RESEARCH DOSSIER
Sex, Gender, and Sexuality: Multidimensional Measurement
Abstract. Sex, gender, and sexuality are multidimensional domains whose scientific measurement requires explicit distinctions among biological characteristics, birth-assigned classifications, gender identity and expression, social position, sexual orientation identity, attraction, behavior, desire, and well-being. This fundamental research monograph investigates how these constructs are defined, recorded, validated, compared, and responsibly interpreted in psychological, population, and health research. Drawing on historical sexology, construct-validity theory, two-step gender measurement studies, survey experiments, cognitive interviews, epidemiologic methods, systematic instrument reviews, and consensus recommendations, it evaluates evidence for the reliability, validity, response-process sensitivity, and limitations of available approaches. It distinguishes short-term response agreement from longitudinal continuity, broadly concordant classifications from subgroup equivalence, and observed associations from causal mechanisms. Separate analyses examine cultural translation, adolescent development, probability sampling, sparse categories, privacy, intersectional heterogeneity, and clinical data systems. The scientific synthesis supports purpose-specific, transparent, and empirically tested multidimensional measurement, while identifying unresolved needs for cross-cultural validation, representative longitudinal designs, independent replication, and responsible data governance. This is a structured critical narrative scientific synthesis, not an original experiment, a registered systematic review, or a new quantitative meta-analysis.
Keywords: sex measurement; gender measurement; sexuality; sexual orientation; gender identity; gender expression; sex assigned at birth; two-step method; transgender identification; attraction; sexual behavior; sexual desire; survey methods; cognitive interviewing; construct validity; psychometrics; reliability; measurement invariance; intersectionality; population sampling; research ethics; privacy
The Research Object and the Logic of Multidimensional Measurement
Sex, gender, and sexuality are related domains of human variation whose scientific indicators answer different questions. A multidimensional measurement system records the attribute relevant to a stated inference: a specific biological characteristic, a birth-record designation, an experienced or socially recognized gender, a pattern of attraction, a sexual identity, a behavior during a specified interval, or a dimension of desire. The same person's responses may vary across dimensions without measurement error. The starting question for this monograph is how research can preserve those distinctions while producing data that remain interpretable, comparable, ethical, and useful for psychological explanation. The National Academies' consensus report recommends collecting gender by default in most human research and recording relevant sex traits when biological mechanisms require them National Academies (2022) Restar et al. (2024).
Scientific measurement is a chain of decisions. A theoretical concept determines the target attribute; a questionnaire, examination, or record supplies an observation; a coding rule turns that observation into an analyzable variable; and an inference connects a modeled association to a population or mechanism. Any link can fail even when the dataset contains no missing values. If investigators use an administrative sex category as a substitute for hormone exposure, a correlation may have precise confidence intervals while measuring the wrong causal variable. If an attraction item is interpreted as identity, a survey can mischaracterize the composition of its sample. Construct validation therefore concerns the meaning and intended use of scores, not simply item consistency Cronbach & Meehl (1955) Campbell & Fiske (1959).
This monograph critically synthesizes consensus recommendations, foundational methodological papers, survey experiments, cognitive interviews, psychometric evaluations, and recent systematic evidence reviews. The review is structured around construct definition, response process, sampling, statistical interpretation, and responsible application; its source selection is purposeful rather than registered or exhaustive. Empirical findings are attributed to the designs and populations from which they arose. Proposed practices are distinguished from results demonstrated in independent studies. The broader analytic objective is to establish how measurements can represent overlapping human characteristics without assuming that every observed relationship signals a single underlying dimension Rich et al. (2025) Ahmad et al. (2026).
Historical Foundations: Classification, Sexology, and Survey Research
Early scientific classifications of sexuality frequently organized people around types and diagnoses, whereas later behavioral research placed greater emphasis on variation in reported practices. Kinsey, Pomeroy, and Martin's 1948 Sexual Behavior in the Human Male is a landmark because it treated sexual experiences as empirical survey material and popularized descriptions of variation along a continuum. Its historical significance does not settle the quality of its sampling frame or the generalizability of particular prevalence estimates. The contrast between interview-based descriptive ambition and probability-based population inference remains fundamental to the field Kinsey et al. (1948) Sell (1997).
By the late twentieth century, investigators were increasingly attentive to the mismatch among sexual acts, attractions, and public identities. Klein, Sepekoff, and Wolf's 1985 multidimensional approach incorporated several dimensions assessed across past, present, and ideal circumstances. The approach showed how a person's orientation profile could be described at multiple temporal and experiential levels, although its original convenience-sample evidence does not establish population norms. Sell's 1997 review documented how competing definitions and operational rules produced divergent samples described by the same categorical labels. Their enduring contribution is a methodological warning: samples defined by identity, attraction, and behavior can differ systematically Klein et al. (1985) Sell (1997).
Historical changes in gender measurement also reflect changes in sampling, administration, and the social conditions of disclosure. Questionnaires that once recorded only female or male as a presumed universal descriptor came to be evaluated for whether they correctly represented transgender experience and gender identities beyond binary categories. The two-question method tested by Tate, Ledbetter, and Youssef made current gender identity and birth-assigned category separately observable. Contemporary consensus guidance and community-informed cognitive testing have expanded this logic into explicit questions about construct relevance, acceptable wording, and the costs of collecting sensitive information Tate et al. (2013) National Academies (2022) Suen et al. (2020).
From Test Scores to Validity Arguments
Cronbach and Meehl's construct-validity program directs researchers to examine whether observations behave as a theoretical network predicts. A gender-role scale may be internally consistent while measuring conformity to locally salient expectations rather than gender identity. A sexual desire scale may reliably measure an individual tendency while failing to represent the particular aspect of attraction required by a study. Reliable repetition and meaningful interpretation are related achievements, but the first never guarantees the second. Interpretation must specify an attribute, a population, an observation process, and the rival explanations that the design can rule out Cronbach & Meehl (1955) Spector et al. (1996).
Campbell and Fiske's multitrait–multimethod analysis provides a practical strategy for evaluating overlap. Agreement across independent indicators of the same construct supports convergence; separation from theoretically distinct constructs supports discrimination. For multidimensional sex–gender–sexuality measures, the logic discourages claims that any positive correlation demonstrates a shared identity. Measures can correlate because respondents experience linked social contexts, because an item contains ambiguous wording, or because the same response style shapes each rating. Independent methods, predictive hypotheses, and design-based comparisons are required to assess whether a proposed distinction survives alternative explanations Campbell & Fiske (1959) Mokkink et al. (2024).
Validity is use-specific. A short gender-identity item may be appropriate for demographic inclusion and insufficient for a hypothesis about discrimination mechanisms. A record of reproductive anatomy may aid a particular screening decision and be inappropriate for predicting social-role experiences. Evidence supporting a survey item in a US adult probability sample does not automatically license its use with younger children, a multilingual refugee cohort, or a clinical registry. Modern measurement evaluation emphasizes content relevance, respondent comprehension, reliability, structural validity, cross-group comparability, and documented limits of generalization National Academies (2022) Ahmad et al. (2026).
Scientific Questions and the Evidence Architecture
The first scientific question is classificatory: which distinctions must be represented to answer a psychological question without merging identity, social position, biology, and behavior? The second is psychometric: how do respondents interpret the stems and options, and can the resulting variables be used consistently across persons and occasions? The third is inferential: under what sampling and modeling conditions do measured associations generalize to a population? These questions are sequenced because the statistical sophistication of a model cannot repair a prior category error in the operational definition National Academies (2022) Rich et al. (2025).
Evidence is organized into complementary layers. Historical works reveal how classification conventions arose. Cognitive interviews disclose response processes and unanticipated meanings. Experiments that vary item order or placement estimate effects of questionnaire design. Repeated-measure studies examine short-term consistency. Probability surveys make population estimation more defensible, while community-based studies reveal categories and experiences that may be rare or poorly captured in general samples. Systematic reviews summarize the tested instrument landscape. Each layer answers a different question and none should be treated as a replacement for the others Suen et al. (2020) Holtzman & Saperstein (2026) Heise et al. (2025/2026) Ahmad et al. (2026).
The evidentiary standard applied throughout is proportional to the claim. A single cross-sectional study can demonstrate that a measurement approach was feasible in its sample; it cannot establish cultural universality. A qualitative interview study can identify threats to interpretation without quantifying their prevalence in an entire country. A systematic review can assess the corpus it searched, yet its findings depend on eligibility rules, publication coverage, and reported psychometric detail. Explicitly documenting these limits strengthens cumulative measurement science and prepares the ground for transparent revisions when new groups or contexts are studied Rich et al. (2025) Mokkink et al. (2024).
Biological Variables, Assigned Categories, and Clinical Relevance
Biological inquiry requires precise measurement of the mechanism under study. Chromosomal pattern, current endocrine profile, organ presence, reproductive history, and medications can be important for different research questions, yet they are not interchangeable proxies. Sex assigned at birth is a historical administrative or clinical classification, not a direct observation of every relevant trait throughout life. Investigators who require a specific physiological variable should identify the variable and collect it with appropriate consent, validated methods, and attention to changes over time. The consensus recommendations expressly link collection of sex traits to a demonstrated scientific purpose National Academies (2022) Restar et al. (2024).
Intersex traits and variations in sex development further illustrate why a binary sex item is an incomplete biological measurement. Identifying an intersex characteristic is a different task from measuring transgender experience, current gender identity, or any given physiological parameter. Adding 'intersex' or 'transgender' as a mutually exclusive third choice alongside 'male' and 'female' collapses categories defined by different criteria and can produce avoidable misclassification. In studies addressing particular medical characteristics, trait-specific documentation may improve relevance and protect against erroneous inference from a broad identity label National Academies (2022) Restar et al. (2024).
Clinical datasets introduce additional distinctions between an answer that helps care and an answer collected solely for research. A patient's current gender and name may support respectful communication, while anatomical or treatment details may be needed for a particular clinical question. Electronic record systems that overwrite all dimensions with one demographic field create difficulties for continuity, privacy, and later analysis. The argument for differentiated records follows from data quality as well as patient welfare, but actual data collection should remain limited by purpose, secure access, and governance arrangements Cahill & Makadon (2014) National Academies (2022).
Gender Identity as a Self-Reported Psychological Construct
Gender identity is a first-person characterization of one's gender, usually recorded through self-description. The act of reporting identity also occurs within a social setting: language, offered labels, perceived surveillance, interviewer characteristics, and anticipated consequences may shape disclosure. A response should therefore be interpreted as the individual's answer to a specified question at a specified time. Researchers can improve interpretability by specifying whether they ask about current identity, an identity history, social recognition, or the relation between gender and birth assignment. These constructs have different psychological implications even when a dataset uses a single word, gender, for all of them National Academies (2022) Restar et al. (2024).
A questionnaire often uses a single-choice item for convenience, yet respondents may understand several categories as relevant. Multiselect formats permit overlapping identifications but require careful coding and adequate sample sizes for analysis. A free-text response offers flexibility while creating translation, confidentiality, and harmonization challenges. Providing an opt-out avoids coercing disclosure, and preserving uncertainty or questioning as a valid response helps distinguish evolving identity from simple item failure. None of these choices is universally superior; they must be assessed against the study's inferential goal and the actual response process Suen et al. (2020) Rich et al. (2025).
Gender identity is not a measure of gender expression, perceived femininity or masculinity, access to gender-affirming care, or exposure to discrimination. When those variables matter, they require their own questions and validation. Gender expression can vary with setting and be evaluated differently by observers; perceived gender has implications for social treatment independent of the individual's self-description. Identity studies gain explanatory accuracy when they model relevant contexts rather than treating identity as a sufficient descriptor of every gendered experience Gower et al. (2018) Meyer (2003).
The Two-Step Approach and Its Original Validation
The two-step approach asks separately about a birth-assigned sex category and current gender identity. Its value is analytic as well as descriptive: the resulting joint responses can identify some forms of transgender experience without requiring respondents to regard 'transgender' as an exclusive gender category. In three studies totaling 990 participants, Tate, Ledbetter, and Youssef compared versions of a single-question method with separate birth-category and current-identity questions. Their reported findings indicated better inclusion and reduced missingness for the two-question approach in the populations studied Tate et al. (2013).
The approach does not make the two questions interchangeable with biological trait measurement or self-described gender modality. The classification constructed by comparing answers depends on response options, respondents' interpretations of their original record, and the particular analytic rule applied. Individuals who use multiple identity labels or question their fit within a category can be misclassified by simplified algorithms. Survey designers should archive both original answers when appropriate and treat derived classifications as variables whose construction can be audited, rather than presenting the derived code as an unmediated personal fact National Academies (2022) Felt et al. (2025).
Population application requires careful attention to age, language, sampling, and the presence of categories that respondents actually use. A two-step method can be sensitive to rare populations, but its utility is conditional on whether nonresponse and classification errors are understood. Researchers should report the exact item wording, option order, whether multiple selections were permitted, how missing values were treated, and which rules converted answers into analytic groups. That transparency permits later reanalysis when labels evolve or classification conventions change Tate et al. (2013) Holtzman & Saperstein (2026).
Comparing Birth-Assignment and Gender-Modality Approaches
An alternative two-step strategy pairs current gender identity with respondents' own descriptions of gender modality, such as whether they identify as transgender, cisgender, or another way. The difference is substantive: gender modality conveys a relationship that a respondent interprets, while birth-assignment data record a historical category. In a 2025 comparison involving 952 US sexual and gender minority adults, Felt and colleagues reported 95% convergence across two classification approaches and identified meaningful cases of divergence. This result supports testing the alternatives against study purpose rather than assuming complete equivalence Felt et al. (2025).
Tordoff and colleagues compared current gender paired with sex assigned at birth against current gender paired with transgender self-identification among 2,275 LGBTQ+ survey participants in Washington state. The methods agreed on the broad classification of transgender and gender-diverse participants at 99.7%, with an unweighted kappa of 0.99. Yet stratifications identified different associations: birth-assignment comparisons were informative for certain sexual-health patterns, while self-identification revealed distinctions in care access and discrimination. High overall agreement thus coexisted with substantive differences for particular research questions Tordoff et al. (2025).
These studies should be read within their recruitment frames. Samples recruited from LGBTQ+ communities can illuminate detailed within-group differences but do not independently provide nationally representative classification error rates. Very high agreement on a broad binary outcome is compatible with meaningful disagreement among smaller subgroups. Inference requires examining confusion matrices, missing categories, subgroup precision, and the costs of erroneous classification. Researchers selecting between the approaches should explain which substantive construct the additional question enables them to investigate Tordoff et al. (2025) Felt et al. (2025).
Questionnaire Architecture as an Experimental Variable
Question wording and layout can change the distribution of self-reports because respondents interpret a question within the sequence presented to them. Holtzman and Saperstein conducted survey experiments in 2026 with an SGM sample of 2,402 adults and a general-population sample of 1,491, comparing terminology for current identity, page separation, and response-option order. Their reported results favored gender rather than sex language for the current-identity item and a page break between two-step items; option order did not produce measurable data-quality changes in their samples Holtzman & Saperstein (2026).
The experiment supplies a design-specific causal test of formatting alternatives, not a rule that every platform or language must yield identical effects. A computer-assisted interview, smartphone form, and interviewer-administered census can create different attention demands and social cues. Item placement may alter memory of prior answers, perceived consistency pressure, or the scope of a question. For sensitive demographic dimensions, these mechanisms justify randomized pretesting before large-scale adoption when feasible Holtzman & Saperstein (2026) Bureau of Labor Statistics (2017).
The practical implication is that standardized wording alone cannot guarantee standardized measurement. Researchers should publish form screenshots or machine-readable questionnaires when privacy and licensing permit, along with skip logic, instructions, page breaks, translation, and coding. If a study changes its presentation over waves, that modification should be treated as a possible measurement discontinuity and evaluated empirically. Comparability is an evidentiary claim requiring study-specific documentation Holtzman & Saperstein (2026) Rich et al. (2025).
Reliability, Temporal Stability, and Developmental Interpretation
Reliability can mean repeatability of an answer, stability of a latent attribute, or agreement among independent raters; those are not equivalent properties. Heise and colleagues examined a revised two-step assessment among 610 gender-diverse youth and 606 parents, using repeated administration over 30–60 minutes. Agreement for both gender and sex questions was at least 95%, with variation across response categories and generally lower reliability for participants selecting less determinate options. Such results support short-interval consistency in studied groups while leaving longer developmental trajectories open Heise et al. (2025/2026).
An identity response that changes across months or years can reflect genuine development, changes in available language, social safety, or differences in questionnaire design. Treating every change as error erases the phenomenon longitudinal research seeks to explain. A study must distinguish measurement reliability at a short interval from stability of identity across life stages. This is especially important in adolescence, when experiences, opportunities for disclosure, and understanding of category labels can change considerably Heise et al. (2025/2026) National Academies (2022).
Repeated surveys should preserve a record of how respondents interpreted each wave's questions. Longitudinal invariance analysis can examine whether items support comparable meanings across time, while qualitative follow-up clarifies why transitions occur. Transition matrices, rather than retrospective assignment to a single lifelong category, can document change. Such analyses require safeguards against small-cell disclosure and careful differentiation between a measured shift in response and a claim about any underlying biological or psychological cause Heise et al. (2025/2026) Suen et al. (2020).
Gender Expression, Social Position, and Contextual Exposure
Gender expression refers to presentation and behavior through which gender is communicated or perceived, including appearance, voice, and other culturally interpreted cues. It may vary across settings and observers, and its measurement must name the observational standpoint. Self-rated expression, anticipated public perception, and an observer-coded presentation are different variables. Conflating them with current identity can hide the mechanisms through which stereotypes or unequal treatment operate Gower et al. (2018) Restar et al. (2024).
Gendered social position includes recognition, expectations, institutional classifications, access to resources, and exposure to gender-based constraints. These are often relational or structural attributes. A single personal identity item cannot measure the distribution of norms in a workplace, school, or health system. Studies examining these mechanisms need multilevel designs, administrative records whose definitions are known, and explicit contextual indicators. Their causal claims depend on temporal order and alternative explanations, not merely on demographic group comparisons Meyer (2003) National Academies (2022).
The multidimensional approach allows researchers to test how identity, expression, and social position interact while preserving each as an identifiable variable. For example, the association between perceived gender presentation and bullying may be different from the association between self-identified gender and bullying. Evidence of such contextual differences should motivate refined hypotheses and survey items, not automatic attribution of a social outcome to an intrinsic property of a group. Researchers should attend to disclosure safety and the risk that externally observed categories reproduce investigator assumptions Gower et al. (2018) Suen et al. (2020).
Sexual Orientation as Identity, Attraction, and Behavior
Sexual orientation is commonly described through dimensions of attraction, identity, and behavior, but a survey must identify which of these constitutes the outcome of interest. A respondent may identify as heterosexual while reporting attraction to more than one gender, or identify as bisexual without recent partners of multiple genders. Survey interpretation improves when investigators report separate estimates for each dimension and describe the reference periods used for behavioral measures. Sell's historical review showed that definitions determine who enters the analytic population and therefore influence findings about health, experience, and social behavior Sell (1997) National Academies (2022).
Identity questions typically ask a person to select a socially intelligible category. Attraction items can ask whether and to what extent sexual or romantic interest is directed toward specified genders, while behavior items specify actions, partners, and time windows. The more detailed the item, the greater the risk that a respondent will read unanticipated assumptions into it. For example, a question about the gender of partners cannot capture an individual's orientation if no partners were reported during the reference period. Instruments should therefore test item comprehension and offer response paths appropriate for people with no sexual experiences or no current sexual attraction Sell (1997) Suen et al. (2020).
These distinctions also affect epidemiology. Estimates derived from an identity item cannot be labeled prevalence of same-gender attraction, and reports of same-gender sexual behavior cannot automatically be classified as gay or lesbian identities. Choice of numerator, denominator, recruitment frame, and age eligibility matters when comparing population estimates. Precision is achieved by matching the measured variable to its descriptive claim, not by treating every sexuality measure as a proxy for a single latent orientation type National Academies (2022) Rich et al. (2025).
Continuous Profiles and Multidimensional Historical Instruments
Kinsey-era descriptive work made variation in same- and other-sex experiences visible within a scheme that represented more than two polar categories. A continuum can be useful for studying relative direction of attractions or experiences, but it can also collapse distinct combinations into an identical summary value. Someone who experiences little attraction to anyone and someone who experiences strong attraction to multiple genders could receive superficially similar positions if the instrument neglects intensity and absence. A scientifically adequate scale must define what its endpoints and intermediate values mean Kinsey et al. (1948) Sell (1997).
Klein, Sepekoff, and Wolf's multivariable approach gave separate attention to attraction, behavior, fantasy, emotional preference, social preference, lifestyle, and self-identification across temporal perspectives. Its historical significance lies in showing that sexuality can be represented as a changing profile of attributes rather than a universal rank order. Each added dimension also creates demands for construct definition and empirical validation; adding variables to an instrument does not by itself establish that their distinctions predict outcomes. Modern researchers should compare conceptual coverage, test–retest properties, response burden, and discriminant validity before adopting a historical measure Klein et al. (1985) Campbell & Fiske (1959).
A profile-based representation can be especially informative for longitudinal designs, because change may occur in one domain while other domains remain relatively stable. For example, identity labels can change with available language and social context, while self-described attraction remains similar, or desire can vary with relationship circumstances. Such patterns need not be interpreted as measurement failure. The research task is to specify the time scale of interest, assess instrument equivalence across waves, and avoid inferring a universal development sequence from descriptive transitions Klein et al. (1985) Rich et al. (2025).
Sexual Desire, Arousal, and Response as Separate Constructs
Sexual desire refers to subjective motivation, interest, or wanting related to sexual activity; arousal can refer to a subjective experience, physiological response, or a composite of multiple indicators. Instruments that use the same word for these different levels of observation can obscure the distinction between wanting, feeling, and bodily reaction. Spector, Carey, and Steinberg's Sexual Desire Inventory operationalized desire through differentiated questionnaire content and reported evidence for its factor structure and reliability. Its contribution concerns a defined self-report construct, and its coefficients cannot be transferred automatically to physiological arousal or to sexual orientation classification Spector et al. (1996) Mokkink et al. (2024).
Sexual desire is affected by interpersonal opportunities, stress, health, attitudes, relational context, and changing life circumstances. A measure asking about desire toward a partner may be uninterpretable for respondents who are unpartnered or whose intimate relationships take other forms. A questionnaire should specify whether the target is general interest, solitary interest, partnered interest, event-specific interest, or change from an individual's usual state. Researchers can then test how the chosen measure relates to well-being and other outcomes without assuming that one score is the complete structure of human sexuality Spector et al. (1996) World Health Organization (2006).
Sexual well-being is a broader outcome domain involving subjective appraisal, safety, consent, relationships, bodily experience, and access to information or care. The World Health Organization's working definition of sexual health emphasizes physical, emotional, mental, and social well-being related to sexuality. Its use as a normative framework does not make every component a psychometric dimension or establish a universal index. Researchers must decide which outcomes are relevant and measure them directly, especially when subjective satisfaction and behavioral frequency can diverge World Health Organization (2006) National Academies (2022).
Asexuality, Romantic Attraction, and Category Coverage
Asexuality and related identities are important examples of why attraction, desire, behavior, and identity require separate questions. Some individuals identify as asexual while reporting romantic attraction, and some report sexual behavior for reasons that differ from ongoing sexual attraction. A survey that requires respondents to choose exclusively between heterosexual and homosexual identities produces an incomplete response space. Similarly, inferring an identity from the absence of sexual activity confuses an observation during a defined period with the person's own description of attraction Sell (1997) National Academies (2022).
Romantic attraction may be measured independently from sexual attraction when the scientific purpose requires it. Yet a design that adds multiple overlapping labels without interviewing prospective respondents risks confusing participants and obscuring the intended construct. Cognitive interviews with sexual and gender minority participants show the importance of wording and response choices that accommodate differences in preferred terminology. A useful questionnaire asks precise questions, permits nonapplicable responses, and documents whether categories are self-selected, algorithmically derived, or independently reported Suen et al. (2020) Rich et al. (2025).
Multidimensional reporting can preserve scientifically meaningful heterogeneity through disaggregated data while still allowing carefully documented broader categories for analyses with limited sample sizes. A recoding process should be reversible wherever privacy and governance permit, because future researchers may recognize distinctions hidden by earlier categorization. Derived variables should never be mistaken for more faithful descriptions of persons than the original responses; their legitimacy depends on an explicit scientific purpose and tested classification rules National Academies (2022) Mokkink et al. (2024).
Behavioral Measurement and the Temporal Reference Problem
A report of sexual behavior is a behavioral observation with a retrospective time window, not a substitute for enduring identity. Survey instruments often use lifetime, past-year, or recent-period questions; these windows answer different questions because partnership status, opportunities, and behavior change over time. Studies should identify which actions qualify, distinguish partnered from solitary behavior when relevant, and describe whether the gender of partners is self-reported or inferred. Survey design also influences what respondents perceive as sensitive or appropriate to disclose Sell (1997) Bureau of Labor Statistics (2017).
Retrospective self-report is shaped by memory, social desirability, question wording, and the level of behavioral detail requested. An item that asks broadly whether the respondent has engaged in sexual behavior may be interpreted differently by people with different experiences. Specific acts can improve operational clarity but may increase burden or raise concerns about privacy. Cognitive pretesting and a scientific rationale for each sensitive item are important because both overgeneralization and invasive specificity can compromise data quality Suen et al. (2020) Bureau of Labor Statistics (2017).
Researchers investigating psychological processes should analyze behavioral measures at the temporal scale on which the explanatory mechanism is expected to operate. Stress and relationship events may be proximal predictors of a recent behavior, whereas developmental experience may be examined through longer histories. Neither approach warrants a deterministic explanation of identity. Longitudinal analysis must further distinguish changes in reporting from changes in behavior and account for differential attrition if the most sensitive questions predict study dropout Rich et al. (2025) National Academies (2022).
Sexuality Instruments and Their Evidence Hierarchy
A named scale cannot be considered scientifically satisfactory solely because it is commonly used. The interpretation of its items, the relevance of each dimension, and its measurement properties need evidence in the population for which conclusions are drawn. Rich and colleagues' systematic review and narrative synthesis of sex, sexual orientation, and gender identity measurement in health research evaluates how the field operationalizes constructs and reports design considerations. Its existence illustrates an important shift from relying on labels alone to auditing the evidence base of the instruments that produce those labels Rich et al. (2025) Mokkink et al. (2024).
Ahmad and colleagues' 2026 systematic review of instruments measuring sex and gender attributes reported 34 identified instruments in 46 papers, of which 8 assessed sex traits and 26 assessed gender-related attributes. Their findings indicated uneven measurement-property evidence and limited validation beyond binary frameworks. These are results about the retrieved instrument literature, not a complete enumeration of all possible questionnaire designs. The review shows how an expanding instrument inventory can coexist with important gaps in tested validity, particularly across population boundaries Ahmad et al. (2026).
Instrument selection must therefore distinguish theoretical fit from empirical availability. An instrument may have excellent factor-analytic results while omitting an essential dimension, or a conceptually inclusive item may have insufficient test–retest evidence in a given setting. A structured decision process examines content validity, response processes, reliability, cross-group functioning, and feasibility together. The COSMIN guidance is useful for assessing the measurement evidence, although its application must be adapted where gender or sexuality constructs do not operate as reflective latent variables Mokkink et al. (2024) Campbell & Fiske (1959).
Cognitive Interviewing and the Psychology of Survey Response
Survey responses emerge from comprehension, retrieval, judgment, and the selection of an available answer. Multidimensional gender and sexuality items are particularly sensitive to these processes because respondents may understand terms differently, recall experiences across different periods, or hesitate to disclose information. Cognitive interviewing asks participants to explain their interpretation of questions and response options, revealing ambiguities that quantitative reliability statistics may not detect. In their qualitative study, Suen and colleagues combined focus groups and cognitive interviews with sexual and gender minority participants, showing why wording and category preferences require direct testing rather than assumptions made by researchers Suen et al. (2020) Bureau of Labor Statistics (2017).
A strong pretest recruits respondents who are likely to encounter the proposed instrument, including individuals whose identities or circumstances may fit poorly into default categories. Participants can be invited to paraphrase stems, explain what each response option means, and identify missing categories or confusing terms. The Bureau of Labor Statistics' work on potential sexual orientation and gender identity questions for the Current Population Survey illustrates how cognitive interviewing can assess the feasibility of adding sensitive demographic items to a major household survey Bureau of Labor Statistics (2017).
Cognitive interviews reveal possible mechanisms of error but cannot directly estimate their population prevalence unless the recruitment and design support that inference. Qualitative findings should guide revised items that can subsequently be tested with probability samples, randomized presentation designs, or structured psychometric methods. Researchers should record the language and cultural context of interviews, since translation can change the relation between everyday words and the psychological constructs the survey intends to measure Suen et al. (2020) Holtzman & Saperstein (2026).
Reliability, Measurement Error, and Classification Uncertainty
Measurement error may arise from random variation in response, systematic misunderstanding, ambiguous categories, incomplete recall, and social pressures. A short-term repeatability coefficient assesses only one subset of these errors. Heise and colleagues' repeated-measure work illustrates the distinction: high agreement over a short interval supports a degree of response consistency but does not resolve the construct meaning of the selected categories or their long-term stability Heise et al. (2025/2026) Cronbach & Meehl (1955).
Classification error can substantially alter estimated associations when one category is small or when the probability of error differs by subgroup. If some respondents are more likely to decline disclosure because of social context, complete-case analysis can systematically underrepresent precisely the people a study seeks to understand. A two-step classification algorithm should therefore be audited through missingness patterns, inconsistent responses, response options, and subgroup counts. Researchers can present results under alternative defensible rules to evaluate the sensitivity of conclusions to classification decisions Tordoff et al. (2025) Felt et al. (2025).
Reliability coefficients, sensitivity, specificity, agreement, and construct-validity evidence answer different questions. Agreement between two classification methods can be high because one category dominates a sample, while the rare subgroup of substantive interest has poor classification accuracy. The comparison of birth assignment and self-identified transgender modality thus requires attention to the full cross-tabulation, not only a summary agreement statistic. Reports should distinguish measured precision from the accuracy of inferences about the population Tordoff et al. (2025) National Academies (2022).
Structural Validity and the Limits of Latent-Variable Models
Factor analysis and item response models are useful when a theory proposes that observable indicators are manifestations of one or more underlying attributes. For example, a multi-item desire questionnaire might support a model distinguishing partnered and solitary interest. However, a set of gender categories, observed partner genders, hormones, and discrimination exposures does not necessarily represent interchangeable indicators of one latent factor. These variables can form a conceptual profile of different kinds of information rather than reflective manifestations of the same hidden trait Spector et al. (1996) Mokkink et al. (2024).
When an index combines sex-related traits, gendered experiences, and sexuality dimensions, its construction needs an explicit account of direction and weighting. A formative index is built by its constituent components and may change meaning when a component is added or removed. Internal consistency is usually inappropriate as a quality criterion for such a composite, because components need not correlate. Instead the evaluation should emphasize domain coverage, predictive relevance, reliability of individual components, and whether the index supports the scientific decision for which it was built Mokkink et al. (2024) Cronbach & Meehl (1955).
Researchers should resist using model fit as proof that an instrument captures an entire domain. Different factor structures can fit the same sample, while wording overlap or response style can create correlations among items. Replication in independent samples, substantive prediction, cognitive interviewing, and theoretical comparison provide additional evidence. A model that reproduces covariance well but collapses identity and social exposure into an arbitrary score can remain difficult to interpret scientifically Campbell & Fiske (1959) Ahmad et al. (2026).
Measurement Invariance Across Groups, Time, and Language
Measurement invariance asks whether observations support comparable interpretations across groups or occasions. For a reflective scale, tests may address factor structure, loading similarity, thresholds or intercepts, and residual variation, depending on the intended comparison. A questionnaire that exhibits adequate internal consistency in one group can still function differently elsewhere because terms, response thresholds, or item relevance differ. In multidimensional gender and sexuality research, such differences can arise from variations in language, age, culture, social disclosure, and institutional settings Mokkink et al. (2024) Rich et al. (2025).
Categorical identity items create special problems for conventional invariance testing. They often lack a continuous latent score, have sparse cells, and use options whose social meanings change over time. Their comparability may be better investigated through cognitive interviews, bridge studies using both old and new options, tests of differential nonresponse, and detailed documentation of recoding rules. Where differential item functioning is estimated, interpretation should identify which substantive characteristic is represented rather than treating the statistical test as self-sufficient evidence of bias Suen et al. (2020) National Academies (2022).
Longitudinal surveys sometimes face a dilemma: preserve an older category set for comparability or revise it to reflect current language and improved coverage. A planned transition can include overlapping forms, explicit mapping tables, and sensitivity analyses. These tools can estimate the degree to which a trend depends on the measurement instrument itself. For rare categories, preserving individual confidentiality may require suppressing public cross-tabs even when underlying bridge data are retained securely for authorized analyses Holtzman & Saperstein (2026) Heise et al. (2025/2026).
Sampling, Nonresponse, and Population Estimation
Population estimates depend on a defined target population, a sampling frame, recruitment probabilities, coverage, response behavior, and weighting. A large volunteer survey of sexual and gender minority adults can provide rich information about identity heterogeneity while remaining unsuitable for direct national prevalence estimates. Probability-based surveys support population inference when selection and nonresponse are properly accounted for, but such surveys may be too small to examine rare multidimensional response profiles with precision. The solution involves fit-for-purpose design and transparent interpretation rather than ranking every dataset by size National Academies (2022) Reisner et al. (2014).
Survey mode may influence who participates and which information is disclosed. In-person, telephone, web, school-based, and clinical surveys create different privacy expectations, interviewer effects, and participation barriers. Reisner and colleagues compared in-person and online respondents to the US National Transgender Discrimination Survey, illustrating why recruitment mode matters for evaluating transgender health research. This evidence does not supply universal correction factors; it demonstrates the need to analyze the actual population reached by each method Reisner et al. (2014).
Small subgroup sizes make the tension between statistical precision and confidentiality particularly acute. An exact percentage based on a handful of respondents is statistically unstable and may permit reidentification when combined with other information. Analysts can report confidence intervals, pool years where justified, predefine broader categories for limited analyses, or use secure-access environments, while documenting which distinctions remain unexamined. Pooling decisions should respect substantive differences and be presented as analytic simplifications rather than as claims that respondents share a single identity National Academies (2022) Cahill & Makadon (2014).
Association, Explanation, and Causal Identification
An observed association between a demographic category and a psychological outcome does not identify the mechanism that produced it. Measured group membership can be linked with multiple experiences, institutions, health determinants, and opportunities. If the actual scientific question concerns harassment or structural exclusion, those exposures should be directly measured. Meyer's minority-stress framework describes stigma-related pressures as possible mechanisms connecting social environments and mental health among sexual minority populations; its application depends on testing context-specific exposure and outcome pathways Meyer (2003) Gower et al. (2018).
Confounding can occur when measured groups differ in socioeconomic conditions, healthcare access, age, recruitment method, or other causes of the target outcome. Temporal ambiguity further limits cross-sectional analyses. Repeated observations, quasi-experimental designs, valid instruments, and causal diagrams can improve inference when assumptions are explicit and plausible. A gender or sexuality classification is often a descriptive variable rather than a well-defined manipulable treatment; causal questions become clearer when they specify an exposure or institutional mechanism that could meaningfully vary Meyer (2003) Restar et al. (2024).
Comparisons among groups also depend on how a binary or multi-category coding scheme determines the contrast. A change from two to six categories may reveal heterogeneity previously hidden in an average effect; it can also increase uncertainty in small cells. The inferential challenge is to preserve the distinctions relevant to the question and to report uncertainty accurately. Treating a model's adjusted coefficient as an inherent property of an identity group obscures the observational nature of the evidence and the role of social context Rich et al. (2025) National Academies (2022).
Reproducibility, Harmonization, and Measurement Documentation
Replication requires access to the operational choices that generated a result. Researchers should document exact question wording, response options, instructions, skip logic, mode, reference periods, recruitment, coding, missingness, and any category harmonization. Sharing only final regression coefficients makes it impossible to reconstruct how an identity or behavior variable was defined. Researchers can release a questionnaire and deidentified codebook while keeping respondent-level information protected, thereby supporting reproducibility without exposing sensitive disclosures National Academies (2022) Rich et al. (2025).
Harmonizing datasets requires distinguishing conceptual equivalence from the convenience of producing a common variable. A study that records attraction toward several genders cannot be directly pooled with a study that only records an identity label without a justified mapping. Researchers should identify comparability levels, preserve metadata about residual uncertainty, and run alternative harmonization scenarios when definitions differ. This approach makes longitudinal or cross-national synthesis more transparent and helps prevent spurious trends produced by coding changes Sell (1997) Mokkink et al. (2024).
Open science practices must be adjusted to the confidentiality risks of small populations. A public codebook, analysis plan, simulated dataset, and aggregate checks may support verification even when raw sensitive responses cannot be shared. The strongest practice separates transparency about method from indiscriminate disclosure of individual records. Prospective registration of a measurement-comparison study can further reduce selective reporting of favorable coding choices, while replication across diverse contexts provides stronger evidence of transportability National Academies (2022) Restar et al. (2024).
Cultural Translation and Contextual Validity
Sex and gender terms do not circulate independently of language and institutional practice. A term that appears self-evident in an English-language US questionnaire may be absent, ambiguous, politically charged, or used differently in another linguistic community. Researchers must investigate how the target population understands local concepts of gender, sexuality, identity, and disclosure. Forward and backward translation may detect textual errors, but comprehension testing and community consultation are needed to determine whether a translated item represents the intended construct. Direct comparability between national prevalence figures should never be assumed from the presence of superficially identical category labels Suen et al. (2020) National Academies (2022).
Sexual identity can be described through formal categories, culturally specific terms, relational practices, and preferences about disclosure. A study that imports one country's category structure without examining local meanings may systematically undercount individuals who would report relevant attractions or behavior under another set of questions. The resulting differences may reflect conceptual coverage or reporting context rather than population differences in the underlying experiences. Researchers can use comparative cognitive interviewing, locally developed categories, and harmonization rules with explicitly stated limitations Sell (1997) Rich et al. (2025).
Cross-cultural design also requires attention to the safety of participation. In some settings, disclosure of sexuality or gender identity may have substantial social or legal consequences. A survey that records exact identity responses together with a small geographic unit may increase risk even if names are omitted. The scientific advantages of collecting a detailed variable should be weighed against its genuine relevance and the possibility of reducing identifiability. Ethics and construct validity intersect because participants' expectations about confidentiality can influence whether they answer at all National Academies (2022) Restar et al. (2024).
Intersectional Measurement and Social Heterogeneity
Every gender and sexuality category intersects with variation in race, ethnicity, class, disability, age, migration, and social position. An aggregate group comparison can obscure differences in discrimination exposure or access to care among people assigned the same label. Analyses that use intersectional categories need adequate empirical support and a theory of how institutional conditions interact. Cross-tabulations alone do not explain mechanisms; carefully measured environments, repeated observations, and context-specific hypotheses are required Meyer (2003) Gower et al. (2018).
Multidimensional measurement can describe overlap across constructs without collapsing all possible combinations into ever smaller analytic cells. Researchers must anticipate that complex category intersections increase the number of parameters and may reduce precision. Partial pooling, preregistered contrasts, and transparent uncertainty estimates can improve stability in appropriate models, while disclosure control may constrain the level of detail released publicly. A responsible study reports where an analysis lacks the information needed to support a group-specific claim National Academies (2022) Rich et al. (2025).
Measurement categories are also affected by the researcher's standpoint. Observed gender presentation, identity reported by an individual, an externally imposed classification, and an institutional eligibility rule may interact to shape experiences of inclusion or exclusion. Separating those dimensions improves conceptual accountability: a difference attributed to 'gender' may arise through an environment's treatment of a person rather than through a stable individual disposition. The test of this proposition requires substantive measurements of the proposed social processes Gower et al. (2018) Meyer (2003).
Development, Adolescence, and Life-Course Measurement
Developmental research adds a temporal dimension to the measurement problem. Adolescents may be acquiring the language used in questionnaires, negotiating disclosure across school and family contexts, and distinguishing identity from presentation or attraction. Questions developed among adults can therefore be confusing or yield different response processes among younger participants. Repeated assessments are particularly valuable because they can distinguish stability of answers at short intervals from the developmental patterns the study seeks to explain Heise et al. (2025/2026) Suen et al. (2020).
Longitudinal results depend on attrition as well as measurement consistency. Participants who experience greater privacy concerns or difficulty finding a response option may be more likely to skip items or leave a study, changing the composition of later waves. Researchers should examine patterns of missing data and use sensitivity analyses appropriate to the plausible missingness mechanisms. When items are revised across waves, bridge samples can test whether apparent changes reflect real reported development or changes in the questionnaire's coverage Heise et al. (2025/2026) Holtzman & Saperstein (2026).
Developmental inquiry should not rely on a presumed universal sequence from attraction to behavior to identity, nor on an assumed timetable of categorical stability. The relevant questions concern individual trajectories, contexts, and the precision of measurements at each age. Consent and assent procedures, confidentiality limits, and the practical consequences of participation require special attention when sensitive questions are asked of minors. The empirical foundation for any particular developmental inference remains the sampled population and longitudinal design rather than a generalized narrative of identity formation National Academies (2022) Rich et al. (2025).
Clinical Records, Public Health, and Responsible Use
Clinical and public-health information systems require measurements that support specific decisions. Current gender identity may improve communication and the evaluation of social determinants of health; sex-related anatomy, hormones, or reproductive history can be relevant for defined clinical questions. Health records that store these as one mutable field risk confusing record history with current biological status and social identity. Cahill and Makadon describe the significance of collecting sexual orientation and gender identity data in health settings for identifying disparities, while the National Academies emphasize that measures must correspond to relevant research aims Cahill & Makadon (2014) National Academies (2022).
Routine data collection raises practical questions about acceptability, staff training, data systems, and patient control over disclosure. Research on patient acceptance of sexual orientation and gender identity questions in community health centers supports the feasibility of respectful collection in studied settings; it does not remove the need to explain purpose or safeguard access to sensitive responses. A clinical form should allow uncertainty, changes, and appropriate correction without requiring the patient to disclose information unrelated to care or research. Researchers must distinguish general guidance about data quality from medical advice for an individual patient Cahill & Makadon (2014) Cahill et al. (2014).
Public-health trend monitoring must interpret changes in reported prevalence alongside questionnaire revisions and changing disclosure conditions. An increase in respondents selecting a newly added category cannot automatically be treated as a biological shift or as evidence that the underlying phenomenon did not previously exist. Population denominators, survey mode, response options, and data suppression practices must be reported. The value of improved measurement is ultimately judged by whether it yields more interpretable estimates while preserving participation, safety, and public trust National Academies (2022) Restar et al. (2024).
Ethics, Privacy, and the Consequences of Classification
Gender and sexuality data can be personally sensitive because they may expose a person's relationships, identity, or circumstances to institutions or other people. The ethical duty extends beyond obtaining consent on an introductory form: researchers should establish why each item is needed, minimize collection, restrict access, assess risks of linkage with other datasets, and plan how granular results can be published safely. Rare combinations of demographic variables may permit inference about respondents even after direct identifiers are removed National Academies (2022) Cahill & Makadon (2014).
Measurement also has distributive consequences. An administrative coding rule can determine eligibility, subgroup visibility, or the interpretation of research findings. A technically consistent code that merges heterogeneous experiences can lead to misleading conclusions about needs and outcomes. Conversely, excessively detailed public reporting can compromise confidentiality. The responsible compromise depends on the analytic purpose, respondent expectations, sample sizes, legal environment, and safeguards; the selected approach should be documented and periodically reexamined Restar et al. (2024) Rich et al. (2025).
Scientific reporting can maintain transparency without exposing raw records. Questionnaires, category dictionaries, transformation code, aggregate validity tests, and deidentified synthetic examples can enable outside scrutiny. Access-controlled analysis environments may be appropriate when datasets contain fine-grained identity combinations or sensitive clinical histories. The obligation to protect participants should be treated as part of measurement quality, because breaches or fear of disclosure may alter recruitment and response processes in later studies National Academies (2022) Suen et al. (2020).
Critical Scientific Synthesis and Unresolved Problems
The strongest conclusion of the available research is methodological: sex-related attributes, gender identity and expression, attraction, behavior, and desire require explicit operational definitions, and the choice among measures changes what can be inferred. Consensus reports establish principled recommendations, experimental questionnaire studies demonstrate selected design effects, and psychometric studies provide context-specific evidence of measurement properties. These sources support multidimensional representation where it serves the research question. They do not establish one universally optimal questionnaire for every clinical, social, or developmental application National Academies (2022) Holtzman & Saperstein (2026) Ahmad et al. (2026).
Several unresolved problems recur across the literature. More independent testing is needed among multilingual and cross-cultural samples, younger and older populations, and people whose experience is poorly represented by dominant response choices. Longitudinal studies must distinguish legitimate changes in self-description from changes in item meaning. Probability surveys need ways to estimate relatively small populations while maintaining credible precision and confidentiality. Comprehensive reviews have identified a shortage of instruments with sufficiently demonstrated measurement properties across diverse populations and applications Rich et al. (2025) Ahmad et al. (2026) Heise et al. (2025/2026).
Future progress can be organized through an evidence-indexed comparison program. Investigators would begin by defining the construct and intended inference, develop candidate items through participatory cognitive interviews, compare alternate wordings experimentally where feasible, evaluate reliability and response-process evidence, and test generalization across settings and time. The results should be linked to accessible item specifications, documented coding decisions, sensitivity analyses, and carefully governed data access. This is a synthesis of existing methodological principles, not the claim of an independently validated new psychological theory Mokkink et al. (2024) Campbell & Fiske (1959) National Academies (2022).
Conclusion
Sex, gender, and sexuality constitute a connected but conceptually heterogeneous field of psychological measurement. Research gains explanatory precision when it records the particular sex-related characteristic, gender experience, expression, social position, attraction, orientation identity, behavior, or desire relevant to the question being tested. The historical development from categorical classification through behavioral surveys to multidimensional profiles and contemporary two-step gender measures demonstrates how changing concepts and instruments shape the scientific objects that appear in datasets. Each operational choice carries empirical, interpretive, and ethical consequences Kinsey et al. (1948) Tate et al. (2013) National Academies (2022).
The evidence supports a practical sequence of scientific work: define the construct and target population, examine participant interpretation, select or develop appropriate measures, assess relevant measurement properties, investigate missingness and classification error, specify temporal and cultural comparability, and analyze associations within a defensible inferential design. Experimental survey studies, methodological evaluations, community-informed interviews, systematic instrument reviews, and consensus guidance contribute complementary forms of evidence. Their limitations should remain visible in scientific conclusions Suen et al. (2020) Holtzman & Saperstein (2026) Ahmad et al. (2026) Mokkink et al. (2024).
A durable research infrastructure preserves the original meaning of observations, the decisions that transformed them into variables, and the boundaries of claims made from them. It also permits revision as populations, terminology, technologies, and scientific questions evolve. Multidimensional measurement thus advances psychological science through more explicit construct maps, better-supported comparisons, and accountable knowledge production. The next generation of evidence will depend on representative longitudinal designs, cross-cultural validation, independent replication, and governance that protects respondents while permitting legitimate scientific scrutiny Rich et al. (2025) Restar et al. (2024) National Academies (2022).
References
- Ahmad, N., Ruetas, A., Shaikh, U., Tylinski Sant’Ana, T., & Mollayeva, T. (2026). Systematic review of instruments for measuring sex and gender attributes: Assessment of measurement properties and utility in research on clinical and functional outcomes. BioMedical Engineering OnLine, 25, Article 84. https://doi.org/10.1186/s12938-026-01563-0
- Bureau of Labor Statistics. (2017). Assessing the feasibility of asking about sexual orientation and gender identity in the Current Population Survey: Results from cognitive interviews. U.S. Department of Labor. https://www.bls.gov/osmr/research-papers/2017/html/st170210.htm
- Cahill, S., & Makadon, H. (2014). Sexual orientation and gender identity data collection in clinical settings and in electronic health records: A key to ending LGBT health disparities. LGBT Health, 1(1), 34–41. https://doi.org/10.1089/lgbt.2013.0001
- Cahill, S., Singal, R., Grasso, C., King, D., Mayer, K., Baker, K., & Makadon, H. (2014). Do ask, do tell: High levels of acceptability by patients of routine collection of sexual orientation and gender identity data in four diverse American community health centers. PLOS ONE, 9(9), e107104. https://doi.org/10.1371/journal.pone.0107104
- Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the multitrait–multimethod matrix. Psychological Bulletin, 56(2), 81–105. https://doi.org/10.1037/h0046016
- Cronbach, L. J., & Meehl, P. E. (1955). Construct validity in psychological tests. Psychological Bulletin, 52(4), 281–302. https://doi.org/10.1037/h0040957
- Felt, D., Beach, L. B., Ashley, F., & Phillips, G. (2025). An exploratory comparison and evaluation of two two-step measures to identify transgender people in survey datasets. Transgender Health, 10(2), 210–219. https://doi.org/10.1089/trgh.2023.0010
- Gower, A. L., Rider, G. N., Coleman, E., Brown, C., McMorris, B. J., & Eisenberg, M. E. (2018). Perceived gender presentation among transgender and gender diverse youth: Approaches to analysis and associations with bullying victimization and emotional distress. LGBT Health, 5(5), 312–319. https://doi.org/10.1089/lgbt.2017.0176
- Heise, M. J., McGhee, C. M., Olson, K. R., Gelman, S. A., & Gallagher, N. M. (2026). Evaluating the reliability of a revised two-step assessment of sex and gender in adolescents and adults. International Journal of Social Research Methodology, 29(4), 587–603. (First published online October 20, 2025.) https://doi.org/10.1080/13645579.2025.2575304
- Holtzman, T., & Saperstein, A. (2026). Formatting the two-step gender measure: Experimental insights from the United States. Demographic Research, 54, 1125–1158. https://doi.org/10.4054/DemRes.2026.54.35
- Kinsey, A. C., Pomeroy, W. B., & Martin, C. E. (1948). Sexual behavior in the human male. W. B. Saunders. https://wellcomecollection.org/works/se8hb577
- Klein, F., Sepekoff, B., & Wolf, T. J. (1985). Sexual orientation: A multi-variable dynamic process. Journal of Homosexuality, 11(1–2), 35–49. https://doi.org/10.1300/J082v11n01_04
- Meyer, I. H. (2003). Prejudice, social stress, and mental health in lesbian, gay, and bisexual populations: Conceptual issues and research evidence. Psychological Bulletin, 129(5), 674–697. https://doi.org/10.1037/0033-2909.129.5.674
- Mokkink, L. B., Elsman, E. B. M., & Terwee, C. B. (2024). COSMIN guideline for systematic reviews of patient-reported outcome measures version 2.0. Quality of Life Research, 33, 2929–2939. https://doi.org/10.1007/s11136-024-03761-6
- National Academies of Sciences, Engineering, and Medicine. (2022). Measuring sex, gender identity, and sexual orientation. National Academies Press. https://doi.org/10.17226/26424
- Reisner, S. L., Conron, K., Scout, N., Mimiaga, M. J., Haneuse, S., & Austin, S. B. (2014). Comparing in-person and online survey respondents in the U.S. National Transgender Discrimination Survey: Implications for transgender health research. LGBT Health, 1(2), 98–106. https://doi.org/10.1089/lgbt.2013.0018
- Restar, A. J., Lett, E., Menezes, N. P., Molino, A. R., Poteat, T. C., Dean, L. T., Glick, J. L., Baker, K. E., & Cole, S. W. (2024). Getting precise about gender and sex measurement: A primer for epidemiologists. American Journal of Epidemiology, 193(12), 1861–1867. https://doi.org/10.1093/aje/kwae144
- Rich, A. J., McGorray, E. L., Felt, D., Kerr, M., Baldwin-SoRelle, C., Beach, L. B., Phillips, G., & Poteat, T. (2025). Sex, sexual orientation, and gender identity measurement in health research: A systematic review and narrative synthesis. LGBT Health, 12(6), 395–406. https://doi.org/10.1089/lgbt.2024.0105
- Sell, R. L. (1997). Defining and measuring sexual orientation: A review. Archives of Sexual Behavior, 26(6), 643–658. https://doi.org/10.1023/A:1024528427013
- Spector, I. P., Carey, M. P., & Steinberg, L. (1996). The sexual desire inventory: Development, factor structure, and evidence of reliability. Journal of Sex & Marital Therapy, 22(3), 175–190. https://doi.org/10.1080/00926239608414655
- Suen, L. W., Lunn, M. R., Katuzny, K., Finn, S., Duncan, L., Sevelius, J., Flentje, A., Capriotti, M. R., Lubensky, M. E., Hunt, C., Weber, S., Bibbins-Domingo, K., & Obedin-Maliver, J. (2020). What sexual and gender minority people want researchers to know about sexual orientation and gender identity questions: A qualitative study. Archives of Sexual Behavior, 49(7), 2301–2318. https://doi.org/10.1007/s10508-020-01810-y
- Tate, C. C., Ledbetter, J. N., & Youssef, C. P. (2013). A two-question method for assessing gender categories in the social and medical sciences. Journal of Sex Research, 50(8), 767–776. https://doi.org/10.1080/00224499.2012.690110
- Tordoff, D. M., Minalga, B., Ó Catháin, N., Fernandez, A., Gross, B., & Glick, S. N. (2025). Comparing two-step approaches to measuring gender identity: The reliability and applications of asking about sex assigned at birth versus transgender self-identification. American Journal of Epidemiology, 194(5), 1255–1263. https://doi.org/10.1093/aje/kwae341
- World Health Organization. (2006). Defining sexual health: Report of a technical consultation on sexual health, 28–31 January 2002. World Health Organization. https://www.who.int/publications/m/item/defining-sexual-health