GDP, GDI, and GDO: an evaluation of output measures for productivity analysis

In 2015, the U.S. Bureau of Economic Analysis (BEA) began publishing a series that was an average of gross domestic product (GDP) and gross domestic income (GDI). This series, which some researchers refer to as gross domestic output (GDO), had been cited in research during the decade prior to 2015 as having exhibited the beneficial quality of lower revisions than either GDP or GDI. The GDO measure also provided a way to integrate source data from both the product and the income sides of the National Income and Product Accounts (NIPAs) into one measure. Acknowledging the usefulness of the measure, BEA responded by publishing it in its quarterly GDP release along with GDP and GDI.

Since that time, all three output measures have become integral to U.S. macroeconomic analysis, as each exhibited unique and useful attributes relevant to real-time as well as historical analysis. The three measures of output are tracked each quarter by the White House Council of Economic Advisers (CEA), the Board of Governors of the Federal Reserve System (FRB), and the U.S. Treasury Department. Also, the three measures are used in the business cycle dating framework of the National Bureau of Economic Research (NBER) and in the modeled quarterly output measures estimated by the Federal Reserve Banks of Philadelphia and New York.

Given the wide usage of this tripartite approach to output estimation and analysis, it will be worthwhile to evaluate and compare the measures of GDP, GDI, and GDO with respect to productivity analysis. This analysis will be conducted by reviewing the extensive research on these three measures that was done in the 2000s and 2010s and considering subsequent data results during the 2020s.

This article provides technical information on the source data and attributes of GDP, GDI, and GDO, background information on why interest grew in GDI and GDO during the 2000s and 2010s, and data results on how the three measures performed as indicators during the 2020s. This article also provides a method for estimating GDI-based and GDO-based nonfarm business labor productivity, and a data analysis of these estimated series from both cyclical and long-term perspectives.

GDP, GDI, and GDO

Labor productivity is defined as output per labor hour worked. The most widely used labor productivity measure published by the U.S. Bureau of Labor Statistics (BLS) is for the nonfarm business sector, because it offers wide (76 percent) coverage of the U.S. economy while excluding sectors for which output is difficult to measure (such as nonprofits and general government) and the volatile farm sector. BEA publishes output for the nonfarm business sector one month after the end of each quarter, in its initial GDP release, and BLS uses this output estimate directly in its nonfarm business output per labor hour series.

The nonfarm business output measure published by BEA, which is based on GDP, comes from the product side of the NIPAs. By saying that GDP comes from the product side, we are indicating that it reflects the value of goods and services produced (less the value of the goods and services used up in that production).¹ In contrast, GDI is from the income side of the NIPAs and is based on the income earned and costs incurred from that production, such as business profits, labor compensation, capital depreciation, and taxes.²

In theory, GDP should equal GDI, because both aim to measure U.S. gross domestic output. However, these measures usually do not equal each other, and this difference is referred to as the statistical discrepancy. The statistical discrepancy—which is often expressed as a percentage of GDP—has had an average size of 0.8 percent of GDP since 1947 and has exhibited notable variation throughout this series, ranging from close to zero to as large as 2.7 percent of GDP. As of the first quarter of 2025, the measure was $126 billion, or 0.4 percent of GDP. The statistical discrepancy reflects differences in the underlying source data and methodology between GDP and GDI. For example, BEA estimates GDP by using U.S. Census Bureau surveys, such as the Monthly Retail Trade Survey and Quarterly Services Survey, as well as financial reports and industry and trade transactions data. In contrast, BEA estimates GDI by using the BLS Quarterly Census of Employment and Wages, the U.S. Census Bureau’s Quarterly Financial Report and its Quarterly Tax Survey, and state government tax collections.³

Another difference between GDP and GDI is that, although early estimates of both measures utilize a combination of comprehensive data, indicator data, and trend-based data, GDP uses a greater proportion of comprehensive data in its early estimates.⁴ Another attribute of GDP is that it is available earlier than GDI (one release earlier than GDI for the first, second, and third quarters of each year, and two releases earlier for the fourth quarter of each year). GDI is published later than GDP due to a lag in the profits and net interest data that are needed for its calculation.⁵ Both measures eventually converge toward being fully based on comprehensive data, after revisions have been incorporated in releases months and years following the initial release.⁶

The average of GDP and GDI, referred to as GDO by some researchers, is computed by averaging the nominal output of GDP and GDI, and then deflating this with the implicit price deflator for GDP.⁷ Similarly, GDI is computed by deflating nominal GDI using the implicit price deflator for GDP.⁸ Given that all three measures use the same deflator, we can say that the difference between GDP, GDO, and GDI solely reflects the difference between the nominal output for these measures.

The BLS Office of Productivity and Technology has stated that it uses the official (GDP-based) BEA nonfarm business output measure for estimating nonfarm business labor productivity for two main reasons, namely because “the product side output measure is conceptually more closely related to what the economy produces” and because “BEA’s source data on the income side are incomplete at the time the GDP statistics are first issued each quarter.”⁹

However, given the increased interest among data users in GDO and GDI in recent decades, it will be worthwhile to evaluate and compare these measures along with GDP, and to consider the usefulness of each of the three measures with respect to output and labor productivity analysis.

Increased interest in GDI and GDO in the 2000s and 2010s

Fundamentally, increased interest in GDO and GDI in the 2000s and 2010s had to do with performance-based metrics associated with these measures in their roles as economic indicators that emerged.

During these years, numerous articles traced a retrospective trend of GDO having exhibited lower revisions than either GDP or GDI over the prior several decades.¹⁰ For example, chart 1 shows the mean absolute revisions (MARs) from third vintage to latest vintage of GDP, GDO, and GDI from the first quarter of 1993 through the fourth quarter of 2013.¹¹ During these years, the MARs were 1.25 percentage points for GDP, 1.07 percentage points for GDO, and 1.49 percentage points for GDI.¹² This result was found by Dennis J. Fixler, Danit Kanal, and Pao-Lin Tien, and these authors asserted that “it is possible to infer that the weighted average series of the two is a more reliable measure of activity than either GDP or GDI alone because some of the measurement errors are averaged out, reducing subsequent revisions in the weighted average.”

Also, according to the BEA Handbook of Methods, “a number of reliability studies concluded that [GDO] would better reflect the economic growth in a particular period by diminishing the known measurement inconsistencies between [GDP and GDI], such as timing differences, gaps in underlying source data, and survey measurement errors.”¹³ Given this observed feature of greater reliability with GDO, researchers at the Federal Reserve Banks of Philadelphia and New York went even further, refining the comingling of GDP and GDI beyond just a direct numerical average of them and creating “blended” model-based measures such as GDP Plus and GDP Solera.¹⁴ The observed record of lower revisions to GDO also led Steven Braun of the White House Council of Economic Advisers (CEA) to recommend estimating a GDO-based productivity series.¹⁵

It should be clarified here that reliability, as measured by size of revision, is a useful concept for the evaluation of economic series, though it is a concept that is different from accuracy. According to BEA, reliability “concerns the repeated estimation of an event,” whereas with accuracy “one is usually referring to the difference between the estimate and some ‘true’ value.”¹⁶ So, while revisions are a measure of reliability—of how consistent or variable a measure is through repeated vintages—revisions are only, according to BEA, a “partial indicator of accuracy” because there remains some error in the estimates even after incorporating revisions.¹⁷ See further discussion of these distinctions in the following section, “Measurement uncertainty and tracking GDP, GDO, and GDI.”

Interest in GDI itself also grew during the 2000s and 2010s, as it indicated certain business cycle and trend turning points earlier than GDP. For example, according to Braun, GDI provided “the first real-time evidence of the post-1995 pickup in productivity growth.”¹⁸

Also, according to Jeremy J. Nalewaik of the FRB, "[GDI] picked up the onset and the severity of the 2007–09 downturn better and sooner than [GDP]."¹⁹ Chart 2 displays the GDP and GDI series as of the December 2008 GDP release and their current values.²⁰ We see that initial estimates of GDP were showing continued economic growth during the early portion of the Great Recession, whereas GDI had been slightly declining. GDP was subsequently revised downward toward initial estimates of GDI, and GDI was also revised down.

The NBER Business Cycle Dating Committee similarly found early estimates of GDI useful in that period. In its December 2008 recession dating announcement, the committee cited the movements in GDI—which corroborated similar movements in payroll employment data—in its decision to declare the peak quarter to be the fourth quarter of 2007, despite GDP continuing to rise into 2008.²¹ More broadly, Nalewaik noted that “real-time GDI has done a substantially better job recognizing the start of the last several recessions than has real-time GDP.”²² Also, Robert Gordon and some private forecasters had begun using GDI in their productivity models.²³

Given the widespread interest from both the government and research communities in GDI and GDO, BEA responded by providing greater prominence to these measures in its GDP news release. GDI was included in table 1 along with GDP beginning with its July 2013 release, and GDO was added two years later, in its July 2015 release.²⁴ The agency has continued to publish these three measures in this way since that time.

A decrease in reliability of GDI and GDO in the 2020s

Although the data through around 2015 did show clear benefits to using GDO and GDI—namely, the superior reliability of GDO as indicated by its lower revisions and the benefits of GDI for improving cyclical analysis—subsequent reliability analysis data published by BEA in the 2020s showed that some extent of these benefits may have been transient and conditional. Specifically, a gradual decrease was observed in the reliability of GDI over the course of several BEA reliability analysis articles that in turn ultimately led to GDO losing its outperformance relative to GDP as of 2024.

Over the course of a series of Survey of Current Business (SCB) articles in 2011, 2018, 2021, and 2024, the GDI MAR over the longest vintage period (from the third release to the latest release) gradually increased, from 1.33, to 1.49, to 1.68, to 1.91 percentage points.²⁵ This reduced reliability in GDI was also reflected in the gap between the GDP MAR and the GDI MAR during these years, which also gradually rose, from 0.21, to 0.24, to 0.33, to 0.65 percentage points.

As GDP and GDI grew increasingly further apart in reliability, the synergistic effect of their combination has become outweighed by GDI’s reliability decline, which has been adding more noise and less signal to the average measure. With the 2024 SCB article, GDO did not outperform GDP in reliability from the third release to the latest release (as its MAR was slightly higher than the GDP MAR), and GDO also underperformed GDP in 6 of the 10 vintage comparisons in that article.²⁶

It becomes even more apparent that GDO's outperformance in reliability is likely conditional upon GDP and GDI staying within a similar range of reliability of each other, when doing a comprehensive examination of all 33 vintage comparisons in the four referenced SCB revisions articles. This analysis reveals a correlation between the size of the gap between the GDP MAR and the GDI MAR and how GDO performs relative to GDP and GDI. In 79 percent of cases in which the gap between the GDP MAR and the GDI MAR was less than 0.4 percentage points, GDO outperformed both measures (this is visible in chart 3, in the frequency of these cases, identified in blue, that are on the left half of the chart). However, in 79 percent of cases in which the gap in MARs was greater than 0.4 percentage points, it was the inverse result: GDP outperformed GDO and GDI in these cases (this is also visible in this chart, in the frequency of these cases, identified in red, that are on the right half of the chart).²⁷

These results show that GDO’s outperformance of GDP in reliability was not an established fact but instead is likely conditional on whether GDP and GDI are both within a similar range of reliability in a given review period. During the 2000s and 2010s, when GDP and GDI exhibited similar reliability, GDO outperformed both measures, though when the gap between these two measures expanded, GDO did not continue to outperform.

Measurement uncertainty and tracking GDP, GDO, and GDI

When analyzing these three estimates of output, it is important to keep in mind that they all exhibit a degree of error and revisions, and that each estimate has relative strengths and weaknesses from both methodological and indicator-performance standpoints. This is why the NBER, the CEA, the FRB, and the Treasury Department analyze all three of these measures in monitoring the U.S. economy, and why BEA notes that “it is useful to look at growth in both GDP and gross domestic income in assessing the current state of the economy.”²⁸

To get a sense of how substantial the revisions in these measures are relative to the typical percent changes in output observed, we can compare the revision amounts to the average percent changes for each measure. As of the 2024 SCB revisions article, GDP, GDO, and GDI exhibited MARs from the third release to the latest release of 1.26, 1.29, and 1.91 percentage points, respectively, between the first quarter of 1999 and the fourth quarter of 2022. These revisions are sizable relative to the average annualized percent changes for these measures over the same period, which are 2.19 percent for GDP, 2.16 percent for GDO, and 2.13 percent for GDI.²⁹ Thus, there is a substantial potential for revision to output measures in a given quarter, reflecting the degree of uncertainty in early estimates.

Furthermore, the effect of output revisions on labor productivity is noteworthy. A recent BLS working paper observed that MARs relative to initial estimates five years prior were 1.64 percentage points for labor productivity, 1.68 percentage points for output, and 0.74 percentage points for hours.³⁰ These MARs indicate that a substantial portion of revisions to labor productivity was from revisions to output, and that the contribution to labor productivity revisions was greater from output than from hours.

As discussed earlier, a majority of source data underlying initial output estimates are based on indicator and trend data, rather than comprehensive data (only a small amount of which is available for, and used in, early estimates).³¹ These indicator and trend data are less reliable than the comprehensive data and underlie the large revisions to early estimates of output. One option to mitigate the measurement uncertainty from this source data availability issue is to wait for the revised data published in the quarters and years following the initial estimates. These later vintages incorporate a greater amount of comprehensive data and are, thus, more reliable. This increase in reliability is indicated by the gradual decrease of the MARs of the three output measures as subsequent releases are published.

And, for those who rely on early estimates, one option is to track GDP, GDI, and GDO together and thereby utilize all available NIPA source data that are encompassed by these three estimates. As we have seen from the results of the cited SCB articles, the performance of these three measures can vary over time and from one historical period to the next. And, because all SCB reliability analyses are retrospective, it is not possible at any given time to know exactly how the estimates are currently performing relative to each other. Though GDP has had consistently lower revisions than GDI over time and also outperformed GDO in reliability in the recent SCB article, GDI has exhibited useful features of its own in certain periods, and GDO has outperformed GDP in reliability in three of the four SCB analysis articles. Accordingly, one way to hedge this uncertainty is to take all three measures into account while also keeping in mind their past performance.

Also, a notable benefit of using all three measures is that although GDO and other blended output measures have exhibited lower revisions than either GDP or GDI during most historical periods, only GDP and GDI allow for decomposition analysis because, according to BEA, “there is no obvious way of distributing the results of the averaging among the major components of GDP and GDI.”³² Given that decomposition analysis is a vital part of understanding the state of the economy through these measures, it is important to continue to use GDP and GDI for this functionality that is available with these two measures, as well as for their other attributes as noted above.

And finally, we should keep in mind that there is a deeper layer of uncertainty with these measures that goes beyond revisions themselves and relates to what William D. Nordhaus refers to as “The Two Map Problem.”³³ Namely, this is that even after all comprehensive data have been incorporated into the measures in the months and years following the initial release, and revisions due to these improvements have been tallied, there will remain both a difference between GDP and GDI—represented by the statistical discrepancy—as well as a difference between these measures and the “true” level of national output, which is unknowable. According to BEA, this issue “arises primarily from error in the source data and secondarily from BEA’s estimating procedures that utilize the source data. On the assumption that later estimates are more accurate than earlier ones, revisions can be viewed as measuring part of the total error in earlier estimates. The rest of the error in the earlier estimates, which is unknown, becomes the total error in the later estimates.”³⁴ Thus, there is some unmeasurable degree of error within these estimates that will remain, which we should also be aware of when analyzing and comparing them.

What this understanding leads to is an acknowledgment that by comparing the MARs of GDP, GDI, and GDO, we are comparing only what BEA has referred to as “a partial indicator of accuracy”—the measurable portion of the error in early estimates.³⁵ We are not able to compare the remaining part of the error that is an unmeasurable but real phenomenon in each of these estimates, and of which we cannot know the relative sizes. This provides further support for continuing to track and monitor all three output estimates, while also keeping the measurable extent of the difference in their respective performance in mind.

Estimating GDO-based and GDI-based labor productivity

This section will outline the method for estimating nonfarm business labor productivity measures that utilize output based on GDO and GDI. This method applies a residual-based approach that is similar to the one used by BEA in estimating its official GDP-based output for the nonfarm business sector.

BEA output for the nonfarm business sector is a chain-type, current-weighted index constructed after excluding from GDP the output of four sectors: general government, nonprofit institutions serving households, households, and farm.³⁶ To create an analogous series by using income-side data, we can use a simplified residual approach that uses the nominal output of these four sectors and the (official, product-side) nonfarm business implicit price deflator (IPD). To do the computation, the nominal output measures of the four sectors are subtracted from nominal GDI and then the result is divided by the nonfarm business IPD. A GDO-based output measure for the nonfarm business sector can similarly be produced by subtracting the nominal output of the four sectors from nominal GDO and then dividing the result by the nonfarm business IPD.

The rationale for taking a residual-based approach for estimating these two alternative nonfarm business output measures is based on the fact that the official nonfarm business output measure is estimated by using a residual-based approach. Since GDP and GDI are conceptually equivalent and much of the output excluded from the nonfarm business sector is income-based, a residual-based approach may also be taken with GDI and GDO as with GDP. This residual-based approach yields income-side and GDO-based nonfarm business output measures that capture the source-data-based differences between these series and the official (product-side) nonfarm business output series.

Also, the rationale for applying the nonfarm business IPD to nominal nonfarm business GDO-based output and income-side output is that this is a similar approach to the BEA deflation method for its GDP, GDI, and GDO measures. BEA applies the GDP IPD to nominal GDO and nominal GDI and produces real GDO and real GDI.³⁷ We can take the same approach by applying the nonfarm business IPD to GDO-based and GDI-based nonfarm business nominal output and produce corresponding real output series. Using this deflation method would keep BLS in accordance with BEA’s methodology for deflating its output measures.

It is possible to use these GDI-based and GDO-based output measures to compute experimental labor productivity and related measures to complement the official GDP-based measures. But it should be noted that while first estimates of GDP-based productivity estimates for a given quarter become available with each preliminary Productivity and Costs release, estimating GDO-based and GDI-based productivity estimates for a given quarter first becomes possible one or two releases after the preliminary release because of publication lags in GDO and GDI. GDI-based and GDO-based experimental productivity series could provide additional information for the analysis of economic growth and research opportunities for those in the governmental and academic communities.

Comparing GDP-based, GDO-based, and GDI-based labor productivity series

Chart 4 tracks the course of the official GDP-based nonfarm business productivity measure and the GDO-based and GDI-based measures during the current business cycle, which began in the fourth quarter of 2019. We see that the GDO-based productivity measure reduces some of the volatility observed in this period—which it achieves by tracking a course between the GDP-based and GDI-based measures—particularly during and following the COVID-19 pandemic recession of 2020.

We see a similar result when comparing these series over a longer period encompassing both the current and last business cycles, in chart 5, with the GDO-based output measure also lessening some of the volatility during this period. In addition, we see that the GDO-based output measure reduced the effect of a sustained gap between the product-side and the income-side productivity measures that emerged from 2007 to 2012 and from 2016 to 2019, charting a course between them. Also, the widening gap between GDP-based and GDI-based productivity during 2007 and 2008 reflects the phenomenon discussed earlier, of GDI having helped make economic observers aware of a stumbling economy before the economic downturn was apparent with the GDP measure. 

In addition to real-time analyses, longer-term analyses of GDI-based and GDO-based estimates of productivity are also possible to calculate back to the first quarter of 1947. Chart 6 shows selected business cycle period growth rates for labor productivity estimates based on the three different output measures.³⁸ We can see that they follow a largely similar path, with each measure decelerating in the 1970s, accelerating between the early 1980s and early 2000s, slowing down after 2007, and then picking up slightly in the current cycle.

Notably, chart 6 reflects GDI capturing the 1990s productivity speedup earlier than GDP, which the CEA noted. We can see that the GDI-based productivity measure accelerated faster than the GDP-based measure and peaked in the 1990s. In contrast, the GDP-based measure peaked in the early-to-mid 2000s. We can also see the GDP-based series outpacing the GDI-based series in the 1970s, reflecting a rising statistical discrepancy in that period. As expected, the GDO-based productivity series displays a rate halfway between the GDP-based and GDI-based measures in all periods.

Over the full long-term period since 1947, the data show a fundamentally similar story for all measures. All three labor productivity measures grew at nearly identical average annual rates over this period: 2.13 percent for the GDP-based measure, 2.14 percent for the GDO-based measure, and 2.15 percent for the GDI-based measure. These similar annual rates show that these three nonfarm business labor productivity estimates follow similar trends over time, though with slight differences over various periods because of the unique attributes of these three series. As we have seen in this section, creating long-term labor productivity series by using three different types of output allows for deeper historical analyses of U.S. productivity, as this approach accounts for all available output source data and encompasses both a product-side and income-side perspective.

Conclusion

GDP and GDI provide different perspectives on growth in the U.S. economy by bringing together source data from different vantage points and with different attributes. GDO, as the average of GDP and GDI, provides a way to synthesize this information into a single indicator that combines all relevant information. Estimating quarterly labor productivity measures by using GDI and GDO offers a complement to the official productivity estimate for the United States that is based on GDP. Over some historical periods, GDO has exhibited smaller revisions relative to those of GDP and GDI, and GDI has been valuable for understanding some cyclical turns and period changes. Given that labor productivity growth is a primary driver of economic growth, estimating labor productivity data on a GDO and GDI basis, in addition to a GDP basis, informs us on the current state of U.S. economic growth in a way that utilizes all available output source data.