Diane Coyle is the Bennett Professor of public policy at the University of Cambridge and director of research at the university’s Bennett School of Public Policy. Her latest book is “The Measure of Progress: Counting What Really Matters” (2025).
Coyle is a member of the U.K.’s Industrial Strategy Council and New Towns Taskforce and an advisor to the Competition and Markets Authority. She was previously a vice chair of the BBC Trust. She was awarded a DBE in 2023 for her contribution to economics and public policy.
After a busy day at work as a prompt engineer for a marketing agency — working from your laptop in your own living room — it’s time to relax. You scroll through TikTok for a while, and then upload a video about the novel you discovered recently from a Threads post. After ordering dinner through DoorDash, you ask ChatGPT to recommend an itinerary for your upcoming vacation, and then make travel reservations using Expedia.
Perhaps this seems a strange way of life, but it is hardly a fantasy. Boosted by the pandemic, online shopping has grown from less than 3% to over a quarter of retail sales in the U.K., and in the U.S. it’s up to 16%. Banking and booking travel online are commonplace. Hundreds of millions of people post entertaining or informative content on the internet, contribute open source software or engage in crowdsourced innovation or scientific projects via Zooniverse and other platforms.
But as we consume online, online consumes us. Even if we set aside the much-debated detriments to mental health and social relations, there are significant costs to living life online: It has become infuriatingly hard to get beyond computerized customer service bots that are unable to answer queries, and filling out government forms or wading through badly designed websites can take ages.
Living in a digital realm poses a serious problem for measuring what’s going on in the economy. By definition, “the economy” — that is, gross domestic product or GDP — excludes activities done at home and those involving no monetary transactions. All those do-it-yourself digital activities take some components of the economy across a boundary into the household. Digital services we don’t pay for — social media, smartphone apps, Google searches — have no impact on GDP because their price is zero.
What’s more, these free services are increasingly replacing things that previously did have a price and were therefore accounted for in GDP. As Nobel-winning economist Gary Becker pointed out long ago, shifts of this kind make it difficult to interpret the published figures for national wealth, economic output and productivity.
In any case, it is far from clear that more productivity, as conventionally measured, is always a good thing. Consider the automatic checkouts steadily colonizing stores. They substitute capital and software for human labor, and so will improve the sales per employee figures for the stores. But they rely on shoppers’ free labor instead. Just as with online banking, which often saves time but occasionally doesn’t, we should be accounting for time spent and saved as well as money spent and saved.
Official statistical agencies rarely collect the kind of data that would be useful for tracking patterns of consumption in our ever-more-digital economy. Nor are they able to fit the new patterns into the existing framework of definitions and categories; it’s like trying to do a complicated jigsaw puzzle when the pieces don’t correspond to the picture on the box.
“Statistics matter. They are the only lens available to governments and citizens to understand and monitor what is happening to the broader economy.”
The same is true of the business side of the economy: the production of goods and services. As the U.S. and other countries are finding out, statistics on global supply chains are sparse, although the Trump administration’s tariffs and haphazard economic policies are making it all too clear which imports were crucial. Even overall trade figures are inadequate, as it is only after two or three years that data is published on how a country’s exports rely on imports from a different country.
Official economic statistics struggle with other aspects of production. Companies like Apple or Nike can be misclassified as manufacturers even if they outsource all their production. Some flagship “manufacturing” companies increasingly make profits from services — maintenance or software or otherwise — that they sell alongside their products, like Rolls-Royce with airplane engines or John Deere with tractors. The printer ink phenomenon, which ties customers to an ongoing service, is widespread. This boosts profits, which can of course be measured, but it doesn’t seem to reflect genuine economic value creation.
The statistical fog will only get worse as generative AI use spreads. There are few broad studies on what businesses are using AI for, but those that do use it seem to be more profitable and are growing employment. (So far at least. This may well change.) Yet if, in the not-too-distant future, my personal AI agent conducts affairs on my behalf with my bank’s AI bot, is that valuable economic activity? How should it be measured in any case?
The absence of basic data collection about the whole range of digital activities in the economy is a problem; statistical agencies still often rely on surveys, which are getting fewer and fewer responses as time goes by. A new architecture of data collection will be required if governments and businesses are going to be able to track basic economic phenomena. Unfortunately, the official agencies are being squeezed everywhere, and particularly in the U.S., so they have scant capacity to innovate their approach.
There is, though, a much more fundamental question: What activities in a digitally transformed economy create value, and who profits from that value?
Businesspeople tend to think in terms of “value added,” which deducts costs of production and sales from revenues. Economists want to adjust both costs and revenues for inflation in the relevant prices. Similarly, we economists want to adjust measurements of consumer expenditure when things get more expensive. Adjusting for price changes produces what, in econ-jargon, are known as “real-terms measures.”
These measures are not real at all; they are hypothetical constructs — ideas. The price indices used to produce them are calculated using complicated formulas that involve some dubious assumptions.
“Countries without good statistics are hobbled in their ability to govern or to understand public grievances.”
For example, one of the simpler formulas used asks how much the products purchased last year would cost today. Another looks at the reverse: how the price of products bought today has changed since last year. If the range of products varies over time, as of course it does, all kinds of statistical techniques are applied to remove old items and patch in new ones. But this means that price indices are incredibly hard to interpret over longer periods of time — such as the 18 years since the launch of the iPhone.
All the formulas (the published price indices use complicated alternatives) also make two fundamental, and incorrect, assumptions. One is that all the products in a price index are substitutes for each other. The other is that patterns of consumption do not change as people’s incomes increase. In a reasonably stable economy, these assumptions can perhaps be excused. But not when the structure of the economy is being transformed.
Take the assumption of substitutability. Textbooks will switch from apples to oranges or coffee to lemonade to justify the algebra involved in constructing a price index. The aim to combine many products into a single number is heroic, so simplifying assumptions have to be made. But of course, nobody can eat their smartphone data plan.
As for assuming that consumption patterns are independent of income, this is so clearly problematic that it has led to research into and sometimes publication of a range of different consumer price indices using the typical mix of goods and services purchased by people at different income levels. Low-income households spend a much higher share of their money on energy and food — although bizarrely, the CPI labeled “core,” often the focus of commentary on economic statistics, excludes these two essentials on the grounds that their prices are more volatile than other items.
There are subtler problems here. GDP uses market prices to add up the amount of products sold and in price indices. There is a presumption that a higher GDP is a good thing. But what if prices reflect market power? Should a measure of the economy intended to reflect how well-off the country is adjust for monopolies? If the GDP growth attributable to the top 1% of the population is removed, U.S. growth since the financial crisis looks much less healthy, and the gap between the U.S. and other Western economies is much reduced.
The period of transformational change being experienced in Western economies is making the standard statistical framework steadily more obsolete. This should not be a surprise — it happens with every technological revolution. It took many decades after the 19th-century Industrial Revolution for governments to start collecting regular statistics on industry, and several more after that effort began to get to the framework Western economies have been using for the past 80 years.
Statistics matter. They are the only lens available to governments and citizens to understand and monitor what is happening to the broader economy. Countries without good statistics are hobbled in their ability to govern or to understand public grievances. Voters in many Western countries are signaling discontent with their governments, often with a striking disconnect between what voters believe is true and what the official economic statistics show. There is something in people’s daily experience that the standard indicators, such as GDP growth and levels of employment, are not capturing.
“Living in a digital realm poses a serious problem for measuring what’s going on in the economy.”
The importance of economic statistics was well understood by early pioneers. William Petty, whose “Political Arithmetick” (1690) was one of the first attempts to measure the U.K.’s economic strength, intended the exercise to provide vital information about military capacity in the war against the French, as indeed it did. The title made the point concisely. The invention of GDP by British and American economists (including John Maynard Keynes) during the Second World War had a similarly effective strategic purpose. It has been described as making a material difference to the outcome of the conflict, as the Allies managed their national resources more effectively than their enemies. Earlier, Simon Kuznets had constructed measures of U.S. national income to inform the federal government’s efforts to counter the Depression.
Similarly, measures of inflation that track changes in the cost of living have always been highly politically charged. Inflation redistributes purchasing power away from people whose wages do not keep up with price rises, while asset owners tend to gain from even higher asset price inflation. Even the formulas become matters of political debate, like when pension or social security payments are linked to a particular measurement, such as the CPI.
Even beyond these headline figures, other statistics — or their absence — can shape decisions in ways that have a significant impact. If the government does not know where companies make and transfer their profits, the corporate tax base will steadily evaporate. If it cares about national economic security and resilience, it should be collecting data on supply networks.
Official statistics are an essential public good, one that should be accessible to all (with due regard for privacy and commercial confidentiality). They provide the information governments need to govern well, are useful to businesses and enable the public to hold to account those with power over their standard of living and quality of life.
It is time for better statistics that track the economy as it is today, not as it was eight decades ago. This ought to mean investing in the collection of statistics. Yet many governments are squeezing the budgets of statistical agencies. This reveals something important about the distribution of power now that the resources going into official statistics are declining just as private corporations invest more.
A new approach is needed. Economists need to step up to the challenge of defining the concepts and measures that will effectively describe the transformed, digitized economy. How, for example, should statisticians be measuring the additional economic value created by agentic AI or huge new datacenters requiring scarce minerals and water to operate? New sources of data will be required — online data, cellphone data, satellite images or urban sensors — and new methods to synthesize this into useful indicators. Statistical agencies will need to find the resources to do research themselves, even if that means diverting funding from their 20th-century activities.
Importantly, though, the politicians who make the ultimate decisions about funding need to accept that economic statistics matter. For some, there is a strong instinct to hide or even distort uncomfortable information, be it too-high inflation or a rapidly increasing debt burden. For example, Argentina notoriously stopped publishing inflation figures for a while in the early 2010s, while Greece fiddled with its GDP figures in the 2000s to be able to borrow more. In neither case did altering the figures change the reality, and reality won out in the end. Throughout Western economies now, politically polarized and facing daunting economic uncertainties, cutting or altering official statistics may seem an appealing option. The Trump administration seems to have made this choice already.
If this is the case, the responsibility to develop a fit-for-purpose statistical framework and set of resources falls to researchers and others with the capacity to help everyone understand today’s reality. Such private efforts will never provide the equivalent of the public good of official statistics, but they can pave the way for a clearer-sighted future, just as pioneers of the past did in their time.
