Sampling bias in Brazilian studies on transgender and gender diverse populations: the two-step measure for assessing gender identity in surveys

Correctly recognizing gender identity in population-based surveys is essential to develop effective public health strategies to improve the living conditions of transgender and gender-diverse populations, as well as to adequately col-lect data on cisgender individuals. This study aims to present the two-step measure as the best strategy for assessing gender identity in Brazilian surveys, thus we performed two separate analyses. Firstly, we conducted a systematic review concerning HIV-related care among Brazilian transgender and gen-der-diverse populations to assess the strategy used to identify participants’ gender identity. Secondly, we re-analyzed data from a recent survey that included Brazilian transgender populations, comparing characteristics and health outcomes from the sample identified by single-item and by the two-step measure. Concerning the systematic review, from 6,585 references, Brazilian research teams published seven articles, and only one study used the two-step measure. Regarding this survey, the two-step measure recognized 567 cisgender and 773 transgender and gender diverse participants among the 1,340 participants who answered the questionnaire, whereas the single-item measure was able to recognize only 540 transgender and gender diverse people. Furthermore, 31 transgender women self-identified as “transgender men” on the single-item measure. Therefore, although scarcely used in Brazil, the two-step measure is a more accurate strategy to recognize gender identity. Gender Identify; Surveys and Questionnaires; Systematic


Introduction
In the early 1990s, transgender appears as an umbrella term encompassing a wide spectrum of gender identity variations that share the incongruence between one's current gender identity and one's gender assigned at birth. Although a subgroup of transgender people may seek medical gender affirmation -such as hormonal therapy and surgical procedures -transgender is not a nosological classification 1,2 . Gender identities as social categories vary significantly across cultures. In Brazil, for example, subgroups of transgender populations may self-identify as travestis and/or trans(sexual) people 3 . Both terms have strong cultural roots and political meanings outside the scope of this study 4 . Travesti is a Latin American gender identity generally understood as a person who performs roles socially perceived as feminine without necessarily changing their primary sexual characteristics 5 .
Bearing in mind the well-known health disparities that transgender populations endure -such as HIV infection prevalence and difficulties to access HIV-related care -4,5,6 focused public health strategies are essential 6 . In many HIV studies, transgender women were included along with men who have sex with men, as it was stipulated that the route for infection was the same -unprotected anal sex 7 . Even today, the same mistake still occurs in compulsory notification for HIV in Brazil, making it impossible to analyze the two groups separately 8 . This misinterpretation of gender identity and sexual orientation in the scientific literature could be partially related to the outdated theory of sexual inversion, or psychological androgyny, which considers that gay men would be sexual inverts, with a "female soul", and that transgender women would simply be a hyper-feminized version of male homosexuality 9 . Today, it is clear that sexual orientation and gender identity are distinct domains 10 .
Named as a hidden population, the transgender populations are difficult to access, therefore studies that include them tend to adopt characteristic sampling methodologies, such as respondentdriven sampling (RDS), facility-based sampling (FBS), and time-location sampling (space -TLS) 7 . RDS is a network-based sampling method developed to enable access to hidden populations 11 . It is a type of chain referral (or snowballing) method in which seeds, also called initial respondents, recruit peers 11 . In each new RDS wave, recruits become recruiters and invite a pre-established number of peers 11 , whereas both FBS and TLS refer to the place where the invitation occurs. In FBS, participants are invited from a variety of facilities that they usually attend, such as health services specialized in gender-affirming processes 12 . In TLS, the recruitment happens at locations where and when the target population may be found 12 . Studies also tend to use samples collected via the internet in groups where this audience is present or using the internet in conjunction with previous methodologies, such as RDS or FBS.
Based on the different approaches used to identify transgender populations, three methods are frequently used. The single-item measure consists of directly asking about the participants' gender identity. Traditionally, demographic surveys use this method and limit the participants' options to "male", "female", and, rarely, "transgender" 13 . Although very simple, this single-item measure has some significant limitations 14 . The single-item measure can mislead information, because a transgender person may only identify as a male or a female. Besides, it does not consider the wide variety of non-binary genders. It also ignores culturally specific gender identities, that is, several segments of the transgender population as well as particularities that could improve the decision-making process in public health will remain invisible. A report of the William's Institute (United States) emphasized that the single-item measure combines gender assigned at birth and transgender identities in a way that limits identification to half the number of participants of other measures 15 . Furthermore, since it is not uncommon for transgender participants to self-identify as simply a woman or a man, misguiding the quality of data collected on cisgender populations as well 16,17,18 .
The two-step measure, developed by a U.S. transgender community-based organization 18 , can overcome some of these challenges. The first question refers to the participants' gender assigned or registered at birth (male or female), whereas the second question inquires about their current gender identity (male, female, transgender men, transgender women, and other gender variations depending on the cultural context).
The main advantage of the two-question method is recognizing a wider spectrum of transgender and gender-diverse identities, including transgender people who prefer not to identify as such (for example, a woman whose gender assigned at birth was male), as well as other culturally-specific identi-Cad. Saúde Pública 2022; 38(1):e00180420 ties 19 . Also, the two-step measure allows participants who are sometimes unaware of gender-specific vocabulary, to disclose their gender identity using terminology that researchers will understand 14,20 . For example, a North-American study conducted interviews with 25 cisgender and 25 transgender people aiming to identify measures that effectively recognize transgender status 20 . Both cisgender and transgender participants had difficulties understanding measures based exclusively on gender, whereas the "gender assigned at birth" was considered a clear and straightforward item. Indeed, four transgender participants chose male-or female-only options to describe their gender identities. Furthermore, a mixed-method study interviewed 36 cis-and transgender participants about the two-step measure 14 , and none reported difficulties in understanding the questions. When applied to a more substantial sample (n > 36,000) encompassing men from Spanish and Portuguese-speaking countries, the two-step measure was able to identify the expected prevalence of transgender people 19 . Therefore, the two-step measure effectively recognizes transgender and gender diverse populations, and it is currently considered to be the gold-standard for health surveillance 21 .
Finally, the third approach, widely used in FBS and TLS strategies, is the assumption that, for occupying transgender venues, one necessarily identifies themselves as transgender: a method we named "trans-assumption". We also considered "trans-assumption" in which the participants who did not receive instructions on defining gender identity, sexual orientation, or transgender women with peer-identification, frequently applied in RDS.
We aim to present the two-step measure as the best strategy for assessing gender diversity in Brazilian studies. Thus, we performed two separate analyses. Firstly, bearing in mind that most Brazilian publications including transgender people focus on HIV-related care 22 , we conducted a systematic review concerning HIV-related care among Brazilian transgender and gender diverse populations to assess the strategy sued to identify participants' gender identity. Then, we re-analyzed data from a recent survey that included Brazilian transgender populations, comparing the characteristics and health outcomes from the samples identified by single-item and by the two-step measure providing data on the accuracy of this method.

Systematic review
To assess strategies used to identify transgender populations, Brazilian studies were re-analyzed concerning HIV-related care among transgender and gender diverse populations, found by a systematic review recently published by our group 7 . Peer-reviewed quantitative studies, published up to April 4th, 2018, concerning transgender women, transgender men, and gender diverse people and HIVrelated care were included 7 . HIV-related care was considered as all interventions aiming to prevent, to treat, or to alleviate the impact of HIV on these populations. The review was registered online on the International Prospective Register of Systematic Reviews (PROSPERO 2017 CRD42017071213). For details regarding the methodology, see Fontanari et al. 7 . As the two-question method was proposed around 2011, all included articles were published after 2015.
Information about recruitment and methodology used to assess gender identity was collected from the studies. Sample strategies were RDS, FBS, TLS, and internet-based survey (Internet). The measures for assessing gender identity were classified as unclear, single-item, two-step, or transassumption, based on what was explicitly mentioned in the article.

Quantitative analysis
To assess the impact of using different methods to recognize transgender people, data were re-analyzed from the Trans Health Research Project -the database utilized by Costa et al. 16 . The Trans Health Research Project, a hospital-and web-based cross-sectional survey, was applied in two Brazilian states, Rio Grande do Sul and São Paulo. Transgender people seeking medical gender affirmation procedures from the two gender identity programs were invited by the researchers by convenience sampling to voluntarily answer the survey. Besides, the survey was available on the Internet during two time peri-Cad. Saúde Pública 2022; 38(1):e00180420 ods: July-October 2014, and January-March 2015. The methods are described in detail in Costa et al. 16,23 and other studies derived from this research.
According to Costa et al. 16 , gender identity was assessed using the two-step measure. Gender identity was questioned both by a closed-ended and a write-in. The closed-ended question was "which of the alternatives below best describes your current gender identity?", and the participant could choose between the options "female", "male", "transgender woman", "transgender man", "travesti", and "other". The alternative "other" offered participants an open-ended write-in, where several participants identified as gender-diverse persons. Participants were asked to disclose their gender assigned at birth by the question "What gender were you assigned at birth, on your original birth certificate?" using the alternatives "male" and "female". Based on the incongruence between their self-reported gender identity and their gender assigned at birth, participants were re-categorized as transgender women, transgender men, or gender diverse persons. That is, participants were categorized as "transgender women" when they selected both "male" to describe their sex assigned at birth and "female" or "transgender women" to characterize their gender identity; as "transgender men" when they checked "female" as their gender assigned at birth and "male" or "transgender men" as their gender identity; and as "gender diverse people" when they chose "female" or "male" as their gender assigned at birth, "others" as gender identity, and, after, described their gender identity (neutral, nonbinary, agender, pangender, genderqueer, two-spirit, third gender) in the write-in space.
To assess the impact of applying different methods to evaluate gender identity, sociodemographic characteristics were described using a sample of transgender people identified both by the single-item measure and by the two-step measure, as well as by a method we named "trans-assumption" -meaning that, if recruited on LGBT venues, participants would necessarily be transgender people.
Data management and statistical analyses were performed using SPSS software version 18.0 (https://www.ibm.com/). Frequencies were estimated according to sociodemographic characteristics. Both test sensitivity and specificity were calculated for the single-item measure considering the two-step measure as standard.

Systematic review
From 6,585 references, 62 articles contained both HIV-related care and transgender people. Among them, Brazilian research teams published seven articles. As described in Box 1, most studies assumed that recruiting a sample from social organizations aiming to provide help to transgender populations, such as lesbian, gay, bisexual, transgender non-governmental organizations (LGBT NGO), would be enough to assure that participants were transgender people -an approach that, as peer-identification, we named as "trans-assumption". For example, Hoagland et al. 24,25 invited participants seeking HIV testing at an LGBT NGO and a mobile testing unit located in an LGBT-friendly venue. We applied one question concerning gender identity, and the participants were dichotomized into "male" and "transgender women". Castro et al. 26 used a similar strategy, inviting participants in front of an LGBT venue and at the site of Arco-Íris Group, a well-known LGBT NGO; however, we did not describe the method used to identity transgender populations.
Furthermore, studies that recruited samples through RDS counted on peer identification to guarantee that the participants were transgender women 27,28 . Therefore, Jalil et al. 27 (p. 2) indicated that "to participate, individuals had to self-identify as transgender women", and Pinheiro Júnior et al. 28 (p. 2) suggested that "each seed and study participant received three coupons to distribute to transgender women they knew by name and had seen in the past 30 days". Grinsztejn et al. 29 did not provide any data on how they identified their sample's gender identity. Notably, the participants did not receive directions on the concept of gender identity, sexual orientation, or transgender women. Cad Table 2 shows 31 (8.5%) transgender women self-identified as "transgender men"; this suggests that, for them, the concept of transgender women differs from the one used in our manuscript. Furthermore, 162 (44.4%) transgender women self-identified as women -meaning that the singleitem measure would not be able to recognize them. Out of the 196 transgender men, 71 (36.2%) selfidentified as men-only and were not embraced by the single-item measure.

Table 1
Sociodemographic and HIV-related data concerning all participants, disregarding their gender identity, as well as participants' gender identity identified by the single-item and the two-step measures. Notably, sensibility, specificity, and accuracy refer to correctly identifying transgender and gender-diverse people. Recognizing transgender and gender-diverse people, however, does not necessarily means identifying correctly transgender men and women. For example, 31 (8.5%) transgender women were misclassified as "transgender men" by the single-item measure.

Discussion
Re-analyzing studies from the last 10 years, we concluded that just one applied the two-step measure. Furthermore, the data reanalysis showed that the different methods used to assess gender identity selected distinct samples. The two-step measure recognized 567 cisgender and 773 transgender and gender diverse participants among the 1,340 participants who started the questionnaire, whereas the single-item measure was able to recognize only 540 transgender and gender-diverse people. Moreover, 31 transgender women self-identified as "transgender men" on the single-item measure. Therefore, although scarcely used in Brazil, the two-step measure is a more accurate strategy to recognize gender identity.
In the reviewed studies, the most used method to assess gender identity was peer-identification, or "trans-assumption", in the context of RDS and TLS, that is, researchers went to places where transgender people usually gather (mainly NGOs) and asked them to self-identify (single-item measure) and/or to refer their peers. Observing the re-analysis of our survey, we found a problem in doing this as many LGB people can identify themselves as being in the transgender spectrum, as well as transgender persons could choose not to reveal their transgender identity or identify as gay or lesbian. As gender identity and sexual orientation are independent, transgender and gender-diverse people may also identify as LGB by attending LGB venues, and joining LGB surveys. The overlap between sexual orientation and gender identity is particularly true in Brazil since gender identity, expression, and sexual orientation are often viewed interchangeably by segments of the LGBT community, as well as the general society. For example, stereotypes of male homosexuality are frequently associated with feminine gender expression, as well as some travestis who could also self-identify as gay men 5,30 . Table 2 Misclassifications of gender identity based on the single-item versus the two-step measure. The aforementioned issue is particularly concerning in the context of TLS sampling. For example, the NGOs used in one of the TLS samplings were, in fact, LGBT NGOs -not just "T" 26 . Therefore, online surveys directed to sexual and gender minorities may also include gay, lesbian, and bisexual people 31 . Applying the RDS method does not exactly overcome this issue, since the concept, as well as strategies to recognize, transgender may vary according to the recruiter and the seeds. Seed quality and subsequent assessment of gender identity are crucial in the RDS method. Without being sure of the participants' gender identity -which is only feasible with the two-step measure -even this more robust methodology may reinforce biases in the sample composition that would need to be considered. From our review, we concluded that the sampling technique is considered enough to identify transgender people; however, data analysis showed the opposite. Because the recruitment occurred on websites geared towards LGBT communities, LGB populations completed our survey. The twostep measure made it possible to identify these participants.

Single-item measure
One study mixed sexual orientation and gender identity by asking only one question as a criterion for the inclusion of participants 29 . The overlap between sexual orientation and gender identity is criticized since it is assumed that all transgender women have sex with men, making different segments of the community invisible, for example, lesbian and bisexual transgender women who are at lower risk for infection, and transgender men 32,33 . Besides, questions related to sexuality were not included in the studies that we reviewed, and it is not possible to know if the samples contain bisexual and homosexual transgender women or men, groups with different degrees of vulnerability for HIV infection. As exposed before, overcoming theories that consider sexual orientation and gender identity at the scientific level, and reaching autonomy of the transgender advocacy from the political movement around sexuality, also reinforced the position that it is fundamental to treat sexuality and gender as distinct domains 34 .
We conclude this paper with a proposal to be used in future studies in Brazil (Figure 1). The first question, as usual, asks about the gender on the civil registration ("a -female" or "b -male", according to the current Brazilian standards). Security is one of the main priorities among the best practices for the collection of data about gender minorities in South America 21 . Considering the high systemic violence against transgender and gender-diverse populations, it is important to have an option to allow the passability as cisgender, not identifying the transgender status to the interviewer in moments where this identification may be risky to the participant's integrity. Therefore, we added "c" as an "I prefer not to answer" response option in the first question. For the second question, we kept "awoman" and "b -man" as the first response options. As stated before, this allows cisgender people to easily choose, for example, "male" in the first question and "man" in the second (or vice versa). At the same time, those two responses allow a transgender person who does not emphasize they "transgender" status to check, for example, "male" in the first option and "woman" in the second. This also allows those who answered "c", "prefer not to answer" in the first question to choose a gender identity that does not allow the identification as transgender.
In the third (c) and fourth (d) alternatives of the second question, we combine different ways in which the Brazilian population identifies itself around the Anglo-Saxon umbrella-term transgender (trans, transgender, transexual). Although we know that there is no consensus regarding these gender identities and that they can encompass different groups within the community, we have chosen to unite them as they denote the group of transgender people who prefer to make their identity visible as such 4 . Besides, following an international guideline 10 , we have suggested making Brazilian-specific gender identities independent. For that, we included travestis as a fifth answer (e) in the second question. In our survey, many people who selected the "other" option later self-identified outside the gender binary. Research in other countries has shown an increase in this category in population surveys, thus we chose to make it an autonomous category 35 . Thus, "f" is an alternative for people who identify outside the gender binaries, such as gender fluid, agender or queer people. Finally, understanding the mutability of the forms of identification and gender itself, anyone who does not identify with the possibilities listed could opt for the last alternative, "other", and declare their specific gender identity "g".
All quantitative data collection implies choices. The method suggested here should not impose academic-oriented identities, but rather identify a broad spectrum of the community flexibly and comprehensively. Therefore, all questions could be edited by the researchers to add different alternatives to cover as broadly as possible how a given community identifies itself at the moment of the   research implementation -for example, a third gender in the civil registration and/or culturallyspecific gender identities. Besides, in the spirit of community-based research, we recommend that members of the local transgender community -where the research will be applied -review these questions 36,37 . As aforementioned, this is a minimal proposal for data collection. However, for research focused on transgender and gender-diverse populations, it may be necessary to ask other complementary questions, such as medical and social procedures for gender affirmation, current or more frequent self-perceived public gender expressions, and the gender identity experienced most frequently (for gender-fluid people). In the context of transgender HIV-related research, we also suggested a question about sexual orientation 7 . This study has one main limitation. We relied only on the information presented in the articles included in our systematic review. By contacting the authors, we could have detailed information on how gender identity was assessed. Thus, the category "trans-assumption" is itself an assumption that the included articles did not use a more sophisticated measure to assess gender identity. However, by unclearly describing the measure used to identify gender identity denotes how the procedure was secondary. Furthermore, it should be noted that, more recently, Brazilian studies adopted the twostep measure, such as Bastos et al. 38 .
In conclusion, we ratify that the two-step measure is more accurate to identify transgender people and that almost all the articles recently published by Brazilian researchers on transgender HIV-related health are likely to present sampling bias. Unlike other countries, such as Canada and Uruguay, Brazil has not yet implemented a large-scale census-type strategy for documenting the living conditions of transgender and gender-diverse populations 39 . Debates on the 2020 census indicate that there is no Cad. Saúde Pública 2022; 38(1):e00180420 intention to assess gender identity in the census research protocol 40 . The two-step question method should be adopted not just in the census, but in studies in the field of collective health dealing with this population, digital information system besides the surveillance notification forms from the Brazilian Unified National Health System (SUS).