Sex-specific metrics for success: Gaps in social word use are bigger for autistic girls than boys

Abstract

Autistic girls are often diagnosed late, missed, or misdiagnosed, which can negatively impact quality of life and mental health. Although research shows the social profiles of autistic girls differ from boys in systematic ways that might explain gaps in diagnosis, little is known about how autistic girls’ social language compares to their same-sex non-autistic peers. This study investigated social words—words that make reference to other people—produced by 138 age- and IQ-matched autistic and non-autistic youth (ages 6–15) during one Autism Diagnostic Observation Schedule, Second Edition task. Girls used significantly more social words than boys across both diagnostic groups. There was a larger gap in social word production between autistic girls and non-autistic girls than autistic boys and non-autistic boys, with non-autistic girls using the most social words. Non-autistic girls’ social language behavior—including their social word production—sets an especially high bar for autistic girls, who often report trying to blend in with other girls. Growing evidence of the distinct social language profiles of autistic and non-autistic girls versus boys should guide researchers and clinicians to assess autism in ways that are sensitive to sex-associated differences and develop interventions that consider the norms of youth’s target social circles.

Lay Abstract

Autism is often diagnosed later in girls and women as compared to boys and men. More research is needed to understand how autism presents differently in girls. This study investigates how autistic and non-autistic youth aged 6 to 15 years use social words (e.g. “friend,” “mom,” “help,” “talk”) during an interview about friends, relationships, and marriage as part of an autism diagnostic assessment. Overall, girls used more social words compared to boys and talked more about friends. Specifically, non-autistic girls used the most social words in comparison with other groups. Highly social language produced by non-autistic girls may make it especially hard for autistic girls to blend in with other girls and could lead them to engage in more camouflaging behaviors to hide their autistic characteristics. With such different average social language behavior from girls and boys, researchers should consider adapting autism assessments and interventions to support the unique needs of autistic girls.

Introduction

Girls are less likely to be diagnosed with autism than boys (Shaw et al., 2025), even when they have comparable autistic characteristics (Dworzynski et al., 2012). Girls are also diagnosed disproportionately later in life (Fusar-Poli et al., 2022; Rutherford et al., 2016). Diagnostic disparities may be attributable to sex differences in the behavioral presentation of autism across development (McFayden et al., 2023) and may contribute to later emotional and social difficulties (Mandy et al., 2022).

Childhood social interactions have many universal elements, but certain domains—including social language (i.e. language used for social purposes)—differ by sex during middle childhood. For example, non-autistic girls generally spend more time engaged in conversation, self-disclose more, and have more extended dyadic interactions than non-autistic boys (Rose & Rudolph, 2006). These sex differences in social language behavior in non-autistic populations mean autistic girls who are trying to fit into neurotypical norms have a unique set of goalposts that may be harder to scale.

Some evidence suggests that verbally fluent autistic girls produce a unique pattern of social language and communication behaviors that reflect population-level sex differences. For example, research shows autistic girls employ language styles and produce content that—consciously or not—could help them to “blend in” with non-autistic peers. These subtle differences include distinct patterns of filled pauses (Parish-Morris et al., 2017), prosodic elements (Cho et al., 2023), cognitively rich content words like “think” and “know” (Boorse et al., 2019), and plural pronouns (Song et al., 2020). A person’s overall tendency to talk about social topics has been considered a marker of social phenotype in autism (Klin, 2000) and has recently been shown to differentiate autistic girls and boys (Cola et al., 2022).

A study by Cola and colleagues (2022) found that autistic girls (N = 25; M = 10.66 years) used more social words than autistic boys (N = 76; M = 10.15 years) during the conversation and reporting sections of the Autism Diagnostic Observation Schedule, Second Edition (ADOS-2; Lord, Rutter, et al., 2012). However, the small exploratory sample of non-autistic children (N = 34) in that study rendered it unclear whether the findings reflected autism-specific differences or a broader trend that also distinguishes non-autistic boys and girls.

To address this knowledge gap, we investigated social word use in a matched sample of autistic and non-autistic girls and boys during the Friends, Relationships, and Marriage section of the ADOS-2. This task elicits discussion of a variety of social relationships and has been used in prior research (Cola et al., 2022; Landau, 2025; Libster et al., 2023; Mattern et al., 2023). Social words were defined as words that make reference to other people (e.g. “classmates,” “them”). The primary question—whether increased social word use in girls is a broader population-level difference that emerges in autistic children too, or whether these differences are specific to autism—has implications for understanding behavior in autistic children’s potential target peer groups, and could inform the development of sex-sensitive social supports.

Methods

Participants

One hundred thirty-eight verbally fluent autistic and non-autistic youth (ages 6–15) participated in research at the Children’s Hospital of Philadelphia. Participant sex was characterized using parent/caregiver-reported assigned sex at birth. Youth participated in a series of studies that included autism diagnostic assessments and IQ testing (S1). Participants with complete data (ADOS-2 recordings and IQ testing) were selected from the larger pool. Participants were excluded if they had a full-scale IQ or verbal IQ ⩽ 70. Diagnostic groups were matched on average age and IQ (Table 1). Autistic girls and boys were group-matched on parent- and clinician-rated autistic traits.

Table 1.

Demographic and clinical characteristics of participants (means, standard deviations, and ranges).

	Autistic (N = 82)		Non-autistic (N = 56)		Effects
Sex ratio	26f, 56m (68% Male)		24f, 32m (57% Male)		p = 0.25
Race	Asian or Pacific Islander: 4.88% (n = 4) Black or African American: 3.66% (n = 3) Multiracial: 6.10% (n = 5) White: 85.36 % (n = 70)		Asian or Pacific Islander: 1.79% (n = 1) Black or African American: 33.93% (n = 19) Multiracial: 10.71% (n = 6) White: 53.57% (n = 30)		–
Maternal Education	High school or less: 2.44% (n = 2) Bachelor’s or less: 48.78% (n = 40) Graduate degree: 36.59% (n = 30) Not reported: 12.20% (n = 10)		High school or less: 7.14% (n = 4) Bachelor’s or less: 48.21% (n = 27) Graduate degree: 33.93% (n = 19) Not reported: 10.72% (n = 6)		–
	Female	Male	Female	Male	Sex	Dx	Sex in AUT
Age (years)	10.69 (1.57) 8.7–14.0	10.26 (2.31) 6.1–15.2	10.35 (2.47) 6.4–15.2	10.23 (2.30) 6.9–15.1	p = 0.47 d = −0.13	p = 0.77 d= −0.05	p = 0.38 d = −0.21
Full-Scale IQ	105.81 (1 1.84) 78–131	106.96 (14.12) 81–148	106.71 (15.16) 82–134	109.78 (10.53) 86–133	p = 0.45 d = 0.13	p = 0.41 d = 0.85	p = 0.72 d = 0.09
Verbal IQ	106.31 (12.71) 79–134	108.14 (14.21) 79–150	108.63 (16.44) 80–148	108.81 (12.71) 86–131	p = 0.69 d = 0.07	p = 0.63 d = 0.08	p = 0.58 d = 0.13
Non-verbal IQ	105.92 (13.58) 73–133	106.55 (14.28) 86–143	103.17 (14.99) 79–132	108.16 (9.93) 89–128	p = 0.28 d = 0.19	p = 0.88 d = −0.03	p = 0.85 d = 0.05
ADOS-2 Total	10.77 (4.99) 3–23	10.98 (4.28) 4–21	1.33 (1.60) 0–7	1.75 (1.55) 0–5	–	–	p = 0.84 d = 0.05
ADOS-2 SA	8.19 (4.14) 3–17	8.41 (3.62) 3–16	1.21 (1.25) 0–5	1.50 (1.39) 0–4	–	–	p = 0.81 d = 0.06
ADOS-2 RRB	2.58 (1.86) 0–7	2.57 (1.68) 0–7	0.13 (0.45) 0–2	0.25 (0.51) 0–2	–	–	p = 0.99 d = 0.003
SCQ Total Score	19.88 (5.85) 8–31	19.16 (7.41) 5–38	2.33 (2.74) 0–10	2.33 (1.81) 0–7	–	–	p = 0.66 d = −0.10

CSS: ADOS-2 calibrated severity score; SA CSS: social affect calibrated severity score; RRB CSS: repetitive behaviors/restricted interests calibrated severity score. Chi-square tests with Yates’ continuity correction tested for diagnostic group differences in sex ratio and maternal educational attainment. p-values and Cohen’s d values for main effects of sex and diagnosis are shown, as well as p and Cohen’s d values of sex differences in the autism group only. One male non-autistic participant was missing SCQ data.

Measures

Participants completed the ADOS-2 Module 3 (Lord, Rutter, et al., 2012), a clinician-administered assessment of autism characteristics. Parents/caregivers completed the Social Communication Questionnaire (SCQ; Rutter et al., 2003). Autism diagnoses were made by doctoral-level clinicians using the clinical best estimate approach (Lord, Petkova, et al., 2012) which prioritizes Diagnostic and Statistical Manual of Mental Disorders (5th ed., DSM-5) criteria.

Language sample

Language data were drawn from the ADOS-2 Module 3 Friends, Relationships, and Marriage section and orthographically transcribed by annotators. LIWC-22 software (Boyd et al., 2022) was used to calculate the overall number of words, and the number of social, friend, and family words produced by participants (S2).

Statistical approach

Data were analyzed using generalized linear regression models (GLM) in R (“lme4” package) with sex, diagnosis, age, and IQ as covariates. Estimated effects, standard errors (SE), z-values, and p-values are provided. Dependent variables were positive, interval, and non-normally distributed (Shapiro–Wilk test ps < 0.001); these data were modeled using a Poisson distribution with a log link. Significance values for planned pairwise comparisons of estimated marginal means were corrected using the Tukey method. GLM modeled relationships between social language and clinical phenotype. For each dependent variable (i.e. social, friend, and family words), we removed outliers ±2 SDs from the mean in each subgroup. To determine whether social word production was associated with clinician-rated phenotype in autism, ADOS-2 Social Affect scores were modeled as a function of social word production.

Dependent variables

Following the transcription and statistical analysis procedures outlined in Cola et al. (2022), preliminary analyses examined total word count controlling for sex, diagnosis, age, and IQ. Results revealed a significant interactive effect of sex and diagnosis on overall word count (p = 0.009), with the non-autistic group producing significantly more words than the autistic group (Table 2). In addition, non-autistic girls produced significantly more total words than autistic girls (estimate: −0.08, SE: 0.01, z = −5.71, p < 0.001), autistic boys (estimate: −0.27, SE: 0.01, z = −23.54, p < 0.001), and non-autistic boys (estimate: 0.24, SE: 0.01, z = 18.83, p < 0.001). Thus, to understand social word use in the context of each individual’s overall volubility, subsequent analyses were conducted on the number of social words standardized per 1000 words. To assess the dimensional relationship between language features and autism characteristics, autism characteristics were measured using ADOS-2 domain scores.

Table 2.

Characteristics of participant speech by sex and diagnosis (means, standard deviations, and ranges).

	Autistic		Non-autistic		Sex*Dx effect
	Female	Male	Female	Male
Total Words	452.72 (265.22) 99–1172	368.13 (222.08) 21–977	484.21 (217.40) 122–1058	382.00 (216.57) 100–1138	p = 0.01 est: −0.05
Social Words (per 1000)	165.96 (41.46) 76–238	144.40 (30.83) 53–206	176.54 (25.43) 136–222	144.97 (30.36) 88–236	p = 0.03 est: −0.06
Friend Words (per 1000)	8.72 (7.60) 0–30	7.44 (7.55) 0–33	10.71 (6.76) 0–23	8.44 (6.30) 0–24	–
Family Words (per 1000)	9.75 (7.43) 0–28	8.91 (10.17) 0–53	8.38 (7.62) 0–30	9.44 (9.14) 0–28	–

Effect sizes for GLM are reported as unstandardized effects. The final GLM model (glm(variable ~ age.z + IQ.z + dx*sex, data =lang.par, family = “Poisson”)) accounts for age (centered), IQ (centered), and examines the interactive effect of sex and diagnosis as the primary predictor variable.

Results

Social words

A GLM controlling for age and IQ revealed a significant interactive effect of sex and diagnosis on participant social word production (estimate: −0.06, SE: 0.02, z = −2.18, p = 0.03). Tukey-corrected pairwise comparisons revealed the effect was driven by non-autistic girls producing more social words than any other subgroup (Figure 1, Table 2). Non-autistic girls produced significantly more social words than autistic girls (estimate: −0.06, SE: 0.02, z = −3.12, p = 0.001), autistic boys (estimate: −0.20, SE: 0.02, z = −10.44, p < 0.001), and non-autistic boys (estimate: 0.13, SE: 0.02, z = 9.06, p < 0.001). Autistic girls produced significantly more social words than autistic boys (estimate: 0.13, SE: 0.02, z = 6.81, p < 0.001) and non-autistic boys (estimate: 0.13, SE: 0.02, z = 5.82, p < 0.001). Autistic boys and non-autistic boys produced similar amounts of social words (estimate: −0.01, SE: 0.02, z = −0.30, p = 0.99).

Figure 1.

Social word use by sex and diagnosis presented as mean values. Error bars represent standard error of the mean. ***p ⩽ 0.001.

Friend words

To assess the types of social words participants produced, we investigated use of friend and family words in each diagnostic group separately, in alignment with procedures outlined in the work by Cola et al. (2022). In the autistic group, a GLM controlling for age and IQ revealed a significant effect of sex on friend word production (estimate: −0.18, SE: 0.08, z = −2.13, p = 0.03), with autistic girls producing significantly more friend words than autistic boys. In the non-autistic group, a GLM revealed significant effect of sex on participant friend word production (estimate: −0.23, SE: 0.08, z = −2.59, p < 0.001), with girls producing more friend words than boys.

Family words

GLMs controlling for age and IQ revealed no significant effect of sex on participant family word production in the autistic (estimate: −0.09, SE: 0.08, z = −1.15, p = 0.25) and non-autistic group (estimate: 0.13, SE: 0.09, z = 1.37, p = 0.17).

Correlations with clinician-rated phenotype

After controlling for age, IQ, and sex, social word production significantly predicted ADOS-2 SA total scores in the autistic group (estimate: −0.004, SE: 0.001, z = −3.67, p < 0.001), indicating that participants who used fewer social words during this task were rated as having greater autism characteristics on this measure. When examined in each sex separately, this effect held in autistic girls (estimate: −0.005, SE: 0.002, z = −2.47, p = 0.01) and autistic boys (estimate: −0.004, SE: 0.002, z = −2.45, p = 0.01). In the non-autistic group, social word production did not significantly predict ADOS-2 SA scores (estimate: 0.005, SE: 0.004, z = 1.11, p = 0.27).

Discussion

This study investigated sex differences in social phenotype, as measured by natural language samples, during one task of a diagnostic assessment in autistic and non-autistic youth. We addressed a critical question: Are sex differences in the social language of autistic children unique to autism (Cola et al., 2022) or do they reflect population-level sex differences that occur in boys and girls regardless of diagnosis? Building on prior research, we found that girls produced more social words than boys, reflecting a population-level sex difference in language use that emerges in both non-autistic and autistic children. Girls also talked more about friends than boys, and social words predicted clinician-rated social phenotype. This study is therefore an important extension of prior research on sex-based variation in social language in autism (Cola et al., 2022) because it demonstrates that sex differences in autistic girls’ and boys’ social language reflect population-level differences that also appear in autism.

Societal expectations

Why do girls—autistic or not—talk more about social topics than boys? One possibility is that social demands and standards differ by sex, and this impacts what girls and boys say and do. In Western cultures, the social contexts of girls and boys differ from an early age across a variety of domains. For example, children demonstrate an awareness of sex/gender-based expectations about behavior by age 12 months, when they show a preference for gender-typical toys (Boe & Woods, 2018). In terms of emotional sex/gender-based expectations, research suggests that girls are expected to display more positive and internalizing emotions, while it is more acceptable for boys to display externalizing emotions (Chaplin, 2015). Girls’ friendships are characterized by social conversation and self-disclosure, while boys’ more often revolve around activities like organized sports (Rose & Rudolph, 2006; Rudolph & Dodson, 2022). Thus, girls and boys develop in different contexts, and the metric for achieving “social success” varies by sex/gender (i.e. for girls, emotional sharing and self-disclosure may be more contextually important than for boys).

The implications for autism are clear. Due to higher demands for social and emotional competence in girls, autistic girls and women may feel pressured to try and achieve gender-specific expectations about social language behavior by hiding their autistic traits or compensating socially (e.g. camouflaging, masking; Bernardin et al., 2021; Hull et al., 2020; Jorgenson et al., 2020), more so than autistic boys and men. With such different average social language behavior from girls and boys, researchers should consider adapting autism assessments and interventions to support the unique needs of autistic girls and women. Specifically, when making determinations about what is “normal,” it may be beneficial for researchers/clinicians to compare autistic girls’ social language to the social language produced by same-sex non-autistic peers, rather than to the language of autistic boys.

Several limitations of the current research should be considered when interpreting our results. First, we utilized a frequency-based lexical analysis, which does not provide information regarding the appropriateness or context of the social words. Second, we assessed language in one conversational context (i.e. a clinical interview about friends, relationships, and marriage). Thus, it is unknown whether our findings would replicate in other, more naturalistic contexts (e.g. conversations with a peer). Third, it is possible that participants’ social word production was impacted by their level of comfort discussing these topics with an unfamiliar adult, which could relate to their experiences with gendered socialization (e.g. girls may be conditioned to talk more readily about social topics than boys). Fourth, we did not collect information about participants’ gender identity, which limited our ability to examine nuances related to gender expression and social communication. Fifth, our sample was largely White, educated, and did not include individuals with ID, which may limit generalizability of our results.

Conclusion

From birth, girls and boys are exposed to sex-differentiated societal pressures and influences that shape behavior. In this study, we demonstrated that observed sex differences in one area of behavior—social language—transcend diagnostic status. We found that girls used more social words than boys, and non-autistic girls used the most social words of all. Heightened sociality in non-autistic girls establishes especially high expectations for autistic girls attempting to blend in with their non-autistic peers. For autistic girls, observed sex differences in the “average” social language behaviors of non-autistic peer groups should illuminate new paths to accurate autism assessment, and inform personalized supports to help autistic girls.

Data availability statement

The datasets generated and/or analyzed during the current study are not publicly available. Requests to access the dataset should be directed to the final author.