Here's the norming ref on the WPPSI-IV:
-- Updated normative sample standardized on 1,700 children ages 2:6�7:7
-- The normative sample was stratified to match current U.S. census data based on sex, race/ethnicity, parent education level, and geographical region for each group
Once they've normed and made their curve, that's it. Anything else becomes procedural modifications, like creating extended norms to address ceilings.
During test creation, questions can be tuned against specialized samples. Like you run the test for 60 kids who are evenly distributed in the 130 to 160 range (based on another test) and you correlate specific questions to those IQ ranges to create questions tied to higher end skill nuances. Testing outside of the normative sample is not included in the final curve.