Lecture 01
Basic Visualization Charts
— Core principles of data visualization — why it matters and how different chart types serve different analytical goals.
Before
Now
TakeawayBars are for comparing categories — sort them whenever ranking matters more than chronological order.
TakeawayStacked bars are good at showing composition, but only the first segment is easy to compare accurately across bars.
TakeawayGrouped bars keep every value individually comparable; stacked bars keep the total comparable. Choose based on which one you need.
TakeawayReach for a pie chart only for a handful of parts of a whole; beyond ~5 slices, a sorted bar chart reads faster and more accurately.
TakeawayA trend line is a summary, not evidence — one bad row can flip it, and only looking at the dots will tell you that it did.
TakeawayArea fill is a line chart with extra visual weight on magnitude — good for one series, reconsider it once you overlay a second.
TakeawayUse lines when the x-axis is continuous (usually time) and the shape of change matters more than any single category's value.
What this lecture covers: the basic chart types — bar, line, pie, scatter and area, one demo each — how to match one to the question you're actually asking, and the handful of rules that keep a chart honest.
What to notice & try
What to notice
- A chart's job is to make ONE comparison fast — pick the type by the question, not by which one looks most impressive.
- Sorting trades 'read this as a timeline' for 'read this as a ranking' — use whichever the question actually needs.
Try this
- Open the Bar Chart demo and toggle Sort Descending — notice how much faster ranking questions become.
- Open the Pie Chart demo with Show Percentages off and try to rank the four quarters — then open the Bar Chart tab and rank the months.
Which month had the best revenue — how fast can you tell?Best for comparing discrete categories. Bar length encodes a quantitative value. Sort bars to highlight ranking.
What to notice & try
What to notice
- Every bar shares the same zero baseline — that's what makes lengths directly comparable.
- Unsorted, finding the best month means scanning all twelve bars; sorted, it's simply the first one.
- Value labels give exact numbers, but too many labels start turning the chart back into a table.
Try this
- Turn on Sort Descending and time how fast you can name the top three months.
- Turn off Show Values — notice how much slower it is to read exact numbers off the axis.
Which month sold the most in total? Now try the harder one: was Marketing bigger in August or in October?Stacked bars show composition — each bar is divided into segments that sum to the total, so one chart carries both the total and its breakdown. The cost is vertical room: the axis has to reach the sum of every segment, not just the largest one, so each series added makes the bars taller.
What to notice & try
What to notice
- Only the bottom segment shares a common baseline — every segment above it is much harder to compare across bars.
Try this
- Pick a month and try to judge the Marketing segment against another month's — then switch to the Grouped tab and do it again.
Was Marketing bigger in August or in October? Compare how long that took you here and on the Stacked tab.Grouped bars place related values side by side, making it easy to compare individual series across categories. Unlike stacked bars, absolute values stay readable for each series. The cost is room: every series added puts another bar in each category, so the same width has to be shared by more and thinner bars.
What to notice & try
What to notice
- Every bar — not just one segment — shares the same baseline, so any two values are directly comparable.
- The trade-off is space: two series mean twice as many bars in the same width, and it gets worse with every series you add.
Try this
- Switch to the Stacked tab with the same idea in mind: which layout answers 'which is bigger, Revenue or Marketing?' faster?
Without reading any labels — can you tell which slice is biggest and which is 2nd?Best for part-to-whole relationships. Use sparingly — humans struggle to compare angles. Donut form is often easier to read.
What to notice & try
What to notice
- Q3 and Q4 are within two percentage points of each other — with the percentages hidden, the picture alone won't settle which is bigger. Angle is one of the least accurate things human vision measures; length is one of the most.
- Donut mode removes the center circle but doesn't fix the underlying angle-judgment problem.
Try this
- Decide whether Q3 or Q4 is bigger from the wedges alone, then switch Show Percentages on to check. Two percentage points is about seven degrees of arc — a gap the same size in bar length would be obvious.
- Switch to the Bar Chart tab with the same data and rank the months — notice how much faster that ranking comes together.
These twelve points are almost perfectly correlated. How much of that can one extra row destroy?Best for showing relationships (correlation) between two quantitative variables. Dot size can encode a third variable.
What to notice & try
What to notice
- Turn Add Outlier on: the fitted slope goes from +0.14 to about zero, so the line now says units and revenue are unrelated — on data where twelve of thirteen points say the opposite.
- The damage comes from where the point sits on the x axis, not from how wrong its value is. A least-squares fit minimises squared distances, so a point far out on its own pulls hardest.
- The trend line summarises the relationship in one stroke — which is exactly why you check the dots before trusting it.
Try this
- Turn Add Outlier on with Trend Line on: the dotted grey line is the fit the twelve real months give on their own, the red one is the fit with the outlier in. Compare the two slopes.
- Leave the outlier on and turn Trend Line off — with only the dots to go on, would you have drawn either line?
Revenue falls from July to October. Does the fill make that drop look bigger or smaller than the plain line on the next tab?Like a line chart but filled — emphasises the magnitude of change. The filled area draws visual attention to cumulative volume.
What to notice & try
What to notice
- The filled area emphasises magnitude — useful for a single series, but overlapping fills get hard to read.
- Overlaying a second series turns this into an implicit comparison — check whether the overlap itself is meaningful or just visual noise.
Try this
- Turn on Overlay Marketing and check whether both filled areas are still individually readable.
Where did revenue grow fastest, and where did it dip?Best for showing trends over a continuous dimension (time). Connected lines emphasise continuity between points.
What to notice & try
What to notice
- It's the slope between two points — not their height — that shows the rate of change.
- Area fill draws the eye to cumulative volume, but can make small dips harder to spot.
Try this
- Turn on Area Fill and check whether the month-over-month dip is still as easy to spot.
Ctrl+Enter to run
No parameters for this demo.
Parameters
Bar Color
Colour carries no data here — one hue keeps attention on bar length, the real encoding.
Labels save the reader from tracing each bar back to the axis.
Sorting trades 'read this as a timeline' for 'read this as a ranking'.
Parameters
Revenue Color
Marketing Color
With more than one segment per bar, labels get crowded fast.
Parameters
Revenue Color
Marketing Color
Two full sets of labels next to each other — check when this gets too busy to read.
Parameters
The quarter names around the pie — without them you can't tell which wedge is which.
Off by default — the angles are all a pie really encodes. Turn it on to check your answer.
Slice Gap
0.02
A thin white gap separates neighbouring wedges; too much of it starts eating the slices themselves.
Parameters
Dot Color
A fitted line always exists mathematically — that alone doesn't mean the relationship is strong.
Adds one implausible thirteenth month — huge volume, tiny revenue — and refits the line.
Parameters
Area Color
Fill Opacity
0.25
Lower opacity keeps a second overlaid series from fully hiding the first.
Parameters
Line Color
Line Width
2.5
A thicker line is easier to see from a distance but can visually smooth over small dips.
Dots mark the actual data points — without them you can't tell where real data ends and interpolation begins.
Error