# What a productivity residual can establish

Economic Explainer · Productivity and Technology
A rise in output per hour is evidence that more output has been recorded relative to working time. It does not identify which investment, management decision or technology caused that change. The distinction becomes especially consequential when a new technology attracts substantial expenditure: an aggregate productivity release can establish an outcome while leaving its explanation unresolved.

The Bureau of Labor Statistics distinguishes labor productivity from total factor productivity. The former relates real output to hours; the latter accounts for a wider set of measured production inputs. BLS also identifies several possible sources of total factor productivity growth, including efficiency, scale, resource reallocation and management. The residual therefore has a broader interpretation than the contribution of a named technology. These are measurement definitions, rather than findings about a particular innovation. [BLS, productivity concepts](https://www.bls.gov/opub/hom/opt/concepts.htm).

## An accounting exercise with explicit assumptions

Consider an entirely hypothetical producer. Use a value-added production boundary, so purchased intermediate inputs have already been deducted from output. Assume that capital services and quality-adjusted labor are the only measured inputs, that their accounting weights are fixed at forty and sixty percent, and that a logarithmic growth-accounting approximation is appropriate. These assumptions define the exercise; they have not been estimated from company accounts.

| Item | Assumed change in natural logarithms | Weight | Contribution |
| --- | ---: | ---: | ---: |
| Real value added | 0.040 | — | 4.0 log percentage points |
| Capital services | 0.050 | 0.40 | 2.0 log percentage points |
| Quality-adjusted labor | 0.010 | 0.60 | 0.6 log percentage points |
| Residual | — | — | 1.4 log percentage points |

The residual is calculated as `0.040 − 0.40 × 0.050 − 0.60 × 0.010 = 0.014`. A log change of 0.040 corresponds to an ordinary growth rate of approximately 4.08 percent; the table deliberately uses log changes throughout. Mixing an exact percentage growth rate with log contributions would introduce an avoidable inconsistency.

Suppose improved measurement subsequently raises capital-services growth from 0.050 to 0.070, with output and labor unchanged. Capital's attributed contribution becomes 2.8 log percentage points and the residual becomes 0.6. Production has not been revised. The allocation of its growth has changed because an input was measured differently. Neither residual can be interpreted without its measurement vintage.

## Capital purchases are not capital services

An expenditure total records purchases over a period. Productive services depend on the stock of assets in operation, their characteristics and how they are used. A newly delivered machine may still require installation. Software expenditure may support several years of production. An asset's purchase price does not, by itself, measure the amount of useful service it provides during the current quarter.

This difference creates a practical identification problem. If an analyst records a technology purchase as evidence of adoption but fails to measure the additional capital services, some contribution may remain in the residual. With a fuller capital measure, part of the same output gain would be attributed to capital deepening. The economic value of the technology need not have changed; its accounting location has.

The treatment of labor requires comparable care. An increase in specialized staff can raise the effective labor input even when total hours are unchanged. If the calculation adjusts for worker composition, that contribution belongs among measured inputs. If it uses hours alone, it may appear elsewhere. A meaningful comparison must disclose which labor measure is being held constant.

## Why timing and aggregation matter

An operational improvement can increase output before the full cost of training and organizational adjustment is visible in the chosen accounts. Conversely, preparation can absorb resources before benefits materialize. A short observation window can therefore record the implementation burden or the eventual gain without capturing the complete sequence. This is a reason to examine investment cohorts and deployment dates, not a basis for assuming that every weak observation conceals a future benefit.

Aggregate results also combine changes within producers with changes in their relative importance. Output may shift toward establishments that already had higher productivity. An aggregate increase then need not imply that the typical continuing establishment improved by the same amount. Entry, exit, outsourcing and industry classification can further alter the boundary being measured.

These possibilities require separate empirical tests. They should not be invoked after the event solely to preserve a preferred account of technological progress. The residual is useful precisely because it defines the part of recorded growth that the specified input accounting has left unexplained.

## Evidence needed for a technology claim

Attribution requires observations on actual deployment, the relevant output and input measures, and a credible account of what would have occurred without adoption. A comparison of adopters with non-adopters must consider why adoption differed. More capable firms may adopt earlier; faster demand growth may finance both investment and measured productivity. Simple correlation cannot distinguish those channels.

A useful robustness check repeats the calculation with alternative, defensible input measures and observation windows. If the conclusion depends entirely on one treatment of capital or one unusual quarter, that dependence should be visible before a policy or investment recommendation is drawn.

A defensible assessment would report the accounting result separately from the causal estimate. It would examine whether output quality, capital services and labor composition were measured consistently; whether comparable units faced similar demand conditions; and whether gains persisted beyond the installation period. The remaining uncertainty belongs in the conclusion. A productivity residual can discipline an explanation of growth, but it cannot supply the missing comparison on its own.

---

American Economic Senate