Productivity Program meeting.

The NBER's Productivity Program met in Cambridge on March 9. Program Director Ernst R. Bern& of MIT organized the meeting. The program was:

Sinan Aral, MIT; Erik Brynjolfsson, MIT and NBER; and Marshall Van Alstyne, Boston University, 'Information Technology and Information Worker Productivity: Task Level Evidence"

Discussant: Susan Helper, Case Western Reserve University and NBER

Nick Bloom, Stanford University and NBER, "The Impact of Uncertainty Shocks: A Firm-Level Estimation and a 9/11 Simulation"

Discussant: Susanto Basu, Boston College and NBER

Timothy Erickson, Bureau of Labor Statistics, and Ariel Pakes, Harvard University and NBER, "An Experimental Component Index for the CPI: From Annual Computer Data to the Monthly Data on Other Goods"

Discussant: Jack Triplett, Brookings Institution 

James Adams, Rensselaer Polytechnic Insitute and NBER: Announcement of the NBER-RPI Scientific Papers Database

"Recent US Productivity Growth: A One-Time Blip or Sustainable?"--a panel discussion moderated by Ernst R. Berndt--presentations by: Kevin J. Stiroh, Federal Reserve Bank of New York, and Dale W. Jorgenson, Harvard University, "A Retrospective Look at the U.S. Productivity Growth Resurgence"; Daniel Sichel, Federal Reserve Board; Robert J. Gordon, Northwestern University and NBER, "Exploding Productivity Growth: Context, Causes, and Implication"; and Barry Bosworth and Jack Triplett, Brookdngs Institution, "The 21st Century Productivity Expansion Is STILL in Services"

Liran Einav, Stanford University and NBER, and Aviv Nevo, Northwestern University and NBER, "Errors in Self-Reported Data: A Cross-Validation of Homescan Data"

Discussant: Alvin J. Silk, Harvard University

In an effort to reveal the fine-grained relationships between IT use, patterns of information flows, and individual information-worker productivity, Aral, Brynjolfsson, and Alstyne study task-level practices at a midsize executive recruiting firm. They analyze both project-level and individual-level performance using: 1) detailed accounting data on revenues, compensation, project completion rates, and team membership for over 1300 projects spanning five years; 2) direct observation of over 125,000 e-mail messages over a period of ten months by individual workers; and 3) data on a matched set of the same workers' self-reported IT skills, IT use, and information sharing. These detailed data allow the researchers to econometrically evaluate a multistage model of production and interaction activities at the firm, and to analyze the relationships among key technologies, work practices, and output. They find that: IT use is positively correlated with non-linear drivers of productivity. Further, the structure and size of workers' communication networks are highly correlated with performance. There is also an inverted-U shaped relationship between multitasking and productivity such that, beyond an optimum, more multitasking is associated with declining project completion rates and revenue generation. Finally, asynchronous information seeking--such as email and database use--promotes multitasking, while synchronous information seeking over the phone is negatively correlated with multitasking. Overall, these data show statistically significant relationships among technology use, social networks, completed projects, and revenues for project-based information workers. The results are consistent with simple models of queuing and multitasking, and these methods can be replicated in other settings, suggesting new frontiers for IT value and social network research.

Uncertainty appears to vary strongly over time, temporarily rising by up to 200 percent around major shocks like the Cuban Missile crisis, the assassination of JFK, and 9/11. Bloom offers the first structural framework to analyze uncertainty shocks. He builds a model with a time-varying second moment, which he solves numerically and estimates using firm-level data. He then uses the parameterized model to simulate a macro uncertainty shock, which produces a rapid drop and rebound in employment, investment, and productivity growth, and a moderate loss in GDP. The temporary impact of a second-moment shock is different from the typically persistent impact of a first-moment shock, highlighting for policymakers the importance of identifying the relative magnitudes in major shocks. Comparing the simulation of an uncertainty shock to the VAR estimations on monthly data and a 9/11 event-study, Bloom finds that both display a surprisingly good match.

The BLS staff recently increased the rate at which they incorporate techniques to correct for selection effects into their component indexes. However, their work--and the work of other researchers--shows very little difference between hedonic and matched-model indices for certain components of the CPI. Erickson and Pakes explore why. They look carefully at the component index for TVs and show that differences between the TV and computer markets, together with the fact that the BLS data are high frequency, make it necessary to use a more general hedonic correction procedure than has been used to date. The computer market is special in that it has both well defined cardinal measures of the major product characteristics and "exiting" goods which have relatively low values. In markets where such measures are absent and where turnover can be at the high-quality end, one needs to allow for selection on unmeasured, as well as measured, characteristics. Also, in high frequency data one needs to correct for differential "sticky price" rates among different goods. The researchers develop an hedonic selection correction that accounts for these phenomena; they show that, when applied to TVs, it yields much larger selection corrections. In particular, they find that matched-model techniques underestimate the rate of price decline by over 20 percent. When they apply the BLS's correction algorithm to their data, they find that it does generate a substantial correction to the matched-model index, but one of only 7.8 percent. Moreover, the BLS staff's recent successful push to modernize their data gathering procedures has made it possible to compute the researchers' index within the BLS's time constraints, making it a real-time alternative to current procedures.

It is now widely recognized that information technology (IT) was critical to the dramatic acceleration of U.S. labor productivity growth in the mid-1990s. The paper by Jorgenson, Stiroh, and Mun S. Ho traces the evolution of productivity estimates to document how and when this perception emerged. Early studies concluded that IT was relatively unimportant. It was only after the massive IT investment boom of the late 1990s that this investment and underlying productivity increases in the IT-producing sectors were identified as important sources of growth. Although IT has diminished in significance since the dot-com crash of 2000, these researchers project that private-sector productivity growth will average around 2.5 percent per year for the next decade, only moderately below the average of the post-1995 period.

Sichel presented some preliminary results based on aggregate data from a paper with Steve Oliner and Kevin Stiroh. That paper looks back at the past ten years of U.S. productivity performance in light of recent critiques of the standard growth accounting methodology that lies at the heart of many analyses of productivity. Specifically, the paper augments the standard framework to account for adjustment costs for capital investment, variable factor utilization, and the role of intangible capital. Regarding intangibles, the paper extends the work of Basu, Fernald, Oulton, and Srinivasan (2003) to develop a new measure of intangible investment and capital related to information technology. Qualitatively, the new measures exhibit a pattern over time similar to that generated by other research (for example, see Corrado, Hulten, and Sichel, 2006). As for the role of information technology (IT), after augmenting the standard growth accounting framework to take these critiques on board, IT still was a key driver of the pickup in labor productivity growth over 1995-2000; since 2000, IT played a smaller, but still sizable, role. As for the continuing strength in the growth of labor and multifactor productivity since 2000, augmenting the standard framework alters the time profile of productivity growth since 1995. Specifically, taking on board the critiques--especially, the inclusion of intangibles--makes the gains over 1995-2000 larger and takes some of the luster off the performance since 2000. As for the outlook for productivity growth, Sichel discussed the crosscurrents affecting U.S. productivity performance, including cyclical dynamics, technical progress, and demand for new IT applications.

Gordon discusses his research which has three goals. The first is to forecast growth in U.S. potential real GDP, not for the full 75-year horizon of the Social Security trustees, but for the more modest but still daunting span of the next two decades. He brings together recent research, both about productivity and about the likely future behavior of the other four factors, especially population growth, that matter for potential output growth. The need to predict future population growth in turn requires an exploration of the determinants of trends in fertility and mortality rates, as well as the likely future trend of net immigration into the United States. The second goal of his paper, connected closely with the first, is to interpret the extraordinary productivity performance of the United States since 1995 and especially since mid-2000. Far from slowing in response to the 2001 recession and the collapse of investment in information and communications technology (ICT) after mid-2000, growth in labor productivity actually accelerated from all average of 2.56 percent a year between 1995:4 and 2000:2 to 3.46 percent a year between 2000:2 and 2003:2. Should a forecast of future productivity growth use as its precedent the average behavior of actual productivity growth over the past two years, the past eight years, or some longer interval? The third goal of the paper, related to the first two, is to provide a new breakdown of past U.S. economic growth into its trend and cyclical components. In assessing long-term growth performance over some historical period, one would not want to include the portion of real GDP growth contributed by a sharp difference in cyclical conditions, for example between the 7.6 percent unemployment rate of mid-1992 and the 4 percent rate of early 2000. This paper bases its cyclical analysis on an identity that links real GDP to productivity, the employment rate, and several other variables. The analysis uncovers important changes in cyclical behavior between the earlier postwar downturns and the two recent jobless recessions and recoveries (1990-3 and 2001-3). One particularly important difference is the strength of productivity growth and the weakness of payroll employment growth in both of the most recent episodes and especially in the latest.

Labor productivity (LP) grew 2.5 percent per year during the 1995-2005 period, nearly double its growth rate over the previous two decades. But services sector LP and multifactor productivity (MFP) grew more rapidly and substantially exceeded productivity acceleration in the goods-producing sector. Bosworth and Triplett show that the services sector contributed three-quarters of U.S.-in-MFP growth after 1995, and that within services, the contribution of MFP to LP growth exceeded the vaunted contribution of IT investment. They also find that the services sector has become even more significant as the primary source of sustained productivity growth after 2000. In this study, they compute LP, MFP, and contributions to growth accounts for 57 industries within the goods and services-producing sectors, using the new NAICS-based dataset. They also show that resource reallocations, which are a newly important factor in productivity analysis, have changed the relation between increases in industry productivity rates and aggregate and sector rates in surprising ways.

While economists have spent much time and energy thinking about sample selection issues, less effort has gone into the process of understanding what bias, if any, is caused by self reporting. Do consumers make mistakes when self-reporting data? How big are these mistakes? And, do these mistakes matter for the bottom line? Einav and Nevo contribute to the literature on cross-validation of data and examine these questions. They match self reported Homescan data, whereby consumers scan all of their purchases at home, with a very detailed and unique dataset of transactions recorded at the cashiers of a retailer. This allows them to construct a matched sample that they then use to address the questions about self reporting. In particular, they match about 200 households and more than 10,000 transactions that appear in both the retailer data and the Homescan data and report the quality of the match. While on some dimensions (for example, purchased quantities) the datasets match remarkably well, there are differences in the shopping trips recorded, likely because of errant use of loyalty cards, and even more dramatic differences in the reported transaction prices, because of the way prices are imputed in the data. Moreover, the researchers find that these price differences are systematic--they do not cancel out with aggregation, and are more likely to be associated with certain demographic groups--and therefore may lead to false conclusions. They illustrate this latter point by showing that running a price regression on an identical set of transactions may lead to different conclusions, depending on the data used to record these transactions. Besides shedding light on the general issue of self-reporting, these results are of interest for what they tell us about the quality of the Homescan data. With the declining cost of collecting and storing transaction-level data, the use of these data has been growing rapidly both in practice (for example, Nielsen has recently doubled the size of their panel) and in academic research. The data are very informative in several dimensions and have been used to study questions of marketing, competition, consumption, and nutrition. More generally, this exercise is a kind of case study demonstrating that selection bias may arise not only at the extensive margin, of whether certain individuals are in or out of the data, but also at the intensive margin, when certain individuals are more likely to be associated with recording errors.