: first bucket, middle bucket, last bucket.

Monthly is not an arbitrary pick. 180 rows spread over twelve months would be 180 daily ticks on a panel about 700 pixels wide, roughly four pixels each. Twelve monthly points are legible from across a room. Set Time grain to Week and the same rows redraw as about 52 points, the caption updates to say so, and the seasonality that was smooth at monthly grain turns into a visible sawtooth. Neither picture is wrong. Only one of them is what you meant, and the caption is how a reader six months from now finds out which one they are looking at.

Now switch Chart to Scatter, leave Measure on units, and set Y measure to revenue. The title becomes revenue against units, all 180 rows appear as individual points, and the caption reads 180 points · r = 0.62 · moderate positive. Underneath, a note repeats it in a full sentence and adds that correlation is not causation. That 0.62 is doing real work here. Revenue is units times unit price, so a perfect 1.00 would only appear if every product cost the same; the gap between 0.62 and 1.00 is the price mix, and seeing it is the reason to plot the scatter at all.

Set Dimension to product while still on Scatter and the cloud splits into colored groups. What looked like one moderately correlated blob resolves into several tight, steeply sloped lines, one per product, each with its own price. That is the whole argument for scatter over a summary statistic: the aggregate said moderate, the picture says several strong relationships stacked on top of each other. Finish with Download chart data and you get the exact plotted pairs as CSV, so the appendix and the slide cannot drift apart.

What this page will not do

The genuine edges, listed because these are the ones people run into within the first minute:

• No second line. One series per graph. There is no multi-line comparison and no stacked area here. Two measures at once is what Scatter is for; a genuine multi-series line belongs in the app's chart builder, which has per-series aggregation and reference lines.
• No trend line and no forecast. The scatter reports r and stops there. No fitted line is drawn, and no dashed continuation runs off the right edge of a time graph. A projection drawn from twelve noisy points looks authoritative and earns none of it.
• No dual axis. Two measures on two vertical scales let you manufacture any crossing point you like by choosing the ranges. If the comparison is real, normalize both series first and plot them on one scale.
• No log scale. The vertical axis is linear. For growth spanning several orders of magnitude, use the app.
• No filtering. Every row in the file is plotted. Filter first in the app, or split the file with the CSV splitter.
• A header row is required. If the first line is data, its values become the column names. Add a header, or promote a row in the app.
• Dates must be parseable. Unrecognized date formats stop the graph with a message rather than falling back to alphabetical order. ISO dates always work.

Comma, semicolon, tab and pipe delimited files all load without configuration, so pasting cells straight from a spreadsheet works. Excel workbooks belong on excel-to-chart and nested JSON on json-to-chart; both flatten to the same rectangle and behave identically from there.

Related