Measuring Productivity in the Context of Technological Change by Diane Coyle 3

WORK IN THE AGE OF DATA

BBVA OPENMIND DATA, IDEAS, AND PROPOSALS ON DIGITAL ECONOMY AND THE WORLD OF WORK BBVAOPENMIND.COM Work in the Age of Data 2 Measuring Productivity in the Context of Technological Change by Diane Coyle 3

Productivity matters because over the long run it is the measure of how much more effectively a society can turn its available resources into valued goods and services. Stated in this general way, it is the ultimate indicator of progress; and, indeed, productivity growth at “modern” rates significantly above zero began with the Industrial Revolution. However, measuring productivity is not straightforward, and linking its behavior to the underlying drivers still less so. Since the mid-2000s there has been Measuring Productivity a slowdown in trend productivity growth in many OECD economies, often described as the “productivity puzzle” precisely be- in the Context of cause its causes are not understood—and particularly because the pace of innovation in fields including digital, biomedicine, and Technological Change materials appears to be at last as rapid as ever (fig. 1). The standard approach is “growth accounting,” the attribution of real terms of Diane Coyle GDP growth to growth in measured inputs of capital and labor and a residual, known as multifactor or total factor productivity (TFP) growth.1 TFP is where technological progress, innovations enabling more output for the same inputs, ought to show up. However, the measured residual also includes the effects of failing to measure all inputs well, or omitting some of them. Over time the measurement of capital and labor inputs has become more sophisticat- ed, with adjustments for the skill level of workers, for example, or the introduction of some types of intangible capital. These im- Technological change is making it harder to provements chip away at the unexplained interpret disappointing productivity figures in residual, which Moses Abramovitz famous- 2 many economies. Although there are likely to ly labeled “the measure of our ignorance.” be many contributory factors, such as post- For example, a recent literature has identi- fied the importance of management quality financial crisis debt overhang and demographic for productivity at the firm level.3 If it were change, technological change complicates the possible to include an aggregate measure interpretation of the evidence in two ways. One of national management quality in the is the delay between companies adopting new growth accounting exercises, as a form of technologies and their impact on productivity intangible capital, this would reduce mea- because of the organizational or management sured TFP. changes and the complementary investments Its residual character thus makes TFP that are also needed. The other is the mismatch somewhat unintuitive. A more intuitive, between how official GDP and productivity and more easily measured, alternative is la- figures are defined and the character of the bor productivity. This is simply real GDP per hour worked. It also makes it easier to see digital economy, such as zero price, advertising- the role of technology, as embodied in capi- funded services, or the switch to cloud tal equipment. For example, a construction computing. A more fundamental question is worker becomes more productive not by whether “productivity” is a useful concept in digging faster with a shovel or taking few- economies consisting to such a large extent of er rest breaks, but by having a mechanical services and intangibles. digger to work with instead. So we should Work in the Age of Data 4

Since the mid-2000s there expect periods of technological change to and bookshops are full of warnings about has been a slowdown in manifest themselves in faster labor produc- the likely effect of the next wave of robotics productivity growth in many tivity growth as well as faster TFP growth. on jobs, but there is absolutely no sign yet OECD economies, a trend However, whichever is selected, pro- of the robot apocalypse as that would cer- often described as the ductivity measures now pose a puzzle. Al- tainly have boosted the labor productivity “productivity puzzle” though the pace of innovation continues, on figures. the face of it, to be very rapid, all measures There are three potential explanations for The technological innovations of productivity growth have slowed signifi- this paradox. of the first half of the 20th cantly, particularly since the mid-2000s. In One prominent perspective is that it the UK, where the slowdown has been paris more apparent than real, and that, in century had profoundly ticularly marked, the level of labor produc- fact, there has been far less technological more important economic tivity is about one-fifth lower than it would innovation than the hype would lead us to consequences than today’s be had the pre-crisis trend continued. But believe. Robert Gordon is a forceful advo- incremental improvements in labor productivity growth has slowed ev- cate of this view. In his book The Rise and digital entertainment or the erywhere across the OECD. Fall of American Growth he argues that digitalization of services Any complex phenomenon will have the technological innovations of the early a number of contributory factors, and in to mid-twentieth century, such as indoor this case the drawn-out effects of the finan- plumbing, electricity, or the initial com- cial crisis on firms’ investment spending, munications technologies of telegraph, decreasing competition in key economic telephone, and radio, had profoundly more sectors, or adverse demographic change in important economic consequences than the OECD countries are all plausible parts today’s incremental improvements in dig- of the productivity story.4 Still, there is a ital entertainment or the digitalization of striking paradox in the combination of existing services and products. While Gor- seemingly rapid innovation in a number of don has a narrow focus in terms of today’s technological fields—advanced materials, frontier technologies, his argument has biomedicine, green energy, autonomous had some support from recent estimates vehicles, small satellites, the “Fourth In- of a sharp decline in research productivity.5 dustrial Revolution” in manufacturing— It is not only Moore’s Law that is showing and dismal productivity figures. The media signs of running out of steam; the number

1975 1980 1985 1990 1995 2000 2005 2010 2015

-2 Annual % Change

Year France Netherlands Fig. 1. Labor productivity growth 1970–2018, selected countries Germany Spain (Source: https://stats.oecd.org) Italy United Kingdom Japan United States Measuring Productivity in the Context of Technological Change by Diane Coyle 5

of researchers in the United States is more tions and hence the GDP and productivity in 5G communications. In recent work on than twenty times greater now than in the statistics. Economic historian Paul David the importance of intangible investments 1930s, and yet there has been a long-run provided a canonical example of the long such as business process redesign, new trend decline in TFP growth for some forty delays between innovation and productivity business models, or additional investments years now. outcomes in his 1999 study of the adoption in human capital for the use of technologies Not surprisingly, given the range of in- of electricity in the early twentieth century. such as artificial intelligence, Erik Bryn- novations reported currently, the technol- Not only was new infrastructure needed for jolfsson and his colleagues characterize this ogy community is dismissive of the claim distribution networks; as David pointed out, as a “productivity J curve,” whereby innova- that the pace of change is slowing. It is hard crystallizing the benefits of electricity also tion leads to lower productivity before any to evaluate the claim that there are severely required major investments such as new improvements are observed.9 Some of the diminishing returns to new ideas. Part of low-level factory buildings configured for current much-hyped innovations, such as the impact of new technologies may never assembly lines in place of old mills built AI and machine learning systems, or the be measured or measurable. For instance, over several stories around their steam en- Internet of Things, are anyway not yet as if today’s renewable energy technologies gines’ drive shafts. Additionally, getting the widely deployed in business as headlines enable the decarbonization of electricity full benefit of the assembly line required in- would suggest so it remains to be seen if generation, this will represent a profound- novations in management techniques and they will live up to expectations. Others— ly important innovation that will never be workflow organization. In that case the lag such as computing, discussed below—are captured in GDP or productivity measure- between original innovations and full pro- widely used, however. ment because the carbon externalities are ductivity impact was some fifty years.6 The final possible explanation concerns not included. A discovery like the use of Research on the initial wave of the dig- the difficulty in measuring productivity. mini-aspirin doses—a cheap and old com- ital revolution also highlighted the impor- This line of explanation also has its skep- pound—to help prevent cardiovascular tance of management and organizational tics. A number of studies have explored the illness will extend many lives but hardly change, leading to either delays, or even potential for measurement error and con- affects the growth statistics at all. failure on the part of some firms to gain cluded that, if anything, the measurement In general, GDP as a value-added mea- any productivity benefit from their invest- difficulties were even greater before the sure does not reflect well innovations that ments in information and communication mid-2000s, so this argument deepens the improve the efficiency of production pro- technology equipment. In their survey of productivity puzzle. Byrne, Fernald, and cesses, as opposed to innovations that give this evidence, Erik Brynjolfsson and Lorin Reinsdorf correct US productivity figures rise to new products or services; few statis- Hitt described this as a role for a form of for several biases such as the need to qual- tical agencies calculate the relevant gross intangible capital, which they argued was ity-adjust prices for ICT equipment, and the output deflated by the appropriate input many times more important than invest- need to measure intangible investments and output prices at each stage of the pro- ment in ICT equipment, and extended better, and yet conclude the measurement duction chain. Over the past twenty years over about a decade before it showed up challenges are no worse than they used to or so there has been a major reorganization in productivity improvements.7 For exam- be so cannot explain the observed slow- of business globally creating extended in- ple, businesses could hold lower stocks of down.10 Looking specifically at the prices ternational supply chains, with each stage inventories, requiring changes in logistics used to deflate nominal GDP, Reinsdorf and more specialized than previously. Adam chains and practices. Workers needed to Schreyer find that there has been meaning- Smith’s insight about the gains from spe- have more flexible job responsibilities and ful overstatement of prices and therefore cialization has been taken to the global more autonomy to take decisions on the understatement of real output and pro- scale. Yet trend productivity growth has basis of information that could now flow ductivity, but again conclude that this was slowed in the OECD nations spearheading to them faster and more cheaply. In their a bigger problem in the past.11 this phenomenon, which can be thought of study of the US productivity boom of the However, others (including me) con- as a significant process innovation—for if it late 1990s, McKinsey also highlighted the clude that there remain some significant were not benefiting companies in some way, role of logistics and inventories, finding that measurement challenges, across a range why would it have become so pervasive? Walmart alone had a measurable impact on of factors affecting either real GDP and This is in effect to restate the productivi- the aggregate figures.8 productivity, or indeed our broader under- ty puzzle, but it suggests a shortfall in our Similarly, additional complementary standing of the structure of the economy. understanding of the economic processes— investments may be needed to reap the These challenges include the following. and measurement, including how to take benefits of some current technological inno- proper account of the use of intermediate vations. Obvious examples are autonomous inputs to production. vehicles requiring significant infrastruc- A second possibility is that the current ture investment and institutional changes Free Digital Goods wave of innovation in many fields will in- in law and insurance products; renewable crease the productivity growth rate—even- energy requiring major investment in the The treatment of digital goods that are free tually. However, important innovations can distribution and transmission networks; or to consumers poses an obvious issue for GDP take a long time to lead to the changes in management and work practice changes in statistics, which are intended to measure to- firms’ activities and consumers’ behavior “Internet of Things” production and service tal monetary transactions in the economy. that are reflected in economic transac- provision, as well as extensive investment What should be done about transactions Work in the Age of Data 6

where no money is changing hands, with At present, then, there is no consensus GDP as a value-added consumers paying, instead, in attention and about the best approach to measuring this measure does not reflect data? The issue is the same as with adver- undoubtedly important economic activi- well innovations that improve tising-funded free-to-air television, but the ty, and the proposals have rather different the efficiency of production scope in the economy is wider now. Ignoring implications for measuring real GDP and processes, as opposed to the goods is not an attractive option because productivity. innovations that give rise to there is some substitution between free and new products or services paid-for digital goods—for example, a service like Spotify has both free and subscription Some of the current much- options for the same service, differing only in Crossing the Production Boundary the advertisements in the former case. hyped innovations, such as AI One possibility is to think of these ad- As this latter argument suggests, one of and the Internet of Things, are vertising-funded zero price goods as a sort the consequences of digitalization is the not yet as widely deployed in of barter transaction. Consumers are paid in increased scope of activities crossing the business as headlines would free digital services in exchange for seeing “production boundary,” whereby transac- suggest so it remains to be advertising; households produce “viewer- tions previously involving monetary trans- seen if they will live up to ship services” that they barter for useful dig- actions in the market economy and GDP are expectations ital services or entertainment. Advertisers substituted by activities within households, in turn pay for the content of these services. which are unmeasured. Such substitutions One estimate of the contribution these bar- across the production boundary occur con- ter transactions make to GDP suggests it stantly: this present digitalization shift is a would have added a tenth of a percentage mirror image of the shift out of household point to US real GDP growth from 1995 to activities into marketed ones in the postwar 2014, with a very modest acceleration post- era, as women increasingly took paid employ- 2005. In other words, the effect on produc- ment and bought in services or conveniences tivity measurement is small and does not as substitutes for unpaid domestic labor—a help account for the slowdown.12 transition that may have flattered measured An alternative is for statisticians to at- productivity growth in the 1960s and 1970s. tempt to estimate directly the value con- Examples of digital shifts from market to sumers gain from these free goods. Erik household sector include online banking or Brynjolfsson and his coauthors have used travel booking (rather than going to the high the kind of contingent valuation methods street), “volunteer” production such as up- previously widely applied to non-mone- loading entertaining videos or posting open- tary environmental goods, and find that source software, or some “sharing economy” consumers (in the US) attach high values to activities. In the absence of updated time-use some free digital goods. The authors sug- data, which would enable the development of gest adding these values to GDP in order household account statistics for digital activ- to calculate a monetary economic welfare ities, it is hard to know the scale of this shift.15 measure, their argument being that the technique provides an estimate of the consumer surplus (welfare gain in excess of the price paid) for these goods.13 The method Price Deflators has attracted interest and is currently being applied in other countries and repeated for There are significant challenges in calcu- the US, not least to test its robustness. lating price indices for sectors experiencing More recently, however, this approach substantial digitally enabled innovation, has been challenged. While agreeing that and it is highly likely that some price indices willingness to pay type methods can provide are overstated, and therefore real output and an estimate of the value of free digital goods, productivity are understated. For example, Heys et al. argue that these are most appro- telecommunications services appear to be priately regarded as intermediate inputs one of the slowest-growing sectors in the into household production—for example UK post-2005. But the previous price index using Google Maps allows faster provision for the sector had taken no account of the of transport services when driving to the massive improvements in quality, such as shops—and therefore add value outside improved compression techniques, faster the market in the household account rather data speeds, reduced latency. Even a modest than adding value to the marketed economy improvement in the price index could turn a GDP is intended to capture.14 zero decrease in prices over five years as re- Measuring Productivity in the Context of Technological Change by Diane Coyle 7

corded by the official index into a decline of attached to each successive embodiment of The number of researchers in more than one-third. Further methodolog- lighting or computing. Similar challenges the United States is more than ical improvements, reflecting the vast in- could even apply to deflators for other sec- twenty times greater now than crease in the volume of telecommunications tors, including “old economy” ones such as in the 1930s, and yet there has traffic, point to even more dramatic price construction, as new methods are incor- 16 been a long-run trend decline declines in the sector. Similarly, calculat- porating features such as digital sensors, in TFP growth for some forty ing the price businesses pay for computing improving the performance along dimen- services to reflect the progressive switch to sions such as energy efficiency, reliability years now cloud computing would indicate substantial or reduced maintenance, none of which is declines in the relevant price index.17 captured in the deflation of nominal output. In general, well-known challenges in The standard theoretical approach of he- constructing consumer price indices when donic adjustment (taking account of certain there are innovative and new goods are measurable quality improvements) is not particularly acute in the digital economy.18 widely applied by statistical offices and also There is a considerable literature looking faces both practical and methodological at how to adjust for quality improvements hurdles.20 Finding methods for calculating in technology goods such as computers or more accurate deflators, in ways statistical smartphones or software. William Nord- agencies can apply in practice, is another haus pointed out the difficulty of measuring active area of research.21 the price over long periods of time of radi- cally changing technologies such as lighting or computing power, because prices are Changing Business Practices and attached to specific products, whereas what Intermediate Goods people get value from is a more fundamental service embodied in different products.19 Classic hip-hop photography on display He calculated the supply side cost of provid- Changes in business practices are making at the launching party of Spotify’s playlist ing these basic goods—although his method the measurement of sectoral productivity “The Hundred” at the Royal Swedish Opera, cannot tell us how much value consumers and international trade statistics more chal- Stockholm, 2018 Work in the Age of Data 8

lenging. Cloud computing is one example. web scraping methods suggest that about The well-known challenges Its price has declined dramatically since the 18% of firms in some sectors of UK manu- in constructing consumer launch and rollout of these services by Am- facturing, and 14% in the case of the US, use price indices when there are azon Web Services and others some seven contract manufacturers.24 innovative and new goods are years ago. Firms substitute from investment particularly acute in the digital in fixed capital equipment (servers, etc.) to economy the purchase of cloud services which are cheaper, higher quality, more secure, and The business investment and constantly being upgraded. Using cloud output statistics of firms are services is, in measurement terms, similar Data to an operational lease on capital equip- based on surveys in which ment owned by another firm. The business Pervading many of these challenges is the they report their capital and investment and output statistics are based treatment of data. In the current framework, operational expenditure. Use on surveys in which firms report their capi- only a small component of the accumulation of the cloud means a switch tal and operational expenditure. Use of the of data is currently incorporated into GDP from the former to the latter, cloud means a switch from the former to the (namely the costs of digitizing and managing with the cloud providers latter, with the cloud providers undertaking a database). Given the explosion of the ac- undertaking the investment the investment expenditure instead. It is in quisition, use and transmission of data, and expenditure instead any case not clear that the large US-domi- the increasing tendency of firms to treat both ciled cloud providers who are market leaders own-account and purchased data as a strate- (Amazon Web Services, Microsoft, Google, gic asset, the current practice with respect to IBM) report their investment spending to investment in data seems too restricted.25 Yet statistical authorities in the separate na- at present there is no consensus about how to tional markets. What is more, in calculating conceptualize, measure, and value the flows sector multifactor productivity, without an of data of different types, and the cross-bor- adjustment for the purchase of capital ser- der aspect of data flows in many of the busi- vices from cloud providers, the productivity ness models described above—and others of the cloud users will be overstated and that based on data including the free digital goods of the cloud providers understated.22 As for models described above—makes these issues the trade statistics, while it is reasonably all the more challenging.26 While there are straightforward to measure imports of ICT standard physical measures of the volume of equipment, it is tricky even to conceptualize data in computing terms (Gigabytes, Zetta- all the cloud service flows, such as when a bytes, etc.) and of communication channel German automaker organizes its global sup- capacity, the economic value will depend ply chains and production via use of cloud on the information content. The economic services from a US-based company with characteristics of data make the valuation data centers in multiple countries.23 particularly difficult because, although there A similar example is the case of “fac- are some market transactions in data allow- toryless manufacturing,” whereby firms ing prices to be discovered, data is a non-ri- retain intellectual property and customer val, public good with externalities, meaning relationships but contract—often over- there is a wedge between market valuation seas—all the manufacturing activity, rely- and economic welfare. The national accounts ing on digital communications and modern are concerned mainly with market transac- logistics. In general, trade figures are hard tions but the wider context is important for to interpret as innovative firms digitally economic policy. This is another area of ac- transmit their IP such as blueprints and tive research including an international de- designs across borders (retaining owner- bate focused on the national accounts about ship) in non-recorded data flows. Mean- taxonomies and classification.27 while, the products resulting from this IP are recorded in trade figures. Furthermore, some big firms thought of as manufacturers may be classified as distributors, with Classification and Data Collection some evidence that the size of the manufacturing sector, often of particular interest Cutting across these conceptual measure- to policy makers, is understated. Existing ment issues is the need for innovation in business surveys do not capture the scope the collection of the raw data used to con- of factoryless manufacturing (and similar struct GDP and productivity measures. models such as toll processing), but novel The classification structure for economic Measuring Productivity in the Context of Technological Change by Diane Coyle 9

statistics in terms of industrial sectors and tional System of National Accounts. Yet the In the current framework, occupations has not kept up with new ac- fact there are so many measurement issues only a small component of tivities and new skills, still reflecting the raises the more fundamental question of the accumulation of data is manufacturing-heavy economy of the 1940s whether “productivity” is the best way to con- currently incorporated into despite subsequent updates, and at pres- ceive of how well the economy is progressing. GDP (namely the costs of ent omitting altogether some industries After all, the OECD economies consist for the digitizing and managing a of interest to policy makers. For instance, most part now of services, not products. database) sectors such as video games have grown As long ago as his 1994 Presidential Ad- in importance but are hard to track in the dress to the American Economic Association, The fact that there are so many existing statistics, and occupational cate- Zvi Griliches observed that “our measure- gories are shifting rapidly. New methods ment and observational tools are becoming measurement issues raises the such as web scraping are being tested as an increasingly inadequate in the context of more fundamental question of alternative to the traditional survey-based our changing economy.”34 Labor productiv- whether “productivity” is the data collection methods.28 An additional ity had flatlined in what he then considered best way to conceive of how problem stems from the extended produc- “unmeasurable” sectors of the economy: con- well the economy tion supply chains noted above. For ex- struction, trade, finance, other services and is progressing ample, many manufacturing services are government. In the subsequent twenty-five understandably classified in the service years, the scope of the “unmeasurable” has sector rather than manufacturing, but with extended. Not only do some of those original insufficient granularity so the size of the unmeasurables account for a greater share manufacturing-centric part of the economy of GDP in many OECD economies, but in ad- is easily understated. One study suggested dition some of the previously “measurable” that the “true” size of the UK’s manufactur- sectors, including communication and man- ing sector could be understated by up to a ufacturing, are giving more trouble. One of half by counting specialist service activi- Griliches’ examples was pharmaceuticals, ties outsourced from manufacturing in with and the difficulty of treating generic and services such as retail or accountancy.29 new drugs prices adequately in construct- It is certain that new data sources and ing a price index and hence real output of the methods will be needed to develop more sector; while it is relatively straightforward accurate measurement of output and pro- to measure the number of pills being taken ductivity, reflecting the structure of modern or shots being given, biomedical innovation economies. Great strides are being made in means the health outcomes per product are using novel big data methods, such as ex- dramatically improved. The GDP statistics tended use of scanner data to improve price were developed in an era of mass production indices,30 web scraping and other novel on- and consumption, so are harder to interpret line data such as text or listings,31 “big data” for the present era of highly differentiated datasets recording massive individual trans- services.35 actions or linkages,32 and satellite data.33 The This points to one of two substantial UK’s Office for National Statistics has estab- difficulties with the current measurement lished a Data Science Campus to develop framework, which is what “real terms” out- innovative methods. Yet initiatives such as put means. The idea of deflating nominal these are in their infancy, and far from being GDP is to remove the part of the expansion systematized by statistical offices. of the dollar or euro total due simply to inflation by calculating what amount of output would leave people with the same level of utility as before. There is a vast literature on Conclusions price indices concerning how best to achieve this constant utility ideal. But it is, need- This list of measurement artifacts and chal- less to say, a heroic abstraction at the best of lenges, extensive as it is, may not in the end times. And when there is rapid innovation add up to a significantly different aggregate and quality change, there is no satisfactory productivity picture, given other import- conceptual solution. Economists often think ant contributory drivers of the long-term that the statistics simply need “hedonic” ad- trends. Even so, there is a growing volume justment, a regression technique to correct of research into economic statistics, much prices of goods for the measurable quality of it due to the digital revolution; some of the improvements (say, air conditioning in cars, issues listed above will be addressed in the faster processing speeds in laptops). This is Workers on a break take a nap near the conveyor belt at the Nike factory in Ho Chi Minh upcoming periodic revision of the interna- somewhat arbitrary in any case. But it can City, 2001 Work in the Age of Data 10

Its economic characteristics potentially lead to implausibly high calcu- nomic welfare. One suggestion, from Charles make data valuation lated growth rates for sectors such as ICTs. Hulten and Leonard Nakamura, is to intro- particularly difficult because, This possibility was flagged up early in the duce an expanded GDP concept, taking the although there are some debate about hedonic adjustment of prices conventional GDP figures based on efficiency by Milton Gilbert, one of the architects of the of resource use in production and augment- market transactions in System of National Accounts, who pointed ing them with additional efficiency in con- data allowing prices to be out that in the extreme the method could sumption, the additional value consumers discovered, data is a non-rival, suggest growing real output of a product gain from given output thanks to innova- public good with externalities, whose physical volume was shrinking to- tion.37 Another, which goes still further away meaning there is a wedge ward zero: if people happily wore nothing from the present focus on production and between market valuation at all on the beach would we argue that the productivity, is to consider how people use and economic welfare “real” output of bathing suits was the same their time and the value they gain from dif- as in the Victorian era, he asked? As the ferent activities, as time use seems a natural great Thomas Schelling once pointed out: metric for a services-based economy.38 Some “[W]hat we call ‘real’ magnitudes are not services will be more “productive” the faster completely real; only the money magnitudes they can be carried out, and these will be the are real. The ‘real’ ones are hypothetical.”36 more routine and potentially automatable. Not surprisingly, price indices will be a Others, more bespoke, will be more valuable major focus of the work contributing to the to consumers if they take more time, and pro- next revision of statistical standards. How- vide higher quality. In this latter case, the An image from a presentation at the 2018 ever, some researchers have suggested more price consumers pay will directly reflect the Computing Conference: “Empower Digital China,” held at the Yunqi Cloud Town radical approaches moving away from the perceived value. These two varieties could International Expo Centre in Hangzhou idea of real GDP as the benchmark for eco- exist in the same conventionally defined sector: consider routine blood tests versus care in the intensive care unit, or a quick shave at the barber versus a designer haircut. It is early days and there will be other suggestions for rethinking the conceptual framework for the economy before economists and statisticians converge on the successor to today’s lens on the economy, including the power- ful arguments for thinking far more broadly about what measurement “beyond GDP” is needed for a valid sense of economic welfare or well-being.39 None of this implies there is no productivity problem. In the ten years-plus since the financial crisis, too few people in the OECD middle classes have seen the steady gains in prosperity that sustain stable soci- eties.40 New innovations are welcome but they must deliver wide benefits for consumers. Addressing the pessimistic sense that the next generation will be worse off than their parents is one of the policy challenges of our times.41 There are plenty of signs that this will be a politically turbulent pe- riod, like previous decades of sustained slow growth in the twentieth century. Measurement questions might seem like a distraction at times like this, but when the questions are as extensive and complex as they are now they point to deeper questions: what do we mean by the idea of progress, and what measures do policy makers need to understand what is happening in the economy and society, and to deliver broad-based, sustainable improvements in citizens’ lives. Measuring Productivity in the Context of Technological Change by Diane Coyle 11

Acknowledgments and C. Syverson, “Artificial uploads/2018/11/ESCoE- intelligence and the modern DP-2018-16.pdf. My thanks to Julia Wdowin for productivity paradox: a 19. W. Nordhaus, “Do her research assistance. clash of expectations and real-output and real-wage statistics,” NBER chapters in measures capture reality? The Ajay Agrawal, Joshua Gans, history of lighting suggests Notes and Avi Goldfarb (eds.), not,” in Timothy F. Bresnahan Diane Coyle is Bennett Professor of Public The Economics of Artificial and Robert J. Gordon (eds.), Policy at the University of Cambridge. Her 1. C. Hulten, “Growth Intelligence: An Agenda, The Economics of New research focuses on the measurement of accounting,” NBER Working National Bureau of Economic Goods, Chicago: University of the digital economy and digital markets. She Paper no. 15341, issued in Research, Inc., Chicago: Chicago Press, pp. 27–70. was previously Professor of Economics at September 2009, available at: Chicago University Press, 20. E. Diewert, “Quality the University of Manchester and has held a https://www.nber.org/papers/ pp. 23–57; E. Brynjolfsson adjustment and hedonics: number of public service roles including Vice- w15341; D. Jorgenson and Z. and K. McElheran, “The rapid A unified approach,” Chair of the BBC Trust (2006–14), member of Griliches, “The explanation for adoption of data-driven Discussion Paper 19- the Competition Commission (2001–09), and productivity change,” Review decision-making,” American 01, Vancouver School of member of the Migration Advisory Committee of Economic Studies 34: Economic Review: Papers Economics, University of (2009–14). She is currently a member 249–283 (1967). and Proceedings 106(5): British Columbia, available of the Natural Capital Committee, and an 2. M. Abramovitz, “The 133–139 (2016). at: https://econ2017.sites. expert adviser to the National Infrastructure search for the sources of 10. D. Byrne, J. Fernald, and olt.ubc.ca/files/2019/03/ Commission and was a member of the Furman growth: Areas of ignorance, M. pdf_paper_diewer_DP19- Review of Digital Markets. Her last book was old and new,” The Journal Reinsdorf, “Does the United 01_QualityAdjustmentand GDP: A Brief but Affectionate History; and her of Economic History 53(2): States have a productivity HedonicsMarch142019.pdf. next, Markets, State and People: Economics 217–243 (June 1993). slowdown or a measurement 21. E. Diewert and R. for Public Policy, will be published in January 3. N. Bloom and J. Van problem?” BPEA conference Feenstra, “Estimating the 2020. She was awarded a CBE for her Reenen, “Measuring and draft, March 10–11, 2016, benefits of new products,” contribution to the public understanding of explaining management available at: https://www. Conference paper available economics in the 2018 New Year Honors. practices across firms and brookings.edu/wp-content/ at: https://www.nber.org/ countries,” The Quarterly uploads/2016/03/ByrneEtAl_ chapters/c14281.pdf R., Journal of Economics ProductivityMeasurement_ (2019). CXXII(4): 1351–1408 (2007). ConferenceDraft.pdf. 22. P. Schreyer, OECD, 4. I. Goldin, P. Koutroumpis, 11. M. Reinsdorf and P. “Measuring productivity: F. Lafond, N. Rochowicz, and Schreyer, “Measuring Measurement of aggregate J. Winkler, “Why is productivity consumer inflation in a digital and industry-level productivity slowing down?,” Working economy,” draft: 9/Nov/2017, growth,” available at: Paper, Oxford Martin School, Fifth IMF Statistical Forum. http://www.oecd.org/sdd/ University of Oxford (2018); 12. L. Nakamura, J. productivity-stats/2352458. P. Ollivaud, Y. Guillemette, Samuels, and R. Soloveichik, pdf (2001). and D. Turner, “Links between “Measuring the ‘free’ digital 23. D. Coyle and D. Nguyen, weak investment and the economy within the GDP and “Cloud computing and slowdown in productivity and productivity accounts,” ESCoE national accounting,” ESCoE potential output growth across Discussion Paper 2017-03, Discussion Paper 2018–19. the OECD,” OECD Economics available at: https://www. 24. D. Coyle and D. Nguyen, Department Working escoe.ac.uk/wp-content/ “No factory, no problem,” Papers, no. 1304, OECD uploads/2017/12/ESCoE- ESCoE Discussion Paper Publishing, Paris, https://doi. DP-2017-03.pdf. forthcoming 2019. org/10.1787/5jlwvz0smq45- 13. J. Brynjolfsson, A. Collis, 25. P. Ollivaud, Y. Guillemette, en (2016). and F. Eggers, “Using massive and D. Turner, “Links between 5. N. Bloom, C. Jones, J. Van online choice experiments to weak investment and the Reenen, and M. Webb, “Are measure changes in well- slowdown in productivity and ideas getting harder to find?” being,” PNAS 116(15): 7250– potential output growth across NBER Working Paper no. 7255 (April 9, 2019). the OECD,” OECD Economics 23782, issued in September 14. R. Heys et al., “GDP Department Working 2017. and welfare, a spectrum Papers, no. 1304, OECD 6. P. David, “The dynamo and of opportunity,” ESCoE Publishing, Paris, https://doi. the computer: An historical Discussion Paper forthcoming org/10.1787/5jlwvz0smq45- perspective on the modern 2019. en (2016). productivity paradox,” The 15. D. Coyle, “Do–it–yourself 26. W. C. Y. Li, M. Nirei, and American Economic Review digital: The production K. Yamana, “Value of data: 80(2): 355–361, Papers and boundary, the productivity There’s no such thing as a free Proceedings of the Hundred puzzle and economic welfare,” lunch in the digital economy,” and Second Annual Meeting Economica, LSE (2018). US Bureau of Economic of the American Economic 16. M. Abdirahman et al., “A Analysis Working Papers, Association (May, 1990). comparison of approaches to Washington, DC, https:// 7. E. Brynjolfsson and L. deflating telecoms services www.bea.gov/research/ Hitt, “Beyond computation: output,” ESCoE Discussion papers/2018/value-data- Information technology, Paper 2017-04. theres-no-such-thing-free- organizational transformation 17. D. Coyle and D. Nguyen, lunch-digital-economy and business performance,” “Cloud computing and (2019),. Journal of Economic national accounting,” ESCoE 27. OECD, Measuring the Perspectives 14(4): 23–48 Discussion Paper 2018–19. Digital Transformation: A (2000). 18. W. Diewert, K. Fox, and Roadmap for the Future, 8. B. Lewis et al., “US P. Schreyer, “The digital OECD Publishing, Paris, productivity growth,” economy, new products and available at: https://www. 1995–2000, McKinsey Global consumer welfare,” ESCoE oecd.org/going-digital/mdt- Institute, October 2001 Discussion Paper 2018–16, roadmap-data.pdf (2019). Report. available at: https://www. 28. A. Lima and H. Bakhshi, 9. E. Brynjolfsson, D. Rock, escoe.ac.uk/wp-content/ “Classifying occupations using Work in the Age of Data 12

web-based job advertisements: An c0554.pdf. application to STEM and creative 37. C. Hulten and L. Nakamura, occupations,” ESCoE Discussion “Accounting for growth in the age Paper DP-2018-08; J. Djumalieva of the Internet: The importance and C. Sleeman, “An open and of output-saving technical data-driven taxonomy of skills change,” Federal Reserve Bank extracted from online job adverts,” of Philadelphia, Working Papers ESCoE Discussion Paper DP- Research Department, available 2018-13. at: https://www.philadelphiafed. 29. J. Hauge and E. O’Sullivan, org/-/media/research-and- “Inside the black box of data/publications/working- manufacturing: Conceptualising papers/2017/wp17-24r.pdf. and counting manufacturing in the 38. G. Becker, “A theory of the economy,” University of Cambridge allocation of time,” The Economic Institute for Manufacturing, Journal 75(299): 493–517 (1965); a report prepared for the UK D. Coyle and L. Nakamura, “Towards Department for Business, Energy a framework for time use, welfare and Industrial Strategy: https:// and household-centric economic www.ifm.eng.cam.ac.uk/uploads/ measurement,” ESCoE Discussion Research/CSTI/Inside_the_Black_ Paper 2019-01. Box_of_Manufacturing_report_ 39. Bennett Institute for Public FINAL_120619.pdf (2019). Policy, “Measuring wealth, 30. A. Chessa, J. Verburg, and delivering prosperity” Wealth L. Willenborg, “A comparison Economy Project Report, 2019, of price index methods for available at: https://www. scanner data,” Ottawa Group, bennettinstitute.cam.ac.uk/ International Working Group on publications/measuring-wealth- Price Indices, available at: http:// delivering-prosperity/; J. Stiglitz, www.ottawagroup.org/Ottawa/ J. Fitoussi, and M. Durand (eds.), ottawagroup.nsf/ For Good Measure: Advancing 4a256353001af3ed Research on Well-being Metrics 4b2562bb00121564/ Beyond GDP, OECD Publishing, 1ab31c25da944ff5ca25822 Paris, https://www.oecd-ilibrary. c00757f87/$FILE/A%20 org/economics/for-good- comparison% measure_9789264307278-en 20of%20price%20index%20 (2018). methods%20for%20scanner%20 40. OECD “Under pressure: The data squeezed middle class,” OECD %20-Antonio%20Chessa,%20 Publishing, Paris. https://doi. Johan%20Verburg,%20Leon%20 org/10.1787/689afed1-en https:// Willenborg%20- www.oecd.org/els/soc/OECD- Paper.pdf (2017). middle-class-2019-main-findings. 31. M. Gentzkow, B. Kelly, and pdf. M. Taddy, “Text as data,” Stanford 41. B. Stokes, “Global publics University Working Paper, https:// more upbeat about the economy,” web.stanford.edu/~gentzkow/ Pew Research Center online research/text-as-data.pdf (2017). report, https://www.pewresearch. 32. V. Carvalho, “From micro to org/global/2017/06/05/global- macro via production networks,” publics-more-upbeat-about-the- Journal of Economic Perspectives economy/#sentiment-about- 28(4): 23–48 (2014); Y. Jin current-economy-and-childrens- and S. Vasserman, “Buying Data prospects-not-always-linked. from consumers: The impact of monitoring programs in US auto insurance,” Harvard University (Job Market Paper) Working Paper, available at: https://scholar.harvard. edu/files/jmp_jin.pdf (2019). 33. D. Donaldson and A. Storeygard, “The view from above: Applications of satellite data in economics,” The Journal of Economic Perspectives 30(4): 171–198 (2016). 34. Z. Griliches, “Productivity, R&D, and the data constraint,” The American Economic Review, 84(1): 1–23 (1994). 35. D. Coyle, GDP: A Brief but Affectionate History, Princeton: Princeton University Press, 2015. 36. T. Schelling, “Design of the accounts,” in A Critique of the United States Income and Product Accounts, Princeton: Princeton University Press, 1958. The quote is from page 332. Available at http://www.nber.org/chapters/ DIANE COYLE HOW TO CITE THIS ARTICLE Coyle, Diane. “Measuring Productivity in the Context of Technological Change.” In Work in the Age of Data. Madrid: BBVA, 2019.

ARTICLE IN SPANISH LISTEN TO THE ARTICLE HERE • Medir la productividad en el contexto del cambio tecnológico.

READ THE FULL BOOK • The Past Decade and the Future of the Global Economy, Barry Eichengreen Towards a New Enlightenment? A Transcendent Decade. • Work in The Age of Data • El trabajo en la era de los datos • Business Models, Information Technology, and the Company of the Future, , Haim Mendelson Reinventing the Company in the Digital Age.

• Big Data and the Future of Business,, Kenneth Cukier Reinventing the Company in the Digital Age.

ALL THE OPENMIND COLLECTION TITLES