British jobs and foreign workers: today's reports on immigration and unemployment
The impact of immigration on British workers is a matter of heated political debate. But anybody seeking light from the economics profession on this topic could be excused if they were somewhat confused by the research published over the last few days. On Monday, Migration Watch published a report highlighting the "remarkable coincidence between the rise in youth unemployment and the surge in immigration from Eastern Europe over the last eight years." Yesterday, new research authored by me and others at NIESR showed no relationship between immigration and unemployment benefit receipt. But later that same day the Government’s independent Migration Advisory Committee (MAC) said that “an extra 100 non-EEA migrants could lead to 23 fewer British residents being employed.” Not unreasonably, the media had some fun at the expense of economists in general. So what should be the reaction of an unbiased reader, economist or otherwise, who wants to know what all this adds up to?’
First the easy bit. We can dispose quickly of the Migration Watch report. As it says, betwen 2004 and 2011 an extra 600,000 Eastern European workers entered the UK labour force, while youth unemployment rose by 400,000. But the vast majority of that rise in youth unemployment took place during 2008 and 2009. During that period, the number of Eastern European workers actually fell slightly. Is this fall a "remarkable coincidence" too, just in the other direction? Or is simply exactly how you would expect the labour market to respond? Even on their own terms, the logic doesn’t add up. Nor is this the only hole in Migration Watch's analysis, as Matt Cavanagh points out here.
This is not the first time that Migration Watch have played fast and loose with the evidence on this very topic. An earlier report, still on their website, has a nice chart (the first one) which shows essentially that (for example) Manchester has more young unemployed people than Windsor, and also more migrants. They argue that this demonstrates a “strong relationship.” Your average GCSE maths student, however, would (I hope) argue that it demonstrates mostly that Manchester has more people than Windsor. I leave it to the reader to decide whose interpretation is more plausible. They manage to make some other pretty basic mistakes too, explained here. [Update, 19/1/2012. This report, and another one using a similarly flawed methodology, have now been removed from the Migration Watch website.]
Much more interesting therefore are the differences between our report and the MAC’s. I think three points are important here.
First, the two studies use the same methodology, known as the spatial correlation approach. Essentially, that amounts to looking, across different areas, at a) how many migrants came to an area over a given time period and b) what happened to unemployment or employment in that area. If, controlling for other things, areas with higher migration saw (say) higher unemployment, one concludes there is at least a prima facie case that there is a relationship, possibly causal.
Second, we use very different data sets. The MAC report uses, as is standard, Labour Force Survey data. The problem with this is that, even though it is a fairly large survey, the number of migrants (especially recent migrants) in many areas is quite small. This means that the study has to look at quite large areas – the 11 UK regions – to get decent sample sizes (it has also been argued that the LFS undercounts some migrants).
By contrast, we use, for the first time in the literature, actual data on National Insurance number registrations - the best and most comprehensive measure of people moving to this country to work. This allows us to look not just at regions but at local authority areas, and with considerably more accuracy. So the standard errors of our coefficient estimates – one measure of how much the data is telling you, and the precision of your results – are only about one tenth those reported by the MAC. We do have to look at a much shorter time period – the MAC data goes back to 1975, ours only to 2002 – but since the MAC in any case choose to split the data by sub-periods, and find different results in the more recent period, this may not be a major loss. It is certainly plausible that the impact of migration on the UK labour market has changed quite a lot in more recent years.
Finally, it is important to look beyond the headlines, and doing so one sees that the different results are more compatible than might be thought. We find, as has accurately been reported, that there is essentially no association between immigration and claimant unemployment overall. But the MAC emphatically did not find the opposite, that immigrants overall have an impact on employment at all times. In fact, what their key table – table 4.1, on page 63 – shows is that overall they, like us, find no statistically significant impact.
They then cut the data various ways – between non-EU migrants and EU migrants, over different time periods, and between periods of economic upturn and downturn. That gives 15 different cuts of the data, and for 11 there is again no statistically significant impact, while in 4 there is. By contrast, in our report, we similarly cut the data in a number of different ways – for none of them do we find any statistically significant impact of any type. So our results are far more consistent than those of the MAC, which could legitimately be read in a number of different ways.
The specific result that the MAC highlighted in their press release, however, is that for non-EU migrants in the period 1996-2010, where they find a statistically significant effect, amounting to a reduction in employment of 23 native workers for each 100 extra non-EU migrants. What should we make of this? For me – and I think for most professional economists – this is a suggestive result, but even if you take it at face value, given that it is only one of many, and applies only to some immigrants for some of the time, not more than that.
Moreover – and the MAC analysts should be praised here for doing their job thoroughly, and publishing the results – these types of analyses are subject to a number of critiques. It is important therefore to subject them to a variety of different “robustness” tests – looking at the same data in slightly different ways. Again, we do this, and our results appear quite robust and consistent. But the MAC’s are not. As they say, with commendable honesty, in section A.59 to A67:
“the results are statistically insignificant when outliers are removed from the data”
“our results are not robust to alterations to the model specification”
“our results are not robust to substituting ..the same ratios lagged..this suggests our results may not be robust to endogeneity bias.”
In addition, it also appears that when the period 1995-2009 is used (ie dropping just 2010) again the results become insignificant. In other words, tweak the data just a a little, and the result is no longer statistically significant. Now this does not mean the results are wrong; this may just reflect the weaknesses in the data I highlighted above. But most economists would place rather little weight on estimates which seemed to fail these fairly standard tests.
Finally, I do want to emphasise – returning to my original question of what an informed reader should take from these seemingly contradictory results – that both our report and the MAC’s are important contributions to the debate. The question of what impact immigration has on native British workers, especially the young, is an important one. Economic theory alone does not provide the answer; careful empirical research, and responsible debate, is required. Today’s reports are not the last word, but so far I think the clear balance of the evidence remains that other factors are far more important; immigration probably has no more than a marginal impact. As the Prime Minister has rightly said: "it’s crude and wrong to say immigrants come to Britain to take all our jobs."