BLS Multifactor Productivity: The Federal Dataset Behind Long-Run Economic Growth Accounting

The Bureau of Labor Statistics publishes two productivity programs. Labor productivity — output per hour of work — is released quarterly and widely quoted. Multifactor productivity — output per combined unit of labor and capital inputs — is released annually and is far less discussed, yet it is the measure that economists actually care about when they want to know whether an economy is becoming genuinely more efficient. MFP is the residual: the growth in output that cannot be explained by growth in measurable inputs. It captures technological progress, organizational efficiency gains, and, inevitably, some measurement error. It is the number at the center of every serious debate about AI, long-run wages, and the sustainability of economic growth.

Two Programs, Two Questions

The BLS productivity programs address two distinct but related questions about how the economy converts inputs into outputs.

The Labor Productivity and Costs (LPC) program publishes quarterly estimates of output per hour of labor for the nonfarm business sector, the business sector, and manufacturing. Because it requires only output data from the Bureau of Economic Analysis and hours data from BLS payroll surveys, it can be released with a roughly five-week lag after each quarter ends. The headline series — nonfarm business output per hour, FRED series OPHNFB — is what financial media refers to when reporting on “productivity.”

The Multifactor Productivity (MFP) program, also known internationally as Total Factor Productivity (TFP), publishes annual estimates that combine output data with measures of both labor input and capital input. Because capital measurement requires detailed investment data from the BEA National Income and Product Accounts, BEA Fixed Asset Accounts, and BLS industry surveys, the annual MFP release typically arrives in the spring following the reference year. The private business sector MFP series extends back to 1947, giving economists nearly eight decades of data with which to study technological change.

The conceptual distinction matters enormously. Labor productivity can rise even in the complete absence of technological progress if workers are given more machines to work with — a process economists call capital deepening. A worker operating a larger excavator moves more earth per hour without becoming more skilled. MFP holds constant the combined input of labor and capital; growth in MFP therefore requires something beyond simply investing more, whether that something is a better organization of production, a new technology, a more educated workforce, or a more efficient allocation of resources across firms.

The Solow Residual and Growth Accounting

The theoretical foundation for MFP measurement is Robert Solow's 1957 paper “Technical Change and the Aggregate Production Function,” which introduced what became known as growth accounting. Solow decomposed observed GDP growth into three additive components: the contribution of capital input growth (weighted by capital's share of income), the contribution of labor input growth (weighted by labor's share of income), and an unexplained residual. The residual — growth not attributable to measurable factor inputs — is the Solow residual, and it is what BLS now measures as MFP.

The arithmetic is straightforward. If output grew at 3% per year, capital contributed 1.2 percentage points (capital's income share of roughly 0.35 times capital input growth of roughly 3.5%), and labor contributed 1.0 percentage point (labor's income share of roughly 0.65 times labor input growth of roughly 1.5%), then MFP growth is 3% − 1.2% − 1.0% = 0.8%. Everything that cannot be assigned to measurable input growth is attributed to MFP. This residual-by-construction nature is both MFP's analytical power and its primary limitation: it absorbs not only genuine technological progress but also measurement error in output and inputs.

The historical MFP record for the US private business sector divides into recognizable episodes. The postwar golden age from 1948 to 1973 saw MFP growth averaging roughly 1.5% per year — a period of rapid diffusion of electrification, the interstate highway system, containerized shipping, and the first generation of mainframe computing. The productivity slowdown from 1973 to 1995 cut MFP growth to roughly 0.3% per year, a collapse that generated an enormous academic literature and no consensus explanation (candidates include the energy price shocks, the shift toward services, a deceleration in applied R&D, and regulatory compliance costs). The information technology revolution produced a productivity revival from 1995 to 2004, with MFP averaging roughly 1.0% per year, concentrated in IT-producing industries and IT-intensive retail and wholesale trade. Since 2004, MFP growth has again slowed, running closer to 0.3–0.5% per year through 2019. The 2020s have reopened the debate.

BLS Measurement Methodology

The BLS MFP methodology involves assembling and combining data from multiple federal statistical systems in ways that require careful attention to definitional consistency.

Output for the private business sector is measured as real gross domestic product minus the output of general government, nonprofit institutions, and the rental value of owner-occupied housing — the three sectors excluded from the private business definition. For industry-level MFP, BLS uses industry gross output (including intermediate inputs consumed) rather than value added, because the industry production function governs how each industry transforms all its inputs, not just primary factors. Output data derives primarily from the BEA National Income and Product Accounts and the BEA Industry Accounts.

Capital input is not simply the stock of physical capital. BLS constructs a capital services index using the Hall-Jorgenson user cost approach: each type of asset is weighted not by its purchase price but by the rental price, or user cost, of a unit of capital services for one period. The user cost reflects the asset's opportunity cost (the rate of return), depreciation rate, and expected capital gain or loss from price changes. A piece of equipment that depreciates rapidly (computers) contributes more capital services per dollar of value than a long-lived structure (a warehouse). BLS aggregates capital services across five major asset types: equipment, structures, intellectual property products (including software and R&D), inventories, and land. The data originates from BEA Fixed Asset Accounts.

Labor input is more than a count of hours. BLS adjusts hours worked for labor composition changes — shifts over time in the education, experience, and gender mix of the workforce. An hour of work supplied by a college-educated, experienced worker embodies more human capital than an hour supplied by a newly hired high school graduate. As the educational attainment of the workforce has risen over decades, a fixed count of hours represents growing effective labor input. This quality adjustment means that raw hours growth understates true labor input growth, and MFP growth therefore understates the contribution of education to output. BLS constructs the labor composition adjustment from Current Population Survey microdata matched to BLS employment and hours series.

The income shares used to weight input contributions — capital's share and labor's share — come from the BEA National Income Accounts. BLS uses a Tornqvist index, which averages adjacent-year shares, smoothing the weights as the functional distribution of income shifts over time. The full MFP release methodology is documented in the BLS Handbook of Methods, Chapter 11.

Private Business Sector and Manufacturing MFP

The flagship BLS MFP series covers the private business sector, which excludes federal, state, and local government; nonprofit institutions serving households; and the imputed rental value of owner-occupied housing. The exclusions reflect a measurement rationale: government output is measured by input cost (because market prices do not exist), which would make government MFP mechanically zero by construction. Including it would dilute the MFP signal from the market economy.

BLS separately publishes MFP for total manufacturing, durable goods manufacturing, and nondurable goods manufacturing back to 1949. Manufacturing MFP has generally outpaced the broader private business sector over the full sample. Within manufacturing, the computer and electronic products sector has recorded extraordinary MFP growth for decades, driven by semiconductor technology improvement. The quality-adjusted price index for semiconductors and related products has fallen at double-digit annual rates since the 1960s; when BLS accounts for quality improvements by measuring output in constant-quality units, output-per-input in this sector surges even as nominal revenues and employment are measured conventionally. This is the statistical mechanism underlying the IT investment story: computing power got exponentially cheaper relative to its performance, so real output in computing grew faster than any input measure.

The 1995–2004 productivity revival is closely associated with IT investment and production. MFP growth in IT-producing industries (semiconductors, computers, telecom equipment, software) ran at several percentage points per year. But the revival also spread to IT-using industries — most notably wholesale and retail trade, where logistics technologies pioneered by Walmart, then Amazon, transformed inventory management and distribution. Economists Erik Brynjolfsson and Lorin Hitt documented the complementary role of organizational change: firms that combined IT investment with reorganized work practices captured far larger productivity gains than those that simply purchased equipment.

Industry-Level MFP

BLS publishes industry-level MFP for approximately 60 detailed NAICS industries, with most series beginning in 1987. Industry MFP data allow researchers to identify where in the economy productivity growth originates, rather than treating the private business sector as a black box.

Capital services at the industry level are broken into the same five asset categories (equipment, structures, intellectual property, inventories, land) with user cost weights computed separately for each industry, because different industries face different depreciation patterns and asset compositions. Software-intensive industries like finance and insurance carry very different capital service profiles than capital-intensive utilities.

Industry MFP data reveal several important empirical patterns. Retail trade productivity grew unusually rapidly from 1987 to 2000, driven by logistics IT, large-format stores, and supply chain integration. Healthcare is the canonical puzzle in the opposite direction: inputs have grown rapidly (more physicians, more equipment, more pharmaceuticals per patient), but measured output — which BLS approximates using hospital admissions, physician visits, and similar counts, adjusted for case mix — has not kept pace. Whether that reflects genuine productivity stagnation, unmeasured quality improvements (patients live longer, recover faster), or output measurement failure is still actively debated. The BEA has experimented with health outcome-adjusted output measures; the gap between those and the conventional measures is large.

Labor Productivity vs. MFP: Why the Distinction Matters for Wages

Labor productivity — output per hour worked — is the right concept for thinking about whether wage growth is sustainable in the short to medium run. When output per worker rises faster than wages, unit labor costs fall and corporate margins expand. When wages rise faster than output per worker, unit labor costs surge and inflationary pressure builds. This is the Fed's core services inflation framework.

But for thinking about the long-run trend in real wages, MFP is the relevant concept. Real wages in a competitive economy will, over long periods, track the marginal product of labor. Capital deepening raises labor's marginal product and therefore real wages, but it does so at a diminishing rate: each additional machine added to a fixed labor force contributes less than the previous machine. Sustained real wage growth requires ongoing MFP growth — genuine technological progress that continuously pushes out the production possibilities frontier.

This distinction has political economy implications. The growth in real wages from 1948 to 1973 was substantially supported by both capital deepening and strong MFP growth. The deceleration in real wage growth after 1973 is closely correlated with the MFP slowdown. When productivity commissions and economists argue that the path to higher living standards runs through productivity policy — R&D investment, human capital, regulatory modernization, technology diffusion — they are arguing about MFP, not labor productivity. Capital deepening alone cannot indefinitely substitute for technological progress.

Unit Labor Costs: The Inflation Signal the Fed Watches

The BLS LPC program publishes Unit Labor Costs (ULC) alongside the labor productivity series. ULC is defined as compensation per hour divided by output per hour, which simplifies to labor compensation per unit of output. Because labor is by far the largest input cost in the services sector, ULC growth is the primary driver of core services inflation ex-shelter — the component of CPI that most directly reflects domestic cost pressures and is therefore the most persistent.

The arithmetic of ULC makes the relationship intuitive. ULC growth equals wage growth minus productivity growth. If hourly compensation grows at 5% and output per hour grows at 2%, ULC grows at 3% — generating persistent inflationary pressure unless profit margins absorb the difference. If productivity accelerates to match wage growth, ULC is stable even without wage deceleration.

The 2021–2022 period illustrated this dynamic sharply. A tight labor market following the pandemic produced rapid wage growth; simultaneously, output per hour fell in 2022 as firms hired workers faster than output recovered. The combination — rising compensation, falling productivity — drove nonfarm business ULC up roughly 6% in 2022. This surge in unit labor costs fed directly into services inflation and validated the Federal Reserve's concern that wage growth was inconsistent with the 2% PCE inflation target. In 2023, the dynamics reversed: productivity rebounded sharply (particularly in Q3 2023, the strongest quarterly productivity reading in years), ULC growth moderated, and core services inflation began to decelerate. The Fed cited this productivity recovery explicitly in its communications about the path of rate cuts.

FRED series for ULC tracking: ULCNFB (nonfarm business unit labor costs, quarterly, index); RCPHBS (real compensation per hour, business sector). The BLS series IDs PRS85006092 (nonfarm business output per hour) and PRS85006112 (nonfarm business unit labor costs) are the primary sources.

The Productivity Paradox and the AI Hypothesis

Robert Solow's 1987 quip — “You can see the computer age everywhere except in the productivity statistics” — named the productivity paradox: despite massive IT investment through the 1970s and 1980s, aggregate MFP growth remained depressed. The paradox was eventually resolved, at least partially, by the 1995–2004 productivity revival, which showed that the productivity benefits of a general-purpose technology can lag the initial diffusion by decades while complementary reorganization of work, business processes, and infrastructure catches up.

The historical evidence on general-purpose technology lags is striking. The steam engine was commercially viable by the 1780s but did not transform factory productivity until the 1850s after the necessary infrastructure — railroads, machine tool industries, skilled mechanics — had been built. Electricity was commercially available by the 1880s, but manufacturing productivity did not clearly accelerate until the 1915–1930 period, after factories had been physically reorganized around electric motor drives rather than steam-era shaft-and-belt layouts. The economist Paul David, who documented the electricity lag, explicitly predicted in 1990 that computers would show the same pattern — and the 1995 productivity revival vindicated his prediction roughly on schedule.

This historical pattern frames the current debate about artificial intelligence and productivity. If large language models and AI-assisted software tools are a genuine general-purpose technology — applicable across sectors, capable of spawning complementary innovations — then the productivity paradigm predicts a lag. Firms must reorganize workflows, retrain workers, redesign products, and develop complementary data infrastructure before the efficiency gains show up in aggregate output statistics. Erik Brynjolfsson and colleagues have argued that AI productivity gains may already be occurring but are concentrated in tasks and outputs that national accounts measure poorly: intangible capital formation, quality-adjusted services, faster problem-solving. Whether AI MFP effects will appear in BLS data by 2026–2028 is the central empirical question in macroeconomics.

The 2023 productivity surge — nonfarm business output per hour rose at roughly a 4% annual rate in the third quarter — was large enough to generate significant commentary. Whether it represents the beginning of an AI-driven acceleration or a cyclical rebound from the 2022 dip remains contested. Annual MFP data for 2023 and 2024, when released by BLS, will provide the fuller picture that includes capital input growth.

International Comparisons and the Great Productivity Divergence

The OECD publishes comparable MFP estimates for member countries, and the Conference Board Total Economy Database covers a broader set of economies including major emerging markets. Cross-country productivity comparisons require care because output must be expressed in common price terms (purchasing power parity), capital measurement methodologies differ across national statistical offices, and industry composition varies.

With those caveats, the international record suggests that US MFP growth has generally outpaced the EU average over the past three decades, and the gap widened notably after 2004. The explanation most prominent in the academic literature focuses on the IT services sector: US firms adopted and productively deployed IT-based services (e-commerce, digital finance, platform businesses) at a pace that European regulatory environments, labor market structures, and incumbent firm strategies did not replicate. The large US technology platform companies — essentially nonexistent in 2000 — generate enormous output per employee and are concentrated almost entirely in US GDP accounts.

Japan's “lost decade” and subsequent stagnation is visible in its MFP record: the OECD data show Japanese business sector MFP growing at essentially zero from the mid-1990s through the 2010s. The zombie firm problem — banks maintaining credit to insolvent firms to avoid recognizing losses — is identified as a major culprit, because it prevented the reallocation of capital and labor from low-productivity to high-productivity firms that is a primary mechanism of aggregate MFP improvement in market economies. FRED provides OECD-sourced MFP index data including BSXRLTT01USQ661S for cross-country comparison.

Accessing BLS Productivity Data

BLS productivity data is available through several access channels depending on analytical requirements.

The BLS Productivity Research Program publishes annual MFP data as Excel downloads at the productivity pages of bls.gov. The files include the MFP index, output index, labor input index, capital input index, and the subcomponents of capital services for the private business sector and manufacturing. These Excel files are the primary source for detailed MFP decomposition work.

The BLS Public Data API at api.bls.gov/publicAPI/v2/timeseries/data/covers quarterly LPC series including output per hour, compensation per hour, and unit labor costs. Key series IDs include PRS85006092 (nonfarm business output per hour), PRS85006112 (nonfarm business unit labor costs), and PRS85006102 (nonfarm business compensation per hour). The API returns up to 10 years of history without a key and 20 years with a free registered key.

FRED (fred.stlouisfed.org) mirrors the major LPC quarterly series with more memorable IDs: OPHNFB (nonfarm business output per hour, seasonally adjusted), ULCNFB (nonfarm business unit labor costs), RCPHBS (real compensation per hour, business sector). OECD-sourced MFP data for cross-country comparison is also available in FRED.

For bulk access, the BLS multifactor productivity program page links directly to the Excel workbooks covering the full historical series, which are more practical than the API for MFP specifically because annual MFP is not represented in the standard BLS series-ID format used by the API.

Python: Quarterly Productivity and Unit Labor Costs

The following script uses the BLS Public Data API to pull 10 years of quarterly data for nonfarm business labor productivity and unit labor costs, computes a four-quarter rolling average for each series, and plots them on dual axes. The dual-axis presentation allows the viewer to see the inverse relationship between productivity and unit labor costs — particularly the 2022 productivity slump coinciding with a ULC surge, and the 2023 productivity recovery coinciding with ULC moderation.

import requests
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib.ticker as mtick
from datetime import datetime

API_URL = "https://api.bls.gov/publicAPI/v2/timeseries/data/"

# Series IDs
# PRS85006092 - Nonfarm business: output per hour (labor productivity), quarterly
# PRS85006112 - Nonfarm business: unit labor costs, quarterly
SERIES = {
    "Labor Productivity (Output/Hour)": "PRS85006092",
    "Unit Labor Costs": "PRS85006112",
}

end_year = datetime.now().year
start_year = end_year - 10

headers = {"Content-Type": "application/json"}
payload = {
    "seriesid": list(SERIES.values()),
    "startyear": str(start_year),
    "endyear": str(end_year),
    "calculations": {"output_type": 1},  # include 12-month pct change
}

resp = requests.post(API_URL, json=payload, headers=headers, timeout=60)
resp.raise_for_status()
data = resp.json()

if data.get("status") != "REQUEST_SUCCEEDED":
    raise RuntimeError("BLS API error: " + str(data.get("message", "")))

id_to_label = {v: k for k, v in SERIES.items()}

frames = []
for series_obj in data["Results"]["series"]:
    sid = series_obj["seriesID"]
    label = id_to_label.get(sid, sid)
    rows = []
    for obs in series_obj["data"]:
        period = obs.get("period", "")
        # Quarterly data uses Q01, Q02, Q03, Q04
        if period.startswith("Q"):
            pct = obs.get("calculations", {}).get("pct_changes", {}).get("1", None)
            if pct is not None:
                quarter_map = {"Q01": "01", "Q02": "04", "Q03": "07", "Q04": "10"}
                month = quarter_map.get(period, "01")
                rows.append({
                    "date": pd.Period(obs["year"] + "-" + month, freq="Q").to_timestamp(),
                    "pct_change_4q": float(pct),
                    "series": label,
                })
    if rows:
        frames.append(pd.DataFrame(rows))

df = pd.concat(frames, ignore_index=True)
df = df.sort_values(["series", "date"])

# Compute 4-quarter rolling average
result_frames = []
for label, grp in df.groupby("series"):
    grp = grp.copy().sort_values("date")
    grp["rolling_4q"] = grp["pct_change_4q"].rolling(window=4, min_periods=2).mean()
    result_frames.append(grp)

df_final = pd.concat(result_frames, ignore_index=True)

# Plot on dual axes
fig, ax1 = plt.subplots(figsize=(13, 6))
ax2 = ax1.twinx()

lp = df_final[df_final["series"] == "Labor Productivity (Output/Hour)"].dropna(subset=["rolling_4q"])
ulc = df_final[df_final["series"] == "Unit Labor Costs"].dropna(subset=["rolling_4q"])

l1, = ax1.plot(lp["date"], lp["rolling_4q"], color="#0b4a8f", linewidth=2.2,
               label="Labor Productivity (4Q rolling avg)")
l2, = ax2.plot(ulc["date"], ulc["rolling_4q"], color="#dc2626", linewidth=2.0,
               linestyle="--", label="Unit Labor Costs (4Q rolling avg)")

ax1.axhline(0, color="#6b7280", linewidth=0.8, linestyle="-")
ax2.axhline(0, color="#6b7280", linewidth=0.8, linestyle="-")

ax1.yaxis.set_major_formatter(mtick.PercentFormatter(decimals=1))
ax2.yaxis.set_major_formatter(mtick.PercentFormatter(decimals=1))

ax1.set_ylabel("Labor Productivity % Change (YoY)", color="#0b4a8f", fontsize=10)
ax2.set_ylabel("Unit Labor Costs % Change (YoY)", color="#dc2626", fontsize=10)

ax1.tick_params(axis="y", labelcolor="#0b4a8f")
ax2.tick_params(axis="y", labelcolor="#dc2626")

lines = [l1, l2]
labels = [l.get_label() for l in lines]
ax1.legend(lines, labels, loc="upper left", fontsize=9)
ax1.grid(axis="y", linestyle=":", alpha=0.4)

ax1.set_title(
    "Nonfarm Business: Labor Productivity vs. Unit Labor Costs (4-Quarter Rolling Avg)",
    fontsize=12, fontweight="bold"
)
fig.tight_layout()
plt.savefig("productivity_vs_ulc.png", dpi=150)
plt.show()
print("Chart saved to productivity_vs_ulc.png")

The BLS quarterly productivity series use period codes Q01–Q04 rather than the monthly M01–M12 codes used by CPI and employment series. The four-quarter rolling average smooths noise from quarter-to-quarter swings in output measurement and reveals the underlying trend. To track annual MFP instead of quarterly labor productivity, download the Excel files from the BLS productivity research program page, which contain the full decomposition into capital, labor, and MFP components.

Unit labor costs rise when wages outpace productivity — making services inflation persistent. The headline inflation measure that combines goods and services is the Consumer Price Index. See BLS CPI: The Consumer Price Index and the Federal Inflation Measurement Behind Every Policy Decision.

MFP measurement draws heavily on BEA output data from the National Income and Product Accounts, which decompose GDP into its expenditure, income, and production components. See BEA GDP and National Accounts: The Federal Dataset That Measures the US Economy.

Industry-level MFP analysis requires county and industry employment and wage data to understand how input composition shifts over time. See BLS QCEW: The County-Level Employment and Wages Dataset Behind Every Local Economic Analysis.