Migrants Where do they come from?

Migrants Where do they come from? Where do migrants come from, and where do they go? Politicians make claim and counter-claim, but what is significant immigration, and what is less significant, and how do you map it so that people can tell which is which? Timothy Martyn Hill and Alan Smith needed to do just that. People move from one place to another. How many of them move has always aroused controversy How do you decide how big is big? Specifically, if you are given a stream of numbers, how do you decide which numbers are significant and which ones can be regarded as make-weights and minor? Two of us, working in the Data Visualisation Department of the Office for National Statistics (ONS), were faced with that problem. The Department s job is to produce graphics that communicate complex data in an easily digestible manner because producing statistics is no good if no one can understand them. One of our tasks was to depict internal migration flows people moving homes between different parts of England and Wales. The way to show it was obvious: a map, an interactive one of course, with all the local authority areas marked out on it. Hovering your cursor over an area produced arrows showing migration to or from that area. The width of each arrow indicated the number of people. This worked fine except that no one could see what it was saying. Every local authority has at least one person moving to or from it from every other local authority and there were 348 local authorities in England and Wales. Consequently the arrows were an impossible tangled mess. (Ecologists have encountered much the same problem with seagulls see Significance, October 213.) The solution to this problem was equally obvious: to highlight the significant migration flows and ignore the rest. So we found ourselves faced with the problem we mentioned at the start: how big is big? How to decide which flows were significant and which ones were not, and do it fast enough to enable the results to be displayed on a mobile phone in real time because that is the way that large chunks of our audience want things. Our job, remember, was not to produce data but to communicate it. We could perhaps have shown just the largest flow into each area; but that would give a distorted impression: an area with two almost equal flows would have one of them showing on our map and the other invisible. Or we could have shown flows of say, more than 1 people. But then small areas would seem to have no immigration or emigration at all. Or we could have shown only flows which contributed to more than 3% of the total for that area. Again, this would be arbitrary and give false impressions: why 3%? How different would the map look if we chose 2% or 4%? And if it did look wildly different, then our method is unsatisfactory. We needed more sophisticated mathematics to guide our choice maths that would give a broadly similar picture no matter what arbitrary cut-off choices we made, and no matter what the distribution of flows might be. Happily, back in 1977, two geography professors, John Holmes and Peter Haggett, had considered a variant of this problem 1. They had looked at flows between cells in a network as, effectively, we were doing. They tested several methods for dividing the sheep from the goats, the significant flows from those that were not, as we were having to do. And they had decided which method was best. Unfortunately 1977 was in the pre-digital age. They dealt with ten-cell arrays at the most. We had 348 local authorities, with flows from each one of them to every other, making 12 -odd flows. Consequently our job was to resurrect the Holmes and Haggett paper, and see if their choice still held for today s data sets that can easily run into billions, and then extract and implement its solutions. And we had to do it against a deadline: the Census 211 release date was fast approaching, and those were the data we were wanting to show. 24 february214 214 The Royal Statistical Society

Reproducing Holmes and Haggett Holmes and Haggett had tested three statistical ways of measuring relative significance. We added three more and laid out criteria to judge them: Do they produce a low number of significant flows? A method that says that most flows are significant is useless. Are they resistant to long-tail distortion? In other words, if a lot of very tiny or even zero flows are added to an area, does this make the flows you had before seem more significant? It ought not to. Adding the data that no one moved from London to the Moon, or to Mars, or to Venus, should make no difference to the significance or otherwise of the other movement data; all too often, though, it does. How do the results change if you double each number in the array before analysis? Or if you square them? Again, the picture should remain broadly unchanged. Perhaps bizarrely, all six of our methods failed on the second criterion: adding lots of zero flows changed the results. Fortunately, two of them failed in the right direction adding lots of zero flows reduced the number of significant flows. Holmes and Haggett had concluded that a method using the sample Pearson product moment correlation worked best for the small arrays that they had. And happily, our tests of Pearson against the much larger arrays that we were working with confirmed that their original choice worked best for big data sets as well. For the mathematics, see www. statslife.org.uk/images/pdf/ migrants-where-do-they-comefrom-ref.pdf. So now we could inspect and, as important, we could display significant flows of people within England and Wales. England and Wales Our analysis was originally carried out on 211 internal migration data between the 348 local authorities in England and Wales. The illustrations are based on data for the year to June 212. Expanding London The analysis confirmed something that the ONS already knew: people migrate from London in a predictable way. They move west down the M4 corridor, east towards Essex, or south down the Eurostar line towards Canterbury, as shown in Figure 1. The orange line shows the flows into Canterbury that are significant. These movements tend to be shortrange. The significant immigrations are from its neighbours and from the London area, just 6 miles away. The maps for Slough, just west of London, and Thurrock, just east, are similar. Neighbours People do not move as much as you might think: in 27, 49% of people in the UK lived within 25 miles of their birthplace (see http://www. ipsos-mori.com/researchpublications/researcharchive/213/ Over-Half-The-Population-Cant- Name-Any-Of-Their-GreatGrandparents.aspx). The analysis showed that many areas import and export significant numbers of people only to and from their nearest neighbours, no matter how prosperous or otherwise they may be. Birmingham (Figure 2) receives significant migration only from its next-door neighbours, even though it is the UK s second city; similarly, people in Grimsby, Liverpool and Oxford significantly migrate Figure 1. Migration to Canterbury. Significant migrations are in orange Figure 2. Migration to Birmingham february214 25

Figure 3. Migration to Cornwall Figure 4. Migration from the Isle of Wight only to their next-door neighbours. Perhaps more surprisingly, despite its reputation as the retirement destination of choice for England and Wales, Cornwall s sole significant source of people (Figure 3) is its neighbour Plymouth. categories, which might well show variability within this pattern. It is for the movements of people overall. 5 From the point of view of the ONS, this was all that was required. But the Pearson algorithm remained and was not tied to a graphical Large, sparsely populated local authorities Consider the larger but sparsely populated local authorities. You get lots of flows, but all of the flows are actually pretty small. What would you expect to see? With no obvious favourites, our method throws up lots of small significant flows. The Isle of Wight exhibits this phenomenon, with significant migration from it to many areas, especially the south coast (see Figure 4). Powys in Wales and Wiltshire in the west country show similar patterns. The general conclusion is, that broadly speaking, when you move, you will probably move somewhere close regardless of where you start from. If you live in a well-populated district you will probably move to one of a few favoured nearby areas, but if you live in a large sparsely populated area then there s no telling where you will move to. Our analysis did not consider age groups, or employment 4 3 2 1-1 Figure 5. Migration into the UK, 1975 211, from the EU15 countries those that belonged to the European Union prior to 24. The balance has almost has always been positive; red bars are significant peaks 26 february214

45 4 35 3 25 2 15 1 5 solid-istanbul/istock/thinkstock Figure 6. EU15 residents in the UK, by country of birth use: you could use it on any series of numbers to decide which was significant. So we used it to analyse migration into the United Kingdom United Kingdom Details of movement into and out of the UK, and foreign-born residents in the UK, were extracted from the International Passenger Survey (IPS) and other data published by the ONS. The IPS data are not an exact match to the Long-Term International Migration data published by the ONS but they are similar and the IPS data could be joined together and extended back to 1975. The data were split into the groupings of countries used by the ONS and the countries/regions of the UK (except Northern Ireland), and then analysed for 1975 211 using the Pearson algorithm. EU15 The EU15 are the fourteen countries that belonged to the European Union (excluding the 7 6 UK) prior to 24. They are Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain and Sweden. For the avoidance of doubt, the ONS does not include the UK in these figures. The balance of migration into the UK from the EU15 has been positive (more have been entering than leaving) since 1975, with the exception of the early 198s. It started to take off in the mid-199s, peaked in the late 199s, then peaked again in the mid-2s. Figure 5 shows in red the significant years of migration according to the Pearson algorithm. The areas of the country significantly affected where most of these immigrants and their predecessors settled were London and the South East. Figure 6 shows that significant numbers of EU15-born residents in the UK are from Germany and the Republic of Ireland. 5 EU8 4 3 2 1 Slovenia Estonia Czech Republic Hungary Figure 7. EU8 residents in the UK, by country of birth Slovakia Latvia Lithuania Poland The EU8 are the eight countries of the EU that joined in 24 from what used to be popularly referred to as the Eastern bloc. They are Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovakia and Slovenia. Prior to that they were included in the other foreign category, so our analysis only covers the years since 24. Unlike the EU15 figures, migration to the UK from these countries is more diffuse and less concentrated. It was significant for every year between 24 and 211, with the exception of the recession of 28 and 29, when february214 27

25 2 15 1 5 New Zealand Canada Australia South Africa Figure 8. Old Commonwealth residents of the UK, by country of birth Commonwealth, old and new Immigration from the Commonwealth is also a live political topic. We can compare the old Commonwealth Canada, Australia, New Zealand and South Africa with the new. The significant country of origin for old Commonwealth-born residents in the UK is South Africa (see Figure 8). The balance of migration from the new Commonwealth the remaining commonwealth countries, though with some anomalies has been positive since 1975 and was reasonably level until the mid-199s, when it started to rise, and has done so ever since. Again the red bars in Figure 9 show significance according to the Pearson algorithm. The significant countries for new Commonwealth-born residents in the UK are India and Pakistan (Figure 1). it dropped off dramatically. Similarly, whereas EU15 migrants were concentrated in London and the South East, EU8 migrants are spread evenly across the country, with the exception of Wales and the North East. However, significant numbers of EU8-born residents in the UK are from one nation: Poland. Four countries Malta, Cyprus, Bulgaria and Romania are members of the EU but are not included in the EU15 or EU8. For completeness, we looked at their figures also. Our analysis covers only the years since 24. Among them, there is significant migration to Britain only from one nation: Romania. Other foreign The balance of migration from the rest of the world was at a low level until the mid-199s, when it suddenly jumped, and has remained high ever since. Significant countries of birth are more widespread: they are Iran, Iraq, Hong Kong, Turkey, Somalia, China, the Philippines and the USA (Figure 11). 14 12 1 8 6 4 Overall Having discussed variation within various categories, we then inspected them grouped together. The pattern by year was unsurprising: positive and at a low level until the mid-199s, then a sudden jump, and remaining high ever since. Similarly, the significant area of immigration was unsurprisingly London. What did come as a surprise was the categories from which significant immigration originated: other foreign and new Commonwealth outweighed the EU categories even combined, and dwarfed the old Commonwealth category. The countries that attract the most attention in debates on migration are not necessarily the most significant. 2 Figure 9. New Commonwealth migration into the UK, by year, 1975 211 Conclusion So we had made for ourselves a powerful graphical tool for flow maps. Since we can animate flow maps, this means we can identify significance and display meaningful information for billions of data points in a intuitively 28 february214

comprehensible manner and in seconds. Even five years ago, such feats were beyond us. Even better, we can generalise this tool to analyse any given series of numbers and apply it to any data. Since the algorithm is easily reproducible, the public can use it in turn to check the pronouncements of politicians and the media. Judgements should be based on accurate knowledge and data, and we hope our statistical tool will help towards this end. 8 7 6 5 4 3 References 1. Holmes, J. H. and Haggett, P. (1977) Graph theory interpretation of flow matrices: A note on maximisation procedures for identifying significant links. Geographical Analysis, 9(4), 388 399. 2 1 Timothy Martyn Hill was, and Alan Smith still is, in the Data Visualisation Department of the Office for National Statistics. Figure 1. New Commonwealth residents in the UK, by country of birth 2 18 16 14 12 1 8 6 4 2 Figure 11. Other foreign residents in the UK, by country of birth february214 29