MEASURING output is the best way we have of taking the temperature of an economy. But the industry standard, gross domestic product (GDP) has a host of weaknesses. It is reliant on an arbitrary definition of what is productive, so it includes childcare by nannies but not by house-husbands and wives. It takes no account of who is doing the producing, meaning an economy could have a single worker or full employment. It ignores the underground economy to a large extent, guaranteeing that production always undershoots reality. But more than any of these, GDP is extremely difficult to measure.

In July 2015 America’s Bureau of Economic Analysis (BEA) announced that GDP had grown by 2.3% in the second quarter. This was below expectations; US stockmarkets slipped in response. But in August and September, the BEA issued revisions that pushed growth up to 3.7% and then to 3.9%. Suddenly, America’s economy was seen to be roaring ahead again, even though nothing extra had been produced. The BEA publishes its first estimate for the last quarter of 2015 on January 29th. Commentators believe it will show that the economy slowed to around 1%. But that figure is almost certain to change, possibly significantly, in future releases. The actual performance of the economy is likely to remain unclear.

The problem is not with the BEA. Among national statistics offices, studies have shown that its revisions are no larger than average. The issue is the near-impossible task which it has been set. Economic data have to be both timely and reliable. Yet these attributes often seem mutually exclusive. Data produced quickly tends to be inaccurate; compiling numbers meticulously takes time. Then are the revisions themselves. These take two forms. First, the data used for initial estimates are based on partial information and trend projections. When more actual numbers begin to trickle in, the aggregate result is changed, as was the case with the BEA’s assessment of second-quarter growth in America. Second, statistics offices habitually review how they compile GDP statistics, to reflect the new sources or concepts. Messing around with a methodology can have major effects: the BEA recently revised down GDP growth in 2011-14, removing $70bn from the economy in an instant. Consequently, America’s recovery from the global financial crisis now looks less impressive.

Chunky revisions to GDP are a thorn in the side of government and business alike. Central bankers typically look 18 months into the future (and so are less concerned with the recent past), but for those working to balance the government’s books, sudden shifts in the level of output are awkward. A narrow surplus can become a wide deficit if the size of the base suddenly shrinks. Companies rely on headline economic indicators to make investment decisions and forecast likely demand. The solution might be a shift in mentality. GDP growth is treated as if it is set in stone, even as it is regularly revised. This is unhelpful. The BEA has recently encouraged broader thinking, by publishing an average of GDP and gross national income, another measure of output that ought to move in step with GDP. It often doesn’t. But by considering the two in tandem, measurement errors in both carry a lighter weight. There is no perfect way to capture the state of an economy. This approach is the best that exists.