I've been busy

It turns out that a Master's degree in Economics is very time consuming - so time consuming I haven't written anything new since I started my course in October 2020. Almost a year later, I've completed the programme - and as part of it, I spent three months working on an independent research project supervised by an academic at my university, UCL. I chose to look at the impact of immigration on robots, and was supervised by Christian Dustmann. If you've read my immigration article, you know I was a fan of his before I even started at UCL. So you can imagine how excited I was to get him as my supervisor!

Despite my best efforts, the 8,000 word paper I wrote is not completely accessible, especially to non-economists. So here's the non-economist's guide to my master's thesis.

(Significant portions of text are lifted directly from my thesis: I am not trying to pass this blog article off as new, original work!)

The Impact of Immigration on Robotisation

Motivation and Policy Relevance

The fear that immigrants may displace native workers, thereby depressing wages and increasing unemployment, is a significant driver of the national debate on immigration in the UK. Two-thirds of Brits think that immigrants take jobs away (European Social Survey, 2014). Rather similar fears are associated with robots, automation and Artificial Intelligence(AI): robots are seen as competitors for jobs currently done by humans, and there are fears that increasing use of robotic technologies could lead to increased unemployment.

While only 7.4% of UK jobs were at high risk of automation, a full 64.9% were at medium risk (ONS, 2017). The employment effects of immigrants and robots are linked because firms optimising production decide between investment in robots, the employment of immigrant or native workers, or a combination of these inputs. An increase in immigration could therefore have different effects on the degree of robotisation. It might be that robots and immigrants are similar inputs, so firms faced with an increase in the number of immigrants might replace robots with immigrant workers, or at least not increase their use of robots. Or it might be that firms use immigrants and robots for different but complementary tasks, so that an increase in the number of immigrants makes the use of robots more profitable, so firms use more robots as a result of immigration.

Policies to restrict immigration are often justified as job-protection for native workers. But if robots are a closer substitute for immigrants than native workers, then restrictions on immigration could hasten robotisation, rather than increasing native employment.

Definitions

I define an immigrant as anyone born outside of the UK, so I include both economic migrants and asylum seekers in my sample.

I define local labour markets as region x industry cells, so for example Agriculture in the South West would be a labour market, distinct from Agriculture in the South East or Car Manufacturing in the South West.

Identifying Causal Effects

Identifying causal effects is hard. It is easy to pick up correlation, which is essentially a statistical term for when two variables co-vary. So if variable x increases, so does variable y. And if variable y decreases, so does variable x. But that doesn't tell us where this correlation comes from: x could cause y, y could cause x, or a third factor, z, could be causing both to vary together. Causality requires correlation, so it's good to start by checking for correlation. But it isn't enough to tell us for sure that a relationship is causal.

In the context of immigration and robots, I could start by comparing local labour market robot numbers and immigrant numbers. If we find more robots in labour markets with more immigrants, then we know that there is correlation between immigration and robotisation. But that doesn't tell us if greater immigration causes greater robotisation, which is what we care about. The obvious reason that this correlation relationship may not be causal is due to a third factor causing both to increase. In this context, one obvious contender is a growing local economy. Immigrants might be more likely to settle where there are plenty of jobs, and firms in a growing local economy might be more likely to buy and install more robots. Economists call this "simultaneity".

To avoid the simultaneity problem, I use a relatively fancy statistical technique called "instrumental variables". There are various reasons that people chose to move to a new country and become immigrants. The reasons that are problematic for me in trying to identify causal effects are those like "plenty of good jobs in London", that might also be correlated to more or less robots in London. So in order to isolate the immigrants who move for reasons unrelated to current market conditions, I use the fact that immigrants tend to settle where previous immigrants from their home country have settled (Bartel 1989). This can be used to create a predicted regional distribution of current immigrants, based on the immigrant settlement patterns from e.g. a decade prior to the year I consider. (I use a decade because it seems fair to think that current regional labour market conditions don't persist for ten years.) So if 10% of Brazilian immigrants in 1986 lived in Outer London, I predict that 10% of Brazilian immigrants in the UK in 1996 will also live in Outer London. I use this predicted immigrant value to isolate the immigrants in my sample who have settled in regions where they would have been predicted to settle based on where other immigrants from their country have settled. The idea is that these immigrants are less likely to be responding to local labour market conditions (that might also affect robot numbers), and more likely to be moving to where their family is, which I can pretty safely assume is unrelated to robot numbers. By using only these immigrants, rather than the full sample, I get rid of those most likely to cause the simultaneity issue. Obviously this is not perfect, but it makes it much more plausible that I am actually picking up a causal effect of immigration on robotisation.

Findings

On average, an increase in the number of immigrants per worker leads to a fall in the number of robots per worker. This supports the idea that immigrants and robots are substitutes, and that firms that have these robots available to them adapt to immigration by using fewer robots, and to a reduced immigrant population by using more robots. Coupled with the previous literature which finds a negligible effect of immigration on native employment and wages in the UK (see e.g. Dustmann, Fabbri and Preston 2005), this suggests that many firms in the industrial sector adapt to immigration through changing their use of technology, specifically robots, rather than by adjusting native employment and wages.

Extensions looking at this effect according to immigrants' occupations or skill levels reveal that some immigrants are more substitutable with robots than others. An increase in low-skilled immigrants per worker leads to a smaller than average fall in robots per worker. An increase in the share of immigrants working in low-skilled occupations leads to a greater than average fall in robots per worker. This supports the hypothesis that immigrants are particularly substitutable with robots due to their concentration in low-skilled occupations. The stronger negative relationship between immigrants in low-skilled occupations rather than whose education suggests they are themselves low-skilled reinforce the evidence of UK immigrants working in jobs for which they are apparently overqualified (Dustmann, Frattini and Preston, 2013) .

The particular substitutability of industrial robots and immigrants working in low-skilled occupations has some interesting policy implications. Certain labour market policies focus on reducing the flow of low-skilled immigrants and those who would work in low-skilled occupations, like the new points-based immigration policy enacted after the UK's withdrawal from the European Union.

My findings suggest that these may lead to large increases in the degree of industrial robotisation, rather than improving the employment conditions of UK natives working in the manufacturing and agricultural sectors.

If you want more

Above is the summary of my work from June-August 2021. I did actually write a full-on paper with tables and graphs and (a little bit of) economic jargon, which can be found here if you're interested!

If you enjoyed the read...

Then why not sign up to my mailing list to get notified when I post anything new?

Back to All Articles