Statistics are usually defended as the cure for bias and guesswork, the hard ground under a decision. The contrarian case is that numbers are made by people, with choices and incentives, and that a society which treats data as automatically objective has simply found a more authoritative way to be wrong.
A trust-in-data question rewards a student who can defend evidence-based reasoning and still expose how numbers mislead. Data really is the best guide we have over hunch and anecdote. Argue where that breaks down: when figures are gamed, when studies fail to replicate, when AI fabricates plausible data, and when 'the numbers say' becomes a way to dodge a value judgement.
The reliability-of-evidence strand recurs in shifting forms. It appeared as statistics guiding the future in 2020 and as scientific advancement breeding complacency in 2021. Underneath both is the same question: how much weight a data-driven society should put on its own numbers.
The current pressure is real. In 2026 generative AI can fabricate plausible figures, citations and studies at scale; the replication crisis has already shown that a large share of published findings do not hold up; trust in institutions and their statistics is contested in election after election. The ground under 'the data shows' is shakier than the phrase suggests.
Framed honestly: SEAB sets the paper, and nobody outside it knows the wording. A question on data, statistics or scientific reliability is a recurring type rather than a certain return. It is worth rehearsing because the contrarian move, defending data while exposing its failure modes, is one most candidates never make.
set at A-Level most recent appearance. Recurs as the reliability-of-evidence strand. The vehicle changes from statistics to science to data; the trust question stays.
These questions test whether you can value evidence without worshipping it. The lazy answer either trusts numbers blindly or rejects them as 'lies, damned lies and statistics'. The strong answer holds that data is the best tool we have and that tools can be misused, mis-measured and gamed, then argues how to tell good data from bad.
They also test whether you understand how a figure is made. A statistic carries choices: what was counted, how it was defined, who funded it, what was left out. A student who can interrogate the production of a number, not just quote it, is doing the exact thinking the question rewards.
Two distinct, defensible routes through the question. A strong script commits to one and uses the other as the concession it answers, rather than sitting on the fence.
For all its flaws, evidence-based reasoning has outperformed intuition and tradition across medicine, policy and planning, so the answer to bad data is better data, not a retreat into gut feeling.
Data is not found but made, with definitions, incentives and errors baked in, and in a world where studies fail to replicate and AI can manufacture plausible figures, treating numbers as automatically objective is its own kind of credulity.
The replication crisis, first named widely after the 2015 Reproducibility Project, has since prompted reforms such as pre-registration and open data, so the failure of trust also drove the method to correct itself.Source: Open Science Collaboration and subsequent reform, as of 2026-06
Generative AI can now produce fabricated statistics, fake citations and plausible-looking studies, which lowers the cost of manufacturing convincing false data and raises the burden on the reader to verify.Source: general technology reporting, 2026, as of 2026-06
Goodhart's principle, that a measure stops being a good measure once it becomes a target, explains why data-driven targets so often get gamed rather than met, from exam metrics to hospital waiting times.Source: established statistical principle, as of 2026-06
A number is not a fact. It is a choice about what to count, dressed in the authority of arithmetic.The contrarian case on data
Book a free trial lesson and see how we teach GP from the exam backwards, building the two-stance habit on issues exactly like this one.