Does the United States have a productivity slowdown or a measurement problem?

Byrne, David M. ; Fernald, John G. ; Reinsdorf, Marshall B. 等

V.B. Globalization

Standard techniques for constructing price indexes do not capture the change in the average price paid by the buyers of a product when they alter their purchasing patterns to buy from a different seller. Similarly, import price deflators in the NIPA do not capture changes in the price paid by buyers when they switch from a domestic source to an offshore producer. As a result, when sourcing moves offshore or to a different import-supplying country, real imports are understated and real output is overstated. (31)

This bias was particularly significant in the late 1990s and early 2000s, when the location of many kinds of manufacturing was shifting rapidly from the United States and other countries with high labor costs to emerging market economies. One impetus for this was China's 2001 accession to the World Trade Organization, which coincided with the start of a large shift in the sourcing for many manufactured goods used in the United States to China. Another was a multilateral free trade agreement that reduced tariffs for IT products to zero for an interval of four years ending in 2000, which accelerated international sourcing changes for IT products (Feenstra and others 2013).

Reinsdorf and Robert Yuskavage (forthcoming) use two approaches to estimate the sourcing bias for imported consumer goods in the 1997-2007 period, and they find a bias in the range of 0.8 to 1 percent a year for durable goods, including computers, and about 0.6 percent a year for imported apparel and footwear; after 2007, the effect is small. However, even if we assume that the bias estimate of 1 percent a year for durable goods can be generalized to similar kinds of imported capital goods and that the bias in the apparel index can be generalized to all textile products, the upward bias in business TFP is only 0.1 percent a year because the affected imports have only a small weight in GDP. This globalization adjustment shows up as a negative contribution of 10 bp from 1995 to 2004 for the "other" category in figure 1.

Another aspect of globalization made possible by reduced communications costs is international trade in services over a wire. The number of American jobs that could potentially be offshored to a country with lower wages is large (Blinder 2009), and the offshoring of services could lead to the same sort of upward bias in measures of productivity that is caused by the offshoring of goods. Thus far, however, the effects have been modest; the BEA's input/output accounts show that the imports of business-process services such as professional, scientific, and technical services, and computer systems services--rose from about 2 percent of total intermediate uses of these services in 1997-98 to about 5 percent in 2010-14.

V.C. The "Sharing" Economy

Nominal GDP includes transactions from the sharing economy, such as car rides via Uber and Lyft. (32) Nevertheless, it is unlikely that the deflator used to compare the new services to previously existing ones correctly measures the decline in the quality-adjusted price experienced by many consumers. Thus, there is probably some (at this point very, very small, but likely growing) downward bias in the growth rate of real GDP.

It would be useful to have official statistics on the nominal output of the various types of services included in the sharing economy. Research indexes of price changes could then be developed to try to calibrate the size of the bias.

VI. Conclusions

The "productivity paradox 2.0" remains alive: Despite ongoing IT-related innovation, aggregate U.S. productivity growth slowed markedly after about 2004. To investigate this paradox, we propose several adjustments to IT-related hardware, software, and services. The good news is that these adjustments would make recent growth in GDP and investment look modestly better than recorded. The bad news is that they would make the paradox even worse--the slowdown in labor productivity is even larger after our durable goods adjustments, while the slowdown in TFP is not much affected. The reason is that mismeasurement was substantial in the 1995-2004 period, as well as more recently, and rising import penetration for computers and communications equipment means that domestic production (which matters for GDP) has fallen over time.

Moreover, the slowdown was broad-based, which suggests that ongoing innovation in IT is not substantially spilling over into other areas. Other measurement challenges--such as digital services, globalization, and fracking--go in the right direction but are small.

Other evidence also suggests that true underlying growth is relatively modest. First, the U.S. productivity slowdown has been mirrored in many parts of the world (Eichengreen, Park, and Shin 2015; Cette, Femald, and Mojon 2016; Askenazy and others 2016). This suggests that underlying macroeconomic factors may be driving the slow pace of growth, given the varied sources and methods used across national statistical systems. Chad Syverson (2016) finds that the slowdown across countries is not correlated with IT production or use, again suggesting that the problem is not related to the mismeasurement of IT goods or services. Second, the decline in economic dynamism--both in the form of fewer start-ups and in the slower reallocation of labor resources in response to productivity shocks supports the idea that productivity-enhancing innovations are diffusing throughout the economy more slowly (Decker and others 2016); also, the subdued pace of investment has slowed the adoption of new technology embodied in capital. And third, Michael Mandel (2016) also finds little evidence of widespread rapid innovation in an analysis of labor market metrics such as occupational employment and help-wanted ads, although there are tremendous occupational changes in some narrow segments ot the economy, such as IT and the extraction of oil and natural gas.

If not mismeasurement, why did productivity growth slow? The slowdown predated the Great Recession, which suggests that the event was not the story--or, at least, not the whole story. Given that the 1970s and 1980s had slow growth similar to the period after 2004, the fast-growth 1995-2004 period looks like the anomaly. With the emergence of the Internet, the reorganization of distribution sectors, and IT investments beginning to pay off, many things came together in a short period. With hindsight, this looks like a one-time upward shift in the level of productivity rather than a permanent increase in its growth rate. Looking forward, we could get another wave of the IT revolution. Indeed, it is difficult to say with certainty what may yet come from cloud computing, the Internet of things, and the radical increase in mobility from smartphones. However, we have not yet seen those gains.

Changes in overall welfare are somewhat harder to assess. Transformative gains related to mobile technologies and the Internet clearly raise welfare. Most of these gains properly belong outside the purview of market sector GDP--and proposals to incorporate them into GDP raise concerns. Still, these innovations are valued by households. That said, the available estimates of the welfare gains (based on the value of leisure time) suggest that "free" digital services add the equivalent of perhaps 0.3 percent of GDP a year to well-being; that is small relative to the roughly 1.75 percent slowdown in labor productivity growth in the business sector from 2004 to 2014.

Nevertheless, much is unknown. Matthew Shapiro and David Wilcox (1996) described the area of the quality adjustment of price indexes as house-to-house combat in national accounting. The analysis must be done product by product, and statistical agencies are usually playing catchup. (33) Digital services and the ensuing new modes of delivery of other types of services are particularly challenging. For example, research on the quality-adjusted price indexes associated with changes in the organization of production caused by the digital economy, such as the substitution of Uber and Lyft for traditional taxi services, would be helpful. Satellite accounts could also help to shed light on gains from digital services and shifting of production outside the market boundary by presenting measures of economic activity that extend beyond the market.

Finally, we conclude by touching on the implications for policymakers. Slow productivity growth, if it persists, implies slow future potential growth. The benefits in the nonmarket sector can offset this somewhat vis-a-vis well-being, but they do not help with taxes or the budget.

ACKNOWLEDGMENTS We thank Martin Baily, Erik Brynjolfsson, Carol Corrado, Erwin Diewert, Jean Flemming, Robert Gordon, Peter Klenow, James Stock, and Hal Varian for helpful comments and conversations. We thank Travis Adams, Genevieve Denoeux, and Arthi Rabbane for excellent research assistance. The title of this paper refers to a phrase in a Wall Street Journal article by Timothy Aeppel (2015). The views expressed in this paper are those of the authors and should not be attributed to the International Monetary Fund or its managers or executive directors, or to the Federal Reserve Bank of San Francisco, the Federal Reserve System's Board of Governors, or the Federal Reserve System.

References

Abraham, Katharine G., and Christopher Mackie, editors. 2005. Beyond the Market: Designing Nonmarket Accounts for the United States. Washington: National Academies Press.

Aeppel, Timothy. 2015. "Silicon Valley Doesn't Believe U.S. Productivity Is Down." Wall Street Journal, July 16.

Aizcorbe, Ana, and Yvon Pho. 2005. "Differences in Hedonic and Matched-Model Price Indexes: Do the Weights Matter?" Working Paper no. WP2006-06. Washington: U.S. Department of Commerce, Bureau of Economic Analysis.

Askenazy, Philippe, Lutz Bellmann, Alex Bryson, and Eva Moreno Galbis, editors. 2016. Productivity Puzzles across Europe. Oxford University Press.

Baily, Martin Neil, and Robert J. Gordon. 1988. "The Productivity Slowdown, Measurement Issues, and the Explosion of Computer Power." Brookings Papers on Economic Activity, no. 2: 347-420.

Basu, Susanto, John G. Fernald, Nicholas Oulton, and Sylaja Srinivasan. 2004. "The Case of the Missing Productivity Growth, or Does Information Technology Explain Why Productivity Accelerated in the United States but Not in the United Kingdom?" NBER Macroeconomics Annual 18: 9-82.

Baumol, William J., and William G. Bowen. 1966. Performing Arts: The Economic Dilemma--A Study of Problems Common to Theater, Opera, Music, and Dance. New York: Twentieth Century Fund.

Becker, Gary S. 1965. "A Theory of the Allocation of Time." Economic Journal 75, no. 299:493-517.

Berndt, Ernst R., and Neal J. Rappaport. 2001. "Price and Quality of Desktop and Mobile Personal Computers: A Quarter-Century Historical Overview." American Economic Review 91, no. 2: 268-73.

--. 2003. "Hedonics for Personal Computers: A Reexamination of Selected Econometric Issues.' Paper presented at R&D, Education, and Productivity: An International Conference in Memory of Zvi Griliches (1930-1999), a meeting of the Conference on Research in Income and Wealth, which is administered by the National Bureau of Economic Research, Paris, August 25-27. http.//www. nber.org/CRIW/papers/berndt.pdf

Bils, Mark, and Peter J. Klenow. 2001. "Quantifying Quality Growth." American Economic Review 91, no. 4: 1006-30.

Blinder, Alan S. 2009. "How Many US Jobs Might Be Offshorable?" World Economics 10, no. 2: 41-78.

Boskin, Michael J., Ellen R. Dulberger, Robert J. Gordon, Zvi Griliches, and Dale Jorgenson. 1998. "Toward a More Accurate Measure of the Cost of Living: Final Report to the Senate Finance Committee from the Advisory Commission to Study the Consumer Price Index." In Getting Prices Right: The Debate over the Consumer Price Index, edited by Dean Baker. Armonk, N.Y.: M.E. Sharpe.

Brynjolfsson, Erik, Yu (Jeffrey) Hu, and Michael D. Smith. 2003. "Consumer Surplus in the Digital Economy: Estimating the Value of Increased Product Variety at Online Booksellers." Management Science 49, no. 11: 1580-96.

Brynjolfsson, Erik, and Joo Hee Oh. 2014. "The Attention Economy: Measuring the Value of Free Digital Services on the Internet." Working paper. https://www. scribd.com/doc/314594157/1-The-Attention-Economy-Measuring-the-Value-of-Free-Goods-on-the-Internet

Byrne, David. 2015a. "Domestic Electronics Manufacturing: Medical, Military, and Aerospace Equipment and What We Don't Know about High-Tech Productivity." FEDS Notes. Washington: Board of Governors of the Federal Reserve System.

--.2015b. "Prices for Data Storage Equipment and the State of IT Innovation. FEDS Notes. Washington: Board of Governors of the Federal Reserve System.

Byrne, David, and Carol Corrado. 2015. "Recent Trends in Communications Equipment Prices." FEDS Notes. Washington: Board of Governors of the Federal Reserve System.

--. 2016. "ICT Prices and ICT Services: What Do They Tell Us about Productivity and Technology?" Working paper.

Byrne, David M., John G. Fernald, and Marshall Reinsdorf. Ongoing work. "Can We Bring Advertising-Supported Activity into Business-Sector Value Added? Unpublished manuscript.

Byrne, David M., Stephen D. Oliner, and Daniel E. Sichel. 2015. "How Fast Are Semiconductor Prices Falling?" Working Paper no. 21074. Cambridge, Mass.: National Bureau of Economic Research.

Byrne, David, and Eugenio Pinto. 2015. "The Recent Slowdown in High-Tech Equipment Price Declines and Some Implications for Business Investment and Labor Productivity." FEDS Notes. Washington: Board of Governors of the Federal Reserve System.

Cette, Gilbert, John Fernald, and Benoit Mojon. 2016. "The Pre-Great Recession Slowdown in Productivity." Working Paper no. 2016-08. Federal Reserve Bank of San Francisco. (Forthcoming in European Economic Review.)

Cole, Rosanne, Y.C. Chen, Joan A. Barquin-Stolleman, Ellen Dulberger, Nurhan Halvacian, and James H. Hodge. 1986. "Quality-Adjusted Price Indexes for Computer Processors and Selected Peripheral Equipment. Survey of Current Business 66, no. 1: 41-50.

Copeland, Adam. 2013. "Seasonality, Consumer Heterogeneity and Price Indexes: The Case of Prepackaged Software." Journal of Productivity Analysis 39, no. 1: 47-59.

Corrado, Carol, Charles Hulten, and Daniel Sichel. 2009. "Intangible Capital and U.S. Economic Growth." Review of Income and Wealth 55 no. 3: 661-85.

Corrado, Carol, and Kirsten Jager. 2015. "Wealth and Investment in Mature Societies." Presentation given at Milestone 3: Midterm Conference, Smart Public INTANgibles (SPINTAN), London, April 23-24. http://www.spintan. net/wp-content/uploads/public/Carol-Corrado-Wealth-and-Investment_carol corrado.pdf

Decker, Ryan A., John Haltiwanger, Ron S. Jarmin, and Javier Miranda. 2016. "Declining Business Dynamism: What We Know and the Way Forward." American Economic Review 106, no. 5: 203-07.

Domar, Evsey D. 1961. "On the Measurement of Technological Change" Economic Journal 71, no. 284: 709-29.

Doms, Mark. 2005. "Communication Equipment: What Has Happened to Prices?" In Measuring Capital in the New Economy, edited by Carol Corrado, John Haltiwanger, and Daniel Sichel. University of Chicago Press.

Dulberger, Ellen R. 1989. "The Application of a Hedonic Model to a Quality-Adjusted Price Index for Computer Processors." In Technology and Capital Formation, edited by Dale W. Jorgenson and Ralph Landau. MIT Press.

Eichengreen. Barry, Donghyun Park, and Kwanho Shin. 2015. "The Global Productivity Slump: Common and Country-Specific Factors." Working Paper no. 21556. Cambridge. Mass.: National Bureau of Economic Research.

Evans, David S., Richard Schmalensee, and Scott R. Murray. 2016. "The Census Bureau Needs to Significantly Revise Reporting and Calculation of Its Online and Physical Retail Sales Figures and Commission an Independent Review." http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2728918

Feenstra, Robert C. 1994. "New Product Varieties and the Measurement of International Prices." American Economic Review 84, no. 1: 157-77.

Feenstra, Robert C., Benjamin R. Mandel, Marshall B. Reinsdorf, and Matthew J. Slaughter. 2013. "Effects of Terms of Trade Gains and Tariff Changes on the Measurement of US Productivity Growth." American Economic Journal: Economic Policy 5, no. 1: 59-93.

Feldstein, Martin. 2015. "The U.S. Underestimates Growth." Wall Street Journal May 18.

Fernald, John G. 1999. "Roads to Prosperity? Assessing the Link between Public Capital and Productivity." American Economic Review 89, no. 3: 619-38.

--. 2014. "A Quarterly, Utilization-Adjusted Series on Total Factor Productivity." Working Paper no. 2012-19. Federal Reserve Bank of San Francisco.

--. 2015. Productivity and Potential Output before, during, and after the Great Recession." NBER Macroeconomics Annual 29: 1-51.

Flamm, Kenneth. 1989. "Technological Advance and Costs: Computers versus Communications." In Changing the Rules: Technological Change, International Competition, and Regulation in Communications, edited by Robert W. Crandall and Kenneth Flamm. Brookings Institution Press.

Goolsbee, Austan, and Peter J. Klenow. 2006. "Valuing Consumer Products by the Time Spent Using Them: An Application to the Internet." American Economic Review 96, no. 2: 108-13.

Gordon, Robert J. 1990. The Measurement of Durable Goods Prices. University of Chicago Press.

--. 2015. "Productivity, Prices, and Measurement." Presentation at Boskin-Fest conference, Stanford University, October 2. https://www.scribd.com/doc/ 314597445/Productivity-Prices-and-Measurement

--. 2016. The Rise and Fall of American Growth: The U.S. Standard of Living since the Civil War. Princeton University Press.

Greenstein, Shane, and Ryan C. McDevitt. 2009. "The Broadband Bonus: Accounting for Broadband Internet's Impact on U.S. GDP." Working Paper no. 14758. Cambridge, Mass.: National Bureau of Economic Research.

Greenwood, Jeremy, Ananth Seshadri, and Mehmet Yorukoglu. 2005. "Engines of Liberation." Review of Economic Studies 72, no. 1: 109-33.

Griliches, Zvi. 1994. "Productivity, R&D, and the Data Constraint." American Economic Review 84, no. 1: 1-23.

Grimm, Bruce T. 1996. "A Quality-Adjusted Price Index for Digital Telephone Switches." Working paper. Washington: U.S. Department of Commerce, Bureau of Economic Analysis.

Grimm, Bruce T., Brent R. Moulton, and David B. Wasshausen. 2005. "Information-Processing Equipment and Software in the National Accounts. In Measuring Capital in the New Economy, edited by Carol Corrado, John Haltiwanger, and Daniel Sichel. University of Chicago Press.

Hatzius, Jan, and Kris Dawsey. 2015. "Doing the Sums on Productivity Paradox v2.0." U.S. Economics Analyst (Goldman Sachs), no. 15/30.

Hausman, Jerry. 1999. "Cellular Telephone, New Products, and the CPI." Journal of Business and Economic Statistics 17, no. 2: 188-94.

Heather, James M., and Benjamin Chain. 2016. The Sequence of Sequencers. The History of Sequencing DNA." Genomics 107, no. 1: 1-8.

Holdway, Michael. 2001. "Hedonic Models in the Producer Price Index (PPI)." Washington: U.S. Bureau of Labor Statistics, http://data.bls.gov/cgi-bin/print.pl/ ppi/ppicomqa.htm

Houseman, Susan, Christopher Kurz, Paul Lengermann, and Benjamin Mandel. 2011. "Offshoring Bias in U.S. Manufacturing." Journal of Economic Perspectives 25, no. 2: 111-32.

Ip, Greg. 2015. "Beyond the Internet, Innovation Struggles." Wall Street Journal, August 12.

Kim, Mina, and Marshall B. Reinsdorf. 2015. "The Impact of Globalization on Prices: A Test of Hedonic Price Indexes for Imports." In Measuring Globalization: Better Trade Statistics for Better Policy 1, edited by Susan N. Houseman and Michael Mandel. Kalamazoo: Upjohn Institute Press.

Mandel, Michael. 2009. "Growth: Why the Stats Are Misleading." Bloomberg Businessweek, June 3.

--. 2016. "Where Is Innovation Falling Short? Using Labor Market Indicators to Map the Successful Innovation Frontier." In New Entrepreneurial Growth Agenda. Kansas City: Ewing Marion Kauffman Foundation, http://www.kauffman.org/neg/section-3

McCallum, John C. 2015. "Disk Drive Prices (1955-2015)." http://www.jcmit.com/ diskprice.htm

McGrattan, Ellen R., and Edward C. Prescott. 2012. "The Great Recession and Delayed Economic Recovery: A Labor Productivity Puzzle?" In Government Policies and the Delayed Economic Recovery, edited by Lee E. Ohanian, John B. Taylor, and Ian J. Wright. Stanford, Calif.: Hoover Institution Press.

Nakamura, Emi, and Jon Steinsson. 2012. "Lost in Transit: Product Replacement Bias and Pricing to Market." American Economic Review 102, no. 7: 3277-316.

Nakamura, Leonard, and Rachel H. Soloveichik. 2015. "Valuing 'Free' Media across Countries in GDP." Working Paper no. 15-25. Federal Reserve Bank of Philadelphia.

Nordhaus, William D. 2002. "Productivity Growth and the New Economy." Brookings Papers on Economic Activity, no. 2: 211-44.

--. 2006. "Principles of National Accounting for Nonmarket Accounts." In A New Architecture for the U.S. National Accounts, edited by Dale W. Jorgenson, J. Steven Landefeld, and William D. Nordhaus. University of Chicago Press.

Nordhaus, William D., and Edward C. Kokkelenberg. 1999. Nature's Numbers: Expanding the National Economic Accounts to Include the Environment. Washington: National Academies Press.

Parker, Robert P, and Bruce T. Grimm. 2000. "Recognition of Business and Government Expenditures for Software as Investment: Methodology and Quantitative Impacts, 1959-98." Paper presented at the meeting of the BEA Advisory Committee, U.S. Department of Commerce, Bureau of Economic Analysis, Washington, May 5. http://www.bea.gov/papers/pdf/software.pdf

Reinsdorf, Marshall. 1993. "The Effect of Outlet Price Differentials on the U.S. Consumer Price Index." In Price Measurements and Their Uses, edited by Murray F. Foss, Marilyn E. Manser, and Allan H. Young. University of Chicago Press.

Reinsdorf, Marshall, and Robert Yuskavage. Forthcoming. "Offshoring, Sourcing Substitution Bias and the Measurement of U.S. Import Prices, GDP and Productivity." Review of Income and Wealth.

Shapiro, Matthew D., and David W. Wilcox. 1996. "Mismeasurement in the Consumer Price Index: An Evaluation. NBER Macroeconomics Annual 11: 93-154.

Schreyer, Paul, and W. Erwin Diewert. 2014. "Household Production, Leisure, and Living Standards." In Measuring Economic Sustainability and Progress, edited by Dale W. Jorgenson, J. Steven Landefeld, and Paul Schreyer. University of Chicago Press.

Schreyer, Paul, Nicola Brandt, and Vera Zipperer. 2015. "Productivity Measurement with Natural Capital." Paper presented at the Society for Economic Measurement Annual Conference, Paris, July 24. http://repository.cmu.edu/cgi/ viewcontent.cgi?article=1007&context=sem_conf

Schultze, Charles L., and Christopher Mackie, editors. 2002. At What Price? Conceptualizing and Measuring Cost-of-Living and Price Indexes. Washington: National Academies Press.

Sichel, Daniel E. 1997. The Productivity Slowdown: Is a Growing Unmeasurable Sector the Culprit?" Review of Economics and Statistics 79, no. 3: 367-70.

Soloveichik, Rachel. 2015a. "Exploring the Boundaries of GDP: Cultivated Assets and Valuing 'Free' Media." Presentation. Washington: U.S. Department of Commerce, Bureau of Economic Analysis, https://bea.gov/about/pdf/ acm/2015/november/exploring-the-boundaries-of-gdp.pdf

--. 2015b. "Valuing 'Free' Entertainment in GDP: An Experimental Approach." Working paper. Washington: U.S. Department of Commerce, Bureau of Economic Analysis, http://bea.gov/about/pdf/acm/2015/november/valuing-free-entertainment-in-gdp-for-aea-paper.pdf

Stallman, Richard. 1985. "The GNU Manifesto." http://www.gnu.org/gnu/manifesto. en.html

Stein, Lincoln D. 2010. "The Case for Cloud Computing in Genome Informatics." Genome Biology 11: Article 207. http://genomebiology.biomedcentral.com/ articles/10.1186/gb-2010-11-5-207

Syverson, Chad. 2016. "Challenges to Mismeasurement Explanations for the U.S. Productivity Slowdown." Working Paper no. 21974. Cambridge, Mass.: National Bureau of Economic Research.

Triplett, Jack E. 1989. "Price and Technological Change in a Capital Good: A Survey of Research on Computers." In Technology and Capital Formation, edited by Dale W. Jorgenson and Ralph Landau. MIT Press.

--. 1999. "Economic Statistics, the New Economy, and the Productivity Slowdown." Business Economics 34, no. 2: 13-17.

Triplett, Jack E., and Barry P. Bosworth. 2004. Productivity in the U.S. Services Sector: New Sources of Economic Growth. Brookings Institution Press.

United Nations and others. 2009. System of National Accounts 2008. New York: European Commission, International Monetary Fund, Organization for Economic Cooperation and Development, United Nations, and World Bank.

Varian, Hal. 2011. "The Economic Value of Google." Presentation at the Web 2.0 Expo, San Francisco, March 29. http://www.web2expo.com/webexsf2011/ public/schedule/detail/17666

Wasshausen, Dave, and Brent R. Moulton. 2006. The Role of Hedonic Methods in Measuring Real GDP in the United States." Paper presented at 31st CEIES Seminar: Are We Measuring Productivity Correctly? in Rome, October 12-13. http://www.bea.gov/papers/pdf/hedonicGDP.pdf

Wetterstrand, Kris A. 2016. "DNA Sequencing Costs: Data from the NHGRI Genome Sequencing Program (GSP)." https://www.genome.gov/27541954/ dna-sequencing-costs/

Zheng, Simon, and Harry Bloch. 2014. "Australia's Mining Productivity Decline: Implications for MFP Measurement." Journal of Productivity Analysis 41, no. 2: 201-12.

Comments and Discussion

COMMENT BY MARTIN NEIL BAILY This is a terrific paper that changed my views on the slowdown in productivity growth. I had thought that a significant fraction of the post-2004 growth slowdown could be explained by measurement errors of two types: first, underestimation of the pace of productivity in the computer and semiconductor industry; and second, the fact that the National Income and Product Accounts do not include the contribution of "free stuff" like Google and Facebook to final output because they are paid for by advertising and are, therefore, considered intermediate production. David Byrne, John Fernald, and Marshall Reinsdorf show persuasively that neither type of measurement error is significant enough to change our estimates of the slowing of growth in about 2004. The authors document, along with Chad Syverson (2016), that the productivity growth slowdown is very large--output would be larger by about $3 trillion today if there had been no slowdown. The errors of measurement would need to be very large indeed to explain much of this loss of output, and the authors show that this is not the case.

It is worth backing up a little and pointing out the strange, paradoxical economic times that form the backdrop to this paper. Productivity growth has been very slow since about 2004, and the slowdown appears to be getting worse, with output per hour in the nonfarm business sector in 2015 only 2.6 percent higher than its 2010 value, according to the Bureau of Labor Statistics (BLS). This is bad news for living standards and economic growth. At the same time, other signs seem to point to rapid technological change. For example, in a survey released in June 2015 of the Fortune 500 CEOs, they named their greatest challenges; the top challenge, listed by 70 percent of the CEOs, was rapid technological change. And a front-page article in the May 3, 2016, Financial Times said, "Surging investment in artificial intelligence is giving the United States an early advantage in the global race to dominate a new era of robotics, according to investors and experts in an industry set to become one of the most strategically important in the coming decades" (Waters and Inagaki 2016). Erik Brynjolfsson and Andrew McAfee (2014) are leading experts on technology, and their 2014 book The Second Machine Age has been an important and influential guide to changing technology. In chapter 2 they write, "Most of the innovations described in this chapter have occurred in just the past few years. They've taken place in areas where ... the best thinking often led to the conclusion that it wouldn't speed up. But then digital progress became sudden after being gradual for so long" (Brynjolfsson and McAfee 2014, p. 37). In short, they suggest that technological change has actually speeded up. Mismeasurement had offered one way to resolve the paradox of slow growth and rapid technological change, and Byrne, Fernald, and Reinsdorf have punctured this balloon.

Although I agree with the bottom fine of this paper, there are some points of disagreement or places where I would have put a different emphasis. Most important, a casual reader of their paper might believe that productivity growth is being well measured overall, but that is not the case. In important segments of the economy, there is no serious effort to capture the impact of technological change on the quality of output. Most notable is health care, where advances in screening and diagnostics, surgical procedures, and medical devices are constantly being introduced. Thanks to the research by Ernst Berndt of the Massachusetts Institute of Technology and others, (1) there has been an effort to capture the benefits of new drugs, but improvements in the quality of hospital care are being missed. The authors are aware of this problem, but their tight focus on the post-2004 slowdown means they do not give it much attention.

Unlike health care, the high-tech industry has seen efforts made to capture quality change in output, and the authors do an admirable job of exploring the various price deflators covering this industry. They make a good case that errors of measurement are not likely to explain much of the slowdown; but as they note, much uncertainty remains. The reader is left wishing that the BLS would do a systematic review of its price index methodology. One of the big changes in recent years has been to make computers and tablets much lighter and more user-friendly, and I do not think this quality change is being captured.

The authors point out that much of the manufacturing of high-tech equipment has moved offshore, so this sector is providing a smaller boost to U.S. productivity. They also mention the outlet substitution bias that has likely occurred as component production has moved offshore. These points are certainly correct, but I wonder if changes in industry structure mean that some U.S. productivity growth is being missed. Much of the design work for computer chips, iPhones, and the like is still being done in the United States, so the quality change that accounts for most of the productivity increase in high-tech industries is still attributable to economic activity located in the United States and not in the countries where the products are manufactured. One reason for this is that the United States' marginal corporate tax rate is higher than those of other countries, so multinational companies minimize the U.S. content of products made and sold internationally. Even without tax distortions, in a world where supply chains are global, it is intrinsically very difficult to correctly account for productivity by country. This measurement problem is not huge, but it probably has been getting worse over time.

When I ask both economists and noneconomists whether they think that the free stuff on their phones or computers is significant enough to shift the needle of productivity growth, I get bipolar answers. Some people believe that all the new stuff is fantastic and is changing their lives, while others dismiss it as trivial. The answers to my nonscientific personal survey are somewhat age-related; older people are usually less enthusiastic than younger people, but not always so. Some older people and some with disabilities find that their lives have been greatly enhanced. Byrne, Fernald, and Reinsdorf are convincing in showing that the magnitude of uncounted free services is just not large enough to make much of a dent in the $3 trillion hole in productivity. They evaluate free services using a time-use framework, which is an entirely reasonable decision given that the literature has used this framework. But I do not find this approach to be all that compelling. All consumption involves time, but we do not try to capture this when we measure productivity. Automobiles, for example, require time to drive, but the contribution of automobile production to productivity depends on quality-adjusted output in relation to inputs. There is no need to estimate the time spent driving.

The tricky part with Google and other similar services, as the authors note, is that consumers do not pay for them directly, but only through advertising. Such free services did not start with Google; television used to be entirely paid for by advertising, and much of it still is. Like Google, television was an intermediate goods industry. Given the value that most consumers place on watching TV, the exclusion of this service from final output in earlier periods meant that productivity growth was understated. Other countries chose to pay for television through license fees, and their TV-related industries were then counted in final output and presumably contributed to measured productivity growth.

[FIGURE 1 OMITTED]

If we lived in a world where the budgets of the statistical agencies were greatly expanded, I would urge them to create satellite accounts to estimate the value of free services. If time-use analysis provided the best approach, then I would be a convert. Another alternative would be to use conjoint analyses from survey data, the approach used by market researchers to assess how consumers value different products and product attributes.

The authors have calculated productivity growth by industry, and they use these results in their paper, but there are some additional lessons worth drawing from these data, and I turn now to describe them. My figure 1 shows the Domar-weighted contribution of each of the major private sector industries to multifactor productivity (MFP) growth in the aggregate over the entire period from 1987 to 2013 for which consistent data are available from the BLS. The first striking fact is that manufacturing contributes about 40 percent of aggregate MFP growth. This is despite the fact that manufacturing is only about 10 percent of employment. From a productivity point of view, manufacturing still matters. The second striking fact is that two large sectors of the economy, construction and services, had zero or negative MFP growth contributions from 1987 to 2013. This raises a red flag that measurement problems may actually be quite important for estimated growth over the whole period. Construction productivity has been a mystery for a very long time. Nonresidential output is very hard to price and measure accurately. On the residential side, I participated in a number of cross-country comparisons of residential productivity in the 1990s with the McKinsey Global Institute, and we never found a country where productivity was higher than in the United States. (2) That is not impossible to square with U.S. productivity that declines over time, but it is odd.

The subindustries within services that drag down the total are education and health care, and I discussed health care above. Having no productivity growth in education is perhaps not surprising, given how little the format of education has changed. The content of what is being learned has changed a lot, however, especially in higher education. And the industry structure and the way teaching takes place both seem poised to change, as new technology-related tools are introduced and competition increases. It would be a step forward if measurement methods for this industry were able to keep up with changing educational methods.

My next two figures both illustrate that the industries whose productivity growth increased after 1995 were generally the industries that showed slower growth after 2004. My figure 2 shows that the pattern holds for the major industry sectors, and my figure 3 shows the same result when all of BLS's subindustries are included. One interpretation of this pattern is that a broad productivity opportunity opened up, creating the scope for a rapid productivity increase. Some industries had business models and market conditions that were conducive to taking advantage of this opportunity, but other industries were less well suited and continued along the old path. After 10 years or so, the impact of the innovative surge was over, and the rapid growers fell back to their prior pattern. The obvious candidate for the productivity opportunity was the rapid improvement and dissemination of information technology. There may also have been other contributory factors, such as a willingness to take risks, intense competition, and strong aggregate demand growth. (3)

[FIGURE 2 OMITTED]

The authors of this paper took on a very hard task, asking whether measurement error can explain the 2004 slowdown in productivity growth, and they came up with a convincing answer: It cannot. In his impressive review of U.S. economic history, Robert Gordon (2016) concludes that the information technology revolution was the last in a series of productivity waves dating back to the Industrial Revolution. The authors of this paper are less pessimistic, and I agree with them. For one thing, it seems unlikely that the productivity wave from information technology has run its course. Moore's law must end eventually, but there are many new ways to take advantage of cheap processing power and low-cost communications. The crowdsourcing of design, robots, and the Internet of Things are three such ongoing advances. Innovations in biotechnology and materials science are in the works. Productivity growth in wholesale and retail trade has slowed, as brick-and-mortar retailers deal with excess capacity because they face competition from online retailers and slow overall consumption growth, but a competitive shakeout of the industry will eventually result in higher productivity. Gordon (2016) may be correct, however, in saying that the era of 3 percent annual productivity growth over multiple decades is over. Future advances are likely to be lumpier, with surges of productivity from time to time, not all the time.

[FIGURE 3 OMITTED]

REFERENCES FOR THE BAILY COMMENT

Brynjolfsson, Erik, and Andrew McAfee. 2014. The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies. New York: W. W. Norton.

Gordon, Robert J. 2016. The Rise and Fall of American Growth: The U.S. Standard of Living since the Civil War. Princeton University Press.

Syverson, Chad. 2016. "Challenges to Mismeasurement Explanations for the U.S. Productivity Slowdown." Working Paper no. 21974. Cambridge, Mass.: National Bureau of Economic Research.

Waters, Richard, and Kana Inagaki. 2016. "Investment Surge Gives U.S. the Early Lead in Rise of the Robots." Financial Times, May 3.

(1.) For a list of Ernst Berndt's publications, many of which relate to price measurement, see http://mitsloan.mit.edu/faculty-and-research/faculty-directory/detail/?id=41392.

(2.) Details of the studies are available on the McKinsey Global Institute website, at http:// www.mckinsey.com/mgi/our-research/productivity-competitiveness-and-growth.

(3.) There may also be some regression toward the mean, although over 10-year periods that effect is likely to be muted.

COMMENT BY

ROBERT J. GORDON The slow productivity growth since 2004, particularly since 2010, represents one of the outstanding economic puzzles of our time. It implies slow current growth of potential real GDP. If it continues, it implies slow future growth of potential GDP and fewer resources to address the nation's problems, including education, infrastructure, and the looming shortfalls in funding for Social Security and Medicare. It would be reassuring for puzzle solving, although disconcerting for the integrity of the nation's statistical system, to learn that the entire post-2004 or post-2010 slowdown in productivity growth was due to well-identified errors in measurement, and that the underlying "true" rate of productivity growth has not decelerated at all.

THE QUESTIONS TO BE ADDRESSED This paper by David Byrne, John Fernald, and Marshall Reinsdorf places its main emphasis on the post-2004 productivity growth slowdown and highlights the $3 trillion in additional business sector real GDP that would have been produced in 2015 if the productivity growth rate of 1995-2004 had continued after 2004. But the post-2004 slowdown is not the only productivity puzzle to be explained. My figure 1 displays the five-year moving average growth rate of quarterly utilization-adjusted total factor productivity (TFP) growth going back to 1952. This data plot identifies four separate eras of TFP growth--consistently rapid, at about 2.0 percent a year, through 1973; then slower and erratic, in the range of 0 to 1.5 percent, from the early 1970s through the mid-1990s; then healthy again for a decade between 1995 and 2004; and finally a sharp slowdown, to about 0.5 percent a year, during most of the past decade.

Average TFP growth rates over selected periods are listed in my table 1. The first two rows contrast the 26 years from 1947 to 1973 with the 42 years since 1973. When the postwar era is divided at 1973, the TFP growth rate slows by more than half, from 2.10 to 0.82 percent a year. This sharp contrast poses the first productivity puzzle: What caused the post-

1973 slowdown? The next three rows divide up the post-1973 interval at 1995 and 2004. The TFP growth rates for 1973-95 and for 2004-15 are almost identical, at about 0.5 percent, in sharp contrast to the growth rate of almost 2.0 percent achieved between 1995 and 2004. This leads to the second puzzle: What caused TFP growth to revive? And to the third puzzle: Why was that revival only temporary? The final two rows of my table 1 divide the 2004-15 era in early 2010 and show that TFP growth during the 2004-10 period was slightly faster than from 1973 to 1995, but during the 2010-15 period was somewhat slower.

[FIGURE 1 OMITTED]

This alternation between relatively fast and relatively slow TFP growth over the four eras of the postwar epoch places a broader perspective on Byrne, Fernald, and Reinsdorf's topic of mismeasurement. Techniques of measurement have been relatively constant since 1947, and thus it is implausible to argue that the slowdown from the first era (1947-73) to the second era (1973-95) happened because measurement became worse by an average of 1.5 percentage points a year. In the same way, it is implausible to argue that the revival to the third era (1995-2004) occurred because measurement became better, at a rate of 1.5 percent a year. Likewise, it is implausible to argue that the slowdown to the fourth era (2004-15) occurred because measurement once again became worse, at a rate of 1.5 percent a year, as had occurred previously, after 1973. These alleged appearances and disappearances of measurement errors of 1.5 percent a year in both directions are implausible, because measurement techniques were relatively constant across the four postwar eras.

MEASUREMENT ERRORS HAVE DIMINISHED IN IMPORTANCE IN THE MARKET ECONOMY Byrne, Fernald, and Reinsdorf place their primary emphasis on the measurement of the private business economy. Their main focus is on mismeasurement in the form of biased deflators for information and communication technology (ICT) equipment in the National Income and Product Accounts (NIPA). They survey recent research on the prices of ICT equipment and conclude that the NIPA deflators systematically understate the rate of decline in the quality-adjusted prices of ICT equipment and thus understate the rate of growth of real ICT investment as well as real GDP and labor productivity. TFP is affected less, because the use of improved deflators not only raises real GDP growth but also raises the growth rate of capital input that is subtracted from output in the calculation of TFP.

However, this price index bias does not help at all in understanding the post-2004 productivity growth slowdown, because the price index bias is roughly constant both before and after 2004. The difference between the rate of change of the authors' liberal ICT price index and the corresponding NIPA index is -5.1 percent a year during the period 1995-2004 and an almost identical -4.5 percent in 2004-14. Both the liberal index and the NIPA index exhibit a sharp deceleration in the rate of the price decline after 2004, which points to a declining rate of technological improvement in ICT equipment as a substantive reason for the productivity growth slowdown.

[FIGURE 2 OMITTED]

The constant post-2004 price index bias is not the end of the story, however, because the relative importance of ICT equipment in the economy has changed in two ways. First, ICT investment is a smaller share of GDP. Comparing the two years 1999-2000 with 2014-15, the GDP share of information processing equipment declined from 2.77 to 1.79 percent, and that of computers and peripherals fell fully by half, from 1.00 to 0.45 percent. When combined with the relatively constant pre- and post-2004 price index corrections, the shrinking ICT shares imply that the price index bias for GDP and labor productivity became smaller after 2004.

The second reason why the price index bias has become less important stems from a sharp shift of ICT investment from domestic production to imports. My figure 2 exhibits the startling shift in computer purchases--from 17.8 percent imported in 2002 to an average of 87.9 percent imported in 2011-13. Consider the implications of the extreme case in which all ICT equipment is imported. An upward price index bias for imported computers would lead to an understatement of the growth rate of both computer investment and computer imports, netting out to zero impact on GDP and labor productivity. The understatement of growth in capital input, however, would lead to an overstatement of TFP growth. Thus the shift to computer imports in the last decade has caused true TFP growth to slow down more since 2004 than in the official NIPA data. This tendency has been exacerbated by the fact that the price indexes for imported computers used in the NIPA decline at a substantially slower rate than the deflators for domestically produced computers, whereas the observed shift of computer purchases to imports suggest the opposite--that the true prices of imported computers have declined faster than prices of domestic production.

These two factors, the declining share of ICT investment in GDP and the shift to imports, together with the substantial upward bias in the price index for imported computers, suggest that since 2004 measurement issues have caused the poor performance of TFP growth to be even worse in reality than in the government's statistics. Byrne, Fernald, and Reinsdorf's treatment concludes, in the third column of their table 2, that measurement errors cause labor productivity growth to be understated by 0.49 percentage point for the period 1995-2004 and by 0.18 point for 2004-14, for a net measurement improvement of 0.31 percentage point. Because of offsetting adjustments to output and capital input, the effect on TFP growth is much smaller and goes in the opposite direction, with measurement errors causing TFP growth to be overstated by 0.08 percentage point during 1995-2004 and by a slightly greater 0.12 percentage point during 2004-14. The conclusion is that improved measurement causes the post-2004 slowdown in both labor productivity and TFP growth to become even worse than in the official data.

These important conclusions of Byrne, Fernald, and Reinsdorf's analysis combine an upward bias in the price indexes of computers with a shrinking share of computer investment and of the domestic production share of computer equipment. If this upward price index bias were larger, their conclusions would be magnified. In his important historical study of the price indexes of computers, William Nordhaus (2007, table 10, p. 153) concludes that the price of computer power during the 1990-2002 period decreased at an annual rate of -57.5 percent, as compared with Byrne, Fernald, and Reinsdorf's alternative price index for computers and peripherals in their table 1, which declines at a much slower annual rate of -27.3 percent during the 1995-2004 period. Though there are conceptual differences between Nordhaus's performance-based measure and Byrne, Fernald, and Reinsdorf's hedonic price indexes, Nordhaus's index has the advantage that it includes data for both mainframe and personal computers, whereas Byrne, Fernald, and Reinsdorf's indexes for the 1990s are based only on personal computers. Compared with mainframes, personal computers achieve a much lower price per calculation, and thus the transition from mainframes to personal computers that took place in the 1980s and 1990s reduced the average price per calculation more rapidly than the price decline for personal computers alone. (1) To the extent that Nordhaus's (2007) approach is a better guide to the overall price behavior of computer investment in the 1990s, there has been an even greater tendency for official data to understate that rapid growth of labor productivity during the 1995-2004 period and to understate the true decline in its growth rate since 2004.

THE WELFARE EFFECTS OF FREE INTERNET CONTENT Free Internet information was available both before and after the 2004 transition from fast to slow productivity growth. The proportion of American households connected to the Internet rose from 5 percent in 1995 to 56 percent in 2004, followed by a more gradual increase to 75 percent by 2013 (Gordon 2016, figure 13-4, p. 455). The mismeasurement hypothesis refers to the difference in the welfare benefits from free Internet content available after 2004, as compared with before 2004. An aspect of the post-2004 improvement is the transition from dial-up to broadband access; the proportion of households with broadband increased from 3 percent in 2000 to 29 percent in 2004 and to 65 percent after 2009 (Gordon 2016, figure 13-5, p. 455). Thus, even if the same amount of time were spent on Internet access before and after 2004, there was a quality change in the form of much faster response times made possible by the spread of broadband. Byrne, Fernald, and Reinsdorf recognize that download speed is not considered a quality change in current deflators for Internet access, but they assert that "only a small amount of extra digital service output is missed."

Most of Byrne, Fernald, and Reinsdorf's treatment of free digital services refers to their role in the market economy. The authors consider alternatives to the current national accounts treatment of advertising-supported media as an intermediate good. Even when free Internet services and other forms of entertainment are treated as final consumption, their role in the market economy can be no larger than their advertising revenue. Because total advertising revenue from all sources, including television and print media, amounts to only 1.3 percent of GDP, and because real advertising revenue has grown faster than business sector GDP by only a small margin, the authors find the impact of free Internet services on market GDP to be close to zero. Intuitively, the growing advertising revenue of Google and Facebook is largely canceled out by the decline in advertising revenue from older forms of media, particularly print publications.

Advocates of the mismeasurement hypothesis have in mind the broader scope of free Internet services as a source of increased consumer welfare, going well beyond market GDP. They point to the rapid increase in Internet usage, particularly of mobile services accessed on smartphones and tablets, as a focus of consumers' leisure-time activity. How important is the increase in consumer welfare resulting from increased Internet use? More than one-third of the U.S. population uses Facebook, and the time each day that its users devote to Facebook reached 50 minutes in 2015 (Stewart 2016). Taken together with other Internet services, most notably YouTube and Google, total daily time devoted to the Internet has been estimated at two hours (Karaian 2015). By comparison, the American Time Use Survey (ATUS) reports that in 2014 Americans on average spent 2.8 hours a day watching television. (2) A problem with the estimation of consumer surplus is that the ATUS does not report on Internet usage as a separate category, with the exception of "household and personal e-mail and messages," which accounted for a trivial 0.03 hour per day. Some mobile phone usage may occur during the ATUS category of time devoted to "socializing and communicating" (0.71 hour per day). Otherwise, the omission of Internet usage could imply that multitasking is the standard mode of behavior, with mobile phones accessed while watching television, eating meals, riding public transit, or standing in line.

To assess the consumer welfare aspects of free Internet services, Byrne, Fernald, and Reinsdorf develop an explicit model based on Gary Becker's (1965) theory of the economics of time, in which total consumption is subject to both a monetary and time budget constraint. However, they do not provide their own estimates of the time value of free Internet services, relying instead on a paper by Erik Brynjolfsson and Joo Hee Oh (2014) that values the incremental consumer surplus from free digital services as "equivalent of adding about 0.3 percentage point per year to business sector output." Note that this addition of 0.3 percentage point exactly cancels out the authors' estimate in their table 2 that the post-2004 decline in business sector labor productivity has been understated by the same 0.3 percentage point a year.

Is this estimate too large or too small? Chad Syverson (2016) provides a survey of different approaches to measuring the consumer surplus of the Internet and compares the resulting surplus estimates to the total missing annual output from the post-2004 productivity slowdown, which he estimates, like Byrne, Fernald, and Reinsdorf, to be about $3 trillion in 2015. Most of the literature surveyed by Syverson provides surplus estimates of about $100 billion a year, a trivial fraction of the missing $3 trillion. But one approach developed by Austan Goolsbee and Peter Klenow (2006), when updated by Syverson's numbers, yields a post-2004 incremental surplus estimate of $842 billion, almost one-third of the missing $3 trillion. This one-third translates into 0.6 percentage point of the total post-2004 slowdown of 1.8 percentage points in the growth rate of business sector productivity. How reasonable is this estimate?

Applying $842 billion to the 80 percent of the population with broadband access yields an annual per capita sum of $3,300. If incremental post-2004 Internet usage, compared with the use of the Internet before 2004, is one hour per day, this would imply an Internet value of $9 per hour. This compares with Syverson's (2016) estimate of the average 2015 after-tax wage of $22. The difference between $9 and $22 makes sense, because it reflects the fact that only half the population is employed, and the leisure time spent on the Internet is inframarginal. The use of one hour per day in this calculation, rather than the two hours reported as average daily Internet use, reflects not only the fact that some time was allocated to the Internet before 2004 but also the multitasking implied by the ATUS time allocation. Much if not most Internet use, according to the ATUS, is not occurring during hours of leisure that previously had zero value, but rather as multitasking during hours that previously had value obtained from socializing or watching television.

Therefore, Byrne, Fernald, and Reinsdorf appear to be too dismissive of the consumer surplus contributed by free Internet services. Their estimate, taken from Brynjolfsson and Oh's (2014) findings, values incremental post-2004 Internet services as worth 0.3 percent of business sector output per year, just enough to offset the authors' table 2 estimate that the decline in the growth rate of business sector productivity has been understated by the same 0.3 percent per year. In contrast, Syverson's (2016) approach values the consumer surplus at 0.6 percent per year, twice as much. This is enough to offset the 0.3 percentage point measurement understatement from the authors' table 2, as well as contributing an additional 0.3 percentage point consumer surplus bonus that provides a partial counterweight to the overall 1.8 percentage point productivity growth slowdown. The relatively large size of this Internet valuation naturally leads to the question of how much consumer surplus was contributed by inventions in the past, including the value of free Internet services introduced in the 1995-2004 decade, such as e-mail, search engines, Wikipedia, and the early phase of e-commerce provided by Amazon, iTunes, and airline websites. The technique of multiplying hours of leisure by an hourly value based on the wage rate would yield a particularly large value of incremental consumer surplus in the early 1950s, as free television broadcasts reached almost every home between 1950 and 1955.

History is full of examples of added consumer surplus that was not included as an increase in GDP. During the period from 1900 to 1940, as motor vehicles replaced horses, real GDP did not value the removal of horse droppings and urine from city streets and rural highways. And real GDP did not value the increase in speed and load-carrying capacity made possible by automobiles, nor their flexibility, which gave birth to a new industry called "personal travel." Moreover, real GDP did not value the increase in consumer surplus as clean running water arrived at the in-home tap and replaced the previous need to carry pails of water into the house from nearby wells or streams. Finally, real GDP did not value the replacement of the outhouse and the need to physically dispose of human waste with the silent efficiency of public sewers. Running water, the electric washing machine and refrigerator, the automobile, and all the other inventions of the late 19th and early 20th centuries are linked into the GDP statistics, which means that zero value is placed on their invention. Real GDP did not value the reduction of infant mortality from 22 percent of new births in 1890 to about 1 percent in 1950. By some estimates, this change created more welfare value than all the other sources of increased consumer welfare taken together.

CONCLUSION Because newly invented products and services have always provided consumer surplus that supplements growth in real GDP by unknown amounts, the authors of this paper are correct to focus their main attention on the measurement issues that arise in the private business sector of the market economy. They convincingly demonstrate that the post-2004 slowdown in the growth rate of labor productivity and TFP has not been due to measurement errors. On the contrary, because the most evident source of measurement bias is in the price indexes for ICT equipment, the understatement of business sector output and productivity was greater in the 1995-2004 period than in the 2004-15 period, both because ICT investment was a greater share of GDP and because a much greater share of ICT equipment was manufactured in the domestic economy, whereas after 2004 there was a sharp increase in the share of such equipment that was imported. The authors thus conclude that the post-2004 slowdown is real. I agree with their interpretation that after 2004, productivity growth returned "back to normal," to a rate roughly the same as that achieved between 1973 and 1995, and that the burst of faster productivity growth between 1995 and 2004 reflected a one-time-only conversion of the economy to a modern world whose ICT equipment and software replaced the previous world's typewriters, calculating machines, and file cabinets.

(1.) Nordhaus (2007, p. 155) provides an example of a 2002 IBM supercomputer that had a price-per-performance ratio roughly 34 times higher than a typical Dell personal computer in 2004.

(2.) See Bureau of Labor Statistics, "Table A-1: Time Spent in Detailed Primary Activities and Percent of the Civilian Population Engaging in Each Activity, Averages per Day by Sex, 2014 Annual Averages" (http://www.bls.gov/tus/tables/al_2014.pdf).

REFERENCES FOR THE GORDON COMMENT

Becker, Gary S. 1965. "A Theory of the Allocation of Time." Economic Journal 75, no. 299:493-517.

Brynjolfsson, Erik, and Joo Hee Oh. 2014. "The Attention Economy: Measuring the Value of Free Digital Services on the Internet." Working paper. https://www. scribd.com/doc/314594157/1-The-Attention-Economy-Measuring-the-Value-of-Free-Goods-on-the-Internet

Byrne, David, and Eugenio Pinto. 2015. "The Recent Slowdown in High-Tech Equipment Price Declines and Some Implications for Business Investment and Labor Productivity." FEDS Notes. Washington: Board of Governors of the Federal Reserve System.

Goolsbee, Austan, and Peter J. Klenow. 2006. "Valuing Consumer Products by the Time Spent Using Them: An Application to the Internet." American Economic Review 96, no. 2: 108-13.

Gordon, Robert J. 2016. The Rise and Fall of American Growth: The U.S. Standard of Living since the Civil War. Princeton University Press.

Karaian, Jason. 2015. "We Now Spend More Than Eight Hours a Day Consuming Media." Quartz, June 1.

Nordhaus, William D. 2007. "Two Centuries of Productivity Growth in Computing." Journal of Economic History 67, no. 1: 128-59.

Stewart, James B. 2016. "Facebook Has 50 Minutes of Your Time Each Day; It Wants More." New York Times, May 5.

Syverson, Chad. 2016. "Challenges to Mismeasurement Explanations for the U.S. Productivity Slowdown." Working Paper no. 21974. Cambridge, Mass.: National Bureau of Economic Research.

GENERAL DISCUSSION Dan Sichel commended the paper for making a good case that mismeasurement in the technology sector is not the right way to understand the productivity slowdown, as the paper convincingly shows that there has not been either (i) a big shift in shares toward the unmeasured or quality-measured sectors, or (ii) a really big step up in the amount of mismeasurement, what he called the "ingredients" of the mismeasurement story. However, he expressed concern that a casual reader might come away with the wrong impression. In particular, while mismeasurement is not a good explanation for the productivity slowdown, mismeasurement remains a big problem, particularly in the technology, education, and health care sectors. For instance, Sichel cited a paper that he coauthored with David Byrne and Stephen Oliner that convincingly documents big measurement problems in the official price indexes for semiconductors. (1)

In addition, Sichel noted that the present paper highlighted a range of other potential measurement problems for the whole span of the technology sector. Though the authors use rough calibrations, he thought that the calibrations were sensible, and that they highlighted the need for what Matthew Shapiro and David Wilcox called the "house-to-house combat of price measurement." (2) That is, rather than simply applying a plausible calibration, one should actually go and quantify the mismeasurement. Even though mismeasurement is not an explanation of the productivity slowdown, technology is a really dynamic, important sector in the economy, and it remains important to support the Bureau of Economic Analysis (BEA) and the Bureau of Labor Statistics (BLS) in their efforts to try to do a better job at correcting it. More broadly, one would not want the casual reader to conclude that, because mismeasurement does not explain the productivity slowdown, there is no need to worry about mismeasurement anymore. Mismeasurement is still a big problem, he concluded, and it needs to be addressed.

Martin Feldstein had three short comments. First, he believed that one big problem was not just the recent decline observed in productivity but also the volatility in the series. Second, he expressed interest in further examining the long-term bias in output measurement. In a piece written for the Wall Street Journal, Feldstein argued that, despite all the best efforts of the BLS, the problems associated with measurement for new and improved products means that the rate of improvement is underestimated. (3) However, he stresses that products are only a small part of the problem, and that the measurement of services is much more important. Roughly 80 percent of private sector employment is in services, and figuring out how to measure improved output in many service sectors, he believes, is going to be hopelessly difficult. And third, he made a point about the interpretation of these statistics. Some mistakenly tend to treat productivity statistics as indicators of well-being, or of consumer utility, when--as the authors correctly emphasize--they are only about market activities. There has long been a tension in the history of the National Income and Product Accounts and in what the BLS does in focusing on market activities where, when people read statistics about, say, income growth, they interpret that in terms of the value to users, rather than just the market aspect of it.

Jason Furman agreed with Feldstein that the paper was excellent, and that it removed whatever sliver of doubt he had that mismeasurement was an important part of the explanation for the productivity slowdown. He followed with two comments. First, he noted that the less weight put on mismeasurement as the explanation for the slowdown, the more optimistic one should be about the growth measured over the next decade, the reason being that the methods used to measure productivity growth are considerably more persistent than the underlying "true" productivity growth itself. He suggested that a lack of capital investment over the last 5 to 10 years was a big part of the low-productivity story, and that he was optimistic that productivity would eventually rebound. Second, he warned against discounting measurement errors not in the form of persistent biases or incomplete sources. As more data become available in the coming years, past productivity numbers will likely need to be revised, and while one should think of those revisions as being unbiased, given the recent large disconnect between strong employment growth and weak output growth, it could be more likely that output and productivity will be revised up than down. If, for instance, productivity growth is presently measured to be 0.5 percent per year, Furman believed that it was more likely in five years the number would be revised up to 1, rather than revised down to 0.

Robert Hall noted that an important distinction needed to be made between productivity and consumer surplus, concepts that discussants in the room seemed to him to be conflating. It is really important to understand that productivity cannot be measured by consumer surplus, and how much consumer surplus is associated with output is a totally separate question. He urged discussants to be sure to keep the two concepts apart. Similarly, Hall noted the important distinction between output per hour and total factor productivity (TFP). Output per hour is just another endogenous variable, he pointed out, whereas TFP is fundamental. He encouraged everyone to keep an eye on what the paper basically does on TFP, and not to confuse it with looking at output per hour, though there is of course a close relationship between the two.

On the question of the future, Hall cited a paper that he recently encountered by Diego Comin and Mark Gertler that stakes out the claim that the reason productivity grew so slowly from 2010 onward was a failure to adopt existing technology, and that as the economy returns to normal in particular, as the things that have held back investment subside--there will be a closing of the gap, because the creation of new technology will have gone on at normal rates. (4) Hall added that a very interesting fact not mentioned in the discussion at all was that research and development spending did not decline at all during the crisis or afterward.

David Romer echoed Furman and Feldstein in stating how terrific the paper was, and added that it crushed any sliver of hope he had that mismeasurement could explain the productivity slowdown. However, he thought the paper conceded too much on one point, which concerned the question of whether we can "rescue" some of the slowdown--not of productivity growth, but in the growth of consumer welfare--by appealing to the consumer surplus created by new technologies. For growth, the relevant question is not whether the recent innovations have increased consumer surplus but whether they are contributing more to consumer surplus than earlier innovations. Since--as discussant Robert Gordon is fond of pointing out--the amount of surplus created by earlier innovations was enormous, he was skeptical that bringing in consumer surplus would eliminate any noticeable part of the growth slowdown. He suggested that in discussing consumer surplus, the authors not just address how much recent innovations have contributed to consumer surplus but also at least mention that what is relevant for growth is changes in those contributions over time.

Justin Wolfers made three points, jokingly claiming that one point he did not believe, one was not true, and one was for introductory economics, so it may be true. The first point, which he did not believe, concerned the question of whether or not any of this really mattered. That is, what would economists do differently if productivity growth was measured appropriately? In terms of cyclical policy, productivity growth is already understood to be sufficiently noisy, and one does not typically look to the productivity numbers to figure out whether the latest slowdown was due to a productivity shock. In terms of long-run policy, whether productivity growth in the past decade was 0.05 percent, 1.5 percent, or 2.5 percent, one would still want to pick up any policy that would add another 1 or 2 points to that. According to Wolfers, most of the things an economist might want to advocate policymakers do would probably not change even if a better job was done at measuring productivity.

The second point Wolfers made related to consumer surplus and economists' apparent obsession with free stuff. There is a view that free services such as Facebook have significantly added to growth and consumer surplus, more so than goods and services that are not free. He pointed out that goods and services do not necessarily have to be free to generate a lot of consumer surplus. As a humorous example, he suggested that the introduction of the discount furniture store IKEA to the United States--which made "Scandinavian clean lines and bright colors" more affordable--arguably generated more of a consumer surplus than Facebook.

Last, Wolfers appealed to introductory economics to make the case to the authors that the idea of creating some sort of "national consumer surplus accounts ' would be bad, and that the authors were right to resist. The diamond--water paradox," an idea often taught in introductory economics, is the apparent contradiction that, although water is on the whole more useful than diamonds in terms of survival, diamonds command a higher price in the market. Suppose six glasses of water are sufficient to sustain life and that the seventh and eighth only serve to ensure "healthy glowing skin." Pricing out the consumer surplus from the first six glasses, one might pay $200,000 for the first six glasses, because otherwise he would be dead. By that logic, the increment to consumer surplus from free services such as Facebook relative to the increment to consumer surplus from the fact that water exists is going to be very small. Rather than try to measure consumer surplus, one could simply measure well-being, and have a "well-being count." The simple way to do that, Wolfers pointed out, is by asking people how happy they are, and it turns out we already know how to do that.

John Haltiwanger spoke next about the micro productivity evidence for the ideas presented in the paper. The productivity slowdown in the high-tech sector, he noted, seems to be supported by the micro evidence. Likewise, the surge in retail trade productivity over the 1990s is evident in both the macro and micro evidence. What is puzzling, Haltiwanger pointed out, is the apparent collapse in retail trade productivity that the authors report around 2004; according to the micro data, there does not seem to have been this collapse. Haltiwanger suggested two findings from the micro data that raise questions about the macro evidence. First, the restructuring from single-unit establishment firms toward large national chains continues unabated over this period. Second, the productivity gap between single-unit establishment firms and large national chains is just as big as it ever was. The bottom line is that the micro dynamics still suggest that retail trade productivity ought to be doing quite well from these reallocation dynamics. According to the macro data, however, apparently it is not.

Melissa Kearney wondered if it was worth taking seriously the idea that the ways in which people use some free digital services might actually increase the labor supply. Many of the discussants had pointed to technology as being a substitute for leisure time, but perhaps access to technology also frees up time for additional work. For example, an individual might use her smart phone for home production--such as family shopping or scheduling--late at night, freeing up time for work during the day. She cited a paper by Lisa Dettling that found that exogenously determined high-speed home Internet access led to an increase in labor force participation for married women. (5) Kearney wondered if the authors had considered anything like this in their analysis.

Jonathan Pingle commended the authors for contributing to what is really important work suggesting a noticeable deceleration in structural productivity growth. However, adding on the implications of Bruce Fallick and others' cohort component model for aggregate labor supply--presented at the Fall 2014 Brookings Papers meeting (6)--and the normal business sector GDP gap, one struggles to talk about potential output growth of 1.5 percent in the United States now, and certainly a deceleration of over 2 percentage points. With many firms, businesses, and policymakers continuing to plan based on backward-looking expectations, that is potentially a big problem.

Joe Beaulieu found, countering Wolfers, that the idea of thinking about consumer surplus is a somewhat interesting but sideline conversation to TFP. He suggested that many of the free digital goods and services talked about essentially take the form of advertising. If there is a huge productivity increase in the advertising industry, that is an intermediate good, which probably means that prices for advertising services have fallen considerably. Therefore, he concluded that productivity properly measured for the rest of the sectors in the economy must be even worse than originally thought. This also raises a second interesting question, which is how one should think about advertising and the implications for the economy. A lot of the advertising industry may involve rent-seeking behavior, which then has interesting implications not only for how one thinks about the economy and what is going on now versus a few years ago but also for how one might actually measure some of these things.

Alan Auerbach raised the question of how the production of multinationals is measured in the U.S. accounts, given that tax-induced profit shifting leads them to understate U.S. profits for tax purposes. Marshall Reinsdorf noted that much of the production could really be taking place in the United States, but that it gets reported overseas. He remarked that a recent paper by the BEA found that roughly 1 percent of GDP is explained by the fact that multinational corporations do most of their production based on where their labor and physical capital are located. He suggested that one might think of this estimate as an upper bound.

David Byrne noted that in addition to failing to reject the hypothesis that there is a productivity slowdown, in order to square the difference between the perspectives of Silicon Valley--that there is in fact a productivity slowdown in the technology sector--with what is actually observed, one should return to a Fortune 500 survey that discussant Martin Baily brought up during his remarks. In that survey, many of the companies said that their most important problem at the moment was adapting to new technology. One way to interpret this is that there is a technical frontier that has moved outward, and that these companies have not figured out what to do with it yet, as it is something that requires a tremendous amount of intangible investment. But another way to look at this situation is that it appears to be somewhat harder now to move from the back to the frontier perhaps than it once was, a finding supported by the work of Haltiwanger and others. This is potentially a different way of looking at the results of the paper.

Sichel had earlier cited a paper coauthored with Byrne and Stephen Oliner, which documented big measurement problems in the official price indexes for semiconductors, and noted that the tenors of that paper and the present paper seemed to be a little different. The right way to think about the Sichel-Byrne-Oliner paper, according to Byrne, from a GDP perspective is that microprocessors (MPUs) and semiconductors act as intermediate goods, so they are only going to show up to the extent that their production affects net trade. As it happens for MPUs--which are one of the most important individual goods that the United States trades--they are big in exports, but they are also big in imports. So, if one makes the price index for MPUs fall faster to a first approximation, it has no effect on GDP--not only because MPUs are intermediate goods but also because of the global value chain. The chips may be fabricated in the United States, tested in Costa Rica, and sent back to the United States to be used in data centers, and the transaction price is captured going both ways. The better way to think about the Sichel-Byrne-Oliner paper and the MPUs, according to Byrne, is to look at the MPU price index as a barometer of what is going on in the frontier of the information technology (IT) sector. The huge slowdown observed in the price index for recent years is very alarming, he noted; what the authors find in the present paper is that, for a number of reasons, that MPU price index is just not that accurate.

It had also been mentioned that IT is less important today than it once was, and Byrne noted that that is certainly true as a share of investment. Nevertheless, IT is still an important portion of production. BEA's investment deflator for IT includes things like computers, equipment software, and other special-purpose equipment. In the 1980s, that high-volume production used to be a much bigger share of IT investment than it is now--about 50 percent then, compared with only about 25 percent today. Investment and production of the low-volume, special-purpose equipment is much more difficult to measure. All this is not to say that mismeasurement is bigger than was found in the present paper. Rather, it is that the confidence interval around the estimate is much bigger than it once was. And it is not that the statistical agencies do not know about these problems; it is that they need more funding for measurement. He joked, "We are not supposed to be making price indexes at the Federal Reserve Board," but that, nevertheless, they do because despite it being well known that there are solutions to the problems of mismeasurement, the funding just is not there for the statistical agencies.

On the question of why more was not said about the dire labor productivity outcomes during the past five years, John Fernald noted that, at least when looking at TFP, it does not actually look worse; the past five years for TFP look about the same as the years before the Great Recession. In growth accounting, one observes that while so many people have been newly hired, capital accumulation has not kept up. One common theme in the room, Fernald noted, was that much is still unknown, and the paper only touches narrowly. He conceded that the measurement of health care, services, and new goods more broadly are all really challenging issues.

One final point Fernald made pertained to the value of the Internet. In terms of growth, the Internet brings no more than a few basis points, which he admitted was probably an overstatement; the authors' best estimate was closer to zero. Part of this, he explained, is because the market portion of the Internet is in advertising support. The big benefits are nonmarket, but these are still important, and things at the border of the market are shifting in interesting ways. Perhaps technology has increased the opportunity cost of working for less-educated workers. There is much need to study this, he concluded, adding that it would be great if the statistical agencies could acquire the resources to be able to develop better satellite accounts, and if researchers were working on ways to measure well-being.

Table 1. Utilization-Adjusted Total Factor
Productivity Growth, 1947-2015

Period            TFP growth (percent)

1947:Q2-1973:Q1   2.10
1973:Q1-2015:Q4   0.82
1973:Q1-1995:Q1   0.52
1995:Q1-2004:Q1   1.99
2004:Q1-2015:Q4   0.48
2004:Q1-2010:Q1   0.73
2010:Q1-2015:Q4   0.21

Source: Federal Reserve Bank of San Francisco (http://www.frbsf.org/
economic-research/indicators-data/total-factor-productivity-tfp/).

(1.) David M. Byrne, Stephen D. Oliner, and Daniel E. Sichel, "Is the Information Tech..jlogy Revolution Over?" Finance and Economics Discussion Series no. 2013-36 (Washington: Board of Governors of the Federal Reserve System, 2013).

(2.) Matthew D. Shapiro, and David W. Wilcox, "Mismeasurement in the Consumer Price Index: An Evaluation," NBER Macroeconomics Annual 11 (1996): 93-154.

(3.) Martin Feldstein, "The U.S. Underestimates Growth," Wall Street Journal, May 18, 2015.

(4.) Diego Anzoategui, Diego Comin, Mark Gertler, and Joseba Martinez, Endogenous Technology Adoption and R&D as Sources of Business Cycle Persistence," Working Paper no. 22005 (Cambridge, Mass.: National Bureau of Economic Research, 2016).

(5.) Lisa J. Dettling, "Broadband in the Labor Market: The Impact of Residential High-Speed Internet on Married Women's Labor Force Participation," Finance and Economics Discussion Series no. 2013-065 (Washington: Board of Governors of the Federal Reserve System, 2013).

(6.) Stephanie Aaronson, Tomaz Cajner, Bruce Fallick, Felix Galbis-Reig, Christopher Smith, and William Wascher, "Labor Force Participation: Recent Developments and Future Prospects," Brookings Papers on Economic Activity, Fall 2014: 197-255.

DAVID M. BYRNE

Federal Reserve Board

JOHN G. FERNALD

Federal Reserve Bank of San Francisco

MARSHALL B. REINSDORF

International Monetary Fund

(1.) Section I and the online appendix discuss data, the timing of the bars in the chart, and the similar pattern in measures of TFP. The online appendixes for this and all other papers in this volume may be found at the Brookings Papers web page, www.brookings.edu/bpea, under "Past Editions."

(2.) In independent work, Syverson (2016) suggests a similar calculation of the missing growth.

(3.) There are also some sources of upward measurement error in growth related to globalization that have become less important. Still, we will usually take "mismeasured" to mean "causing GDP growth to be understated."

(4.) Nordhaus (2006) sketches principles of national accounting for nonmarket as well as market goods and services.

(5.) Some academic research found even larger effects--for example, Bils and Klenow (2001)--while Schultze and Mackie (2002) argued for a smaller number.

(6.) A separate debate is whether the productivity slowdown of the 1970s was itself due to mismeasurement. Griliches (1994) points out that the post-1973 slowdown was concentrated in poorly measured industries. Gordon (2016) argues instead that the post-1973 slowdown reflects the unusual strength of the 1920-70 period rather than anything specific that happened in the 1970s. Relatedly, Fernald (1999) estimates that building the Interstate Highway System substantially boosted productivity growth in the 1950s and 1960s, but then its effects ran their course. Triplett (1999) reviews arguments that the post-1973 slowdown was illusory.

(7.) A possibly more optimistic perspective on recent developments comes from noting that TFP growth has continued since the Great Recession at its pre-1995 pace. This pace of TFP growth may be normal--it was, perhaps, the 1995-2004 period that was exceptional. Furthermore, in recent years TFP may be more relevant than labor productivity, whose weakness since 2010 partly reflects transitory factors associated with weak capital deepening.

(8.) "Other well-measured" includes most of manufacturing (except computers and electronics equipment), agriculture, mining, utilities, transportation, broadcasting, and accommodations. Nordhaus (2002) also considers wholesale and retail trade as well measured, but we have broken that out separately.

(9.) Value added weighting of value added TFP growth is essentially equivalent to doing so-called Domar weighting of gross output residuals (Domar 1961). The fixed weights are based on nominal expenditures, not quantities. In the data, the rise in the nominal share of services reflects both faster growth in quantities and faster growth in prices.

(10.) Our focus in this section is on the contribution of IT capital services to productivity and its implications for TFP growth. Parallel measurement problems exist for IT consumer durables, which we do not discuss explicitly. However, we account for understatement of GDP from the mismeasurement of IT through our adjustments to domestic production, whether for the consumer or business market.

(11.) See research for communications equipment (Byrne and Corrado 2015), computers (Byrne and Pinto 2015; Byrne and Corrado 2016), and microprocessors (Byrne, Oliner, and Sichel 2015).

(12.) With appropriate data on characteristics, hedonic regressions are a useful tool for quality-adjusting prices, but the absence of hedonic adjustment does not necessarily indicate that a price index is biased. Other techniques may also account for quality improvements (Wasshausen and Moulton 2006).

(13.) Research for the remaining category, peripherals, is sparse. The BEA investment price index fell 12 percent a year on average in the 1990s, but 4 percent on average since then. Aizcorbe and Pho (2005) examine scanner data for eight categories of peripherals for the years 2001-03. Although we note that the geometric mean of price indexes for these categories falls 15 percent per year, we chose not to adjust the peripherals index based on this short time series.

(14.) Navigational equipment and audiovisual equipment are classified as communications equipment in the BEA investment taxonomy.

(15.) Although the sequencing of a human genome is not final output, improvements in the tools used to conduct science are the likely foundation of falling prices for health services in the future. Heather and Chain (2016, p. 6) present the history of DNA sequencing equipment, and they note that "over the years, innovations in sequencing protocols, molecular biology and automation increased the technological capabilities of sequencing while decreasing the cost, allowing the reading of DNA molecules that are hundreds of base pairs in length, massively parallelized to produce gigabases of data in one run." On the role of high-performance computing in genetics, also see Stein (2010).

(16.) Byrne and Corrado (2016) have added estimates of an alternative price index for software since this paper was written. Their price index accelerates by roughly the same amount (1.9 percent) as the price index we employ (1.7 percent). Consequently, the contribution of IT price mismeasurement to the productivity slowdown would not change if we employed their index. Their price index falls 3 percentage points faster in both periods, implying a somewhat greater contribution to labor productivity of capital deepening and smaller contribution of TFP both before and after 2004, but roughly the same acceleration of TFP.

(17.) Though not original to them, Basu and others (2004) make this point in the context of intangible investment. Dale Jorgenson had made this observation to Fernald when software investment was added to the U.S. GDP in 1999.

(18.) Note, as well, that the slower pace of aggregate TFP growth would be distributed unevenly. Suppose the mismeasurement reflects faster true TFP growth in domestic equipment and software goods. Then TFP growth in the other industries must be slower than measured. Intuitively, this happens because growth in their capital input is more rapid than measured, but growth in their output is the same as measured.

(19.) The careful reader will note that labor productivity growth for 1995-2004 is about 0.1 percentage point higher in column 3 than column 2, as is capital growth. So why does TFP growth fall, even though the labor productivity effect looks larger than the adjusted contribution of capital (capital's share times capital growth)? The reason is that, with intangibles, capital's share is also adjusted upward, and so the effect on TFP involves not just the adjustment to capital growth but also the adjustment to capital's share multiplied by (the new) capital growth rate. This effect can be a few tenths.

(20.) In "The GNU Manifesto," Richard Stallman (1985) describes his vision that "in the long run, ... nobody will have to work very hard just to make a living. People will be free to devote themselves to activities that are fun, such as programming." (We thank Hank Farber for pointing us to this quotation.)

(21.) This is according to the American Time Use Survey (http://www.bls.gov/tus/tables/ al_all_years.xlsx).

(22.) As a nonprofit institution serving households, Wikipedia's output, about $0.2 billion in 2011, is counted as personal consumption. The $25.2 billion thus overstates the adjustment that could be made to GDP by $0.2 billion.

(23.) The online appendix shows that the Laspeyres and Paasche quantity indexes that are averaged to obtain the Fisher index are upper and lower bound measures of the relative change in consumer surplus.

(24.) We thank Joo Hee Oh for sending these data. Our calculation corresponds to Brynjolfsson and Oh's (2014) equation 14 on the money benefits.

(25.) Another revenue source for providers of free digital services is the valuable information that the users of these services reveal about themselves, but this is a small revenue source compared with advertising.

(26.) In another project we are working on (Byrne, Fernald, and Reinsdorf ongoing work), we discuss a way to bring the extra value added into the business sector, as opposed to being in the household sector. Their approach requires special treatment of the advertising revenue, so that some of the output that is currently viewed as intermediate consumption by the ad buyers can instead be viewed as consumed by households. This would make business sector nominal and real value added larger, but the effect on TFP growth is still close to zero. A separate issue is that a more explicit agreement for consumers to watch the ads in order to receive the services would be required for the ad watching to qualify as a barter transaction under the international guidelines of the System of National Accounts (United Nations and others 2009, para. 3.51 and 3.53, pp. 43-44).

(27.) A well-known example--the need to distinguish between increases in welfare and productivity gains--occurs with changes in terms of trade. Favorable shifts in exports and imports increase the opportunity to gain from trade, allowing real consumption to rise as the economy moves to a different point on the production possibility frontier.

(28.) The Census Bureau defines e-commerce as purchases made over the Internet or other electronic network or via email. The e-commerce shares for products that are easy to order online (such as books) are even larger, because some products (such as gasoline and building supplies) presumably involve little e-commerce (https://www.census.gov/retail/mrts/www/ data/excel/tsnotadjustedsales.xls). Evans, Schmalensee, and Murray (2016) conjecture that the Census figures underestimate e-commerce.

(29.) Schreyer, Brandt, and Zipperer (2015) discuss an alternative approach to measuring multifactor productivity (MFP) for mining that includes services of natural resource assets. The Australian Bureau of Statistics publishes an experimental measure of MFP for mining that includes services of subsoil natural resource assets in inputs. In the tables released in December 2015, this raises the estimated growth rate of mining MFP between 2000-01 and 2014-15 from -4 percent a year to -1 percent a year. Similarly, Zheng and Bloch (2014) find that adjusting for inputs of natural resources, declining returns to scale, and capacity utilization raises MFP growth for the mining industry of Australia between 1974-75 and 2007-08 from -0.2 percent a year to 2 percent a year.

(30.) These data were downloaded at http://www.abs.gov.aU/AUSSTATS/abs@nsf/ DetailsPage/5260.0.55.0022014-15.

(31.) This problem is examined by Houseman and others (2011) and Mandel (2009).

(32.) Where they are in the source data is not clear. However, the Quarterly Services Survey indicates slowing nominal growth of the local transportation measure that includes taxis, which is where one would expect to find the new kinds of local transportation services.

(33.) Wasshausen and Moulton (2006) discuss how statistical agencies incorporate quality adjustments.

Table 1. Prices and Weights for Information Technology Investment (a)

Measure                           1947-78   1978-95   1995-2004 2004-14

IT investment share of
  business fixed investment         12.2      23.6      30.7      29.3
IT investment price indexes
  National Income and
    Product Accounts                 0.2      -2.2      -6.1      -1.4
  Conservative alternative (b)      -1.8      -4.4      -9.2      -4.4
  Liberal alternative (c)           -3.9      -6.5     -11.2      -5.9
Share of IT investment
  Computers and peripherals         13.1      22.8      20.8      14.5
  Communications equipment          36.9      26.6      22.6      17.0
  Other information systems
    equipment                       38.3      26.7      17.3      20.4
  Software                          11.7      23.9      39.3      48.2
Price deflators
  Computers and peripherals (d)
    National Income and
      Product Accounts             -18.1     -14.6     -19.3      -6.6
    Alternative                    -18.1     -19.0     -27.3     -18.6
  Communications equipment
    National Income and
      Product Accounts               1.9       1.4      -5.4      -2.7
    Alternative                     -3.0      -2.7     -11.2     -10.3
  Other information systems
      equipment
    National Income and
      Product Accounts               2.3       2.9      -0.6       0.5
    Alternative                     -1.7      -2.2      -8.9      -4.9
  Software (d)
    National Income and
      Product Accounts              -0.7      -1.2      -1.1       0.1
    Alternative                     -4.8      -4.4      -2.5      -0.8

Sources: U.S. Bureau of Economic Analysis; Byrne and Corrado
(2016).

(a.) All values are expressed as percents.

(b.) Incorporates alternative computer and communications equipment
prices.

(c.) Incorporates alternative software and special-purpose
equipment prices.

(d.) Price indexes begin in 1958.

Table 2. Adjustments to Business Sector Growth Accounting (a)

                                 (0)
   Measure                    Published
  of growth       Period     baseline (c)

Labor            1978-95      1.50
  productivity   1995-2004    3.26
                 2004-14      1.44
                   2004-10    1.92
                   2010-14    0.71

Capital-to-      1978-95      2.20
  hours ratio    1995-2004    3.68
                 2004-14      1.80
                   2004-10    3.14
                   2010-14   -0.22

Total factor     1978-95      0.53
  productivity   1995-2004    1.82
                 2004-14      0.49
                   2004-10    0.44
                   2010-14    0.58

                             Annual percentage point
                             change relative to baseline (b)

                                 (1)          (2)           (3)
   Measure                   Conservative   Liberal      Liberal +
  of growth       Period         (d)          (e)     intangibles (f)

Labor            1978-95      0.12           0.21      0.30
  productivity   1995-2004    0.27           0.38      0.49
                 2004-14      0.13           0.19      0.18
                   2004-10    0.17           0.25      0.24
                   2010-14    0.06           0.11      0.10

Capital-to-      1978-95      0.27           0.52      0.66
  hours ratio    1995-2004    0.54           0.89      1.02
                 2004-14      0.44           0.70      0.55
                   2004-10    0.46           0.74      0.54
                   2010-14    0.41           0.63      0.58

Total factor     1978-95      0.04           0.05     -0.01
  productivity   1995-2004    0.09           0.09     -0.08
                 2004-14     -0.04          -0.07     -0.12
                   2004-10    0.00          -0.02     -0.12
                   2010-14   -0.10          -0.14     -0.12

Sources: Fernald (2014); Corrado and Jager (2015).

(a.) Averages start in 1978 because of the availability of
intangibles data.

(b.) Each column involves a separate, experimental adjustment to
selected components of capital investment. The entries show the
percentage point adjustment to business sector growth accounting
components, relative to the unadjusted estimates in column 0.

(c.) Baseline (the business sector) measured as the percent change
at an annual rate.

(d.) Alternative deflators for computers and communications.

(e.) Column 1, plus alternative deflators for specialized equipment
and software.

(f.) Column 2, plus intangibles from Corrado and Jager (2015).