Exercise · Intermediate
Every number on it is correct
A cluster presentation ranks twelve communes on malnutrition, colours the worst three red, and truncates the axis so the gap looks decisive. Nothing on the slide is factually wrong. Find the four decisions that made it dishonest and produce the version that is not.
A slide is going to the nutrition cluster tomorrow. It shows a bar chart of global acute malnutrition across twelve communes, sorted worst to best, with the top three bars in red and the rest in grey. The y-axis runs from 5% to 16%. The title reads “Three communes in crisis”.
Every number plotted is correct. You have been asked to check it.
The file
muac-screening-artibonite-2024.v1.csv — 4,218 screenings across twelve communes.
What to find
One: what the axis is doing. Recompute the chart with a zero-based axis and describe, in one sentence, how much of the apparent gap between the top three and the rest survives.
Two: what the colour is doing. The slide uses red for three communes. Say what red means to this audience, and whether these three communes are that.
Three: what is missing. The chart carries twelve point estimates and nothing else. Compute the 95% intervals and the number of screenings behind each commune, then say which of the slide’s claims survive.
Four: what the title asserts. “Crisis” is a word with a definition in this sector. Find it, check the three communes against it, and say whether the title is supportable.
Produce
The corrected figure, with a caption in the shape lesson 6 requires — the finding, the denominator, and one thing the chart does not support. Plus the one sentence you would say when you present it in place of “three communes in crisis”.
Check yourself
| Expected | |
|---|---|
| Communes | 12 |
| Screenings, lowest and highest commune | 189 and 768 |
| Highest GAM | Anse-Rouge, 15.3% |
| Lowest GAM | Dessalines, 5.6% |
| Adjacent pairs with overlapping intervals | 11 of 11 |
| Communes whose lower bound clears the district median of 8.0% | 3 |
| Anse-Rouge interval | 10.9% to 20.6% |
| Widest interval | L’Estère, 10.0 points |
| Narrowest interval | Gonaïves, 3.8 points |
If your intervals are symmetric around the estimate, you have used the Wald formula; use an exact binomial interval instead, and say why in your answer.
Three questions
Two sentences each.
1. The three highlighted communes turn out to be the right three to highlight, and the slide highlighted them for the wrong reason. Explain the difference, and say what would have happened on a dataset where the two criteria disagreed.
2. The axis runs from 5% to 16%. State the magnification that produces, and say why a bar chart in particular cannot survive it.
3. Your corrected chart is less dramatic than the original and says something the original did not. Write the sentence you would use to introduce it to a cluster coordinator who has already seen the first version.
The point
Nothing on the original slide is false. The bars are in the right places, the communes are correctly ranked, and the three communes at the top really are the three with the highest prevalence. A fact-check finds nothing.
Four decisions did the damage and none of them is a number. The axis magnified a gap. The palette borrowed a meaning from a severity scale. The missing intervals turned an uncertain ordering into a league table. And a word with a technical definition was used for its everyday one.
The corrected chart makes a smaller claim and can be defended. Which is the trade this course exists to make explicit: a chart that says less than the data supports is a bad chart, and one that says more is a worse one.