Experimental Total Factor Productivity for Detailed Manufacturing Industries with Labor Composition and Capitalized R&D

On September 29, 2026, the Bureau of Labor Statistics (BLS) released experimental total factor productivity (TFP) and costs measures for 86 detailed manufacturing industries at the 4-digit NAICS level. The data cover 2005 to 2023. These series incorporate three changes beyond the TFP series currently published for these industries. First, the index of labor input is modified by incorporating a labor composition series into the existing measure of hours worked. Second, the index of capital services is modified by including research and development (R&D) as a capitalized asset. Third, industry output is expanded to include own-account R&D and software. BLS is researching the reliability of these estimates to measure TFP more accurately at the detailed industry level. Additional information will be available in a forthcoming article.

Comparison with official BLS TFP measures

Table 1 compares the experimental measures with the official BLS TFP measures.

Introduction to labor composition

Labor is a key input in the production of goods and services. Labor input has two dimensions, the quantity of labor and the quality of labor. The quantity of labor is the number of hours worked. The quality of labor reflects the skill composition of the workforce, or how effective workers are during those hours. For example, an hour worked by a newly hired employee is not likely to produce as much output as an hour worked by a highly experienced one. The effectiveness of the experienced worker’s hours is presumably greater than that of the new employee’s. The skill composition of the workforce is accounted for in labor input through an adjustment referred to as labor composition.

Incorporating labor composition allows different groups of workers to be treated as heterogeneous, with the hours worked by each group weighted by its effectiveness. In practice, workers are differentiated by their age, education level, and sex, and their skill level is estimated by their hourly wage. Accounting for labor composition therefore incorporates changes in the educational attainment and experience of the workforce into productivity estimates.

In the official BLS measures of total factor productivity for detailed industries, labor input is estimated as the sum of hours worked by all workers, which captures only the quantity of labor. Consequently, the hours worked of all workers, regardless of their skill levels, are treated equally. In this experimental data release, BLS incorporates the labor composition adjustment to hours worked in the measure of TFP for detailed manufacturing industries. This method estimates labor composition at a detailed industry level using the U.S. Census Bureau’s American Community Survey (ACS) as well as the BLS Current Population Survey (CPS) and then aggregates up to the higher-level major sectors and industries that are currently published by the BLS productivity program. Details on the calculation of labor input and labor composition are available in the Productivity Handbook of Methods and the 2022 technical note.

The industry detail available in the ACS determines the detail of the labor composition measures. The ACS uses Census industry codes. Some of these industry definitions combine two or more 4-digit NAICS industries (e.g., Census code 2290 maps to NAICS 3251 and 3259) or provide information only at a broader industry level (e.g., Census code 3895 maps to NAICS 337). Shares are used to allocate labor hours and costs to individual 4-digit industries. However, these shares do not provide information on the age, sex, and education of the workers in each industry. As a result, industries covered by the same Census industry code are assumed to have the same labor composition growth.

Effect on labor input in semiconductors and other electronic components

Semiconductor and other electronic component manufacturing (NAICS 3344) provides an example of how the labor composition adjustment affects labor input.

The experimental labor input index, which incorporates the labor composition adjustment, is higher than the official labor input index, which only measures hours worked. From 2005 to 2023, composition-adjusted labor input declined by 0.3 percent annually, compared with a 0.8 percent decline in hours worked. This indicates that there was an overall increase in the skill composition of workers.

Introduction to intangible capital assets

Capital input is measured as the flow of services from the stock of fixed assets used repeatedly or continuously in production for more than a year. In other words, productive services of all types of capital assets flow from the cumulative historical acquisition of these assets.

Capital assets can be either tangible or intangible. Tangible capital assets have physical substance, and the BLS industry productivity program measures four broad categories: equipment, structures, inventories, and land.

Intangible capital assets lack physical substance. They are generally a store of knowledge or information which is expected to generate output into the future. The 2008 System of National Accounts groups several types of intangible capital assets into a category called “intellectual property products” (IPP). These assets include software and databases, R&D, and entertainment, literary, and artistic originals. The official BLS capital measures for detailed manufacturing industries include software as the only intangible capital asset.

Entertainment, literary, and artistic originals are not a significant source of capital services in manufacturing industries. However, R&D is a significant investment in a small but important subset of industries, such as pharmaceutical, semiconductor, and aerospace equipment manufacturing. A proper accounting of this investment presents the opportunity for us to better understand productivity, efficiency, and output growth in these industries.

Incorporation of R&D as a capitalized asset

The BLS productivity program calculates measures of capital services for manufacturing at the 4-digit NAICS level using investment data from the annual economic surveys of the U.S. Census Bureau. Historically, data on capital expenditures from these surveys have been limited to tangible capital categories (structures, equipment, and inventories).

To incorporate capitalized software and R&D, these measures use the Fixed Asset Accounts of the U.S. Bureau of Economic Analysis (BEA). The BEA Fixed Asset Accounts publish investment data at a broad level of manufacturing industry aggregation (generally 3-digit NAICS industries) for intellectual property products from 1901 to the current year. These products include 17 types of R&D.^[1] R&D investment data from the Fixed Asset Accounts are disaggregated from a 3-digit NAICS basis to a 4-digit NAICS basis using shares of own-account R&D.^[2]

Measures of capitalized R&D and labor composition at the detailed industry level are subject to certain limitations. Estimates for detailed industries, especially those smaller industries which employ fewer workers or which conduct less R&D, are subject to more volatility and error than estimates for more aggregate sectors.

Effect on capital input in pharmaceuticals and medicine

Pharmaceutical and medicine manufacturing (NAICS 3254) provides an example of how including R&D affects capital input.

The pharmaceuticals industry invests more money into R&D than any other manufacturing industry. According to the U.S. National Science Foundation (BERD table 25), in 2023 U.S. pharmaceutical manufacturers paid for (and performed themselves) more than $114 billion of R&D domestically. This figure represents one-third of all R&D of this type among the entire manufacturing sector.

When we capitalize pharmaceutical R&D, that vast amount of annual investment accumulates over time into capital stock that serves as an input into production.

The incorporation of R&D has a dramatic effect on the growth rate of the industry’s capital index. Without R&D, capital grew at an annual rate of 1.2 percent from 2005 to 2023. With R&D, capital grew at an annual rate of 4.6 percent.

Modification of industry sectoral output measures

Own-account R&D and software are produced by firms for their own use. Including this production in output makes the output measure consistent with the treatment of these assets as capital. BEA estimates the value of this own-account production using a cost-based approach described in Measuring R&D in the National Economic Accounting System.

BLS uses these BEA estimates to expand its official industry output measures. The ratios of own-account R&D and software production to gross output in the BEA data are applied to BLS estimates of industry gross output. The resulting estimates are adjusted for price changes and combined with the official output measures to calculate the experimental output indexes.

In pharmaceutical and medicine manufacturing, output declined at an annual rate of 1.9 percent from 2005 to 2023 under the official TFP measures. Including own-account production reduced the annual rate of decline to 1.3 percent.

Combined effects on total factor productivity

For 57 of the 86 detailed manufacturing industries, the experimental annual percent change^[3] in TFP from 2005 to 2023 differs from the official rate by 0.2 percentage point or less. Changes in output and combined inputs can offset each other, so a small difference in TFP growth does not necessarily mean that the adjustments were small. Pharmaceuticals and semiconductors are examples of industries with larger differences in TFP growth due to these adjustments.

TFP is calculated as output divided by combined inputs. Growth in TFP is approximately equal to growth in output less growth in combined inputs. The following industry comparisons bring together the labor, capital, and output adjustments described above.

Pharmaceutical and medicine manufacturing

Pharmaceutical and medicine manufacturing shows one of the largest differences between the experimental and official TFP measures.

Output declined under both measures, but the decline was smaller under the experimental measures. Combined inputs grew by 1.2 percent annually under the experimental measures, compared with a decline of 0.8 percent under the official measures.

The cumulative changes made in the experimental series cause measured TFP in pharmaceuticals to fall at a rate of 2.5 percent annually, compared with a decline of 1.1 percent in the official TFP index. This result reflects the large increase in measured capital shown earlier. Because combined inputs grew while output declined, the TFP index shows a steep drop.

The TFP index below shows how the difference in growth rates accumulated over the 2005-2023 period.

Semiconductors and other electronic components

With one of the highest rates of official TFP growth among the detailed manufacturing industries, semiconductor and other electronic component manufacturing is a useful case for examining the experimental measures.

Using the official TFP measures, output grew at an annual rate of 1.8 percent from 2005 to 2023. Over the same period, combined inputs fell by 0.2 percent annually. TFP growth (2.1 percent per year) was therefore the main source of output growth. However, the picture changes when we account for the previously unmeasured factors of labor composition and capitalized R&D. Including these two elements raises the growth of measured combined inputs by 0.7 percentage point per year and reduces TFP growth by about the same amount.

Contributions to output growth

The experimental dataset also includes contributions to output and labor productivity growth. These measures provide additional detail on the sources of growth across the detailed manufacturing industries, including the roles of labor composition, R&D, and TFP.

Table 2 shows how capital, hours worked, labor composition, intermediate inputs, and TFP contributed to the annual percent change in output from 2005 to 2023 in the previously discussed industries.

In semiconductors and other electronic components, the positive contribution of labor composition partly offsets the negative contribution of hours worked. In pharmaceuticals and medicine, output declined despite a positive contribution from capital, almost all of which came from R&D. The negative contributions from intermediate inputs and TFP more than offset the positive contributions from capital input and labor input.

Chart 6 shows how capital input, labor input, intermediate inputs, and TFP contributed to the annual percent change in output from 2005 to 2023 in the 10 detailed manufacturing industries with the fastest output growth. Intermediate inputs made the largest contribution in six of these industries, TFP in three, and capital input in one. Semiconductors and other electronic components, which had the second-fastest output growth, is one of the three industries in which TFP made the largest contribution.

Notes

Last Modified Date: September 29, 2026