Cluster analytic identification of autistic preschoolers.

Rescorla (1988) fed Child Behavior Checklist (CBCL) scores into a cluster program. The goal was to see if math could spot autistic preschoolers without a clinician watching.

The sample was small. The study does not give exact numbers. Kids were already in preschool programs.

The numbers fell into clear piles. One pile held almost all autistic children. Two other piles held typical kids and kids with other problems.

Inside the autism pile, the program split kids again by anxiety level and daily living skills. This gave two autism sub-groups ready for different help.

Mélinia et al. (2026) repeated the idea with 188 preschoolers and added developmental tests. They found the same three main piles, showing the 1988 pattern still holds.

De Kegel et al. (2016) looks like a clash. They say CBCL alone is too weak for screening (only 59-70 % accuracy). The gap is age and aim: Rescorla (1988) used clusters inside a clinic group, while Alexandra tried one quick cut-off on a wide age range.

So et al. (2013) built a 10-item CBCL/TRF short scale after seeing the 1988 clusters. Their scale keeps the good part (high negative predictive value) while fixing the false-positive problem.

You can borrow the cluster idea today. Run CBCL scores through free cluster software. If a child lands in the high-anxiety autism cloud, plan extra coping skills before you start table work. The method costs nothing and gives you a data-driven reason to individualize the first month of therapy.