Economics, Business & Finance: Chart deception analysis

What you'll be able to do: Find what a chart's construction is doing to its argument, even when every number on it is accurate, which is the difference between reading a graph and reading it critically.

Why this is an economics problem specifically

A graph in this subject is never decoration. The slope of a supply curve is a causal claim. The choice of what to put on each axis, where the axis starts, and what date range to show is a set of decisions with an argument built into each one, and a chart can be completely accurate, every number correct, and still misleading by construction. This is not a subject where you're hunting for fabricated data. It's a subject where honest numbers get arranged to imply a conclusion the numbers don't actually support.

The standard tricks, by name

Truncated axis. A bar chart of a tax rate moving from 35% to 39% looks dramatic if the y-axis starts at 30 and modest if it starts at 0. Same data, opposite visual impression.

Nominal instead of real. A chart of GDP, wages or house prices "rising steadily" over decades in nominal terms can be roughly flat, or falling, once adjusted for inflation. Whichever version supports the story gets shown.

Cherry-picked date range. An "inflation is coming down" chart that starts right after the peak looks like a trend; extend it back before the spike and the picture changes completely. The same trick runs in the other direction for "the economy is collapsing" charts that start at a recent high.

Totals instead of per-capita or per-GDP. National debt charts showing a rising total, with no reference to a growing population or a growing economy to service it, make every country's debt chart look alarming, because totals almost always rise.

Mismatched dual axes. Two lines on one chart, each on its own scale, can be made to cross, diverge or track each other by choosing the two scales, independent of whether the underlying quantities have any real relationship.

Correlation dressed as causation. A scatterplot with a trend line implies a causal story the chart itself provides no evidence for; nothing in the picture rules out a confound or reverse causation (see method 09).

Stacked areas that hide a shrinking share. A category can shrink as a proportion of the whole while the chart shows its absolute value rising, because the total grew faster, the stacked area looks stable or growing when the honest claim is decline.

The test

Give the AI a description of a real or hypothetical chart and ask what construction choices are doing work.

Here's a chart: [describe or paste it, what's plotted, the axis ranges, the date range, nominal or real terms]. What choices in how this was built would change the visual impression if made differently, starting the axis at zero, using real instead of nominal terms, extending or shifting the date range, or switching totals to per-capita? For each, say which direction the chart would shift if that choice were reversed.

The output you want is not "this chart is misleading": it's a specific list of construction choices and, for each, which way an honest alternative would move the visual argument.

Worked example

A chart titled "Corporate profits at record highs" showing the dollar total of aggregate profits rising over twenty years, axis starting at zero, nominal dollars.

Reconstructed: in real terms, per unit of GDP, the picture is a share that has moved within a historical range rather than broken out of it: "record" is true of the nominal total and not obviously true of the profit share of the economy. Neither chart is fabricated. They support different claims, and only one of them is the claim the headline makes.

Across the disciplines

Macroeconomics. Debt-to-GDP versus raw debt totals; real versus nominal wage growth; seasonally adjusted versus raw unemployment figures, three pairs where the "wrong" version is a normal, common way to present data, not a rare trick.

Finance. A fund's returns chart starting at its best historical month; a stock chart on a linear scale over a multi-decade span, which visually crushes early growth that was proportionally just as large as recent growth, log scale is usually the honest choice for long price histories, and it's routinely omitted because it flattens the dramatic recent slope.

Marketing. A "customers love us" chart of satisfaction scores that omits the sample size or response rate, letting a self-selected, enthusiastic minority stand in for the customer base.

Management. A performance dashboard with a y-axis rescaled each quarter, so every quarter's bar looks similarly tall regardless of the actual trend underneath.

Pitfalls

1. Treating "misleading" as "fabricated." Almost none of these charts contain a wrong number. The deception is entirely in the construction.
2. Fixating on one trick and missing the others. A chart can combine several, nominal terms and a cherry-picked start date and a truncated axis, stacked for effect.
3. Assuming the "honest" version has no argument of its own. Real, per-capita, zero-based, full-range charts are more informative defaults, not neutral ones, every construction choice implies something.
4. Stopping once you've spotted the trick, without asking what the honest chart would actually show. The reconstruction is the useful half.
5. The tell: you can name the trick in a chart someone else made and have never checked one of your own for the same thing.

Try this today

Find a chart from a news article about the economy, a company earnings report, or a policy debate. Identify the axis range, whether it's nominal or real, and the date range shown. Ask what the chart would look like with each of those choices reversed, and whether the headline claim survives.