A paper published in Demographic Research, ‘Adapting chain referral methods to sample new migrants: possibilities and limitations’ reviewed the experience of attempting to sample two representative populations of new migrants in London. Together with my co-author, Renee Luthra, I considered the lessons from that experience in a blog for openpop.org blog.
In recent research we investigated using Respondent Driven Sampling, a chain referral sampling and analysis technique, to survey newly arrived immigrants to the UK. Newly arrived migrants are a population of particular interest for demographic research, but poorly covered by existing surveys. However, despite the apparent promise of RDS for migration research, we struggled to achieve referrals. This was largely due to lack of well-developed networks for such newcomers.
However, there were differences across our two groups of interest — Poles and Pakistanis. Using an adapted approach, which placed control over contact in the hands of the researcher rather than the respondent, we had some success with reaching Pakistani referrals. We concluded that our adapted researcher-led approach could be promising in sampling more stable and more trusting populations, particularly where intrinsic interest in the survey is likely to be low.
Newly arrived migrants are of particular interest for understanding migration flows, who arrives, who stays, and how migrants adapt over time. To address such questions requires representative samples of new migrant populations. Yet recent immigrants are difficult to capture even in specialist surveys, and particularly where population registers do not offer convenient sampling frames. We realised a different approach would be needed to access a large, representative sample of recent migrants to the UK.
Respondent driven sampling (RDS), potentially offers the opportunity to achieve population-level inference of recently arrived migrant populations, and has begun to be used with some success in certain immigrant studies. We therefore thought it offered potential for achieving representative samples of our two populations of Pakistani and Polish immigrants in London, who had arrived within the previous 18 months.
Nevertheless we judged from the outset that there might be differences between our target populations and those for whom RDS was originally developed (e.g. those at risk of HIV infection). Specifically, we thought that our newly arrived Pakistanis and Poles would be likely to be less-well networked, to show cleavages (e.g. in terms of gender or visa status –for Pakistanis subject to visa restrictions), to have less intrinsic motivation to participate, but on the other hand to be less hidden.
Of these, limited networks alongside some reluctance to refer seemed to present the biggest obstacles in early days of fieldwork. We therefore adapted our chain-referral approach somewhat to adjust to these features of our population. Our biggest adaptation – or innovation – was to develop a researcher-led chain referral approach, collecting information about potential referrals directly from the referrer and putting responsibility for the initiation of contact in the hands of the researcher rather than leaving initiation of contact to the person referred, with the incentivised encouragement of the referrer.
Our researcher-led approach met with some success. Figure 1 shows the increase in referrals after we implemented researcher-led referral in mid 2011. However, we can also see that there was a big difference by group, with Pakistani referrals (the purple line) picking up much more than Polish referrals (the red line). Figure 2 also shows how the resulting chains were then much longer for Pakistanis than Poles.
Figure 1: progress of sample recruitment over survey period: and following introduction of researcher-led approaches in June 2011
Fig 2: length of referral chains
Comparing Pakistanis recruited by ‘traditional RDS’ and to those recruited by our researcher-led approach, we found that, not only did our approach result in an increase in referrals it also enabled us to reach population members with smaller network sizes, showing similar recruiter-recruit transition probabilities across many demographic and social characteristics to the traditional approach. It thus indicated that it could be a promising option for reaching migrant populations – and even recently arrived migrants – in certain circumstances.
Those circumstances are where the population network is less dense than in typical applications of RDS and more fragmented, but there are fewer privacy concerns. Our Pakistani group constituted such a population. These recent arrivals were nearly all on student or family visas, they were thus relatively geographically constrained compared to EU migrants, there was a general willingness to participate and provide contact information, yet the network was strongly fragmented by visa status and interest in the survey was generally low. It is likely that many of the documented recently arrived immigrant groups in Europe will present similar characteristics to the Pakistanis in London, and hence there is potential for further applications of the researcher-led approach.
By contrast, our Polish group was not only more mobile, they also, interestingly, exhibited greater degrees of mistrust. Moreover, with orientations that were often focused on return migration or transnational lives their engagement with the survey was further reduced. Thus referral recruitment continued to work poorly with them when researcher-led. For such populations referral-based approaches are less appropriate.
However, among both groups there was a large proportion with no networks of recently arrived co-nationals. These then would never have been picked up if we had not continued with seed recruitment, even if referrals had been more successful.
If a population can be expected to be reasonably stable, moderately well-connected and reasonably geographically concentrated, it may be worth considering some element of chain referral within the sample, whether researcher-led for less hidden but more disengaged populations or respondent-led for less trusting but more densely connected populations. However, the more densely networked and geographically concentrated such populations are, the more homogenous they are likely to be, with less to tell us about diverse migration flows. If diversity and mobility – and lack of connectedness – of populations are themselves central to the desired sample, it is unlikely that complete reliance on chain-referral methods is advisable. More ad hoc methods may be needed to supplement referral in order to provide more comprehensive coverage.
Platt, L., Luthra, R., Frere-Smith, T. (2015). ‘Adapting chain referral methods to sample new migrants: Possibilities and limitations’. Demographic Research 33:665-700.
Photo credit: Epic Fireworks