This has been a cruel month for statisticians. On June 3, the US Bureau of Labor Statistics admitted it had messed up the weights for its Current Population Survey, affecting the employment, unemployment and labour market participation figures for April. The following day, the UK’s Office for National Statistics came clean with its own weighting disaster, in which it had overstated April’s inflation statistics.
The question for users of those statistics is: what should we do? The problem is especially acute for central banks, which make decisions that they boast are “data dependent” and which respond to specific, fine details in economies’ cycles.
Bogus Labour Snapshot
The BLS has been in the wars recently because of increasing concerns about the quality of its data. Although the US statistical agency said its weighting error in the employment figures was minimal and would be corrected in the May data, it was far from its first embarrassment in recent weeks.
The following day, the BLS announced that staff shortages had reduced the quality of the official US CPI inflation data, causing a suspension of price collection in some locations and more data than usual being imputed. The effects would be small, but could “increase the volatility of subnational or item-specific indexes”, it said. Last month it said it would stop collecting and publishing many items in its producer price index. Given the sudden interest in the “taco trade”, the suspension of tortilla manufacturing price indices will hurt.
Painful as these errors and omissions have been for the BLS, its price and employment survey data are still highly regarded. The same should not be said for perhaps its flagship labour market product: non-farm payrolls. Each month analysts predict the gain in jobs numbers to the nearest thousand. The Federal Reserve also watches the figures extremely closely. Then everyone reacts positively or negatively depending on whether these expectations were met or not. By the following month, those figures are forgotten and the process starts again.
The problem is that the BLS’s preliminary estimate for the change in non-farm payrolls is poor. Since 2023, for example, non-farm payroll growth has almost always been revised down. This happens in the following two months after the initial figures are released, and again some time later when the data is benchmarked against a more accurate survey, the Quarterly Census of Employment and Wages. This uses administrative data from US states’ unemployment insurance programmes. The revisions are large, at almost 50,000 a month between the latest data and the first estimate. That is more than a 20 per cent average downward revision to date.
The latest QCEW has again shown jobs growth to be slower in the period between April and December 2024 than in the original non-farm payroll figures, so the downward revisions will grow. Analysts at Barclays reckon that once the latest benchmarking process has been completed early next year, the monthly job gains for the latest year will be revised down from about 150,000 a month to 80,000.
Had those figures been published as a contemporary record last year, there would probably have been even more misplaced concern about rising unemployment and a coming US recession. Barclays says that the most likely cause is a fall in immigration reducing the sustainable pace of jobs growth.
Of course, that judgment is for the Fed to make. But it would benefit from better data in doing so.
Obviously Not Sound
Across the Atlantic, the UK’s Office for National Statistics would love to have the BLS’s problems. With a review into its culture and leadership pending, its chief statistician Ian Diamond quit suddenly in May.
This has not stopped errors and unreliable survey data. The most embarrassing came in the inflation figures for April. Data on vehicle taxation was incorrectly weighted when given to the ONS, but no one appeared to check whether the figures passed the sniff test. Outside observers quickly said they did not and the statistical agency only confessed after the FT highlighted these concerns.
This error will be corrected in the May inflation figures, published on Wednesday. As the chart shows, they are far from trivial.
The surprise about the price data blunder was that it came from the part of the ONS thought to be reasonably well functioning. The known disaster area is the jobs data, where the ONS is battling with a broken Labour Force Survey. This prevents the Bank of England from knowing what is happening to participation in the labour market, where the LFS is the only source of information.
The ONS has recently been boasting that there have been “clear improvements” in the data and survey response rates, shown in the chart below. I’ll leave you to judge whether a 35 per cent response rate for the first wave of interviews, falling to less than 14 per cent by the fifth, is good enough.
If the ONS wants to avoid getting known as “Only Nearly Statistics”, its suggestion to its regulator last week that a badge of quality be removed from another of its products, the Wealth and Assets Survey, didn’t help. The WAS is used to assess how much wealth there is in the UK and who holds it.
Dirty laundry
The UK and the US’s statistical agencies should be praised for airing their dirty laundry so publicly. Economic statistics are getting harder to collect and, as UBS chief economist Paul Donovan says: “Just because some statistical agencies do not publicly admit their errors does not mean the errors do not exist.” China regularly deletes data series it finds uncomfortable, for example.
While not an error, EU statistics can give mad results. The latest GDP growth figures for the first quarter doubled from 0.3 per cent to 0.6 per cent after the first revision. Most of this surprise jump came from Ireland, whose quarterly growth rate was first estimated at 3.2 per cent but then jumped to 9.7 per cent. Yes, you read those figures correctly. And, for US readers, these are not annualised.
It’s all about front running tariffs, particularly in the pharmaceutical sector, alongside the regular problem of Ireland acting as something of a tax haven for US companies “locating” business activity there. The latter does not reflect genuine economic activity in Europe. And as Ireland’s Central Statistics Office highlights, a better measure of underlying activity called “modified domestic demand” grew only 0.8 per cent.
What can we do?
I’ll come back and look at some specific suggestions in future articles, but here is a quick guide to navigating the more difficult world of data we now confront.
-
Do not get excited by or rely solely on a single statistic to make important decisions. In a world of dodgy data, you need to see corroborating evidence and broad trends to take decisions. In monetary policy, that might make you late.
-
Governments should not skimp on funding statistical agencies, which are extremely cheap relative to the costs of data errors. They should also change laws and bang heads together so that the vast quantities of quality administrative data they hold can be used more easily for economic statistics.
-
Do not use one source of data, but seek to extract a common signal from multiple sources. All central banks are now doing this. The models required will differ, depending on the problem that needs addressing, but modern econometrics helps generate unbiased indicators because humans are always prone to cherry-picking from a menu of competing statistics. Examples of this include the FT core inflation indices and the Chicago Fed’s new unemployment nowcast, “Churn”.
What I’ve been reading and watching
A chart that matters
The US CPI inflation figures for May were benign, with little sign of tariffs driving prices higher. The Fed will find this encouraging, but is still likely to think that it is too early to declare that tariffs’ effects will disappear somewhere in the supply chain.
The monthly annualised change in banana prices shot up, as did prices of major appliances and toys. This is far from an inflationary surge in overall prices, but this early sign of aggressive pricing behaviour in a few areas should make us cautious.
Central Banks is edited by Harvey Nriapia
