Measurement · 5.1

Productivity, and what the number contains

Productivity, and what the number contains. What is actually the case, and how it compares with what is repeated.

A practical software reference for this part of the discussion is Monitask's overview of employee time tracking app.

Labour productivity is output divided by hours. Both terms are harder to measure than the phrase suggests, and the number is routinely read as something it does not describe.

What it is not

Not how hard anybody is working. Productivity rises mainly through capital, technology and reallocation, and a worker in a highly productive economy is not exerting more effort than one in a less productive one. Treating the figure as a report on diligence is the commonest misreading.

The output problem

In manufacturing, output is units and quality change can be estimated. In services it is value added, deflated by a price index, and the deflator carries most of the difficulty.

Worse, in much of the public sector output has historically been measured by inputs: the value of health or education services taken as what was spent producing them. Where that convention is used, measured productivity is approximately constant by construction, and any statement about it is a statement about an accounting rule.

Several statistical agencies have moved to direct output measures for parts of the public sector, which is a substantial improvement and covers some services rather than all.

The composition effect

Average productivity can rise because everybody got better or because the least productive firms closed. A recession that removes marginal firms raises the average without anything improving.

The same applies to hiring: taking on inexperienced workers lowers measured productivity in exactly the situation everybody would call good.

The remark that keeps being quoted

In 1987 an economist observed that the computer age was visible everywhere except in the productivity statistics. The paradox it named was eventually partly resolved, partly by better measurement and partly by the gains arriving later than expected.

It is worth remembering when the same claim is made about any new technology, which it currently is.

The recent puzzle

Productivity growth in most high-income countries slowed markedly after 2008 and did not recover on the previous trend. Explanations offered include measurement error, low investment, weak demand, misallocation of capital, and the exhaustion of earlier general-purpose technologies.

None commands agreement. It is one of the largest open questions in macroeconomics and this entry does not resolve it.

Individual productivity

The aggregate concept does not transfer to a person. There is no meaningful denominator for an individual in most jobs and no separable numerator, and an organisation that reports individual productivity is reporting whatever proxy it chose.

Which is the subject of the rest of this part.

What the number is good for

Comparing an economy or an industry with itself over time, cautiously, and after checking that the deflator and the coverage did not change. It is a serious statistic constructed with care and it will not answer a question about a team.

The reading rule

When somebody attributes a productivity figure to effort, ask what the numerator was, how it was deflated, and whether the denominator counts hours paid or hours worked. Those three questions dispose of most confident claims made with this statistic.

The denominator

Hours worked and hours paid are different series and the gap is not constant. Unpaid overtime, which the entry on overtime describes as badly measured, sits entirely inside that gap.

Where a country's productivity figure uses hours paid, an increase in unpaid hours shows up as a productivity improvement. Nobody intends this and the arithmetic is unavoidable.

Comparing countries

Output per hour and output per worker rank countries differently, sometimes dramatically, because average hours differ so much. A country with high output per hour and short hours can sit below one with lower hourly productivity and long hours on the per-worker measure.

Both are correct. Quoting whichever supports the argument is standard practice and is usually done without noticing.

Quality change

If this year's product is better than last year's at the same price, real output rose and a naive measure will not see it. Statistical agencies adjust for this in some categories through hedonic methods, most extensively in computing equipment.

The adjustment is contested in both directions: too aggressive in some categories, absent in others, particularly in services where quality change is real and unmeasurable. Anybody comparing productivity across decades is relying on those adjustments whether or not they know it.

The firm distribution

Productivity differences between firms in the same narrow industry in the same country are very large and persistent, frequently a factor of two or more between the upper and lower quartile.

That dispersion is one of the most robust findings in the field, and it means an average conceals a distribution wide enough that the average describes very few actual firms.

For broader context, consult OECD productivity indicator.