Data on stocks and flows of international migration are necessary to understand migrant patterns and trends and to monitor and evaluate migration-relevant international development agendas. Many countries do not publish data on bilateral migration flows. At least six methods have been proposed recently to estimate bilateral migration flows between all origin-destination country pairs based on migrant stock data published by the World Bank and United Nations. We apply each of these methods to the latest available stock data to provide six estimates of five-year bilateral migration flows between 1990 and 2015. To assess the resulting estimates, we correlate estimates of six migration measures from each method with equivalent reported data where possible. Such systematic efforts at validation have largely been neglected thus far. We show that the correlation between the reported data and the estimates varies widely among different migration measures, over space, and over time. We find that the two methods using a closed demographic accounting approach perform consistently better than the four other estimation approaches.
This paper came out a day before the United Nations Population Division updated their World Population Prospects (WPP). As all the estimates were based on WPP2017 our claim for the estimates being based on the most up to date input data was only true for around 24 hours.
In order to keep the estimates a bit more current I have added new set based on the WPP2019 to the Figshare collection. Below is a quick summary of the changes in the updated estimates to those in the paper.
The plot below shows the relationship between the bilateral flow estimates based on WPP2017 and WPP2019 from each period and estimation method.
For the stock differencing methods there are no changes in the estimates. They do not rely on WPP data. The migration rates approach uses the total absolute net migration data from WPP. All bilateral flows from WPP2019 are slightly higher than their WPP2017 counterparts, though it is barely noticeable in the plot above. The demographic accounting methods use the birth, death and population estimates from the WPP. In each method there are some sizable differences for the flows generated by updated revisions to the WPP demographic data, in particular for those based on the closed demographic accounting approach.
These difference results in revisions to the overall totals of migration flows. Below is an animation of the changes to Figure 2 in the paper
From this plot it is easier to see the changes in the migration rates estimation method. I was surprised that the changes in the estimates were occurring in all periods, not just the most recent period (2010-2015). To investigate I took a look at the changes in the WPP data.
The countries where the largest changes in the bilateral estimates occured (from the closed demographic accounting methods) can be detected by looking at the revisions in net migration between WPP2017 and WPP2019. Net migration is the best measure to track their changes as it correlates perfectly with the net migration in the WPP and is the residual of the input data (births, deaths and population) for the estimates based on the demographic accounting methods. Below are the changes in the complete time series of net migration in nine countries where the largest absolute differences (in any period) between the two WPP versions occur.
At first I was a bit surprised by the scale of the changes. In some periods the revisions to net migration are greater than a million. I dug a little deeper into net migration in previous WPP revisions to find that similar revisions are not unusual. Below are the revisions of absolute net migration between past WPP versions that exceed one million (back to WPP2000, the earliest WPP data I can get my hands on).
The impact of the revision in WPP data on the validation exercise in the paper is minimal. Below is an update of Figure 4 in the paper.
The correlations change by few hundredths of a decimal. These small changes, despite what is shown in the first plot above, are due to the limited amount of reported migration flows statistics (at the global level) to carry out our validation exercise. In the 45 countries that we used (based on the United Nations Population Division collection) the revisions in the WPP data were relatively minor, hence only small changes in their estimates and the correlations with the reported data.
Another month, another update in the input data - this time the UN International Migrant Stock (IMS) data. I have added another set of flow estimates based on the IMS2019 and WPP2019 to the Figshare collection (the original flow estimates in the paper were based on IMS2017). I do not expect there will not be a need to update the estimates again until at least 2021.
I have added a few more plots below to once again give some summaries of the changes in the updated estimates to those in the paper and from the last update.
The plot below shows the relationship between the bilateral flow estimates based on IMS2017 and WPP2017 (as in the paper) and IMS2019 - WPP2019 (this update) from each period and estimation method.
In all methods (columns) there are changes some the estimates, which tend to be larger in more recent periods (lower rows) and estimation methods based on demographic accounting (columns to the right). These patterns are likely due to larger revisions in the most recent stock data and the use of updated demographic data in the demographic accounting methods - not required in the stock differencing approaches.
The revisions to the overall totals of migration flows, shown in Figure 2 in the paper, are animated below, transitioning from 1) the estimates in the paper to 2) the first update of the estimates from changes in the demographic data to 3) the most recent update for changes in the stock data.
The 2010-2015 estimates are, on the whole, suggesting that the total global flows remained at similar levels to 2005-2010. Earlier versions of the estimates had suggested a decline. As a result the crude global migration rate falls by only a small margin for most estimation methods during 2010-2015, except for Pseudo-Bayesian estimation of flows where the rate increases a touch.
The largest changes in the bilateral flow estimates can be partially detected by looking at the revisions in migrant stock data between IMS2017 and IMS2019. In the stock differencing methods these changes are directly related to the change in the estimated flow sizes between the bilateral country pair. In the demogrpahic accounting approaches the impact of the revision is less direct on the estimated flows, as the method allows for return and onwards migration to match changes in migrant stock data. Below is scatter plot of the changes in the IMS data by continent.
There are a few features to note. First, and unsuprisingly, the largest revisions are occuring in the most recent data (2015). Second, the biggest changes are in North American data sources. Below is a table of the bilateral pairs where the absoulte revision to the migrant stocks are greater than 100,000. Third, in some areas there are noticable patterns to the changes - the diagonal lines parrallel to the \(y=x\) line - which I guess is related to updates in the data used to imputation missing bilateral stocks.
The impact of the revision in stock data on the validation exercise in the paper is slightly larger than the previous update, but still not very dramatic. Below is a animated version of Figure 4 in the paper, showing the correlations between the flow estimates and reported data for various migration measures, for the original estimates and subsequent updates based on new WPP and IMS data.
As with the first update, the small changes in the correlations, despite some large revisions in the migrant stock data are due to the limited amount of reported migration flows statistics (at the global level) to carry out our validation exercise.