← Data Science learning path0%
06
Week 6 · 37 slides + Lab 6
Data Visualisation
Visual encoding, baselines, histograms and density, colour and accessibility, and honest claims.
Not started yet
Start Quiz mastery0%
What you’ll take away
- One mark, one meaning; one channel, one field.
- Bars from zero; lines with labelled limits.
- Value goes to area; radius grows by the square root.
- Question first, then variables, then chart.
- Name the denominator and the unit of change.
- Bins change the picture, not the data.
- Height is density; area is share.
- Fences are rules; whiskers are data.
- A gap means unobserved; a headline needs dates.
- Would the message survive without its colours?
- One question, one highlight, one shared scale.
- Pool counts, not percentages.
- Point to the evidence; protect the individual.
- Every interval needs a name.
Lecture topics
Marks, channels and visual encoding; Baselines: bars start at zero, lines may focus; Angles and areas: pie charts and bubbles; Chart choice follows the question; Counts, rates and honest wording; Histograms: bins and bin width; Count, proportion and density; Box plots, fences and whiskers; Line charts: time order, gaps and windows; Colour with a purpose; Highlighting and small multiples; Pooled rates in a grouped chart; Aggregates, privacy and chart critique; Optional: uncertainty, log scales and plotting code
Lessons
- Marks, channels and visual encodingA chart maps each record or summary to a mark, and each field to a visual channel such as position, length or colour. Reading a chart starts with knowing what one mark means.7 min
- Baselines: bars start at zero, lines may focusA bar shows a value by its length, so it must start at zero. A line shows change by position and slope, so a focused axis can be honest if its limits are clear.7 min
- Angles and areas: pie charts and bubblesA pie splits one whole into shares, so its angles follow the percentages. A bubble's area, not its radius, should match its value.7 min
- Chart choice follows the questionThe question and the variable types decide the chart. Keep the natural order of ordered categories, and make overlapping points visible.7 min
- Counts, rates and honest wordingA count says how many; a rate says what share of the eligible cases. Comparing rates needs percentage points or relative change, clearly named.7 min
- Histograms: bins and bin widthA histogram counts numerical values in touching intervals. Different bin widths show the same data differently, so a real pattern should survive sensible choices.7 min
- Count, proportion and densityOne histogram can show counts, proportions or densities. When bins have different widths, use density so that each bar's area shows its share of the data.7 min
- Box plots, fences and whiskersA box plot shows the median, the middle half of the data and unusual values. Fences decide what is flagged; whiskers stop at real observations.7 min
- Line charts: time order, gaps and windowsA line joins observations in time order. Missing weeks should appear as gaps, and the chosen time window must be stated because it can change the story.7 min
- Colour with a purposeColour should do one clear job: separate categories, show order, show direction from a midpoint, or highlight one item. The message must still work without colour.7 min
- Highlighting and small multiplesTo answer one question about one series, highlight it and keep the others grey for context. To compare groups side by side, use small panels that share the same axis.7 min
- Pooled rates in a grouped chartPooled rates add counts before dividing, so they depend on group sizes. A grouped bar chart can show that one mode leads in every group yet trails overall.7 min
- Aggregates, privacy and chart critiqueA good critique checks the unit, calculation, scale, identity and claim using actual numbers. Published aggregates can reveal individuals when filters overlap.7 min
- Optional: uncertainty, log scales and plotting codeError bars can show very different things, so each must be named. On a log axis, equal distances mean equal multiplication factors.7 min
Practise this week · Lab 6: Exploratory Data Analysis7 guided sections · Count the right unit, describe distributions, calculate correlation by hand and explain a Simpson reversal.
Key terms
- Mark
- The point, bar, line or area that represents a record or a summary.
- Channel
- A visual property of a mark, such as position, length, angle, area, colour or shape, that carries a value.
- Visual encoding
- The rule linking each data field to a channel.
- Truncated axis
- A bar-chart axis that starts above zero and cuts off part of every bar.
- Focused axis
- A narrowed line-chart range chosen to reveal small changes, with labelled limits.
- Pie chart
- A circle divided into slices whose angles show shares of one total.
- Bubble chart
- A chart whose circle areas encode values.
- Area scaling
- Making a circle's area, not its radius, proportional to its value.
- Ordinal category
- A category with a natural order, such as Low to High.
- Jitter
- Small, disclosed random offsets added only for display so overlapping points become visible.
- Small multiples
- A grid of small charts that share axes, one per group.
- Percentage point
- The unit of an absolute difference between two percentages.
- Relative change
- The difference divided by the starting value.
- Dashboard flag
- A highlighted change that prompts investigation rather than proving a cause.
- Half-open interval
- An interval such as [40,60) that includes its left end and excludes its right end.
- Bin width
- The length of each histogram interval.
- Proportion
- A count divided by the total number of observations.
- Density
- A count divided by (total observations × bin width), so that area equals proportion.
- Area principle
- In a density histogram, height × width gives the share of observations in a bin.
- Flag
- A point drawn separately because it lies beyond a fence.
- Gap
- A break in a line where no value was recorded.
- Fixed cohort
- The same group of people followed across time.
- Categorical palette
- Distinct hues for unordered groups.
- Sequential palette
- Pale-to-strong shades for ordered values.
- Diverging palette
- Two hues moving away from a meaningful midpoint such as zero.
- Redundant encoding
- Showing the same information with a second cue such as labels or marker shapes.
- Highlighting
- Drawing one series in a strong colour and the others in grey.
- Common scale
- Identical axis ranges across panels so that positions can be compared.
- Pooled rate
- Total successes divided by total cases across groups.
- Grouped bar chart
- Bars for each subgroup placed side by side within each category.
- Differencing
- Learning an individual's value by subtracting two overlapping published totals.
- Minimum cell size
- A rule that suppresses or merges groups too small to publish.
- Chart critique
- A review that cites specific numbers, scales and labels.
- Standard error
- The estimated uncertainty of a sample mean, s/√n.
- Confidence interval
- A range for a population mean, such as mean ± t × SE.
- Log scale
- An axis that places values by their logarithm, so each equal step multiplies by the same factor.
Practice this topic84 questions · earn XP and build your streak