Search

Respondent-driven sampling (RDS) is a chain-referral network-sampling method, specifically designed for studying rare, marginalized, or otherwise "hidden" populations that are difficult to sample using traditional probability-sampling methods. Despite its wide application, challenges are present in RDS data collection and analysis. This dissertation presents improvements in implementing RDS as a sampling technique and an analytical tool, thereby enabling more reliable and insightful inferences about the characteristics of hidden populations that lack easily constructed sampling frames. Unlike standard probability samples, sample sizes are not easily controlled in RDS studies. RDS operates by generating chains of respondents who are invited to participate by previous respondents through coupon-based referrals. As recruitment chains start from a small sample of initial seeds and proceed beyond researchers' control, the resulting sample size becomes a random variable. Chapter II explores the sample size distribution in RDS as a function of seed numbers, coupon numbers issued to respondents, and response probabilities to invitations. Because RDS is usually used where traditional sampling frames are absent, Chapter III investigates population size estimation using RDS data. We develop a population size estimator within a Bayesian inferential framework, leveraging information on duplicate occurrences (when an individual is recruited more than once) and participants' recruitment sequences and network sizes. Through simulation and case studies, we demonstrate that our method yields credible and more reasonable lower-bound population size estimates than existing competing methods, especially in common small sampling fraction settings. Chapter IV delves into the individual network size (degree) measures in RDS studies. RDS relies on degree values reported by participants to adjust for their unequal sampling probabilities to generalize findings. However, this self-reporting often exhibits substantial measurement errors, evident in the unusual frequency of multiples of five and implausibly large values. To address this, we propose a novel individual degree estimator based on a latent variable model of the true degree. This model effectively accounts for response errors through a reporting mechanism and incorporates recruitment information and external demographic profiles, resulting in improved inference. Through case studies and simulations, our method delivers more accurate and reliable degree estimates than competing methods.

Barriers to safe, effective, and affordable methods of abortion include restrictive laws, unwilling or untrained providers, long wait times, high costs, lack of services in the public sector, and abortion stigma. These barriers prevent individuals from exercising their fundamental human right to choose if and when to have a child, and threaten bodily and reproductive autonomy. Self-managed abortion with medication—defined as ending one’s own pregnancy outside of a formal healthcare setting using misoprostol alone or in combination with mifepristone—can be safe and effective; however, there is a lack of research on its incidence, safety, and effectiveness, and there are considerable methodological challenges in its assessment.This dissertation aims to address key methodological challenges in the measurement of self-managed abortion via the following approaches: development of a conceptual framework for measuring abortion complications; assessment of a novel application of respondent-driven sampling (RDS) to study the incidence of abortion; and estimation of the effectiveness of self-managed medication abortion in probabilistic bias-adjusted models. Chapter 1 describes the development of a framework for measuring complications and outcomes from medication abortion based on self-report, informed by thematic analysis of in-depth interviews from people with various abortion experiences in Argentina, Indonesia, Mexico, Nepal, Nigeria, and Sierra Leone. Findings from this analysis demonstrated that individuals describe and quantify their experiences with bleeding and cramping in varied ways, and highlights the need for a person-centered framework that emphasizes the individual’s preferences around medical care seeking. Chapter 2 is a methodological assessment of an RDS study on the incidence of abortion in Soweto, South Africa, and explores the implications of potential violations of RDS assumptions on incidence estimation. In this study, several key assumptions of RDS were not met, yielding potentially biased estimates of abortion incidence. Chapter 3 utilizes data from a prospective observational study assessing the effectiveness of self-managed medication abortion among callers to accompaniment groups and safe abortion hotlines in Argentina and Nigeria, and demonstrates a Monte Carlo Sensitivity Analysis approach to adjusting for misclassification and selection bias. After adjusting for potential misclassification, selection bias, and enrollment of ineligible participants, self-managed medication abortion remains highly effective, conditional on our assumptions around the chosen bias parameters.Findings from this dissertation will contribute to the development of a self-report questionnaire to measure complications and outcomes from medication abortion; highlights the pitfalls of respondent-driven sampling and offers potential remedies; and can be used to inform future analyses of the effectiveness of self-managed medication abortion based on observational data.

Coercive work occurs when people are required to work under threat of criminal legal repercussions should they fail or refuse to do so. As a requirement of their parole supervision, people on parole must search for and obtain employment; should they fail, they may face criminal legal sanctions including parole violation and reincarceration. In this dissertation, I document the prevalence of coercive work for people on parole in Los Angeles County, identify the mechanisms through which coercive work operates, and illustrate the exploitative employment conditions of coercive work. I use a novel “Hybrid-RDS” methodology to recruit participants and collect and analyze survey data from 520 people on parole in Los Angeles County and to estimate the prevalence of coercive work experiences among the county’s total parole population. To better understand the mechanisms of coercive work, these innovative “Hybrid-RDS” methods are supplemented with additional data from 40 qualitative interviews with the same population, ethnographic observations, and an analysis of official administrative documents from the California Department of Corrections and Rehabilitation and others. The findings demonstrate that, though employment is not a formal condition of parole, people on parole largely perceive employment to be a requirement – one that is backed by the threat of sanctions for failure or refusal to comply. This parolee perception is constructed through official parole rules and written documents, verbal communications by parole agents around expectations and what work is “acceptable,” and parolee past experiences of coercion, threats of sanctions, parole violations, and reincarcerations related to work. Though these parole conditions play a critical role in constructing the requirement that people on parole find employment, other conditions of parole function as barriers to work, creating a “double bind” around employment for people on parole; parole conditions constrict parolee movement and agency to such an extent that they often obstruct people’s attempts to acquire the work they are required to obtain. In the face of these requirements, barriers, and threats around employment, parolees are channeled down into exploitative working conditions in the low end of the labor market or in informal work, often outside the bounds of legal labor protections around wages, faithful representations of job duties, and worker safety. Because of their legally precarious status and the employment “double bind,” people on parole often accept these problematic working conditions in response to parole-initiated pressures to work and the perpetual threat of punitive repercussions and a return to prison.

Background: Despite a high prevalence of cigarette smoking and smoking-related morbidity and mortality among U.S. American Indians (AI), few studies have investigated tailored community and workplace interventions, including cessation programs. The purpose of this study was to determine the association of (1) the presence of a workplace smoke-free policy and (2) perceived social support among family and friends for quitting smoking with current smoking in a representative sample of urban AI adults. Methods: Data collected using respondent-driven sampling (RDS) from the 2011 Tribal Tobacco Use Prevalence Study was used to analyze the risk of being a current smoker when exposed to (1) workplace smoke-free policies and (2) perceived social support using two log multinomial regression models, adjusted for potential confounders based upon exposure-specific Directed Acyclic Graphs. Study-eligible participants were: self-identified AI, aged 18+, and residents of Hennepin or Ramsey Counties, Minnesota with a separate residence from the recruiter. Data were weighted to account for variation in participant network size using RDS Analysis Tool, V. 5.6., and participant demographics using U.S. Census 2010, yielding 940 of 964 completed interviews with full demographics. Results: Lack of a workplace smoke-free policy was associated with a 36 percent increase in risk of being a current smoker (PR = 1.36, 95% CI = 1.29 to 1.42) compared to a former smoker and a 41 percent decrease in risk of being a former smoker (PR = 0.59, 95% CI = 0.55 to 0.64) compared to a nonsmoker. Additionally, exposure to quite a bit or a lot of social support for quitting or staying smoke-free was associated with an 11 percent decrease in the risk of being a current smoker (PR = 0.89, 95% CI = 0.86 to 0.91) and an eight percent decrease in risk of being a former smoker (PR = 0.92, 95% CI = 0.88 to 0.95) compared to being a nonsmoker, respectively. There was also an eight percent decrease in risk of being a current smoker (PR = 0.92, 95% CI = 0.90 to 0.94) compared to a former smoker with exposure to quite a bit or a lot of social support for quitting or staying smoke-free versus no social support. Conclusions: Further study of Minnesota AI work environments and strategies for encouraging employers to implement and enforce complete smoking bans is warranted. AI smoking cessation programs may benefit from inclusion of family and friend support mechanisms.

Respondent-driven sampling (RDS) is a network sampling methodology used worldwide to sample key populations at high risk for HIV/AIDS who often practice stigmatized/illegal behaviors and are not typically reachable by conventional sampling techniques. In RDS, study participants recruit their peers to enroll, resulting in a sampling mechanism that is unknown to researchers. Current estimators for RDS data require many assumptions about the sampling process, including that recruiters choose people from their network uniformly at random to participate in the study. However, this is likely not true in practice. We believe that people recruit based on observable covariates, such as age, frequency of interaction, geography, socioeconomic status, or social capital.To model preferential recruitment, I develop a sequential two-sided rational-choice framework, referred to as the RCPR model. At each wave of recruitment, each recruiter has a utility for selecting each peer, and symmetrically each peer has a utility for being recruited by each recruiter. Each person also has utilities for selecting themself (not recruiting or not participating). People in the network behave in a way that maximizes their utility given the constraints of the network and the restrictions on recruitment. Although a person's utility is not observed, it can be modeled as a linear combination of observable nodal or dyadic covariates plus unobserved pair-specific heterogeneities. This framework allows generative probabilistic network models to be created for the RDS recruitment process. The models can incorporate observable characteristics of the population and have interpretable parameters. It greatly increases the sophistication of the modeling of the RDS sampling mechanism. Inference can be made about the preference coefficients by maximizing the likelihood of the observed recruitment chain given the observed covariates. As the likelihood is computationally intractable, I develop a Bayesian framework where inference is made feasible by approximating the posterior distribution of the preference coefficients via a Markov chain Monte Carlo algorithm. Each update step samples new values of the preference coefficients and utilities via Metropolis-Hastings, subject to constraints. New prevalence estimates can be calculated be generating many recruitment chains from the population using the RCPR coefficients, then directly obtaining the first-order and second-order inclusion probabilities. This framework allows the incorporation of covariates we think effect recruitment into the sample weights.

Respondent-driven sampling, or RDS, is used to draw samples from hard-to-reach or marginalized populations and to make inferences about the populations based on the samples. Such sampling begins with an initial, or "seed,"' sample from the population of interest. It then exploits the networked structure of those populations, relying on the members themselves to recruit further members of the population for the sample. A number of estimators for RDS have already been developed. Each estimator is motivated by a model that makes a number of assumptions about seed selection, respondent behavior, and certain properties of the underlying social network itself. In a series of articles, Gile and Handcock have used the statnet package in R to simulate respondent-driven sampling under a variety of conditions.They then use these simulations to examine the bias and variance of different estimators when assumptions are not perfectly or at all fulfilled.The goal of this project is twofold. First, the original R code to simulate respondent-driven sampling is quite slow. In the statnet package RDSdevelopment, we have written C code to duplicate and extend the functionality of the R code. The new code is considerably faster than the old R code. Second, we examine the sensitivity of current estimators to a previously unstudied aspect of respondent behavior: how accurately respondents report their number of contacts in the network, also known as their "degree." The two networks we consider are a previously simulated network, fauxmadrona, and the Project 90 network. The latter is constructed from an actual population of heterosexuals at high-risk for HIV living in Colorado Springs, CO, in the late 1980s and early 1990s. We carry out simulations on both networks using statnet in which we simulate imperfect recall of degree. Under conditions of imperfect recall, estimators tend to increase in variance as recall erodes. There is also change in bias, though the direction of change varies from estimator to estimator and is not monotonic with the increase in recall error. Finally, we introduce a variation on the current estimators by replacing reported degree by reported rank of degree in each estimator formula. Rank of degree is calculated by a carrying out a rank transformation on the reported degree distribution of each sample of individuals from the networked population. A rank transformation of a set of numbers maps each member of a set onto its rank with respect to the other members of the set. A number of rank transformations are possible. The rank transformation known as "standard competition" degrades estimator performance on the fauxmadrona and the Project 90 network. The dense rank transformation leaves estimator performance mostly unchanged, at least on the networks treated by the thesis.

Data that can be represented as a network, where there are measurements both on units and on pairs of units, are becoming increasingly prevalent in the social sciences and public health. Homophily in network data, or the tendency of units to connect based on similar nodal attribute values (i.e. income, HIV status) more often than expected by chance is receiving strong attention from researchers in statistics, medicine, sociology, public health and others. Respondent-Driven Sampling (RDS) is a link-tracing network sampling strategy heavily used in public health worldwide that is cost efficient and allows us to survey populations inaccessible by conventional techniques. Via extensive simulation we study the performance of existing methods of estimating population averages, and show that they have poor performance if there is homophily on the quantity surveyed. We propose the first model-based approach for this setting and show its superiority as a point estimator and in terms of uncertainty intervals coverage rates, and demonstrate its application to a real life RDS-based survey. We study how the strength of homophily effects can be estimated and compared across networks and different binary attributes under several network sampling schemes. We give a proof that homophily can be effectively estimated under RDS and propose a new homophily index. This work moves towards a deeper understanding of network structure as a function of nodal attributes and network sampling under homophily.

People who inject drugs (PWID) are greatly affected by blood borne viruses (BBV). Behavioural surveillance, which consists of ongoing cross-sectional surveys of at-risk populations, is a core component of BBV prevention. Respondent-driven sampling (RDS) is a relatively new sampling method, developed to sample hard-to-reach populations, such as PWID. This thesis aimed to describe the role of RDS in the behavioural surveillance of PWID. An RDS study of PWID was conducted in 2009 in Sydney, Australia. Five seeds were selected to start the recruitment, which led to a sample of 261 participants over 16 waves. To achieve the thesis aim, I examined the benefits and challenges of conducting RDS with minimal and existing resources, the ability to use RDS data to calculate unbiased population estimates, the characteristics of the RDS study’s sample as compared with those from separate service-based sampling studies, and whether RDS network data can add to the usefulness of a behavioural surveillance system. While there were benefits to conducting RDS with existing resources, there were also challenges related to crowd control and privacy issues. An examination of RDS requirements and assumptions found a lack of random recruitment and a low proportion of coupons returned, which suggested that population estimates were biased. In a comparison of RDS participants with participants from two separate service-based sampling studies, no marked differences in demographics or risk practices were found. As well, compared to pharmacy-based sampling, RDS was 80% less likely to capture participants with small injecting networks, a group found to be of interest to behavioural surveillance. Finally, using the network information gathered through RDS, it was found that closeness, Aboriginal origin status and hepatitis C testing were related to sharing injecting equipment within dyads, which were formed by linking participants’ information with that of their recruiter. The results indicate that in locations where there is good coverage by services for PWID, RDS should be best viewed as a complementary tool, where it can be used on an intermittent basis to evaluate existing sampling methods and to add to the understanding of how network characteristics shape risk practices.