Alphabet soup

Plenty of headlines today for a new report from the always interesting and usually very reliable Resolution Foundation about the impact of migration on wages etc.

Link to RF website here

Readers of Aldous Huxley’s original Brave New World will remember that in an imagined future the World State selectively breeds its population into five castes ranked from Alpha to Epsilon to fulfil different economic roles in a consumption-driven society. This is unironically echoed in typical immigration controls with their tiered visas, points-based systems, shortage occupation lists etc. More ironically for our febrile debate on the best basis for our future relationship with the EU is that the character Bernard, suspected of doubting the benevolence of the World State, is threatened with exile to ….. Iceland.

Back to the report though, which estimates the likely effect of lower migration on ‘native’ earnings by seeing how migration actually affected ‘native’ wages in the past. As it says

This is very difficult to do. Following the main approach pursued in the relevant academic literature we look to see if changes in the share of migrants in an area is related to changes in the wages … for natives in the same area, controlling for a range of other factors that may also have an effect.

It is indeed difficult for all sorts of reasons, both conceptual and data-related, but that’s an issue for another time so lets take the basic methodology at face value for now. This observes how wages and share of migrants actually changed, using data from the Labour Force Survey and the ONS Average Weekly Earnings series, and compares this reality with imagined scenarios of what might have happened if migration had been lower that it actually was. In these imagined scenarios or ‘counterfactuals’ it has to be assumed that individual migrants in the survey data never arrived and so they are dropped from the sample analysed to provide a comparison. However, they were dropped on a random basis. This doesn’t seem the right approach.

Say that in a year where 50,000 high-skilled and 100,000 low-skilled people had actually come to the UK, the counterfactual required migration to have been halved to only 75,000 people in total arriving. Dropping cases randomly will mean that the comparison will be with a scenario where 25,000 high-skilled and 50,000 low-skilled would have arrived. But with greater controls on migration, that isn’t likely to be what would have happened. Instead, low-skilled people are likely to have been excluded to a far higher degree than the high-skilled. Even if the past imaginary controls had not done that, it seems beyond doubt that any likely future controls would do so.

This is really a bit of a fundamental flaw as the random approach will, I think, necessarily bias the results in a downward direction i.e. show a lower impact on the wage of the lower-skilled than would otherwise have been. In reality, any downward pressure on wages is likely to have been exerted more by a stereotypical Polish poultry-plucker rather than by a stereotypical French financier, and absent free movement it is the former who would have been excluded rather than the latter and thus a random basis is not appropriate. Of course it might be that to have got the numbers down sufficiently in the counterfactuals would have meant excluding more than just the lower-skilled, but that just requires dropping all the lower-skilled and then moving step by step up the skill levels until enough cases have been dropped.

There is also the issue of the people who are not covered in the Labour Force Survey. By design it only captures people who are here for more than six months, so if the lady with the LFS clipboard knocks on the door of a house and finds that it contains the stereotypical six seasonal workers she will put her questionnaire away and move on. But as has been recently discovered, there are really quite a lot of such people here, with the ONS examination of HMRC data finding that short-term migration accounts for much of the very large difference between National Insurance Numbers issued and official statistics on the number of long-term migrants.

Link to ONS report here

A lot of short-term circular migration means that a number of jobs are essentially permanently filled by a rotating cast of workers whose presence and of course wages will not be picked up in the Labour Force Survey or other relevant official statistics. They are thus not part of the ‘changing migrant share’ and so any downward pressure they might exert on wages is quite overlooked. This is a ‘confounder’ and obviously likely to be a issue particularly in lower-skilled and poorer-paid occupations.

Generally, people might wonder at the apparent disconnect between consistent research findings that a greatly increased labour supply has made a difference of only a few pennies per hour to pay even in the most affected sectors, yet their employers claim (as Resolution Foundation and many others correctly report) that the very survival of their business is put at risk by the possibility of any diminution in that supply.

One possible explanation is that rather more than a few pennies are involved and so for its use of rather questionable counterfactuals and not taking account of likely confounders with some regret I can’t give this report more than a Gamma ….

P.S. and Updates

The above is not intended as any criticism of the author (who kindly answered my initial questions before I wrote it) or indeed of RF. The methodological issue is something that could have been done differently, and it is just a bit frustrating (to this reader at least) that the opportunity wasn’t taken to model a range of different assumptions for the counterfactual as even if they had all given similar outcomes, that would have given a degree of assurance about the result. As it is, there is at least the possibility that it is very sensitive to the mix of skills or occupation assumed. Whether it is or not is really important for Brexit policy going forward!

As to the potential body of unobserved workers and their wages, that is one of the limitations of all work based on the Labour Force Survey, so this is a cautionary note rather than criticism. As regular readers will know, I regularly suggest that government do more to use its own data to inform itself (especially if they won’t share it with anyone else!)

I’ve also been asked why I didn’t mention the research by Nickell and Salaheen published by the Bank of England a few months back. The answer is that that would have required a rather longer and different post which is on my (long) list of things to do.