Read a market on evidence, not on the first chart you find.

Free public data is genuinely good at three things: showing you direction (is a market growing or shrinking), relative size (is this country's trade volume larger or smaller than that one's), and structure (which sectors, partners or demographics dominate). Every source in the ledger below can do this reliably, because it's what national statistical offices, central banks and multilateral bodies are built to publish.

It is much weaker at three other things it's frequently asked to do anyway. First, precise point-in-time figures for a narrow slice (a specific city, a specific product code, a specific month) — the finer you slice, the more a single source's sampling and reporting lag start to matter. Second, the informal economy — surveys and administrative filings systematically undercount activity that never registers with a tax authority or customs post, which matters enormously in markets where informal trade is large. Third, causal explanation — a data source can show you that two things moved together; it cannot, by itself, tell you that one caused the other.

The discipline this page argues for is simple to state and easy to skip under deadline pressure: know which of the three things you actually need before you go looking, and treat any single number as a claim to be checked, not a fact to be repeated.

Sampling bias is the simpler trap: a survey or dataset systematically over- or under-represents part of the population it claims to describe. India's household consumption and labour-force surveys, for instance, sample households — which is the right method, but means the informal, unregistered edges of the economy are structurally harder to capture than formally employed, urban respondents.

Survivorship bias is subtler and easy to miss because the data that would correct it has, by definition, already vanished. The canonical illustration is Abraham Wald's 1943 work for the U.S. Army's Statistical Research Group: asked to recommend where to add armor to bombers based on the damage pattern of planes returning from missions, Wald argued for reinforcing the spots with the fewest hits — because planes hit in the engines or other critical areas were the ones that never made it back to be counted. The damage visible in the surviving sample was, precisely because of survivorship, the least informative part of the picture.

Worth being honest about the story itself: Wald's own 1943 memorandum doesn't use the phrase "survivorship bias" or tell the vivid "bullet holes" anecdote in the form it's usually retold today — that framing is a later, simplified reconstruction. The rigorous technical treatment of what Wald actually did is Mangel & Samaniego's 1984 paper in the Journal of the American Statistical Association (source 9 below). The lesson still holds even with the popular telling softened: when you only observe the survivors of a market, a channel, or a customer segment, ask what left the sample before you could measure it.