MisleadingCharts
Back to the gallery

The two 89%s that a second axis drew the same size

Showing the misleading chart

Be the first to star this exhibit

A sourcing slide we drew for an invented flavour house plots U.S. vanilla imports on two y-axes: the price per kilo on the left, tonnes on the right, each axis fitted to its own line. From the 2018 peak to 2024 the price is down 89% and the tonnage up 89%, so demand looks like it answers one-for-one. The two 89%s are not the same size: the price was divided by 8.9, the tonnage multiplied by 1.9, and the bill fell 79%.

01The claim

Since the 2018 peak, vanilla’s import price has fallen 89% and America’s vanilla buying has risen 89%: the tonnage answers the price move for move, so the elasticity of demand is −1.0. A sourcing slide like this one, which we drew for an invented flavour house from real UN Comtrade figures, uses that to argue that bean savings should be passed straight through to extract list prices, because every 1% off the price wins 1% more volume.

02The trick

Every number on the slide is right. The price went from $467.82 a kilo in 2018 to $52.79 in 2024, down 88.7%; the tonnage went from 1,320 t to 2,494 t, up 88.9%. The trick is that two y-axes let each line choose its own ruler. The left axis runs $0 to $500 and the right one 1,150 t to 2,550 t, each fitted snugly around its own series, so both lines are stretched to fill the same box. Over these thirteen years the price spans a 20.9-fold range and the tonnage a 2.05-fold one, and the slide gives both exactly the same height. That is what makes the X. The matching 89%s finish the job, because a percentage fall and a percentage rise are not measured on the same ruler. A fall of 89% divides by 8.9; a rise of 89% multiplies by 1.9. To undo the price fall would take a rise of 786%. The −1.0 elasticity is one of those percentages divided by the other, so it inherits the mismatch. Take the log ratio instead, the usual way to measure an elasticity between two years because it treats ups and downs alike, and the same endpoints give about −0.29. A second axis did not invent the inverse movement: the two series do tend to move opposite ways (r ≈ −0.69 across the thirteen years). What it invented was the size of the response. With two free rulers, any pair of series can be drawn to swing equally far, and the eye reads equal swings as equal-sized moves. The window helps: 2018 is the price peak, and 2024 is the tonnage high, the year after its low.

03The fix

Put both series on one ruler. Divide each by its own 2018 value and plot both on a single logarithmic axis, where equal distances mean equal ratios. The tonnage line now stays between 0.92 and 1.9 times its 2018 level and finishes at ×1.9. The price line drops to ÷8.9, about 3.4 times as far. Then multiply the slide’s own numbers, because price times tonnes is the bill: 0.11 of the price times 1.89 of the tonnage is 0.21 of the spending. At a true −1.0 that product would be 1 and spending would stay flat. Instead it fell 79%, from $617.5M in 2018 to $131.7M in 2024. The squeeze years point the same way. From 2015 to 2018 the price rose 7.0-fold and tonnage fell 37%, so American buyers paid 4.4 times as much ($139.7M to $617.5M) and did not walk away, a log ratio of about −0.24. On these numbers U.S. buying looks far less responsive than −1.0, which undercuts the slide’s read-out. Keep that conclusion modest, though. A unit value mixes grades, and 2024 came straight after the series low of 2023, which looks more like buyers rebuilding stocks than a demand curve. A year on, 2025 came in at $50.39 a kilo (−4.5%) and 1,933 t (−22.5%): the tonnage moved a fifth while the price barely moved. To show the levels honestly, give each quantity its own small chart on its own zero-based axis, in its own unit. To show the comparison, use one ratio axis. Neither needs a second y-axis. Whenever a slide puts a price and a quantity on two axes and reads a slope off the picture, ask what each axis multiplies by from bottom to top.