Aim

To draw a bar graph of India's CWG-2018 medal tally (Gold 26, Silver 20, Bronze 20, Total 66) with matplotlib and to read it using sound perceptual principles.

CO Mapping: CO1, CO2, CO3

Theory

A bar chart encodes each category's value as the length of a bar rising from a common zero baseline. Cleveland & McGill's classic graphical-perception experiments ranked position along a common scale and length as the judgements humans decode most accurately — far better than angle (pie charts) or area/colour. That is why bars are the default choice for comparing quantities across categories.

Two integrity rules follow directly:

1. The value axis must start at 0. Length is the encoding; truncating the axis (say, starting at 15) would make Gold look several times Silver instead of 1.3×. The distortion is measured by Tufte's lie factor — visual effect size divided by data effect size, which should stay ≈ 1.
2. Do not mix parts with wholes carelessly. "Total" (66) is the sum of the other three bars. Plotting it alongside them is acceptable in a scoreboard, but a reader must recognise it is a different kind of quantity — otherwise the chart appears to claim Total is simply the best category.

Colour here works as identity encoding: gold, silver and bronze hues match the medals, and steel-blue marks Total as different — a legend-free labelling trick, not decoration.

Dataset

Procedure

1. Build the medals DataFrame with columns Type and Count; print it as the data record.
2. Select the non-interactive backend with matplotlib.use("Agg") before importing pyplot, so the script also runs on headless/online compilers.
3. Call plt.bar(medals["Type"], medals["Count"], color=[...]), passing four colours — one per bar, in data order.
4. Add title, xlabel ("Medal Type") and ylabel ("Count"); call plt.tight_layout().
5. Save with plt.savefig("cwg_medal_tally.png"); the try/except prints a friendly message if no plotting backend exists.

Interpretation of Results

Gold (26) stands tallest of the three medal bars — India converted finals into wins at a high rate, with silver and bronze tied at 20. Because every bar rises from zero, the 26:20 ratio is directly readable as a length ratio. The Total bar towers at 66 but conveys no new information — it is a redundant encoding of the sum; an annotation ("Total: 66") often serves better than a fourth bar. Analytical reading of any bar chart means: compare lengths, check the baseline, and ask whether every bar answers the same question.

Common Mistakes

1. Truncating the y-axis to "zoom in", which visually exaggerates small differences — the cardinal sin of bar charts.
2. Reaching for a pie chart: the angles for 26/20/20 are nearly indistinguishable, while bar lengths are compared trivially.
3. Importing pyplot before setting the Agg backend on a server, which crashes with a display error.

🎯 Viva Questions

1. Why must bar charts start at zero? Length is the encoding; a cut baseline distorts every length ratio.
2. Which perceptual channels are decoded most accurately? Position on a common scale, then length (Cleveland & McGill).
3. Why is the Total bar special? It is the sum of the parts — a whole plotted next to its own components.
4. What does the Agg backend do? Renders figures to raster files without a display — ideal for headless environments.
5. When would a horizontal bar chart be better? Long category labels or many categories.
6. What is Tufte's lie factor? The ratio of the effect shown in the graphic to the effect in the data; honest charts keep it ≈ 1.