JOURNAL OF SEX RESEARCH, 52(4), 396–411, 2015

Copyright # The Society for the Scientific Study of Sexuality

ISSN: 0022-4499 print=1559-8519 online
DOI: 10.1080/00224499.2015.1016476

ANNUAL REVIEW OF SEX RESEARCH SPECIAL ISSUE

Nonprobability and Probability-Based Sampling Strategies in Sexual Science

Joseph A. Catania and M. Margaret Dolcini
Hallie E. Ford Center for Healthy Children and Families, School of Social and Behavioral
Health Sciences, College of Public Health and Human Sciences, Oregon State University
Roberto Orellana
School of Social Work, Portland State University
Vasudah Narayanan
Westat Corporation

With few exceptions, much of sexual science builds upon data from opportunistic
nonprobability samples of limited generalizability. Although probability-based studies are
considered the gold standard in terms of generalizability, they are costly to apply to many
of the hard-to-reach populations of interest to sexologists. The present article discusses recent
conclusions by sampling experts that have relevance to sexual science that advocates for
nonprobability methods. In this regard, we provide an overview of Internet sampling as a use-
ful, cost-efficient, nonprobability sampling method of value to sex researchers conducting
modeling work or clinical trials. We also argue that probability-based sampling methods
may be more readily applied in sex research with hard-to-reach populations than is typically
thought. In this context, we provide three case studies that utilize qualitative and quantitative
techniques directed at reducing limitations in applying probability-based sampling to hard-to-
reach populations: indigenous Peruvians, African American youth, and urban men who have
sex with men (MSM). Recommendations are made with regard to presampling studies,
adaptive and disproportionate sampling methods, and strategies that may be utilized in
evaluating nonprobability and probability-based sampling methods.

Observational, experimental, and clinical studies in the Public Opinion Researchers (AAPOR) (AAPOR et al.,
field of human sexuality rely heavily on opportunistic 2013; Baker et al., 2013). We provide an overview of
nonprobability samples. Although data from such Internet sampling as a nonprobability sampling method
samples have limited generalizability, nonprobability of value to sex researchers. In addition, we provide three
sampling can be a relatively low-cost, efficient method case studies that utilized both qualitative and quantitative
of gathering human sexuality data. Probability-based techniques directed at reducing limitations in probability-
sampling is often considered the gold standard for recruit- based sampling of hard-to-reach populations: low
ing representative samples and generating results that are socioeconomic status (SES) ethnic minority households,
generalizable. However, probability-based sampling can indigenous persons in the Amazon, and urban men who
be cost prohibitive, particularly in sampling hard-to-reach have sex with men (MSM).
and rare populations. The present article considers these
observations in the context of recent recommendations
AAPOR: Sampling Recommendations and
by sampling experts from the American Association of
Commentary
Correspondence should be addressed to Joseph A. Catania, Hallie
Nonprobability-based sampling methods were
E. Ford Center for Healthy Children and Families, School of Social
and Behavioral Health Sciences, College of Public Health and Human recently reviewed by a task force of sampling specialists
Sciences, 2631 SW Campus Way, Oregon State University, Corvallis, from AAPOR. A summary report with additional com-
OR 97330. E-mail: catania1951@comcast.net mentary is provided in the Journal of Survey Statistics
 SAMPLING STRATEGIES IN SEXUAL SCIENCE

and Methodology (Baker et al., 2013); the full report is Evaluating Sample Quality
available from AAPOR and colleagues (2013). The task
Strategies for judging sample quality are not uniform.
force had a number of observations relevant to the field
A key concept in judging sample quality is ‘‘fit for
of sexual science, which we summarize and comment
purpose’’ (AAPOR et al., 2013). That is, does the
on here.
sampling method fit the purpose of the study proposed
by the investigators? Moreover, quality should not be
Nonprobability-Based Samples judged by cooperation rates alone; other external bench-
marks may be used to judge quality (e.g., Vallilant, 2013).
Nonprobability sampling, based on newer approaches Two broad categories of users have been described. They
(e.g., sample matching methods; see Baker et al., 2013), differ in their research purposes and, consequently, in
may provide modelers with valid data for examining the their sampling needs (AAPOR et al., 2013).
relationships among variables without pretense to gener- Modelers. Modelers could achieve the goals of
alizability. Internet sampling is an extremely cost- modeling theoretical relationships with data from non-
efficient sampling method in this regard. An important probability samples. Modelers need statistical power
consideration when using ‘‘opportunistic’’ data is how and variable distributions that approximate those at the
well variables are distributed. In some cases, opportun- population level.
istic samples may misrepresent ‘‘true’’ distributions and,
therefore, may misidentify relationships among Subscribers. Subscribers, such as government agen-
variables. This is shown in work by Fendrich, Avci, cies and departments of public health, require highly
Johnson, and Mackesy-Amiti (2013), who illustrated accurate data generated from probability-based samples
that household probability samples show significant at the population level (e.g., neighborhood, city, nation).
relationships between mental health and sexual Departments of public health, for example, need to
health variables that are found to be nonsignificant know the ‘‘true’’ prevalence of a sexual health problem
in studies based on opportunistic samples (see also (e.g., STIs) and the distribution of the problem in the
Mackesy-Amiti, Fendrich, and Johnson, 2010). When population to more efficiently prioritize limited
employing opportunistic samples, caution must also be resources for surveillance, treatment, and prevention.
used in interpreting differences among groups. For Indexing sample quality in probability samples typically
example, gender differences in studies based on oppor- relies on standardized statistical indices (e.g., cooperation
tunistic samples may reflect differences in sample repre- rates). There is a challenge in indexing quality for prob-
sentativeness rather than true gender differences (e.g., if ability samples of some hard-to-reach populations (e.g.,
men and women in the opportunistic sample dispropor- MSM). When these population segments are not adequa-
tionately derive from different social classes). A value of tely represented in census data, one loses the ability to
Internet samples in addressing these issues is that data employ poststratification weighting that adjusts for differ-
collection can be cost-effectively monitored and distri- ences between the sample and census demographic data. In
butions tailored to the investigators’ needs. these cases, other external indicators may be useful.
Quality indicators for nonprobability samples
typically include external indices such as health or
Probability-Based Samples
behavioral data that, based on legal-regulatory require-
Probability-based samples offer the strongest approach ments, are ‘‘universally’’ collected. In the field of sexual
to making population estimates, which in the field of health, comparisons of sample data to such public
sexual health may include estimates of, for instance, sexual health measures offer an index of quality (e.g., all repor-
dysfunctions, sexual trauma, sexually transmitted infection table STIs=HIV for an area). Although public health
(STI), reproductive health, clinical help seeking, and their reporting systems have sources of error (as does the
antecedents. Thus, probability-based samples are impor- U.S. Census), given sufficient resources and motivation
tant from a public health perspective for making accurate to support the reporting process, they may be taken as
prevalence estimates but may not be cost effective when reasonably accurate. Another method of judging sample
testing theories or developing new clinical treatments. quality is to test alternative explanations for the sam-
Opportunistic samples, on the other hand, are problematic ple’s demographic profile. A third strategy is based on
for making population-level estimates of sexual health out- predictive validity. For instance, in the field of political
comes. In comparing different opportunistic sample polling, the comparison of estimates from nonprobabil-
frames, Zwahlen and colleagues (2007) found STI rates ity samples to predict election outcomes, a known out-
both decreasing and increasing over time depending on come, provides a measure of sample quality (e.g.,
which sample was selected (see also Guo et al., 2011). Vallilant, 2013). If data from a nonprobability sample
Meyer and Colten (1999) compared opportunistic and can be used to predict future STI rates, then the sample
probability-based samples of MSM and found significant may be inferred to be of good quality.
differences, including an underrepresentation of closeted In the following sections, we first discuss Internet sam-
men in opportunistic samples. pling, a method that has enormous potential for conducting

397
 CATANIA, DOLCINI, ORELLANA, AND NARAYANAN

sex research. Then we discuss three case studies of hard-to- participants self-select, reducing representativeness, and
reach populations that illustrate, among other strategies, it is challenging to validate whether respondents are who
the importance of conducting presampling qualitative they say they are (e.g., a minor pretending to be an
research at the community level. These case studies illus- adult). Population penetration is also limited with
trate solutions that can be brought to bear in executing regard to in-home Internet access among some demo-
highly challenging probability surveys. In discussing graphic groups. Nevertheless, if representativeness is
Internet sampling and our case studies, attention is paid not the primary goal and the focus is statistical modeling
to strategies for judging sample quality. of theoretical relationships, then Internet sampling may
provide investigators with a significant alternative strat-
egy to sampling students and patients.
Internet Sampling
Household Internet Access: United States
Historically, the AIDS epidemic and reproductive
health concerns have led to funding for studies using The Current Population Survey, a nationally rep-
probability-based sampling methods (e.g., National resentative survey of the United States, found that
Survey of Family Growth, National Health and Social 69% of households in 2012 had home access to the Inter-
Life Survey, National Sexual Health Study, Add Health net (United States Census Bureau, 2012). However,
Survey). However, the bulk of sex research consists of home Internet access is not uniform across demographic
studies or experiments that rely on lower-budget groups (U.S. Census Bureau, 2012). For instance,
nonprobability samples. College students and medical= among adults 65 years and older, only 48% have home
mental health clinic samples are often employed in sex Internet access (versus a high of 79% for adults aged
research. However, the growth of the Internet and, sub- 35 to 44 years). Among ethnic groups, Hispanics (any
sequently, Internet sampling have led to another means race) have the lowest access (55% versus non-Hispanic
of generating large nonprobability samples with which Whites, 75%; Asians, 75%; Blacks, 58%). In terms of
to conduct sex research. A number of studies have social class, households with incomes below $25,000
been conducted in the field of sexual health using per year and persons with low education levels have lim-
Internet samples (e.g., Alessi & Martin, 2010; Barresi ited home access (45% versus 90% for those earning
et al., 2010; Evans, Wiggins, Mercer, Bolding, & $150,000= per year; less than a high school education,
Elford, 2007; Feldacker, Torrone, Triplette, Smith, & 30% versus a bachelor of art degree or higher, 90%).
Leone, 2011; Pequegnat et al., 2006; Ross, Tikkanen, Regional differences are less extreme (South 66% access
& Månsson, 2000; Whitehead, 2007). to Northeast 72% access), but some states (Mississippi
Here we discuss strategies for conducting Internet and Louisiana) have particularly low home access rates
surveys and explore their utility in conducting modeling (<59%). Investigators should be aware of these gaps in
work. It should be noted at the outset that the Internet household Internet access, which are changing over
can be used both as a strategy for sampling and as a time. In the United States, the Current Population Sur-
mode of data collection. As a data collection mode, vey is conducted every two years between census years
for instance, it can be appended to probability-based and is used to estimate household Internet coverage.
samples in which data are collected in one of two ways: Household coverage is the relevant estimate. Work
(1) initially through a mailing address–based or tele- addresses should be removed from the sample to avoid
phone (e.g., random-digit dial) survey followed by a double coverage (obtaining interviews from the same
Web-based survey to collect additional data at a rela- person at work and home) if possible. Moreover,
tively lower cost or (2) using a Web-based survey as privacy and employment risk problems may arise for
the main mode of data collection. We are not discussing participants when surveys are completed at work.
computer-based interviews as a data collection mode but
are focusing on the Internet as a sampling tool. In
Strategies for Internet Sampling
addition, it is important to note that Internet sampling
is not a single method. The methods described here Internet sampling typically refers to recruiting study
are meant only as a guide and are in no way prescriptive. participants from among members of online=Internet
The Internet has a number of clear sampling panels, visitors to online forums, social networking sites,
advantages: (a) it is inexpensive; (b) it can generate non- and chat rooms.
representative national, regional, or local level data; and One of the methods gaining popularity is the use of
(c) it can be used to focus on specific subpopulations. Internet panels. These panels are maintained by compa-
Further, data may be obtained in a very short period nies that specialize in recruiting individuals to partici-
of time; data flows can be monitored in something close pate in such panels. Companies that maintain this type
to real time; and responses can be processed through of panel, such as GfK or SurveyMonkey, have pro-
data management software that will build the data set prietary methods for recruiting and maintaining the
quickly and efficiently. The primary disadvantage is that panels. The main attraction to use such panels is time

398
 SAMPLING STRATEGIES IN SEXUAL SCIENCE

and lower cost. Surveys using these panels can be But the guidance provided here should aid investi-
mounted within a week. gators in gathering the information necessary to
understand the drivers of cost.

Investigator-Generated Internet Samples

Additional Internet Survey Considerations
Study participants can also be recruited through a
Selected sites can be used to access hard-to-reach
variety of other methods, including by placing banner
populations. For example, Gay.com sees more than
advertisements on select Web sites (often referred to as
326,000 unique visitors daily and more than 5.1 million
River sampling), placing recruiters in chat rooms, or
visitors in a month (Alexa Internet Inc., 2014). This level
posts on social networking sites. This sample frame
of access dwarfs sampling from commercial listings of
is otherwise referred to as an opt-in sample. In most
gay men or standard community venue sampling (e.g.,
cases, study participants elect to enroll in a study in
bars, clubs). In general, because the Internet is the most
response to an online advertisement, hence opting in
common and geographically extensive venue for sexual
to participate. This kind of opt-in Internet sample is
information and sex negotiations (Pequegnat et al.,
inherently nonprobabilistic. In general, these surveys
2006), it seems to be a natural corollary to use the Inter-
follow several generic steps:
net to sample sexually diverse population segments.
The Internet also opens the possibility of targeting
. Define the target audience in terms of, for instance, particular population characteristics such as geographic
social demographic characteristics, sexual orien- residence, age group, and gender. Important sociodemo-
tation, and geographic boundaries (e.g., national graphic and geographic variations can be studied that
versus local). These characteristics help researchers otherwise may not be accessible to the investigator
determine the eligibility criteria for study parti- through local or state-level venues. Many online adver-
cipants, Internet sites to target for recruitment, tising networks provide researchers the advantage of
and recruitment methods. For example, if the sam- reaching multiple Web sites without the difficulties of
ple is restricted to MSM, 18 to 34 years old, and a dealing with each site individually. These ad networks
few zip codes, it might be efficient to go through an can target the ad to be visible in selected zip codes and
online advertising network. These ad networks can cap the number of impressions seen by a single Internet
target the ad to be visible in select zip codes. protocol (IP) address, thereby avoiding oversaturating
. Determine the unit of analysis: Is the intention to one potential participant and missing other potential
produce one estimate for the entire sample, or does participants. In this regard, banner advertisement
the researcher wish to compare between geo- placement has become extremely sophisticated. The
graphic areas, demographic groups, or other placement can be targeted by geographic area and time
characteristics? For instance, if the sample is of the day; this placement can be changed within
national (e.g., across all 50 States and the District hours. The ads can also be changed, if necessary, to
of Columbia), does the researcher want one make midcourse corrections. These networks can pro-
national estimate? Or does the researcher wish to vide researchers with data such as click-through rates
compare among states and hence needs an estimate by, for instance, advertisement type, site, and zip code.
for each state and the capital district? Hence, researchers can use real-time statistics to fine-
. Definition of the unit of analysis has a direct bear- tune the sample as it is being collected.
ing on sample size. As the units of analysis Finally, a number of strategies may be employed to
increase, the sample size also increases. And there limit extreme sampling biases. Bias can be mitigated
is a tension among analytic needs, sample size, by (a) varying times of the day when banner ads are
and cost. So the researcher might decide that while placed on a site that will increase demographic diversity
an urban–rural comparison for each state would be (Pequegnat et al., 2006) and (b) recruiting large samples
interesting, it might only be possible to conduct an with geographic diversity. For instance, those who work
urban–rural analysis at the national level. This during the day may visit Web sites only in the evenings.
calls for a rough draft of the analysis plan before However, those who are part-time or night-shift workers
the sample design is finalized. This may seem to or unemployed may be online more frequently during
be an obvious statement, but many studies encoun- daytime hours. Varying the times of placing banner
ter rough waters because the analysis plan is not ads, therefore, will help in recruiting both employed
closely tied to the sample design. and unemployed persons.
. Although the unit of analysis is an important
driver of sample size, an equally important driver
Cost Comparisons
is available funds. Costs vary depending on sam-
pling method, choice of panel, banner ads on spe- The relative cost reduction for an Internet survey
cific sites, use of an advertising network, and so on. versus the least expensive household probability

399
 CATANIA, DOLCINI, ORELLANA, AND NARAYANAN

methods (telephone, address-based surveys) is difficult Quality Data from Internet Sampling
to estimate, but generally the cost is considered to be
As noted previously, sampling quality can be cast
substantially lower, depending on a number of factors
into a broad framework conceptualized as ‘‘fit for pur-
(de Leeuw, 2010). The differences between address-based
pose’’ (AAPOR et al., 2013). In this regard, two broad
and Internet surveys begin to shrink if the Internet survey
categories of users—modelers and subscribers—have
includes a significant amount of new programming for
been described. Nonprobability sampling that is
managing and monitoring the survey and survey results.
inexpensive, allows for access to wide variations in
However, programming Web surveys, even complex sur-
respondent characteristics, and can generate data very
veys, requires a low level of effort if standard commer-
quickly may fit the purpose of many modelers in the
cially available methods are used. Moreover, standard
field of human sexuality research. Modelers need stat-
Web survey software is sophisticated enough to handle
istical power and variable distributions that approxi-
most complex surveys. The cost of data collection includes
mate those at the population level. Further, borrowing
the amount of staff time to monitor data collection
from qualitative methodologists, it is possible to reach
and possibly the cost of hosting the survey on a server.
saturation on participant characteristics using nonprob-
The latter cost is usually minimal. Other cost factors for
ability samples. That is, most variants of a phenomenon
consideration are listed in the following sections.
may be represented in the data set, even though their
prevalence levels may not be those of the general popu-
Target population. For hard-to-reach populations lation. That is, an investigator, using Internet sampling,
Internet surveys are substantially less costly (ignoring can study most or all variants on a phenomenon while
data quality differences) than telephone or address- not being able to make prevalence estimates. The wider
based surveys. Screening costs (see discussion on Urban the sampling net, the better able the investigator will be
Men’s Health Study) to identify a low-prevalence to examine a full range of variants.
population (e.g., MSM of color) drive the costs for Statistically, however, investigators may be chal-
telephone or address-based surveys. In contrast, because lenged by variable distributions that do not represent
we know, for instance, the Web sites that MSM visit, it their true distribution in the population. For instance,
is highly productive and cost efficient to place banner if sexual satisfaction is normally distributed in the
ads on sites frequented by MSM of color. The most population, and our sample yields a highly skewed
efficient method is to go through established online distribution toward those who are unsatisfied, then cor-
advertising services, for example, the Gay Ad Network relations between independent variables and satisfaction
(GAN). Services like GAN reach more than 300 sites scores may be attenuated or nonsignificant; alterna-
frequented by MSM. tively, some variables may have much larger correlations
than would be observed in a representative sample.
Sample size. In telephone and address-based sur- Relationships between independent and dependent vari-
veys, sample size affects subsequent costs such as those ables may then be misidentified if variable distributions
for data collection and data processing. Web surveys are heavily distorted. If you can compare the Internet
are not sensitive to sample sizes. For example, the server data to data from a probability sample, then it is poss-
costs will not be any different for a survey sample of 100 ible not only to describe the limitations of the modeling
versus a sample of 5,000; server costs may register once work but also to adjust variables statistically to better
you hit 50,000 surveys. approximate their true distributions.
A study in Great Britain examined the distributions
Data processing. In either a Web survey or a of sexual health variables collected in an Internet sample
computer-assisted telephone interview, much of the data versus a probability-based sample (Evans et al., 2007).
entry and cleaning is programmed in the survey or inter- Evans and colleagues (2007) found few variable differ-
view. Mailing address–based surveys require a higher ences between study samples (MSM, 18 to 44 years
level of effort for data preparation and cleaning. Even old) with the exception of age, student status, urban
if the survey questionnaire is set up for scanning, every residence, and general health (i.e., Internet participants
return has to be manually reviewed for margin notes were more likely to be younger, urban students in better
to determine if they have a bearing on the response. health than the general population). A few sexual
Most importantly, skip logic and out-of-range responses variables reached statistical significance. For instance,
cannot be programmed for paper questionnaires the proportions reporting anal intercourse (76.9%
requiring manual checking. versus 63.3%) were significantly different but were not
Although it is not possible to simply provide a so large as to be worrisome (13% difference). Internet
spreadsheet with a rate calculator, an informed respondents also reported significantly more STIs
decision based on the types of factors discussed pre- (16.9% versus 4.8%). This finding is important, parti-
viously and a cost comparison of vendors will yield clear cularly for studies that wish to oversample people with
cost options. STIs, in suggesting there may be a significant cost

400
 SAMPLING STRATEGIES IN SEXUAL SCIENCE

benefit in using Internet sampling. A recent meta- Ethical Concerns with Internet Sampling
analysis comparing opportunistic samples for online
Protecting the privacy and confidentiality of Internet
versus offline studies of MSM found that online surveys
participants is a common concern (Curtis, 2014). In gen-
obtained higher prevalence levels of unprotected anal
eral, the entire pipeline from investigator to respondent
intercourse among MSM (33.9% versus 24.9%; Yang,
and back again needs to be secure. One strategy is to
Zhang, Dong, Jin, & Han, 2014), which is consistent
initially recruit potential participants using less secure
with Evans and colleagues’ (2007) findings. These
commercial methods with a link to the investigators’
findings suggest that higher-risk MSM more consistently
site; at this juncture potential participants are provided
respond to surveys online, and this is true across
informed consent and, if eligible and willing to partici-
European, Asian, and U.S. studies.
pate, are given an additional Internet address and
In reviewing the literature based on Internet samples,
password providing entry to the investigators’ secure
it is important to keep in mind that the results are not
server. All survey questions are completed on the secure
generalizable and not reliable in the way that results
server. The survey and participants’ responses never
from probability-based samples are. Thus, because one
reside on the participants’ computers. Data are
investigator found high STI rates in an Internet sample
encrypted. Using this type of strategy coupled to the less
does not mean that the next Internet sample frame,
secure commercial recruitment routes has substantial
even with the same selection characteristics, will obtain
advantages that, if done correctly, may reduce ethical
the same result. However, given the ability to monitor
problems. An ongoing problem is knowing whether
Internet data in real time and adjust sampling rates,
respondents are appropriately eligible, particularly if
one can achieve oversamples much more efficiently
they are below the age of consent. However, this prob-
and at lower cost relative to probability-based sampling,
lem has parallels with those encountered in conducting
commercial-venue sampling, or other such strategies.
telephone surveys and may provide only minimal risk
Moreover, Internet sampling can be used for almost
for participants, depending on the nature of the survey.
any kind of study, from qualitative studies to recruiting
for clinical trials.
A number of factors might reduce the quality of an Hard-to-Reach Populations: Probability-Based
Internet sample. It is difficult to control, for instance, Sampling
individuals from opting in multiple times to obtain
incentives through different computer addresses. Internet sampling is the newest entrant into the field of
Further, as noted previously, the individual may falsify opportunistic sampling that begins to address the issue of
responses to meet eligibility requirements. Some online obtaining large samples at low costs. Moreover, the
panels have built-in verification processes to guard Internet has been used successfully to access hard-to-reach
against the multiple opt-ins from one individual. Many populations (e.g., MSM), as well as rare populations (e.g.,
of the verification methods are, however, proprietary men with Peyronie’s disease), and garner large samples
and not available in the public domain. Resources for analytic purposes. However, gains have also been
permitting, researchers may opt for telephone contact made over the past two decades in employing household
with the recruited sample to verify the identity of the probability-based sampling methods to survey hard-to-
individual. reach populations. Although substantially more costly
than Internet sampling, these methods have begun to
evolve from a common set of approaches, to surmount-
Targeting Hard-to-Reach and Special Populations: ing the social ecological barriers (e.g., mistrust of
Internet Sampling outsiders), to surveying those who live in bounded com-
The social media and social support dimensions of munities (e.g., gay neighborhoods, African American
the Internet allow investigators to sample a variety neighborhoods, isolated rural communities). In this
of populations that might be considered challenging or regard, we next present three case studies that highlight
too costly without substantial funding. For instance, the common methods employed in sampling hard-to-
Rosen and colleagues (2008) used Internet recruitment reach populations and noting variations in approaches
to obtain samples of men with Peyronie’s disease. and ability to verify sample quality. We begin with
Similarly, other investigators have successfully used a presentation of work conducted in the Amazon rain-
the Internet to recruit large samples of MSM and con- forest. We then discuss sampling of African American
duct rapid-intervention evaluations (Feldacker et al., youth. Finally, we examine a survey of urban MSM.
2011; Pequegnat et al., 2006; Ross et al., 2000). Online
advertising networks and Web sites that serve sexual
minorities and those with sexual health problems make Sampling in the Peruvian Amazon
it relatively easy to sample these population segments.
Identifying such groups through a probability sample Sampling in rural settings presents challenges often
would, in contrast, be time and resource intensive. not encountered in many urban settings. Most sexual

401
 CATANIA, DOLCINI, ORELLANA, AND NARAYANAN

health research with indigenous populations throughout from Lima. In these port cities, new migrants join a
Latin America has relied on purposive sampling techni- highly diverse indigenous population that inhabits more
ques because many indigenous villages, especially in than 1,786 distinct villages throughout the Amazon
remote regions, lack street names or household numeri- basin (Instituto Nacional de Estadı́stica e Informatica,
cal identifiers, telephone service, or other forms of tra- 2008). In Peru, indigenous people are estimated to
ditional identifiers often used in probabilistic sampling. account for up to 45% of the population (Montenegro
However, the urgent need to estimate and examine the & Stephens, 2006). In the Peruvian Amazon region there
burden of disease among underserved populations in are 13 linguistic families and more than 60 ethnic groups
the developing world should compel investigators to (Instituto Nacional de Estadı́stica e Informatica, 2008)
develop and fine-tune innovative ways of obtaining spread over a large geographic area. Prior research indi-
representative samples from settings where health data cates that indigenous people remain some of the most
are difficult to obtain. marginalized populations in the Americas, with extreme
As part of a large population-based study of STIs in poverty, low educational attainment, and poor access to
Peru (Garcı́a et al., 2012), a team of investigators was health care and social services (Auerbach, Parkhurst, &
presented with the opportunity to partner with indigen- Cáceres, 2011).
ous communities in the Peruvian Amazon to conduct a Early in the sample development process it became
sexual health survey and HIV=STI counseling and test- apparent that prior research in the Amazon did not pro-
ing. This was an opportunity to ascertain HIV=STI vide a proven protocol for engaging with indigenous
prevalence data, as well as social and behavioral infor- communities in a health research context. To address
mation from a historically underserved population in this latter problem, investigators initiated rapid ethno-
Peru. The motivation for this work was based on obser- graphic research based on principles of participatory
vations by earlier HIV=STI studies in Peru (see Garcı́a action research (Minkler, 2000, 2005). This approach
et al., 2012) that indicated substantial interactions offers a framework to partner with communities to
between indigenous groups and nonindigenous dwellers improve the quality of the research, including the sam-
in larger riverport towns in Amazonian departments pling strategies (e.g., Morin, Maiorana, Koester, Sheon,
(states or provinces) (see also Cárcamo et al., 2012). & Richards, 2003). Four community advisory boards
For the new study, investigators wished to develop a (CAB) were established in four river port towns in the
sampling plan to recruit a representative sample. Prior Amazon region. CAB members were recruited from
studies utilized convenience sampling (e.g., Bartlett indigenous communities, indigenous organizations, the
et al., 2008; Zavaleta et al., 2007), and to the investiga- health sector, local government, and community-based
tors’ knowledge, probability-based sampling had never organizations. CABs also included individuals living
been conducted in this region. This case study describes with HIV=AIDS, MSM, sex workers, and river trans-
in detail the sampling strategies used in a mixed- portation workers. Building on the strengths and
methods study in the Peruvian Amazon. The objective resources of the CABs and the research team, a colla-
of this case is to provide ideas to local and global health borative partnership was established to develop research
researchers responsible for conducting research in set- protocols, including sampling and recruitment strategies
tings where traditional means for obtaining random for the qualitative and quantitative phases of the study.
and representative samples are nonexistent. In each port city, CAB members discussed and
recommended three to five different indigenous villages
that the research team could travel to for several days
Identifying the Geographic Sampling Area
each to conduct fieldwork. Villages were purposively
Peru has the second-largest territory of Amazonia selected by the CAB and research team based on accessi-
after Brazil and is considered the fourth-largest country bility, available transportation, and having a reliable
in tropical forest extension on earth (Orta-Martı́nez & way to contact them to ask permission to visit. In
Finer, 2010). The Amazon region is characterized by addition, the investigators sought to maximize the ethnic
vast areas of jungle and extensive transportation net- diversity of the participants by selecting villages belong-
works that rely heavily on the Amazon River and its tri- ing to different ethnic groups. A total of 27 indigenous
butaries. The region has been subjected to constant villages were included in the study. Most villages were
migration of poor people from the Pacific Coast and accessible only by boat and ranged between 1 and 12
the Andes Mountains to the west and to ever-increasing hours in distance from the river ports. To visit these
resource extraction (Swenson, Carter, Domec, & villages, the investigators relied on their indigenous part-
Delgado, 2011). Medium-sized towns have grown along ners, who would relay messages of the purpose of their
the major rivers and serve as ports for the movement of work to the village leaders. All visits were approved by
people and goods. The team of investigators identified the Apu, the village chief, who would arrange with all
four river port towns to use as a base for the work in community members to be present on the accorded
the indigenous communities. These river ports were day of arrival. This process, as we will see later, was
accessible via airplane flights over the Andes Mountains enormously successful.

402
 SAMPLING STRATEGIES IN SEXUAL SCIENCE

Qualitative Phase of the Study household was then given a location and number on the
map. A list of all numbered households was then
The CAB worked with the research team to recruit
prepared, the number of eligible members in each
participants based on CAB contacts and participant
household was recorded, and all eligible persons in
referrals in two general settings: indigenous villages
the community were enumerated. Eligible household
(N ¼ 16) and four river ports. Women and men were
members were men and women ages 18 to 29 living in
eligible to participate if they self-identified as indigen-
the selected community for at least six months prior to
ous, were 18 years of age or older, were conversant in
the interview. If more than 30 eligible persons were
Spanish, and were able to consent to participate in the
identified in a community, a subset was randomly selec-
rapid ethnographic study. Each potential participant
ted for participation. Selected household members were
was met for a brief screening to assess for impairment
then interviewed to confirm eligibility and obtain
related to substance use or cognitive ability. The investi-
consent to participate.
gators read the informed consent aloud for the partici-
The informed consent procedure used in the qualitat-
pant to make sure that every section was reviewed and
ive phase was replicated for the survey interview. Each
understood. To ensure that the participant understood
participant was read the consent form aloud to ensure
the information given in the consent process, the inves-
that all aspects of the study were reviewed and under-
tigators asked if any clarification was needed after each
stood. The investigators highlighted the fact that partici-
section was read. The protocol, instruments, and
pation was voluntary and that they could terminate the
consents were reviewed and approved by the ethics
interview at any time without penalty. Participants were
committee of the institutional review board at the
given the opportunity to ask clarifying questions after
researchers’ university.
each section of the consent form was read.
A total of 40 individual in-depth interviews and nine
focus groups (n ¼ 98) were conducted. In general, parti-
cipants’ livelihoods depended on small-scale commerce, Sample Quality
agriculture, fishing, and other natural resource extrac-
A total of 644 eligible individuals agreed to partici-
tion enterprises. Several men worked as laborers in
pate (285 men, 359 women). With the exception of four
large-scale timber extraction and in the river transpor-
individuals who interrupted their interviews, all parti-
tation system as deckhands and boat loaders. The
cipants completed the survey and provided biological
CAB aided investigators in developing an appropriate
samples for HIV=STI testing. The cooperation rate
interview protocol, which, among other relevant
was 99%. Biological findings indicated that there was a
topics, also obtained information to aid in designing a
high prevalence of STIs and STI-risk behaviors among
probability-based sampling strategy for the quantitative
indigenous people living in the Amazonia region of Peru
survey. To triangulate the information obtained through
(Alva et al., 2012; Orellana, Alva, Cárcamo, & Garcı́a,
the in-depth interviews and focus groups, the investiga-
2013). However, there are no external, public health
tors also conducted systematic observations in all study
data to which the findings can be compared. Given the
settings, took ample field notes, and conducted informal
high cooperation rate, we have substantial confidence
interviews with key informants. A key component of
in the quality of the sample from which these data were
the qualitative phase of the study was the prolonged
derived.
engagement with members of the indigenous communi-
ties. This process resulted over several months in build-
ing partnerships and trust. Summary: Sampling the Amazon
The strategies employed in the present case example
have been used in contexts where many sample building
Developing a Probability-Based Sample Frame:
blocks are not available and must be created to pursue
Mapping and Enumeration
sample selection of high quality. The unusually high
Census data were not available for the villages in the cooperation rate (99%) was likely due to the prolonged
sample frame; nor were the sampling units (households) engagement with the communities, which increased
identifiable by addresses. However, the rapid ethno- mutual trust and allowed for those willing to participate
graphic study helped the team estimate the number of to complete the interviews and HIV=STI testing. In
households and residents for each participating com- addition, the use of local CABs also led to closer work-
munity. Based on these data, we began developing a ing relationships with village leaders, which contributed
probability sampling plan and approximate sample substantially to our success. By partnering with local
sizes. To verify these estimates, we engaged the chief authorities and community members, investigators were
of the community or other village authorities to aid able to build trust that facilitated the rapid ethnographic
the research team in drawing a map portraying the work, which supported the success of the field survey.
distribution of households and reference points in the Building those relationships was key in mapping villages
community (e.g., water wells, soccer fields, rivers). Each and then listing all household members. Because this

403
 CATANIA, DOLCINI, ORELLANA, AND NARAYANAN

was the first study of its nature in the Peruvian Amazon, and investigators (Armstrong, Crum, Rieger, Bennett,
the investigators were unable to compare the data to & Edwards, 1999; Freimuth et al., 2001; Furr, 2002;
other empirical sources. Nevertheless, many of the Gamble, 1997; Harris, Gorelick, Samuels, & Bempong,
behavioral and epidemiological findings of the study 1996; LaVeist, Nickerson, & Bowie, 2000; Matthews,
were shared with the CAB and the community, and the Sellergren, Manfredi, & Williams, 2002).
findings were seen by both as a reasonable represen-
tation of their experiences.
Local Context and Pre-Sampling Work
We identified neighborhoods in a Western city of the
Sampling Ethnic Minority Youth and Their United States with a large proportion of low-income
Friendship Networks African American youth and high rates of sexually
transmitted infections and teenage pregnancy (see Dolcini
Most studies of adolescent sexuality in the United et al., 2005; Dolcini et al., 2013). The presampling work
States have their basis in nonprobability samples drawn focused on several issues: (a) developing and maintaining
from school and clinic populations (Add Health being community relationships, (b) identifying the appropriate
a notable exception). These venue samples have limited social groups to target and defining the characteristics
generalizability. School-based studies, for instance, do of these groups, and (c) conducting pilot tests of study
not include youth who are truant or have dropped out procedures. The first step in this process is essential to
of school (a significantly more common problem among building a connection to the community and acceptance
low-income youth in large urban areas; see Hochschild, of the research work. Prior work supports the view that
2003). Furthermore, factors such as busing to schools out- developing relationships with community partners is
side of neighborhoods and attendance at charter or private a key first step in working at the community level
institutions lead to conditions wherein some portion of the (see Kellam, 2012; Stokols, 2006). This portion of the
neighborhood population may be missed in school-based fieldwork involves four processes: (a) recognizing that
studies because such studies tend to rely on sampling from developing community relationships involves a signifi-
a specific subset of schools (e.g., only public schools; cant time commitment, (b) building trust through mutual
only private schools). Thus, approaches that can provide cooperation in the research process, (c) recognizing and
coverage for an entire neighborhood are desirable. engaging community talent, and (d) giving back to the
National-level studies have been conducted based community in informational and material ways.
on household probability samples, but this approach
has shortcomings as well. National studies provide Building community relationships. Introductions
an overview of the general population but do not reflect were made to community leaders by a key stakeholder
variation that occurs at the local level. Sexual activity recruited by the research team because this person had
occurs within communities, and many ethnic minority long-standing relationships both in the community and
communities are structured around neighborhoods. in the local department of public health. The key stake-
The following case study provides a strategy for obtain- holder served to legitimize the research team and helped
ing a household probability sample of adolescent begin the process of building trust. This process took
African American youth residing in low-income neigh- considerable time. Over a period of approximately six
borhoods. This population segment has high rates of months, researchers met with community leaders and
sexual health problems (e.g., Dolcini et al., 2013) and staff at community-based agencies to develop a shared
so is of public health significance in the United States. conception of what the important issues were in the
Moreover, we demonstrate how sampling individuals community with regard to youth and sexual health.
can be used to develop subsequent samples of friendship All meetings took place in the community to emphasize
networks, which contribute to sexual socialization of the centrality of the community in the research and
youth in their respective communities (Beale, Ausiello, demonstrate respect for community members’ time.
& Perrin, 2001; Maxwell, 2002; Romer et al., 1994). In Through this process, community leaders and the
addition, the structure of these peer socialization influ- researchers were able to establish a level of trust that
ences is also known to vary by venue and community enabled the work to develop.
(e.g., Brown, 2004; Dolcini & Adler, 1994; Dolcini,
Catania, & Harper, 2005; Way, 2006). A key feature of Establishing a community advisory board. Over time,
our sampling strategy was the presampling qualitative– the presampling process also included development of
community participatory work conducted to address a community youth advisory group and advisory
many of the social issues of working in African members from organizations providing youth services
American communities (see Brown, 2004). Notably, or general services in the community. These community
there is a history of unethical research being conducted advisory boards (CABs), among other activities,
with African Americans that has led to a much greater commented on recruitment scripts that were important
degree of mistrust of medical and research institutions to accessing households and gaining adult permission to

404
 SAMPLING STRATEGIES IN SEXUAL SCIENCE

interview their adolescent children. CABs also provided recruitment. A sample frame was developed for the
input on the interview instruments, preliminary data, targeted neighborhoods (neighborhoods had 100%
and interpretation of the final data. phone coverage at that time, with an estimated 1,200
adolescent residents) that was geographically approxi-
Community field sites. To facilitate community par- mate, and the sample required further geographic
ticipation and familiarity with the research project, we screening. A probability sample of households was
established a field site centrally located in the primary screened to determine eligibility (zip code, African
commercial center of the community. The site was access- American, a resident adolescent with two close friends
ible by public transit, shared a building with other local aged 13-21 years in the neighborhood). Parental consent
businesses (i.e., those entering could be entering for was secured prior to recruitment of the adolescent. A
reasons unrelated to the research), and was in neutral brief interview was conducted and then youth were
territory (i.e., so youth from various segments of the com- interviewed in-person at the community site to increase
munity would be willing to come to this location). CABs privacy. Selected adolescents (seeds; N ¼ 91) enumerated
provided input on how and where to set up the community their close friends (N ¼ 312 friends). Each seed then
research site as well as providing referrals to local talent. A recruited close friends who met study criteria (resided
majority of project staff were hired from the community in the neighborhood, age 13 to 21, met the definition
and trained to conduct various research activities. of close friend) up to a maximum of three friends; if
the number of close friends exceeded three, then the
Community feedback and participation. To maintain friends were randomly selected. This phase of network
community visibility, we developed a regular newsletter sampling has parallels to respondent-driven sampling,
distributed to community leaders, organizations, and albeit the selection process was randomized.
participants; and the research team attended community
events (e.g., health fairs, retirement parties). When avail-
able, we offered data summaries to community leaders Sample Quality
and maintained an open door to share data in support We examined sample quality at two levels—for the
of local needs (e.g., for community-based grant applica- initial sample of adolescents and for the friendship net-
tions, reports). When our investigation ended, we work sample. The cooperation rate for the initial sample
offered equipment to the community (e.g., computers, frame of youth was excellent (70%) and comparable to
chairs, desks). Together, these efforts reflect the process that achieved in sampling from other African American
of giving back to the community both information and neighborhoods (Catania, Coates, Peterson, & Dolcini,
material goods. 1993; Kanouse et al., 1991). Among friends, approxi-
mately 86% were successfully recruited into the study.
Presampling Research: Clarifying Sampling Units Data provided by the seed and their friends confirmed
their close friend relationships. We compared the initial
Within the study neighborhoods, households with
fully enumerated list of friends with the subsampled
adolescents and an adolescent resident were the initial
interviewed friends and found that they paralleled each
sampling units. However, we also wanted to sample
other in terms of residence, friendship length and quality,
the social networks that youth were embedded in within
and proportion associating on weekends. There was
those neighborhoods. School-based research had sug-
a minor but nonsignificant difference on gender, with
gested that social crowds (i.e., large reputation-based
sampled friends slightly more likely to be female [approxi-
social groups, such as jocks, dopers) have a significant
mately 5% (v2 ¼ 1.45, p > .10)]. These findings suggest that
influence on adolescents, including ethnic minority our sampling procedures generated good representation of
youth (Brown & Lohr, 1987; Dolcini & Adler, 1994). respondents’ neighborhood friendship groups. Data on ref-
However, basing our sampling on these school studies erence group norms within the friendship groups (Dolcini
would have been problematic. Presampling qualitative et al., 2013) paralleled data observed in general population
research (focus groups) indicated that the crowd-based (Laumann, Gagnon, Michael, & Michaels, 1994), suggest-
social groups were irrelevant in the neighborhoods, ing that the sampled friendship networks were not
but membership in close friendship groups was com- unusually skewed in terms of normative structures.
mon. Thus, the research shifted to sampling friendship We obtained external validation by comparing school
groups (i.e., cliques). Pilot studies refined the friendship drop-out rates from our sample to those from the city’s
recruitment process and aided in developing culturally school database. The high school drop-out rate was only
relevant measures (Dolcini et al., 2005). slightly lower in our sample than in the general African
American population (city records ¼ 5.5%; our sample ¼
Sampling Households and Friendship Networks
4.0%, after excluding those with HS diploma or GED).
Although the primary household sampling strategy Contrasting sample biological data on STIs to public
was based on a random digit-dial method, many of the health records showed highly comparable results for
procedures employed would be similar for in-person Chlamydia and Gonorrhea (unpublished data).

405
 CATANIA, DOLCINI, ORELLANA, AND NARAYANAN

Community versus School Samples of Social on MSM in industrialized nations and developing
Networks: Value Added countries. A systematic review of studies representing
advances, debates, and comparisons in sampling meth-
Historically, school-based studies often limited
ods shows four methods to be in use: (a) opportunistic
respondents to identifying friends who attend the same
sampling (e.g., Guo et al., 2011; Gustafson et al., 2013;
school and are in the same grade (for an exception, see
Paquette & De Wit, 2010); (b) venue-based (time–space)
Huang et al., 2013). Sampling from neighborhoods
sampling (e.g., Guo et al., 2011; Jenness et al., 2011;
opens the investigator to the broader context in which
Pollack, Osmond, Paul, & Catania, 2005, 2006;
sexuality occurs for urban African American youth. In
Wei, McFarland, Colfax, Fuqua, & Raymond, 2012);
this regard, the neighborhood data indicate that African
(c) respondent-driven sampling (e.g., Carballo-Diéguez
American youth are more likely to have their closest
et al., 2011; Dennis et al., 2013; Guo et al., 2011;
friends in the neighborhood and be less connected
Johnston et al., 2013; Paquette & De Wit, 2010;
to school social networks than is found for White
Ramirez-Valles, Heckathorn, Vázquez, Diaz, &
youth (see also Dubois & Hirsch, 1990). Moreover, the
Campbell, 2005; Raymond & McFarland, 2009; Risser,
majority of friendship networks in our sample included
Padgett, Wolverton, & Risser, 2009; Wei et al., 2012);
some nonschool friends (57%) and nearly one-quarter
and (d) household probability-based samples (e.g.,
were exclusively nonschool friends, highlighting the
Catania et al., 2001; Fendrich et al., 2013; Mackesy-
importance of moving outside school environments
Amiti et al., 2010; Osmond, Pollack, Paul, & Catania,
when examining the impact of friendship groups on sex-
2007; Pollack et al., 2005, 2006; Schwarcz et al., 2007).
ual behavior and health (Dolcini et al., 2005). Moreover,
The few comparative studies that have been conducted
although we found that youth were more likely to spend
have found that venue samples, compared to com-
time with school networks on weekdays during school
munity probability-based sexual health surveys of
time=activities, they spent more time with their neigh-
MSM, may underestimate closeted MSM and distort
borhood networks after school and on weekends. Prior
data on sexual development (Pollack et al., 2005,
work suggests it is weekdays after school and weekends
2006). Comparison of venue-based sampling and
that provide the greatest opportunities for engaging in
respondent-driven sampling indicate that respondent-
health risk behaviors (Cohen, Farley, Taylor, Martin,
driven sampling generates more diverse samples of
& Schuster, 2002; Fortenberry, Orr, Zimet, & Blythe,
ethnic minority MSM than venue-based samples (Wei
1997; Hahn et al., 1990). Expanding sample selection
et al., 2012). The case study presented here is based on
beyond specific venues to the large community opens a
the Urban Men’s Health Study I (UMHS I) (Catania
broader window on the social conditions that influence
et al., 2001; Osmond et al., 2007; Catania et al., 2006),
sexual behavior and sexual health.
a probability-based household sample that utilized
telephone data collection. The methodological issues
Summary: Sampling Adolescent Communities discussed here may be applied to other data collection
A number of recommendations emerge from this modes with minor variations.
prior work. Presampling fieldwork in combination with UMHS I obtained data on MSM in the areas of
community participatory strategies has the potential to sexual health, sexual development, social migration,
increase study acceptance and avoid conceptual errors substance use, and other health and mental health out-
that might develop from work based on school settings comes. Data were collected in four cities (Los Angeles,
alone. Second, a combination of probability-based sam- Chicago, New York, and San Francisco; see Catania,
pling and network-respondent driven sampling can be Canchola, Pollack, & Chang, 2006; Catania et al., 2001;
used to generate a sample that represents social groups Osmond et al., 2007). Prior to sample development, we
in a community. Third, sample quality can be deter- conducted mapping studies (qualitative and quantitat-
mined by comparisons to prior research findings and ive) and made quantitative estimates of the number of
publicly available surveillance data and other records. MSM residing within different geographic subareas of
Moreover, there appears to be added value to sampling each city.
social networks from the community. Finally, research-
ers must be sensitive to community issues and the time Pre-Sampling Quantitative and Qualitative Studies
and skills necessary for working in ethnic minority com-
munities, particularly those in which there is distrust of Mapping research. We initially mapped MSM
large organizations and government. residences in 26 HIV epicenters of the United States
and selected four cities for study. Mapping utilized
U.S. Census, commercial, and public health data to
Sampling Urban MSM: Urban Men’s Health Study determine the number of MSM residences in geographic
subareas of each city (zip codes). The goal of this work
A variety of sampling approaches with diverse data was to identify areas of high-, moderate-, and low-
collection methods have been utilized to gather data density MSM residential areas. AIDS case data by zip

406
 SAMPLING STRATEGIES IN SEXUAL SCIENCE

code proved to be the most useful mapping data Based on these data, if we wished to oversample only
source. We subsequently conducted a multipurpose African American MSM (estimated at 20% of the
rapid ethnographic study in each city. Research teams MSM population) we would have to screen approxi-
spent one to two weeks in each city and communicated mately 300,000 households to obtain a sample of 3,000
by telephone and e-mail for a longer period of time. African American MSM [.05 (est. overall prevalence of
Teams conducted informal interviews with local MSM)  .20 ¼ .01 sampling rate; .01  300,000 ¼ 3,000
residents, commercial interests, and health providers in cases]. Such numbers are staggering to consider.
obtaining their perspectives on where MSM lived, Two sampling strategies were employed to reduce
identifying specific community health needs, building a costs and meet our sampling goals. Disproportionate
network of verification contacts within the community and adaptive sampling techniques were used (Blair,
so prospective respondents could verify the authenticity 1999; Capell & Schiller, 1989; Catania et al., 1996;
of the study (build trust), and providing information on Hansen, Hurwitz, & Madow, 1953; Kalton, 1993;
the forthcoming study. Small amounts of funding were Sudman, 1976). Disproportionate sampling (Kalton,
given to organizations that contributed significant time 2003) provides a standardized method of sampling var-
to these efforts. Teams also completed drive-throughs iations in MSM household geodensity (i.e., areas are
of various ‘‘high-density’’ areas to get a sense of whether sampled inversely proportional to the square root of
they were ‘‘mini gay ghettos’’ or more reflective of the their costs). Thus, all areas are sampled at known rates
larger heterocommunity. These data provided on-the- of selection, which allows for statistical weighting of the
ground qualitative verification of MSM residential sample and, importantly, keeps costs within budget.
densities (e.g., judged by the number of gay commercial Because our sampling process is based on multiple esti-
establishments). mates, it is important to be able to correct the sampling
process in situ if estimates for a given geographic area
Geopopulation size estimates. To sample MSM prove wildly incorrect. Adaptive sampling allows you
from urban areas, and go beyond the usual ‘‘gay ghetto’’ to systematically adjust the sample frame in relation to
neighborhoods, it was necessary to estimate the number the reality obtained while collecting data in a given
of MSM households, the geographic distribution of geographic area (Blair, 1999; Thompson, 1990). Both
those households in each city, and the number of eligible sampling strategies allowed us to obtain excellent cover-
MSM within those households. To estimate the overall age and a large sample size while staying within budget.
MSM population size for each city, we used data from Nevertheless, as most readers are probably aware, the
Binson and colleagues (1995) that aggregated national overall cost of these types of surveys is high. For studies
household probability samples from consecutive surveys with sample sizes of 1,000 to 3,000 persons, the costs will
(e.g., General Social Survey) to generate estimates of the be in the hundreds of thousands of dollars. However,
prevalence of MSM across different geographic regions these costs can be reduced if adaptive and dispropor-
of the United States. Using the appropriate estimate tionate sampling can be applied. The cost savings
for cities of the size of each of the four priority cities depend on a wide variety of factors, and other authors
and the proportion of MSM AIDS cases in each city, have discussed these. Thompson (1990) has discussed
we estimated variation in the size of the MSM popu- sampling efficiencies when using adaptive sampling
lation in each city (i.e., we estimated a range of values that, in turn, reduce costs. Kalton (2003) discussed
based on the 95% confidence interval). City estimates the conditions under which disproportionate sampling
were then apportioned to each zip code of the city based produces more than modest gains in efficiency (e.g.,
on the proportion of MSM AIDS cases in each zip code. at least one of the sampling strata [usually the largest
We also estimated the average number of MSM per fraction] must have a prevalence of characteristic
household to be two. With these numbers we were [e.g., MSM] that is much larger than in the general
then able to estimate the number of MSM households population [a condition met by UMHS]).
per zip code.
Sample Quality
Sample Construction
A variety of strategies were used to examine sample
It is expensive to conduct probability samples of quality. The cooperation rate was reasonably high
hard-to-reach populations. Moreover, to sample from (78%: 2,881 interviewed=3,700 eligible MSM). Further,
low-density residential areas the costs become over- study results were verified by public health indices,
whelming. Our goal was to sample from geographic predictive checks, and independent replication (Catania
areas with densities ranging from 1% to 40%. The et al., 2006; Catania et al., 2001; Osmond et al., 2007;
process of screening households for MSM residents Schwarcz et al., 2007): (a) AIDS prevalence estimates,
(from which only one person was randomly selected incidence rates, and back-calculation modeling based
for interview) is the limiting factor. UMHS screened on UMHS data aligned with public health records;
more than 95,000 households to interview 2,881 MSM. and (b) an independent household probability survey

407
 CATANIA, DOLCINI, ORELLANA, AND NARAYANAN

in San Francisco, obtained HIV prevalence levels with friends (Catania et al., 2006). Limiting disclosure
confidence intervals overlapping those from the UMHS to only one select social group (e.g., friends) may
San Francisco subsample. create the impression that ethnic minority MSM
The absence of U.S. Census data on sexual minorities are closeted (also see Mills et al., 2001).
(excepting for data on unmarried same-sex-partnered
households; (e.g., see Lieb et al., 2011) limits our ability
to evaluate the demographic profile of the sample or Summary: UMHS
apply poststratification weights. Because our final The qualitative and quantitative sampling strategies
sample was approximately 80% Caucasian, the concern employed in the UMHS can be applied to other
has been raised that ethnic=racial minority men were sample–interview mode combinations. We recommend
underrepresented. This view rests heavily on speculation to (a) utilize a mixed-methods approach to presampling
and the misperception that adult, urban MSM represent work that enables investigators to prepare the
the demographic profile of their city of residence. community for the upcoming survey, build trust with
We have examined this issue extensively (Catania et al., the community, and engage the community with regard
2001) and provide evidence indicating the following: to their needs, (b) reduce sampling costs by employing
known procedures such as adaptive and disproportion-
. MSM migrate in large numbers to major cities ate sampling methods, and (c) use multiple external
from elsewhere in the country to be with men and analytic data strategies to provide measures of
of the same sexual orientation within the safety sample quality.
provided by the gay communities; the consequence
of this migration is that the urban demographic
profile of MSM resembles that of the nation’s Overall Summary and Recommendations
demographics more than the city of residence.
UMHS found that approximately 82% of MSM Nonprobability Sampling
(in 1996) residing in the four study cities were in-
migrants, and substantially more in-migrants were Internet sampling will likely not fully replace student
White (particularly among young adult MSM). and clinic samples, but for relatively low cost and high
These results confirm earlier observations (Bell & efficiency it can meet the needs of modelers. A key
Weinberg, 1978; Murray, 1992). The 1996 Census- feature of Internet sampling is the ability to adjust in
Current Population Survey data (80% of adult real time the characteristics of the sample to construct
males residing outside of central cities were White; variable distributions that reflect desired characteristics.
Catania et al., 2006) and migration studies (Frey, It is important to keep in mind that some segments of
1978) that show that more highly educated persons the population have low household Internet access=use
migrate, confirms that in-migrants from outside (e.g., elderly, impoverished), but this may change with
large urban areas are more likely to be Caucasian time. Maintaining a secure data pipeline is a key ethical
(and better educated). consideration in conducting Internet sampling, while
. For an urban sample of MSM to reflect the demo- also recognizing that Internet sampling poses risks not
graphic profile of a city at the time in which they dissimilar to those experienced in using other methods
were sampled (1996) would require that the bulk (e.g., telephone sampling).
of the population be born in the city of residence
at the time surveyed. We found through review
Probability Sampling
of Census records that the racial=ethnic profile of
the UMHS sample frame better reflected profiles Obtaining probability-based sampling of difficult-to-
of the parents of men born between 1920 to reach populations has always been a challenge in sex
1978, the birth years of UMHS respondents, than research, as well as other fields. These challenges
they reflected current city profiles. increase when populations are not represented in Census
. The quality of the sampling and recruitment data or other well-managed data sources, and=or when
procedures are also evident in the number of semi- primary sampling units cannot be easily identified. The
and fully closeted men interviewed. UMHS was case studies presented here provide a number of solu-
able to interview men with relatively high levels tions to conducting high-quality sampling in popula-
of closetedness: 20% of the sample self-rated as tions with these types of challenges. Presampling work
moderate to highly closeted (out to less than half is essential and typically includes (a) rapid ethnographic
of family or friends), including a small subsample studies involving qualitative data collection; (b) recruit-
of men (<2%) who were not out to any family, ing a local ‘‘guide’’ to interface between investigators
friends, neighbors, or coworkers. Moreover, and community leaders and gatekeepers, and subse-
among African American MSM, 27% were not quently, the larger community; (c) building trust and
out to neighbors, but only 4% were not out to respect; (d) obtaining input from the community and

408
 SAMPLING STRATEGIES IN SEXUAL SCIENCE

providing the community with feedback; (e) hiring local Beale, A. C., Ausiello, J., & Perrin, J. M. (2001). Social influence on
talent to conduct fieldwork; and (f) conducting both health-risk behaviors among minority middle school students.
Journal of Adolescent Health, 28(6), 474–480. doi:10.1016=
qualitative and quantitative mapping work to identify S1054-139X(01)00194-X
key sampling areas and units. In addition, the use of dis- Bell, A. P., & Weinberg, M. S. (1978). Homosexualities: A study of
proportionate and adaptive sampling strategies can be diversity among men and women. New York, NY: Simon and
employed to reduce costs and increase coverage, parti- Schuster.
cularly when some sampling areas have extremely low Binson, D., Michaels, S., Stall, R., Coates, T. J., Gagnon, J. H., &
Catania, J. A. (1995). Prevalence and social distribution of men
levels of eligible MSM. who have sex with men: United States and its urban centers.
Journal of Sex Research, 32(3), 245–254. doi:10.1080=0022
4499509551795
Sample Quality Blair, J. (1999). A probability sample of gay urban males: The use of
two-phase adaptive sampling. Journal of Sex Research, 36(1),
Sample quality judged by cooperation rates and
39–44. doi:10.1080=00224499909551965
comparison to external ‘‘gold standards’’ are typical Brown, B. B. (2004). Adolescent relationships with peers. In R. Lerner
strategies for evaluating the quality of a sample frame; & L. Steinberg (Eds.), Handbook of adolescent psychology (2nd ed.,
hence the data generated. When Census data are pp. 363–394). Hoboken, NJ: John Wiley & Sons.
lacking, investigators may resort to using public health Brown, B. B., & Lohr, M. J. (1987). Peer-group affiliation and ado-
lescent self-esteem: An integration of ego-identity and symbolic-
and other public records for evaluation, as well as infer-
interaction theories. Journal of Personality and Social Psychology,
ential strategies (e.g., analysis of demographic profiles, 52(1), 47–55. doi:10.1037=0022-3514.52.1.47
predictive validity). Capell, F. J., & Schiller, G. (1989). Efficiency of general population
screening for persons at an elevated risk of HIV infection: Evi-
dence from a statewide telephone survey of California adults. In
F. J. Fowler (Ed.), Fifth conference on health survey research meth-
References ods (pp. 167–172). Keystone, CO: U.S. Dept. of Health and
Human Services.
Alessi, E. J., & Martin, J. I. (2010). Conducting an Internet- Carballo-Diéguez, A., Balan, I., Marone, R., Pando, M. A., Dolezal,
based survey: Benefits, pitfalls, and lessons learned. Social Work C., Barreda, V., . . . Ávila, M. M. (2011). Use of respondent driven
Research, 34, 122–128. doi:10.1093=swr=34.2.122 sampling (RDS) generates a very diverse sample of men who have
Alexa Internet Inc. (2014). Retrieved from http://www.alexa.com/ sex with men (MSM) in Buenos Aires, Argentina. PloS One, 6(11),
siteinfo/gay.com e27447. doi:10.1371=journal.pone.0027447
Alva, I. E., Orellana, E. R., Blas, M. M., Bernabe-Ortiz, A., Cárcamo, C. P., Campos, P. E., Garcı́a, P. J., Hughes, J. P., Garnett,
Cotrina, A., Chiappe, M., . . . Zunt, J. R. (2012). Htlv-1 and-2 G. P., & Holmes, K. K. (2012). Prevalences of sexually
infections among 10 indigenous groups in the Peruvian Amazon. transmitted infections in young adults and female sex workers in
American Journal of Tropical Medicine and Hygiene, 87, Peru: A national population-based survey. The Lancet Infectious
954–956. doi:10.4269=ajtmh.2012.12-0289 Diseases, 12(10), 765–773. doi:10.1016=S1473-3099(12)70144-5
American Association of Public Opinion Research, Baker, R., Brick, Catania, J. A., Binson, D., Canchola, J., Pollack, L. M., Hauck, W., &
J. M., Bates, N. A., Battaglia, M., Couper, M. P., . . . Tourangeau, Coates, T. J. (1996). Effects of interviewer gender, interviewer
R. (2013, June). Report of the AAPOR Task Force on choice, and item wording on responses to questions concerning
Non-Probability Sampling. Retrieved from https://www.aapor. sexual behavior. Public Opinion Quarterly, 60(3), 345–375.
org/AAPORKentico/AAPOR_Main/media/MainSiteFiles/ doi:10.1086=297758
NPS_TF_Report_Final_7_revised_FNL_6_22_13.pdf Catania, J. A., Canchola, J., Pollack, L., & Chang, J. (2006). Under-
Armstrong, T. D., Crum, L. D., Rieger, R. H., Bennett, T. A., & standing the demographic characteristics of urban men who have
Edwards, L. J. (1999). Attitudes of African Americans toward sex with men. Journal of Homosexuality, 51(3), 33–51.
participation in medical research. Journal of Applied Social doi:10.1300=J082v51n03_03
Psychology, 29, 552–574. doi:10.1111=j.1559-1816.1999.tb01401.x Catania, J. A., Coates, T. J., Peterson, J., & Dolcini, M. M. (1993).
Auerbach, J. D., Parkhurst, J. O., & Cáceres, C. F. (2011). Addressing Changes in condom use among black, hispanic, and white hetero-
social drivers of HIV=AIDS for the long-term response: Concep- sexuals in San Francisco: The AMEN cohort survey. Journal of
tual and methodological considerations. Global Public Health, Sex Research, 30(2), 121–128. doi:10.1080=00224499309551691
6(Suppl. 3), S293–S309. doi:10.1080=17441692.2011.594451 Catania, J. A., Osmond, D., Stall, R. D., Pollack, L., Paul, J. P.,
Baker, R., Brick, M. J., Bates, N. A., Battaglia, M., Cooper, M. P., Blower, S., . . . Fisher, L. (2001). The continuing HIV epidemic
Dever, J. A., . . . Tourangeau, R. (2013). Summary report of the among men who have sex with men. American Journal of Public
AAPOR Task Force on Non-Probablilty Sampling. Journal of Health, 91(6), 907–914. doi:10.2105=AJPH.91.6.907
Survey Statistics and Methodology, 1(2), 90–105. doi:10.1093= Cohen, D. A., Farley, T. A., Taylor, S. N., Martin, D. H., & Schuster,
jssam=smt008 M. A. (2002). When and where do youths have sex? The potential
Barresi, P., Husnik, M., Camacho, M., Powell, B., Gage, R., LeBlanc, role of adult supervision. Pediatrics, 110(6), e66. doi:10.1542=
D., . . . Koblin, B. (2010). Recruitment of men who have sex with peds.110.6.e66
men for large HIV intervention trials: Analysis of the explore Curtis, B. L. (2014). Social networking and online recruiting for HIV
study recruitment effort. AIDS Education and Prevention, 22(1), research: Ethical challenges. Journal of Empirical Research on
28–36. doi:10.1521=aeap.2010.22.1.28 Human Research Ethics, 9(1), 58–70. doi:10.1525=jer.2014.9.1.58.
Bartlett, E. C., Zavaleta, C., Fernández, C., Razuri, H., Vilcarromero, de Leeuw, E. D. (2010). Mixed-mode surveys and the Internet. Survey
S., Vermund, S. H., & Gotuzzo, E. (2008). Expansion of HIV and Practice, 3(6). Retrieved from http://www.surveypractice.org/
syphilis into the Peruvian Amazon: A survey of four communities index.php/SurveyPractice/article/view/150
of an indigenous amazonian ethnic group. International Journal of Dennis, A. M., Murillo, W., de Maria Hernandez, F., Guardado,
Infectious Diseases, 12(6), e89–e94. doi:10.1016=j.ijid.2008.03.036 M. E., Nieto, A. I., de Rivera, I. L., . . . Paz-Bailey, G. (2013).

409
 CATANIA, DOLCINI, ORELLANA, AND NARAYANAN

Social network-based recruitment successfully reveals HIV-1 of men who have sex with men. American Journal of Epidemi-
transmission networks among high-risk individuals in El ology, 177(10), 1157–1164. doi:10.1093=aje=kws358
Salvador. Journal of Acquired Immune Deficiency Syndromes Hahn, G., Charlin, V. L., Sussman, S., Dent, C. W., Manzi, J., Stacy,
(1999), 63(1), 135–141. doi:10.1097=QAI.0b013e318288b246 A. W., . . . Burton, D. (1990). Adolescents’ first and most recent
Dolcini, M. M., & Adler, N. E. (1994). Perceived competencies, peer use situations of smokeless tobacco and cigarettes: Similarities
group affiliation, and risk behavior among early adolescents. and differences. Addictive Behaviors, 15(5), 439–448. doi:10.1016=
Health Psychology, 13(6), 496–506. doi:10.1037=0278-6133. 0306-4603(90)90030-2
13.6.496 Hansen, M., Hurwitz, W., & Madow, W. (1953). Survey methods and
Dolcini, M. M., Catania, J. A., & Harper, G. W. (2005). Friends theory, 2 vols. New York, NY: John Wiley and Sons.
in the ‘‘hood’’: Should peer-based promotion target out-of-school Harris, Y., Gorelick, P. B., Samuels, P., & Bempong, I. (1996). Why
networks? Journal of Adolescent Health, 26, 267.e7–267.e16. African Americans may not be participating in clinical trials.
doi:10.1016=j.jadohealth.2004.10.003 Journal of the National Medical Association, 88(10), 630–634.
Dolcini, M. M., Catania, J. A., Harper, G. W., Watson, S. E., Ellen, Hochschild, J. L. (2003). Social class in public schools. Journal of Social
J. M., & Towner, S. L. (2013). Norms governing urban African Issues, 59(4), 821–840. doi:10.1046=j.0022-4537.2003.00092.x
American adolescents’ sexual and substance-using behavior. Huang, G. C., Unger, J. B., Soto, D., Fujimoto, K., Pentz, M. A.,
Journal of Adolescence, 36(1), 31–43. doi:10.1016=j.adolescence. Jordan-Marsh, M., & Valente, T. W. (2013). Peer influences:
2012.09.002 The impact of online and offline friendship networks on ado-
Dubois, D. L., & Hirsch, B. J. (1990). School and neighborhood lescent smoking and alcohol use. Journal of Adolescent Health,
friendship patterns of Blacks and Whites in early adolescence. 54(4), 508–514. doi:10.1016=j.jadohealth.2013.07.001
Child Development, 61(2), 524–536. doi:10.1111=j.1467-8624. Instituto Nacional de Estadı́stica e Informatica. (2008). Ii censo de
1990.tb02797.x comunidades indigenas de la Amazonia Peruana 2007 [Second
Evans, A. R., Wiggins, R. D., Mercer, C. H., Bolding, G. J., & census of indigenous communities of the Peruvian Amazon
Elford, J. (2007). Men who have sex with men in Great Britain: 2007]. Lima, Peru.
Comparison of a self-selected internet sample with a national Jenness, S. M., Neaigus, A., Murrill, C. S., Gelpi-Acosta, C., Wendel,
probability sample. Sexually Transmitted Infections, 83(3), 200–205. T., & Hagan, H. (2011). Recruitment-adjusted estimates of HIV
doi:10.1136=sti.2006.023283 prevalence and risk among men who have sex with men: Effects
Feldacker, C., Torrone, E., Triplette, M., Smith, J. C., & Leone, P. A. of weighting venue-based sampling data. Public Health Reports,
(2011). Reaching and retaining high-risk HIV=AIDs clients 126(5), 635–642.
through the Internet. Health Promotion Practice, 12(4), 522–528. Johnston, L. G., Prybylski, D., Raymond, H. F., Mirzazadeh, A.,
doi:10.1177=1524839909349178 Manopaiboon, C., & McFarland, W. (2013). Incorporating the
Fendrich, M., Avci, O., Johnson, T. P., & Mackesy-Amiti, M. E. service multiplier method in respondent-driven sampling surveys
(2013). Depression, substance use, and HIV risk in a probability to estimate the size of hidden and hard-to-reach populations: Case
sample of men who have sex with men. Addictive Behaviors, studies from around the world. Sexually Transmitted Diseases,
38(3), 1715–1718. doi:10.1016=j.addbeh.2012.09.005 40(4), 304–310. doi:10.1097=OLQ.0b013e31827fd650
Fortenberry, J. D., Orr, D. P., Zimet, G. D., & Blythe, M. J. (1997). Kalton, G. (1993). Sampling considerations in research on HIV risk
Weekly and seasonal variation in sexual behaviors among and illness. In D. K. Ostrow & R. C. Kessler (Eds.), Methodologi-
adolescent women with sexually transmitted diseases. Journal of cal issues in AIDS behavioral research (pp. 53–74). New York,
Adolescent Health, 20(6), 420–425. doi:10.1016=S1054-139X NY: Plenum Press.
(96)00275-3 Kalton, G. (2003). Practical methods for sampling rare and mobile
Freimuth, V. S., Quinn, S. C., Thomas, S. B., Cole, G., Zook, E., & populations. Proceedings of the Annual Meeting of the American
Duncan, T. (2001). African Americans’ views on research and Statistical Association, August 5–9, 2001. Retrieved from https://
the Tuskegee syphilis study. Social Science and Medicine, 52(5), www.amstat.org/sections/srms/Proceedings/y2001/Proceed/
797–808. doi:10.1016=S0277-9536(00)00178-7 00454.pdf
Frey, W. H. (1978). The changing impact of migration on the population Kanouse, D., Berry, S., Gorman, E., Yano, E., Carson, S., &
composition of origin and destination metropolitan areas. Paper Abrahamse, A. (1991). AIDS-related knowledge, attitudes, beliefs,
presented at the Population Association of America, Atlanta, and behaviors in Los Angeles County. Santa Monica, CA: Rand.
Georgia. Kellam, S. G. (2012). Developing and maintaining partnerships as the
Furr, L. A. (2002). Perceptions of genetics research as harmful to foundation of implementation and implementation science:
society: Differences among samples of African-Americans and Reflections over a half century. Administration and Policy in Mental
European-Americans. Genetic Testing, 6(1), 25–30. doi:10.1089= Health and Mental Health Services Research, 39(4), 317–320.
109065702760093889 doi:10:1007=s10488-011-0402-8
Gamble, V. N. (1997). Under the shadow of Tuskegee: African Laumann, E., Gagnon, J., Michael, R., & Michaels, S. (1994). The
Americans and health care. American Journal of Public Health, social organization of sexuality: Sexual practices in the United
87(11), 1773–1778. doi:10.2105=AJPH.87.11.1773 States. Chicago, IL: University of Chicago Press.
Garcı́a, P. J., Holmes, K. K., Cárcamo, C. P., Garnett, G. P., Hughes, LaVeist, T. A., Nickerson, K. J., & Bowie, J. V. (2000). Attitudes
J. P., Campos, P. E., & Whittington, W. L. (2012). Prevention about racism, medical mistrust, and satisfaction with care among
of sexually transmitted infections in urban communities African American and White cardiac patients. Medical Care
(Peru PREVEN): A multicomponent community-randomised Research and Review, 57(Suppl. 1), 146–161. doi:10.1177=
controlled trial. The Lancet, 379(9821), 1120–1128. doi:10.1016= 1077558700574007
S0140-6736(11)61846-1 Lieb, S., Fallon, S. J., Friedman, S. R., Thompson, D. R., Gates, G. J.,
Guo, Y., Li, X., Fang, X., Lin, X., Song, Y., Jiang, S., & Stanton, B. Liberti, T. M., & Malow, R. M. (2011). Statewide estimation of
(2011). A comparison of four sampling methods among men racial=ethnic populations of men who have sex with men in the
having sex with men in China: Implications for HIV=STD US. Public Health Reports, 126(1), 60–72. doi:10.2307=41639324
surveillance and prevention. AIDS Care, 23(11), 1400–1409. Mackesy-Amiti, M. E., Fendrich, M., & Johnson, T. P. (2010).
doi:10.1080=09540121.2011.565029 Symptoms of substance dependence and risky sexual behavior in
Gustafson, P., Gilbert, M., Xia, M., Michelow, W., Robert, W., a probability sample of HIV-negative men who have sex with
Trussler, T., . . . Gustafson, R. (2013). Impact of statistical adjust- men in Chicago. Drug and Alcohol Dependence, 110(1), 38–43.
ment for frequency of venue attendance in a venue-based survey doi:10.1016=j.drugalcdep.2010.01.016

410
 SAMPLING STRATEGIES IN SEXUAL SCIENCE

Matthews, A. K., Sellergren, S. A., Manfredi, C., & Williams, M. Risser, J., Padgett, P., Wolverton, M., & Risser, W. (2009).
(2002). Factors influencing medical information seeking among Relationship between heterosexual anal sex, injection drug use,
African American cancer patients. Journal of Health Communi- and HIV infection among Black men and women. International
cation, 7(3), 205–219. doi:10.1080=10810730290088094 Journal of STD and AIDS, 20(5), 310–314. doi:10.1258=
Maxwell, K. A. (2002). Friends: The role of peer influence across ado- ijsa.2008.008394
lescent risk behaviors. Journal of Youth and Adolescence, 31(4), Romer, D., Black, M., Ricardo, I., Feigelman, S., Kaljee, L.,
267–277. doi:10.1023=A:1015493316865 Galbraith, J., . . . Stanton, B. (1994). Social influences on the
Meyer, I. H., & Colten, M. E. (1999). Sampling gay men: Random sexual behavior of youth at risk for HIV exposure. American
digit dialing versus sources in the gay community. Journal of Journal of Public Health, 84(6), 977–985. doi:10.2105=
Homosexuality, 37(4), 99–110. doi:10.1300=J082v37n04_06 AJPH.84.6.977
Mills, T. C., Stall, R., Pollack, L., Paul, J. P., Binson, D., Canchola, J., Rosen, R., Catania, J., Lue, T., Althof, S., Henne, J., Hellstrom, W., &
& Catania, J. A. (2001). Health-related characteristics of men who Levine, L. (2008). Impact of Peyronie’s disease on sexual and
have sex with men: A comparison of those living in ‘‘gay ghettos’’ psychosocial functioning: Qualitative findings in patients and con-
with those living elsewhere. American Journal of Public Health, trols. Journal of Sexual Medicine, 5, 1977–1984. doi:10.1111=
91(6), 980–983. doi:10.2105=AJPH.91.6.980 j.1743-6109.2008.00883.x
Minkler, M. (2000). Using participatory action research to build Ross, M. W., Tikkanen, R., & Månsson, S.-A. (2000). Differences
healthy communities. Public Health Reports, 115(2–3), 191–197. between internet samples and conventional samples of men who
doi:10.1093=phr=115.2.191 have sex with men: Implications for research and HIV interven-
Minkler, M. (2005). Community-based research partnerships: tions. Social Science and Medicine, 51(5), 749–758. doi:10.1016=
Challenges and opportunities. Journal of Urban Health, S0277-9536(99)00493-1
82(Suppl. 2), ii3–ii12. doi:10.1093=jurban=jti034 Schwarcz, S., Scheer, S., McFarland, W., Katz, M., Valleroy, L.,
Montenegro, R. A., & Stephens, C. (2006). Indigenous health in Latin Chen, S., & Catania, J. (2007). Prevalence of HIV infection and
America and the Caribbean. The Lancet, 367(9525), 1859–1869. predictors of high-transmission sexual risk behaviors among
doi:10.1016=S0140-6736(06)68808-9 men who have sex with men. American Journal of Public Health,
Morin, S. F., Maiorana, A., Koester, K. A., Sheon, N. M., & 97(6), 1067–1075. doi:10.2105=AJPH.2005.072249
Richards, T. A. (2003). Community consultation in HIV preven- Stokols, D. (2006). Toward a science of transdisciplinary action
tion research: A study of community advisory boards at 6 research. American Journal of Community Psychology, 38(1–2),
research sites. Journal of Acquired Immune Deficiency Syndromes, 63–77. doi:10.1007=s10464-006-9060-5
33(4), 513–520. doi:10.1097=00126334-200308010-00013 Sudman, S. (1976). Applied sampling. New York, NY: Academic Press.
Murray, S. O. (1992). Components of gay community in San Swenson, J. J., Carter, C. E., Domec, J.-C., & Delgado, C. I. (2011).
Francisco. In G. Herdt (Ed.), Gay culture in America (pp. 107– Gold mining in the Peruvian Amazon: Global prices, deforesta-
146). Boston, MA: Beacon Press. tion, and mercury imports. PloS One, 6(4), e18875. doi:10.1371=
Orellana, E. R., Alva, I. E., Cárcamo, C. P., & Garcı́a, P. J. (2013). journal.pone.0018875
Structural factors that increase HIV=STI vulnerability among Thompson, S. K. (1990). Adaptive cluster sampling. Journal of the
indigenous people in the Peruvian Amazon. Qualitative Health American Statistical Association, 85(412), 1050–1059. doi:10.
Research, 23(9), 1240–1250. doi:10.1177=1049732313502129 1080=01621459.1990.10474975
Orta-Martı́nez, M., & Finer, M. (2010). Oil frontiers and indigenous U.S. Census Bureau. (2012). Computer and Internet access in the
resistance in the Peruvian Amazon. Ecological Economics, 70(2), United States: 2012. Retrieved from http://www.census.gov/
207–218. doi:10.1016=j.ecolecon.2010.04.022 hhes/computer/publications/2012.html
Osmond, D., H., Pollack, L. M., Paul, J. P., & Catania, J. A. (2007). Vallilant, R. (2013). Commentary: Summary report of the AAPOR Task
Changes in prevalence of HIV infection and sexual risk behavior Force on Nonprobability Sampling. Journal of Survey Statistics and
in men who have sex with men in San Francisco: 1997–2002. Methodology, 1(2), 105–111. doi:10.1093=jssam=smt010
American Journal of Public Health, 97(9), 1677–1683. doi:10.2105= Way, N. (2006). The cultural practice of close friendships among urban
AJPH.2005.062851 adolescents in the United States. In X. Chen, D. French, & B. H.
Paquette, D., & De Wit, J. (2010). Sampling methods used in Schneider (Eds.), Peer relationships in cultural context (pp. 403–
developed countries for behavioural surveillance among men 425). New York, NY: Cambridge University Press.
who have sex with men. AIDS and Behavior, 14(6), 1252–1264. Wei, C., McFarland, W., Colfax, G. N., Fuqua, V., & Raymond, H. F.
doi:10.1007=s10461-010-9743-7 (2012). Reaching Black men who have sex with men: A
Pequegnat, W., Rosser, B. S., Bowen, A. M., Bull, S. S., DiClemente, comparison between respondent-driven sampling and time-
R. J., Bockting, W. O., . . . Horvath, K. (2006). Conducting location sampling. Sexually Transmitted Infections, 88(8),
Internet-based HIV=STD prevention survey research: Considera- 622–626. doi:10.1136=sextrans-2012-050619
tions in design and evaluation. AIDS and Behavior, 11(4), Whitehead, L. C. (2007). Methodological and ethical issues in
505–521. doi:10.1007=s10461-006-9172-9 Internet-mediated research in the field of health: An integrated
Pollack, L. M., Osmond, D. H., Paul, J. P., & Catania, J. A. (2005). review of the literature. Social Science and Medicine, 65(4),
Evaluation of the center for disease control and prevention’s 782–791. doi:10.1016=j.socscimed.2007.03.005
HIV behavioral surveillance of men who have sex with men: Yang, Z., Zhang, S., Dong, Z., Jin, M., & Han, J. (2014). Prevalence of
Sampling issues. Sexually Transmitted Diseases, 32(9), 581–589. unprotected anal intercourse in men who have sex with men
doi:10.1097=01.olq.0000175419.02839.d6 recruited online versus offline: A meta-analysis. BMC Public
Pollack, L. M., Osmond, D. H., Paul, J. P., & Catania, J. A. (2006). Health, 14, 508. doi:10.1186=1471-2458-14-508
Letter to the editor [Response to letter from Lansky et al.] Zavaleta, C., Fernández, C., Konda, K., Valderrama, Y., Vermund,
Sexually Transmitted Diseases, 33(4), 274–275. S. H., & Gotuzzo, E. (2007). Short report: High prevalence of
Ramirez-Valles, J., Heckathorn, D. D., Vázquez, R., Diaz, R. M., & HIV and syphilis in a remote native community of the Peruvian
Campbell, R. T. (2005). The fit between theory and data in Amazon. American Journal of Tropical Medicine and Hygiene, 76,
respondent-driven sampling: Response to Heimer. AIDS and 703–705.
Behavior, 9(4), 409–414. doi:10.1007=s10461-005-9031-0 Zwahlen, M., Spoerri, A., Gebhardt, M., Mäusezahl, M., Boubaker,
Raymond, H. F., & McFarland, W. (2009). Racial mixing and HIV K., & Low, N. (2007). Surveillance systems for sexually
risk among men who have sex with men. AIDS and Behavior, transmitted diseases in Switzerland. Sexually Transmitted
13(4), 630–637. doi:10.1007=s10461-009-9574-6 Diseases, 34(2), 76–80. doi:10.1097=01.olq.0000223248.96376.3b

411
Copyright of Journal of Sex Research is the property of Routledge and its content may not be
copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for
individual use.