Empirically derived subtypes of pervasive developmental disorders: a cluster analytic study.
Four data-driven autism flavors exist—pick teaching targets that match the flavor, not the label.
01Research in Context
What this study did
The team fed scores from common autism tests into a computer. They used cluster math to spot hidden groups.
Data came from the kids with PDD. All had DSM-III-R labels like autism or PDD-NOS.
What they found
Four clear clusters popped out. Each cluster had its own social, talk, and play style.
One group talked on time but flapped a lot. Another group had almost no words and poor eye contact.
How this fits with other research
Cordova et al. (1993) saw the same mess of skills two years earlier. Their KIDIES clips showed kids act very different with mom versus teacher.
Meier et al. (2012) later asked if high autism is just social fear. The 1995 clusters say no—some kids fear little yet still line up toys for hours.
Safer-Lichtenstein et al. (2019) warn that most autism studies only test white boys with high IQ. The 1995 sample was small and also pale, so we need bigger, browner re-runs of the cluster work.
Why it matters
Stop saying "my autism kid." Say "my cluster-2 kid" or "my cluster-4 kid." Each group needs its own plan. Cluster-1 loves scripts—use them for teaching. Cluster-3 needs pictures, not words. Match the plan to the profile and you waste fewer teaching hours.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →Graph your client’s last ADOS and VB-MAPP scores; see which 1995 cluster they land in; pick one skill that fits that cluster’s profile.
02At a glance
03Original abstract
A cluster analytic study was conducted to empirically derive behaviorally homogeneous subtypes of pervasive developmental disorders (PDD). Subjects were clustered based on a broad range of behavioral symptoms which characterize autism. Behavioral variables were measured using several of the standardized psychometric instruments most commonly employed in assessing autistic individuals. The cluster solution indicated the presence of four distinct groups. Validity checks generally confirmed significant between-group differences on independent measures of social, language, and stereotyped behaviors. In addition, the four-group cluster solution was compared to previously developed typological systems of PDD (i.e., subcategories based on IQ early onset, styles of social interaction, and DSM-III-R diagnosis). Results generally supported both the behavioral homogeneity of the four subgroups and also several important between-group differences. The potential utility of using cluster analyses to explore subtypes of PDD is discussed.
Journal of autism and developmental disorders, 1995 · doi:10.1007/BF02178188