Here are two numbers from the same government table, covering the same subject, in the same year. Parental responsibility: 553. Parental responsibility: 5. Both sit in the Ministry of Justice's family court figures for 2023 (Source: Ministry of Justice, Family Court Statistics -- Family_Court_Tables__Jul-Sep_2024_ -- Table_3).

Feed those two into any automated comparison and it spits out a rise of 10,960%. Which is arithmetically correct: 553 is 110 times 5. It is also completely meaningless, because there is no year gap between them. Nothing surged. Nothing collapsed. Two different lines in one spreadsheet happen to carry a similar label, and a machine did what machines do.

I'm writing about it anyway, because this is how bad statistics get born. Somewhere, right now, a dashboard is rendering that percentage as a red arrow pointing skywards. Somebody will screenshot it. Somebody else will build an argument on top of it. The original table never claimed a 10,960% anything. The claim was manufactured in transit.

Parental responsibility orders are not a trivial corner of the justice system. They are the legal mechanism by which an unmarried father, a step-parent or a second female parent gets recognised in law as a parent: the right to be consulted on schooling, on medical treatment, on where a child lives. For the families involved, it is the difference between being a parent on paper and being a stranger with a school run. A few hundred of these pass through the family courts, and the exact count depends entirely on which category you're reading and how the courts have coded it.

That's the real problem. The label alone doesn't tell you which is which. To understand why one row reads 553 and another reads 5, you need to open the workbook, find the notes, work out whether you're looking at applications or orders, first instance or appeal, one order type or a subcategory of it. Most people will never do that. They will take the top-line figure, or worse, take a percentage someone else calculated from it.

Add the lag. The most recent annual figure here is for 2023, published in a quarterly release covering July to September 2024. By the time anyone builds policy on it, the family court has moved on by the better part of two years. Slow data with ambiguous labels is a gift to anyone who wants to argue anything.

The Ministry of Justice is not doing anything unusual. Quarterly spreadsheets with dozens of tabs and footnotes buried three sheets deep are the standard format across Whitehall. That format was designed for statisticians. It is now consumed by scrapers, chatbots and news aggregators that cannot read footnotes.

So a small ask, aimed at every department publishing numbers: label the rows so a stranger can tell them apart, and put the base year next to every figure. Until then, the rule for the rest of us is simple. Before you believe a percentage, find out what it was a percentage of. In this case, it was five.