Data Visualisation Design

This workbook turns the course into a working method you apply to one real dataset. Choose a dataset now, something with at least one category column and one number column, and carry it through every exercise so you finish with a portfolio-ready data story rather than abstract notes. Each section maps to one course module, moving from perception through chart choice, colour and clutter, to a finished dashboard and narrative.

Ground every later decision in perception: rank your encodings, decide what should pop, and choose the dataset you will carry through the course.

Pick a single real dataset you genuinely care about and will use for the entire course. It needs at least one category column and one numeric column; a time column is a bonus. Public sources like Our World in Data, your own business numbers, or a sports table all work. Open it and answer the prompts before doing anything visual.

Take any chart you find in the wild this week, a news graphic, a report, a dashboard, and audit it against the Cleveland-McGill hierarchy. The goal is to start seeing charts as encodings ranked by accuracy, not as pictures.

For your own dataset, write the single sentence the chart must say, then decide which preattentive attribute will carry it. The point is to prove you can name what should be loud before you style anything.

Move from perception to practice: pick the chart family from the question, build the right workhorse chart, and stress-test it for honesty.

Before choosing a chart, name the question. Fill this in for your dataset so the chart is the answer to a stated question rather than a gallery pick.

Make the same data three different ways, for example a pie, a sorted bar chart, and a dot plot, then judge which lets a reader answer your question fastest and most accurately. This makes the accuracy hierarchy real in your own hands.

Run your chosen chart through the common ways charts distort data and record the verdict for each. Treat this as the honesty contract: the visual proportions must match the data proportions.

Make the chart clear and accessible: match the palette type to the data, prove it survives colour blindness and greyscale, and strip everything that is not data.

Choose colour deliberately by first classifying the data, then picking the matching palette type from a tested source. Fill this in before colouring anything.

Take your chart and prove it works for everyone. Simulate it under colour-vision deficiencies and convert it to greyscale, then fix any failure with a redundant cue. This is the single most important accessibility step.

Apply data-ink discipline to your chart: remove non-data ink until the data dominates, then replace the legend with direct labels and rewrite the title as a message. Compare a before and after screenshot.

Assemble finished work: tidy the data, build in a real tool, lay out a dashboard by priority, and frame a single honest data story.

Cleaning beats styling: get the data into tidy shape and pick the right tool before you build. Fill this in so the chart step takes minutes, not hours.

Take your tidy data into Datawrapper or Flourish and produce a finished, publishable chart: correct chart type, zero-baselined bars, colour-blind-safe palette, direct labels, message title, and a source line. Export both an embed and a static image.

Design the dashboard as one answer, not a wall of charts. Decide the audience and priority before placing anything, following overview first, then details on demand.

Turn your best chart into an explanatory data story built around one message, framed honestly, and guided by annotation. This is the capstone that proves the full method on your dataset.