Camouflaging in autistic men and women

The authors said their camouflaging score was a proxy and proposed the fixes themselves: compare it with other quantifications of camouflaging, recompute it as a regression residual once a large enough sample exists, and build an alternative measure from social imitation using the Self-Monitoring Scale or the Multidimensional Iowa Suggestibility Scale.

Question

Do discrepancy, residual and self-report camouflaging scores in the same autistic adults agree, and do any of them predict how non-autistic observers rate those people?

Hypothesis

Discrepancy and residual camouflaging scores predict observers' thin-slice ratings of social presentation, beyond measured autistic traits, better than self-reported intent does.

Attempt type

evaluation
Vocabulary note: This is a psychometric validation study in human participants. 'Evaluation' is the nearest value because what is under test is the behaviour of measurement instruments rather than a biological hypothesis, but none of the five values covers psychometrics.

Design

One cohort of 250 autistic adults without intellectual disability, each given the three inputs the original study used (Autism Diagnostic Observation Schedule, Second Edition, module 4; Autism Spectrum Quotient; Reading the Mind in the Eyes Test), so that both the original difference score and a within-sample regression residual of observed presentation on self-reported traits and mentalizing can be computed from identical data. Each person also completes the self-report Camouflaging Autistic Traits Questionnaire, an informant version of it, and the Revised Self-Monitoring Scale, and is filmed in two scenarios of differing social-evaluative demand. Ten-second clips are rated for first impressions by at least 127 naive non-autistic raters, following Belcher et al. Primary analysis is a multi-method correlation matrix with bootstrapped intervals, plus pre-registered tests of each index's incremental prediction of observer ratings beyond trait and mentalizing scores. Duration about two years.

Controls

• designated A non-autistic comparison group of equal size, assessed on every measure including the filmed interaction and the observer ratings.
  Rules out: That the indices track general social skill or social anxiety rather than anything specific to autistic presentation.
• within-subject Every participant is filmed twice, once in a job-interview scenario and once talking with a trusted person, following the context-discrepancy method of Ai et al.
  Rules out: That an index showing no spread is measuring a stable trait, when in fact it is insensitive to the social demand that is supposed to drive camouflaging.
• internal-positive A social-skill measure and a social-anxiety measure entered as discriminant comparators that each camouflaging index must not simply duplicate.
  Rules out: That an index presented as camouflaging is really indexing social competence or fear of negative evaluation, the confound Arnold et al. flag and the one Hannon et al. tested.

Sensitivity

With 250 participants, an observed correlation of 0.25 carries a 95% interval of about 0.13 to 0.36 (Fisher z, computed in this session), which separates a convergence of r = 0.20, the self-report to discrepancy estimate in Hannon et al. 2022, from one of r = 0.46, their informant-report to discrepancy estimate. 194 participants give 80% power for r = 0.20 at alpha 0.05, two-tailed. The design is therefore sized to estimate convergence precisely rather than merely to reject zero.

Would count as

Supported → positive · opposed → negative-no-association
Inconclusive if: Agreement among the naive raters is too low to form a stable criterion. Morrison et al. 2019 found that variability in first impressions of autistic adults is driven more by the rater than by the person rated, so a low-reliability criterion would make every index look invalid for reasons that have nothing to do with the indices.

Already done

• A self-report camouflaging instrument was built and validated two years after this paper, with masking, compensation and assimilation subscales. (Hull et al. 2019, Journal of Autism and Developmental Disorders, doi:10.1007/s10803-018-3792-6)
  So: The self-report arm of this design is an existing instrument, not something to be invented, so the study becomes a comparison rather than a development exercise.
• A systematic review found 16 distinct camouflaging measurement tools, all at a preliminary stage of psychometric evaluation, and then tested convergence: self-report and informant-report questionnaires correlated r = 0.47 with each other, and with a discrepancy measure r = 0.20 (95% CI -0.06 to 0.43) and r = 0.46 (0.23 to 0.64) respectively. No measure correlated with social skills. (Hannon, Mandy and Hull 2022, Autism Research, doi:10.1002/aur.2850)
  So: The concurrent validity the paper called untested has been tested once and came out weak for self-report, which is why this design needs an external criterion rather than another questionnaire-to-questionnaire comparison.
• In 174 autistic and high-autistic-trait young adults, self-report camouflaging correlated only with the self-report-based discrepancy score, not with the mentalizing-based one, and undiagnosed high-trait adults scored higher on discrepancy measures than diagnosed autistic adults. (Milner et al. 2022, Autism Research, doi:10.1002/aur.2873)
  So: Discrepancy and self-report indices are measuring different things, effectiveness versus intent, so the design must keep them separate rather than treat disagreement as error.
• In 674 Dutch adults, the Camouflaging Autistic Traits Questionnaire showed no measurement invariance between autistic and non-autistic people, and its correlation with an Autism Spectrum Quotient minus Autism Diagnostic Observation Schedule discrepancy score ranged from r = 0.04 to 0.28 in the 90 autistic adults who had both. (van der Putten et al. 2022, Research in Autism, doi:10.1016/j.rasd.2022.102072)
  So: Confirms weak convergence in a second, larger, pre-registered sample, and warns that any cross-group comparison in this design needs invariance tested rather than assumed.
• In 80 autistic and non-autistic adults filmed in conversation and rated from ten-second clips by 127 non-autistic peers, self-reported camouflaging intent did not predict first impressions, although better first impressions were associated with a later age of autism diagnosis. (Belcher et al. 2022, Journal of Autism and Developmental Disorders, doi:10.1007/s10803-021-05221-3)
  So: The external-criterion study has been run for self-report intent and came out null, which both supplies the protocol for this design and narrows its novelty to the discrepancy and residual indices.
• A context-discrepancy assessment of camouflaging, framed as impression management, was validated in 23 autistic and 25 non-autistic adults across two scenarios with different social-evaluative pressure; autistic adults reported greater inauthenticity, monitoring and anxiety as demand rose. (Ai, Yu, Jacob and Lai 2026, Molecular Autism, doi:10.1186/s13229-026-00715-2)
  So: Supplies the two-context manipulation used here as a within-subject control, and shows the field has moved from a single trait score to a context-sensitive one.

Searched: Searched Europe PMC and OpenAlex for camouflaging combined with discrepancy, residual, measure, impression management, self-monitoring, suggestibility, imitation, first impressions and thin slice, and read the abstracts of the three convergence studies; I found no published camouflaging measure built from the Self-Monitoring Scale or the Multidimensional Iowa Suggestibility Scale, and no abstract that states a regression-residual camouflaging score has actually been computed, although Hannon et al.'s review counts 16 tools without itemising them in the abstract.

Feasibility

A clinical autism research group with assessors trained and licensed on module 4 of the Autism Diagnostic Observation Schedule, plus an online panel for naive raters · 18 to 30 months · moderate, because the cost driver is trained assessor time for 500 diagnostic observations plus double coding, not equipment or reagents
Blockers: Licensing and trained raters for module 4 of the Autism Diagnostic Observation Schedule; Recruiting 250 autistic adults willing to undergo a full diagnostic observation and be filmed; Ethics and consent for re-use of participant video by naive raters; No published test-retest reliability for the module 4 algorithm score in autistic adults, so the ceiling on a discrepancy score's reliability is unknown

Strongest objection

A specialist would say the design smuggles in a contested criterion. How a stranger rates a ten-second clip is not what autistic people mean by masking, which they describe as effort, monitoring and inauthenticity. On that view an index that fails to predict observer ratings may still be the valid one, and Belcher et al. have already shown self-reported intent does not predict first impressions without anyone concluding the questionnaire is invalid. A second objection is statistical: a difference score inherits the unreliability of both components, so weak convergence may be measurement error rather than construct divergence, and the study would need the reliability of each input before it could tell those apart.

Not derivable from what is published

• Test-retest reliability of the module 4 algorithm score in autistic adults, needed to bound the reliability of any discrepancy or residual score
• Whether any published camouflaging measure already uses the regression-residual method the authors proposed
• Inter-rater reliability of thin-slice first-impression ratings at the clip length chosen
• Published variance of Revised Self-Monitoring Scale scores in autistic adults

The authors named the three designs that would settle direction, and since then the field has done the third one many times over and the first two barely at all, so the useful experiment is the longitudinal one with the within-person separation the cross-sectional models cannot give.

Question

In autistic adults, does camouflaging predict later depressive symptoms once the reverse path is modelled at the within-person level?

Hypothesis

In a three-wave cohort, the within-person path from camouflaging to later depressive symptoms is larger than the path from depressive symptoms to later camouflaging.

Attempt type

evaluation
Vocabulary note: This is a prospective observational cohort with repeated measures in human participants. None of the five values covers a longitudinal human study, and 'wet-lab' or 'computational' would be plainly wrong, so 'evaluation' is used as the least misleading.

Design

300 autistic adults without intellectual disability, recruited from clinics and research registries, stratified by sex assigned at birth and by gender identity recorded separately. Three waves six months apart. At each wave: the Camouflaging Autistic Traits Questionnaire with its three subscales, a depression measure, a generalised anxiety measure, and an online Go/No-Go task yielding signal-detection sensitivity and response criterion as in the original study. At wave one, 14 days of daily diary reporting of camouflaging episodes and mood, to estimate the within-person variance the model needs. Primary analysis is a random-intercept cross-lagged panel model that separates within-person cross-lagged paths from stable between-person differences, with a pre-registered contrast of the camouflaging-to-depression path against its reverse. Secondary models substitute signal-detection sensitivity for depression and test sex and gender moderation as a latent interaction.

Controls

• within-subject The random-intercept specification uses each participant as their own comparison across waves, so only within-person change contributes to the cross-lagged paths.
  Rules out: Confounding by stable between-person characteristics such as autistic trait level, co-occurring attention deficit hyperactivity disorder, education and perceived stigma, which inflate the cross-sectional correlations this design is meant to improve on.
• known-positive-task The Go/No-Go task must reproduce the already reported group difference in signal-detection sensitivity between autistic adults and a non-autistic comparison sample before any camouflaging to sensitivity path is interpreted.
  Rules out: That a null camouflaging to sensitivity path reflects a task that is not working in an online administration, rather than an absent relationship.
• designated A non-autistic comparison cohort followed on the same schedule with the same instruments.
  Rules out: That any camouflaging to depression path is a general property of effortful impression management rather than something specific to autistic people, which matters because Ai et al. found camouflaging and mental health are linked in the general population too.

Sensitivity

Two parts, one derivable and one not. Derivable: the sex and gender moderation claims in the original paper can be costed exactly. Detecting the reported depression difference between autistic men (r = 0.533) and autistic women (r = 0.030) at 80% power needs about 53 per sex (Fisher z, computed in this session), against the 30 per sex the paper had, which is why that interaction came out at p = 0.068. Detecting the signal-detection difference between women (r = 0.432) and men (r = 0.233) needs about 313 per sex, so no cohort of realistic size will settle that one and this design should not promise it. Not derivable: the exact power for the cross-lagged paths, because it depends on the within-person variance and six-month stability of the camouflaging questionnaire, which no published study reports. 300 participants give 80% power for a simple correlation of about 0.17, and the wave-one diary component is included specifically to supply the variance components needed to replace that placeholder with a real figure before the main analysis is locked.

Would count as

Supported → positive · opposed → refuted
Inconclusive if: Attrition leaves fewer than about 200 participants with all three waves, or the within-person variance in camouflaging over six months is near zero, in which case there is no within-person change for a cross-lagged path to explain and the model returns stable between-person differences dressed as dynamics.

Already done

• A systematic review and meta-analysis of 22 studies and 5897 autistic participants found moderate positive meta-correlations between camouflaging and anxiety, depression and social anxiety and a small negative one with wellbeing, but reported that every included study was underpowered for small effects, many used insufficiently validated measures, and that longitudinal studies are still needed to establish causality. (Khudiakova, Russell, Sowden and Surtees 2024, Research in Autism, doi:10.1016/j.rasd.2024.102492)
  So: Confirms the association is real and the direction is still unestablished seven years on, so the contribution of this design is direction and not association.
• In a large online sample of autistic adults, self-reported camouflaging predicted generalised anxiety, social anxiety and depression, but only to a small extent beyond autistic traits and age, and there was no camouflaging by gender interaction. (Hull et al. 2021, Molecular Autism, doi:10.1186/s13229-021-00421-1)
  So: Directly fails to reproduce this paper's male-specific depression association, so the moderation test must be pre-registered as a confirmatory contrast rather than offered as an expected result.
• A path model in 225 autistic adults explained about 25 percent of the variance in camouflaging from psychological and sociocultural predictors, with fear of negative evaluation the strongest, and remained cross-sectional. (Zhuang et al. 2025, Autism, doi:10.1177/13623613241262110)
  So: The structural-equation option the authors listed has been taken up repeatedly, so proposing another cross-sectional model would add nothing; only the temporal designs are still open.
• In 71 autistic adults completing an eight-week modified group cognitive behavioural therapy for social anxiety, social anxiety fell with moderate effect sizes while camouflaging scores did not change, although individual reductions in social anxiety correlated with reductions in the compensation and assimilation subscales. The study was uncontrolled. (Roisenberg, Boulton, Thomas and Guastella 2025, Autism Research, doi:10.1002/aur.70103)
  So: The intervention option the authors listed has been attempted once without a control arm and with a null on the camouflaging measure, which both motivates a longitudinal design first and warns that the questionnaire may not move over two months.
• A position piece argues for a transactional impression management framework in which camouflaging is a property of person-by-context transactions, and calls for separate measures of motivation, ability, effortfulness and perceived effectiveness. (Khudiakova, Alexandrovsky, Ai and Lai 2025, Autism Research, doi:10.1002/aur.3299)
  So: Implies a single six-monthly score may be the wrong unit of analysis, which is why the daily diary is built into wave one rather than left as an extension.
• A protocol for 150 non-autistic, self-identifying autistic and diagnosed autistic adolescents aged 15 to 18 combines self-report, caregiver report, seven days of daily diaries with embedded cognitive tasks, electroencephalography, hair cortisol and a three-month follow-up. (Klein et al. 2026, BMJ Open, doi:10.1136/bmjopen-2025-111221)
  So: A team is already collecting the intensive longitudinal data in adolescents, so an adult cohort should coordinate measures with it and can borrow its variance components rather than duplicate the pilot.

Searched: Searched Europe PMC for camouflaging combined with longitudinal, prospective, structural equation, mediation, two time points, follow-up, over time and waves, and for camouflaging with intervention; I found many cross-sectional path and mediation models, one uncontrolled pre-post therapy study, one protocol with daily diaries and a three-month follow-up, and no published multi-wave cohort study of camouflaging and mental health in autistic adults.

Feasibility

A clinical research group with a standing autistic adult cohort or a national research registry and an online assessment platform · 30 to 36 months including the final wave and analysis · moderate, because every measure is self-report or an online task, so the cost is coordination, retention payments and a risk protocol rather than assays
Blockers: Retention across 18 months in a population with high mental health burden; No published minimal important difference for the camouflaging questionnaire, so a non-zero but tiny path cannot be interpreted clinically; A safeguarding protocol for suicidality disclosure at every wave; Whether the Go/No-Go version used in the original study is still available in a comparable online form

Strongest objection

The strongest objection is that camouflaging may not be a stable person-level quantity at all. If it is a transaction with each social context, then asking someone twice a year how much they camouflage measures a self-concept rather than a behaviour, and a cross-lagged model on that self-concept will produce a clean coefficient that means very little. A reviewer who holds the transactional view would say the daily diary should be the primary design and the six-monthly waves the secondary one, and would also point out that the autistic-to-non-autistic comparison cohort invites a measurement invariance problem that has already been reported for this questionnaire.

Not derivable from what is published

• Within-person variance and six-month test-retest stability of camouflaging questionnaire scores in autistic adults, needed for exact power on the cross-lagged paths
• A minimal important difference for the camouflaging questionnaire
• The meta-analytic correlation values behind the 2024 review, whose full text is not in Europe PMC, so its effects are described here only as the abstract states them
• Whether an online Go/No-Go administration reproduces the sensitivity values reported in the original cohort

The authors could not tell whether the female advantage they measured was specific to autistic people, and they specified what would settle it: the same external minus internal operationalization applied to non-autistic people, using a measure of social communicative behaviour other than the diagnostic observation schedule that is sensitive to individual differences in the non-autistic range.

Question

Among non-autistic adults, is there a sex or gender difference in observation-based camouflaging, and is it smaller than the difference seen among autistic adults?

Hypothesis

Non-autistic women and men differ by no more than d = 0.3 on an observation-based camouflaging residual, while autistic women exceed autistic men by about d = 0.75.

Attempt type

evaluation
Vocabulary note: A cross-sectional comparative study in human participants with blinded behavioural coding. 'Evaluation' is the nearest of the five values; none describes a group-comparison study with an equivalence test as its primary analysis.

Design

Four groups of 180, matched on age and full-scale intelligence quotient: autistic women, autistic men, non-autistic women, non-autistic men. External presentation is coded not from the diagnostic observation schedule, which has no spread in non-autistic adults, but from the Contextual Assessment of Social Skills, a short conversation with a confederate scored for vocal expressiveness, gestures and overall involvement, as used by Corbett et al. Internal status is the Autism Spectrum Quotient and the Reading the Mind in the Eyes Test, as in the original. Camouflaging is the standardized residual of coded presentation regressed on the two internal measures across the whole sample, so one scale applies to all four groups. Primary analysis is a group by sex and gender analysis of covariance on the residual, with the non-autistic contrast judged by two one-sided equivalence tests against a margin of d = 0.3. Measurement invariance of the internal measures across groups is tested first and reported whatever it shows.

Controls

• designated The two non-autistic groups are the comparison the original study lacked, assessed on exactly the same instruments and coded by the same blinded raters.
  Rules out: That the autistic female to male difference is one instance of a general female advantage in social self-presentation rather than something about being autistic.
• adversarial-label Coders are blind to diagnostic status, clip order is randomized, and a subset of clips is presented twice under a deliberately incorrect group label to quantify coder bias directly.
  Rules out: That the group and sex pattern is produced by coders expecting autistic men to present worse, which is the examiner stereotype confound the original paper itself raised as a limitation of the diagnostic observation schedule score.
• internal-positive The self-report camouflaging questionnaire is administered to all four groups, and the established pattern must reproduce before the new residual measure is interpreted: autistic women above autistic men by roughly d = 0.27, and no difference between non-autistic women and men.
  Rules out: That a null equivalence result reflects an unrepresentative sample rather than an absent difference, since a sample that cannot reproduce the known self-report pattern cannot be trusted on a new measure.

Sensitivity

180 per group gives 80% power for two one-sided equivalence tests at alpha 0.05 against a margin of d = 0.3 (137 per group needed) and 80% power for a conventional two-sided test of d = 0.3 (174 per group needed), both computed in this session. The autistic female to male contrast is amply powered: the pooled discrepancy-method effect of d = 0.755 from the 2025 meta-analysis needs only 28 per group. The limit should be stated in the protocol: ruling out a non-autistic difference as small as d = 0.2 would need about 392 per group, which this design cannot do, so a result inside the equivalence bound means 'no difference larger than small', not 'no difference'.

Would count as

Supported → positive · opposed → refuted
Inconclusive if: The coded presentation measure shows a ceiling among non-autistic adults, so the residual has almost no variance in those two groups. That is the exact failure the diagnostic observation schedule has, and it would mean the study reproduced the original limitation with a different instrument.

Already done

• In 306 autistic and 472 non-autistic adults, autistic women scored higher than autistic men on self-reported camouflaging, with no gender difference among non-autistic adults on any subscale, controlling for age and autistic traits. (Hull et al. 2020, Autism, doi:10.1177/1362361319864804)
  So: The core question is already answered for self-report, which is why this design is restricted to the observation-based operationalization and framed as equivalence rather than discovery.
• A meta-analysis of 15 studies ran 17 separate analyses and found no difference between non-autistic women and non-autistic men on total self-reported camouflaging (pooled d = -0.077, 95% CI -0.248 to 0.094, k = 6, p = 0.379), while autistic women exceeded autistic men (d = 0.273, k = 11) and, on the discrepancy method, by d = 0.755 (95% CI 0.264 to 1.246, k = 4). Only one discrepancy-method analysis was possible and it covered autistic groups only. (Cancino-Barros, Villacura-Herrera and Castillo 2025, Scientific Reports, doi:10.1038/s41598-025-06137-z)
  So: Supplies both the effect size this design must beat and the exact residual gap: there is no pooled estimate of a non-autistic sex or gender difference on any discrepancy measure, because nobody has computed one in adults.
• In 84 children and adolescents with and without an autism diagnosis, two operationalizations were run with a neurotypical comparison group: autistic and neurotypical females showed similar social reciprocity whereas autistic males showed lower reciprocity than neurotypical males, and autistic males and females had similar theory of mind despite the females' higher reciprocity. (Wood-Downie et al. 2021, Journal of Autism and Developmental Disorders, doi:10.1007/s10803-020-04615-z)
  So: The design this issue asks for has been executed in young people, so an adult version must justify itself on age rather than on novelty, and can borrow the choice of an observed reciprocity measure in place of the diagnostic schedule.
• Autistic and neurotypical adolescents were compared on self-reported camouflaging, and an age by diagnosis interaction emerged once age was modelled, with autistic females resembling neurotypical peers on masking and assimilation items. (Jorgenson, Lewis, Rose and Kanne 2020, Journal of Autism and Developmental Disorders, doi:10.1007/s10803-020-04491-7)
  So: Shows the non-autistic comparison is routinely included now, and that age must be a covariate rather than an afterthought in any group by sex design.
• A context-discrepancy measure of impression management was validated in autistic and non-autistic adults, with the same scenarios administered to both groups and effects expressed as the change between a high-demand and a low-demand context. (Ai, Yu, Jacob and Lai 2026, Molecular Autism, doi:10.1186/s13229-026-00715-2)
  So: Offers an alternative to the residual approach that sidesteps the ceiling problem by using within-person change across contexts, and a reviewer may prefer it to the between-group residual proposed here.
• In autistic and general-population samples aged 10 to 83, camouflaging declined across adulthood among non-autistic people but stayed elevated among autistic people, and the questionnaire's factor structure did not fit the Swedish data well, with validity problems concentrated below age 15. (Lundin Remnelius and Bolte 2024, Journal of Autism and Developmental Disorders, doi:10.1007/s10803-023-05909-8)
  So: Means the non-autistic comparison must be age-matched tightly, because a non-autistic group drawn older than the autistic group would manufacture a group difference.

Searched: Searched Europe PMC and OpenAlex for camouflaging with non-autistic, neurotypical, gender differences, children and adolescents, and read the 2025 meta-analysis full text from Europe PMC to extract the pooled non-autistic sex and gender effect; the self-report form of this question is answered and a child version of the observation-based form has been run, while I found no adult study that administered an observation-based external measure to non-autistic men and women in order to compute a discrepancy or residual camouflaging score.

Feasibility

A lab with an established confederate-based social interaction paradigm and a trained blinded coding team, ideally one already using the Contextual Assessment of Social Skills · 24 to 36 months · moderate to high, because 720 in-person interaction sessions with confederates and double-coded video is labour-intensive even though no instrument is expensive
Blockers: Recruiting 360 non-autistic adults willing to complete trait questionnaires and a filmed interaction; Coder training and reliability on the interaction measure across 720 sessions; No published variance for the interaction codes in non-autistic adults, so the ceiling risk cannot be priced before a pilot; Deciding whether sex assigned at birth or gender identity is the grouping variable, which the original paper fuses as sex/gender throughout

Strongest objection

A residual computed across a mixed sample forces one regression on two populations whose relation between internal status and external presentation may differ in form, not just in level, in which case the residual is not comparable across groups and the equivalence test is meaningless. Measurement invariance has already failed between autistic and non-autistic adults for the self-report questionnaire, and there is no reason a residual should behave better. A second objection is that the question has aged: the field has moved from asking whether camouflaging is a female phenotype to asking what social contexts demand it, so a reviewer may call this a well-powered answer to a question that is no longer the live one.

Not derivable from what is published

• Variance of Contextual Assessment of Social Skills codes in non-autistic adults, needed to confirm the residual has usable spread in the comparison groups
• Whether the Reading the Mind in the Eyes Test is an acceptable internal measure in non-autistic adults, given its ceiling and its dependence on vocabulary, which the original paper itself flagged
• Which grouping variable the equivalence margin should apply to, sex assigned at birth or gender identity
• Any pooled estimate of a non-autistic sex or gender difference on a discrepancy measure, because the 2025 meta-analysis could run only one discrepancy analysis and it covered autistic groups only

The authors proposed this themselves as their immediate next step: follow camouflaging over development, and look for the camouflaging to anxiety link in children and young people that was absent in their adults. The anxiety half has since been answered cross-sectionally, so what remains is the within-person developmental course.

Question

Does camouflaging increase within individual autistic young people across adolescence, and does within-person increase precede increases in anxiety rather than follow them?

Hypothesis

Camouflaging rises within autistic young people across ages 10 to 20 but not in non-autistic peers, and a rise in one year predicts higher anxiety the next more than the reverse.

Attempt type

evaluation
Vocabulary note: An accelerated longitudinal cohort study in children and young people. As with the other designs, none of the five values describes a human developmental cohort, and 'evaluation' is used as the least misleading.

Design

An accelerated longitudinal design: three age cohorts recruited at 10, 13 and 16 years, 150 autistic young people per cohort plus 50 non-autistic per cohort, each followed annually for four years so that the design spans ages 10 to 20 while every participant contributes a within-person slope. Each year: caregiver-report camouflaging at all ages and self-report camouflaging from age 15, given the reported validity problems below that age; a discrepancy index built as in Ross et al. from caregiver-reported autistic traits minus clinician-observed presentation; a child anxiety measure; and a brief executive function battery. Analysis is a parallel-process multilevel growth model with camouflaging and anxiety estimated jointly, testing whether within-person change in camouflaging at one year predicts anxiety the next above the reverse path. Measurement invariance across age and informant is tested before the growth model and reported either way.

Controls

• designated Non-autistic cohorts at the same three starting ages, followed on the same annual schedule.
  Rules out: That any rise is ordinary adolescent self-presentation. This control is load-bearing here because camouflaging has been reported to decline with age in non-autistic people while staying elevated in autistic people, so an autistic slope has to be read against a non-flat comparison.
• within-subject Annual repeated measures on the same individuals, with between-person differences absorbed by random intercepts.
  Rules out: That a cross-sectional age gradient reflects cohort, diagnostic-era or referral effects rather than development, which is the limitation every existing age finding in this literature carries.
• internal-positive At baseline the cohort must reproduce two established cross-sectional findings before any slope is interpreted: higher camouflaging in autistic girls than autistic boys, and camouflaging predicting internalizing symptoms with age, gender and intelligence quotient controlled.
  Rules out: That a flat or null trajectory reflects an unrepresentative sample or a broken measure rather than an absence of developmental change.

Sensitivity

This cannot be set from what is published, and saying so is the honest output. A growth model's power depends on the within-person variance and the variance of individual slopes in camouflaging across adolescence, and no published study reports either; the cross-sectional age findings are all between-person, and the one cross-cultural validation reports test-retest reliability without within-person variance components. Two things would settle it: the within-person variance from the daily diary and three-month follow-up data in the 2026 adolescent protocol, or a two-wave one-year pilot of about 60 autistic adolescents run before the main cohort is costed. One anchor is available as a floor rather than a power statement: a camouflaging to internalizing association was detectable in 733 autistic children and adolescents with age, gender and intelligence quotient controlled, so a cross-sectional association of that kind is within reach at that scale.

Would count as

Supported → positive · opposed → negative-no-association
Inconclusive if: Measurement invariance across age or across informant fails, so that the same questionnaire does not measure the same thing at 11 and at 19 and no single trajectory can be estimated. This is a live risk rather than a formality: the self-report questionnaire's factor structure has already been reported as a poor fit below age 15.

Already done

• In 733 autistic children and adolescents aged 4 to 17, camouflaging predicted internalizing symptoms including anxiety and depression with age, gender and intelligence quotient controlled, and adolescents showed a larger discrepancy between caregiver-reported and clinician-observed autistic traits than younger children did. (Ross, Grove and McAloon 2023, Autism Research, doi:10.1002/aur.2859)
  So: The anxiety half of this issue is answered, and the caregiver minus clinician discrepancy index is already defined and used at scale, so this design adopts it rather than inventing one.
• In autistic and non-autistic adolescents, social camouflaging was associated with internalizing symptoms in both groups. (Bernardin, Lewis, Bell and Kanne 2021, Autism, doi:10.1177/1362361321997284)
  So: Shows the association is not autism-specific in adolescence, which is why a non-autistic comparison cohort is required rather than optional.
• In autistic and general-population participants aged 10 to 83, camouflaging declined through adulthood among non-autistic people but remained elevated among autistic people, while the questionnaire's factor structure fitted the data poorly and validity problems were concentrated below age 15. (Lundin Remnelius and Bolte 2024, Journal of Autism and Developmental Disorders, doi:10.1007/s10803-023-05909-8)
  So: Sets both the comparison trajectory and the hard constraint that self-report cannot carry the younger cohorts, which is why caregiver report runs at every age here.
• In 58 autistic adolescents aged 13 to 18, fewer executive function difficulties predicted greater use of camouflaging and of the compensation subscale specifically, with a small total variance explained. (Hull, Petrides and Mandy 2021, Autism Research, doi:10.1002/aur.2407)
  So: Supports including an executive function battery and echoes this paper's signal-detection finding in women, so the cognitive strand can be tested developmentally rather than assumed.
• In 115 autistic and non-autistic adolescents matched on age and social anxiety severity, masking and impression management related to social anxiety severity through social cognitions rather than to autistic traits, and factor analysis showed construct overlap between camouflaging items and social anxiety safety behaviours. (Lei, Leigh, Charman, Russell and Hollocks 2024, Journal of Child Psychology and Psychiatry, doi:10.1111/jcpp.13884)
  So: Warns that a camouflaging to anxiety path in young people may be partly item overlap, so the design must report the two measures' shared items and test the path with overlapping items removed.
• A protocol for 150 adolescents aged 15 to 18 collects camouflaging by self-report, caregiver report and seven days of daily diaries, with electroencephalography, hair cortisol and a three-month follow-up. (Klein et al. 2026, BMJ Open, doi:10.1136/bmjopen-2025-111221)
  So: This is the nearest thing to the authors' next step now in the field, and it will supply the within-person variance components this design cannot derive, so the cohort should be planned to follow it rather than in parallel.

Searched: Searched Europe PMC and OpenAlex for camouflaging with adolescents, youth, children, longitudinal, age effects and developmental course, and read the abstracts of the adolescent studies; the camouflaging to anxiety link at younger ages is established cross-sectionally in several samples including one of 733 young people, the age trajectory has been described cross-sectionally from ages 10 to 83, and I found no published within-person longitudinal study of camouflaging across adolescence, only one protocol with daily diaries and a three-month follow-up and one qualitative study of retrospective change over time.

Feasibility

A developmental autism cohort with clinic access, annual in-person follow-up capacity and an established relationship with families · 5 to 6 years from first recruitment to the final wave · high, because 600 families with annual clinician-observed assessment over four years is the most expensive design of the four here
Blockers: Measurement invariance across ages 10 to 20 and across self and caregiver report; No camouflaging measure validated below age 15; Attrition across four annual waves in a developmental sample; The ethical question of asking autistic young people to report repeatedly on masking, which may itself prompt reflection that changes the behaviour being measured

Strongest objection

A developmentalist would say the design cannot separate two things it treats as one: growth in the ability to camouflage and growth in the social demand to do so. Both rise across adolescence, and annual self and caregiver report will not tell them apart. The same reviewer would add that if masking is a response to stigma, the real changes cluster around school transitions and diagnosis events rather than spreading evenly across years, so an event-anchored design with bursts of intensive sampling would detect what annual sampling smooths away. A third objection is that caregiver report of camouflaging is close to incoherent, since successful camouflaging is by definition what the caregiver does not see.

Not derivable from what is published

• Within-person variance and slope variance of camouflaging scores across adolescence, needed for any power statement about trajectories
• A camouflaging measure validated below age 15
• Measurement invariance of the caregiver-report camouflaging questionnaire across the 10 to 20 range
• The degree of item overlap between the camouflaging and anxiety measures chosen, which determines how much of any path is shared content