Why the labour market stats are not all that they seem

There are worrying signs - but we need to be wise to misleading graphs that make things worse than they seem

Nearly every month we see the publication of the latest labour market statistics, and this week we saw yet more coverage about how badly the labour market is doing, the rising levels of inactivity, and record levels of long-term sickness.

But often these statistics are not all that they seem, because the charts that journalists and analysts produce can be misleading:

1. They sometimes show the absolute numbers of people who are inactive. I simply do not understand why someone would do this - it confuses population growth with what’s happening in the labour market (as I’ve previously said re benefits claims).

2. They often manipulate the graphs to make recent changes look larger, most commonly by looking only at the trend since Covid. This trend is important, but we can only understand it in the context for how this looks over a longer period. Otherwise small changes in historic terms are blown out of all proportion to seem like catastrophic declines. There’s also lots of selective choices about which date ranges to focus on, again which often serves to make things seem worse than they are.

3. They usually ignore population ageing. We know that the workforce is older than it used to be (and pension ages have been rising too) - this is something we need to be aware of, but we need to avoid confusing it with whether people of a given age are less likely to be in work.

4. They focus exclusively on inactivity. But in my view, it’s not that important how someone labels their reasons for not working (whether this is unemployment if they’re seeking work, or inactivity if they’re not looking for work, and whether the reason they give for not looking for work is long-term illness or something else) - indeed, more people move from inactivity→work than from unemployment→work (see p16 of this IES briefing).¹ The main brute fact is whether they’re working or not. Not everyone agrees with this, but at the very least, any analysis of inactivity needs to be complemented by a clear analysis of what’s been going on with employment per se.

(I’m not going to criticise particular people for this, because lots of these people have done amazing jobs at communicating the labour market to policymakers and the public, and have much more expertise than me to boot. But still: I hope it’s a useful prompt for everyone to reflect on what they do!).

In the rest of this post, I want to present a series of graphs that avoid these issues, and I think give you a better picture of what’s been going on. I’m still using the official labour market statistics using the Labour Force Survey (LFS), but just presented in a different way. I focus on non-employment, which is obviously just the inverse of the employment rate, but I prefer presenting in this way to make it clear that it covers both unemployment and inactivity (and I’ll return to the issue of different types of non-employment in a future post). As ever, please comment below if you disagree…

The overall rate of non-employment

One of the first worrying signs is that the overall rate of non-employment has risen recently, from 23.8% before Covid to 25.5% in the most recent period.² This represents a bit of a jump, as the non-employment rate was only 24.5% nine months ago, and 25.0% three months ago.³

However, our understanding of this is a bit different if we put this in historical perspective, as shown below. The rise since Covid is clearly visible - but pales in comparison to the rises in earlier recessions. Moreover, the non-employment rate now is incredibly low in historical terms - it’s higher than 2017-19, but is otherwise lower (mostly much lower) than at any point since records began in 1971.

It’s particularly useful to split this by gender, shown below. This is pretty basic stuff - but it’s important to emphasise it again and again, because it changes our understanding of what has been going on:

• For women, non-employment rates are flat since Covid, coming after a long reduction in non-employment. This earlier fall reflects changing gender norms over time. But (contrary to some charts) it’s clear that there’s not an iron law that women’s non-employment keeps reducing at a fixed rate - it was pretty flat in the 2000s before the financial crash, for example.

• For men, non-employment rates have indeed risen since Covid, after which there was a fall followed by a recent rise. But what’s striking is that male non-employment rates have been in a pretty narrow window of 20-25% ever since the early 1990s. In contrast, in 1973 the male non-employment rate was about 8%! Male employment rates took a hammering over the 1970s and 1980s, from which they have never recovered.

But what has happened to young people?

Lots of the recent attention has been around the situation of young people, which is what we’ll focus on for the rest of the post. Note that when we split things by age (particularly detailed age groups), we’re also mostly getting rid of the effects of population ageing, which is helpful.

If we first look at the situation for broad age groups (shown below), we do indeed see that the main issues are for young people. Youth non-employment rates are now 50.1%, up from 44.7% pre-Covid; and what’s worse is that they basically recovered after Covid until early 2023, after which they’ve suddenly deteriorated. This isn’t unprecedented, because youth non-employment was even higher in the early 2010s. But it is true to say that youth non-employment is now almost as high as it ever has been. (But wait: more on this below).

As for older ages - well there’s extremely little change among 25-49 year olds, who are still close to record lows in non-employment. Among 50-64 year olds, there has been a slight rise since Covid, but again we’re still close to record lows. (Not shown in the graph, but there’s little change in non-employment rates for those aged 65+; they rose slightly, but then fell, and are again now close to record lows).

If we further split young people by age, we see that 16-17 year olds are basically where they were before Covid (but things actually improved for them until 2023, after which non-employment has risen). The picture for 18-24 year olds is basically the same as in the chart above, but with a smaller Covid spike in 2020.

The age patterns vary for men and women a bit, and you can see the split by age and gender here. The biggest difference is for 25-49 year olds, where non-employment rates have gone up since Covid for men (by 2.0pp) but fallen for women (by 0.7pp). The rise in non-employment since Covid in young people is also slightly higher in men than women (6.0pp vs. 4.8pp). Trends among 50-64 year olds are similar for men and women, though it’s more unusual to see rising non-employment for women (as this comes after a long period of declining non-employment). But still - even where there have been slight rises in non-employment, levels are pretty low in historic terms.

Young people, education and work

One of the problems with this picture, though, is that ‘non-employment’ includes full-time education. It makes sense to exclude this, but some of the coverage this week did this by focusing on inactivity among those not in education. (If you look at inactivity among those not in full-time education, you’ll see that this is at record highs since records began in 1992 - but I’m not convinced that this is a useful statistis…).

A better way of dealing this is to look at how many young people are not in full-time education or employment (the so-called NEETs). This is shown in the ONS labour market statistics, but it’s very easy to create from them (though we can’t split this by gender frustratingly). The top line in the chart below is the same as in the previous chart - it’s the overall rate of non-employment in young people. Below that, though, the red line shows young people who are neither employed nor in full-time education.

This shows that there has been no change in young people not in employment/education since the eve of Covid. There was a 2.5% fall to mid-2022 (then a discontinuity), then a possible 1.4% rise from late-2022 - but the recent figures are fluctuating and there’s no clear trend.⁴ This shows that there has been a recent rise in 16-24 year olds not in full-time education or employment, to 15.4% (from a low of 12.0% in summer 2022), and this is higher than just before Covid. But Moreover, this is a lot lower than in the early 2010s (let alone the early 1990s), and probably⁵ similar to the level in the 2000s.

Final thoughts

Putting this all together - there are some genuinely concerning trends in the latest labour market statistics, with a noticeable rise in the past year in young people not in education nor employment, as well as smaller rises since Covid in non-employment among 50-64 year olds and 25-49 year old men.

However, these levels are still relatively low in the context of the last 30 years. It’s sensible for analysts to figure out what’s going on and for policymakers to respond - but there is simply no call for sensationalist coverage about worklessness among young people, nor among the economy as a whole.

1. Just to be clear - the probability of moving into work is higher for unemployed people (who are seeking work and available to start soon) than inactive people (who are either not seeking work or unavailable to start). But it is quite possible that we see changes in whether people report looking for work that do not translate into any meaningful changes in the labour market, e.g. because of what people are required to do by the benefits system for compliance.

2. The LFS presents data for quarters, so '‘just before Covid’ refers to Dec 2019-Feb 2020, and ‘latest period’ refers to Dec 2023-Feb 2024. For plotting these on the graphs, I have used the 1st day of the middle month in each quarter (so the latest release is plotted as 1st Jan 2024).

3. The official labour market statistic releases don’t generally include confidence intervals to capture uncertainties around sampling errors. In general, the all-ages figures are very precise (so the recent changes are likely to be statistically significant), but the figures get more and more uncertain as the group we’re talking about changes - so for example, the uncertainty around 16-17 year olds is much greater.

4. This sentence was added 19th April (a day after the original post), and the following sentence deleted (shown in strikethrough), after I edited the graph to show the discontinuity in the data in mid 2022.

5. This is true if we assume that the discontinuity (caused by recent LFS reweighting) does not affect the figures in the 2000s. My guess is that it doesn’t, but it’s hard to know for sure.