The NHBS-IDU1 results stand out from RAVEN, Kiwi and HARS data in the high estimated proportion of participants from downtown Seattle residents and in the markedly older age distribution. Lacking a definitive gold standard, we cannot determine with assurance which of these populations, if any, accurately reflects the characteristics of Seattle area IDU. The various data sources in this report have strengths and weaknesses. RAVEN’s random number based sampling reduced volunteer bias, and the multiple recruitment settings reduced the influence of any single site. IDU who had no contact with the institutions where RAVEN recruitment occurred, however, would have been missed. With its jail-based recruitment, Kiwi provided access to the substantial proportion of IDU experiencing incarceration,41
but would have missed IDU who were not arrested. As HIV/AIDS is a reportable infection, HARS data would be expected to identify essentially all persons diagnosed with HIV/AIDS,42
but is likely to be affected by patterns of HIV testing, and to over-sample IDU at higher risk for HIV transmission (such as IDU/MSM) and older IDU. The RDS methodology of NHBS-IDU1 has a body of theory-based investigations asserting its capacity to produce unbiased estimates of population characteristics but this has not been convincingly verified by empirical data.
The closer concordance among the three other sources of data constitutes an argument that the NHBS estimates for these characteristics are the less representative portrayal of Seattle area IDU. We offer the hypothesis that RDS coupon distribution did not effectively penetrate the full universe of injector networks in the Seattle area. The similarity in age and racial distribution between the NHBS-IDU1 population and downtown needle exchangers suggests that NHBS-IDU1 recruitment occurred disproportionately among networks of downtown IDU. Our data on recruitment probabilities within and across groups support the idea of incomplete network penetration in NHBS-IDU1 by documenting lower recruitment probabilities, and hence network barriers, between IDU across differing areas of residence, races and injection drugs.
The criteria generally considered important for valid RDS recruitment appear to have been fulfilled in the Seattle NHBS-IDU1 survey: recruitment chains were long and the preponderance of participants were recruited in the fourth or higher waves, the sample population approached estimated equilibrium values for the characteristics analyzed, recruitment occurred across interviewing sites and few participants reported being recruited by strangers. The numbers of participants in the present report and the number of survey sites are comparable to what has been published in other studies. Our findings raise the question whether the efficient penetration of the injector networks throughout a large metropolitan area might require a wider dispersal of interview sites and larger numbers of participants than has been the practice in RDS studies.
In addition to differing recruiting methods, the four sources of data were conducted over an 11-year time span. The differing patterns of drug preference in the different studies may reflect increasing amphetamine use in the Seattle area over the time period of our data,43
as has been seen in other areas.44
The increasing proportions of Hispanic participants across the studies likely reflects the continuing growth of the Hispanic population in the King County, increasing from 2.9% in 1990 to 6.7% in 2005.45
The higher proportion of females in RAVEN may be a product of a higher likelihood that females take part in the drug treatment programs that were a source of a substantial proportion of RAVEN participants.
Other reports have found discrepancies between RDS-recruited IDU study populations and those recruited by other methods. RDS-derived IDU populations were compared with contemporaneous samples derived from targeted sampling methods in Detroit, Houston and New Orleans, finding no significant difference between the different sample populations in gender or age distributions but differences in racial distributions in Houston and New Orleans.18
An RDS-generated study population of drug users in New York found similar age and gender distributions to those in two previous studies but a different racial makeup, possibly as a result of population changes over time.26
Outside the United States, RDS-recruited IDU study populations were compared to participants recruited by earlier studies by indigenous field worker sampling in Volgograd, Russia and Barnaul, Estonia.15
In both locations, significant differences were found between the sample populations in age, gender, education, and needle sharing. A St. Petersburg study found a substantially higher proportion of females than seen in 2 previous studies.20
Also, a web-based RDS survey of Cornell University students found substantial discrepancies between RDS-adjusted estimates for race and gender proportions and their true distribution.24
Two RDS studies in MSM provide data relevant to our findings. Ma et al. reported on MSM in Beijing surveyed by RDS in three consecutive years. Substantial differences were observed among the surveys in a number of key study variables.13
While these differences may be a product of rapid changes over time, or a change in how network size data was elicited, it is also possible they reflect material variation in repeat samples recruited by RDS methodology. Kendall et al. compared a RDS generated MSM study population in Forteleza, Brazil with previous surveys which used snow ball sampling and venue-based recruitment.12
The RDS population had characteristics, most notably lower SES, less consistent with the other study populations than those populations were with one another. To explain this, the authors note that the club venues at which recruitment occurred might have biased the sample to MSM of higher SES and noted that the RDS sample more closely resembled the census characteristics of Forteleza. On the other hand, they remark that higher SES MSM may have hesitated to travel to the two interview sites in the central city.
The discrepancies among the different sources of data on Seattle area IDU make it difficult to determine the degree to which, if any, accurately reflects the characteristics of the underlying IDU universe. We offer a hypothesis that incomplete penetration of injector networks lies behind the observed differences between the NHBS-IDU1 estimates for age and area of residences and those of the other sources of data. We acknowledge that there are other plausible sources of bias among the sources of data we examined. Given the limited empirical evidence of RDS functioning currently available we would recommend caution in drawing overly broad conclusions from any individual study. Claims that RDS efficiently accesses all but the most isolated networks need to be validated in practice. Further empirical data on RDS functioning in a variety of settings are called for.