1. Bibliographic Information

1.1. Title

Anxiety disorders, gender nonconformity, bullying and self‐esteem in sexual minority adolescents: prospective birth cohort study

1.2. Authors

The authors are Abbeygail Jones, Emily Robinson, Olakunle Oginni, Qazi Rahman, and Katharine A. Rimes. Their affiliation is the Institute of Psychiatry, Psychology and Neuroscience, King's College London, London, UK.

1.3. Journal/Conference

This paper was published in the Journal of Child Psychology and Psychiatry, Volume 58, pages 1269–1277. This is a highly reputable and influential journal in the field of child and adolescent mental health, known for publishing rigorous empirical research.

1.4. Publication Year

The paper was published in 2017, with a correction added on June 7, 2017, after its initial online publication.

1.5. Abstract

Sexual minority adolescents (youth not exclusively heterosexual) are at increased risk of anxiety disorders at 17.5 years compared to heterosexual youth, as assessed using a computerised structured clinical interview in a UK birth cohort (ALSPAC). Sexual minority status at 15.5 years was associated with higher early childhood gender nonconformity, lower self-esteem, and greater reported bullying. While adjusting for ethnicity, maternal occupation, and childhood gender nonconformity had minimal effect on anxiety risk, accounting for bullying and self-esteem reduced but did not eliminate the association. Results demonstrate bullying between ages 12–16 and lower self-esteem partially contribute to elevated anxiety risk in sexual minority youth.

1.6. Original Source Link

The original source link provided is: /files/papers/69056b090b2d130ab3e047eb/paper.pdf. Based on the context and the journal information, this appears to be the PDF link to the officially published version of the paper.

2. Executive Summary

2.1. Background & Motivation

The core problem the paper aims to solve is understanding why sexual minority adolescents (youth who do not exclusively identify as heterosexual) experience higher rates of anxiety compared to their heterosexual peers. This problem is significant because previous research, primarily relying on self-report questionnaires, consistently indicated this disparity, but lacked the rigor of standardized diagnostic tools in a prospective, longitudinal setting.

Specific challenges or gaps in prior research include:

1. Reliance on symptom questionnaires: Most previous studies measured anxiety using self-report questionnaires, which can indicate symptoms but not necessarily a clinical diagnosis. There was a lack of research utilizing standardized diagnostic tools.

2. Lack of prospective data: Many studies were cross-sectional or retrospective, making it difficult to establish causality or the temporal sequence of risk factors.

3. Limited investigation of mechanisms: While minority stress theory (Meyer, 2003) suggests stigmatization and discrimination contribute to psychopathology, the specific psychosocial mechanisms (childhood gender nonconformity, bullying, self-esteem) mediating this risk in a longitudinal context were not fully explored with diagnostic outcomes.

4. Absence of sex-separated analysis: Previous findings suggested potential sex differences in anxiety, self-esteem, bullying, and gender nonconformity, necessitating separate analyses for girls and boys.

   The paper's entry point or innovative idea is to address these gaps by using a large, prospective UK birth cohort study (the Avon Longitudinal Study of Parents and Children, ALSPAC) and a standardized, computerised clinical interview (Clinical Interview Schedule-Revised, CIS-R) to diagnose anxiety disorders. This allows for a more robust investigation of the association between sexual minority status and clinical anxiety, and the prospective examination of potential contributing factors like childhood gender nonconformity (CGN), bullying, and self-esteem over time.

2.2. Main Contributions / Findings

The paper makes several primary contributions:

1. First use of structured diagnostic tool in a prospective cohort: It is the first study to use a standardized diagnostic tool (CIS-R) within a prospective birth cohort to assess anxiety disorders in sexual minority adolescents, confirming previous findings from questionnaire-based studies with higher diagnostic rigor.

2. Confirmed increased risk of anxiety: It rigorously demonstrates that sexual minority adolescents are at a significantly increased risk (approximately 2.5 times higher odds) of being diagnosed with an anxiety disorder at 17.5 years compared to heterosexual youth, for both girls and boys.

3. Identified mediating factors: The study shows that bullying experienced between ages 12-16 and lower self-esteem at 17.5 years partially contribute to this elevated anxiety risk. Adjusting for these factors reduced, but did not eliminate, the association between sexual minority status and anxiety.

4. Clarified role of childhood gender nonconformity (CGN): While CGN was found to be associated with later sexual minority orientation (consistent with previous research), it had a minimal effect on the association between sexual orientation and the risk of an anxiety disorder, contrary to the study's hypothesis for its direct mediating role.

5. Sex-separated analysis: The study consistently analyzed data for girls and boys separately, revealing that the patterns of association and partial mediation by bullying and self-esteem were similar across sexes, though there were some nuances in the significance of CGN.

   These findings solve the problem of understanding the scope and some of the underlying mechanisms contributing to mental health disparities in sexual minority youth by providing robust, longitudinally observed, and clinically diagnosed evidence. They highlight specific psychosocial targets (bullying and self-esteem) for prevention and intervention strategies.

3. Prerequisite Knowledge & Related Work

3.1. Foundational Concepts

• Sexual Minority (SM): This term refers to individuals who identify with a sexual orientation other than heterosexual (e.g., gay, lesbian, bisexual, pansexual, asexual) or who experience same-sex attraction or behavior. In this study, sexual minority adolescents were defined as youth not exclusively heterosexual.
• Anxiety Disorders: These are a group of mental health conditions characterized by persistent, excessive, and often irrational fear or worry that significantly interferes with daily life. Common types include generalized anxiety disorder, agoraphobia, social phobia, specific phobia, and panic disorder.
• ICD-10 (International Classification of Diseases, 10th Revision): This is a globally recognized diagnostic tool developed by the World Health Organization (WHO) for categorizing diseases, health problems, and mental health conditions. It provides diagnostic criteria that clinicians use to diagnose disorders. The CIS-R in this study produces ICD-10 diagnoses.
• CIS-R (Clinical Interview Schedule-Revised): A computerised structured clinical interview used to assess psychiatric disorders. It is a standardized tool designed to be reliable for diagnosing various mental health conditions, including anxiety disorders, based on ICD-10 criteria. Its structured format and computerization aim to reduce interviewer variability and improve diagnostic consistency.
• Gender Nonconformity (GN or CGN): Refers to behaviors, interests, or traits that do not align with culturally expected gender roles for a person's assigned sex at birth. Childhood gender nonconformity (CGN) specifically refers to these characteristics observed during childhood. Examples include boys preferring activities or toys typically associated with girls, or girls preferring activities or toys typically associated with boys. It is important to note that gender nonconformity is distinct from gender identity or gender dysphoria, though related.
• Minority Stress Theory (Meyer, 2003): This theory posits that sexual minorities experience unique forms of stress due to social stigma, prejudice, and discrimination. This minority stress can lead to negative mental health outcomes either directly (e.g., experiencing victimization) or indirectly through various psychological processes (e.g., internalized homophobia, concealment, expectations of rejection, lower self-esteem, rumination).
• Longitudinal Study / Birth Cohort Study: A research design that involves repeated observations of the same variables over long periods, often decades. A birth cohort study specifically follows a group of individuals born during a particular period from birth through their lives. The ALSPAC (Avon Longitudinal Study of Parents and Children) is an example, allowing researchers to study developmental trajectories and risk factors prospectively.
• Logistic Regression: A statistical method used to model the probability of a binary outcome (e.g., presence or absence of an anxiety disorder) based on one or more predictor variables. It estimates the relationship between a dependent binary variable and one or more independent variables.
• Odds Ratio (OR): In logistic regression, the odds ratio quantifies the strength of association between an exposure (e.g., sexual minority status) and an outcome (e.g., anxiety disorder). An OR of 1 indicates no association. An OR greater than 1 suggests an increased likelihood of the outcome with the exposure, while an OR less than 1 suggests a decreased likelihood. For example, an OR of 2.55 means the odds of having an anxiety disorder are 2.55 times higher for the exposed group compared to the unexposed group.
• Confidence Interval (CI): A confidence interval provides a range of values within which the true odds ratio (or other parameter estimate) is likely to fall. A 95% CI means that if the study were repeated many times, 95% of the calculated intervals would contain the true population parameter. If the CI for an OR does not include 1, the association is considered statistically significant.
• P-value: In hypothesis testing, the p-value indicates the probability of observing a test statistic as extreme as, or more extreme than, the one calculated from the sample data, assuming the null hypothesis (e.g., no association between variables) is true. A small p-value (typically less than 0.05) suggests that the observed data are unlikely under the null hypothesis, leading to its rejection and supporting the alternative hypothesis (e.g., a significant association).

3.2. Previous Works

The paper builds upon a body of existing literature, often highlighting their limitations to underscore its own unique contributions:

• Systematic Review by Ploderl & Tremblay (2015): This review indicated that sexual minority adolescents generally report increased levels of anxiety compared to heterosexual adolescents. This discrepancy was found across different sexual orientation identifications and for both females and males, though effect sizes were generally larger for males. The current paper notes that this review, like many others, relied on questionnaire ratings or self-report of professional diagnosis, lacking the structured diagnostic tool used here.
• Minority Stress Theory (Meyer, 2003) and its mechanisms (Hatzenbuehler, 2009): Previous work extensively theorized that minority stress due to stigmatization and discrimination leads to psychopathology. This paper seeks to empirically test specific pathways (victimization, self-esteem, CGN) within this theoretical framework.
• Victimization and distress (Birkett et al., 2015; Burton et al., 2013; Poteat et al., 2014):
  • Birkett et al. (2015) reported that higher levels of victimisation predicted greater levels of later distress in sexual minority adolescents, measured by the Brief Symptoms Inventory.
  • Burton et al. (2013) found victimisation to be a significant mediator of the effect of sexual orientation on depressive symptoms and suicidality.
  • Poteat et al. (2014) showed homophobic victimisation predicted increased anxiety in heterosexual adolescents over a school year. The current study notes that no previous longitudinal studies had investigated the contribution of victimisation to the increased risk of anxiety specifically in sexual minority adolescents using a diagnostic outcome.
• Self-esteem as a mediator (Hershberger & D'Augelli, 1995; Ploderl & Fartacek, 2005):
  • Hershberger and D'Augelli (1995) reported that low self-esteem mediated the relationship between victimisation and poor mental health in sexual minority youth, but this effect was contingent on family support.
  • Ploderl and Fartacek (2005) found that controlling for self-esteem attenuated the relationship between sexual minority status and suicidality. The current paper highlights that the role of self-esteem in anxiety among sexual minority youth had yet to be studied longitudinally.
• Childhood Gender Nonconformity (CGN) and later outcomes (Bailey & Zucker, 1995; Steensma et al., 2012; Alanko et al., 2009; Roberts et al., 2012; Van Beusekom et al., 2016):
  • CGN has been associated with adult homosexuality in retrospective studies (e.g., Bailey & Zucker, 1995) and one prospective study (Steensma et al., 2012).
  • Alanko et al. (2009) found recalled CGN to be associated with increased anxiety and depression, especially among sexual minorities.
  • Roberts et al. (2012) found CGN partially accounted for abuse and post-traumatic stress disorder (PTSD) in sexual minorities.
  • Van Beusekom et al. (2016) found self-reported gender nonconformity to be associated with social interaction anxiety, partially mediated by homophobic victimisation. The current study points out that many of these studies relied on retrospective CGN measures (prone to recall bias) or were cross-sectional. There was a lack of prospective studies investigating the role of CGN (reported during childhood) and anxiety disorders in sexual minority adolescents using objective child-reported measures.

3.3. Technological Evolution

The field of mental health research, particularly concerning marginalized groups, has evolved towards greater methodological rigor. This paper's work fits within this evolution by:

• Moving from self-report to diagnostic tools: Early research often relied on symptom questionnaires, which are useful for screening but don't provide a clinical diagnosis. The use of structured clinical interviews like the CIS-R represents an advancement in diagnostic accuracy and standardization, providing a more robust measure of anxiety disorders.
• Embracing longitudinal, prospective designs: Cross-sectional studies can identify correlations but struggle to establish causality. Prospective birth cohort studies, like ALSPAC, are at the forefront of epidemiological research, allowing researchers to track individuals over time, assess exposures before outcomes occur, and better understand developmental trajectories and causal pathways. This minimizes recall bias and strengthens causal inference.
• Integrating psychosocial factors: The shift from simply identifying disparities to investigating the complex psychosocial mechanisms (minority stress, bullying, self-esteem, CGN) reflects a more nuanced understanding of mental health, moving beyond descriptive epidemiology to explanatory models.
• Statistical advancements: The use of multiple imputation addresses missing data in longitudinal studies, a common challenge that can lead to biased estimates and reduced statistical power. Logistic regression is a standard and appropriate tool for analyzing binary outcomes like a diagnosis.

3.4. Differentiation Analysis

Compared to the main methods in related work, this paper's approach offers several core differences and innovations:

• Gold Standard Outcome Measure: Unlike prior studies that used symptom questionnaires or self-reported diagnoses, this paper employs a computerised structured clinical interview (CIS-R) to generate ICD-10 diagnoses of anxiety disorders. This provides a gold standard measure of clinical anxiety, significantly increasing the diagnostic validity and rigor of the findings.

• Prospective Birth Cohort Design: The use of the ALSPAC birth cohort allows for prospective data collection, meaning risk factors (like CGN, bullying) and sexual orientation were assessed at earlier time points than the anxiety outcome. This temporal ordering strengthens the ability to infer causality and minimizes recall bias, especially for childhood gender nonconformity measures (mother-reported and child-reported).

• Longitudinal Assessment of Mediators: The study prospectively examines the roles of bullying and self-esteem as potential mediators between sexual minority status and anxiety, with bullying reported between 12-16 years and self-esteem at 17.5 years (though the latter is concurrent with anxiety outcome, it follows sexual orientation assessment).

• Childhood Gender Nonconformity (CGN) from multiple reporters: It uses both mother-reported CGN from early childhood (30, 42, 57 months) and child-reported CGN at 8 years, providing a more comprehensive and less biased assessment than retrospective self-reports typically used in prior studies.

• Separated Sex Analyses: The analysis is conducted separately for girls and boys, following recommendations to identify potential sex differences and avoid missing nuances in risk factors, which many previous aggregated studies might have overlooked.

• Robust Statistical Handling of Missing Data: The application of multiple imputation helps to address the attrition inherent in longitudinal studies, thereby reducing bias and increasing statistical power, which is a common methodological challenge in large cohorts.

  In essence, this paper differentiates itself by combining methodological strengths (prospective design, diagnostic outcome, objective early life measures) to provide a more definitive and granular understanding of anxiety risk in sexual minority youth compared to the predominantly self-report, cross-sectional, or retrospective nature of much of the prior literature.

4. Methodology

4.1. Principles

The core idea behind this study's methodology is to leverage the strengths of a large, prospective birth cohort study (ALSPAC) to rigorously investigate the association between adolescent sexual minority orientation and the diagnosis of anxiety disorders using a standardized clinical assessment. The study then systematically explores whether this association is mediated or explained by childhood gender nonconformity (CGN), bullying experiences during adolescence, and self-esteem, by introducing these factors sequentially into logistic regression models. A key principle is to conduct analyses separately for girls and boys to account for potential sex-specific differences in these relationships, as recommended by previous research. By taking a prospective approach, the study aims to establish a clearer temporal sequence between potential risk factors and mental health outcomes, thereby strengthening causal inference compared to cross-sectional or retrospective designs.

4.2. Core Methodology In-depth (Layer by Layer)

4.2.1. Sample

The study utilized secondary data from the Avon Longitudinal Study of Parents and Children (ALSPAC), a large UK birth cohort.

• Recruitment: All pregnant women in the Avon Health Authority with due dates between April 1, 1991, and December 31, 1992, were invited to participate. Written informed consent was obtained.
• Initial Cohort: This resulted in a core cohort of 14,541 pregnancies, leading to 14,062 live births and 13,988 babies alive at 1 year.
• Expanded Cohort: Additional recruitment expanded the sample to 15,247 pregnancies (15,458 foetuses for analyses after age seven), with 14,775 live births and 14,701 alive at 1 year.
• Clinic Attendance: A random 10% sub-sample attended clinics at the University of Bristol at various intervals between 4 and 61 months of age.
• Study Sample: The analyses presented in this paper are based on 4,564 participants who completed the Clinical Interview Schedule-Revised (CIS-R) at 17.5 years of age, which represents approximately 65% of the original cohort invited to the research clinic at that age.
• Representativeness & Attrition Bias: The ALSPAC cohort was generally representative of the UK general population, although non-White minority ethnic groups were under-represented. To assess attrition bias (potential systematic differences between participants who remained in the study and those who dropped out), the CIS-R respondents were compared to the rest of the original sample on main study variables.
  • Demographics: The CIS-R sample had a higher proportion of White participants and mothers in professional/skilled occupations. This was tested using a chi-square test (a statistical test to determine if there is a significant association between two categorical variables):
    • For ethnicity: $\chi^2(1, 12150) = 7.2$, $p = .007$.
    • For maternal occupation: $\chi^2(1, 11127) = 100.6$, $p < .001$.
  • Gender Nonconformity (CGN):
    • Girls who completed the CIS-R reported lower gender nonconformity at 8 years than girls who did not (measured by Childhood Activities Inventory, CAI). This was tested using a t-test (a statistical test to compare the means of two groups): $M = 39.7$ ($SD = 12.4$) vs. $M = 40.7$ ($SD = 12.6$), $t(3,567) = 2.4$, $p = .015$.
    • Boys who completed the CIS-R had higher mother and child-reported gender nonconformity than boys who did not (measured by Pre-school Activities Inventory, PSAI and CAI).
      • PSAI: $M = 54.0$ ($SD = 3.1$) vs. $M = 54.5$ ($SD = 3.1$), $t(4,045) = 5.4$, $p < .0005$.
      • CAI: $M = 59.3$ ($SD = 11.2$) vs. $M = 60.3$ ($SD = 11.7$), $t(3,478) = 2.6$, $p = .009$.
  • No significant differences were found for other analysis variables. This detailed attrition analysis helps researchers understand potential biases in the generalizability of their findings.

4.2.2. Ethical Approval

Ethical approval for the ALSPAC study was obtained from the ALSPAC Law and Ethics Committee. For this specific research project, approval was also obtained from the King's College London College Research Ethics Committee (ref. PNM/14/15-67).

4.2.3. Measures

Anxiety outcome: CIS-R

• Assessment: The outcome variable was an ICD-10 diagnosis of any anxiety disorder at 17.5 years. This was determined using the computerised Clinical Interview Schedule-Revised (CIS-R).
• Reliability: The CIS-R is fully standardised, and both its computerised and interview versions are considered reliable measures of psychiatric disorders.
• Diagnoses Included: The outcome variable was binary (presence or absence of a disorder), indicating at least one of the following ICD-10 diagnoses: generalised anxiety disorder, agoraphobia, social phobia, specific phobia, or panic disorder.

Sexual orientation

• Assessment Time-point: Assessed only at 15.5 years.
• Question Format: Participants were asked to choose from a list, "the description that best fits how you think about yourself."
• Sample Size: Of the 4,564 CIS-R respondents at 17.5 years, 3,600 had responded to the sexual orientation question.
• Categorization: Due to small sample sizes, a dichotomous 'heterosexual versus nonheterosexual' variable was computed.
  • Heterosexual: 86.8% ($n=3,126$) identified as "100% heterosexual (straight)."
  • Nonheterosexual: 11.3% ($n=405$) were coded as nonheterosexual if they identified as:
    • "mostly heterosexual, but also attracted to own sex" ($n=333$, 9.3%)
    • "mostly homosexual, but also attracted to opposite sex" ($n=17$, 0.5%)
    • "100% homosexual (gay)" ($n=9$, 0.3%)
    • "bisexual (equally attracted to both sexes)" ($n=46$, 1.3%)
• Exclusions: Those who responded "not sure" ($n=57$, 1.6%) or "not sexually attracted to either sex" ($n=12$, 0.3%) were excluded, as previous research suggests these individuals often later identify as heterosexual or do not consider themselves sexual minorities.

Gender behaviour

Pre-school Activities Inventory (PSAI):

• Assessment Time-points: When the child was 30, 42, and 57 months old.
• Reporter: Mothers completed the PSAI.
• Scale Content: Consists of 24 items: 12 stereotypically masculine (e.g., "plays with tool set") and 12 stereotypically feminine (e.g., "plays with jewellery").
• Reliability & Validity: Good reliability (Cronbach's alpha .84 for male items, .90 for female items) and construct validity.
• Item Categories: Items are divided into Toy scale (7 items), Activity scale (11 items), and Character scale (6 items).
• Response Format: Parents reported if their child displayed behaviors "never", "hardly ever", "sometimes", "often", or "very often."
• Scoring: Responses were scored from 1 to 5, respectively. The Character scale was excluded from calculations because its items (e.g., avoiding risks, exploring, interest in insects, avoiding dirt) might overlap with anxiety disorder symptoms.
• Score Calculation: The PSAI is scored by summing items, then subtracting feminine items from masculine items. A transformation is applied to yield scores with a mean close to 50, where scores above 50 indicate more masculine behaviour (i.e., less gender nonconformity for boys, more gender nonconformity for girls, if higher masculinity is considered less conforming for girls).
• Analysis Variable: A mean of transformed Toy and Activity scale scores across the three time-points (30, 42, 57 months) was used for analysis.

Childhood Activities Inventory (CAI):

• Assessment Time-point: At 8 years old.
• Reporter: Children completed the CAI in face-to-face interviews.
• Scale Content: A 16-item version of the PSAI with age-appropriate additions.
• Reliability: Reported split-half reliability of 0.64. CAI scores at 8 years were associated with PSAI scores in earlier childhood.
• Interview Method: The child was presented with two envelopes for each item:
  • A blue statement suggesting the child did not engage in the behavior (e.g., "Some children play with dolls").
  • A red statement suggesting the child did engage in the behavior (e.g., "Other children don't play with dolls").
  • The child posted the statement into one of two slots ("sort of true for him/her" or "really true for him/her").
• Scoring: Items were scored as:
  • 1 = really true they do not identify with behavior.
  • 2 = sort of true they do not identify with behavior.
  • 3 = sort of true they do identify with behavior.
  • 4 = really true they do identify with behavior.

Bullying

• Assessment Time-point: At 16 years.
• Question Format: A questionnaire asked participants if they had experienced "bullying by another person" since the age of 12 years.
• Variable Type: A dichotomous variable (Yes/No) was created.
• Rationale: This variable was chosen because it was assessed after sexual orientation but referred to a time period prior to the anxiety outcome, making it a suitable potential mediating factor.

Self-esteem: Bachman self-esteem scale

• Assessment Time-point: At 17.5 years.
• Reporter: Completed by study children in an online survey.
• Scale: The Bachman revision of the Rosenberg's Self-Esteem Scale.
• Reliability & Validity: Good internal consistency ($Cronbach alpha = .75$) and construct validity.
• Content: 10 statements.
• Response Format: Rated on a scale from 1 "almost always true" to 5 "never true."
• Interpretation: Higher scores indicate higher self-esteem.
• Rationale: Selected as it was measured later than sexual orientation, positioning it as a potential mediator.

Demographic covariates

• Purpose: Included to adjust statistical models for potential confounding effects.
• Maternal occupation:
  • Categorization: Dichotomised into "$Skilled/managerial/professional$" versus "Partly skilled/unskilled" based on the Dale & Marsh (1993) classification.
• Ethnicity:
  • Categorization: Dichotomised into "White" versus "Non-white" due to the small numbers of non-White individuals in the sample.

4.2.4. Data analysis

Missing values

• Challenge: High levels of sample attrition and missing data are common in longitudinal designs.
• Solution: Multiple imputation was used to address power and sample size issues and reduce bias.
• Method: The mi impute chained command in Stata was used. This command performs multivariate imputation for both continuous and categorical variables with arbitrary missing value patterns.
• Imputations: Seventy imputations were carried out, determined by the level of missing data in the analysis variables.
• Assumption: Pre-imputation analysis confirmed that data were missing at random.
• Imputation Model: A small imputation model was implemented, including all analysis variables and a select number of auxiliary variables that (a) were related to the missingness of analysis variables and (b) were independently associated with the outcome variable.
• Outcome Variable Imputation: The analysis was based on imputed cases with complete outcome variable data (anxiety), as current recommendations advise against using imputed outcome measures.

Statistical analysis

• Software: Analyses were performed using Stata version IC 14.1.
• Primary Analysis: Logistic regression analyses were conducted to predict the odds ratios of participants having an anxiety disorder at 17.5 years, based on sexual orientation at 15.5 years, while adjusting for covariates.
• Sequential Model Building: Five sequential logistic regression models were conducted in addition to an initial unadjusted model:
  • Step 0 (Unadjusted Model): Tested the direct association between anxiety disorder (dependent variable) and sexual orientation (independent variable). $ \log\left(\frac{P(\text{Anxiety}=1)}{1-P(\text{Anxiety}=1)}\right) = \beta_0 + \beta_1 \cdot \text{Sexual Orientation} $ Where $P(\text{Anxiety}=1)$ is the probability of having an anxiety disorder, $\beta_0$ is the intercept, and $\beta_1$ is the log odds coefficient for Sexual Orientation.
  • Step 1 (Demographic Covariates): Maternal occupation and ethnicity were added to the model. These covariates were adjusted for at every step of the analysis. $ \log\left(\frac{P(\text{Anxiety}=1)}{1-P(\text{Anxiety}=1)}\right) = \beta_0 + \beta_1 \cdot \text{SO} + \beta_2 \cdot \text{Maternal Occ.} + \beta_3 \cdot \text{Ethnicity} $
  • Step 2 (Mother-reported CGN): Mother-reported CGN (between 30 and 57 months) was added. $ \log\left(\frac{P(\text{Anxiety}=1)}{1-P(\text{Anxiety}=1)}\right) = \beta_0 + \beta_1 \cdot \text{SO} + \beta_2 \cdot \text{Maternal Occ.} + \beta_3 \cdot \text{Ethnicity} + \beta_4 \cdot \text{Mother-reported CGN} $
  • Step 3 (Child-reported CGN): Child-reported CGN (at 8 years) was added. $ \log\left(\frac{P(\text{Anxiety}=1)}{1-P(\text{Anxiety}=1)}\right) = \beta_0 + \beta_1 \cdot \text{SO} + \beta_2 \cdot \text{Maternal Occ.} + \beta_3 \cdot \text{Ethnicity} + \beta_4 \cdot \text{Mother-reported CGN} + \beta_5 \cdot \text{Child-reported CGN} $
  • Step 4 (Bullying): Bullying (between 12 and 16 years) was added. $ \log\left(\frac{P(\text{Anxiety}=1)}{1-P(\text{Anxiety}=1)}\right) = \beta_0 + \dots + \beta_6 \cdot \text{Bullying} $
  • Step 5 (Self-esteem): Self-esteem (at 17.5 years) was added as the final covariate. $ \log\left(\frac{P(\text{Anxiety}=1)}{1-P(\text{Anxiety}=1)}\right) = \beta_0 + \dots + \beta_6 \cdot \text{Bullying} + \beta_7 \cdot \text{Self-esteem} $
• Sex-separated Analysis: All analyses were carried out separately for girls and boys.
• Multiple Imputation Handling: Logistic regression with multiply imputed data was performed using the mi estimate command. This command runs logistic regression on the original data and each of the 70 imputed datasets, then pools the results to produce a single, robust output.
• Variance Check: After each step, the variance of results between and within the imputation models was checked using the vartable option of mi estimate to ensure consistency. The pattern of findings for the non-imputed data set was also compared to the imputed data for robustness.

5. Experimental Setup

5.1. Datasets

The study exclusively used secondary data from the Avon Longitudinal Study of Parents and Children (ALSPAC).

• Source: A large, ongoing UK birth cohort study based in the Avon Health Authority.

• Scale: The original cohort involved over 14,000 pregnancies, with the analytical sample for this study comprising 4,564 participants who completed the CIS-R at 17.5 years.

• Characteristics:

  • Prospective: Data were collected longitudinally from the prenatal period into adolescence and adulthood. This allows for assessing exposures (e.g., CGN, sexual orientation, bullying) at earlier time points and outcomes (e.g., anxiety disorders) at later time points.
  • Comprehensive Data: ALSPAC collects a vast array of demographic, health, lifestyle, and psychosocial data through questionnaires, interviews, and clinical assessments.
  • Age Range: The data relevant to this study span from early childhood (30 months for mother-reported CGN) through adolescence (15.5 years for sexual orientation, 12-16 years for bullying) to late adolescence/early adulthood (17.5 years for anxiety disorder and self-esteem).
  • Representativeness: The cohort was broadly representative of the general UK population, although non-White minority ethnic groups were under-represented.

• Domain: Public health, developmental psychology, psychiatry, epidemiology.

• Data Sample Example (Conceptual): A hypothetical data sample for a single participant might look like this:

  • Participant ID: 12345
  • Sex: Female
  • Maternal Occupation: Skilled/Managerial/Professional
  • Ethnicity: White
  • Mother-reported CGN (mean score): 43.2 (indicating relatively typical feminine behavior for a girl, or slightly masculine for a boy)
  • Child-reported CGN (score at 8 years): 39.6 (indicating relatively typical feminine behavior for a girl)
  • Sexual Orientation (15.5 years): Nonheterosexual
  • Bullying (12-16 years): Yes
  • Self-esteem (17.5 years): 27.7 (a lower score indicating lower self-esteem)
  • Anxiety Disorder (17.5 years): Present

• Rationale for Choice: ALSPAC was chosen because it provides a unique opportunity to conduct a prospective birth cohort study with rich longitudinal data from early childhood through adolescence, allowing for the investigation of complex developmental pathways leading to mental health outcomes. Its large sample size, despite attrition, provides statistical power. The availability of mother-reported and child-reported CGN and a structured clinical interview for anxiety disorders at appropriate ages were crucial for addressing the research questions with high methodological quality.

5.2. Evaluation Metrics

The paper uses several standard statistical metrics to evaluate relationships between variables and the predictive power of its logistic regression models.

• Odds Ratio (OR):

  1. Conceptual Definition: The Odds Ratio quantifies the strength and direction of association between a specific exposure (e.g., sexual minority status) and a binary outcome (e.g., presence of an anxiety disorder). It indicates how much more likely (or less likely) the outcome is in the exposed group compared to the unexposed group.
  2. Mathematical Representation: For a logistic regression model, the Odds Ratio for a predictor $X_i$ is derived from its estimated coefficient $\beta_i$. $ \text{OR}_i = e^{\beta_i} $
  3. Symbol Explanation:
     • $\text{OR}_i$: The Odds Ratio for the $i$-th predictor.
     • $e$: Euler's number (the base of the natural logarithm), approximately 2.71828.
     • $\beta_i$: The estimated coefficient (log odds) for the $i$-th predictor in the logistic regression model.

• P-value:

  1. Conceptual Definition: The p-value assesses the statistical significance of an observed result. It represents the probability of obtaining data at least as extreme as the observed data, assuming that the null hypothesis (e.g., no effect or no difference) is true. A small p-value suggests that the observed data are unlikely under the null hypothesis, leading to its rejection.
  2. Mathematical Representation: The p-value is typically derived from a test statistic (e.g., Wald statistic in logistic regression, t-statistic in t-tests, chi-square statistic in chi-square tests). The specific formula depends on the test, but the interpretation is universal.
     • For a Wald test in logistic regression, the test statistic $Z = \frac{\beta_i}{\text{SE}(\beta_i)}$, where $\text{SE}(\beta_i)$ is the standard error of the coefficient. The p-value is then calculated from the standard normal distribution.
  3. Symbol Explanation:
     • $Z$: Wald test statistic.
     • $\beta_i$: Estimated coefficient.
     • $\text{SE}(\beta_i)$: Standard error of the estimated coefficient.

• Confidence Interval (CI):

  1. Conceptual Definition: A Confidence Interval provides a range of plausible values for an unknown population parameter (e.g., the true Odds Ratio). A 95% CI means that if the study were repeated many times, 95% of the calculated intervals would contain the true population parameter. If the CI for an Odds Ratio does not include 1, the OR is considered statistically significant.
  2. Mathematical Representation: For an Odds Ratio, a $100(1-\alpha)\%$ Confidence Interval is typically calculated as: $ [e^{\beta - Z_{\alpha/2} \cdot \text{SE}(\beta)}, \quad e^{\beta + Z_{\alpha/2} \cdot \text{SE}(\beta)}] $
  3. Symbol Explanation:
     • $e$: Euler's number.
     • $\beta$: The estimated coefficient (log odds).
     • $Z_{\alpha/2}$: The critical value from the standard normal distribution corresponding to the desired confidence level (e.g., for a 95% CI, $\alpha=0.05$, so $Z_{0.025} \approx 1.96$).
     • $\text{SE}(\beta)$: The standard error of the estimated coefficient.

• R-squared (Pseudo R-squared):

  1. Conceptual Definition: In logistic regression, R-squared (often called pseudo R-squared) is an approximate measure of the proportion of variance in the dependent variable that can be explained by the independent variables. Unlike linear regression, there isn't a single, universally accepted R-squared for logistic regression; several pseudo R-squared measures exist (e.g., Nagelkerke R-squared, McFadden R-squared). It gives an indication of the model's overall fit. A higher value indicates a better fit.
  2. Mathematical Representation (Example: Nagelkerke R-squared): $ R^2_{\text{Nagelkerke}} = \frac{1 - (L_0/L_M)^{2/N}}{1 - L_0^{2/N}} $ where $L_0$ is the likelihood for the null model (intercept only), $L_M$ is the likelihood for the fitted model, and $N$ is the sample size.
  3. Symbol Explanation:
     • $R^2_{\text{Nagelkerke}}$: The Nagelkerke pseudo R-squared value.
     • $L_0$: The likelihood of the data under the null model (a model with only an intercept, assuming no predictors).
     • $L_M$: The likelihood of the data under the fitted model (the model with predictors).
     • $N$: The sample size.

• Chi-square ($\chi^2$):

  1. Conceptual Definition: The chi-square test is used to determine if there is a significant association between two categorical variables (e.g., ethnicity and CIS-R completion status). It compares the observed frequencies in categories to the frequencies that would be expected if there were no association.
  2. Mathematical Formula: $ \chi^2 = \sum \frac{(O_i - E_i)^2}{E_i} $
  3. Symbol Explanation:
     • $\chi^2$: The chi-square test statistic.
     • $\sum$: Summation over all categories.
     • $O_i$: The observed frequency in category $i$.
     • $E_i$: The expected frequency in category $i$ (under the assumption of no association).

• Cohen's d:

  1. Conceptual Definition: Cohen's d is a standardized measure of effect size that quantifies the difference between two means in standard deviation units. It provides a measure of the magnitude of the observed difference, independent of sample size.
  2. Mathematical Formula (for two independent groups with similar sample sizes): $ d = \frac{\bar{x}_1 - \bar{x}_2}{s_p} $ where $s_p$ is the pooled standard deviation: $ s_p = \sqrt{\frac{(n_1-1)s_1^2 + (n_2-1)s_2^2}{n_1+n_2-2}} $
  3. Symbol Explanation:
     • $d$: Cohen's d effect size.
     • $\bar{x}_1$, $\bar{x}_2$: The means of group 1 and group 2, respectively.
     • $s_p$: The pooled standard deviation.
     • $n_1$, $n_2$: The sample sizes of group 1 and group 2, respectively.
     • $s_1^2$, $s_2^2$: The variances of group 1 and group 2, respectively.

• t-statistic (t):

  1. Conceptual Definition: The t-statistic is used in a t-test to determine if the means of two groups are significantly different from each other. It measures the difference between sample means relative to the variability within the samples.
  2. Mathematical Formula (for independent samples t-test with unequal variances, Welch's t-test): $ t = \frac{\bar{x}_1 - \bar{x}_2}{\sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}} $
  3. Symbol Explanation:
     • $t$: The t-statistic.
     • $\bar{x}_1$, $\bar{x}_2$: The means of group 1 and group 2, respectively.
     • $s_1^2$, $s_2^2$: The variances of group 1 and group 2, respectively.
     • $n_1$, $n_2$: The sample sizes of group 1 and group 2, respectively.

5.3. Baselines

In this study, the concept of "baselines" is not in terms of competing models or algorithms, but rather comparison groups and sequential statistical adjustments.

• Primary Comparison Group: The heterosexual youth cohort serves as the baseline comparison group against which the sexual minority youth are compared regarding anxiety disorder risk and other characteristics. This is representative as heterosexuality is the statistical norm, allowing for the quantification of disparities in minority populations.
• Sequential Model Baselines: The study uses a step-wise logistic regression approach where each preceding step acts as a baseline model for the subsequent step.

  • Step 0 (Unadjusted): The baseline model is the unadjusted association between sexual orientation and anxiety.

  • Step 1 (Demographics): The effect of sexual orientation is then compared to this baseline after adjusting for maternal occupation and ethnicity.

  • Subsequent Steps (CGN, Bullying, Self-esteem): Each new covariate added (CGN, bullying, self-esteem) provides a new baseline against which the reduction in the Odds Ratio for sexual orientation (and the R-squared increase for the overall model) is evaluated. This allows the researchers to assess the partial contribution of each factor to the observed disparity.

    This method is effective for validating the relative contributions of different factors in explaining an outcome within a single, complex longitudinal model, rather than comparing entirely different predictive models.

6. Results & Analysis

6.1. Core Results Analysis

The study's results consistently demonstrate a higher risk of anxiety disorders among sexual minority adolescents, with bullying and lower self-esteem partially explaining this disparity.

Sample Characteristics

As shown in Table 1, the sample comprised more females ($n=2,567$, 56.2%) than males ($n=1,996$, 43.7%). A higher percentage of girls (14%) identified as nonheterosexual compared to boys (8.2%).

Key differences between heterosexual and nonheterosexual participants:

• Anxiety Diagnosis: For both sexes, nonheterosexual youth were more likely to have an anxiety disorder diagnosis at 17.5 years.
  • Girls: 21.5% nonheterosexual vs. 10.2% heterosexual (Cohen's d = 0.24, a small effect size).
  • Boys: 11.6% nonheterosexual vs. 5.1% heterosexual (Cohen's d = 0.15, a small effect size).
• Bullying: Nonheterosexual youth were more likely to report bullying between ages 12-16.
  • Girls: 26.0% nonheterosexual vs. 17.4% heterosexual (Cohen's d = 0.15).
  • Boys: 28.9% nonheterosexual vs. 12.6% heterosexual (Cohen's d = 0.27). The effect size for boys was larger, indicating a more pronounced difference.
• Self-esteem: Nonheterosexual youth had lower self-esteem.
  • Girls: Mean score 26.3 for nonheterosexual vs. 27.7 for heterosexual (Cohen's d = 0.21).
  • Boys: Mean score 28.4 for nonheterosexual vs. 29.7 for heterosexual (Cohen's d = 0.22).
• Gender Nonconformity (CGN): Mother-reported CGN was higher for nonheterosexual participants in both sexes (i.e., more masculine behavior for girls, less masculine behavior for boys).
  • Girls: Mean score 44.0 (nonheterosexual) vs. 43.2 (heterosexual), t = -4.0, p < .001 (Cohen's d = -0.26). A lower score on the PSAI indicates more feminine behavior, so a higher score for nonheterosexual girls means more masculine behavior.
  • Boys: Mean score 52.9 (nonheterosexual) vs. 54.2 (heterosexual), t = 5.2, p < .001 (Cohen's d = 0.43). A higher score on the PSAI indicates more masculine behavior, so a lower score for nonheterosexual boys means less masculine behavior, i.e., higher gender nonconformity.
• Child-reported CGN at 8 years: For boys, nonheterosexual youth had significantly higher self-reported CGN (i.e., less masculine behavior) than heterosexual boys. For girls, there was a non-significant trend in the same direction.
• Demographic Covariates: There were no significant differences in ethnicity or maternal occupation between heterosexual and nonheterosexual participants for either sex.

Logistic Regression Analyses

Table 2 presents the results of the sequential logistic regression analyses for girls and boys separately.

• Step 0 (Unadjusted Association):

  • Sexual minority orientation at 15.5 years was significantly associated with an anxiety disorder diagnosis at 17.5 years for both sexes.
    • Girls: $OR = 2.55$ (CI 1.85-3.52), $p < .001$.
    • Boys: $OR = 2.48$ (CI 1.40-4.39), $p = .002$.
  • This initial finding strongly validates the paper's hypothesis that sexual minority youth face a substantially elevated risk of clinical anxiety. The pseudo R-squared was .02 for girls and .01 for boys, indicating that sexual orientation alone explained a small but significant portion of the variance in anxiety.

• Step 1 (Adjusting for Demographics):

  • Adjusting for maternal occupation and ethnicity had minimal effect on the ORs for sexual orientation.
    • Girls: $OR = 2.53$ (CI 1.84-3.50), $p < .001$.
    • Boys: $OR = 2.48$ (CI 1.40-4.39), $p = .002$.
  • Maternal occupation and ethnicity themselves were not significant predictors of anxiety in these models after adjusting for sexual orientation (p > .05). This suggests socioeconomic status and ethnicity did not confound the primary association.

• Step 2 & 3 (Adjusting for Childhood Gender Nonconformity - CGN):

  • Neither mother-reported nor child-reported CGN were significantly associated with anxiety disorder risk when added to the models initially.
  • Adjusting for CGN (both mother and child-reported) had minimal impact on the association between sexual orientation and anxiety disorder.
    • Girls: $OR = 2.54$ (CI 1.84-3.52), $p < .001$.
    • Boys: $OR = 2.44$ (CI 1.37-4.33), $p = .002$.
  • This finding challenges the hypothesis that CGN would directly mediate or explain a substantial portion of the anxiety risk in sexual minority youth.

• Step 4 (Adjusting for Bullying):

  • Bullying between 12-16 years was a significant independent risk factor for anxiety disorder diagnosis for both sexes.
    • Girls: $OR = 2.25$ (CI 1.65-3.05), $p < .001$.
    • Boys: $OR = 2.62$ (CI 1.61-4.27), $p < .001$.
  • Crucially, adjusting for bullying reduced the odds ratios for sexual orientation and anxiety diagnosis.
    • Girls: OR decreased from 2.54 to 2.34 (CI 1.69-3.25), $p < .001$.
    • Boys: OR decreased from 2.44 to 2.10 (CI 1.17-3.79), $p = .013$.
  • The pseudo R-squared for the models increased to .04, indicating that bullying explained additional variance. This suggests that bullying partially mediates the link between sexual minority status and anxiety, consistent with the minority stress theory.
  • For girls, mother-reported CGN became significantly associated with anxiety ($OR = 0.96$, $p = .037$), indicating a protective effect of higher masculinity (less nonconformity) on anxiety for girls when bullying is considered. This is an interesting, nuanced finding.

• Step 5 (Adjusting for Self-esteem):

  • Low self-esteem (higher scores on the Bachman scale indicate higher self-esteem, so $OR < 1$ indicates low self-esteem is a risk factor) was significantly associated with anxiety disorder diagnosis for both sexes.

    • Girls: $OR = 0.89$ (CI 0.88-0.91), $p < .001$.
    • Boys: $OR = 0.91$ (CI 0.88-0.94), $p < .001$.

  • Adjusting for self-esteem further reduced the odds ratios for sexual orientation and anxiety diagnosis.

    • Girls: OR decreased from 2.34 to 2.14 (CI 1.52-3.01), $p < .001$.
    • Boys: OR decreased from 2.10 to 1.93 (CI 1.06-3.54), $p = .032$.

  • The pseudo R-squared increased substantially to .12 for girls and .08 for boys, indicating self-esteem is a strong predictor and accounts for a significant portion of variance.

  • Both sexual orientation and bullying remained significant independent predictors even after adjusting for self-esteem, confirming their partial contributions. Bullying's association with anxiety also slightly reduced after accounting for self-esteem.

  • For girls, child-reported CGN now showed a significant association with anxiety ($OR = 1.01$, $p = .023$), suggesting that higher child-reported CGN (i.e., less typical feminine behavior) was a risk factor for anxiety when self-esteem and bullying are controlled. This shows the complex interplay of factors and how their significance can change in multivariate models.

    In summary, the results strongly validate that sexual minority adolescents have a heightened risk of anxiety disorders. Bullying and lower self-esteem partially mediate this risk, but a significant independent association remains, suggesting other unexplored factors. CGN showed complex and less direct associations. The consistency of findings across sexes, despite some minor differences in CGN's effects, strengthens the overall conclusions. The variance tests for multiple imputation confirmed a consistent pattern of results, adding to the robustness of the findings.

6.2. Data Presentation (Tables)

The following are the results from [Table 1] of the original paper:

The following are the results from [Table 2] of the original paper:

6.3. Ablation Studies / Parameter Analysis

The sequential logistic regression models (Steps 0-5) effectively function as an ablation study to assess the relative contributions and mediating effects of demographic covariates, childhood gender nonconformity (CGN), bullying, and self-esteem.

• Baseline (Step 0): Establishes the fundamental association between sexual minority status and anxiety disorder. The Odds Ratios (OR) of 2.55 for girls and 2.48 for boys represent the initial, unattenuated risk. The small pseudo R-squared values (.02 for girls, .01 for boys) reflect that sexual orientation is a significant, but not the sole, predictor.
• Demographic Controls (Step 1): Adding maternal occupation and ethnicity had negligible impact on the sexual orientation ORs and R-squared values. This indicates that socioeconomic status and ethnicity do not confound the relationship between sexual minority status and anxiety in this cohort, or at least not significantly enough to explain the observed disparity. This effectively "ablates" the influence of these demographic factors.
• Childhood Gender Nonconformity (CGN) (Steps 2 & 3): Incorporating mother-reported and child-reported CGN also had minimal effect on the sexual orientation ORs. This finding is contrary to the initial hypothesis that CGN would significantly mediate the risk. It suggests that while CGN is associated with sexual minority orientation itself, it does not directly explain a substantial portion of the anxiety risk in this context. The R-squared increased only slightly (to .03 for girls, .01 for boys), reinforcing CGN's limited direct explanatory power for anxiety. However, the subsequent emergence of CGN as a significant predictor for girls in later steps (after bullying and self-esteem are added) suggests a complex, possibly indirect, relationship rather than a direct mediating one. For instance, for girls, mother-reported CGN became protective (OR < 1) at Step 4, and child-reported CGN became a risk factor (OR > 1) at Step 5. This hints at conditional effects or suppression rather than simple mediation.
• Bullying (Step 4): The introduction of bullying dramatically impacted the models.
  • Bullying itself emerged as a strong independent risk factor for anxiety (ORs > 2 for both sexes).
  • The ORs for sexual orientation decreased notably (from 2.54 to 2.34 for girls, and 2.44 to 2.10 for boys). This "ablation" of bullying's effect demonstrates that bullying partially explains the elevated anxiety risk among sexual minority youth.
  • The R-squared also increased more substantially (to .04 for both), indicating that bullying explained a greater proportion of variance than demographics or CGN.
• Self-esteem (Step 5): Adding self-esteem further refined the understanding.

  • Low self-esteem was a very strong independent risk factor for anxiety (ORs ~ 0.9, meaning each unit increase in self-esteem reduces odds of anxiety by ~10-11%).

  • The ORs for sexual orientation further decreased (from 2.34 to 2.14 for girls, and 2.10 to 1.93 for boys). This indicates that low self-esteem also partially mediates the anxiety risk.

  • The R-squared values saw the largest jump (to .12 for girls, .08 for boys), highlighting self-esteem as a major explanatory factor.

  • Importantly, even after accounting for bullying and self-esteem, sexual minority status remained a significant predictor of anxiety (ORs ~ 2 for both sexes). This suggests that other factors related to minority stress (e.g., internalized homophobia, discrimination, expectations of rejection) or unidentified mechanisms continue to contribute to the increased risk.

    In essence, this sequential analysis acts as a robust method to disentangle the complex relationships, showing that while bullying and low self-esteem are critical pathways, they do not fully account for the observed mental health disparities in sexual minority adolescents. The varying impact and significance of CGN at different steps highlight the need for careful interpretation of mediation in complex longitudinal models.

7. Conclusion & Reflections

7.1. Conclusion Summary

This prospective birth cohort study provides robust evidence that sexual minority adolescents are at a significantly increased risk of developing anxiety disorders by 17.5 years, compared to their heterosexual peers. This finding is particularly strong due to the use of a standardized clinical diagnostic tool (CIS-R) and a longitudinal design. The study identified bullying experiences during adolescence (ages 12-16) and lower self-esteem at 17.5 years as significant psychosocial factors that partially contribute to this elevated anxiety risk for both girls and boys. While childhood gender nonconformity (CGN) was associated with later sexual minority orientation, it had minimal direct impact on the association between sexual orientation and anxiety disorder diagnosis. Even after accounting for these important mediating factors, the increased risk for sexual minority youth remained statistically significant, suggesting that other unique stressors or vulnerabilities are still at play. These findings underscore the importance of addressing victimisation and self-esteem in clinical assessments and interventions for anxiety disorders among sexual minority adolescents.

7.2. Limitations & Future Work

The authors acknowledge several limitations, which also point to future research directions:

• Under-reporting of Minority Sexual Orientation: The stigma associated with nonheterosexuality might have led to under-reporting of minority sexual orientation, especially given the data collection period (2006-2008) when societal attitudes were less accepting than currently. This could lead to an underestimation of the true prevalence and potentially attenuate the observed associations.

• Single Time-point Assessment of Sexual Orientation and Identity: Sexual orientation was assessed only at 15.5 years and defined by identity and attraction, not behavior. Sexual identity can evolve, especially among females (sexual fluidity), meaning some individuals might have changed their identity after this assessment, or the categorization might not fully capture the complexity of sexual fluidity.

• Combined Nonheterosexual Group: Due to small sample sizes for specific sexual minority subgroups (e.g., gay, lesbian, bisexual), all non-exclusively heterosexual individuals were combined. This obscures potential differences in risk factors and outcomes between these distinct subgroups (e.g., bisexual individuals may face unique stressors). Future research with larger, more diverse samples is needed to allow for subgroup analyses.

• Attrition Bias related to CGN: The CIS-R sample showed some attrition bias regarding CGN (girls who completed CIS-R reported lower CGN, boys who completed CIS-R reported higher CGN). This could bias the CGN results, making it difficult to generalize.

• Timing of Self-esteem Measurement: Self-esteem was measured at the same time-point (17.5 years) as the anxiety outcome. This limits the ability to perform mediation analyses to definitively establish that low self-esteem precedes and causes increased anxiety in this context. It's possible that anxiety itself reduces self-esteem. Future prospective studies should measure self-esteem at an intermediate time-point between sexual orientation and anxiety outcome.

• Unstandardized and Limited Bullying Measure: Bullying was assessed using an unstandardized measure ("bullying by another person") and only up to age 16. Bullying experiences between 16-17.5 years, or more specific forms of bullying (e.g., homophobic bullying), could also contribute to anxiety at 17.5 years. A stronger study design would involve standardized measures and multiple, contiguous time-points for assessing bullying.

• Under-representation of Non-White and Low Socioeconomic Status Groups: The ALSPAC cohort under-represented non-White participants and those with mothers in unskilled occupations. This limits the generalizability of findings to more diverse populations, especially considering that minority stress theory suggests multiple stigmatized identities could lead to greater adverse psychological outcomes.

  Future work should therefore focus on:

• Conducting mediation analyses with appropriately timed measures of self-esteem.

• Utilizing larger, more diverse cohorts to enable subgroup analyses within sexual minority populations and to understand the intersection of sexual orientation with ethnicity and socioeconomic status.

• Investigating adolescent gender nonconformity and its association with anxiety disorders.

• Exploring other potential mediators of minority stress (e.g., internalized homophobia, perceived discrimination, lack of social support).

7.3. Personal Insights & Critique

This paper represents a significant step forward in understanding mental health disparities in sexual minority youth. Its use of a prospective birth cohort and a structured clinical interview for anxiety diagnosis sets a high bar for methodological rigor, overcoming key limitations of much previous research. The confirmation of a ~2.5 times higher odds ratio for anxiety disorders among sexual minority adolescents provides compelling evidence for this disparity, moving beyond self-reported symptoms to clinical diagnoses.

Inspirations:

• Importance of Longitudinal Data: The study powerfully demonstrates the value of longitudinal data in disentangling complex relationships, especially in sensitive areas like minority stress. The ability to track CGN from early childhood and bullying through adolescence before anxiety diagnosis is invaluable.
• Nuance of Mediation: The partial mediation by bullying and self-esteem is crucial. It highlights concrete, modifiable targets for intervention, informing clinicians, educators, and policymakers on where to focus efforts. The fact that a significant risk remains after accounting for these factors also underscores the multifaceted nature of minority stress and the need for comprehensive support systems.
• Methodological Sophistication: The careful handling of attrition bias through multiple imputation and comparison of imputed vs. non-imputed results exemplifies best practices in longitudinal research.

Potential Issues and Areas for Improvement (Critique):

• Defining "Sexual Minority": The dichotomization of sexual orientation into "heterosexual" vs. "nonheterosexual" is a practical necessity due to sample size but inherently limits the understanding of diverse experiences within the sexual minority community. As noted by the authors, specific subgroups like bisexual individuals or mostly heterosexual individuals may face unique challenges. Future research should strive for more granular distinctions.
• Concurrent Measurement of Self-Esteem: While acknowledged as a limitation, the concurrent measurement of self-esteem with anxiety at 17.5 years means the causal direction is still ambiguous. Low self-esteem could precede and contribute to anxiety, or anxiety could erode self-esteem. Future studies should measure self-esteem at an earlier, intermediate time point to clarify this.
• Limited Scope of Gender Nonconformity: The CGN measures focused on gender-typed behaviors and interests, not gender identity or gender dysphoria. Given the increasing recognition of transgender and gender-diverse identities, future research should incorporate measures that capture these aspects, as gender identity can be a significant source of minority stress and interact with sexual orientation in complex ways. The changing significance of CGN for girls in later models suggests a more intricate relationship than a simple direct or mediating effect.
• Generality of Bullying Measure: The unstandardized and broad "bullying by another person" measure, while expedient, doesn't distinguish between general bullying and homophobic/transphobic bullying. The latter is likely a more potent and specific stressor for sexual minority youth. More precise measures would offer clearer insights.
• Remaining Unexplained Risk: The persistence of a significant, albeit reduced, odds ratio for sexual minority status even after accounting for bullying and self-esteem highlights that other factors are contributing. These could include internalized homophobia, discrimination in broader social contexts (e.g., employment, healthcare), family rejection, or the constant vigilance required to navigate a heteronormative world. Future research should aim to quantify these remaining minority stress factors.

Applicability to Other Domains: The methodological rigor, particularly the sequential logistic regression for evaluating mediating factors, could be applied to other areas of public health and social science research investigating disparities (e.g., racial/ethnic disparities in health outcomes, socioeconomic disparities in educational attainment). The emphasis on longitudinal data and diagnostic outcomes (where applicable) provides a template for robust research designs. The findings also underscore the need for inclusive public health policies and school-based interventions that target bullying and foster self-esteem for all adolescents, with specific attention to the unique vulnerabilities of sexual minority youth.