A Time-Series Examination of the Quality of Industry-Level U.S. Productivity Data

A very large number of productivity analyses have focused on Total Factor Productivity (TFP), the volume of aggregate output produced per unit of aggregate input, as the measure of choice. For example, industry-level TFP data series have been widely used to investigate many important economic issues, including whether productivity gains have been concentrated in a few industries and whether such gains were linked to the use of information technology (Stiroh 2002), whether automation is labor-displacing (Autor and Salomons 2018), whether the recent rise in the capital share can be attributed to increasing automation (Aghion, Jones, and Jones 2019), how GDP growth has been impacted by sectoral trends in TFP and labor growth (Foerster et al. 2022), the contributions of individual industries to U.S. aggregate TFP growth (Jorgenson, Ho, and Samuels 2019), and the reasons for the productivity gap between Europe and the United States in the late 1990s and early 2000s (van Ark, O’Mahony and Timmer 2008). Recently, growing concerns about environmental degradation and climate change have spurred interest in “environmentally-adjusted” TFP indicators, which take into account the production of undesirable by-products and externalities, as well as how intensely natural resources are used (OECD 2020b). For the agricultural sector in particular, studies based on TFP have analyzed public investments (Fuglie, Wang, and Ball 2012; Fuglie 2018; Ortiz-Bobea et al. 2021), international trade (Garcia-Verdu et al. 2019; Yuan et al. 2021), and the design of policies aimed at decoupling productivity growth from environmental pressure (OECD 2020a), among other issues. In the United States, agricultural TFP measures have been extensively used to evaluate returns to public investments (Fuglie and Heisey 2007; Alston et al. 2011; Jin and Huffman 2016), identify the drivers of productivity growth (Capalbo 1988; Schimmelpfennig and Thirtle 1999; Huffman and Evenson 2006; Alston et al. 2010; Andersen, Alston and Pardey 2012; O’Donnell 2012, 2014; Plastina and Lence 2018), evaluate convergence in productivity across states (McCunn and Huffman 2000; Ball, Hallahan, and Nehring 2004; Poudel, Paudel, and Zilberman 2011), assess spillovers between agriculture and other sectors of the economy (Lence and Plastina 2020), and gauge the impact of weather and climate on aggregate productivity (Njuki, Bravo-Ureta, and O’Donnell 2018; Sabasi and Shumway 2018; Chambers and Pieralli 2020; Ortiz-Bobea, Knippenberg, and Chambers 2018; Plastina, Lence, and Ortiz-Bobea 2021; Ortiz-Bobea et al. 2021). Given the vast literature that has applied TFP to analyze issues concerning productivity, it is not surprising that significant efforts have been devoted to the development of proper measures of the individual components of TFP (OECD 2001; Fuglie, Wang, and Ball 2012; Fuglie 2015; Shumway et al. 2017; USDA-ERS 2021), as well as to the evaluation of the relative merits of alternative aggregation methods (Szulc 1964; Eltetö and Köves 1964; Jorgenson and Griliches 1967; Caves, Christensen, and Diewert 1982a, 1982b; Bjurek 1996; Balk and Althin 1996; O’Donnell 2012, 2016; Färe and Zelenyuk 2021). Contrastingly, there has been a dearth of studies exploring the quality of real-world TFP data series. Interestingly, studies analyzing productivity usually rely on a single source of TFP data, even in cases where more TFP sources are available. Typically, no robustness analyses are conducted to assess the extent to which inferences hold using alternative TFP data sources. Implicitly, such studies assume that the underlying TFP data being used is of sufficiently high quality to yield valid inferences. However, Alston (2018) and Andersen, Alston, and Pardey (2011) --among the few studies analyzing more than a single TFP source-- provide evidence that calls this assumption into question. The lack of studies concerning the quality of real-world TFP series provides the main motivation of the present investigation. We contribute to the literature by examining the industry-level TFP series for the United States obtained from three alternative sources, namely, (1) Jorgenson, Ho, and Samuels (JHS), (2) the U.S. Bureau of Labor Statistics (BLS), and (3) the U.S. Bureau of Economic Analysis (BEA). These three sources are of special interest because they are highly regarded and their series have been used extensively by researchers to analyze productivity (e.g., Stiroh 2002, Autor and Salomons 2018, Aghion, Jones, and Jones 2019, Foerster et al. 2022, Jorgenson, Ho, and Samuels 2019, van Ark, O’Mahony and Timmer 2008). Besides providing an empirical assessment of the relative quality of the aforementioned series, our study contributes to the literature by proposing a general method to examine the quality of alternative time series reportedly measuring the TFP of a particular entity or sector. The main goal of our study is to spur interest in the exploration of the quality of real-world TFP data series, with the aim of finding ways to enhance them and uncovering series whose quality may be deemed questionable. Our preliminary results show that, out of the 61 industry series for which TFP data from different sources are being compared, between 34 (for JHS vs. BEA) and 46 (for BEA vs. BLS) industries have inconsistent series across sources. In other words, only 31% to 64% of the industries have TFP data consistent between source pairs. These results strongly suggest that empirical analyses based on a single data source may not be sufficiently robust to draw strong inferences and implications. The results also demonstrate the need to devote greater attention to improving the reliability of TFP data.