Still an awesome chart!

Today the Resolution Foundation released an excellent and fascinating paper looking at data behind Branco Milanovic’s ‘elephant chart’, and asking what conclusions can be realistically drawn from it.

My tweets may have played some small role in popularising the chart – the reporting of an opinion I tweeted that it is ‘the most important of the last decade’ is footnoted a couple of times in the paper, and it even appeared in a Dutch Trade Minister’s speech a couple of weeks ago.

Here’s the tweet, which I drew on my iPad of a picture that I snapped while reading Milanovic’s terrific book:

I have been thrilled to have had a role in drawing attention to the work that Milanovic’s and Lakner have conducted, a summary of which can be read here.

But judging from my Twitter feed this morning you would think that the Resolution Foundation paper has found major methodological flaws that makes a nonsense of the chart.

Not so.

To reiterate, the elephant chart is an amazing chart. It paints in one graphic a picture of the global economy that is (in my opinion) unrivalled. It is so good precisely *because* it combines income changes and demographic changes; it tells the story of recent global history and the degree to which the rise of China (and India) have changed the world. It also illustrates that lower income households in the West have largely not participated in the remarkable global growth over the two decades. The original Milanovic & Lakner paper  dwelt at length on the compositional issues behind the chart, and there was even an animated gif that @MaxCRoser put together using country-specific compositional data contained therein which is pretty stunning, and was widely distributed before inexplicably disappearing from Twitter. Luckily I have been able to find and embed a copy below: global-income

The Resolution report does the world a service by further drawing out the data behind the chart, discussing what policy implications can and cannot be drawn from the chart, and reminding people of the compositional issues.

Compositional issues are hard. And this is probably the most striking chart in the report for people interested in the impact of compositional issues. It even makes its way into the FT.


This chart is complicated. As its title suggests, it shows ‘Growth in average per capita household income of each percentile group (rolling average) if there had been no income growth and changes were solely due to uneven population growth’.

I think that this means that it imagines a situation where:

  1. every worker is put into an country-specific income decile in 1988, and the mean income of the decile is then fixed in real terms;
  2. net new workers per country are distributed evenly across these country-specific deciles;
  3. all countries for which there is data have their country-specific deciles chucked into a spreadsheet in 2008;
  4. the income required to make it into each percentile in 2008 is then compared to the income required to make it into each percentile in 1988.

The outcome is then charted. (Apologies if I’ve misinterpreted.)

If this understanding is correct, the chart then shows the impact on the elephant chart of

  1. poorer countries having had faster population growth than richer countries;
  2. starting the exercise in 1988 when the global income distribution had the following shape:


Or, as the author of the Resolution report, Adam Corlett, puts it:

“For example, the poorest decile of people in the US were in the 70-75th percentile range in 1988. But population growth among poorer countries would have pushed those Americans up into the 75th-80th percentile range by 2008. The bottom US decile would be replaced in the 70-75th percentile part of the global distribution by the richest urban Chinese, but the latter’s average income was around $1,500 compared to the former’s $2,600: producing a fall in the average income of those percentiles.”

And so, when holding constant the country population share the elephant chart looks as follows (red line):


Is this a ‘truer’ version of the original elephant chart? Not at all. In fact, holding constant population shares stable is a nonsense: it no longer shows a picture of profound changes that have been experienced across the global economy as a whole. The author both understands this and makes no attempt to conceal it, because he is trying to do something important and useful to which this new chart is particularly well-suited: to correct a misperception as to what the original elephant chart showed.

Specifically, it issues a corrective to claims that working class developed market incomes have *stagnated* in real terms during the period. And here, despite being one of the people who actually read Milanovic & Lakner, I am at least partly culpable. In a blog for voxeu among other places, I have used the elephant chart and stated “there is a large section of people who are well-off in global terms who have largely not participated in global growth over the past 20 years. That section is populated largely by the Western lower middle and working classes”.  The words may not be technically incorrect, but to say that they fail to draw attention to the compositional issues underlying the chart is more than fair. Furthermore, they mask the fact that while I have looked at US income distribution in some depth, and think myself relatively familiar with the data that is published in the UK, the impact of Japanese income data on the overall chart had passed me by.

When we look at the constant country population share chart from the Resolution paper we can see that my claim that a large section of Western lower middle and working classes largely not participating in global growth might more accurately (but not massively more accurately) be described as having experienced a cumulative real income growth of c25% (although this will vary *meaningfully* by country – with Japanese lower income deciles experiencing contraction, US lower income deciles experiencing low but positive growth, and Western European lower deciles experiencing c45% cumulative income growth). These levels of cumulative income growth have been lower than the income growth at the top of each of the income distributions for the respective developed market block (leading in many developed countries to higher levels of income inequality), and lower than the income growth of the global median or global poor (leading to lower levels of income inequality across the globe, principally due to the rise of China). And so while real incomes have risen for lower middle and working classes in absolute terms, the bottom 80% labour share of GDP in the UK and US has declined as a proportion of GDP (defined as the labour share of GDP multiplied by the proportion of labour income received by the bottom 80% of the income distribution, see chart below), while the relative cost of labour in the West vs the rest of the world has reduced. (It is also notable that the big decline in the UK occurred in the 1980s, with an evening out thereafter.)


I don’t think that all these things are disconnected, and I’m pretty sure that the Resolution Foundation report isn’t arguing that they are disconnected. Instead, it is seeking to quash a meme that real incomes have stagnated for developed market middle class workers on account of globalisation. It does so with aplomb. As Torsten Bell, also of the Resolution Foundation, blogsboth those saying globalisation automatically benefits everyone and those saying that developed world middle classes have seen no income growth are wrong. Perhaps most crucially, where individual countries lie in between those extreme positions is to a significant degree down to policy choices.”

Given that I read Milanovic & Lakner’s compositional discussions in the original paper, why did I not major on these compositional discussions? Well, I didn’t think that they change the use to which I put the chart (discussing the global labour glut that came about with China joining the global trading system, the associated substitution of capital for labour, and the loss of labour bargaining power in a variety of developed market economies). Although I was not aware of the contribution the Japan had made to the distribution until reading the report.

I tend not to be accused of oversimplifying things. But given the reaction to the Resolution Foundation report, I certainly feel complicit in propagating the end-product of a complex piece of analysis without due caveats.

However, it’s still an awesome chart.

Visualising Wage Changes II

In the last blog I bemoaned my lack of understanding of available data on the UK labour market, and also on the way in which this data is traditionally visualised (which can impede its discussion). It seems that the ONS has some really interesting data but hides it well.

I appreciated very much the crowd-sourced pointers on the data front that the blog generated. And also am indebted to @Nevillehill who downloaded and sent me enough ONS data to freeze my bewildered Mac.

As is often the case when looking at data, I am left with new questions. And as is often the case with me, some of these questions sound pretty stupid. The stupidest question that I am left with is this: how do you – as an individual – know whether your earnings are growing quickly or slowly? But I think it is an interesting question.

What do I mean? Well, let’s look at the data.

The first chart below shows the pace of wage growth on a per annum basis split by age.* This shows that the earnings of a brand new 25 year old rose in 2007 were 3.0% than the earnings of a brand new 25 year old in 2006, while the earnings of a brand new 25 year old in 2011 were 1.7% below the earnings of a brand new 25 year old in 2010.

Increase in median earnings for a brand new person of a given age (age cohorts 18-55)

median earnings brand new person

But we are only a brand new 25 year old once. After that we are a brand new 26 year old. Then a brand new 27 year old. We get to be one of these things one time only and so while knowing how much more the people a year younger than you will get paid for the role you used to do might be interesting, it doesn’t relate much to you and cannot be observed in your paycheque. What about the change in earnings that we as individuals can observe in our paycheques? What has happened to them?

The second chart below shows the median step up in earnings experienced by people as they have become older over recent years. In this chart we are 18 years old once, then 19 years old once, then twenty years old once. And the change in earnings experienced between ages is plotted.

Median increase in earnings experienced in given year, given starting age in that year (age cohorts 18-55)

median earnings in a given year

This shows that a median 25 year old personally increased their earnings by 6.2% in 2007 as they moved from being 25 years old to 26 years old (compared to the 3% bump up in earnings that the next batch of 25 year olds would have had over you when you were 25), and in 2011 a median 25 year old personally increased their earnings by 1.9% as they similarly aged (compared to the 1.7% reduction in earnings that the next batch of 25 year olds would have had compared to you when you were 25).

Of course, from a traditional macroeconomic perspective, what individuals experience over time as individuals has never really mattered: it is what the workforce collectively experience that counts.

But to this I would say two things.

Firstly, as per the previous blog, some nod towards the demography of the workforce might be worthwhile in considering the impact on aggregate wage changes if institutions frame their consideration of employment and compensation practices with reference to either age (which would be illegal) or experience (which would not be). As people age so their earnings growth has traditionally flattened out, and having a bunch of older people whose earnings are not growing fast just because they are young could in aggregate make a difference to wage growth numbers (depressing them). I am not a labour market economist and am sure that the smart folk who are had this all in hand decades ago, but none of this sort of good stuff comes through to market-types like me on a day-to-day basis and any pointers would be well-received. (Apologies if I haven’t been paying attention.)

Secondly, I have sympathies with classical economists who believe that the tug of war between capital and labour is one of the most important things going on in a political economy. From this perspective employers have a high level of information about wage developments across age and skill cohorts for their given industry (or at least company), but each employee has information only about wage developments as they apply to their own age and skill cohort. And given that you are only the same age once, labour cannot tell whether they are getting a good or a bad deal by observing their own personal experience. In the second chart above, the median 25 year old turning into a median 26 year old in 2011 had a wage increase of 1.9% rather than a wage cut of 1.7%. The median experience of earnings growth for young (less well-compensated) workers is hardly ever as bad as economy-wide average earnings growth, while median older (better compensated) workers will experience consistently lower levels of earnings growth than the economy-wide average. Absent unions, younger workers have no first-hand data as to whether their above-average earnings growth represents an historically good deal or bad deal: wage growth in the early years of their employment is not only entirely consistent with aggregate wage deflation, but taking smaller positive wage gains than has traditionally been associated with early-stage career development could be the main driver of wage deflation. This is by no means a new thing. But it casts a different light (for me) upon the public conversation on income and wealth inequality that appears to inform ever-more of the political debate in the UK.




* I split the age cohorts for which the ONS provides data smaller cohorts sized containing only people of a given age (eg 24 yrs old, 25yrs old etc), using the (wrong) assumptions that: (1) there are the same number of individuals of every age across the population; (2) that the cohort average wage was the wage received by the smaller (made up) cohort in the middle of the larger (actual) cohort; (3) that wages rose or fell to in linear increments by age between the actual data points from the ONS. I’m gambling that these wrong assumptions are insufficiently wrong to meaningfully change either chart or any substantive point made in the blog (but will bow to anyone with better data that contradicts this assertion).

Visualising UK wage changes

The Bank of England has been quite publicly troubled by the lack of wage inflation. And across the market, folks are looking hard at dismal charts that look a bit like this:

UK Average Weekly Earnings, 3mth YoY

weekly average wages UK Sep-14

Here we see a familiar picture of annual average wage growth in the UK: a labour market in the doldrums. This sits in contrast to a lot of the survey data regarding how easy or hard it is to recruit staff without paying up, Bank Agent reports on the same topic, and proprietary data from recruitment firms that some of my super-smart colleagues analyse when evaluating the worth of said firms. The Bank has started to drop references to low labour market churn perhaps suppressing wage inflation numbers into the last couple of MPC minutes, echoing some work that @nevillehill has been conducting at Credit Suisse.

What would help me a lot in thinking about this is some better data visualisation. As a human being I instinctively think of stagnant wages being associated with wages for people individually stagnating. But my brain tells me that stagnant wages at an aggregate level does not at all mean that individual households will experience stagnant wages. With zero aggregate wage inflation, the average Brit will experience rapid salary increases through their twenties and thirties (rising at around 7% per annum in their 20s, and 4% per annum in their 30s if pre-tax income is a guide to wages), experiencing an income peak in their mid-forties, and then through a mixture of changing working hours, working practices and plain redundancies, see their incomes trail off from there and into retirement.

So how would I like to see wage data displayed? As someone who grew up professionally in the bond market, I like these visualised using a wage-age curve (like a yield curve, it shows a snapshot, which needs refreshing over time; instead of curve roll-down there is age-roll-up in early professional years etc). I’m showing pre-tax income by age cohort because I couldn’t find the wage data by cohort easily on the ONS website. The first chart shows the basic pre-tax income of taxpayers by age cohort in 2010-11 (blue) and then again in 2011-12 (red), with the changes between years shown on the green dots at the bottom referencing the left-hand axis. The second chart shows the same data, but sizes the cohorts as a proportion of the taxpayers, and this corresponds to the size of bubbles (so showing the relative importance of these cohorts for aggregate wage data).

uk wage curve 1uk wage curve 2

Does data visualisation matter? To me it absolutely does. If monetary policy is contingent upon changes in wages, I should absolutely be focused on the degree to which wage inflation is changing due to demographic changes that can be anticipated (eg, everyone gets the same lifecycle in wages, but there are more older folks/ younger folks/ middle-aged folks etc), changing sectoral or skill-based labour market footprint, or due to changing compensation practices among firms. Changing the way that we look at data helps achieve that goal.

But unfortunately this brings me to the ONS website. Upon which it is probably best to dwell no further.

With the unemployment rate collapsing, some folks (understandably) raise the question as to whether we risk soon breaching the Non-Accelerating Inflationary Rate of Unemployment (NAIRU), and worry that at some point soon wage inflation will explode to the upside. But conversations with @Goodrichwatts have led me to the view that it may be more complicated than that: NAIRU is going to be contingent on the type of job added. If all new jobs are for burger-flippers NAIRU is probably going to be lower than anything we have historically seen; if new jobs from here all require PhDs NAIRU is going to be somewhat higher. It becomes a question of McNAIRU vs DrNAIRU if you like. I have my view as to where the balance lies between McNAIRU and DrNAIRU, but it is forward-looking and qualitative.