analysis#
The analysis package reproduces the four data-driven autism classes of Litman et al. (2025) and tests whether they hold within strata of age at diagnosis and diagnostic era.
Hypotheses
- \(H_0^A\) and \(H_0^D\): are the class profiles invariant, and is any drift small?
- \(H_0^B\): do the class proportions shift with diagnostic era and age at diagnosis?
- \(H_0^C\): is the supported number of classes four in every stratum?
- \(H_0^E\): does the drift have a direction along the axis?
- \(H_0^F\): is the drift spread evenly across the seven phenotype categories?
- \(H_0^G\): is the era drift spread evenly across instruments regardless of referent?
- \(H_0^H\): is the era drift an artefact of measurement timing?
- \(H_0^I\): does the genotype-to-class mapping hold across diagnostic timing?
- \(H_0^J\): does genetic drift track phenotypic drift?
Technical guides
- The pipeline and its cache
- Running the pipeline
- The cohort interface
- Parsing the SSC milestone ages
- Choosing the stratification bins
- Aligning stratum classes to the reference
- Measuring how far a class drifts
- Computing the invariance effect size
- Screening orderings with the displacement atlas
- Conditioning on demographics
- The score-based invariance test
- Attributing a class’s movement
- Splitting the era drift by referent
- Measuring prevalence drift
- Testing the number of classes
Appendix
Reference