Rung 3 — What's Related

Five dives, built and ready to run. Data in data/, rebuildable with python3 fetch.py.

Year 11's question is can you judge what other people did, so every dive here is two charts of ONE dataset that disagree. A lone chart can only be described; judgment needs a comparison, and the district bank never supplied one.

Q1 the trap. Which of these two chose its baseline, bins, or window to make its point? Q2 the finding. Where do they disagree? Q3 the turn. Which would you send to somebody you respect?

Nothing on this page is a fake chart. Both renders are always honest renders of the same real file. That is what makes the rung hard.

---

3.1 · the choropleth against the proportional symbol

pair_ca_counties_area_population_aqi.csv — 58 California counties: population, land area, density, unhealthy-air days. Census + EPA, 2024.

Draw it twice. Shade each county by unhealthy-air days. Then drop a circle on each county sized by the same number.

Q2. Which counties dominate the shaded map? Which dominate the circle map?

A choropleth spends its ink on land, and land does not breathe.

Nearly half the map is about one Californian in twenty.

San Bernardino County gets 4.9 times the ink of Los Angeles County — 20,068 square miles against 4,060 — while holding 23% of its population. It is the largest county in the United States. On a shaded map of California it is the loudest thing on the screen.

The same file also carries 3.4's pair: plot density against unhealthy days as a scatter, then look at the choropleth again. The scatter shows fifty-eight points with a relationship; the choropleth shows a shape of California. The unit hides everything happening inside it — Los Angeles County alone runs from the coast to the desert.

---

3.2 · one series, two windows

window_global_temperature_annual.csv — NASA GISTEMP, 1880–2025. window_15_year_trends.csv — every 15-year window in the record, with its trend.

The full record warms at +0.083 °C per decade. Then hand them the second file.

Q1. Every one of those is the same thermometer record, honestly computed. Somebody choosing the window can produce anything from −0.249 to +0.384. A sign flip, or four and a half times the real trend. Q2. Find the fifteen years you would quote if you wanted to argue warming has stopped. Now find the fifteen you would quote to argue it is accelerating. Both are in the file. Q3. You have now cherry-picked, twice, on purpose. What would you need to see in somebody else's chart to catch them doing it?

A correction worth teaching from. This dive was designed around the famous 1998–2012 "hiatus" — and the data refused. In current GISTEMP that window runs +0.136 °C/decade, above the full-record trend. Later revisions and thirteen more years overtook the famous cherry-pick. The lever is real; the folk example is stale. The dive is stronger for letting students find their own window than for handing them somebody's.

---

3.3 · the box plot against the bar of means

spread_la_311_daily_by_month.csv — 365 days of LA 311, with a month column.

Draw a bar chart of the monthly means. Then draw a box plot of the same 365 days.

Q2. Which month was strangest in 2022?

On the bar of means, September ranks seventh of twelve at 3,620 — the middle of the pack, nothing to see. December is the lowest month and nobody would look twice.

On the box plot, September has a point at 95 against a month whose maximum is 5,754.

That 95 is the twenty-two-hour outage from rung 2. A whole session's finding — the hole and the spike that were one event — is completely invisible on the bar of means, and it is invisible not because anyone lied but because an average is a machine for removing exactly that.

Q3. You did the outage dive. Somebody who only ever saw the monthly bars did not, and never could. What else has been averaged away from you?

A mean is a summary, and every summary is a deletion. This is the cheapest way to show it, because they already know what was deleted.

---

3.4 · covered by 3.1

Density against unhealthy air, as a scatter, beside the same counties as a choropleth. Same file, and the note is in 3.1. The canopy-and-heat version named in the bank remains unbuilt — it needs a tract-level join and Census ACS wants an API key.

---

3.5 · a real correlation, and a nonsense one

correlation_co2_temperature_exoplanets.csv — 1992–2025, three real series from three real publishers: NOAA CO2, NASA GISTEMP, the NASA Exoplanet Archive.

Two scatters. Same x axis.

Q1. One of these is physics. The other is that both numbers happen to have gone up since

1. From the correlation alone, which is which?

Q2. The nonsense one is nearly as strong. How much weaker would it have to be before you stopped believing it — and is that a number or a feeling? Q3. Somebody shows you r = 0.9 for a claim you already agree with. What do you ask them?

This is the rung's anchor. Every student arrives believing a strong correlation is evidence. Here are two, a hundredth of a point apart, and one of them is that humanity got better at finding planets while it got better at burning carbon.

No mechanism, no argument. A correlation is a question, not an answer.

---

The MAKE dive

One dataset. Two honest charts that argue opposite things. Then one paragraph: which is honest for what question, and which you would sign your name to.

window_global_temperature_annual.csv and pair_ca_counties_area_population_aqi.csv both support this with no cleaning. Flourish, per ../../_planning/the-two-tools.md — all public data, nothing collected from a student.

The trap to expect: most tables will produce one chart they think is honest and one they think is a lie. Push back. Usually both are honest and they answer different questions, and the ones that are genuinely indefensible are indefensible for a reason the student can name. That distinction is the whole rung.

---

Not built, and why

Provenance

SOURCES.md. Publishers this rung adds: NASA GISTEMP, US Census Bureau (county population estimates; the Gazetteer land-area file). All keyless bulk downloads.