We have better metrics than the labour income share. Let’s use them!
A decline in the labour income share (LIS) is not definitive evidence of capital replacing human labour. Neither is a decline in the LIS good evidence of increasing income inequality. Yet I’ve seen a declining LIS used as “evidence” to support both claims1, and what’s more to suggest there’s a link between the two.
In this post, I argue the LIS is not a suitable metric for either automation or inequality.
What is the labour income share?
Put simply, labour income is wages and salaries. Anything else is “capital income”. The LIS is calculated as labour income as a proportion of labour + capital income. You might calculate it, for example, as total wages and salaries paid in an economy divided by nominal GDP. In that case, capital income is simply a residual – everything that isn’t wages and salaries.
In practice, it is difficult to reliably separate labour income from capital income. Income received by self-employed people is particularly tricky. A sports professional, for example, might receive a salary – think of it as payment for games and training sessions. But what about the income received from product endorsement? This seems more like a rent on intangible assets – ie, capital income. But I’m sure you could make good cases for other interpretations.
Government activity, taxes and transfers, imputed rent, and depreciation rates on capital, complicate a simple labour and capital income share measure. And lead to different treatment by researchers and different results.
Researchers and statistical agencies taking different approaches and studying different parts of the economy, over different time periods, arrive at quite different trends in the LIS. For example:
- Gutiérrez and Piton report that if they exclude housing income and account for self-employment, apparent declines in the LIS over 1970–2015 became increases for most of the countries they studied. By their measure, the LIS increased in all major European economies, except France.
- For New Zealand, Rosenberg (2017) uses official data on labour share from wages and self-employment from 1939. He finds a rise in income share until the 1970s and a steep fall in the 1980s. The OECD (2012) reports a small fall in the LIS over 1990–2009; Conway, Meehan and Parham (2015) describe a falling LIS over 33 years for 11 industries; Fraser (2018) finds a slight fall between 1996 and 2016 for 16 industries; Reddell (2017) highlights a decline 1972–2002 and subsequent rise 2002–16 for the whole economy; and Partridge and Wilkinson (2019) argue that the decline in the LIS halted around 1991 and it subsequently trended back up.
I have used Michael Reddell’s approach in the graph below.
Compensation of employees as a share of GDP (excluding taxes and subsidies)
What factors can influence the LIS?
Here’s three examples.
Nominal GDP goes up and down with the terms of trade – changes in the prices of exports and imports. A rise in export prices affects capital income more strongly than labour income, and so we would expect the LIS to decline in response.
Similarly, the LIS changes as the relative quantities of labour and capital change. The LIS should, for example, rise as labour market participation increases or the unemployment rate falls. Changes to official interest rates are ambiguous. Higher interest rates raise the returns to financial capital (reducing the LIS) but also make capital equipment more costly (decreasing its returns, and hence increasing the LIS). The net effect on the LIS may thus depend on distribution of capital and its ownership within and outside the country.
Policy changes can also influence the LIS. Take KiwiSaver, for example. Employee contributions do not change labour income, but in the longer run they create capital income for those same employees. Employer contributions2, on the other hand, decrease their capital income and increase labour income in the short term. They also contribute to the employee’s capital income in the longer run. I expect the introduction of KiwiSaver to have increased the LIS in the short run. But over time, capital earnings will swamp annual employer contributions, reducing the LIS to below its original starting point. So, we’ll end up with a lower LIS and higher incomes for workers in their retirement.
The LIS is sensitive to these and many other factors. An observed change in the LIS can only be reliably attributed to change in a single factor if all other potential factors are controlled for.
Is the LIS a good proxy for income inequality?
For the LIS to be any kind of proxy for income inequality over time, you’d need to make a lot of assumptions. For example, suppose:
- People can be neatly divided into two groups: “workers” (receiving labour income) and “capitalists” (receiving capital income).
- The relative proportions of workers and capitalists don’t change over time (and people don’t move between categories).
- Within group-inequality is constant over time.
- There is between-group inequality to start with.
Then a change in LIS would mean a change in income inequality for the population. But real life doesn’t neatly match these assumptions.
Income inequality is better measured as annual Gini coefficients for various types of income, or comparisons of income growth rates across income deciles. While such measures have their own issues, they are at least designed for the purpose and are subject to ongoing testing and refinement.3
Is the LIS a good proxy for the automation of human labour?
I’ll flip this around. How might we measure automation?
Consider a firm with $1000 a week to spend. They can choose to spend it on extra labour or extra capital equipment. Logically, they will choose the one that adds the most to their output. Put another way, if from a firm’s perspective the productivity of capital exceeds the productivity of labour, then we would expect to see that firm investing in capital equipment. And vice versa.
The wider economy is a collection of such firms. Improved or cheaper automation technology might be expected to increase the productivity of capital, making it more attractive for firms to invest. Some capital investments make labour more productive too.
To understand what’s going on with technological diffusion in the economy, you need to measure labour and capital stocks; and the productivity of labour and capital. Statistical agencies already measure these at an industry and economy-wide level. The LIS can’t usefully add much, as it conflates changes in the quantity of labour and capital with their productivity.
My take outs
- The labour income share is not a well-defined metric. Reasonable people have found differing ways to measure it, creating different datasets. Trends observed in one dataset are not necessarily present in others.
- The LIS is not a magic metric, able to do double duty. It is poorly suited for tracking changes in income inequality. Similarly, it is unsuited to tracking the automation of human labour.
- The LIS is particularly unhelpful as a measure of the impact of technological change on human work. We have better measures. Let’s use them!
1. The LIS has also been used as a proxy for real wages. But real wages can increase even while the LIS is declining.
2. To the extent that the incidence of the contribution falls on the employer.
3. Measuring inequality is certainly not straight-forward – see the APC’s research report on inequality in Australia – a stocktake of the evidence. For New Zealand, I’d start with the work of John Creedy and Normal Gemmell and that of Matt Nolan and Brian Perry.