ChakataStat
Download

Module 4 — Graphs

Question: what does the distribution actually look like? Numbers summarize; pictures reveal shape, outliers and comparisons at a glance.

Open the sample (Ctrl+Shift+O). Charts land as cards in the Output View, interleaved with tables; right-click a card to copy it as an image straight into a document, or Export image… to save it — see Taking a chart out of ChakataStat at the end of this module.

Histogram — one scale variable's shape

Graphs → Histogram… (Ctrl+Shift+H), variable BMI, Run.

A roughly bell-shaped pile centred in the mid-20s with a right tail — real BMI data look like this too. Re-run with the bins slider at different values: more bins show more detail and more noise. The bins are display only — they change no data. (Turning BMI into actual categories is a transform — module 8.)

Boxplot — a scale variable across groups

Graphs → Boxplot… (Ctrl+Shift+B), variable SystolicBP, category SmokingStatus, Run.

Each box is a group's median and quartiles, whiskers and outlier points. The Current box should sit visibly higher — the picture version of the test you will run in module 6.

Line Chart — a trend across a scale variable

Graphs → Line Chart… (Ctrl+Shift+L), Y series SystolicBP, X axis Age, split by SmokingStatus, Run.

health_study is one row per participant, not one row per visit, so this connects each smoking group's points in age order rather than tracing a single patient over time — the mechanics are the same chart you would use for a real time series (repeated visits, a stock price, a sensor log). Turn on point markers to see the individual participants along each line.

The Chart Builder

Graphs → Chart Builder… (Ctrl+Shift+G) is the general dialog: pick the chart type, assign variables to its roles. Rebuild the boxplot above in it, then try a Scatterplot of BMI (X) against SystolicBP (Y) — the cloud should visibly climb, previewing module 7's correlation.

Beyond the four chart types this module builds, the Graphs menu also offers Bar Chart, Error Bar and Q-Q Plot, and analyses draw their own charts as part of their output (a forest plot from a meta-analysis, a survival curve from Kaplan-Meier). The Graphs reference is the full catalogue.

Editing a chart after you have drawn it

Every chart card has an Edit chart… button in its header. Press it on the boxplot and the chart editor opens with a live preview beside the controls that chart actually supports — for a boxplot: the title, the y-axis title and range, the colours, gridlines, reference lines and figure height. A line chart or a bar chart offers more (legend placement, marker shape and size, bar stacking), because those charts have more to set.

The editor changes how a chart looks, never what it measures: no control in it can alter a variable, a filter or a statistic. To change the analysis, re-run the command.

Change the title to something a reader would understand, then press Apply.

Now look at the Syntax tab. Your edit is in the command:

boxplot variable=SystolicBP, category=SmokingStatus, title="Systolic pressure by smoking status"

That is the part worth pausing on. ChakataStat did not save your styling somewhere beside the chart — it rewrote the chart's own line in the journal, so re-running the script tomorrow brings the chart back styled exactly as you left it. Presentation is part of the analysis you recorded, not a layer on top of it. Editing a chart covers the editor in full.

Taking a chart out

Right-click a card and choose Export image…. Three formats:

  • PNG — a picture at a resolution you choose. For a slide or an email.
  • SVG — vector: every line and label still separate and still editable in a drawing program. The format most journals prefer.
  • PDF — vector, as a one-page document that opens the same way everywhere.

The two vector formats have no resolution to set, which is the point of them: they redraw sharp at any size, so the resolution control greys out when you pick one. Exporting one chart has the detail.

Exercise

  1. Histogram of FastingGlucose_mmolL. There is a small second bump well to the right of the main pile — what is it? (Codebook: a diabetic-range cluster.) Note that the declared-missing 99s are not in the plot.
  2. Boxplot of BMI by Sex. Do the medians differ much?
  3. Scatterplot Age (X) vs HeartRate (Y). Describe what "no relationship" looks like — you will want this picture in mind whenever a correlation comes out near zero.

The syntax trail

histogram variable=BMI
boxplot variable=SystolicBP, category=SmokingStatus, title="Systolic pressure by smoking status"
line_chart y=[SystolicBP], x=Age, split=SmokingStatus
scatter x=BMI, y=SystolicBP

Note the boxplot line: it carries the title you typed in the editor. Every chart edit lands in the trail the same way.