Siegel, 1998; Smith et al., 2000). IQ is an important variable, particularly for approaches that claim “recovery,” because “recovery” implies intellectual functioning within the average range. However, these results are difficult to interpret for a number of reasons. First, variability among children and variability within an individual child over time make it nearly impossible to assess a large group of children with autism using the same test on numerous occasions. Within a representative sample of children with autism, some children will not have the requisite skills to take the test at all, and some will make such large gains that the test is no longer sufficient to measure their skills. This is a difficulty inherent in studying such a heterogeneous population as children with autistic spectrum disorders.
The challenge to find appropriate measures and to use them wisely has direct consequences in measuring response to treatment. For example, there is predictable variation in how children perform on different tests (Lord and Schopler, 1989a). Children with autism tend to have the greatest difficulty on tests in which both social and language components are heavily weighted and least difficulty with nonverbal tests that have minimum demands for speed and motor skills (e.g., the Raven’s Coloured Progressive Matrices [Raven, 1989]). Comparing the same child’s performance on two tests, given at different times—particularly a test that combines social, language, and nonverbal skills, or a completely nonverbal test—does not provide a meaningful measure of improvement. Even within a single test that spans infant to school-age abilities, there is still variation in tasks across age that may differentially affect children with autism; this variation is exemplified in many standard instruments such as the Stanford-Binet Intelligence Scales (Thorndike et al., 1986) or Mullen Scales of Early Development (Mullen, 1995).
Generally, IQ scores are less stable for children first tested in early preschool years (ages 2 and 3) than for those tested later, particularly when different tests are used at different times. In one study (Lord and Schopler, 1989a), mean differences between test scores at 3 years or younger and 8 years and older were greater than 23 points. These findings have been replicated in other populations (Sigman et al., 1997). Thus, even without special treatment, children first assessed in early preschool years are likely to show marked increases in IQ score by school age (Lord and Schopler, 1989b), also presumably reflecting difficulties in assessing the children and limitations of assessment instruments for younger children.
Studies with normally developing children have indicated that there can be practice effects with developmental and IQ tests, particularly if the administration is witnessed by parents who may then, not surprisingly, subsequently teach their children some of the test items (Bagley and McGeein, 1989). Examiners can also increase scores by varying breaks,