Effectiveness of Respondent-Driven Sampling for Conducting Health Studies Among Undocumented Immigrants at a Time of Heightened Immigration Enforcement

Abstract

Aims:

This paper assessed the effectiveness of Respondent Driven Sampling (RDS) in recruiting undocumented Latinx immigrants for a prevalence health study at a time of heightened immigration enforcement.

Methods:

RDS was used to collect and analyze data from clinical interviews with 254 undocumented Latinx immigrant adults, enabling inference to a population of 22,000.

Findings:

45% of the sample reported having a chronic medical condition. The desired sample size was achieved and exceeded with 3 initial recruits and 10 waves of recruitment across 9 weeks. There was substantial cross-group mixing for recruitment in terms of sex and recency of immigration, which facilitated the emergence of diversity within recruitment chains. Primary factors that contributed to effective recruitment were location, flexibility, on-site childcare, and detailed explanation of the recruitment process.

Conclusion:

RDS is an effective recruitment method to study the health of undocumented Latinx immigrants, which is essential to informing intervention and policy.

Introduction

Undocumented immigrants (UIs) are at risk for diminished health outcomes in large part due to their immigration legal status. Specifically, existing research has demonstrated that UIs are at high risk for negative mental health outcomes (e.g., depression, anxiety, and post-traumatic stress disorder) and physical health outcomes (e.g., unmanaged/poorly managed diabetes, cardiovascular disease, kidney disease) [1–3]. Several reasons for this have been proposed with the most notable being restricted access to healthcare and greater, more prolonged exposure to high levels of distress as a result of their immigration legal status [4,5]. However, studies documenting reliable, population-based prevalence rates of health conditions and their determinants among UIs are largely non-existent in the current literature. This is partly due to barriers that limit the participation of UIs in research. Examples of such barriers are limited study opportunities and poor understanding of the research process, reluctance to disclose legal status, fear of stigmatization and/or retaliation, and demanding lifestyles and work schedules [6]. Identification of this immigrant population’s health status is needed to (a) ameliorate the negative health effects of harsh living environments, (b) recommend best practices for providers or organizations that come into contact with this population, and (c) inform interventions, advocacy, and policy efforts.

Most studies of UIs in the United States (U.S.) have relied on convenience samples that do not generate population estimates [7]. The primary methodology used to recruit UIs into studies includes the use of flyers and referral lists provided by clinicians, community-based agencies, social services organizations, schools/universities, immigration agencies, electronic bulletins, and migrant shelters [7]. Snowball sampling, a technique in which participants assist researchers in recruiting participants for a study, is also a common strategy. Although the aforementioned methods are useful for recruiting UIs into health studies, they are not without limitations. For example, recruitment from specific settings (e.g., church) may not provide a diverse sample given its exclusion of immigrants who do not frequent those settings [8]. Despite its ability to recruit broad social networks of UIs, random sampling is hard to accomplish via snowball sampling given that individuals with larger social networks may be overrepresented in the sample which may lead to bias in interpreting and generalizing study findings.

In epidemiological research, probability sampling is the gold standard for producing generalizable results. An effective probability sampling method to study hidden populations is respondent-driven sampling (RDS), a methodology that is based off of a mathematical model of the social networks that connect participants in a study [9]. In using RDS, researchers reach members of hidden populations by relying on the social networks of participants to create a structured referral system that in turn uses successive waves for participant recruitment. The first wave of participants recruits a second wave, with second wave then recruiting a third wave, and this method continuing on until the desired sample is achieved, and the later samples no longer mirror the demographic characteristics of the initial samples. The statistical theory of RDS proposes that with a large number of referral waves, the composition of the sample will become independent of the participants from which recruitment began. Thus, resulting in a statistical process that overcomes biases in the selection of non-random participants used in the first recruitment wave [9].

To obtain probability-sampling, RDS modifies traditional snowball sampling by increasing the breadth of the social networks captured by the sample and limiting the number of coupons that a participant receives (maximum of three). It also uses a systematic weighting scheme that generates population-based estimates based on a respondent’s social network size so that the probability of selection is proportional to the size of each participant’s social network (degree). Thus, each participant is weighted by the inverse of their probability of selection so that units with small chance of being selected (i.e., those with smaller social networks) are weighted more heavily than those with larger social networks [10]. Thus, although RDS is similar to other chain-referral sampling methods in that it begins with a convenience sample, the structured process used in recruitment allows for researchers to obtain less biased estimates of the desired population. A detailed description of RDS and its methodology has been described by Tyldum & Johnston [10].

The social network structure that is critical in the decision-making process for UIs to immigrate to the U.S. could suggest that RDS is an effective method for conducting improved health research with this hidden population. New immigrants rely on their social networks for economic, informational, and emotional assistance prior to, during, and post-immigration [11]. During times of high risk (e.g., anti-immigrant climate, changes in immigration policies), social networks of UIs may be hesitant to engage in the referral process or may not be dense enough to support RDS recruitment. In this paper, we present findings of the effectiveness of RDS in recruiting undocumented Latinx immigrants for a health study, who resided in high-risk neighborhoods during 2014, a year of record number of deportations in the U.S. [12]. High-risk neighborhoods are defined as extremely conservative areas with substantial Immigration and Custom Enforcement (ICE) presence, and strong opposition/punitive actions against UIs [13]. Specifically, this paper focuses on (a) describing the process and logistical considerations in adapting the use of RDS to recruit undocumented Latinx immigrants for a health study at a time of high risk; (b) outlining RDS recruitment patterns; and (c) identifying how social networks in this population influence RDS effectiveness.

Methods

Formative Research

Consistent with RDS recommendations, we conducted extensive formative research to (a) learn about the social networks of UIs, (b) to inform the logistics of our study, and (c) to identify the initial participants (i.e., seeds) for the recruitment process [10]. Formative research included a systematic review of the literature [7], field observations, seven focus groups with UIs in the target region, 20 in-depth key informant interviews, ethnographic mapping, and pilot testing of the health interview. To avoid biasing the RDS design, none of the participants in the formative research participated in the RDS study. Instead, key informants provided contact information for three UIs that served as seeds to start the RDS recruitment process. All phases of this study, including formative research and RDS, were approved by the {blinded for review} Institutional Review Board.

Recruitment and Data Collection

To estimate sample size, an a priori power analysis was conducted using OpenEpi, Version 3.01 [14]. Our target sample size was 190 participants and was designed to detect a 14% prevalence of diagnosing a disorder with 95% confidence that the prevalence estimate would be within 7% of the true prevalence value [15]. Sixty-seven additional participants were recruited beyond the target sample size to increase the number of recruitment waves. Participants were undocumented Latinx adults residing in the target high-risk region. To assess eligibility based on immigration legal status, a rule-out system was used [15]. Participants were asked a series of yes/no questions outlining current legal statuses with the exception of undocumented status. Respondents who answered “no” to all questions were coded as undocumented. The target region was a medium-sized city located near the California-Mexico border. Community estimates suggest that between 15% to 25% of the population in this region is undocumented [16].

Data were collected between November 2014 and January 2015. Recruitment began with three seeds varying in demographic and immigration characteristics, which accelerated the rate at which the study sample reached equilibrium (i.e., the point in recruitment when the proportion of a sample characteristic is independent from the seed characteristics) (see Table 1) [17]. Equilibrium is attained when the proportions remain stable within 2% of the sample proportion [17]. After joining the study, the seeds and subsequent eligible participants were given a one-page step-by-step visual diagram that explained the recruitment process and three uniquely coded, non-replicable invitations to recruit no more than three other undocumented Latinx immigrants who they knew well and would be likely to participate. Each invitation displayed a serial number used to link respondents to seeds and referral chains. Invitations also included the study name (Proyecto Voces), study location, appointment times, a contact number, and an expiration date (two weeks from when the invitation was issued). Participants could either drop-in for interviews or make an appointment. For safety, neither the description of the study, nor the requirements for participation were provided in the invitation. Instead, recruiters explained the study, the incentives, and the recruitment process using the one-page step-by-step diagram. As the sample reached the desired sample size, the number of referral invitations given was gradually reduced until no more invitations were distributed. The study terminated once everyone with a valid invitation was interviewed. Participants received a primary incentive of $30 after completing the clinical interview. They also received a secondary incentive of $10 for each recruited participant that they referred to the study and who completed an interview. Participants could refer a maximum of three participants for the study. After the interview, participants completed a debrief questionnaire and were given health information materials including referrals to safe, low-cost, and nearby Spanish health services.

Table 1.

Seed characteristics (N =3)

	Seed 1	Seed 2	Seed 3
Sex	Female	Female	Male
Age	52	27	37
Marital status	Single/Divorced	Living as married	Married
Education	≥ High school	< High School	≥ High school
Employment	Part-time	Not working	Full time
Type of employment	Sales	Homemaker	Construction
Household income	≤ $500	$2001 – $3000	≥ $3001
Household size	2	5	3
English proficiency	Little	None	Good
Age of arrival to U.S.	27 years	15 years	23 years
Years in U.S. category	> 20 years	11 to 20 years	11 to 20 years

Tracking of invitations and clinical interviews

Excel was used to track invitations given to participants. The link between the recruiter and the recruit is documented by matching the recruitment invitation to the invitation’s serial numbers. Data entered into the Excel sheet included invitation number, date of the interview, sex of participant, number of invitations given, invitation expiration date, dates of primary and secondary incentive collection, and a unique code created by the participant to cash secondary incentives. Once the code was verified, participants received payment for recruited peers.

Face-to-face clinical interviews were conducted in Spanish by trained research assistants under direct supervision of trained clinicians. Data were collected using a computer assisted personal interviewing system (CAPI), which was used to minimize missing data, error, and in increase efficiency [18, 19]. Interviews lasted between 1 and 3 hours, depending on the extent of health concerns reported. Information gathered included demographics, immigration history, and physical/mental health evaluations. To prevent duplicate participation, the number of interviewers available to interview on any given day was limited and the coupons were made so that they would be difficult to reproduce. Verbal informed consent was obtained prior to participation. The study was conducted at a safe location identified during formative research.

Data Analysis

RDS Analyst was used to calculate recruitment matrices based on relevant characteristics of the recruiter and recruit, sample proportions, population prevalence estimates and 90% confidence intervals [20]. The diagram of recruitment chains was created using NetDraw version 2.118 [21] (Figure 1).

Figure 1.

Recruitment chains.

RDS Assumptions.

RDS is based on functional and analytical assumptions based on social science statistics [22]. First, RDS is based on reciprocity of ties. In other words, respondents should know one another as members of the hidden population so that recruitment ties are reciprocal. The target population must be socially networked and be able to recruit others within that social network. We assessed this by evaluating closeness and type of relationship between participants and their recruiter. Second, networks should be dense enough to sustain a chain-referral process to meet the desired sample size. This assumption is met by specifying a time window need to sustain the referral process (e.g., expiration date on invitations) and was verified by visually inspecting the wave recruitment process. Third, sample size needs to be small in relation to the population size. If the sample size required to create valid estimates is large and the population size is small, achieving the target sample size would be difficult. Thus, we determined the required sample size based on the size of the population estimate in the target region. Fourth, respondents are recruited from their network at random. It is difficult to ensure random recruitment in RDS, although non-random recruitment will not necessarily bias the RDS estimator as long as recruitment is not correlated with any variable important for estimation [23]. Random recruitment was verified empirically by assessing homophily, a measure of the likelihood that recruiters recruit individuals with similar characteristics to themselves [24]. Homophily scores range from −1 to 1, with a score of 1 indicating that 100% of participants in that subgroup were recruited entirely from within their subgroup, whereas a score of −1 indicates that 100% of participants in that subgroup were recruited from peers outside of their subgroup. Scores closer to zero are desirable meaning that participants were more likely to recruit peers at random [24]. We used recruitment matrices to describe recruitment patterns based on sex, country and region of origin, and recency of immigration/time living in U.S. as indicators of social connectedness.

Measures

Three questions were used to map recruitment chains and to calculate RDS weights that produced population estimates from the sample. These included (a) estimated size of the respondent’s personal network that is undocumented and that the respondent has seen or interacted within the past month (social network size); (b) relationship of the participant to the referral source; and (c) length of time that the participant has known the referral source. These questions provided a basis for weighing chain-referral data that would otherwise have been biased by differential recruitment [25]. In addition, socio-demographic and immigration information, as well as comprehensive health assessment questionnaires using self-report were collected. Immigration related information included questions pertaining to length of time in the U.S., acculturation, history of deportation, history of trauma and migration-related loss across different stages of the migration process, migration living difficulties, and perceived effect of undocumented status on self-image. Health assessment questionnaires included information about physical and mental health history and current symptoms. Additional details about some of the clinical measures used have been described elsewhere [12]. All questions were translated in Spanish using established methodology [26].

Results

Participants

We recruited 254 undocumented Latinx immigrants with 10 waves of recruitment over a 9-week period. A total of 549 invitations were distributed, of which 47% were redeemed. Of the 257 redeemed invitations, 256 were eligible, 254 provided consent, and 254 completed the interview. Participants ranged in age from 18 to 74, with the majority being middle-aged (54%). The average age was 38 years (SD = 11.2). The majority of the sample was female (69%), partnered (68%), had children (86%), had less than high school education (65%), and were employed (52%). Among employed participants, more than half (56%) reported working in construction, cleaning, or maintenance industries. Most participants spoke little to no English (67%) and were of Mexican origin (98%). Years spent living in the U.S. ranged from less than one to 54 years with an average of 16 years (SD = 7.9). Most participants immigrated to the U.S. between the ages of 11 and 30 years and reported living in mixed status families (72%) (Table 2).

Table 2.

Sample and population characteristics.

Factor	Sample (n = 254)		Population (N = 22,000)
	n	%	%	95% CI
Sex
Women	176	69.3	66.7	56.4	77.1
Men	78	30.7	33.3	22.9	43.6
Age (years)
18 – 25	35	13.8	16.2	86.0	23.8
26 – 35	63	24.8	21.8	14.3	29.2
36 – 45	105	41.3	38.3	28.0	49.0
≥ 46	51	20.1	23.7	14.8	32.3
Education
< HS (no graduation)	165	65.0	65.5	56.5	74.8
≥ HS graduate	89	35.0	34.6	25.2	43.5
Employment
Not working outside home	123	48.4	49.6	40.1	59.2
Working (Full/part-time)	131	51.6	50.4	40.8	59.9
Marital status
Single	81	31.9	29.0	20.8	37.1
Married/other	173	68.1	71.0	62.9	79.2
Parental status
No children	35	13.8	16.2	9.4	22.9
≥ 1 child	219	86.2	83.9	77.1	90.6
Immigration Factors
Time in U.S.
≤ 10 years	59	23.2	17.6	11.4	23.7
11 – 20 years	126	49.6	52.8	44.0	61.7
> 20 years	67	26.4	29.6	19.8	39.4
Mixed status family
Yes	182	71.77	64.8	55.2	74.4
		1.7
Deportation
Yes	55	21.7	24.8	16.8	33.1
Physical Health
Chronic Medical Condition*
Yes	115	45.3	45.2	38.3	52.2
Heart disease
Yes	6	2.4	2.2	0.3	4.1
Hypertension
Yes	36	14.2	14.0	9.1	18.7
Hyperlipidemia
Yes	35	13.8	14.2	9.5	18.8
Asthma
Yes	10	3.9	4.6	1.8	7.5
Allergies
	50	19.7	19.3	14.2	24.3
Yes
Respiratory disease
Yes	9	3.5	3.3	1.2	5.4
Diabetes
Yes	21	8.3	8.4	4.6	12.1
Digestive disorders
Yes	45	17.7	18.1	12.7	23.4
Neurologic conditions
Yes	3	1.2	1.7	0.3	3.7
Cancer
Yes	2	0.8	0.6	0.00	0.02
Health Satisfaction
Dissatisfied	29	11.4	15.2	7.7	22.8
Neither satisfied/dissatisfied	80	31.5	32.2	23.8	40.5
Satisfied	145	57.1	52.7	43.2	62.1

Note.

^*Chronic health conditions were assessed via self-report using the World Health Organization Composite International Diagnostic Interview (CIDI) Chronic Conditions Module Version 3.0. (24) Also, none of the participants reported having HIV or hepatitis; thus, these medical conditions are not reported in this table.

Health Status

Overall, 45.2% of the participants self-reported having a chronic medical condition. On average, participants reported having 0.9 chronic medical conditions (SD=1.19). Of the medical conditions reported, the most prevalent were allergies (19.3%), digestive disorders (18.0%), hyperlipidemia (14.2%), hypertension (14.0%), diabetes (8.4%), asthma (4.6%), respiratory disorders (3.3%), and heart disease (2.2%). Of note, 52.7% reported being satisfied with their health (Table 2).

Field Experiences

There were 18 recruits in week 1, with an average of 36 recruits per week between weeks 2 and 7. Peak recruitment occurred in week 6 with 52 recruits and significantly slowed during weeks 8 and 9 with an average of 9 recruits during those two weeks. Primary factors that contributed to recruitment success were (a) the location of the interview site, which was easily accessible by public transportation and had a low record of immigration raids; (b) flexibility in scheduling appointments, including accepting drop-ins and extended hours to accommodate busy life/work schedules; (c) the provision of on-site childcare; and (d) the provision of a step-by-step printed diagram explaining the recruitment process to recruiters. Most participants were accompanied by a family member or friend (61%), and most agreed to refer additional participants to the study (97.2%). Reasons provided for completing the study included wanting to help others (26.2%), curiosity (23.1%), the monetary incentive (22.6%), looking for resource referrals (15.9%), and recommended to do so by a close friend/relative (12.2%). A small number of participants expressed that their peers did not accept the invitation out of fear that study staff were affiliated with immigration authorities.

There were significant differences in the productivity of the three seeds’ recruitment chains (see Figure 1). Chain 1 comprised of 123 participants (48.4% of the sample) over 10 waves, Chain 2 of 54 participants (21.3%) over 8 waves, and Chain 3 of 77 participants (30.3%) over 9 waves. Most participants were recruited by friends (60.7%), family members (30.8 %), neighbors (6%) and co-workers (1.6%). Less than 1% were recruited by strangers. Most participants had a close relationship with their recruiter (86.6%), with the majority interacting with their recruiter at least once a week (82.3%). Also, most participants had known their recruiter for 5 years or longer (56.7%), with less than 13% of participants having known their recruiter for less than a year.

Cross group recruitment and homophily

Cross-sex recruitment was substantial (41%), implying that recruitment chains did not become trapped within a single sex group. Although the majority of participants in the study were women, women recruited men 30% of the time, while men recruited women 67% of the time. The tendency to recruit across gender was reflected in the low homophily scores. Men had a homophily score of H = 0.11, meaning that men recruited as though 11% of their social ties were to other men, and 89% of their social ties were formed independently of sex. The corresponding score for women was H = 0.07. The mean network size for men was 20 and 15 for women (Table 3).

Table 3.

Recruitment patterns by sex.

Sex of Recruiter	Sex of recruit		Total
	Man	Woman
Man	24 (33%)	49 (67%)	73
Woman	53 (30%)	125 (70%)	178
Total	77	174	251*
Homophily	0.11	−0.07
Mean network size	20.4	15.10

^*Initial seeds (3) not included in analyses because they were recruits from formative research.

Regarding recruitment patterns by nationality, the majority of the sample was of Mexican origin. Mexican nationals recruited participants of Mexican origin 98% of the time (H = 0.35). The few Central American participants in the study (2%) also recruited Mexican nationals (H = 0.23). The mean network size was 17 for Mexican nationals and was 14 for Central American nationals. Given the diversity in the cultural and contextual backgrounds of Mexican immigrants from different Mexican regions, we explored recruitment patterns by Mexican region of origin which revealed a tendency to recruit Mexicans from the same country region. For instance, immigrants from central Mexico recruited participants from central Mexican states 63% of the time, while immigrants from southern Mexico recruited participants from southern Mexican states 48% of the time. Likewise, immigrants from northern Mexico recruited participants from northern Mexican states 44% of the time. The tendency to recruit along similar Mexican regions was reflected in the moderately elevated homophily scores. Mexican immigrants from central Mexico had a homophily score of H = 0.21, meaning that these immigrants recruited as though 21% of their social ties were to Mexican immigrants from their same central region, and 79% of their social ties were formed independently of where immigrants were from. The corresponding score for Mexicans from northern Mexico was H = 0.32 and was H = 0.30 for Mexicans from southern Mexico. The mean network size was 8 for Mexicans from northern and southern Mexico, and was 6 for Mexicans from central Mexico.

Recruitment patterns by recency of immigration showed a proportional within-group recruitment by time living in the U.S. Established immigrants, who comprised an estimated 76% of the population, recruited other established immigrants 78% of the time. Similarly, recent immigrants, who comprised an estimated 24% of the population, recruited other recent immigrants 29% of the time. Nonetheless, there was substantial cross-group recruitment by recency of immigration (33%). The tendency to recruit across recency of immigration categories was reflected in the low homophily scores. Recent immigrants (<10 years in U.S.) had a homophily score of H = 0.09, whereas the corresponding score for established immigrants (≥10 years in U.S.) was H = 0.01. The mean network size was somewhat similar for both groups (18 for recent immigrants; 16 for established immigrants).

Discussion

This study described and assessed the effectiveness of RDS in recruiting undocumented Latinx immigrants for a health study at a time of heightened immigration enforcement. To our knowledge, this study is the first to provide population-based prevalence estimates for various health outcomes among UIs. For instance, results from this study showed that 45% of participants self-reported having a chronic medical condition, which is slightly above the prevalence estimate for the general population in the target area (34% to 42.6%) [26]. Likewise, the prevalence estimate for self-reported diabetes in this study was 8.4%; also, higher than the prevalence of diabetes among Latinxs in the target region (7.5%) [26]. In this study, we relied on self-report; thus, the aforementioned estimates are likely conservative and actual estimates may be higher, which is of concern. Given that UIs have limited access to healthcare, findings from this study emphasize the need for policies (e.g., access to purchasing insurance in the marketplace) and non-traditional sources of service delivery (e.g., collaboration with faith-based organizations, non-profits, and established community partnerships) that may facilitate access to healthcare for this population.

Fear of deportation and limited knowledge about the importance of research is a significant barrier to the participation of UIs in research studies. However, our results suggest that RDS provides an effective method for rapidly recruiting a large sample of UIs for health studies, even during times of heightened immigration enforcement. Using only three seeds, quick recruitment of 254 undocumented Mexican immigrants was achieved over a 9-week period. The cost of conducting this study was relatively low with a research team of only six people and approximately $10,000 spent on incentives. Our recruitment efforts showed that careful identification of diverse seeds during formative research was essential for recruiting a sample of UIs that varied in sociodemographic and immigration characteristics. Our study demonstrated that RDS yields a sample with sufficient sociometric depth to attain diversity after a relatively small number of recruitment waves (n=10). Moreover, essential to the success of this RDS study was the close collaboration with leaders in the target community that began a year prior to the start of the study. The use of a step-by-step printed diagram explaining the recruitment process was also essential. The aforementioned recommendations are particularly relevant for future investigators that wish to conduct health research in undocumented communities.

UIs are an at-risk population for diminished physical and mental health given the chronic and multiple stressors that these immigrants face, along with having limited access to health and social services [7, 12]. Yet, information about the health status of UIs and the many barriers that they face in accessing healthcare is limited. This study shows that RDS is an effective methodology for filling the existing gaps in the literature, for elucidating the health concerns of UIs, and for identifying the facilitators that improve the wellbeing of this vulnerable population. For instance, RDS can be helpful in providing population-based estimates for health concerns and chronic illnesses affecting this population (e.g., obesity, anemia, asthma, diabetes, cardiovascular disease), along with identifying their impact on UIs functional ability and mental health. The incorporation of epigenetic biomarkers into the aforementioned studies would also facilitate the study of mechanisms influencing the risk and resilience processes in this marginalized population, which is essential for informing advocacy and policy efforts. Another effective use of RDS with UIs would be in the identification of barriers to healthcare access, along with a deeper understanding of preferences for alternative sources of care and prevalent beliefs of health and illness that may be culturally and contextually sensitive (e.g., home remedies, healers or curanderos, healing practices) to the health needs of UIs. Lastly, RDS may be used to study other topics relevant to this population, such as the effects of context (e.g., trauma, family dynamics, social and work environments) on the wellbeing of UIs. The aforementioned knowledge is essential to informing the development of interventions, resources, practices, and policies needed to reduce risk and prevent further harm in this immigrant population.

Limitations

The adequate level of cross-group mixing by sex and recency of immigration facilitated the emergence of diversity within each recruitment chain and reduced the number of waves needed to achieved equilibrium. However, cross-group mixing in terms of nationality was low, which resulted in a sample predominately of Mexican origin. Nonetheless, our sample is consistent with the demographic distribution of undocumented Latinxs in the target area. Additional studies to assess the effectiveness of RDS for recruiting a more diverse sample of UIs from different nationalities are needed. This could be done by reproducing the present study in areas known to vary in the nationality of their Latinx population. Further, this study mostly represented established UIs with U.S.-born families. These estimates may differ for recent immigrants and those without U.S.-born families. Finally, health outcomes in our study were assessed via self-report and actual estimates may be higher.

Conclusion

UIs are rarely included in health research. During times of heightened immigration enforcement, fear among UIs increases, pervades in the community, and creates a hostile environment. This hostile environment hinders UIs’ willingness and desire to participate in research studies and to exchange information with people outside of their social network. This study provides evidence on the effectiveness and cost efficiency of RDS to overcome the aforementioned challenges and to facilitate recruitment for the study of health among UIs. This is essential for informing interventions, advocacy, policy efforts, and best practices among healthcare providers.