Walk the garden

One tick in the row

Each choice you made above was defensible, and so were all the alternatives: five target years; three ways of binning studies through time; an optional interval around the target year to account for estimates referring to a temporal range rather than a specific year; two overlap rules; and six possible measures of spread. Together, these choices generate 360 analytical pathways for exactly the same question. Whatever your pathway concluded, a colleague making different but equally defensible decisions could have concluded otherwise.

This is Borges' Garden of Forking Paths, brought into scientific discourse by Gelman and Loken and walked here through a multiverse analysis in the sense of Steegen and colleagues. When several analytical pipelines are defensible, run them all and let the distribution of answers—not a single answer—be the finding.

The strip above shows the shares reported by the DAWN team in their recent Nature Water paper (Puy et al. 2026): only 17% of pathways find uncertainty in global irrigation water withdrawals declining (10% once projections to 2100 are excluded), while 40% find it increasing and 34% find no stable trend. In other words, the most robust outcome of fifty years of research has not been less uncertainty, but more of it. This raises a more fundamental possibility: there may be no single “true” value toward which estimates can converge. Progress may instead require purposefully expanding uncertainty until it reaches the boundaries of what can be known about global irrigation water use given current data, models and assumptions.