data_dictionary.md

Data dictionary

The statistical pipeline (code/statistical_tests.R) consumes eight semicolon-delimited CSV files. These files contain individual-level and clinical data and are not redistributed in this repository.

To run the analysis, place the files in a folder and point the script at it:

TOMASD_DATA_DIR=/path/to/data Rscript code/statistical_tests.R

Cohort: N = 126 children (one row per subject). Group label values across files: asd, adhd, ad, NT (sometimes also coded as clin vs ctrl, or as the original French abbreviations TSA/TDA/TA in data_QD.csv). All wide-format files share row order so they can be joined on row index.

Opponent indexing convention

The five opponents reported in the manuscript map to the CSV columns as follows:

Manuscript label	Wide CSV column	Long CSV factor level	Description
Random-biased (RB)	hs_0 / pc_0	RB	Stochastic opponent biased to one side
0-ToM	hs_1 / pc_1	X0tom / X0ToM	Reinforcement-learning-style opponent
1-ToM	hs_2 / pc_2	X1tom / X1ToM	Models depth-0 opponent
2-ToM	hs_3 / pc_3	X2tom / X2ToM	Models depth-1 opponent
Sequential (mean)	hs_seqMoy / pc_seqMoy	moyseq / seqMoy	Mean across deterministic sub-sequences A/B/C

Prefixes: hs_ = Hide-Seek (social framing); pc_ = A Picky Chick (non-social framing).

---

Files (wide format, one row per subject)

perf_nomissing_wgroup.csv

Per-subject win rate against each opponent in each framing.

column	type	description
(rowname)	int	subject row index (1..N)
hs_0–hs_3, hs_seqMoy	float	Win rate, social, per opponent (see mapping above)
pc_0–pc_3, pc_seqMoy	float	Win rate, non-social, per opponent
BrainUs_ID	string	Pseudonymous platform UUID
group	string	asd / adhd / ad / NT
group_bin	string	clin / ctrl
group_asdbin	string	ASD / nonASD

ktom.csv

Per-subject sophistication phenotype: posterior expected ToM depth k, computed from the softmax over k-ToM model evidences.

column	type	description
Social	float	Sophistication, social framing
Non-social	float	Sophistication, non-social framing
Groupe	string	Diagnostic group
Groupe_bin	string	clin / ctrl
Gr_asdbin	string	ASD / nonASD

ftom.csv

Per-subject flexibility phenotype: 1 minus the dot product of softmaxed model evidences across framings (high = different model best supports behavior in social vs non-social).

column	type	description
Flexi	float	Flexibility index (0–1)
Groupe	string	Diagnostic group
Gr_asdbin	string	ASD / nonASD

alternance_nomissing.csv

Per-subject alternance scores per opponent and framing, z-scored within the cohort.

⚠️ Naming caveat. The CSV column is labeled alternance, but the aggregated group means in aggregated/phenotypes.csv show ASD higher than NT — consistent with the manuscript's perseveration finding, not its inverse. The values appear to follow the manuscript convention (positive = perseveration). The R script preserves this and labels its y-axis "Perseveration/alternance Score". Confirm with the analysis authors before relying on the sign for new analyses.

column	type	description
ID	int	Subject index
hs_0–hs_3, hs_moyseq	float	Alternance z-score, social, per opponent
pc_0–pc_3, pc_moyseq	float	Alternance z-score, non-social, per opponent
framing...7, framing...14	string	"social" / "non social" sanity columns
group, group_bin, group_asdbin	string	as above

clinicalscales_nomissing.csv

Per-subject parent-reported clinical scales (de-identified, sequential ID).

column	type	description
X / (rowname)	int	Subject index
group_bin, Groupe	string	Group
cbcl_anxiousdistress, cbcl_socialproblems, cbcl_thoughtproblems, cbcl_totalproblems	float	CBCL subscale scores
BRIEF	float	BRIEF Global Executive composite
CONNERS	float	Conners-3 composite
SCARED	float	SCARED total
SCQ	float	SCQ Lifetime total
*_nona	float	Same scales after missing-data imputation

data_QD.csv

Master sheet combining demographics, clinical battery, raw performance, and derived phenotypes. Used by the R script for the clinic data frame and the final merge with phenotypes.

⚠️ Privacy. Contains potentially re-identifiable columns (ID BrainUs, decimal age, sex, ADOS dates) and is released only under the data-access agreement. Variable names mix English, French (e.g. FlexibilitÈ.brief), and unicode (𝑘 ̂_Social); see column headers for the authoritative list.

---

Files (long format, repeated-measures rows)

perf_nomissing_wgroup_long.csv

Performance reshaped for the 3-way mixed ANOVA (Group × Framing × Opponent). Two rows per subject (one per framing); opponent is a column.

column	type	description
ID / id	int	Subject index (repeated across framings)
framing	string	"social" / "non social"
group	string	Diagnostic group
RB, X0ToM, X1ToM, X2ToM, seqMoy	float	Win rate per opponent (see mapping)

alternance_nomissing_long.csv

Alternance reshaped likewise. Same row structure as the perf long format.

column	type	description
ID / id	int	Subject index
framing	string	"social" / "non social"
group	string	Diagnostic group
RB, X0tom, X1tom, X2tom, moyseq	float	Alternance z-score per opponent

---

Provenance

The CSVs are produced from the MATLAB fitting pipeline in code/ (k-ToM model inversion via the VBA toolbox), followed by per-subject extraction of the three phenotypes. The extraction scripts that bridge the MATLAB outputs to these CSVs (and the alternance/perseveration estimation in particular) are not in this repository.