The International Passenger Survey
The International Passenger Survey (IPS) is the only comprehensive source of UK international migration statistics. It has been taken in its present form since 1964 and comprehensive tabulations have been published annually by the Office for National Statistics as International Migration since 1974. It is based on interviews with a sample of passengers arriving and departing at UK ports, airports and the channel tunnel. The results are used for a variety of purposes including estimates of tourism and of components of the balance of payments. As noted in the text, the definition used to identify a migrant is a person who is arriving or departing for an intended stay of at least a year, having previously spent at least a year overseas (for arrivals) or in the UK (for departures). The advantages of the IPS are that it measures both immigration and emigration and that it does not depend on visa status. The disadvantage is that it is based on a small sample and is therefore subject to sampling error.
Passengers arriving from and departing to overseas countries are randomly sampled in shifts between 6 a.m. and 11 p.m. To boost the number of migrants in the survey, some shifts proceed to the full interview only when the individual is identified as a migrant. The results are then grossed up to the aggregate using a complex weighting system based on factors that reflect the probability of being sampled and with an allowance for ‘out of hours’ arrivals and departures. To illustrate, in 2001 there were 256,000 interviews of which approximately 1% were identified as migrants. These are about 0.5% of the grossed-up figure for total migrant flows (International Migration 2001, p. 34–5). Sampling errors are calculated for total migration and its components. In 2001 the standard error was 4.1% for the migrant inflow and 5.2% for the migrant outflow. Not surprisingly the standard errors are far larger for sub-aggregates. For inflows, the average standard error for 15 origin regions is 17%, whereas for inflows the average standard error by destination region is 24.4% (International Migration 2001, p. 30). Thus there is a large margin of error in any particular sub-aggregate figure for a single year. Nevertheless, because, by construction, these errors are random it is still valid to use the data as a left-hand side variable in regression analysis.
Non-response to questions is generally quite small (effectively zero for categories such as citizenship and last or next residence). But there are some systematic biases. There are those who say they intend to stay for less than a year but end up staying longer (visitor switchers) and those who say they intend to stay for more than a year but leave earlier. In addition, there are asylum seekers (and others) who often enter (and leave) clandestinely and to that extent are not captured in the survey. Finally, there are migrants to and from the Republic of Ireland who are not surveyed and therefore do not appear in the IPS data. The ONS uses other sources to adjust the total inflows and outflows from the IPS to obtain total gross and net migration figures, but until very recently, those adjustments have not been carried through to the sub-aggregate data. Thus any analysis of sub-aggregate data over the longer term has to be conducted using the unadjusted IPS figures.
For 1999 onwards, certain improvements were made to the system of weighting the IPS responses, and routes to and from Ireland are now included in the survey. But more radical adjustments were made to the data for the 1990s following the 2001 census. The census produced a lower population figure in 2001 than the annual estimates constructed from births, deaths and net migration predicted. Part of this was due to the over correction of the undercount in the previous benchmark, the 1991 census (evidently a result non-response due to the poll tax). But the ONS also believes that net immigration in the 1990s was overestimated and has revised total net immigration downwards by 351,000 for the decade. Most of this revision applies to the adjustments to the IPS (visitor switchers, asylum seekers, etc.) rather than to the IPS itself. But IPS net immigration is also revised down by 12,600 per annum, largely as a result of changes to the weighting adjustments
The recent revisions suggest that IPS inflows were underestimated and outflows were overestimated during the 1990s. But those revisions have not been carried back to the 1980s and earlier, and it is far from clear why the flaws in the survey adjustments should apply only to the 1990s. Given that the survey methodology has not changed, it seems safer to use the data as reported on a consistent basis rather than including the adjustments for the 1990s.