You do not need to be a statistician. You need to be willing to click one link further than the person who wrote the article. 

Why bother checking at all? 

The cost of being wrong is lopsided, and that is the whole of the case for bothering. A misunderstood statistic about market participation costs you nothing. A misunderstood statistic about what a platform charges, or about how often people in your position lose money, feeds directly into a decision involving your money. And because checking is much faster than people imagine. The procedure below takes about ten minutes and catches the large majority of errors in circulation, most of which are not fabrications but misreadings that have been passed along by people who never opened the source. 

Step one: is there a source at all? 

Look for a named document, not a named organisation. According to the FCA is not a source, it is a gesture. FCA Financial Lives 2024 is a source, because it names a specific publication you can find. If the article names only an institution, or says research shows, you have already found the problem. 

The absence of a specific source does not prove the figure is wrong. What it proves is that the writer did not check it either, which means you are looking at a number that has travelled some unknown distance from wherever it started. Treat it as a rumour with a decimal point. 

Step two: does the source say what the article says? 

Open it. This is the step everyone skips and it catches more errors than all the others combined. 

Use the document’s own search to find the figure, then read the sentence around it and the definition it depends on. Frequently the number is real and the framing has drifted. 

The commonest drift is between related-but-different quantities. Accounts becoming people. Holders becoming active traders. A range becoming a midpoint. Adults becoming investors. None of these requires bad faith and all of them change what the sentence means. 

Step three: what is the date? 

Of the underlying data, not of the article. A piece published last month can rest on a survey conducted three years ago, and in a market where leverage rules changed in 2019 that difference can be decisive. Pre-intervention figures describe a different product. 

Check for a more recent edition while you are there. Regulators and statistical agencies publish on cycles, and a figure that was current when an article was written may have been superseded twice since. The newer number is sometimes much less dramatic, which is occasionally why the older one is still circulating. 

Six checks, about ten minutes. Any one of them failing is reason to stop using the figure, not reason to conclude the writer was dishonest. 

Step four: does the arithmetic hold? 

Do the division yourself where you can. If a piece says GBP 75 million was lost by more than 90,000 consumers, the implied average is around GBP 833, and if the article characterises that as devastating losses you now know the shape of the distribution matters more than the headline. The arithmetic often tells you something the prose does not. Percentages are where errors cluster. Check whether components sum to something sensible, whether a percentage change has been confused with a percentage point change, and whether a rate is being applied to the right base. These are ordinary mistakes made by tired people, and they survive because readers assume somebody checked. 

Step five: who benefits from this number being believed? 

Not a reason to reject it, and a reason to check it harder. A broker’s own figure about customer satisfaction, a regulator’s figure about harm prevented by its own intervention, and a research site’s figure demonstrating the value of research, all warrant an extra look. 

Having an interest in a number does not make it wrong. It makes optimism at the margins a good deal likelier. 

Apply it symmetrically. Figures that confirm what you already believe deserve the same scrutiny as figures that annoy you, and in practice receive far less. That asymmetry is the largest single source of persistent misinformation among otherwise careful people. 

When the primary source is the problem, and what a passing one looks like 

The procedure leads you accurately to a flawed number, and there is no sixth step that rescues you. If a regulator’s survey has a sampling problem, or a broker’s disclosure uses an idiosyncratic definition of an active client, the citation trail is intact and the figure is still wrong. Verification establishes provenance, not truth. 

Selection is the second gap and the larger one. A perfectly sourced, perfectly dated, perfectly defined figure can still be one of ten the writer looked at and the only one that supported the argument. Nothing detects what was left out, which is why reading the whole source rather than the quoted line is worth more than the checklist above and gets done far less often. 

A source that passes names the document, gives the collection period, states the unit, and survives your opening it. Compilations can pass too, provided each figure carries its own provenance. The HMRC, FCA and Ombudsman Service releases behind the UK trading statistics compiled by The Investors Centre are named and public, which is the only reason any of it can be checked by a reader rather than taken on trust. Run these five steps on any figure you intend to rely on, including theirs. 

Which numbers deserve the ten minutes? 

Not all of them, and trying to check everything is how the habit gets abandoned by Wednesday. Spend it on the figures that change what you do: what a platform charges, what proportion of people in your position end a period down, what a tax wrapper is actually worth. Leave the ones that are decoration, and most participation statistics are decoration. 

A workable rule for deciding: if you would repeat the number to somebody else, check it first. Repetition is what gives a figure a second life, and the person you repeat it to will treat you as the source rather than as the sixth link in a chain. That is how a misreading acquires a career. 

 …