Making News Headlines out of Statistical Noise

The press story around the newest CPI figures is excitable and significantly over-interprets the data.

"UK inflation figures showed a surprise upward change between May and June, CPI rising 3.6 per cent" says The Standard in yesterday's lede on the inflation story. But this massively misinterprets the reality.

First, CPI is presented usually as a rolling annual figure, so it compares prices with the same time last year. But I suppose that a headline screaming 'Prices surge by 0.2% in a month' wouldn't cause such frothing.

Second, the CPI doesn't measure prices per se, but only samples prices across a selection of locations and items, introducing sampling bias, measurement errors and statistical uncertainties. This is the reason that the published figures are always rounded to the nearest decimal place, but even so the figures are often revised in subsequent months as more data comes in.

Third, and this is inexcusable, the published CPI figures are calculated from the rounded figures, compounding the rounding errors from previous data. This is in contravention of international standards, and it adds more unneccessary innacuracy to the published headline numbers.

The rounding errors alone introduce a plus or minus error of up to 0.05% for each index figure, which is doubled when accounting for typical revisions to data, to say nothing about the sampling error and biases, and compounded again when comparing two index figures.

Which is to say, that when the CPI jumps from 3.4% to 3.6% it might represent a real increase, or it might just be a statistical artefact of the data and calculation processes.