--- canonical_url: "https://csls.ca/ipm-archive/international-productivity-monitor-50th-issue/the-anatomy-of-the-us-manufacturing-productivity-slowdown-evidence-from-firms-and-industries/" title: "The Anatomy of the U.S. Manufacturing Productivity Slowdown: Evidence from Firms and Industries" author: name: "mangom" url: "https://csls.ca/author/mangom/" date_published: "2026-06-09T09:55:06+00:00" date_modified: "2026-07-17T10:29:59+00:00" post_type: "ipm" summary: "Abstract U.S. manufacturing labour productivity growth fell from roughly 3.5 per cent per year over 1987–2007 to near zero over 2010–2022. Growth in total factor productivity (TFP) also fell to near zero over the same period. This article examines the sources of that slowdown by decomposing manufacturing productivity growth into contributions from leader and follower […]" ipm_issue_period: - "Spring 2026" ipm_keyword: - "Federal Reserve New York" - "firms" - "frontier firms" - "ideas harder to find" - "industries" - "Lashkari" - "Pearce" - "productivity slowdown" - "R&D" - "research productivity" - "U.S. manufacturing" ipm_research_theme: - "Sectoral and firm-level dynamics" ipm_country_focus: - "United States" featured_image: "https://csls.ca/wp-content/uploads/2026/06/IPM_50_banner.webp" status: "publish" --- # The Anatomy of the U.S. Manufacturing Productivity Slowdown: Evidence from Firms and Industries Canonical URL: https://csls.ca/ipm-archive/international-productivity-monitor-50th-issue/the-anatomy-of-the-us-manufacturing-productivity-slowdown-evidence-from-firms-and-industries/ ![](https://csls.ca/wp-content/uploads/2026/06/IPM_50_banner.webp) # IPM issue 50 Spring 2026 [Download full PDF issue](https://csls.ca/wp-content/uploads/2026/06/IPM_Issue_50_Final.pdf) Celebrating 25 years of IPM research [Issue 50 summary](https://csls.ca/ipm-archive/international-productivity-monitor-50th-issue/) ![](https://csls.ca/wp-content/uploads/2026/06/ipm-article-icon-research.svg) IPM Research Article ## The Anatomy of the U.S. Manufacturing Productivity Slowdown: Evidence from Firms and Industries Danial Lashkari and Jeremy Pearce International Productivity Monitor, Issue 50, Spring 2026 [Download full PDF article](https://csls.ca/wp-content/uploads/2026/06/IPM_Pearce_Final.pdf) Cite this article × Cite this article as: Lashkari, Danial, and Jeremy Pearce. 2026. "The Anatomy of the U.S. Manufacturing Productivity Slowdown: Evidence from Firms and Industries." *International Productivity Monitor*, No. 50 (Spring 2026): 79–98. https://csls.ca/ipm-archive/international-productivity-monitor-50th-issue/the-anatomy-of-the-us-manufacturing-productivity-slowdown-evidence-from-firms-and-industries/ Table of contents [Abstract](#abstract) [1. Introduction](#1-introduction) [2. Data and Measurement](#2-data-and-measurement) [2.1 Data Sources](#2-1-data-sources) [2.2 Time Periods, Industries, and Firm Groups](#2-2-time-periods-industries-and-firm-groups) [2.3 Productivity and R&D](#2-3-productivity-and-randd) [3. Productivity Trends across Firms and Industries](#3-productivity-trends-across-firms-and-industries) [3.1 Aggregate Slowdown](#3-1-aggregate-slowdown) [3.2 Within-Group Trends](#3-2-within-group-trends) [3.3 Decomposing the Aggregate Slowdown](#3-3-decomposing-the-aggregate-slowdown) [3.4 Comparing to Andrews et al.](#3-4-comparing-to-andrews-et-al) [4. R&D and Research Productivity](#4-randd-and-research-productivity) [4.1 Trends in R&D Intensity](#4-1-trends-in-randd-intensity) [4.2 Estimating the R&D–Productivity Connection](#4-2-estimating-the-randd-productivity-connection) [4.3 Estimation Results](#4-3-estimation-results) [5. Conclusion](#5-conclusion) [References](#references) [Footnotes](#footnotes) ## Other articles in this issue: [The Potential for Sustained Productivity Impetus from GenAI](https://csls.ca/ipm-archive/international-productivity-monitor-50th-issue/the-potential-for-sustained-productivity-impetus-from-genai/) [The Productivity J-Curve from an International Perspective: Is the United States Unique?](https://csls.ca/ipm-archive/international-productivity-monitor-50th-issue/the-productivity-j-curve-from-an-international-perspective-is-the-united-states-unique/) [Does the Import Invasion Explain the Disappearance of Productivity Growth in U.S. Manufacturing?](https://csls.ca/ipm-archive/international-productivity-monitor-50th-issue/does-the-import-invasion-explain-the-disappearance-of-productivity-growth-in-us-manufacturing/) [Productivity Growth in the U.S. Medical Care Sector: An Analysis Using the BEA’s Health Care Satellite Account](https://csls.ca/ipm-archive/international-productivity-monitor-50th-issue/productivity-growth-in-the-u-s-medical-care-sector-an-analysis-using-the-bea-health-care-satellite-account/) [Addressing Canada’s High Cost of Living: The Role of Productivity and Bargaining Power](https://csls.ca/ipm-archive/international-productivity-monitor-50th-issue/addressing-canadas-high-cost-of-living-the-role-of-productivity-and-bargaining-power/) Copy link Email LinkedIn ## Abstract U.S. manufacturing labour productivity growth fell from roughly 3.5 per cent per year over 1987–2007 to near zero over 2010–2022. Growth in total factor productivity (TFP) also fell to near zero over the same period. This article examines the sources of that slowdown by decomposing manufacturing productivity growth into contributions from leader and follower firms within frontier and laggard industries. We find that the slowdown is broad-based: both for labour productivity and TFP, productivity growth declined among both leaders and followers, across alternative weighting methods, and multiple industry groupings. Standard models of economic growth treat research and development (R&D) as the primary channel through which firms generate productivity growth. The broad-based nature of the slowdown raises the question of whether the translation of R&D expenditures into productivity gains has weakened. We estimate an R&D production function at both the industry and firm levels and find that the elasticity of productivity with respect to R&D is consistently larger in the earlier period than in the full sample, even as R&D expenditure has risen across firms and industries. These results suggest that the slowdown reflects declining research productivity rather than reduced innovation effort. ## 1. Introduction Historically, the U.S. manufacturing sector has been a pivotal driver of aggregate productivity growth, with labour productivity, defined as output per hour, growing at over 2 per cent per year through most of the twentieth century. More recently, however, manufacturing has experienced a pronounced productivity slowdown. Labour productivity fell from roughly 3.5 per cent growth per year over 1987–2007 to near zero over 2010–2022. Growth in total factor productivity (TFP, which is the residual output after accounting for labour, capital, and intermediate inputs), fell from roughly 1.3 per cent to near zero over the same periods. This slowdown is part of a broader deceleration in aggregate labour productivity across the U.S. economy (Fernald, 2015; Syverson, 2017; Byrne *et al.*, 2016; Sharpe and Chittoor, 2025; Atalay *et al.*, 2025) and is particularly puzzling given that the manufacturing sector still accounts for the majority of private-sector research and development (R&D) expenditures (Lashkari and Pearce, 2024). The existing literature points to several possible sources. Industry-level analyses emphasize the role of leading sectors, in particular Computer and electronic products (North American Industry Classification System, or NAICS, 334), whose rapid productivity growth in the late 1990s and early 2000s has collapsed after the Great Recession (Atalay *et al.*, 2025; Syverson, 2017). Firm-level accounts, including those that emphasize divergence between frontier firms and others, suggest that the slowdown should be concentrated in follower firms that are falling behind. In many of those theories, the frontier firms continue to advance (Andrews *et al.*, 2015; Aghion *et al.*, 2023; Andrews *et al.*, 2019; Akcigit and Ates, 2023; Olmstead-Rumsey, 2022). This “best-versus-rest” narrative has become an influential framing of the productivity slowdown, linking it to rising dispersion between frontier and laggard firms. In contrast, the competing “ideas getting harder to find” hypothesis implies a broad-based slowdown affecting all firms and industries, perhaps roughly in proportion to their research intensity (Bloom *et al.*, 2020). Two natural questions emerge from this literature. First, is the slowdown broad-based across categories of firms and industries, or is it concentrated in particular groups? Second, how is the slowdown reflected in R&D activity: is R&D spending itself declining, or is R&D becoming less effective at generating productivity growth? This article addresses both questions by extending our previous work along several dimensions. In Lashkari and Pearce (2026), we introduce a decomposition framework that links firm-level information from Compustat to industry-level aggregates from the U.S. Bureau of Labor Statistics (BLS) and study the trends in R&D intensity and productivity. The present article extends that analysis in three ways. First, we broaden the industry split beyond NAICS 334 to a wider set of frontier industries. Second, we broaden the definitions of leader firms to include both revenue and productivity definitions of leadership. Third, we move beyond documenting the R&D–productivity disconnect and directly estimate the changes in the relationship between R&D spending and subsequent productivity growth at both the industry and firm levels. Our first finding is that the slowdown is broad-based across frontier and lagging industries and across firms. When leader firms within each industry are defined by employment or revenue, which are both relatively persistent measures of firm size, both leaders and followers slow down. This result holds across different weighting methods, both for labour productivity and TFP, and across multiple industry splits. The main exception emerges when classifying leader firms by productivity, a more volatile measure. Under this alternative definition, the most productive firms do not slow down, but these firms are not particularly large in terms of employment and revenue, which limits their contribution to the aggregate slowdown (Section 3.4). Turning to R&D patterns, we provide evidence that the productivity payoff to R&D spending has weakened over time. We first show that R&D intensity broadly rose across firms and industries, even as productivity growth declined. Because R&D is the primary channel through which firms invest in future productivity, this disconnect motivates a direct test of whether the translation of R&D expenditures into productivity gains has weakened. Estimating an R&D production function at both the industry and firm levels, we find that the elasticity of productivity with respect to R&D is consistently larger in the pre-period (1987–2006) than in the full sample (1987–2022), a pattern that holds across both productivity measures (TFP, labour productivity), different specifications, and multiple weighting schemes. These patterns are consistent with the “ideas getting harder to find” hypothesis and suggest that the slowdown may stem from shifts in the nature of the R&D process rather than changes in innovation effort. ## 2. Data and Measurement This article begins by documenting aggregate trends and then decomposes them using more detailed firm- and industry-level data. We decompose aggregate manufacturing productivity into contributions across two groups of firms within each industry, which we label leader and follower firms, and across two groups of industries, which we refer to as frontier and laggard industries. We consider three definitions for leader firms: employment (the baseline, capturing a fairly persistent proxy for size of inputs), revenue (highly correlated with employment but proxying size of output), and productivity (similar to the definition considered by Andrews *et al.* (2019) in the ‘best-versus-rest’ narrative). We use the average of the previous and current period (period-average) weights as the baseline and find similar results using either current-period and previous-period weights. We study the top-four frontier industries defined by labour productivity growth in the pre-period, 1987–2007, which are Textile mills (NAICS 313), Computer and electronic product manufacturing (NAICS 334), Electrical equipment, appliance and component manufacturing (NAICS 335) and Transportation equipment manufacturing (NAICS 336). In the Appendix, we provide additional details for NAICS 334 as the sole frontier industry. We evaluate both labour productivity and TFP throughout the analysis. The reason for these different splits is to understand how pervasive the slowdown is given different productivity definitions and different cuts of the data. ### 2.1 Data Sources Our analysis bridges industry-level and firm-level data for U.S. manufacturing (NAICS 31–33) over the 1987–2022 period. At the industry level, we draw on the BLS detailed industry productivity accounts (Bureau of Labor Statistics, 2025a,b) at NAICS 4-digit level, which provide labour productivity and TFP indices (both 2017=100), hours worked, sectoral output, and output price deflators. The BLS indices incorporate detailed-industry deflators, including hedonic adjustments for NAICS 334. At the firm level, we use Compustat North America (Standard & Poor’s, 2025), drawing on revenue, employees, PP&E (Property, Plant, & Equipment), R&D (Research & Development), and cost of goods sold, restricting the sample to firms with non-missing employment and revenue and dropping bare 2-digit NAICS codes. For PP&E, we use historical prices consistent with financial statements and the existing R&D data come from U.S. Bureau of Economic Analysis (BEA) fixed-asset investment by industry (U.S. Bureau of Economic Analysis, 2025) and from Compustat at the firm level. We deflate Compustat revenues by BLS NAICS 3-digit gross-output deflators, capital by major industry capital deflators, and R&D by college-educated male wages (Bloom *et al.*, 2020; U.S. Census Bureau, 2025). Details are in Appendix A. To bridge the two data sources, we build explicit aggregations from the firm to the industry level and from the industry level to the level of the entire manufacturing sector. The firm-to-industry aggregation uses employment-share weights for labour productivity and revenue-share weights for TFP; industry-to-aggregate aggregation uses BLS hours shares for labour productivity and nominal output shares for TFP (see Appendix B.1 for details on the log-additive aggregation procedure). We construct aggregate manufacturing TFP as the weighted average of NAICS-4 digit industry TFP growth rates, with each industry weighted by its period-average share of total nominal manufacturing gross output. This procedure more closely aligns with how BLS calculates (sectoral output) TFP, and closely tracks the aggregate in Figure 1.1 As we show below, the BLS and Compustat measures are closely aligned, which enables more granular analysis: the firm-level data can be used to decompose the industry-level trends without introducing measurement discrepancies. ### 2.2 Time Periods, Industries, and Firm Groups This analysis primarily focuses on two time periods: the high-growth period (the pre-period) in manufacturing (1987–2007) and the low-growth period (2010–2022). We exclude the Great Recession years 2008–2009. We thus reference 1987 as the initial date for the analysis. We then partition the data along two dimensions: an industry dimension that separates high-growth frontier industries from the rest of manufacturing, and a firm dimension that separates leaders from followers within each industry. The firm and industry partitions serve as the main cuts of our analysis. The baseline industry split defines the top four NAICS 3-digit sectors by pre-period labour productivity growth as frontier industries, with the rest of manufacturing as laggard industries; see Section 3.2. As a robustness check, we also consider a narrower split that isolates Computer and electronic product manufacturing, which is by far the fastest-growing industry over 1987–2007 (Atalay *et al.*, 2025), as the sole frontier industry (Appendix C.2). As for the two firm types, within each 4-digit NAICS code-year cell, we classify the top 10 per cent of firms as leaders (or the single largest firm if fewer than 10 are present); the remainder are followers. We consider three ranking variables. Employment (the baseline) captures scale and is persistent over time; revenue captures market position and is highly correlated with employment. Both are more persistent than productivity rankings (Pearce and Wu, 2025). Productivity (labour productivity or TFP) captures the technological frontier but is much less persistent: smaller firms are often the most productive. For defining leaders, we base our measure on an average of the current and previous period. ### 2.3 Productivity and R&D We measure labour productivity at the firm level as deflated revenue per employee. At the industry level, we use the BLS labour productivity index (2017=100) and the BLS Total Factor Productivity index. These measures are relevant for our industry analysis. For firm-level TFP, we follow the same approach used by De Ridder *et al.* (2026), who estimate a gross-output production function with two inputs: a variable input (cost of goods sold, absorbing both materials and labour payments) and a fixed input (net PP&E). Specifically, we define: ln TFPft = ln Yft – β^ X,j ln Xft – β^ K,j ln Kft (1) where Yft is deflated revenue, Xft is deflated cost of goods sold, Kft is deflated net PP&E, and the elasticities (β^X,j,β^K,j) are estimated at the NAICS 3-digit industry codes following the De Ridder *et al.* (2026) procedure. As for the industry-level productivity measures, we directly use the BLS TFP Index rather than aggregating firm-level estimates, so that industry-level productivity in the R&D regressions (Section 4) is measured independently of the Compustat data. We measure R&D intensity as real R&D per worker, or R&D relative to revenue. These measures are discussed in greater detail in Section 4. For R&D expenditures, the BEA only provides details at the NAICS 3-digit level as far as we are aware. In order to get more granular industry analysis, we extend to NAICS 4-digit level using Compustat data. ## 3. Productivity Trends across Firms and Industries This section documents the aggregate manufacturing productivity slowdown and decomposes its drivers at the industry and firm levels. Using the four-group framework introduced in Section 2, we examine whether the slowdown is concentrated in particular groups or spread across the distribution. The central finding is that the slowdown is pervasive: under size-based rankings (employment or revenue), both leaders and followers slow down. Section 3.4 examines the alternative productivity-based ranking, which yields a different pattern. ### 3.1 Aggregate Slowdown We begin by documenting the aggregate slowdown. Figure 1 plots labour productivity and TFP indices for both BLS and Compustat, each initialized to 100 in 1987. For our BLS series (Figures 1A and 1C), we plot our industry-level aggregation, with current shares and fixed shares as well as the two BLS aggregates for labour productivity and TFP. We find that our aggregation shows a very similar trend to the BLS aggregate manufacturing method. ![Figure 1](https://csls.ca/wp-content/uploads/2026/06/cslc/ipm/issue-50/pearce/IPM-Pearce-Figure-1-T.png) The picture is striking: labour productivity grew at an annualized rate of roughly 3.5 per cent per year over the 1987–2007 period, approximately doubling in level by 2007, and then stagnated, with annualized growth falling to near zero over the 2010–2022 period. TFP grew at roughly 1.3 per cent per year through 2007 and then flattened. The slowdown is visible in every panel of Figure 1: all productivity indices rise steeply through the pre-period, and the change in slope around 2007 marks the onset of the slowdown across both productivity measures and both data sources. The two data sources are closely aligned under both measures, confirming that the Compustat sample of publicly traded firms captures the same broad trends as the BLS universe. Table 1 quantifies the slowdown. Annualized labour productivity growth fell by roughly 4 percentage points from the pre-period (1987–2007) to the post-period (2010–2022), whether measured from BLS or Compustat. TFP slows by 1.1–1.4 percentage points, consistent with the labour productivity pattern but more muted. Both BLS and Compustat TFP are near zero or slightly negative in the post-period. Which firms and industries account for the flattening? ![Figure 2](https://csls.ca/wp-content/uploads/2026/06/cslc/ipm/issue-50/pearce/IPM-Pearce-Figure-2-T.png) ![Figure 3](https://csls.ca/wp-content/uploads/2026/06/cslc/ipm/issue-50/pearce/IPM-Pearce-Figure-3-T.png) ### 3.2 Within-Group Trends We start by focusing on industry- and firm-level splits in accumulated growth from 1987–2022. This section focuses on overall trends and average growth within each group before turning to each group’s contribution to aggregate productivity growth, which depends on its size. Figures 2 and 3 plot each group’s own labour productivity and TFP trajectories under period average weights for employment- and revenue-ranked leaders, using the top four frontier industries versus the rest. The common trend is immediately apparent: all four groups (leader and follower firms in both the frontier and laggard industry) rise together through the pre-period and flatten together after 2007. No single group pulls away or collapses on its own. The narrower 334-only industry split yields qualitatively similar results (Appendix C.2). Section 3.4 examines what happens when leaders are instead defined by productivity. The qualitative results are the same under different weightings and leading firm definitions (see Appendix C.1 and Figure 12 for additional robustness checks). Table 2 reports within-group growth rates for the top four versus rest industry split. Under employment rankings (Table 2), all four groups slow down. The frontier industries (both leaders and followers) grew rapidly in the pre-period and then sharply decelerated. Leaders and followers in laggard industries also slowed. For TFP, the pattern is qualitatively similar: frontier firms slowed substantially while laggard-industry firms slowed more modestly. The slowdown is present in both leaders and followers in laggard industries. ### 3.3 Decomposing the Aggregate Slowdown The trajectory figures show that all groups slow down in productivity, but these groups differ in their contribution to aggregate growth in manufacturing. To assess how much each group contributes to the aggregate slowdown, we next turn to decompositions of aggregate productivity to each of these groups. Figure 4 presents the central result for the top four industries versus rest split. Each panel shows the decomposition of aggregate labour productivity. The visual message mirrors the trajectories: all four bands rise together and flatten together. Both leaders and followers contribute to the rise and to the subsequent stagnation. Figure 5 repeats the exercise for TFP; the same pattern holds throughout. The narrower 334-only split yields qualitatively similar results (Appendix C.2). These patterns persist across different weighting methods (see Appendix C.1, Figures 9–10 for current-period, previous-period, and period-average weights side by side). ![Figure 4](https://csls.ca/wp-content/uploads/2026/06/cslc/ipm/issue-50/pearce/IPM-Pearce-Figure-4-T.png) ![Figure 5](https://csls.ca/wp-content/uploads/2026/06/cslc/ipm/issue-50/pearce/IPM-Pearce-Figure-5-T.png) We decompose aggregate productivity growth using the log-additive approximation, which enables us to aggregate the separate contribution of each group ignoring the reallocation term. The method is consistent with the fact that aggregate log productivity growth is close to a weighted sum of group-level log productivity growth, ΔlogLPt≈∑gωgt,ΔlogLPgt, where ωgt is group g‘s share of aggregate hours. Each group’s contribution to aggregate growth is then its hours share times its own productivity growth. Slowdown shares, defined as each group’s share of the preto-post change in aggregate growth, sum to one by construction. This is only an approximation, but the small differences between initial-share productivity growth and evolving shares in Table 1 indicate it is unlikely to generate strong divergence. As a robustness check, we also report a level-additive (exact) decomposition in Appendix B.1. The qualitative conclusions are the same under both methods. These results are robust to the choice of weighting scheme.2 The qualitative results are the same under all three schemes (see Appendix C.1, Figures 9–10). To assess the extensive margin directly, Appendix C.1 reports a Foster *et al.* (2001) decomposition that explicitly separates the contributions of entering, exiting, and continuing firms. All groups contribute to the slowdown. Leader firms in the frontier industries account for about a quarter of the labour productivity slowdown and nearly half the TFP slowdown, but leaders and followers in laggard industries together account for roughly 70 per cent of the labour productivity slowdown (see Table 10 in Appendix C.2 for the full breakdown). We next focus on a split that has been discussed in the literature, that of looking at the most productive firms rather than the largest in a given industry.