Are we under-estimating the association between autism symptoms?: The importance of considering simultaneous selection when using samples of individuals who meet diagnostic criteria for an autism spectrum disorder.
ASD symptom links look weaker in clinic-only samples, so check the sampling method before you trust correlation numbers.
01Research in Context
What this study did
Murray et al. (2014) built fake data sets and real clinic files.
They asked: do we lose true links between autism traits when we only look at kids who already have a diagnosis?
The team ran computer models that mimicked how clinics pick children for an ASD label.
What they found
The links between autism symptoms looked much weaker inside the diagnosed group.
In plain words, if you only test kids who cleared the diagnostic hurdle, you may think traits barely relate.
The real-world symptom ties are stronger than they appear in clinic-only samples.
How this fits with other research
Demello et al. (1992) warned that DSM-III-R calls too many kids autistic.
Louise shows one cost of that over-selection: it hides how traits truly hang together.
Chuthapisith et al. (2012) later proved the 3Di-Thai screener keeps English cut-offs.
Their work extends Louise’s point: good tools must keep the same thresholds across cultures or the same bias creeps back in.
Why it matters
When you read an article saying "social and repetitive scores barely correlate," check who was in the sample.
If every child already met strict DSM cut-offs, the numbers may under-sell the link.
Before you plan treatment based on weak trait ties, look for data that include the full range of symptom levels.
Get CEUs on This Topic — Free
The ABA Clubhouse has 60+ on-demand CEUs including ethics, supervision, and clinical topics like this one. Plus a new live CEU every Wednesday.
Pull the last ASD trait-correlation paper you cited and see if the sample was pre-diagnosed—if yes, treat the r-values as likely under-estimates.
02At a glance
03Original abstract
The magnitude of symptom inter-correlations in diagnosed individuals has contributed to the evidence that autism spectrum disorders (ASD) is a fractionable disorder. Such correlations may substantially under-estimate the population correlations among symptoms due to simultaneous selection on the areas of deficit required for diagnosis. Using statistical simulations of this selection mechanism, we provide estimates of the extent of this bias, given different levels of population correlation between symptoms. We then use real data to compare domain inter-correlations in the Autism Spectrum Quotient, in those with ASD versus a combined ASD and non-ASD sample. Results from both studies indicate that samples restricted to individuals with a diagnosis of ASD potentially substantially under-estimate the magnitude of association between features of ASD.
Journal of autism and developmental disorders, 2014 · doi:10.1007/s10803-014-2154-2